PLoS Biology

Fonseca-Garcia, Citlali;Pettinga, Dean;Caddell, Daniel;Ploemacher, Hannah;Louie, Katherine;Bowen, Benjamin P.;Park, Joelle;Sanchez, Jesus;Zimic-Sheen, Alen;Traxler, Matthew F.;Northen, Trent R.;Coleman-Derr, Devin

doi: 10.1371/journal.pbio.3003526pmid: 41296785

Introduction Plants exist as holobionts, intimately associated with complex microbial communities that play critical roles in plant development, health, and stress tolerance [1]. These plant-associated microbiomes, particularly those colonizing the rhizosphere and root endosphere, can influence a wide range of plant phenotypes including nutrient acquisition, pathogen resistance, and abiotic stress resilience [2,3]. Under environmental stress, such as drought, plants undergo shifts in root exudation and physiology that reshape their microbiomes in ways that have been proposed to buffer the host against damage [4,5]. Understanding the composition and functional traits of microbial taxa that associate with plant roots under stress conditions is essential for improving crop resilience through microbiome-informed strategies [6,7]. One group of microbes repeatedly found to increase in relative abundance in the roots of drought-stressed plants is the Actinobacteria, particularly members of the genus Streptomyces [5,8–10]. Levels of Streptomyces increased in the root microbiome across a wide range of more than 30 angiosperm plant hosts, and overall levels of Streptomyces abundance correlated with positive host phenotypic outcomes during drought stress [9]. Some recent studies have also shown that bioinoculation with specific Streptomyces species in an agricultural setting mitigated drought stress, potentially through enhancing antioxidant enzymes and optimizing osmotic regulation [11]. Collectively, these prior studies demonstrate that the genus Streptomyces as a whole exhibits drought enrichment (DE), and may be an important source of plant growth promotion during drought. While Streptomyces enrichment under drought is well-documented, less is known about the specific functional diversity within this genus that may explain its ecological success in the rhizosphere and its variable effects on plant hosts. Evidence from prior studies suggested several potential causes for this DE: Streptomyces form spores, enabling survival during harsh conditions, they exhibit high osmotic stress tolerance, produce siderophores, and metabolize diverse plant-derived carbon substrates [12,13] Streptomyces species are renowned for their metabolic versatility, including the production of antibiotics, siderophores, osmoprotectants, and plant hormones, the presence and expression of these traits can vary dramatically even among closely related strains [14–16]. In drought-stressed rhizospheres, these functional differences could influence microbial fitness, competitive interactions, and ultimately, the degree of benefit conferred to the plant. Given that the genus Streptomyces is well-known to be fast-evolving [17–19], we hypothesized that Streptomyces DE in sorghum roots is not a uniform, lineage-level phenomenon, but instead driven by isolate-specific traits varying across a genetically diverse pool of endophytic strains. Furthermore, we propose that DE and its underlying causes—such as osmotic stress tolerance, secondary metabolite production, siderophore production, and utilization of drought-associated root exudates—could be uncoupled from the strains’ capacity to promote plant growth. To test these hypotheses, we began by performing a metagenomic analysis of the impact of drought on the sorghum root microbiome using both V3-V4 and full-length 16S rRNA amplicon sequencing, with an emphasis on understanding genotypic variation in Streptomyces and its relationship to DE. Using these data, we identified a collection of closely related Streptomyces root endophyte isolates obtained from Sorghum bicolor, an important feedstock and bioenergy crop [20], and subjected them to genomic, exometabolomic, and phenotypic characterization. We demonstrated that the single most dominant Streptomyces amplicon sequence variant (ASV) observable through V3-V4 16S rRNA sequencing in the sorghum root represents a wide diversity of isolates with substantial variation in genomic content and traits purported to be associated with the DE phenomenon. Through pangenomic analysis, we showed that both DE and plant growth promotion vary widely among this collection of isolates, and that neither trait correlates with the other or the underlying phylogenetic distance between them, underlining the importance of considering isolate-level genomic and phenotypic variation when exploring and predicting crop-microbe interactions. Results The genus Streptomyces has been repeatedly exhibited strong increased relative abundance under drought, relative to irrigated conditions (referred to as DE) through a variety of metagenomics sequencing techniques [5,8–10], but the degree to which this DE holds for individual root-associated Streptomyces remains unclear. To explore this, we performed a field experiment in which the root microbiomes of drought-stressed and control-irrigated sorghum were sampled and analyzed via 16S rRNA sequencing using both V3-V4 (Illumina) and full-length (PacBio) primer sets (Fig 1A). Interestingly, Actinobacterial amplification was lower in both conditions when the full-length 16S rRNA primers were used (S1 Table). In silico analysis of Actinobacteria, Proteobacteria, Bacteroidetes, and Firmicutes sequences in the reference database used for taxonomic classification revealed that a much smaller fraction of Actinobacterial sequences were “amplifiable” with the PacBio full-length primer set than with the V3-V4 primer set, potentially accounting for the discrepancy (S2 Table); numerous studies have demonstrated that use of different 16S rRNA primer sets leads to significant and reproducible differences in the resulting taxonomic profiles [21–23]. Importantly, however, both the V3-V4 and the full-length 16S datasets demonstrated significant Actinobacterial DE in the roots of drought-stressed sorghum. Further analysis of the V3-V4 16S rRNA dataset revealed a single dominant Streptomyces ASV (ASV650) in the root endosphere of sorghum as the most abundant (Fig 1B); it represented 12.3% of the total relative abundance within the roots of watered plants and 16.7% (P = 0.017, S1 Table) within drought-stressed roots. A comparison of the V3-V4 and full-length 16S rRNA ASVs reveals that ASV650 shares 100% nucleotide identity across the V3-V4 region with five distinct ASVs detected using the full-length 16S primer sets (Fig 1C). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. High-resolution taxonomic profiling reveals hidden diversity among root endophytic Streptomyces under drought and control conditions. (A) 16S rRNA amplicon sequencing of the root endosphere under drought and control conditions using both V3-V4 and full-length 16S rRNA primer sets. (B) Phylogenetic tree of all the Actinobacteria ASVs identified by V3-V4 16S rRNA amplicon sequencing (outer ring) or full-length 16S rRNA amplicon sequencing (inner ring) of the root endosphere of field-grown sorghum at order level. The red arrows highlight the single dominant Streptomyces ASV identified via V3-V4 and the five corresponding ASVs identified via full-length sequencing. The tree shown here represents topology only (branch lengths not to scale) and is intended to illustrate taxonomic relationships rather than evolutionary distances. (C) Plot of single nucleotide polymorphism (SNP) locations across the full-length 16S rRNA region of the five ASVs identified via PacBio sequencing, reveals 100% sequence conservation within the V3-V4 region and additional genotypic diversity elsewhere within the 16S rRNA gene. The data underlying this Figure can be found in https://doi.org/10.5281/zenodo.17554086. https://doi.org/10.1371/journal.pbio.3003526.g001 An exploration of the enrichment patterns of these five full-length ASVs (referred to as AC2 through AC36) demonstrated that all exhibit some level of DE in the root microbiome, although the level of enrichment varied (S1 Table). Only one of these ASVs (AC2) was detectable in roots of both drought and control treatments (watered sorghum); three of the other ASVs were only detected following drought stress, but are observed in multiple samples, with an average relative abundance varying between 1.94% and 3.15% of the total root microbiome readcounts. The final ASV, AC36 was only detected in a single drought-stressed root sample, and accounted for only 0.011% of total relative abundance. When combined, these five ASVs explained 1.39% of total relative abundance under watered conditions, and 9.68% under drought conditions, corroborating the pattern of enrichment observed for ASV650. Collectively, these data revealed that while both 16S rRNA sequencing approaches exhibit a general trend of DE for Streptomyces in the root microbiome, levels of enrichment differed between ASVs in the full-length 16S rRNA dataset. As full-length 16S rRNA sequencing revealed additional diversity in Streptomyces present within the sorghum root not observable via shorter amplicon sequencing, we hypothesized that root-associated Streptomyces isolates matching these ASVs may reveal further diversity in response and allow for deeper functional and physiological characterization of their growth in the rhizosphere. To enable this, we explored a previously described sorghum root endophyte isolate collection developed in our lab [24]; a set of 170 Streptomyces from this collection were genotyped via full-length 16S rRNA sequencing and screened for 100% nucleotide identity homology matches to the dominant ASV650 observed in our original V3-V4 dataset. In total, we identified 28 isolates from the collection that fit this criteria, representing exact matches to four of the five AC ASVs identified by full-length 16S rRNA sequencing (isolates matching AC36 were not identified). Visual inspection of all 28 Streptomyces isolates grown on spore induction media revealed large variation in pigmentation phenotypes (Fig 2A), suggesting additional layers of metabolic and/or physiological variation could be present within each AC group. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Phenotypic and metabolomic differentiation among Streptomyces isolates sharing a dominant root-derived V3-V4 16S rRNA genotype. (A) Image of the growth (on spore induction media, or SIM) for all 28 Streptomyces isolates with 100% identity to the dominant ASV identified in the root via V3-V4 16S rRNA sequencing. Isolates are arranged by the AC group (full-length 16S rRNA genotype) that they belong to, highlighting within-group differences in pigmentation patterns. Isolates selected for additional metabolomic, genomic, and phenotypic characterization (n = 12) are indicated by red circles around the corresponding petri dish. (B) Ordination of polar exometabolomic profiling of the 12 strains following growth on liquid tap water–yeast extract (TWYE) medium showing distinct clustering of AC group 2 and 5 (replication n = 4). AC groups are presented by colors: AC2, purple; AC5, light blue; AC6, green mist; AC8, green. Streptomyces isolate IDs are indicated in the corresponding plate. The data underlying this Figure can be found in https://doi.org/10.5281/zenodo.17554086. https://doi.org/10.1371/journal.pbio.3003526.g002 Next, from each group of isolates matching the four AC ASVs (referred to as AC groups), we selected three isolates for exometabolomic characterization following growth on three distinct media types (Fig 2A, total strains n = 12) versus uninoculated control media. Data from minimal media (TWYE) revealed clustering by AC group for isolates belonging to AC2 and AC5, with less distinction between isolates belonging to AC6 and AC8, (Fig 2B and Fig A in S1 Text). Additionally, these data suggest that metabolite profiles for some individual isolates (e.g., DC14) are different from others within their AC group. To test how growth on more rhizosphere-relevant carbon substrates might impact these results, all 12 isolates were next profiled using exometabolomics following growth on media containing ground sorghum root tissue from either drought-stressed or control-irrigated plants. These data demonstrate that isolates fed drought-stressed root tissue have distinct metabolic profiles from those fed control root tissue, and lend further support to the previous observation that isolates belonging to group AC2 and AC5 have distinct exometabolomic profiles from the other strains (Fig B in S1 Text). Next, we hypothesized that the observed differences in metabolic profiles among these 12 isolates may correlate with differences in other phenotypes, including those hypothesized to be associated with DE in the root environment. To test this, we assessed variation among the 12 isolates for several microbial phenotypes with proposed links to the observed DE of Actinobacteria including: general osmotic stress resistance [25], iron acquisition ability [5,26], and carbon resource utilization [24]. Importantly, the phenotypes showed strong between-isolate variability (S3 Table). This included single-locus traits, such as levels of the individual bacterial siderophore Ferrioxamine E-F (measured through exometabolomics, Figs C, D in S1 Text), as well as more complex phenotypic traits likely explained by multiple loci, such as osmotic stress tolerance and total siderophore production (Figs E, F in S1 Text). To assess whether phenotypic variation between these 12 isolates correlated with phylogenetic relatedness, whole genome sequencing of the 12 strains was performed using a combination of PacBio and Illumina sequencing, and both pangenomic and phenotypic information was plotted in Anvi’o (Fig 3) [27]. An exploration of the genomic data supports the general topology of the phylogeny established through full-length 16S rRNA sequencing, with most members belonging to a given AC group sharing greater genomic homology with other members of the group than those outside the group. Of note, we observed the greatest differences in genomic content and organization between clusters AC2 and AC5, while isolates belonging to AC6 and AC8 appeared more similar to one another. Next, the genomes for each strain were screened for copy number variation in genes putatively associated with DE, including iron acquisition, carbon resource utilization, and osmotic stress response (Fig 3). Genes associated with osmotic stress tolerance included ProP, which encodes an osmoregulatory proton symporter [28]; OmpR, a transcription factor regulating the production of outer membrane porins [29]; Kdp, member of an ATP-dependent potassium ion uptake system [30]; Trk, a potassium ion transporter [31]; and FabG, an essential reductase in the fatty acid biosynthesis pathway [32]. Importantly, for both the genomically-derived phenotypes (Fig 3, in blue) and the previously lab-quantified phenotypes (Fig 3, in orange) we found evidence of both inter- and intra-AC group variation. Collectively, these data suggest that phylogenetic relatedness among isolates does not predictsimilarity for DE-associated phenotypes. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Pangenomic visualization of the genomic organization of all 12 Streptomyces isolates representing each of the four AC groups. Hierarchical clustering is based on orthogroup presence/absence using Euclidean distance and Ward’s linkage method. Colored rings show orthogroup presence/absence per isolate, the outer ring shows the number of contributing genomes, and the second outer ring shows the number of genes present in each orthogroup. The metadata table shows 1) experimental phenotype data (orange) of total siderophore production (SP, halo size on CAS-LB media), growth under osmotic stress (OST, 1.0M sorbitol), and the relative production of four individual siderophores in response to media supplemented with drought- vs. well-watered root tissue (EF1 and EF2, ferrioxamine EF; H, O-methyl desferrioxamine H, D4, derrioxamine D4), and 2) genome analysis data (dark blue), showing the number of gene copies that match with specific COG function search terms related to iron acquisition (iron and Sid, siderophores), carbon resource utilization (G-3-P, glycerol-3-phosphate), and osmotic stress response (ProP, FabG, OmpR, Kdp, Trk). Scale values represent Z-scores, where darker blue indicates more gene copies, and darker orange indicates more siderophore production, more growth under osmotic stress, or a larger halo. The data underlying this Figure can be found in https://doi.org/10.5281/zenodo.17554086. https://doi.org/10.1371/journal.pbio.3003526.g003 Based on these results, we hypothesized that Streptomyces DE within the root environment may follow a similar pattern of trait variability across the genus [9,10,24]. To test this, we selected a larger cohort of 48 Streptomyces from our sorghum root-endophyte isolate collection for genomic characterization and DE phenotyping. This new cohort included all 12 isolates analyzed previously, plus an additional 36 isolates from our collection to maximize taxonomic breadth (based on V3-V4 16S rRNA sequence data). To provide an accurate estimate of phylogenetic relatedness among isolates, whole genome sequencing was performed (Fig 4); these data revealed 17 unique species (ANI > 95%) and 21 unique strains (ANI > 99%) across the cohort. To estimate functional diversity across the cohort, we estimated the core (n = 2,123) and pangenome (n = 17,557). To illustrate the heterogeneity of gene content at different taxonomic classifications, we compared the numbers of shared and unshared orthogroups—a set of genes descended from a single gene in the last common ancestor of the considered genomes [33] between genomes within the genus, species, and strain level. Even within strains, we observed high variability in shared gene content from a minimum of 5,835–8,085 orthogroups, and as many as 1,660 unshared orthogroups. Genome fluidity comparisons—a measure of gene-level dissimilarity across the pangenome [34]—within species ranged from 0.04 to 0.12 and within strain from 0.03 to 0.06, suggesting that differentiation by ANI underestimates functional differentiation. Collectively, the presented pangenome captured high levels of genetic diversity, even within closely related Streptomyces strains isolated from a single field site. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. Phylogenomic and comparative genomic analysis of 48 Streptomyces isolates. (A) Phylogenetic tree of all 48 Streptomyces isolates based on alignment of 138 single-copy genes and rooted by a Nocardioides outgroup. Relevant isolates are annotated with AC Group. Corresponding ANI heatmap aligned to tree tips shows levels of relatedness among isolates calculated with fastANI. Black boxes within the heatmap indicate unique strains (>99% ANI). (B) Pan- and core genome rarefaction simulations fit to power law and exponential decay curves, respectively using pagoo R package. (C) Shared and (D) unshared orthogroup counts between pairs of isolates compared at different taxonomic levels across the collection. Orthogroups identified by Orthofinder and taxonomic classification determined by pairwise ANI (Genus: <95%, Species: 95%–99%, Strain: >99%). Scale bar represents the mean number of substitutions per site. The data underlying this Figure can be found in https://doi.org/10.5281/zenodo.17554086. https://doi.org/10.1371/journal.pbio.3003526.g004 Next, to assess DE scores for this cohort, each individual Streptomyces isolate (n = 48) was grown in mono-association with young sorghum seedlings in sterilized calcined clay under either watered or drought conditions for a period of 4 weeks, at which point plants were harvested and root colonization for the isolate was assayed by qPCR using Actinobacteria-specific primers. A single DE score was calculated for each isolate by averaging replicates (n = 4). In support of our hypothesis, we observed high DE variance (1.16) across the 48 isolates with scores ranging from −2.48 (depleted) to 1.73 (enriched) (Fig 5A). The observation that many tested Streptomyces isolates exhibit depletion under drought when grown with sorghum roots lacking a normal rhizosphere microbiome stands in contrast to many observations made in native field contexts [9,10,24], and demonstrates that individually assessed Streptomyces phenotypes may vary considerably for this reportedly conserved trait. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. Drought enrichment of Streptomyces isolates and associated gene function reveal candidate mechanisms for root colonization under drought stress. (A) Drought enrichment (DE) scores for all 48 Streptomyces isolates measured by lineage-specific 16S rRNA qPCR amplification of bacterial DNA from sorghum root extracts after 4 weeks of growth under drought or irrigation treatment in gnotobiotic microbox containers. DE, log2(Drought/Water) where “Drought” refers to the mean abundance of 4 replicates under drought and “Water” refers to mean abundance of 4 irrigated replicates. Red bars represent average drought enrichment and blue bars represent depletion. (B) Orthogroups with significant, positive DE-copy number associations were tested for COG term enrichment relative to all analyzed orthogroups by hypergeometric test. Among the most significantly enriched COG terms Lipid transport and metabolism, and Inorganic ion transport and metabolism were multiple orthogroups with (C) essential fatty acid biosynthesis gene FabG (n = 8) and (D) osmolyte-proton symporter ProP (n = 7) annotations, respectively. The data underlying this Figure can be found in https://doi.org/10.5281/zenodo.17554086. https://doi.org/10.1371/journal.pbio.3003526.g005 We next hypothesized that the combination of complete genomes and variable DE scores for this cohort of related and environmentally co-habitating Streptomyces may enable us to identify genetic modules positively correlated with increased DE in the root. To identify pangenomic variation contributing to the observed variation in DE, orthogroup copy number variation was assessed across the pangenome. Orthogroups with suitable variance (>0.2) in copy number were tested for association with the DE phenotype using phylogenetic regression. For each orthogroup, a linear regression for DE as a function of orthogroup copy number was fit while simultaneously estimating the degree to which phylogenetic signal (Pagel’s λ) explains variation among the model residuals. Among analyzed orthogroups, only 4 out of 2,735 exhibited λ > 0. In total, 342 clusters passed our filtering thresholds for robust regression results including only a single cluster with any degree of phylogenetic signal estimated in its model. These results suggest an overwhelming lack of phylogenetic signal among individual orthogroup copy number variation correlating with DE. Next, results were tested for functional enrichments among the top DE-associated orthogroups. Significant results included two categories: inorganic ion transport and metabolism and lipid transport and metabolism. Across both of these high-level categories were two notable osmolyte-related orthogroups (Fig 5B). First, within the Lipid transport and metabolism Clusters of Orthologous Groups (COG) category, we detected multiple instances of FabG (COG1028), an enzyme essential for fatty acid biosynthesis. In total, 8 distinct orthogroups annotated as FabG displayed strong association between copy number and DE (Fig 5C). Second, we also detected association between gene copy number and DE in 7 independent orthogroups for ProP (COG0477), a proton symporter necessary for osmolyte transport (Fig 5D). Taken together, these results suggest that biosynthetic machinery and transporters for root and rhizosphere-enriched osmolytes may increase bacterial fitness under drought in the root environment. Finally, it has been proposed that the phenomena of Streptomyces DE in the root may support host growth and stress tolerance, and individual isolates of Streptomyces have been shown to confer such benefits to their hosts [24]. Given the high variability in DE responses observed across our 48 isolates, we next sought to determine if the observed degree of DE correlated with beneficial impact on host performance across this cohort of isolates. To test this, we analyzed shoot biomass data for each plant grown in the previous DE experiment (Fig 6). Importantly, we observed no evidence for correlation between plant growth promotion and DE (water: Spearman’s ρ = 0.16 P = 0.26; drought: Spearman’s ρ = −0.2, P = 0.17) or plant drought tolerance promotion with DE (Spearman’s ρ = 0.00 P = 0.99). Examination of the phylogenomic context of both DE and plant growth promotion demonstrated that neither trait appeared phylogenetically conserved among closely related isolates. For example, divergent DE was noted between sister isolates DC01 and SAI126. These isolates were among the highest and lowest DE performers, respectively. Interestingly, these strains also had strongly contrasting patterns of plant host growth promotion under watered conditions and drought tolerance promotion. Notably, SAI126 treatment—which exhibited negative DE—reduced plant biomass under watered conditions, yet positively impacted plant biomass and water content under drought. In contrast, DC01—which exhibited positive DE—increased plant biomass under watered and droughted conditions, but reduced water content under drought. Such variation among closely related bacteria suggests a lack of phylogenetic signal for DE, plant growth promotion, and plant drought tolerance promotion. This was confirmed with formal tests for two estimates of the effect of phylogeny upon DE trait variation (Fig 6) (Pagel’s λ = 7.3 * 10−5, P = 1; Blomberg’s K = 2.6 * 10−6, P = 0.35). Similar results were obtained in tests of the effect of phylogeny on plant growth promotion under watered conditions—Pagel’s λ = 7.3 * 10−5, P = 1; Blomberg’s K = 3.2 * 10−6, P = 0.35—and drought—Pagel’s λ = 7.3 * 10−5, P = 1; Blomberg’s K = 8.5 * 10−6, P = 0.12. Lastly, we found no effect of phylogeny on the impact of percent water content of the plant host under drought—Pagel’s λ = 7.3 * 10−5, P = 1; Blomberg’s K = 6.1 * 10−6, P = 0.3. Collectively, these data demonstrate a lack of connection between the ability to confer growth benefit or drought tolerance to the host, the degree of DE, and the phylogenetic context of the isolate. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 6. Phylogenomic analysis reveals that Streptomyces-mediated drought enrichment and plant growth promotion are variable and not phylogenetically conserved. Phylogenetic tree of all 48 Streptomyces isolates based on alignment of 138 single copy genes and rooted by a Nocardioides outgroup. Corresponding annotation columns at tree tips include AC Group, DE, ∆ Biomass (g) under watered conditions, ∆ Biomass (g) under drought conditions, and ∆ Water Content (%) under drought conditions. “DE” refers to enrichment of the isolate in sorghum roots in drought relative to watered conditions in a mono-association microbox experiment: DE = log2(Drought/Water) where “Drought” refers to the mean abundance of 4 replicates under drought and “Water” refers to mean abundance of 4 irrigated replicates as measured by lineage-specific qPCR. Isolates with significant mean differences (FDR < 0.05) between “Drought” and “Water” abundances by t test and adjusted for multiple testing with Benjamini-Hochberg method are outlined in black. ∆ Biomass (g) represents the mean difference in dry shoot biomass under Water – column “W” – and Drought – column “D” – conditions, respectively, relative to mock-treated control plants. The final column represents the mean difference in % Water Content of shoot tissue under drought conditions relative to mock-treated control plants. Tiles outlined in black represent significant differences (FDR < 0.05) relative to mock-treated plants by t test and adjusted for multiple testing with Benjamini–Hochberg method. Scale bar represents the mean number of substitutions per site. The data underlying this Figure can be found in https://doi.org/10.5281/zenodo.17554086. https://doi.org/10.1371/journal.pbio.3003526.g006 Discussion Our study provides a high-resolution genomic and phenotypic characterization of root-associated Streptomyces diversity and highlights a key limitation of using amplicon sequencing based characterization of the microbiome as the sole means of inquiry. We demonstrate that both traditional and current gold standard 16S rRNA sequencing approaches (variable-region and full-length, respectively) fail to fully capture the Streptomyces genetic variation present within the rhizosphere, obscuring important phenotypic and functional differences in the genus that are relevant to its survival in this environment and its impact on host physiology. In other studies, 16S rRNA markers have been found insufficient to infer metabolic differentiation among Streptomyces strains [35,36], potentially because loci responsible for secondary metabolite production are subject to lateral gene transfer [37], resulting in rapid evolution. It has also been observed that isolates of Streptomyces with 100% 16S rRNA nucleotide identity can have as low as 84.42% ANI [38], suggesting that a single ASV could theoretically capture the average abundance over the entire genus in a 16S dataset. The data in our study also suggest that even full-length 16S rRNA ASV-based phylogenies of root-associated Streptomyces are inconsistent with genome-wide ANI-based taxonomic assignments, something that has also recently been observed in a broad survey of publicly available genomes belonging to this genus [39]. These issues with 16S rRNA resolution are not unique to Streptomyces and have been reported in diverse microbiome studies from many fields. In medicine, 16S rRNA sequencing resolution fails to distinguish between pathogenic and non-pathogenic isolates of genera commonly identified in clinical samples, including Bacillus and Anaplasma [40–42]. In marine biology, the most abundant clades are widely distributed across the world, and although ecotypes are closely related, they display distinct niche adaptations to various parameters such as temperature, salinity, and hydrostatic pressure [43]. For example, the Synechococcus clades III and IV thrive in distinct oceanic conditions, but could not be resolved based on the V6 hypervariable region of the SSU rRNA gene [44]. In contrast, despite high phylogenetic variability, functional composition can be highly similar, asshown in bacterial communities associated with macroalga Ulva australis [45]. Collectively, these observations highlight the importance of integrating high-resolution, genome-resolved approaches, such as whole-genome sequencing or metagenomics, alongside 16S rRNA profiling to more accurately assess microbial diversity, resolve functionally relevant taxa, and better understand the ecological and physiological roles of microbes within complex communities. In this study, we assessed Streptomyces DE at both the isolate and metagenome scale. Based on our field data and many other field studies of droughted rhizospheres [8,10,24,46], we expected consistent, positive DE among the Streptomyces isolates tested in this study. In fact, only 16 of 48 isolates displayed positive DE scores when grown in mono-association with host sorghum. This discrepancy is likely explained by severaI key differences in the experimental design of the prior and current studies. First, the highly controlled and artificial abiotic environments of the gnotobiotic experiments described above likely differ from the field conditions wherein Streptomyces enrichment has been routinely observed. These differences include the complexity and carbon content of the soil substrate, soil pH and nutrient availability, diurnal temperature fluctuations, and the duration and intensity of drought [24]. Second, and perhaps more importantly, the microbial composition of the rhizosphere in our lab-based experiment differs dramatically from field studies. In the present study, we applied each isolate to sorghum seedlings in mono-association under gnotobiotic conditions such that no competition or coordination with other microbes (bacteria or fungi) occurs. It is therefore feasible that the observed Streptomyces DE in native contexts requires the presence of other microbes. For example, an increase in Streptomyces abundance during drought may require a direct or indirect (host-mediated) biological interaction with other members of the rhizosphere microbiome. Recent work has shown that some strains of Streptomyces can respond to chemical signals and other microbes in their environments to alter growth habit and increase rates of expansion [47,48]. Alternatively, observed increases in Streptomyces abundance in the field may result from the fact that 16S rRNA amplicon sequencing in a community context constrains the sum of all microbes quantified such that a decrease in abundance of one microbe necessitates the increase in the abundance of all others [49,50]. Indeed, in a prior field-based study [24], a quantification of phylum-level absolute abundance revealed that Actinobacteria collectively exhibit an absolute decrease in abundance following severe drought, but show significantly greater resilience to this stress than three other phyla tested (Proteobacteria, Firmicutes, and Bacteroidetes). Taken together, these observations suggest that the apparent enrichment of Streptomyces under drought observed in field studies may reflect a combination of relative abundance shifts due to the depletion of more drought-sensitive taxa and context-dependent ecological interactions that cannot be replicated in isolation, underscoring the importance of studying microbial dynamics within their native community and environmental frameworks. Our study also explored the extent to which DE is shared among closely related Streptomyces, revealing a lack of phylogenetic conservation in this trait. Such incongruence of phenotype and genotype has been observed with other traits in Streptomyces [51]. This phenomenon may be explained by distinct genomic characteristics of Streptomyces. Evolutionary studies in the genus suggest that extremely high recombination rates exceed the impact of mutation, decays clonality, and challenges traditional tree shaped phylogenies [52–54]. Such recombination events have been documented at both small and large timescales [54,55] with major implications for evolution within the genus. In addition, extensive horizontal gene transfer via integrative and conjugative elements has been documented among isolates derived from a millimeter-scale rhizosphere population [37,56]. This highly dynamic transfer among closely related isolates has the potential to impact fitness—both positive and negative—and facilitate adaptation to environmental stressors, including drought. In summary, rampant recombination and horizontal gene transmission decays clonality among Streptomyces and likely challenges phylogenetic conservation of DE and other traits among our cohort. Despite the lack of phylogenetic signal for DE, our analyses successfully identified orthogroups strongly associating copy number with DE. Among the most enriched functions were multiple FabG orthogroups. This gene produces a β-ketoacyl–acyl carrier protein reductase, an essential enzyme of fatty acid synthase (FAS) II which is responsible for the first reductive step in the elongation cycle of fatty acid biosynthesis [32,57]. FAS II activity is required to produce fatty acids, the primary role of which is hydrophobicity of membrane lipids—critical barriers responsible for cellular integrity, particularly in periods of abiotic stress such as drought. ProP, another gene represented by multiple significant orthogroup associations is a proton symporter. This protein senses osmotic shifts and responds by importing osmolytes [28] including proline, betaine, pipecolic acid, and taurine. Notably, each of these osmolytes was recently shown to accumulate in the roots and rhizosphere of S.bicolor RTx430 under drought conditions [58], suggesting a potential connection between altered host metabolite production under drought and the strong colonization of Streptomyces isolates with greater capacity to exploit this shift in the metabolite environment. Finally, our findings reveal a lack of consistent correlation between drought-induced microbial enrichment (DE) and plant growth-promoting rhizobacteria (PGPR) activity in sorghum, suggesting that enrichment alone may not be a reliable indicator of functional benefit to the host. Similar conclusions have been drawn in other systems, where microbial enrichment did not translate to measurable plant benefits, highlighting the need for functional validation [10,59–61]. While individual isolates belonging to the genus Streptomyces have often been associated with drought resilience and PGPR functions [8,62,63], our study demonstrates not only that root-associated Streptomyces isolates from this system did not universally exhibit these beneficial abilities, but also that those capable of benefiting their hosts are phylogenetically dispersed, and that even closely related isolates may exhibit contrasting patterns of help and harm with respect to host phenotype. This observation underscores the complexity of rhizosphere community dynamics, and suggests that microbial enrichment in the root may be governed as much by abiotic filtering, niche competition, and stochastic colonization as by targeted host recruitment. In practical terms, this means that microbiome engineering efforts aimed at improving sorghum resilience to drought must move beyond genus- or species-level classifications and instead focus on the strain-specific gene content and metabolic potential. Moreover, the presence of potentially antagonistic or neutral Streptomyces strains in the root microbiome may complicate inoculant design, as competitive interactions could suppress the establishment of beneficial strains. Ultimately, leveraging the full potential of Streptomyces in sorghum or related crops will require a systems-level understanding of microbiome composition, functional gene distribution, and environmental context. Additional studies aimed at developing this proper framework are critical for accurately interpreting microbiome responses to environmental stress and for designing effective microbiome-based interventions in crop systems. Materials and methods Field experimental design and sample collection Sorghum bicolor cultivar RTx430 plants were grown in a field located at the University of California Kearney Agricultural Research and Extension (KARE) Center located in Parlier, California (36.6008°N, 119.5109°W), as described previously [58]. Sorghum seeds were sown into pre-watered plots. Starting in the third week, control treatment plants were watered for 1 h three times per week by drip irrigation (1.89 l/h flow rate). After 8 weeks, which coincided with the average onset of flowering across all plants, root samples were harvested from a subset of field samples prior to watering. All field samples were collected between 11 AM and 12 PM as described in [58]. Roots were vortexed three times for 1 min in epiphyte removal buffer (ice-cold 0.75% KH2PO4, 0.95% K2HPO4, 1% Triton X-100 in ddH2O; filter sterilized at 0.2 μm with Corning brand filters) and patted dry. All samples were immediately flash-frozen in LN2 in the field and stored at −80°C until sample processing. Microbial isolates All 48 Streptomyces isolates used in this study originated from the same field plots at the KARE Center. All bacterial isolations were performed on the roots of sorghum cultivar RTx430 grown in this field annually over a period of three years as described in [24]. Whole genome sequencing and assembly was performed using two distinct workflows. Three isolates (SAI104, SAI190, and SAI211) were sequenced using Oxford Nanopore Technologies (ONT) v14 library prep chemistry on R10.4.1 flow cells followed by assembly with Miniasm v0.3 [64], Flye v2.9.1 [65], and polished via Medaka v1.8.0 (ONT 2022). All other isolates (n = 45) were sequenced with PacBio Sequel SMRT and assembled into draft genomes with HGAP v4 (0.2.1) [66]. Taxonomic identification was performed using GTDB-tk 2.4.0 using reference database r220 [67,68]. Phylogenetic tree construction was performed using a multiple sequence alignment of 138 known single copy genes from the Actinobacterial lineage identified with Hidden Markov Models (HMMs) using the GToTree v1.7.08 workflow [69–73]. Orthogroups across all genomes were identified with Orthofinder [33] and used to calculate core and pangenome estimates with Micropan [74] and subsequently visualized with Pagoo [75]. Microbox experiments To explore Streptomyces growth in planta, isolates were grown on Tryptic Soy Agar (3 g TSB powder and 15 g agar per liter) plates at 30°C for 3 days. Colonies were harvested and suspended in 1× phosphate-buffered saline (PBS) at 1 mg/ml of cell suspension. Sorghum cultivar RTx430 seeds were surface-sterilized with 10% bleach for 15 min and washed five times with sterile water. Seeds were placed on filter paper hydrated with 6 mL of sterile water in petri dishes, sealed with Micropore tape (3M) and incubated at 30°C in constant darkness for 48 hours. Germinated seedlings were inoculated in bacterial suspensions for 5 min, planted at a depth of 5 cm, inoculated again with 1mL of bacterial slurry and covered with calcined clay (Greens Grade, Profile). Each 5 L microbox (SacO2, Belgium) housed four seedlings planted in 1 kg of calcined clay mixed with 400 mL of 1× Hoagland’s Solution. All plants were grown at 26–28°C, 16 hours:8 hours light/dark cycle, 50% humidity. For drought treatments, drought-assigned boxes were placed in a sterilized laminar flow cabinet after 1 week of growth, box lids were removed, and moisture was drawn from the boxes overnight for 16 hours. This dry-down process was repeated the following day. Water was withheld from drought-treatment boxes for the remainder of the experiment. For irrigated treatments, each plant was irrigated with 6mL of sterile water every 7 days with no dry-down. After 5 weeks of growth, all treatments were phenotyped and harvested. Shoot fresh weight, dry weight, and % water content—((fresh weight-dry weight)/fresh weight) * 100—were measured and root tissues were harvested for microbial colonization quantification. Drought enrichment (DE) quantification DNA was extracted from 35 mg of root tissue from each plant using the DNeasy PowerSoil Kit (QIAGEN, cat #47014). Isolate colonization was quantified from DNA extracts via quantitative PCR using Qiagen Quantitect SYBR Green PCR (Cat.#204143) with Actinobacterial lineage-specific 16S rRNA primers Actino235 5′-CGCGGCCTATCAGCTTGTTG-3′ and Eub518 5′-ATTACCGCGGCTGCTGG-3′ [76]. Thermal cycling used the following protocol: 95°C 15 min, and 36 3-step cycles (95°C for 15 s, 55°C for 55 s, and 72°C for 45 s) followed by a plate read; a melt curve was generated by heating from 65 to 95°C with 0.5°C increments for 5 s. Cq values from these reactions were compared to isolate DNA standard curve to interpolate 16S rRNA abundance. For each isolate, DE was calculated as such: DE = log2(Drought/Water) where “Drought” refers to the mean abundance of 4 replicates under drought and “Water” refers to mean abundance of 4 irrigated replicates. This yields a score where drought-enriched bacteria have positive scores and drought-depleted bacteria have negative scores. Phylogenetic signal for DE was tested using phytools::phylosig [77]. DNA extraction, amplification, and amplicon sequencing DNA extraction was performed using the protocol for collection of root endosphere samples using a Qiagen DNeasy Powersoil DNA extraction kit with 150 mg as starting material in the provided collection vials. The V3-V4 region of the 16S rRNA gene was PCR amplified from 25 ng of genomic DNA using dual-indexed 16S rRNA Illumina iTags 341F (5′-CCTACGGGNBGCASCAG-3′) and 785R (5′-GACTACNVGGGTATCTAATCC-3′) supplemented with PNAs [78] designed to target host-derived amplicons from chloroplast and mitochondria 16S rRNA sequences (0.75 µM of each, PNABIO, Thousand Oaks, CA). Barcoded 16S rRNA amplicons were quantified using a Qubit dsDNA HS assay kit on a Qubit 3.0 fluorometer (Invitrogen, Carlsbad, CA), pooled in equimolar concentrations, purified using Agencourt AMPure XP magnetic beads (Beckman Coulter, Indianapolis, IN), quantified using a Qubit dsDNA HS assay kit on a Qubit 3.0 fluorometer (Invitrogen), and diluted to 10 nM in 30 μl total volume before being submitted to the QB3 Vincent J. Coates Genomics Sequencing Laboratory facility at the University of California, Berkeley for sequencing using Illumina Miseq 300 bp pair-end with v3 chemistry. The full-length 16S rRNA gene (~1,500 bp) was amplified using the universal primer pair 27F (5′-AGAGTTTGATCMTGGCTCAG-3′) and 1492R (5′-TACGGYTACCTTGTTACGACTT-3′). Barcoded full-length 16S rRNA amplicons were quantified, pooled, and purified as described above for the V3-V4 region. Full-length 16S rRNA pool was finally quantified and diluted to 50 nM in 30 µL total volume before being submitted to the Department of Energy’s Joint Genome Institute (DOE JGI) for sequencing using a single SMRT II cell Pacific Biosciences. Amplicon sequence processing and analysis for field experiment V3-V4 16S rRNA amplicon sequencing reads were demultiplexed in QIIME2 [79] applying a minimum predicted accuracy of Q30. Chimera detection and removal were performed using DADA2 [80] and high-quality ASVs were assigned taxonomy using the August 2013 version of GreenGenes 16S rRNA gene database as described previously [58]. Circular consensus sequences (CCS) from the full-length 16S rRNA amplicons were generated from raw subreads using PacBio’s SMRT Link software, applying a minimum of three full passes and a minimum predicted accuracy of Q30. Then, CCS were passed to DADA2 to generate ASVs and taxonomy was assigned also using the GreenGenes 16S rRNA gene database. All subsequent 16S statistical analyses were performed in R-v3.6.1 [81]. To account for differences in sequencing read depth across samples, samples were normalized by dividing the reads per ASV in a sample by the sum of usable reads in that sample, resulting in a table of relative abundance frequencies, which were used for analyses. Statistical significance was determined using the Holm–Sidak method, with alpha = 0.05, where each row was analyzed individually, without assuming a consistent standard deviation. Metabolite extraction, identification, and analysis To evaluate the metabolomic profiles of Streptomyces isolates, exometabolites were collected from cultures of 12 DC strains grown in liquid tap water–yeast extract (TWYE) medium. The TWYE medium consisted of 0.25 g/L yeast extract, 0.5 g/L K₂HPO₄, and 18 g/L agar, prepared in tap water and supplemented with 5 mL/L of nystatin (5 mg/mL) to inhibit fungal contamination. TWYE medium was further supplemented with ~0.08g finely ground root tissue harvested from sorghum cultivar RTx430 plants subjected to either well-watered or drought conditions. Root tissue was obtained from sorghum grown in microboxes under controlled conditions, with drought treatment applied during the final 2 weeks of growth, as described above. For each condition, 3 mL liquid TWYE cultures were inoculated with 5 µL of a 1 × 10⁵ spores/µL suspension (four tubes per strain) and incubated at 30°C with shaking at 200 rpm for 6 days. Cells were pelleted by centrifugation in 2 mL microcentrifuge tubes. One milliliter of the clarified supernatant (spent media) was carefully transferred to a fresh 2 mL tube. This transfer step was repeated once to ensure removal of residual cells. The clarified supernatants were then snap-frozen in liquid nitrogen and lyophilized to dryness. In preparation for LC–MS, lyophilized supernatants were resuspended with 300 μl of methanol, vortexed and sonicated 10 min, centrifuged 5 min at 5,000 rpm, and supernatant centrifuge-filtered 2.5 min at 2,500 rpm (0.22 µm hydrophilic PVDF, Millipore, Ultrafree-CL GV, #UFC40GV0S), and then 150 µl was transferred to LC–MS glass autosampler vials. Internal standards used and the untargeted liquid chromatography–mass spectrometry conditions were performed as described in [58]. Metabolite identification was based on exact mass and comparison of retention time (RT) and MS/MS fragmentation spectra to those of standards run using the same chromatography and MS/MS method. A Feature-Based Molecular Networking workflow was performed using MZmine 2 and GNPS [82–84]. An MZmine workflow was used to generate a list of features (MzRT values obtained from extracted ion chromatograms containing chromatographic peaks within a narrow m/z range) and filtered to remove isotopes, adducts, and features without MSMS (S4 and S5 Tables). For each feature, the most intense fragmentation spectrum was uploaded to GNPS: Global Natural Products Social Molecular Networking, a web-based mass spectrometry identification tool. When a sample mass spectrum matches one deposited within the GNPS database, a putative identification was made. Totals of 3,750 polar metabolites and 15,107 non-polar metabolites were predicted including positive and negative ion modes across different treatments. Principal components analysis of metabolite profiles was performed using prcomp::stats function in R. All other metabolite analyses were performed using MetaboAnalyst-v4.0 [85]. Microbial phenotyping Spore induction. Spore induction in Streptomyces isolates was performed using MS agar, a nutrient-limited medium optimized for sporulation. MS agar was prepared by dissolving 20 g of mannitol and 20 g of soya flour in 1 L of tap water, supplemented with 20 g of agar, and autoclaved twice at 120°C for 20 min each to enhance media clarity and reduce clumping from soya particulates. After cooling to ~55°C, the medium was poured into sterile Petri dishes and allowed to solidify. Streptomyces strains were streaked onto the surface of the dried plates with 200 µl of an overnight culture in TSB (30 g TSB powder per liter) and incubated at 30°C for 5–10 days. Sporulation was monitored visually by the appearance of aerial hyphae and colored, powdery spore masses. Siderophore detection using CAS-LB agar. Siderophore production was assessed using a Chrome Azurol S (CAS)-LB agar assay based on the method of [86], with minor modifications. The CAS dye solution was prepared by dissolving 60 mg of CAS in 50 mL deionized water, followed by the addition of 1 mM FeCl3·6H2O prepared in 10 mM HCl. A solution of 91 mg Hexadecyltrimethylammonium bromide (CTAB) in 40 mL deionized water was then added slowly under constant stirring, yielding a dark blue complex. This solution was filter sterilized and stored at 4°C protected from light. LB agar medium (10 g tryptone, 5 g yeast extract, 10 g NaCl, 15 g agar per liter, pH 7.0) was prepared and autoclaved, then cooled to approximately 50°C before adding the CAS dye solution at a 1:10 ratio (e.g., 50 mL dye per 500 mL agar). The mixture was stirred gently and poured into Petri plates. Streptomyces isolates were spot-inoculated onto the CAS-LB plates and incubated at 28°C for 72 hours. A yellow to orange halo around colonies indicated siderophore production, resulting from iron chelation from the blue CAS–Fe3+ complex. Osmotic tolerance assay. Osmotic tolerance of microbial isolates was evaluated using sorbitol-supplemented tryptic soy broth (TSB) in both solid and liquid formats, with bacterial growth quantified via Bradford protein assay. For solid media, 10% TSB agar (3 g TSB powder per liter) was supplemented with sorbitol to final concentrations of 0.5 M, 1 M, and 1.5 M, autoclaved and poured into Petri plates. For liquid assays, TSB was similarly prepared with sorbitol concentrations of 0.5 M, 1 M, and 1.5 M, and aliquoted with 3 ml into sterile culture tubes. Bacterial isolates were pre-cultured overnight in standard TSB normalized to 1 × 105 spores and inoculated into each condition. In solid media, 5 μL of bacterial suspension was spotted onto the surface and incubated at 30°C for 48–72 h, with growth scored qualitatively. In liquid media, cultures (5 tubes per strain per condition) of 3 ml were incubated at 30°C with shaking (200 rpm) with 5 μL of bacterial suspension. One tube per condition was harvested at 0, 24, 48, 72, and 96 hours and cells were pelleted by centrifugation at 12,000 rpm for 10 min, washed once with PBS, and store at −20°C. Cells were lysed with 500 µl of 1M NaOH and total soluble protein content was measured using the Bradford assay [87]. Ten μL of lysate added to 200 μL of Bradford reagent in a 96-well plate and bovine serum albumin (BSA) was used as standard. Absorbance was read at 595 nm using a microplate reader (Infinite Nano M+, TECAN). TBS 053, Bacillus megaterium, isolate was used as positive control and TBS 091, Paenibacillus lautus, as negative control of the assays. Time-course protein quantification under increasing osmotic stress enabled the evaluation of growth dynamics and adaptation to sorbitol-induced osmotic pressure. Pangenome analysis Orthogroups were identified across all 48 genomes using OrthoFinder 3.0.1b1 [33]. CDS for each orthogroup identified by OrthoFinder were aligned with MAFFT 7.505 [88]. hmmbuild and hmmemit from HMMER3 [89] were used to produce an HMM and consensus sequence from each multiple sequence alignment to represent the orthogroup. Finally, these cluster consensus sequences were functionally annotated with eggNOG-mapper v2 [90,91]. Only clusters with copy number variance across all genomes >0.25 and represented in at least ¼ of the genomes (n = 2,735) were then subjected to phylogenetic generalized least squares analysis via nlme::gls (Pinheiro, Bates, and R Core Team 2025). Associations between cluster count and DE were tested while simultaneously estimating Pagel’s λ ape::corPagel [92] to account for the effect of phylogenetic structure. In addition, a model for each gene was also fit using lambda fixed to 0—reflecting no phylogenetic signa—and 1—reflecting a Brownian motion process. Model fits were assessed by Akaike Information Criterion (AIC) to identify and select the best Pagel’s λ for each orthogroup. Finally, results were filtered for standard error <0.25, slope estimate >0.25, and normality of model residuals with Shapiro–Wilk test. Remaining results were tested for functional enrichments in Clusters of Orthologous Groups (COG) [93] relative to the background of all analyzed orthogroups (n = 2,735) by hypergeometric test via enrichPlot::enricher [94]. The pangenome visualization performed using Anvi’o v8 [27] used imported FASTA files, processed into a pangenome database using the snakemake pangenomics workflow [95,96]. In this workflow, open reading frames were determined using Prodigal [70], nucleotide sequences were aligned using MUSCLE [97], and amino acid sequences were aligned using DIAMOND [98]. The average nucleotide identity (ANI) was calculated using pyANI [99]. Field experimental design and sample collection Sorghum bicolor cultivar RTx430 plants were grown in a field located at the University of California Kearney Agricultural Research and Extension (KARE) Center located in Parlier, California (36.6008°N, 119.5109°W), as described previously [58]. Sorghum seeds were sown into pre-watered plots. Starting in the third week, control treatment plants were watered for 1 h three times per week by drip irrigation (1.89 l/h flow rate). After 8 weeks, which coincided with the average onset of flowering across all plants, root samples were harvested from a subset of field samples prior to watering. All field samples were collected between 11 AM and 12 PM as described in [58]. Roots were vortexed three times for 1 min in epiphyte removal buffer (ice-cold 0.75% KH2PO4, 0.95% K2HPO4, 1% Triton X-100 in ddH2O; filter sterilized at 0.2 μm with Corning brand filters) and patted dry. All samples were immediately flash-frozen in LN2 in the field and stored at −80°C until sample processing. Microbial isolates All 48 Streptomyces isolates used in this study originated from the same field plots at the KARE Center. All bacterial isolations were performed on the roots of sorghum cultivar RTx430 grown in this field annually over a period of three years as described in [24]. Whole genome sequencing and assembly was performed using two distinct workflows. Three isolates (SAI104, SAI190, and SAI211) were sequenced using Oxford Nanopore Technologies (ONT) v14 library prep chemistry on R10.4.1 flow cells followed by assembly with Miniasm v0.3 [64], Flye v2.9.1 [65], and polished via Medaka v1.8.0 (ONT 2022). All other isolates (n = 45) were sequenced with PacBio Sequel SMRT and assembled into draft genomes with HGAP v4 (0.2.1) [66]. Taxonomic identification was performed using GTDB-tk 2.4.0 using reference database r220 [67,68]. Phylogenetic tree construction was performed using a multiple sequence alignment of 138 known single copy genes from the Actinobacterial lineage identified with Hidden Markov Models (HMMs) using the GToTree v1.7.08 workflow [69–73]. Orthogroups across all genomes were identified with Orthofinder [33] and used to calculate core and pangenome estimates with Micropan [74] and subsequently visualized with Pagoo [75]. Microbox experiments To explore Streptomyces growth in planta, isolates were grown on Tryptic Soy Agar (3 g TSB powder and 15 g agar per liter) plates at 30°C for 3 days. Colonies were harvested and suspended in 1× phosphate-buffered saline (PBS) at 1 mg/ml of cell suspension. Sorghum cultivar RTx430 seeds were surface-sterilized with 10% bleach for 15 min and washed five times with sterile water. Seeds were placed on filter paper hydrated with 6 mL of sterile water in petri dishes, sealed with Micropore tape (3M) and incubated at 30°C in constant darkness for 48 hours. Germinated seedlings were inoculated in bacterial suspensions for 5 min, planted at a depth of 5 cm, inoculated again with 1mL of bacterial slurry and covered with calcined clay (Greens Grade, Profile). Each 5 L microbox (SacO2, Belgium) housed four seedlings planted in 1 kg of calcined clay mixed with 400 mL of 1× Hoagland’s Solution. All plants were grown at 26–28°C, 16 hours:8 hours light/dark cycle, 50% humidity. For drought treatments, drought-assigned boxes were placed in a sterilized laminar flow cabinet after 1 week of growth, box lids were removed, and moisture was drawn from the boxes overnight for 16 hours. This dry-down process was repeated the following day. Water was withheld from drought-treatment boxes for the remainder of the experiment. For irrigated treatments, each plant was irrigated with 6mL of sterile water every 7 days with no dry-down. After 5 weeks of growth, all treatments were phenotyped and harvested. Shoot fresh weight, dry weight, and % water content—((fresh weight-dry weight)/fresh weight) * 100—were measured and root tissues were harvested for microbial colonization quantification. Drought enrichment (DE) quantification DNA was extracted from 35 mg of root tissue from each plant using the DNeasy PowerSoil Kit (QIAGEN, cat #47014). Isolate colonization was quantified from DNA extracts via quantitative PCR using Qiagen Quantitect SYBR Green PCR (Cat.#204143) with Actinobacterial lineage-specific 16S rRNA primers Actino235 5′-CGCGGCCTATCAGCTTGTTG-3′ and Eub518 5′-ATTACCGCGGCTGCTGG-3′ [76]. Thermal cycling used the following protocol: 95°C 15 min, and 36 3-step cycles (95°C for 15 s, 55°C for 55 s, and 72°C for 45 s) followed by a plate read; a melt curve was generated by heating from 65 to 95°C with 0.5°C increments for 5 s. Cq values from these reactions were compared to isolate DNA standard curve to interpolate 16S rRNA abundance. For each isolate, DE was calculated as such: DE = log2(Drought/Water) where “Drought” refers to the mean abundance of 4 replicates under drought and “Water” refers to mean abundance of 4 irrigated replicates. This yields a score where drought-enriched bacteria have positive scores and drought-depleted bacteria have negative scores. Phylogenetic signal for DE was tested using phytools::phylosig [77]. DNA extraction, amplification, and amplicon sequencing DNA extraction was performed using the protocol for collection of root endosphere samples using a Qiagen DNeasy Powersoil DNA extraction kit with 150 mg as starting material in the provided collection vials. The V3-V4 region of the 16S rRNA gene was PCR amplified from 25 ng of genomic DNA using dual-indexed 16S rRNA Illumina iTags 341F (5′-CCTACGGGNBGCASCAG-3′) and 785R (5′-GACTACNVGGGTATCTAATCC-3′) supplemented with PNAs [78] designed to target host-derived amplicons from chloroplast and mitochondria 16S rRNA sequences (0.75 µM of each, PNABIO, Thousand Oaks, CA). Barcoded 16S rRNA amplicons were quantified using a Qubit dsDNA HS assay kit on a Qubit 3.0 fluorometer (Invitrogen, Carlsbad, CA), pooled in equimolar concentrations, purified using Agencourt AMPure XP magnetic beads (Beckman Coulter, Indianapolis, IN), quantified using a Qubit dsDNA HS assay kit on a Qubit 3.0 fluorometer (Invitrogen), and diluted to 10 nM in 30 μl total volume before being submitted to the QB3 Vincent J. Coates Genomics Sequencing Laboratory facility at the University of California, Berkeley for sequencing using Illumina Miseq 300 bp pair-end with v3 chemistry. The full-length 16S rRNA gene (~1,500 bp) was amplified using the universal primer pair 27F (5′-AGAGTTTGATCMTGGCTCAG-3′) and 1492R (5′-TACGGYTACCTTGTTACGACTT-3′). Barcoded full-length 16S rRNA amplicons were quantified, pooled, and purified as described above for the V3-V4 region. Full-length 16S rRNA pool was finally quantified and diluted to 50 nM in 30 µL total volume before being submitted to the Department of Energy’s Joint Genome Institute (DOE JGI) for sequencing using a single SMRT II cell Pacific Biosciences. Amplicon sequence processing and analysis for field experiment V3-V4 16S rRNA amplicon sequencing reads were demultiplexed in QIIME2 [79] applying a minimum predicted accuracy of Q30. Chimera detection and removal were performed using DADA2 [80] and high-quality ASVs were assigned taxonomy using the August 2013 version of GreenGenes 16S rRNA gene database as described previously [58]. Circular consensus sequences (CCS) from the full-length 16S rRNA amplicons were generated from raw subreads using PacBio’s SMRT Link software, applying a minimum of three full passes and a minimum predicted accuracy of Q30. Then, CCS were passed to DADA2 to generate ASVs and taxonomy was assigned also using the GreenGenes 16S rRNA gene database. All subsequent 16S statistical analyses were performed in R-v3.6.1 [81]. To account for differences in sequencing read depth across samples, samples were normalized by dividing the reads per ASV in a sample by the sum of usable reads in that sample, resulting in a table of relative abundance frequencies, which were used for analyses. Statistical significance was determined using the Holm–Sidak method, with alpha = 0.05, where each row was analyzed individually, without assuming a consistent standard deviation. Metabolite extraction, identification, and analysis To evaluate the metabolomic profiles of Streptomyces isolates, exometabolites were collected from cultures of 12 DC strains grown in liquid tap water–yeast extract (TWYE) medium. The TWYE medium consisted of 0.25 g/L yeast extract, 0.5 g/L K₂HPO₄, and 18 g/L agar, prepared in tap water and supplemented with 5 mL/L of nystatin (5 mg/mL) to inhibit fungal contamination. TWYE medium was further supplemented with ~0.08g finely ground root tissue harvested from sorghum cultivar RTx430 plants subjected to either well-watered or drought conditions. Root tissue was obtained from sorghum grown in microboxes under controlled conditions, with drought treatment applied during the final 2 weeks of growth, as described above. For each condition, 3 mL liquid TWYE cultures were inoculated with 5 µL of a 1 × 10⁵ spores/µL suspension (four tubes per strain) and incubated at 30°C with shaking at 200 rpm for 6 days. Cells were pelleted by centrifugation in 2 mL microcentrifuge tubes. One milliliter of the clarified supernatant (spent media) was carefully transferred to a fresh 2 mL tube. This transfer step was repeated once to ensure removal of residual cells. The clarified supernatants were then snap-frozen in liquid nitrogen and lyophilized to dryness. In preparation for LC–MS, lyophilized supernatants were resuspended with 300 μl of methanol, vortexed and sonicated 10 min, centrifuged 5 min at 5,000 rpm, and supernatant centrifuge-filtered 2.5 min at 2,500 rpm (0.22 µm hydrophilic PVDF, Millipore, Ultrafree-CL GV, #UFC40GV0S), and then 150 µl was transferred to LC–MS glass autosampler vials. Internal standards used and the untargeted liquid chromatography–mass spectrometry conditions were performed as described in [58]. Metabolite identification was based on exact mass and comparison of retention time (RT) and MS/MS fragmentation spectra to those of standards run using the same chromatography and MS/MS method. A Feature-Based Molecular Networking workflow was performed using MZmine 2 and GNPS [82–84]. An MZmine workflow was used to generate a list of features (MzRT values obtained from extracted ion chromatograms containing chromatographic peaks within a narrow m/z range) and filtered to remove isotopes, adducts, and features without MSMS (S4 and S5 Tables). For each feature, the most intense fragmentation spectrum was uploaded to GNPS: Global Natural Products Social Molecular Networking, a web-based mass spectrometry identification tool. When a sample mass spectrum matches one deposited within the GNPS database, a putative identification was made. Totals of 3,750 polar metabolites and 15,107 non-polar metabolites were predicted including positive and negative ion modes across different treatments. Principal components analysis of metabolite profiles was performed using prcomp::stats function in R. All other metabolite analyses were performed using MetaboAnalyst-v4.0 [85]. Microbial phenotyping Spore induction. Spore induction in Streptomyces isolates was performed using MS agar, a nutrient-limited medium optimized for sporulation. MS agar was prepared by dissolving 20 g of mannitol and 20 g of soya flour in 1 L of tap water, supplemented with 20 g of agar, and autoclaved twice at 120°C for 20 min each to enhance media clarity and reduce clumping from soya particulates. After cooling to ~55°C, the medium was poured into sterile Petri dishes and allowed to solidify. Streptomyces strains were streaked onto the surface of the dried plates with 200 µl of an overnight culture in TSB (30 g TSB powder per liter) and incubated at 30°C for 5–10 days. Sporulation was monitored visually by the appearance of aerial hyphae and colored, powdery spore masses. Siderophore detection using CAS-LB agar. Siderophore production was assessed using a Chrome Azurol S (CAS)-LB agar assay based on the method of [86], with minor modifications. The CAS dye solution was prepared by dissolving 60 mg of CAS in 50 mL deionized water, followed by the addition of 1 mM FeCl3·6H2O prepared in 10 mM HCl. A solution of 91 mg Hexadecyltrimethylammonium bromide (CTAB) in 40 mL deionized water was then added slowly under constant stirring, yielding a dark blue complex. This solution was filter sterilized and stored at 4°C protected from light. LB agar medium (10 g tryptone, 5 g yeast extract, 10 g NaCl, 15 g agar per liter, pH 7.0) was prepared and autoclaved, then cooled to approximately 50°C before adding the CAS dye solution at a 1:10 ratio (e.g., 50 mL dye per 500 mL agar). The mixture was stirred gently and poured into Petri plates. Streptomyces isolates were spot-inoculated onto the CAS-LB plates and incubated at 28°C for 72 hours. A yellow to orange halo around colonies indicated siderophore production, resulting from iron chelation from the blue CAS–Fe3+ complex. Osmotic tolerance assay. Osmotic tolerance of microbial isolates was evaluated using sorbitol-supplemented tryptic soy broth (TSB) in both solid and liquid formats, with bacterial growth quantified via Bradford protein assay. For solid media, 10% TSB agar (3 g TSB powder per liter) was supplemented with sorbitol to final concentrations of 0.5 M, 1 M, and 1.5 M, autoclaved and poured into Petri plates. For liquid assays, TSB was similarly prepared with sorbitol concentrations of 0.5 M, 1 M, and 1.5 M, and aliquoted with 3 ml into sterile culture tubes. Bacterial isolates were pre-cultured overnight in standard TSB normalized to 1 × 105 spores and inoculated into each condition. In solid media, 5 μL of bacterial suspension was spotted onto the surface and incubated at 30°C for 48–72 h, with growth scored qualitatively. In liquid media, cultures (5 tubes per strain per condition) of 3 ml were incubated at 30°C with shaking (200 rpm) with 5 μL of bacterial suspension. One tube per condition was harvested at 0, 24, 48, 72, and 96 hours and cells were pelleted by centrifugation at 12,000 rpm for 10 min, washed once with PBS, and store at −20°C. Cells were lysed with 500 µl of 1M NaOH and total soluble protein content was measured using the Bradford assay [87]. Ten μL of lysate added to 200 μL of Bradford reagent in a 96-well plate and bovine serum albumin (BSA) was used as standard. Absorbance was read at 595 nm using a microplate reader (Infinite Nano M+, TECAN). TBS 053, Bacillus megaterium, isolate was used as positive control and TBS 091, Paenibacillus lautus, as negative control of the assays. Time-course protein quantification under increasing osmotic stress enabled the evaluation of growth dynamics and adaptation to sorbitol-induced osmotic pressure. Spore induction. Spore induction in Streptomyces isolates was performed using MS agar, a nutrient-limited medium optimized for sporulation. MS agar was prepared by dissolving 20 g of mannitol and 20 g of soya flour in 1 L of tap water, supplemented with 20 g of agar, and autoclaved twice at 120°C for 20 min each to enhance media clarity and reduce clumping from soya particulates. After cooling to ~55°C, the medium was poured into sterile Petri dishes and allowed to solidify. Streptomyces strains were streaked onto the surface of the dried plates with 200 µl of an overnight culture in TSB (30 g TSB powder per liter) and incubated at 30°C for 5–10 days. Sporulation was monitored visually by the appearance of aerial hyphae and colored, powdery spore masses. Siderophore detection using CAS-LB agar. Siderophore production was assessed using a Chrome Azurol S (CAS)-LB agar assay based on the method of [86], with minor modifications. The CAS dye solution was prepared by dissolving 60 mg of CAS in 50 mL deionized water, followed by the addition of 1 mM FeCl3·6H2O prepared in 10 mM HCl. A solution of 91 mg Hexadecyltrimethylammonium bromide (CTAB) in 40 mL deionized water was then added slowly under constant stirring, yielding a dark blue complex. This solution was filter sterilized and stored at 4°C protected from light. LB agar medium (10 g tryptone, 5 g yeast extract, 10 g NaCl, 15 g agar per liter, pH 7.0) was prepared and autoclaved, then cooled to approximately 50°C before adding the CAS dye solution at a 1:10 ratio (e.g., 50 mL dye per 500 mL agar). The mixture was stirred gently and poured into Petri plates. Streptomyces isolates were spot-inoculated onto the CAS-LB plates and incubated at 28°C for 72 hours. A yellow to orange halo around colonies indicated siderophore production, resulting from iron chelation from the blue CAS–Fe3+ complex. Osmotic tolerance assay. Osmotic tolerance of microbial isolates was evaluated using sorbitol-supplemented tryptic soy broth (TSB) in both solid and liquid formats, with bacterial growth quantified via Bradford protein assay. For solid media, 10% TSB agar (3 g TSB powder per liter) was supplemented with sorbitol to final concentrations of 0.5 M, 1 M, and 1.5 M, autoclaved and poured into Petri plates. For liquid assays, TSB was similarly prepared with sorbitol concentrations of 0.5 M, 1 M, and 1.5 M, and aliquoted with 3 ml into sterile culture tubes. Bacterial isolates were pre-cultured overnight in standard TSB normalized to 1 × 105 spores and inoculated into each condition. In solid media, 5 μL of bacterial suspension was spotted onto the surface and incubated at 30°C for 48–72 h, with growth scored qualitatively. In liquid media, cultures (5 tubes per strain per condition) of 3 ml were incubated at 30°C with shaking (200 rpm) with 5 μL of bacterial suspension. One tube per condition was harvested at 0, 24, 48, 72, and 96 hours and cells were pelleted by centrifugation at 12,000 rpm for 10 min, washed once with PBS, and store at −20°C. Cells were lysed with 500 µl of 1M NaOH and total soluble protein content was measured using the Bradford assay [87]. Ten μL of lysate added to 200 μL of Bradford reagent in a 96-well plate and bovine serum albumin (BSA) was used as standard. Absorbance was read at 595 nm using a microplate reader (Infinite Nano M+, TECAN). TBS 053, Bacillus megaterium, isolate was used as positive control and TBS 091, Paenibacillus lautus, as negative control of the assays. Time-course protein quantification under increasing osmotic stress enabled the evaluation of growth dynamics and adaptation to sorbitol-induced osmotic pressure. Pangenome analysis Orthogroups were identified across all 48 genomes using OrthoFinder 3.0.1b1 [33]. CDS for each orthogroup identified by OrthoFinder were aligned with MAFFT 7.505 [88]. hmmbuild and hmmemit from HMMER3 [89] were used to produce an HMM and consensus sequence from each multiple sequence alignment to represent the orthogroup. Finally, these cluster consensus sequences were functionally annotated with eggNOG-mapper v2 [90,91]. Only clusters with copy number variance across all genomes >0.25 and represented in at least ¼ of the genomes (n = 2,735) were then subjected to phylogenetic generalized least squares analysis via nlme::gls (Pinheiro, Bates, and R Core Team 2025). Associations between cluster count and DE were tested while simultaneously estimating Pagel’s λ ape::corPagel [92] to account for the effect of phylogenetic structure. In addition, a model for each gene was also fit using lambda fixed to 0—reflecting no phylogenetic signa—and 1—reflecting a Brownian motion process. Model fits were assessed by Akaike Information Criterion (AIC) to identify and select the best Pagel’s λ for each orthogroup. Finally, results were filtered for standard error <0.25, slope estimate >0.25, and normality of model residuals with Shapiro–Wilk test. Remaining results were tested for functional enrichments in Clusters of Orthologous Groups (COG) [93] relative to the background of all analyzed orthogroups (n = 2,735) by hypergeometric test via enrichPlot::enricher [94]. The pangenome visualization performed using Anvi’o v8 [27] used imported FASTA files, processed into a pangenome database using the snakemake pangenomics workflow [95,96]. In this workflow, open reading frames were determined using Prodigal [70], nucleotide sequences were aligned using MUSCLE [97], and amino acid sequences were aligned using DIAMOND [98]. The average nucleotide identity (ANI) was calculated using pyANI [99]. Supporting information S1 Text.. Fig A. Phenotypic and metabolomic differentiation among Streptomyces isolates sharing a dominant V3-V4 16S rRNA ASV. (A) Image of the growth (on spore induction media, SIM) for all 28 Streptomyces isolates with 100% identity to the dominant ASV identified in the root via V3-V4 16S rRNA sequencing with original hand labeling from Fig 1A. (B) Ordination of non-polar exometabolomic profiling of the 12 strains following growth on liquid tap water–yeast extract (TWYE) medium showing distinct clustering of AC group 2 and 5 (replication n = 4). AC groups are presented by colors: AC2, purple; AC5, light blue; AC6, green mist; AC8, green. Streptomyces isolates ID are indicated in the corresponding plate. The data underlying this Figure can be found in https://doi.org/10.5281/zenodo.17554086. Fig B. Ordination plot of nonpolar exometabolomics data analyzed from the spent media of all identified Streptomyces strains matching V3-V4 ASV in the isolate collection following growth on root tissue. The color of each shape indicates the AC group it belongs to: AC2, purple; AC5, light blue; AC6, green mist; AC8, green. Blank control samples containing only drought root tissue (brown circles) or control irrigated root tissue (brown triangles) are shown at top right in the plot (replication n = 4). Streptomyces isolates ID are indicated in the corresponding plate. The data underlying this Figure can be found in https://doi.org/10.5281/zenodo.17554086. Fig C. Production of representative individual siderophores putatively identified and measured by exometabolomics. DC strains were under growth on TWYE (yellow), drought-stressed root tissue (orange), or non-stressed root tissue (blue). Cells were pelleted by centrifugation and the clarified supernatants were then snap-frozen in liquid nitrogen and lyophilized to dryness. Lyophilized supernatants were resuspended with methanol, sonicated and then transferred to LC–MS glass autosampler vials for untargeted liquid chromatography–mass spectrometry identification. The data underlying this Figure can be found in https://doi.org/10.5281/zenodo.17554086. Fig D. Mirror match plot comparing experimental (query) and library MS/MS spectra of siderophores putatively identified by exometabolomics. The mirror plot visualizes the spectral alignment between an experimental MS/MS spectrum (top in black) and a reference spectrum from the GNPS library (bottom in green). Shared fragment ions are indicated by aligned peaks, and spectral similarity is quantified using the cosine score. The precursor ion m/z, retention time (RT), ionization mode, cosine similarity, and number of matched fragment peaks are used to assess the quality and confidence of the match. Fig E. Total siderophore production as measured by CAS-LB agar assay based on the method of Schwyn and Neilands (1987). Streptomyces isolates were spot-inoculated onto the CAS-LB plates and incubated at 28°C for 72 hours. A yellow to orange halo around colonies indicated siderophore production, resulting from iron chelation from the blue CAS–Fe3+ complex. The experiment was repeated twice with three technical replicates each. Fig F. Osmotic tolerance assay of Streptomyces isolates under increasing sorbitol-induced stress. Streptomyces isolates were grown in tryptic soy broth (10% TSB) supplemented with sorbitol at final concentrations of 0.5, 1.0, and 1.5 M. Bacterial growth was monitored over time (0, 24, 48, 72, and 96 hours; n = 3) using Bradford protein assays to quantify total biomass. This time-course approach enabled the assessment of growth dynamics and adaptation to osmotic stress. Bacillus megaterium (TBS 053) and Paenibacillus lautus (TBS 091) were included as positive and negative controls, respectively. The data underlying this Figure can be found in https://doi.org/10.5281/zenodo.17554086. https://doi.org/10.1371/journal.pbio.3003526.s001 (PDF) S1 Table. Significant differences of ASVs identified in both full-length and V3–V4 16S rRNA sequencing datasets. Differentially abundant ASVs were identified using normalized ASV abundance data from both full-length and V3–V4 16S rRNA amplicon sequencing of root microbiomes. Statistical comparisons were performed using Holm–Sidak corrected significance testing (α = 0.05). https://doi.org/10.1371/journal.pbio.3003526.s002 (CSV) S2 Table. In silico predicted primer coverage based on the Silva SSU Ref NR (release 132) database compared with observed ASVs. Predicted values indicate the proportion of reference sequences matching each primer set, while observed values represent the proportion of ASVs recovered in this study. https://doi.org/10.1371/journal.pbio.3003526.s003 (CSV) S3 Table. PERMANOVA analysis of the phenotyping data collected from the osmotolerance and siderophore production assays. The analysis was conducted using the Euclidean distance of control and DC isolate samples per phenotyping condition. https://doi.org/10.1371/journal.pbio.3003526.s004 (CSV) S4 Table. Library-based spectral matches (GNPS DB_results). This table summarizes high-confidence spectral matches between experimental MS/MS spectra and reference spectra in the GNPS library. These results support the identification of known metabolites present in the sample. https://doi.org/10.1371/journal.pbio.3003526.s005 (XLSX) S5 Table. Summary of molecular clusters generated by GNPS molecular networking. This table provides information on the consensus MS/MS clusters identified through molecular networking analysis using GNPS. Each cluster represents a group of MS/MS spectra with similar fragmentation patterns, corresponding to a putative unique metabolite or structurally related compounds. The table includes the cluster index, consensus precursor m/z, retention time, number of spectra per cluster, and the number of unique sample files contributing to each cluster. https://doi.org/10.1371/journal.pbio.3003526.s006 (XLSX) Acknowledgments We thank the staff of the Kearney Agricultural Research Center for their help in sample collection and field preparation.

Maybrun, Chloe L.;Oderberg, Isaac M.;Gaviño, Michael A.;Cooke, Thomas F.;Choi, Kyungyong;Han, Jongyoon;Reddien, Peter W.

doi: 10.1371/journal.pbio.3003482pmid: 41218061

Introduction Many organisms can regenerate appendages or significant parts of body axes [1,2]. During regeneration, tissues must be formed in the correct position within the body and properly oriented relative to remaining tissues. This attribute of regeneration poses a unique patterning challenge in biology. How positional information is integrated across the axes of new and pre-existing, adult tissues is therefore a central and understudied question in the field of regeneration. Planarians can regenerate from a large array of injuries, including from small fragments of the body [3]. The planarian body plan is organized along three axes: anterior–posterior (AP), dorsal–ventral (DV), and medial–lateral (ML). Positional information refers to factors that influence the regional identity and organization of cell types, organs, and appendages [4]. Positional information across body axes, such as during animal development, promotes organization of the body plan [5–8]. Positional information in adult planarians includes the products of genes referred to as position control genes (PCGs). PCGs are constitutively and regionally expressed along one or more body axes and either have an abnormal patterning phenotype when inhibited or are predicted to be part of planarian patterning pathways by homology [9]. Most PCGs show constitutive regional expression predominantly in planarian muscle [9]. In development, most animals utilize canonical Wnt signaling to control pattern formation on the AP axis, whereas Bmp signaling is used to pattern the DV axis [5–7]. The planarian AP axis is patterned largely by canonical Wnt signaling [10]. A system of posteriorly expressed Wnt ligands and anteriorly expressed Wnt inhibitors establishes a posterior-to-anterior gradient of β-catenin-1 signaling activity that is required for the establishment and maintenance of regional tissue identity [11–20]. RNAi of β-catenin-1 causes heads to regenerate in place of tails after transverse amputation and induces the formation of ectopic heads in uninjured animals [11–13]. Two FGFRL-Wnt modules and Src regulate planarian head and trunk patterning [21–23]. Bmp signaling controls planarian DV patterning [10]. The expression of the Bmp ligand-encoding bmp4 gene is restricted to dorsal muscle in a medial-to-lateral gradient [9,24–26]. RNAi of bmp4 or the downstream pathway component smad1 results in progressive ventralization in uninjured animals [25–27]. During normal regeneration, major regions of body axes are generated with the correct orientation relative to each other in the blastema (regenerative outgrowth) and to pre-existing patterns within the amputated fragment, suggesting the existence of mechanisms to coordinate axial regeneration. bmp4 is required for regeneration of pattern along the AP axis [28]. The possibility that DV pattern informs the location of AP axis establishment was proposed in theoretical models of planarian patterning [29,30] and is supported by the following observations. The anterior pole, a cluster of specialized muscle cells that organizes pattern during head regeneration [31–33], forms at the DV median plane [34]. This region of the animal, sometimes called the dorsal–ventral boundary (DVB), is defined by a zone of epidermal and sub-epidermal gene expression at the lateral edge of the animal, where the dorsal and ventral sides meet. This DVB location is set by a patterning process on the DV axis involving dorsal Bmp signaling [25–27]. We also note that ectopic head formation after β-catenin-1 RNAi occurs only at the animal lateral edge (body margin) [11–13]. DV confrontation can promote axis formation in certain patterning contexts in other organisms. In Drosophila, for instance, interactions between dorsal and ventral cells in the wing imaginal disc drive wing growth [35,36]. Genetic manipulations that create an ectopic boundary between dorsal and ventral cells in Drosophila can lead to formation of an ectopic outgrowth of the wing blade. DV contact during wound closure has been proposed to promote planarian blastema formation [37,38]. Whether diverse organisms across the animal kingdom such as those capable of whole-body regeneration have a DVB present in the adult stage that regulates regeneration, and the mechanistic roles of any such adult DVBs in regenerative contexts remain poorly addressed. The presence of a region at the planarian DV median plane with a candidate regulatory role in regeneration therefore makes planarians an attractive venue for studying the regenerative role and nature of an adult DVB. In this work, we find that the DVB is sufficient and necessary for planarian head formation in a low Wnt environment. By manipulating the pattern of the DVB, it is possible to induce heads to grow out of the dorsal side of the animal or to block head positional information expression from either the lateral animal edge in a Wnt-low environment or at anterior-facing wounds (AFWs). Our findings support a hierarchical model of pattern integration across body axes in regeneration, with the lateral DV median plane orchestrating growth and patterning of large AP axis regions. Results The AP and DV axes maintain pattern largely independently in homeostasis The maintenance of adult positional information in planarians is important for tissue turnover and for regeneration after injury [10]. To maintain positional information with suitable spatial relationships between axes during homeostatic tissue maintenance, positional information along one axis could, in principle, require input from another axis. Inhibition of β-catenin-1 results in homeostatic anteriorization [11–13] (S1A Fig). However, DV asymmetric tissues [12] and both dorsally biased (bmp4, nlg-8) and ventrally biased (admp, nlg-7) PCG expression remained unchanged after β-catenin-1 RNAi (Figs 1A, S1B, and S1C), suggesting that Wnt signaling is not overtly required for the maintenance of DV pattern during adult tissue turnover. bmp4, the inhibition of which leads to homeostatic ventralization [25–27] (S1D Fig), has been proposed to integrate patterning of the DV and AP axes by suppressing Wnt [39]. In agreement with findings in Clark (2023) [39], we found that the domain of wnt1+ cells (the posterior pole) was expanded after 45 days of bmp4 RNAi (Figs 1B and S1E). The fz4-1 expression domain was similarly lengthened in bmp4 RNAi animals. However, no overt changes in the wntP-2 or ndl-3 expression domains were apparent (Fig 1B). The anterior pole, labeled by notum, was expanded, albeit to a lesser extent than the posterior pole (Figs 1B, S1F, and S1G). However, no dramatic changes in anterior PCG expression domains (sFRP-1, ndl-5) were observed (Figs 1B and S1F), as previously reported [39]. Furthermore, the relative order of PCG expression domains along the AP axis remained largely unchanged after bmp4 RNAi, suggesting that Bmp signaling is not strictly required for the maintenance of broad AP pattern, but might limit the extent of the posterior-most patterning domain of the AP axis. These results point towards the possibility that axis integration might predominantly be accomplished during regeneration or during ectopic head formation in a low Wnt environment, with pattern thereafter largely stably maintained by each axis independently during cell turnover. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Ectopic heads form at the DVB after β-catenin-1 RNAi. A) Dorsally biased (bmp4, nlg-8) and ventrally biased (admp, nlg-7) PCG expression after β-catenin-1 RNAi (21d). B) PCG expression domains in bmp4 RNAi animals (45d). Blue arrow denotes anterior or posterior boundary of PCG expression domain. Diagrams show the mean PCG domain length ± standard deviation relative to body length in bmp4 RNAi animals. Six to seventeen animals per PCG were measured. Data points from individual animals are listed in S1 Data. Significant domain shifts are marked: *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001 by two-tailed t-tests. C) The epidermal laminB+ DVB is positioned ventral to the dorsally high bmp4 expression domain in muscle and is coincident with the known location of lateral admp expression in muscle. Anterior and posterior PCG expression is concentrated near the DVB during regeneration (96 hpa). D) Fluorescence intensity along a line is plotted and smoothed data is represented with a thick stroke. The foci of anterior (ndl-4) and posterior (fz4-1) PCG expression show peaked behavior on the dorsal and ventral sides juxtaposing the laminB+ DVB. Individual data points are listed in S1 Data. E) Ectopic anterior PCG (sFRP-1) expression near the DVB after β-catenin-1 RNAi (14d). D, dorsal; V, ventral. F) Ectopic anterior PCG (sFRP-1, ndl-4) expression after β-catenin-1 RNAi (18d) is detected in longitudinal (myoD+), circular (nkx1-1+), and lateral DV (nk4+) muscle at the lateral animal edge. G) Co-expression of anterior PCGs (sFRP-1, ndl-4) and the patterning molecule admp in muscle at the DVB after β-catenin-1 RNAi (21d). Number of admp+ cells that co-localize with sFRP-1 and ndl-4 transcripts (magenta) out of total admp+ cells (green) is shown. n = 4 animals. Colored boxes, area depicted in photos. Scale bars, 100 μm (A, B, E), 50 μm (C, F, G). https://doi.org/10.1371/journal.pbio.3003482.g001 Organization of PCG expression at the DVB in regeneration During head regeneration, anterior pole cells labeled by notum transcripts form a focus at the DVB [34]. We used epidermal cells expressing laminB, which mark the lateral DV median plane [38,40], as a reference point for the DVB. At 96 hours post-amputation (hpa), this DVB location was positioned ventral to the dorsally high bmp4 domain in muscle and was coincident with the known location of admp expression in muscle [9,41,42] at the lateral DV median plane (Figs 1C and S1H). Expression of the anterior PCGs sFRP-1, ndl-4, and ndl-5 was centered around the DVB (Figs 1C and S1I) and showed peaked behavior on the dorsal and ventral sides that juxtapose the laminB+ DVB location (Figs 1D and S1J). The mature posterior pole labeled by wnt1 is dorsally biased [13,16,43] but formed a cluster near the DVB at 96 hpa during tail regeneration (Figs 1C and S1I). Expression of the posterior PCGs wnt11-1 and fz4-1 was centered around the laminB+ DVB (Figs 1C and S1I) and showed peaked behavior on the dorsal and ventral sides juxtaposing the DVB (Figs 1D and S1K), similar to the case of anterior PCGs. These results connect patterning processes involving PCG expression at the head and tail tips to the location of the DVB and support a potential mechanism for axis integration [34] wherein the DVB serves as a critical landmark to guide proper regeneration along the AP axis. Ectopic heads form at the DVB after β-catenin-1 RNAi Ectopic heads form at the animal periphery after β-catenin-1 RNAi, with the occasional exception of a head near the planarian pharynx, during a process of homeostatic tissue turnover [11–13]. β-catenin-1 RNAi resulted in ectopic focal expression of the anterior PCGs sFRP-1 and ndl-4 laterally, centered at the laminB+ DVB (Figs 1E and S1L). Anterior PCGs showed peaked ectopic expression on the dorsal and ventral sides juxtaposing the DVB in some regions (S1L Fig). Ectopic anterior PCG (sFRP-1, ndl-4) expression occurred in multiple muscle subtypes present near the DVB [44,45], including in longitudinal muscle (muscle that is oriented along the AP axis), circular muscle (muscle that is oriented along the ML axis), and lateral DV muscle (muscle that is oriented along the DV axis at the body margin) (Figs 1F and S1M). Ectopic anterior PCG (sFRP-1, ndl-4) expression in β-catenin-1 RNAi animals occurred in a large fraction (62.0%) of muscle cells expressing the DVB PCG admp [9,41,42] (Fig 1G). Ectopic anterior PCGs (sFRP-1, ndl-4) were co-expressed with additional muscle-enriched genes at the DVB, including lactadherin (dd_5463, identified below) and netrin-1 [46,47] (S1N Fig). Although anterior PCG (sFRP-1, ndl-4) expression was concentrated in foci at the DVB after β-catenin-1 RNAi, sporadic cells expressing anterior PCGs were also detected dispersed (not in foci) in the animal (S1A Fig). Hereafter, we analyze specifically the anterior PCG expression that occurred in a concentrated manner at the DVB in β-catenin-1 RNAi animals. A spatial atlas of gene expression, including genes encoding cell surface and signaling molecules, in muscle at the DVB Given the potential significance of the planarian DVB in patterning, and the fact that its molecular determinants are poorly understood, we sought to define a spatial atlas of gene expression for muscle at the DVB. We performed single-cell RNA sequencing (scRNA-seq) of the lateral edges of animals to enrich for DVB cells (S2A and S2B Fig). All major muscle subtypes at the DVB were represented in our data (S2C Fig). Known markers of the body margin from prior literature (admp [41,42], egf-6 [48], wnt-5 [14,16], unc5-B [46], ephR-2 [25], netrin-1 [46,47], follistatin [49,50], noggin-1 [25,41], and netrin-3 [44,47,51]) displayed expression predominantly in longitudinal muscle at the DVB (S2C Fig). Some DVB markers also displayed expression in lateral DV muscle (nk4+) and/or circular muscle (nkx1-1+) (S2C Fig), but subclustering data from circular muscle did not separate cells by their DV position of origin (S2D Fig). Subclustering scRNA-seq data from longitudinal muscle cells identified four total subclusters. Transcripts known to be expressed at the DVB were enriched in subcluster 1 (Figs 2A, S2E, and S2F), suggesting genes with enriched expression in this subcluster could represent DVB-expressed genes. Thirty-two total genes had significantly enriched expression in subcluster 1; nine of these genes were previously characterized as being expressed at the body margin, with 23 new DVB genes. We assessed the expression of these 32 genes by FISH, together with laminB as a common spatial reference marker. This generated a spatial atlas of gene expression in muscle at the DVB (Figs 2B–2D, S2G, and S2H). Previously identified DVB markers were expressed in either dorsal (admp, egf-6, wnt-5, unc5-B, ephR-2), ventral (netrin-1, follistatin), overlapping (noggin-1), or both dorsal and ventral (netrin-3) domains of the DV axis relative to laminB (Fig 2B—2D). In addition to these known DVB markers, nine additional genes (lactadherin (dd_5463), ephR-3 (dd_27668, dd_39545), troponin-2 (dd_7974), frem-1 (dd_5187), netR, dd_30210, delta-3, coupTF-1 (dd_33286), and zfp-4 (dd_29263, dd_21349)) were also expressed in particular domains relative to laminB at the DVB (Figs 2B—2D, S2I, and S2J). Lactadherin is a secreted protein that can bind anionic phospholipids [52]. Ephrin receptors mediate the effects of Ephrins, a family of membrane-bound signaling ligands that promote a variety of cellular responses including attraction, repulsion, and migration [53]. frem-1 encodes an extracellular matrix protein that contributes to epithelial-mesenchymal integrity in development [54]. Netrin receptors mediate the attractant and repellent properties of secreted Netrins that guide cell and axon migration in development [55]. dd_30210 encodes a novel protein with no characterized PFAM domains. delta-3 encodes a membrane-bound Notch signaling ligand. Notch signaling regulates diverse developmental processes, including cell fate determination [56]. COUP transcription factors encode orphan nuclear receptors that regulate diverse cellular processes in development [57]. zfp-4 encodes a novel zinc finger protein. Using our scRNA-seq data, we calculated the Pearson correlation coefficient for all pairwise comparisons of the 18 genes in the DVB atlas with expression detected by FISH. Within longitudinal muscle cells at the DVB, the expression of many genes showed a high degree of correlation (r > 0.5) (S2K Fig), indicating that individual muscle cells co-express multiple markers of the DVB. There were 14 genes with transcripts enriched in longitudinal muscle subcluster 1 (S2E and S2F Fig) that did not display clear DVB-restricted expression by FISH, either because expression was broad or not detected. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Spatial DVB gene expression atlas in muscle. A) Cartoon shows region of the animal isolated for scRNA-seq. Subclustering of longitudinal muscle cells from scRNA-seq data reveals a DVB cell population in UMAP plots. Novel markers of the DVB are labeled in black. B) FISH characterization of the DVB region. Images show the expression of genes along the DV axis relative to the spatial domain of the canonical epidermal marker of the DVB, laminB. C) FISH images show the expression of genes that label dorsal (nlg-8) and ventral (nlg-7) muscle, lateral DV muscle (nk4), and a DVB epidermis domain (equinox) relative to laminB. D) Cartoon representation of expression domains at the DVB. Dorsal (nlg-8) and ventral (nlg-7) PCGs are shown as graded columns. Scale bars, 50 μm (B, C). https://doi.org/10.1371/journal.pbio.3003482.g002 Taken together, the molecular data indicate that the DVB, often defined anatomically, can be viewed as an intricately patterned domain of gene expression that in muscle includes a large array of genes encoding signaling factors, cell-surface molecules, and transcription regulators (Figs 2D, S2G, and S2H). DVB muscle is thus a prominent but poorly understood pattern element of the planarian body plan. Ectopic DVB is permissive for the formation of an anterior region of the AP axis after β-catenin-1 RNAi To determine whether the DVB is sufficient to promote the formation of an anterior region of the AP axis after β-catenin-1 RNAi, we endeavored to generate animals with ectopic DVB. Transplantation of tissue with reversed DV polarity has been shown to induce ectopic DVB, labeled by epidermal expression of laminB, in goblet-shaped outgrowths [38,58]. By contrast, similar tissue transplantation without DV reversal leads to tissue integration without DVB formation. We used this approach to generate outgrowths with ectopic DVB in the post-pharyngeal region of the animal (Figs 3A and S3A). These outgrowths contained DVB muscle gene expression signatures, including for genes encoding the secreted molecules admp, lactadherin, and frem-1, indicating that both epidermis and underlying muscle acquire an ectopic DVB identity in outgrowths (Figs 3B and S3B). These outgrowths had AP character typical of the location of the transplant [38] (S3C Fig) and lacked anterior identity (Fig 3C). After allowing dorsal-to-ventral (D-V) transplant outgrowths to form, we initiated β-catenin-1 RNAi. Whereas head structures are normally constrained to the periphery of β-catenin-1 RNAi animals, D-V transplant outgrowths developed into goblet-shaped head-like structures marked by the presence of eyes in the middle of the animal (Fig 3A). These outgrowths (both dorsal and ventral outgrowths), after β-catenin-1 RNAi, expressed anterior PCGs (sFRP-1, ndl-4) and contained opsin+ photoreceptor cells (Figs 3C and S3D). By contrast, dorsal-to-dorsal (D-D) transplant animals simply healed the wound (n = 13/13) and these transplantation sites did not express anterior PCGs or transform into heads following β-catenin-1 RNAi (Fig 3A and 3D). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Ectopic DVB generated by transplantation is competent to induce formation of an anterior region of the AP axis after β-catenin-1 RNAi. A) Transplant animals that received tissue grafted with reversed DV polarity develop goblet-shaped outgrowths. After β-catenin-1 RNAi (10–14d), these outgrowths form head-like structures. Yellow arrowhead, eye. B) D-V transplant outgrowths show epidermal (laminB) and muscle (admp) DVB gene expression. C) D-V transplant outgrowths express anterior PCGs (sFRP-1, ndl-4) and form eyes (opsin) after β-catenin-1 RNAi (10–14d). Yellow arrowhead, eye. D) D-D transplants do not result in ectopic DVB (laminB) or anterior PCG (sFRP-1, ndl-4) gene expression after β-catenin-1 RNAi. E) DVB transplant animals develop dorsal outgrowths. After β-catenin-1 RNAi (10d), these outgrowths express anterior PCGs (sFRP-1, ndl-4) and form eye cells (opsin). Colored boxes, area depicted in photos. Scale bars, 250 μm (A), 100 μm (B, D), 50 μm (C, E). https://doi.org/10.1371/journal.pbio.3003482.g003 To complement the use of transplantation to generate ectopic DVB, we next transplanted the DVB region itself. Transplantation of a small posterior DVB-containing segment of the animal margin to the midline of a recipient animal induced the formation of a dorsal outgrowth in the recipient. These outgrowths expressed anterior PCGs (sFRP-1, ndl-4) and contained opsin+ photoreceptor cells after β-catenin-1 RNAi (Fig 3E), indicating that ectopically positioned DVB is competent to promote anterior patterning after Wnt inhibition. As an independent method to generate ectopic DVB, we utilized smad1 RNAi. smad1 encodes a downstream mediator of Bmp signaling and has a prominent role in planarian DV patterning [25]. Inhibition of bmp4 or smad1 leads to homeostatic ventralization [25–27], with the appearance of ectopic laminB+ DVB cells on the dorsal surface of animals, in some cases in organized stripes (Fig 4A). After allowing smad1 RNAi animals to develop ectopic DVB, we inhibited β-catenin-1 with RNAi. These smad1; β-catenin-1 double RNAi animals expressed anterior PCGs (sFRP-1, ndl-4) and developed opsin+ photoreceptor cells at sites of ectopic DVB, indicative of head patterning (Figs 4A and S3E). The ectopic anterior PCG expression observed traced the variable and irregular pattern of the ectopic DVB on the dorsal side of these animals (Fig 4A). smad1; β-catenin-1 double RNAi animals ultimately formed heads that emerged from the dorsal surface of the animal (Fig 4B). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. Ectopic DVB generated by DV axis perturbation is competent to induce formation of an anterior region of the AP axis after β-catenin-1 RNAi. A) smad1 RNAi animals develop ectopic DVB (laminB) that is anteriorized (sFRP-1 ndl-4+) after β-catenin-1 RNAi (10d). B) smad1 RNAi causes head-like structures to form on the dorsal surface of animals after β-catenin-1 RNAi (21d). Yellow arrowhead, eye. C) smad1 RNAi animals develop outgrowths at the site of dorsal injury. After β-catenin-1 RNAi (12d), these outgrowths form head-like structures. Yellow arrowhead, eye. D) Outgrowths in injured smad1 RNAi animals show epidermal DVB (laminB) gene expression. After β-catenin-1 RNAi (12d), these outgrowths express the anterior PCG sFRP-1. Colored boxes, area depicted in photos. Scale bars, 100 μm (A, D), 250 μm (B, C). https://doi.org/10.1371/journal.pbio.3003482.g004 We realized it could be possible to use smad1 RNAi to induce head formation by design in a desired location. We performed a dorsal injury in the pre-pharyngeal region three days after initiating smad1 RNAi, triggering outgrowth formation at the injury site that displayed ectopic laminB+ DVB gene expression. After subsequent β-catenin-1 inhibition, the outgrowths were converted to head-like structures (Fig 4C and 4D). Together, these results are consistent with a model in which muscle at the planarian DVB is competent to activate anterior PCGs and support formation of an anterior region of the AP axis in a low Wnt environment. By combining methods to generate ectopic DVB at designated locations with Wnt pathway inhibition, head location and shape can be controlled to yield a large array of body plans. The DVB region is required to promote anterior PCG expression in a low Wnt environment To test whether the DVB is required to initiate head formation in the context of β-catenin-1 RNAi, we inflicted either a lateral or internal injury after RNAi (Figs 5A and S4A). Anteriorization after β-catenin-1 RNAi requires new cell production [18,20], and loss of tissue elicits a localized increase in mitoses at the wound site [59,60]. However, only tissue removal at the lateral edge of β-catenin-1 RNAi animals was sufficient to reliably induce anterior PCG expression at the wound (Fig 5A). An internal hole-punch that removed tissue did not cause anterior PCG expression in most cases (n = 51/59) (Figs 5A and S4B), suggesting that signals at the DVB are necessary to robustly initiate anterior patterning. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. The DVB region is required to promote anterior PCG expression in a low Wnt environment. A) β-catenin-1 RNAi animals after a lateral wedge cut develop a focus of anterior PCG (sFRP-1, ndl-4) expression at the wound. B) Treatment with the MEK inhibitor PD blocks lateral anterior PCG (sFRP-1, ndl-4) expression after β-catenin-1 RNAi and DVB removal (n = 12/22). Anterior PCGs are expressed in foci at the intact DVB on the uninjured side of PD-treated β-catenin-1 RNAi animals (n = 22/22). C) bmp4; smad1 RNAi post-pharyngeal fragments express anterior PCGs (sFRP-1, ndl-4) at the anterior-facing wound despite no DVB (laminB) regeneration. However, anterior PCG expression is limited to the lateral edge of the fragment near one or two pre-existing, intact DVB domains. bmp4; smad1 RNAi fragments with both DVB domains removed show little anterior PCG expression. Colored boxes, area depicted in photos. Scale bars, 100 μm (A–C), 50 μm (B, inset). https://doi.org/10.1371/journal.pbio.3003482.g005 We next surgically removed the DVB (parasagittal amputation) and blocked its regeneration by inhibiting Erk signaling using PD0325901 (PD) (Figs 5B, S4C, and S4D). Erk signaling is required to promote wound-induced signaling that culminates in a regenerative response [61]. Erk signaling was not required, however, for the homeostatic outgrowth that accompanies ectopic head formation after β-catenin-1 RNAi (S4E Fig). DMSO-treated β-catenin-1 RNAi animals regenerated the DVB, labeled by laminB, and expressed anterior PCGs (sFRP-1, ndl-4) in large foci on the side that was regenerating, whereas PD-treated β-catenin-1 RNAi animals did not regenerate the DVB and little anterior PCG expression was detected on this side in most cases (n = 12/22) (Fig 5B). In some cases (n = 9/22), anterior PCG (sFRP-1, ndl-4) expression was detected on the injured side of PD-treated β-catenin-1 RNAi animals, despite no regeneration of epidermal DVB (laminB) (S4F Fig) or muscle DVB (lactadherin, frem-1) gene expression (S4G Fig). However, this expression did not form foci or local outgrowths indicative of head formation (S4F Fig). It is possible that anterior PCG expression can be initiated at low levels in the absence of the DVB, and that the DVB acts to promote elevated and sustained anterior PCG expression that culminates in strong foci. Importantly, anterior PCGs were expressed on the uninjured side of both DMSO- and PD-treated animals, where they could form foci, demonstrating that Erk signaling is not generally required for anteriorization during homeostatic turnover (Fig 5B). Together, these results suggest that the DVB is required for robust promotion of anterior patterning and head formation after β-catenin-1 RNAi. To further determine whether the DVB is required to promote anterior patterning, we utilized the independent context of head regeneration. Anterior PCGs can become expressed during head regeneration between 24 and 48 hpa [16,44]. Muscle DVB genes became expressed at the edge of head blastemas in a patterned manner by 36 hpa, similar in time to the appearance of anterior PCG foci at the wound (S5A Fig). We utilized inhibition of Bmp signaling to prevent DVB regeneration after transverse amputation. Bmp signaling is required for DV patterning and pathway inhibition can prevent regeneration of the DVB [25–27]. Short-term bmp4 RNAi tail fragments can express wound-induced genes and rescale positional information despite no DVB regeneration or blastema outgrowth [28]. Interestingly, bmp4 RNAi has been observed to result in asymmetric poles, displaced laterally [28]. We generated control or bmp4; smad1 RNAi post-pharyngeal fragments. Control post-pharyngeal fragments expressed anterior PCGs (sFRP-1, ndl-4) at the midline of the AFW near the regenerated DVB (laminB) (Fig 5C). By contrast, bmp4; smad1 RNAi post-pharyngeal fragments did not regenerate the DVB at their wounds, leaving the fragments with only lateral domains of DVB (pre-existing prior to injury). Anterior PCGs were expressed at the AFW of these bmp4; smad1 RNAi fragments, but this expression only occurred at the lateral fragment edge near the pre-existing, intact DVB in all cases (n = 26/26) (Fig 5C). In some cases, anterior PCGs were expressed on only one side of the fragment at the DVB (left side (n = 10/26) or right side (n = 8/26)). In other cases, two foci of anterior PCG expression were observed, with one near the left DVB and one near the right DVB (n = 8/26). We next generated bmp4; smad1 RNAi post-pharyngeal fragments and also removed one side of lateral DVB. Anterior PCGs were expressed at the AFW in most cases (n = 31/34; n = 3/34, anterior PCGs were not detected). Notably, in these 31 animals, anterior PCG expression was restricted to the lateral edge only, exclusively near the side with remaining DVB (n = 25/25, right DVB removal and n = 6/6, left DVB removal) (Fig 5C). These results suggest that the DVB is required for sustained anterior PCG expression in regeneration. To assess this possibility, we generated post-pharyngeal bmp4; smad1 RNAi fragments that also had both sides of the lateral DVB removed. In most cases, bmp4; smad1 RNAi fragments without any DVB contained few (n = 7/15) or no detectable (n = 6/15) anterior PCG-expressing cells at 7 days post-amputation (dpa) (n = 3/15 had detectable levels of sFRP-1, ndl-4). These findings support the hypothesis that the DVB is needed for anterior PCG expression in regeneration. The DVB has an anterior pole-independent role in the promotion of anterior positional information expression after Wnt inhibition The planarian anterior pole coalesces at the DVB [34] and is required to promote sustained anterior PCG expression during head regeneration [31–33]. Perturbations that affect the distribution of DVB cells also alter the location of anterior pole formation [34], suggesting that the DVB plays an active role in its positioning. In head regeneration, anterior pole formation and anterior PCG expression occur on similar timescales [16,34,44]. However, after β-catenin-1 RNAi, anterior PCG (sFRP-1, ndl-4) expression was detected at the DVB prior to anterior pole transcripts being apparent by FISH (Fig 6A). By 7 days of β-catenin-1 RNAi, 37 ectopic anterior PCG foci were detected across 9 animals. Thirty-three foci across 7 animals and 41 foci across 8 animals were detected at 14 days and 21 days of RNAi, respectively. FoxD encodes a Forkhead-family transcription factor that is expressed in a subpopulation of neoblasts (stem cells) during head regeneration and is required for the specification of neoblasts to form anterior pole progenitors [31,32]. Mature anterior pole cells also express FoxD [31,32]. Ectopic FoxD expression was not reliably detected until day 14 (n = 10 foci across 7 animals) to day 21 (n = 11 foci across 7 animals) after the initiation of β-catenin-1 RNAi (Fig 6A). We verified these results using an RNA probe to notum, which encodes a secreted Wnt inhibitor expressed in the anterior pole [17,31]. notum expression at wounds requires β-catenin-1 [17]. However, notum transcripts could still be detected in the anterior pole after β-catenin-1 RNAi [19] (S6A Fig). In agreement with the FoxD results, ectopic notum expression was not detected until day 14 (n = 1 focus across 6 animals) to day 21 (n = 7 foci across 9 animals) after initiation of β-catenin-1 RNAi (Fig 6A). Although transcript detection could be limited by FISH sensitivity, this order of events suggests that it is the expression of anterior PCGs at the DVB that is first associated with head formation, and that pole formation might follow and facilitate pattern resolution during head formation. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 6. The DVB has an anterior pole-independent role in the promotion of anterior PCG expression after β-catenin-1 RNAi. A) β-catenin-1 RNAi animals ectopically express anterior PCGs (sFRP-1, ndl-4) by 7d of RNAi, prior to detection of ectopic anterior pole transcripts (FoxD, notum) at 14–21d. Blue arrowhead, ectopic anterior pole. Blue asterisk, ectopic anterior PCG focus. B) FoxD; β-catenin-1 RNAi animals express anterior PCGs (sFRP-1, ndl-4, ndl-5) in ectopic foci at the DVB. C) The anterior pole is positioned at the DVB in the head. D) Top: Anterior pole formation is preceded by anterior PCG expression at the DVB after β-catenin-1 RNAi. Bottom left: Ectopic DVB on the dorsal surface of the animal (smad1 RNAi) leads to the ectopic dorsal expression of anterior PCGs after β-catenin-1 RNAi. Similar results were obtained after D-V or DVB transplantation (see Fig 3C and 3E). Bottom right: DVB removal (parasagittal cut, PD treatment) blocks lateral anterior PCG expression after β-catenin-1 RNAi. Similar results were obtained at anterior-facing wounds of bmp4; smad1 RNAi animals (see Fig 5C). A, anterior. P, posterior. E) Model: DV patterning acts to establish the location of the DVB at the midpoint of the DV axis, at the lateral animal margin. The DVB then serves as a landmark to allow growth and patterning of regions of the AP axis, such as during head regeneration or head formation in a low Wnt environment. This allows the DV axis to control orthogonal AP axis region growth and pattern formation. Scale bars, 100 μm (A, B), 50 μm (A, inset). https://doi.org/10.1371/journal.pbio.3003482.g006 To test whether the anterior pole is required for anterior PCG expression during homeostatic transformation following β-catenin-1 RNAi, we inhibited pole formation by FoxD RNAi, then fed animals β-catenin-1 double-stranded RNA (Figs 6B and S6B). FoxD inhibition did not dramatically affect anterior PCG (sFRP-1, ndl-4, ndl-5) gradients in control animals (S6C Fig) and FoxD; β-catenin-1 double RNAi animals still induced expression of anterior PCGs at the DVB (Fig 6B), suggesting that the DVB plays an anterior pole-independent role in the promotion of anterior positional information expression after Wnt inhibition. The AP and DV axes maintain pattern largely independently in homeostasis The maintenance of adult positional information in planarians is important for tissue turnover and for regeneration after injury [10]. To maintain positional information with suitable spatial relationships between axes during homeostatic tissue maintenance, positional information along one axis could, in principle, require input from another axis. Inhibition of β-catenin-1 results in homeostatic anteriorization [11–13] (S1A Fig). However, DV asymmetric tissues [12] and both dorsally biased (bmp4, nlg-8) and ventrally biased (admp, nlg-7) PCG expression remained unchanged after β-catenin-1 RNAi (Figs 1A, S1B, and S1C), suggesting that Wnt signaling is not overtly required for the maintenance of DV pattern during adult tissue turnover. bmp4, the inhibition of which leads to homeostatic ventralization [25–27] (S1D Fig), has been proposed to integrate patterning of the DV and AP axes by suppressing Wnt [39]. In agreement with findings in Clark (2023) [39], we found that the domain of wnt1+ cells (the posterior pole) was expanded after 45 days of bmp4 RNAi (Figs 1B and S1E). The fz4-1 expression domain was similarly lengthened in bmp4 RNAi animals. However, no overt changes in the wntP-2 or ndl-3 expression domains were apparent (Fig 1B). The anterior pole, labeled by notum, was expanded, albeit to a lesser extent than the posterior pole (Figs 1B, S1F, and S1G). However, no dramatic changes in anterior PCG expression domains (sFRP-1, ndl-5) were observed (Figs 1B and S1F), as previously reported [39]. Furthermore, the relative order of PCG expression domains along the AP axis remained largely unchanged after bmp4 RNAi, suggesting that Bmp signaling is not strictly required for the maintenance of broad AP pattern, but might limit the extent of the posterior-most patterning domain of the AP axis. These results point towards the possibility that axis integration might predominantly be accomplished during regeneration or during ectopic head formation in a low Wnt environment, with pattern thereafter largely stably maintained by each axis independently during cell turnover. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Ectopic heads form at the DVB after β-catenin-1 RNAi. A) Dorsally biased (bmp4, nlg-8) and ventrally biased (admp, nlg-7) PCG expression after β-catenin-1 RNAi (21d). B) PCG expression domains in bmp4 RNAi animals (45d). Blue arrow denotes anterior or posterior boundary of PCG expression domain. Diagrams show the mean PCG domain length ± standard deviation relative to body length in bmp4 RNAi animals. Six to seventeen animals per PCG were measured. Data points from individual animals are listed in S1 Data. Significant domain shifts are marked: *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001 by two-tailed t-tests. C) The epidermal laminB+ DVB is positioned ventral to the dorsally high bmp4 expression domain in muscle and is coincident with the known location of lateral admp expression in muscle. Anterior and posterior PCG expression is concentrated near the DVB during regeneration (96 hpa). D) Fluorescence intensity along a line is plotted and smoothed data is represented with a thick stroke. The foci of anterior (ndl-4) and posterior (fz4-1) PCG expression show peaked behavior on the dorsal and ventral sides juxtaposing the laminB+ DVB. Individual data points are listed in S1 Data. E) Ectopic anterior PCG (sFRP-1) expression near the DVB after β-catenin-1 RNAi (14d). D, dorsal; V, ventral. F) Ectopic anterior PCG (sFRP-1, ndl-4) expression after β-catenin-1 RNAi (18d) is detected in longitudinal (myoD+), circular (nkx1-1+), and lateral DV (nk4+) muscle at the lateral animal edge. G) Co-expression of anterior PCGs (sFRP-1, ndl-4) and the patterning molecule admp in muscle at the DVB after β-catenin-1 RNAi (21d). Number of admp+ cells that co-localize with sFRP-1 and ndl-4 transcripts (magenta) out of total admp+ cells (green) is shown. n = 4 animals. Colored boxes, area depicted in photos. Scale bars, 100 μm (A, B, E), 50 μm (C, F, G). https://doi.org/10.1371/journal.pbio.3003482.g001 Organization of PCG expression at the DVB in regeneration During head regeneration, anterior pole cells labeled by notum transcripts form a focus at the DVB [34]. We used epidermal cells expressing laminB, which mark the lateral DV median plane [38,40], as a reference point for the DVB. At 96 hours post-amputation (hpa), this DVB location was positioned ventral to the dorsally high bmp4 domain in muscle and was coincident with the known location of admp expression in muscle [9,41,42] at the lateral DV median plane (Figs 1C and S1H). Expression of the anterior PCGs sFRP-1, ndl-4, and ndl-5 was centered around the DVB (Figs 1C and S1I) and showed peaked behavior on the dorsal and ventral sides that juxtapose the laminB+ DVB location (Figs 1D and S1J). The mature posterior pole labeled by wnt1 is dorsally biased [13,16,43] but formed a cluster near the DVB at 96 hpa during tail regeneration (Figs 1C and S1I). Expression of the posterior PCGs wnt11-1 and fz4-1 was centered around the laminB+ DVB (Figs 1C and S1I) and showed peaked behavior on the dorsal and ventral sides juxtaposing the DVB (Figs 1D and S1K), similar to the case of anterior PCGs. These results connect patterning processes involving PCG expression at the head and tail tips to the location of the DVB and support a potential mechanism for axis integration [34] wherein the DVB serves as a critical landmark to guide proper regeneration along the AP axis. Ectopic heads form at the DVB after β-catenin-1 RNAi Ectopic heads form at the animal periphery after β-catenin-1 RNAi, with the occasional exception of a head near the planarian pharynx, during a process of homeostatic tissue turnover [11–13]. β-catenin-1 RNAi resulted in ectopic focal expression of the anterior PCGs sFRP-1 and ndl-4 laterally, centered at the laminB+ DVB (Figs 1E and S1L). Anterior PCGs showed peaked ectopic expression on the dorsal and ventral sides juxtaposing the DVB in some regions (S1L Fig). Ectopic anterior PCG (sFRP-1, ndl-4) expression occurred in multiple muscle subtypes present near the DVB [44,45], including in longitudinal muscle (muscle that is oriented along the AP axis), circular muscle (muscle that is oriented along the ML axis), and lateral DV muscle (muscle that is oriented along the DV axis at the body margin) (Figs 1F and S1M). Ectopic anterior PCG (sFRP-1, ndl-4) expression in β-catenin-1 RNAi animals occurred in a large fraction (62.0%) of muscle cells expressing the DVB PCG admp [9,41,42] (Fig 1G). Ectopic anterior PCGs (sFRP-1, ndl-4) were co-expressed with additional muscle-enriched genes at the DVB, including lactadherin (dd_5463, identified below) and netrin-1 [46,47] (S1N Fig). Although anterior PCG (sFRP-1, ndl-4) expression was concentrated in foci at the DVB after β-catenin-1 RNAi, sporadic cells expressing anterior PCGs were also detected dispersed (not in foci) in the animal (S1A Fig). Hereafter, we analyze specifically the anterior PCG expression that occurred in a concentrated manner at the DVB in β-catenin-1 RNAi animals. A spatial atlas of gene expression, including genes encoding cell surface and signaling molecules, in muscle at the DVB Given the potential significance of the planarian DVB in patterning, and the fact that its molecular determinants are poorly understood, we sought to define a spatial atlas of gene expression for muscle at the DVB. We performed single-cell RNA sequencing (scRNA-seq) of the lateral edges of animals to enrich for DVB cells (S2A and S2B Fig). All major muscle subtypes at the DVB were represented in our data (S2C Fig). Known markers of the body margin from prior literature (admp [41,42], egf-6 [48], wnt-5 [14,16], unc5-B [46], ephR-2 [25], netrin-1 [46,47], follistatin [49,50], noggin-1 [25,41], and netrin-3 [44,47,51]) displayed expression predominantly in longitudinal muscle at the DVB (S2C Fig). Some DVB markers also displayed expression in lateral DV muscle (nk4+) and/or circular muscle (nkx1-1+) (S2C Fig), but subclustering data from circular muscle did not separate cells by their DV position of origin (S2D Fig). Subclustering scRNA-seq data from longitudinal muscle cells identified four total subclusters. Transcripts known to be expressed at the DVB were enriched in subcluster 1 (Figs 2A, S2E, and S2F), suggesting genes with enriched expression in this subcluster could represent DVB-expressed genes. Thirty-two total genes had significantly enriched expression in subcluster 1; nine of these genes were previously characterized as being expressed at the body margin, with 23 new DVB genes. We assessed the expression of these 32 genes by FISH, together with laminB as a common spatial reference marker. This generated a spatial atlas of gene expression in muscle at the DVB (Figs 2B–2D, S2G, and S2H). Previously identified DVB markers were expressed in either dorsal (admp, egf-6, wnt-5, unc5-B, ephR-2), ventral (netrin-1, follistatin), overlapping (noggin-1), or both dorsal and ventral (netrin-3) domains of the DV axis relative to laminB (Fig 2B—2D). In addition to these known DVB markers, nine additional genes (lactadherin (dd_5463), ephR-3 (dd_27668, dd_39545), troponin-2 (dd_7974), frem-1 (dd_5187), netR, dd_30210, delta-3, coupTF-1 (dd_33286), and zfp-4 (dd_29263, dd_21349)) were also expressed in particular domains relative to laminB at the DVB (Figs 2B—2D, S2I, and S2J). Lactadherin is a secreted protein that can bind anionic phospholipids [52]. Ephrin receptors mediate the effects of Ephrins, a family of membrane-bound signaling ligands that promote a variety of cellular responses including attraction, repulsion, and migration [53]. frem-1 encodes an extracellular matrix protein that contributes to epithelial-mesenchymal integrity in development [54]. Netrin receptors mediate the attractant and repellent properties of secreted Netrins that guide cell and axon migration in development [55]. dd_30210 encodes a novel protein with no characterized PFAM domains. delta-3 encodes a membrane-bound Notch signaling ligand. Notch signaling regulates diverse developmental processes, including cell fate determination [56]. COUP transcription factors encode orphan nuclear receptors that regulate diverse cellular processes in development [57]. zfp-4 encodes a novel zinc finger protein. Using our scRNA-seq data, we calculated the Pearson correlation coefficient for all pairwise comparisons of the 18 genes in the DVB atlas with expression detected by FISH. Within longitudinal muscle cells at the DVB, the expression of many genes showed a high degree of correlation (r > 0.5) (S2K Fig), indicating that individual muscle cells co-express multiple markers of the DVB. There were 14 genes with transcripts enriched in longitudinal muscle subcluster 1 (S2E and S2F Fig) that did not display clear DVB-restricted expression by FISH, either because expression was broad or not detected. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Spatial DVB gene expression atlas in muscle. A) Cartoon shows region of the animal isolated for scRNA-seq. Subclustering of longitudinal muscle cells from scRNA-seq data reveals a DVB cell population in UMAP plots. Novel markers of the DVB are labeled in black. B) FISH characterization of the DVB region. Images show the expression of genes along the DV axis relative to the spatial domain of the canonical epidermal marker of the DVB, laminB. C) FISH images show the expression of genes that label dorsal (nlg-8) and ventral (nlg-7) muscle, lateral DV muscle (nk4), and a DVB epidermis domain (equinox) relative to laminB. D) Cartoon representation of expression domains at the DVB. Dorsal (nlg-8) and ventral (nlg-7) PCGs are shown as graded columns. Scale bars, 50 μm (B, C). https://doi.org/10.1371/journal.pbio.3003482.g002 Taken together, the molecular data indicate that the DVB, often defined anatomically, can be viewed as an intricately patterned domain of gene expression that in muscle includes a large array of genes encoding signaling factors, cell-surface molecules, and transcription regulators (Figs 2D, S2G, and S2H). DVB muscle is thus a prominent but poorly understood pattern element of the planarian body plan. Ectopic DVB is permissive for the formation of an anterior region of the AP axis after β-catenin-1 RNAi To determine whether the DVB is sufficient to promote the formation of an anterior region of the AP axis after β-catenin-1 RNAi, we endeavored to generate animals with ectopic DVB. Transplantation of tissue with reversed DV polarity has been shown to induce ectopic DVB, labeled by epidermal expression of laminB, in goblet-shaped outgrowths [38,58]. By contrast, similar tissue transplantation without DV reversal leads to tissue integration without DVB formation. We used this approach to generate outgrowths with ectopic DVB in the post-pharyngeal region of the animal (Figs 3A and S3A). These outgrowths contained DVB muscle gene expression signatures, including for genes encoding the secreted molecules admp, lactadherin, and frem-1, indicating that both epidermis and underlying muscle acquire an ectopic DVB identity in outgrowths (Figs 3B and S3B). These outgrowths had AP character typical of the location of the transplant [38] (S3C Fig) and lacked anterior identity (Fig 3C). After allowing dorsal-to-ventral (D-V) transplant outgrowths to form, we initiated β-catenin-1 RNAi. Whereas head structures are normally constrained to the periphery of β-catenin-1 RNAi animals, D-V transplant outgrowths developed into goblet-shaped head-like structures marked by the presence of eyes in the middle of the animal (Fig 3A). These outgrowths (both dorsal and ventral outgrowths), after β-catenin-1 RNAi, expressed anterior PCGs (sFRP-1, ndl-4) and contained opsin+ photoreceptor cells (Figs 3C and S3D). By contrast, dorsal-to-dorsal (D-D) transplant animals simply healed the wound (n = 13/13) and these transplantation sites did not express anterior PCGs or transform into heads following β-catenin-1 RNAi (Fig 3A and 3D). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Ectopic DVB generated by transplantation is competent to induce formation of an anterior region of the AP axis after β-catenin-1 RNAi. A) Transplant animals that received tissue grafted with reversed DV polarity develop goblet-shaped outgrowths. After β-catenin-1 RNAi (10–14d), these outgrowths form head-like structures. Yellow arrowhead, eye. B) D-V transplant outgrowths show epidermal (laminB) and muscle (admp) DVB gene expression. C) D-V transplant outgrowths express anterior PCGs (sFRP-1, ndl-4) and form eyes (opsin) after β-catenin-1 RNAi (10–14d). Yellow arrowhead, eye. D) D-D transplants do not result in ectopic DVB (laminB) or anterior PCG (sFRP-1, ndl-4) gene expression after β-catenin-1 RNAi. E) DVB transplant animals develop dorsal outgrowths. After β-catenin-1 RNAi (10d), these outgrowths express anterior PCGs (sFRP-1, ndl-4) and form eye cells (opsin). Colored boxes, area depicted in photos. Scale bars, 250 μm (A), 100 μm (B, D), 50 μm (C, E). https://doi.org/10.1371/journal.pbio.3003482.g003 To complement the use of transplantation to generate ectopic DVB, we next transplanted the DVB region itself. Transplantation of a small posterior DVB-containing segment of the animal margin to the midline of a recipient animal induced the formation of a dorsal outgrowth in the recipient. These outgrowths expressed anterior PCGs (sFRP-1, ndl-4) and contained opsin+ photoreceptor cells after β-catenin-1 RNAi (Fig 3E), indicating that ectopically positioned DVB is competent to promote anterior patterning after Wnt inhibition. As an independent method to generate ectopic DVB, we utilized smad1 RNAi. smad1 encodes a downstream mediator of Bmp signaling and has a prominent role in planarian DV patterning [25]. Inhibition of bmp4 or smad1 leads to homeostatic ventralization [25–27], with the appearance of ectopic laminB+ DVB cells on the dorsal surface of animals, in some cases in organized stripes (Fig 4A). After allowing smad1 RNAi animals to develop ectopic DVB, we inhibited β-catenin-1 with RNAi. These smad1; β-catenin-1 double RNAi animals expressed anterior PCGs (sFRP-1, ndl-4) and developed opsin+ photoreceptor cells at sites of ectopic DVB, indicative of head patterning (Figs 4A and S3E). The ectopic anterior PCG expression observed traced the variable and irregular pattern of the ectopic DVB on the dorsal side of these animals (Fig 4A). smad1; β-catenin-1 double RNAi animals ultimately formed heads that emerged from the dorsal surface of the animal (Fig 4B). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. Ectopic DVB generated by DV axis perturbation is competent to induce formation of an anterior region of the AP axis after β-catenin-1 RNAi. A) smad1 RNAi animals develop ectopic DVB (laminB) that is anteriorized (sFRP-1 ndl-4+) after β-catenin-1 RNAi (10d). B) smad1 RNAi causes head-like structures to form on the dorsal surface of animals after β-catenin-1 RNAi (21d). Yellow arrowhead, eye. C) smad1 RNAi animals develop outgrowths at the site of dorsal injury. After β-catenin-1 RNAi (12d), these outgrowths form head-like structures. Yellow arrowhead, eye. D) Outgrowths in injured smad1 RNAi animals show epidermal DVB (laminB) gene expression. After β-catenin-1 RNAi (12d), these outgrowths express the anterior PCG sFRP-1. Colored boxes, area depicted in photos. Scale bars, 100 μm (A, D), 250 μm (B, C). https://doi.org/10.1371/journal.pbio.3003482.g004 We realized it could be possible to use smad1 RNAi to induce head formation by design in a desired location. We performed a dorsal injury in the pre-pharyngeal region three days after initiating smad1 RNAi, triggering outgrowth formation at the injury site that displayed ectopic laminB+ DVB gene expression. After subsequent β-catenin-1 inhibition, the outgrowths were converted to head-like structures (Fig 4C and 4D). Together, these results are consistent with a model in which muscle at the planarian DVB is competent to activate anterior PCGs and support formation of an anterior region of the AP axis in a low Wnt environment. By combining methods to generate ectopic DVB at designated locations with Wnt pathway inhibition, head location and shape can be controlled to yield a large array of body plans. The DVB region is required to promote anterior PCG expression in a low Wnt environment To test whether the DVB is required to initiate head formation in the context of β-catenin-1 RNAi, we inflicted either a lateral or internal injury after RNAi (Figs 5A and S4A). Anteriorization after β-catenin-1 RNAi requires new cell production [18,20], and loss of tissue elicits a localized increase in mitoses at the wound site [59,60]. However, only tissue removal at the lateral edge of β-catenin-1 RNAi animals was sufficient to reliably induce anterior PCG expression at the wound (Fig 5A). An internal hole-punch that removed tissue did not cause anterior PCG expression in most cases (n = 51/59) (Figs 5A and S4B), suggesting that signals at the DVB are necessary to robustly initiate anterior patterning. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. The DVB region is required to promote anterior PCG expression in a low Wnt environment. A) β-catenin-1 RNAi animals after a lateral wedge cut develop a focus of anterior PCG (sFRP-1, ndl-4) expression at the wound. B) Treatment with the MEK inhibitor PD blocks lateral anterior PCG (sFRP-1, ndl-4) expression after β-catenin-1 RNAi and DVB removal (n = 12/22). Anterior PCGs are expressed in foci at the intact DVB on the uninjured side of PD-treated β-catenin-1 RNAi animals (n = 22/22). C) bmp4; smad1 RNAi post-pharyngeal fragments express anterior PCGs (sFRP-1, ndl-4) at the anterior-facing wound despite no DVB (laminB) regeneration. However, anterior PCG expression is limited to the lateral edge of the fragment near one or two pre-existing, intact DVB domains. bmp4; smad1 RNAi fragments with both DVB domains removed show little anterior PCG expression. Colored boxes, area depicted in photos. Scale bars, 100 μm (A–C), 50 μm (B, inset). https://doi.org/10.1371/journal.pbio.3003482.g005 We next surgically removed the DVB (parasagittal amputation) and blocked its regeneration by inhibiting Erk signaling using PD0325901 (PD) (Figs 5B, S4C, and S4D). Erk signaling is required to promote wound-induced signaling that culminates in a regenerative response [61]. Erk signaling was not required, however, for the homeostatic outgrowth that accompanies ectopic head formation after β-catenin-1 RNAi (S4E Fig). DMSO-treated β-catenin-1 RNAi animals regenerated the DVB, labeled by laminB, and expressed anterior PCGs (sFRP-1, ndl-4) in large foci on the side that was regenerating, whereas PD-treated β-catenin-1 RNAi animals did not regenerate the DVB and little anterior PCG expression was detected on this side in most cases (n = 12/22) (Fig 5B). In some cases (n = 9/22), anterior PCG (sFRP-1, ndl-4) expression was detected on the injured side of PD-treated β-catenin-1 RNAi animals, despite no regeneration of epidermal DVB (laminB) (S4F Fig) or muscle DVB (lactadherin, frem-1) gene expression (S4G Fig). However, this expression did not form foci or local outgrowths indicative of head formation (S4F Fig). It is possible that anterior PCG expression can be initiated at low levels in the absence of the DVB, and that the DVB acts to promote elevated and sustained anterior PCG expression that culminates in strong foci. Importantly, anterior PCGs were expressed on the uninjured side of both DMSO- and PD-treated animals, where they could form foci, demonstrating that Erk signaling is not generally required for anteriorization during homeostatic turnover (Fig 5B). Together, these results suggest that the DVB is required for robust promotion of anterior patterning and head formation after β-catenin-1 RNAi. To further determine whether the DVB is required to promote anterior patterning, we utilized the independent context of head regeneration. Anterior PCGs can become expressed during head regeneration between 24 and 48 hpa [16,44]. Muscle DVB genes became expressed at the edge of head blastemas in a patterned manner by 36 hpa, similar in time to the appearance of anterior PCG foci at the wound (S5A Fig). We utilized inhibition of Bmp signaling to prevent DVB regeneration after transverse amputation. Bmp signaling is required for DV patterning and pathway inhibition can prevent regeneration of the DVB [25–27]. Short-term bmp4 RNAi tail fragments can express wound-induced genes and rescale positional information despite no DVB regeneration or blastema outgrowth [28]. Interestingly, bmp4 RNAi has been observed to result in asymmetric poles, displaced laterally [28]. We generated control or bmp4; smad1 RNAi post-pharyngeal fragments. Control post-pharyngeal fragments expressed anterior PCGs (sFRP-1, ndl-4) at the midline of the AFW near the regenerated DVB (laminB) (Fig 5C). By contrast, bmp4; smad1 RNAi post-pharyngeal fragments did not regenerate the DVB at their wounds, leaving the fragments with only lateral domains of DVB (pre-existing prior to injury). Anterior PCGs were expressed at the AFW of these bmp4; smad1 RNAi fragments, but this expression only occurred at the lateral fragment edge near the pre-existing, intact DVB in all cases (n = 26/26) (Fig 5C). In some cases, anterior PCGs were expressed on only one side of the fragment at the DVB (left side (n = 10/26) or right side (n = 8/26)). In other cases, two foci of anterior PCG expression were observed, with one near the left DVB and one near the right DVB (n = 8/26). We next generated bmp4; smad1 RNAi post-pharyngeal fragments and also removed one side of lateral DVB. Anterior PCGs were expressed at the AFW in most cases (n = 31/34; n = 3/34, anterior PCGs were not detected). Notably, in these 31 animals, anterior PCG expression was restricted to the lateral edge only, exclusively near the side with remaining DVB (n = 25/25, right DVB removal and n = 6/6, left DVB removal) (Fig 5C). These results suggest that the DVB is required for sustained anterior PCG expression in regeneration. To assess this possibility, we generated post-pharyngeal bmp4; smad1 RNAi fragments that also had both sides of the lateral DVB removed. In most cases, bmp4; smad1 RNAi fragments without any DVB contained few (n = 7/15) or no detectable (n = 6/15) anterior PCG-expressing cells at 7 days post-amputation (dpa) (n = 3/15 had detectable levels of sFRP-1, ndl-4). These findings support the hypothesis that the DVB is needed for anterior PCG expression in regeneration. The DVB has an anterior pole-independent role in the promotion of anterior positional information expression after Wnt inhibition The planarian anterior pole coalesces at the DVB [34] and is required to promote sustained anterior PCG expression during head regeneration [31–33]. Perturbations that affect the distribution of DVB cells also alter the location of anterior pole formation [34], suggesting that the DVB plays an active role in its positioning. In head regeneration, anterior pole formation and anterior PCG expression occur on similar timescales [16,34,44]. However, after β-catenin-1 RNAi, anterior PCG (sFRP-1, ndl-4) expression was detected at the DVB prior to anterior pole transcripts being apparent by FISH (Fig 6A). By 7 days of β-catenin-1 RNAi, 37 ectopic anterior PCG foci were detected across 9 animals. Thirty-three foci across 7 animals and 41 foci across 8 animals were detected at 14 days and 21 days of RNAi, respectively. FoxD encodes a Forkhead-family transcription factor that is expressed in a subpopulation of neoblasts (stem cells) during head regeneration and is required for the specification of neoblasts to form anterior pole progenitors [31,32]. Mature anterior pole cells also express FoxD [31,32]. Ectopic FoxD expression was not reliably detected until day 14 (n = 10 foci across 7 animals) to day 21 (n = 11 foci across 7 animals) after the initiation of β-catenin-1 RNAi (Fig 6A). We verified these results using an RNA probe to notum, which encodes a secreted Wnt inhibitor expressed in the anterior pole [17,31]. notum expression at wounds requires β-catenin-1 [17]. However, notum transcripts could still be detected in the anterior pole after β-catenin-1 RNAi [19] (S6A Fig). In agreement with the FoxD results, ectopic notum expression was not detected until day 14 (n = 1 focus across 6 animals) to day 21 (n = 7 foci across 9 animals) after initiation of β-catenin-1 RNAi (Fig 6A). Although transcript detection could be limited by FISH sensitivity, this order of events suggests that it is the expression of anterior PCGs at the DVB that is first associated with head formation, and that pole formation might follow and facilitate pattern resolution during head formation. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 6. The DVB has an anterior pole-independent role in the promotion of anterior PCG expression after β-catenin-1 RNAi. A) β-catenin-1 RNAi animals ectopically express anterior PCGs (sFRP-1, ndl-4) by 7d of RNAi, prior to detection of ectopic anterior pole transcripts (FoxD, notum) at 14–21d. Blue arrowhead, ectopic anterior pole. Blue asterisk, ectopic anterior PCG focus. B) FoxD; β-catenin-1 RNAi animals express anterior PCGs (sFRP-1, ndl-4, ndl-5) in ectopic foci at the DVB. C) The anterior pole is positioned at the DVB in the head. D) Top: Anterior pole formation is preceded by anterior PCG expression at the DVB after β-catenin-1 RNAi. Bottom left: Ectopic DVB on the dorsal surface of the animal (smad1 RNAi) leads to the ectopic dorsal expression of anterior PCGs after β-catenin-1 RNAi. Similar results were obtained after D-V or DVB transplantation (see Fig 3C and 3E). Bottom right: DVB removal (parasagittal cut, PD treatment) blocks lateral anterior PCG expression after β-catenin-1 RNAi. Similar results were obtained at anterior-facing wounds of bmp4; smad1 RNAi animals (see Fig 5C). A, anterior. P, posterior. E) Model: DV patterning acts to establish the location of the DVB at the midpoint of the DV axis, at the lateral animal margin. The DVB then serves as a landmark to allow growth and patterning of regions of the AP axis, such as during head regeneration or head formation in a low Wnt environment. This allows the DV axis to control orthogonal AP axis region growth and pattern formation. Scale bars, 100 μm (A, B), 50 μm (A, inset). https://doi.org/10.1371/journal.pbio.3003482.g006 To test whether the anterior pole is required for anterior PCG expression during homeostatic transformation following β-catenin-1 RNAi, we inhibited pole formation by FoxD RNAi, then fed animals β-catenin-1 double-stranded RNA (Figs 6B and S6B). FoxD inhibition did not dramatically affect anterior PCG (sFRP-1, ndl-4, ndl-5) gradients in control animals (S6C Fig) and FoxD; β-catenin-1 double RNAi animals still induced expression of anterior PCGs at the DVB (Fig 6B), suggesting that the DVB plays an anterior pole-independent role in the promotion of anterior positional information expression after Wnt inhibition. Discussion Positional information along adult planarian body axes is controlled by two largely independent signaling pathways: Wnt for the AP axis and Bmp for the DV axis [10]. This feature is reminiscent of Drosophila development, where AP and DV axes are established by largely autonomous, orthogonal patterning systems [62]. Utilization of mechanistically separate molecular pathways to pattern orthogonal tissue axes might represent a common strategy to achieve stable patterns in biology. The planarian AP and DV axes maintain pattern largely independently during homeostatic cell turnover. At the same time, some mechanism of coordination between axes could be deployed in diverse animal contexts to ensure a suitable spatial relationship. Understanding how coordination of tissue axes is achieved is thus fundamental to problems of animal development and regeneration. We propose that orthogonal coordination of AP and DV patterns in adult planarians occurs primarily during outgrowth formation, such as during head regeneration or homeostatic head formation after Wnt inhibition, as opposed to during pattern maintenance in tissue turnover. Bmp signaling does restrict the posterior extent of Wnt gene expression domains in planarians during tissue turnover [39]. However, we suggest that the primary linkage between patterning of distinct planarian body axis regions during regeneration involves DV-patterning processes that form the DVB at the lateral edge of the DV median plane, or body margin. The DVB then regulates AP-axis growth and patterning, restricting growth and anterior positional information expression to the DVB in a Wnt-low environment (Fig 6C–6E). The linkage of AP-axis patterning processes to the position of the DVB is also intimately connected to the ML axis. The anterior pole coalesces specifically at the intersection of three landmarks (the DVB, the midline, and at AFWs) during head regeneration [34]. Transplantation of tissue with reversed DV polarity can lead to the formation of an ectopic head and/or tail in resultant outgrowths [38,63]. Rojo-Laguna (2019) shows that in AP polarized outgrowths the midline marker slit becomes patterned along the emergent ML axis, whereas outgrowths that lack AP polarity display aberrant expression of slit [63]. It has been proposed that a PAK family kinase coordinates pattern regeneration along the planarian AP and ML axes [64]. After β-catenin-1 RNAi, ectopic midlines are established in ectopic heads that emerge from the animal lateral edge [19]. Although the role of the DVB in promoting anterior PCG expression after β-catenin-1 RNAi is independent of the anterior pole, the anterior pole can subsequently coalesce at the DVB, where it likely acts to refine anterior pattern. It is possible that focal anterior PCG expression at the DVB influences the emergence of an anterior pole at the DVB. Early embryonic patterning can involve influence of one body axis on another. However, border-like signaling centers that are spatially restricted along one axis (such as a planarian DVB-like border) and that influence growth and patterning of another axis are typically not present during initial embryonic axis pattern establishment across animals (e.g., in Caenorhabditis elegans, Drosophila, zebrafish, Xenopus, and mouse). Instead, early pattern-initiating factors (e.g., asymmetric maternal factors, sperm entry location, cortical rotation) lead to patterning processes (such as morphogen gradient formation or organizer formation) that establish axial patterns [65,66]. In Xenopus embryos, as in adult planarians, Wnt and Bmp signaling pattern orthogonal AP and DV axes [5–7]. The organizer in Xenopus is formed on the prospective dorsal-posterior side of the embryo, following cortical rotation after fertilization, by nuclear β-catenin and vegetal factors (Xnr proteins) [65,67]. There, the organizer controls DV mesoderm patterning and marks the site of gastrulation. This location thereafter expresses Wnt ligands that promote posterior patterning of the AP axis (e.g., AP neuroectoderm patterning) [68]. It has been suggested that integration of Wnt and Bmp signaling can involve Smad1 phosphorylation by GSK3 [69]. Bmp signaling, which regulates DV axis pattern, can also influence convergent extension that promotes lengthening of the AP axis in zebrafish [70]. Patterning centers formed at the midpoint of a tissue axis, however, do have prominent roles in certain later stages of development. For example, the DV boundary in the Drosophila wing disc is involved in coordinating growth along the proximodistal axis [71]. In vertebrate limb development, the apical ectodermal ridge resides at the boundary between the dorsal and ventral ectoderm where it promotes proximodistal outgrowth and patterning [72–74]. One possibility is that outgrowths with patterned axes that emerge later in development, such as in appendage formation, or in adulthood, such as during regeneration, are predominant contexts where boundary-like signaling centers have axis-organizing roles. Because these outgrowths need to establish and coordinate axial pattern, they might rely on boundary-like signaling centers with positions determined by embryonic patterning, thereby harnessing existing tissue positional cues to direct new growth. The capacity for whole-body regeneration (ability to form large body axis regions after amputation) is widespread in the Bilateria, including in platyhelminthes, annelids, molluscs, nemertean worms, echinoderms, hemichordates, and urochordates [2]. Whether DVB-like regions exist and promote regeneration in other animals with whole-body regeneration, similar to the case of the planarian DVB, is unknown and will be important to assess. Adult cues, like the DVB, might prove to be fundamental to growth and coordination of axial patterning in regenerative outgrowths broadly in animals. In addition to DVB-like signaling centers, conflict between dorsal and ventral tissue regions can lead to outgrowth in certain experimental contexts. For instance, DV misalignment, or conflict, generated by transplantation can lead to supernumerary limbs during development [75] and during regeneration [76,77]. Transplantations generating positional conflicts of varied types can lead to regenerative outgrowths in diverse organisms, from crickets [78] to salamanders [79] to planarians [3]. Various models have been proposed to account for these observations, such as those involving intercalation of positional values between juxtaposed positions combined with distalization in the case of certain appendages [80,81] and a boundary model for limbs involving production of distalizing factors at the intersection of AP and DV boundaries [82]. Because of the experimental capacity to induce ectopic DVB and to generate a Wnt-low environment in planarians, unprecedented control over an adult animal body plan can be experimentally achieved. Ectopic heads can be induced to form at desired locations by experimental design. Head shape can also be influenced by the pattern of ectopic DVB, such as being induced to form in goblet-like shapes at DVB rings or along dorsal ridges of DVB. Utilizing a pattern element from one axis to determine the patterning outcome on another axis provides one solution to the substantial challenge in regeneration of replacing tissues with a spatial organization that integrates new tissues with pre-existing patterns in the remaining body. Methods Animal husbandry Asexual Schmidtea mediterranea strain CIW4 animals were cultured in 1× Montjuic water (1.6 mmol/l NaCl, 1.0 mmol/l CaCl2, 1.0 mmol/l MgSO4, 0.1 mmol/l MgCl2, 0.1 mmol/l KCl, and 1.2 mmol/l NaHCO3 prepared in Milli-Q water) at 20°C in the dark. Animals were fed homogenized beef liver and starved at least 7 days prior to experiments. 10× single-cell mRNA sequencing and analysis Tissue fragments from the lateral edges of ~20 large animals were amputated using a scalpel and collected in 1× Montjuic water. Montjuic water was replaced with 0.25% Trypsin-EDTA (1×) and tissue fragments were incubated for 6 min with vigorous pipetting to allow for cell dissociation. The sample was then centrifuged at 500g for 5 min, resuspended in ice-cold CMFB [calcium–magnesium free solution with 0.1% bovine serum albumin (BSA) (400 mg/L NaH2PO4, 800 mg/L NaCl, 1,200 mg/L KCl, 800 mg/L NaHCO3, 240 mg/L glucose, 0.1% BSA, 15 mM HEPES, pH7.3)], and passed through a 40 μm filter. Debris was removed from the dissociated cells with a spiral microfluidic device. The device was a gift from Kyungyong Choi and Jongyoon Han at MIT. It was used as described in [83]. Briefly, the cells were suspended in 20 mL of CMFB and circulated through the device at 16 mL/min in closed-loop separation mode, returning the output of the inner wall outlet to the sample tube, and discarding the output from the outer wall outlet. Once the volume in the sample tube was reduced to approximately 5 mL, the sample was washed at the same pump rate with an additional 40 mL of CMFB, which was added 1 mL at a time to the sample tube so as to keep the sample volume at approximately 5 mL. After washing, the sample was concentrated by further pumping until the input tube was nearly empty. The sample input tube was then disconnected and the sample (approximately 3 mL) was collected by directing both the outer and inner outlet tubes to a sample collection tube and pumping at 1 mL/min until the device was empty. After sorting, the sample was stained with trypan blue to count viable cells for the 10× single-cell mRNA sequencing procedure. Cells were processed by the Whitehead Institute Genome Technology Core (WIGTC) using 10× Genomics Chromium Controller and Chromium Single Cell 3′ Library & Gel Bead Kit following standard manufacturer’s protocol. The sample was sequenced on a NovaSeqSP (100 × 100 paired-end reads). Sequencing reads were mapped using a GTF file of Smed_v6 (https://planmine.mpinat.mpg.de/planmine/model/bulkdata/dd_Smed_v6.pcf.contigs.fasta.zip) genes in the context of the Smes_g4 (https://planmine.mpinat.mpg.de/planmine/model/bulkdata/dd_Smes_g4.fasta.zip) genome. This GTF file was generated by using BLAT to map all Smed_v6 transcripts to the Smes_g4 genome and each transcript was assigned to a single genome location based on the best alignment score. Transcripts were then collapsed using genome location prior to mapping using the Cell Ranger 7.2.0 pipeline. Cells were assessed for nUMI, nGene, and percent mitochondrial transcript content, which were represented in violin plots. Percent mitochondrial content was based on mitochondrial genes reported in [28] which are represented in v_6 of the Dresden transcriptome (dd_Smed_v6_258_0_1, dd_Smed_v6_289_0_1, dd_Smed_v6_292_0_1, dd_Smed_v6_297_0_1, dd_Smed_v6_344_0_1, dd_Smed_v6_505_0_1, dd_Smed_v6_753_0_1, dd_Smed_v6_957_0_1) and on the highly abundant mitochondrial transcripts (mtRNA_1, mtRNA_2) from [84]. Any cells with nFeature_RNA < 500 or nFeature_RNA > 4,000, or nCount_RNA < 1,000 or nCount_RNA > 10,000, were removed from the dataset prior to analysis. Doublets were identified using scDblFinder (https://bioconductor.org/packages/release/bioc/html/scDblFinder.html) and removed after basic QC filtering. 10× analysis was performed using Seurat 5.1.0 where cells were visualized using the uniform manifold approximation and projection (UMAP) algorithm. The number of dimensions used with RunPCA, RunUMAP, and FindNeighbors was determined using JackStraw with a p-value cutoff of 0.05. Clusters were determined via FindClusters using the Leiden algorithm. UMAP plots of identities were created using Seurat’s DimPlot function. UMAP plots of gene expression were created using Seurat’s FeaturePlot function. Positively enriched genes were identified using Seurat’s FindMarkers function. Human best BLASTx hits were identified using the Human GENCODE Gene Set v39. RNAi For RNAi experiments, double-stranded RNA (dsRNA) was synthesized by in vitro transcription reactions (Promega) using PCR-generated templates with flanking T7 promoters, followed by ethanol precipitation, and annealed after resuspension in water. The concentration of dsRNA varied in each prep between 4 and 7 μg/ml. dsRNA was mixed in a 1:2 ratio with liver and 1−2 μl of this mixture (liver containing dsRNA) per animal was used for feedings. C. elegans unc-22 was used as the control condition. Homeostatic β-catenin-1 RNAi animals were fed three times in 2 weeks and fixed at 21 days (Figs 1A, S1A, and S1C). For S1B Fig, β-catenin-1 RNAi animals were fed twice and fixed after 7 days. Homeostatic bmp4 RNAi animals were fed six times in 3 weeks, then once per week until fixation at 45 days (Figs 1B and S1D–S1G). For epidermal DVB FISH, muscle fiber type FISH, and muscle DVB FISH, β-catenin-1 RNAi animals were fed three times and fixed at 14−21 days (Figs 1E—1G and S1L–S1N). Transplant animals were fed β-catenin-1 dsRNA once 10−14 days after surgery and fixed 10−14 days later (Figs 3A—3E and S3D). smad1 RNAi animals were fed 1-part smad1 dsRNA mixed with 1-part control dsRNA. After 2−3 weeks, when animals had developed dorsal DVB ridges, animals were fed β-catenin-1 dsRNA once and fixed 10 days later (Figs 4A and S3E). For panel 4B, RNAi was administered as described in [85] and animals were fixed after 77d of smad1 RNAi and 21d of β-catenin-1 RNAi. For smad1 RNAi dorsal injury experiments, animals were fed 1-part smad1 dsRNA mixed with 1-part control dsRNA and injured 3 days later. Animals were fed β-catenin-1 dsRNA 10 days following injury and fixed 12 days after that (Fig 4C and 4D). For lateral versus internal injury experiments, β-catenin-1 RNAi animals were fed once, injured 3 days later, and fixed 3 days after injury (Figs 5A, S4A, and S4B). For most Erk inhibition experiments, β-catenin-1 RNAi animals were fed once, injured 3 days later, and fixed 7 days after injury (Figs 5B, S4C, S4D, and S4F). For the muscle DVB FISH of β-catenin-1 RNAi, PD-treated animals, animals were fed with β-catenin-1 dsRNA twice, injured 5 days after the initiation of RNAi, and fixed 5 days after injury (S4G Fig). For the homeostatic Erk inhibition experiment, animals were fed β-catenin-1 dsRNA twice, incubated in PD immediately following the first feeding, and fixed 10 days after the initiation of RNAi (S4E Fig). bmp4; smad1 RNAi animals were fed four times in 2 weeks. Animals were amputated 7 days after the last feeding and fragments were fixed 7 days after amputation (Fig 5C). For timecourse experiments, homeostatic β-catenin-1 animals were fed two (7 days) to three (14 days, 21 days) times and fixed 5−14 days after the last feeding (Figs 6A and S6A). FoxD RNAi animals were fed six times in 3 weeks. β-catenin-1 dsRNA was then fed two times in 1 week and animals were fixed after 21 days (Figs 6B, S6B, and S6C). Transplantation procedures Animals were soaked in 0.2% chloretone for 2 min and placed on moist WhatmanTM filter paper (GE Healthcare, Life Sciences) on a cold block to limit movement. A ~250 µm piece of tissue at the midline posterior to the pharynx was excised from a recipient animal, and the animal was allowed to recover in Holtfreter’s solution for 2 min. For D-V transplants, after a second incubation in chloretone, donor tissue (from a matched AP and ML location) was transplanted into the recipient with reversed DV polarity. For DVB transplants, a ~250 µm piece of tissue from the donor animal lateral edge (at a matched AP location) was transplanted into the midline posterior to the pharynx of a recipient. The recipient animal was then covered with cigarette rolling paper soaked in Holtfreter’s solution and placed at 10°C in the dark. After 24 hours, animals were placed in planarian water and returned to 20°C. Erk signaling inhibition PD0325901 (in short, PD) was dissolved in DMSO, used at 10 μM, and replaced daily. Animals were fed β-catenin-1 dsRNA, incubated in PD immediately following feeding, and amputated 3–5 days later. Animals were fixed for further analysis 5–7 days after parasagittal amputation. Fluorescence in situ hybridizations RNA probes were synthesized in vitro and whole-mount FISH was performed. Briefly, animals were killed in 5% NAC and treated with proteinase K (2 μg/ml). Following overnight hybridizations, samples were washed twice in pre-hybridization buffer, 1:1 pre-hybridization-2 × SSC, 2 × SSC, 0.2 × SSC, PBS with Triton-X (PBST). Subsequently, blocking was performed in a 10% Western Blocking Reagent (Roche, 11921673001) PBST solution for DIG and DNP probes, or in a 5% Western Blocking Reagent and 5% Horse serum PBST solution for FITC probes. Samples were incubated overnight at 4°C in antibody. Antibody washes were then performed for one hour, followed by tyramide development. Peroxidase inactivation with 1% sodium azide was done for 90 min at room temperature. Microscopy and image analysis Brightfield images were taken with a Zeiss Discovery Microscope. Fluorescent images were taken with a Leica SP8 or Leica Stellaris confocal microscope. Fiji/ImageJ and Adobe Photoshop were used to perform brightness and contrast adjustments. All FISH images shown are representative of all images taken in each condition and are maximum intensity projections. PCG domain lengths were quantified in a condition-blind manner using FIJI/ImageJ. A line was drawn at the posterior or anterior boundary of prominent PCG signal and the distance from this line to the head or tail tip was measured and normalized to animal length. To generate line plots of signal intensity, fluorescence intensity was measured along a line of width 100 pixels drawn through the midline from the dorsal to ventral side of the animal using FIJI/ImageJ. Raw data was smoothed using a Gaussian filter. Animal husbandry Asexual Schmidtea mediterranea strain CIW4 animals were cultured in 1× Montjuic water (1.6 mmol/l NaCl, 1.0 mmol/l CaCl2, 1.0 mmol/l MgSO4, 0.1 mmol/l MgCl2, 0.1 mmol/l KCl, and 1.2 mmol/l NaHCO3 prepared in Milli-Q water) at 20°C in the dark. Animals were fed homogenized beef liver and starved at least 7 days prior to experiments. 10× single-cell mRNA sequencing and analysis Tissue fragments from the lateral edges of ~20 large animals were amputated using a scalpel and collected in 1× Montjuic water. Montjuic water was replaced with 0.25% Trypsin-EDTA (1×) and tissue fragments were incubated for 6 min with vigorous pipetting to allow for cell dissociation. The sample was then centrifuged at 500g for 5 min, resuspended in ice-cold CMFB [calcium–magnesium free solution with 0.1% bovine serum albumin (BSA) (400 mg/L NaH2PO4, 800 mg/L NaCl, 1,200 mg/L KCl, 800 mg/L NaHCO3, 240 mg/L glucose, 0.1% BSA, 15 mM HEPES, pH7.3)], and passed through a 40 μm filter. Debris was removed from the dissociated cells with a spiral microfluidic device. The device was a gift from Kyungyong Choi and Jongyoon Han at MIT. It was used as described in [83]. Briefly, the cells were suspended in 20 mL of CMFB and circulated through the device at 16 mL/min in closed-loop separation mode, returning the output of the inner wall outlet to the sample tube, and discarding the output from the outer wall outlet. Once the volume in the sample tube was reduced to approximately 5 mL, the sample was washed at the same pump rate with an additional 40 mL of CMFB, which was added 1 mL at a time to the sample tube so as to keep the sample volume at approximately 5 mL. After washing, the sample was concentrated by further pumping until the input tube was nearly empty. The sample input tube was then disconnected and the sample (approximately 3 mL) was collected by directing both the outer and inner outlet tubes to a sample collection tube and pumping at 1 mL/min until the device was empty. After sorting, the sample was stained with trypan blue to count viable cells for the 10× single-cell mRNA sequencing procedure. Cells were processed by the Whitehead Institute Genome Technology Core (WIGTC) using 10× Genomics Chromium Controller and Chromium Single Cell 3′ Library & Gel Bead Kit following standard manufacturer’s protocol. The sample was sequenced on a NovaSeqSP (100 × 100 paired-end reads). Sequencing reads were mapped using a GTF file of Smed_v6 (https://planmine.mpinat.mpg.de/planmine/model/bulkdata/dd_Smed_v6.pcf.contigs.fasta.zip) genes in the context of the Smes_g4 (https://planmine.mpinat.mpg.de/planmine/model/bulkdata/dd_Smes_g4.fasta.zip) genome. This GTF file was generated by using BLAT to map all Smed_v6 transcripts to the Smes_g4 genome and each transcript was assigned to a single genome location based on the best alignment score. Transcripts were then collapsed using genome location prior to mapping using the Cell Ranger 7.2.0 pipeline. Cells were assessed for nUMI, nGene, and percent mitochondrial transcript content, which were represented in violin plots. Percent mitochondrial content was based on mitochondrial genes reported in [28] which are represented in v_6 of the Dresden transcriptome (dd_Smed_v6_258_0_1, dd_Smed_v6_289_0_1, dd_Smed_v6_292_0_1, dd_Smed_v6_297_0_1, dd_Smed_v6_344_0_1, dd_Smed_v6_505_0_1, dd_Smed_v6_753_0_1, dd_Smed_v6_957_0_1) and on the highly abundant mitochondrial transcripts (mtRNA_1, mtRNA_2) from [84]. Any cells with nFeature_RNA < 500 or nFeature_RNA > 4,000, or nCount_RNA < 1,000 or nCount_RNA > 10,000, were removed from the dataset prior to analysis. Doublets were identified using scDblFinder (https://bioconductor.org/packages/release/bioc/html/scDblFinder.html) and removed after basic QC filtering. 10× analysis was performed using Seurat 5.1.0 where cells were visualized using the uniform manifold approximation and projection (UMAP) algorithm. The number of dimensions used with RunPCA, RunUMAP, and FindNeighbors was determined using JackStraw with a p-value cutoff of 0.05. Clusters were determined via FindClusters using the Leiden algorithm. UMAP plots of identities were created using Seurat’s DimPlot function. UMAP plots of gene expression were created using Seurat’s FeaturePlot function. Positively enriched genes were identified using Seurat’s FindMarkers function. Human best BLASTx hits were identified using the Human GENCODE Gene Set v39. RNAi For RNAi experiments, double-stranded RNA (dsRNA) was synthesized by in vitro transcription reactions (Promega) using PCR-generated templates with flanking T7 promoters, followed by ethanol precipitation, and annealed after resuspension in water. The concentration of dsRNA varied in each prep between 4 and 7 μg/ml. dsRNA was mixed in a 1:2 ratio with liver and 1−2 μl of this mixture (liver containing dsRNA) per animal was used for feedings. C. elegans unc-22 was used as the control condition. Homeostatic β-catenin-1 RNAi animals were fed three times in 2 weeks and fixed at 21 days (Figs 1A, S1A, and S1C). For S1B Fig, β-catenin-1 RNAi animals were fed twice and fixed after 7 days. Homeostatic bmp4 RNAi animals were fed six times in 3 weeks, then once per week until fixation at 45 days (Figs 1B and S1D–S1G). For epidermal DVB FISH, muscle fiber type FISH, and muscle DVB FISH, β-catenin-1 RNAi animals were fed three times and fixed at 14−21 days (Figs 1E—1G and S1L–S1N). Transplant animals were fed β-catenin-1 dsRNA once 10−14 days after surgery and fixed 10−14 days later (Figs 3A—3E and S3D). smad1 RNAi animals were fed 1-part smad1 dsRNA mixed with 1-part control dsRNA. After 2−3 weeks, when animals had developed dorsal DVB ridges, animals were fed β-catenin-1 dsRNA once and fixed 10 days later (Figs 4A and S3E). For panel 4B, RNAi was administered as described in [85] and animals were fixed after 77d of smad1 RNAi and 21d of β-catenin-1 RNAi. For smad1 RNAi dorsal injury experiments, animals were fed 1-part smad1 dsRNA mixed with 1-part control dsRNA and injured 3 days later. Animals were fed β-catenin-1 dsRNA 10 days following injury and fixed 12 days after that (Fig 4C and 4D). For lateral versus internal injury experiments, β-catenin-1 RNAi animals were fed once, injured 3 days later, and fixed 3 days after injury (Figs 5A, S4A, and S4B). For most Erk inhibition experiments, β-catenin-1 RNAi animals were fed once, injured 3 days later, and fixed 7 days after injury (Figs 5B, S4C, S4D, and S4F). For the muscle DVB FISH of β-catenin-1 RNAi, PD-treated animals, animals were fed with β-catenin-1 dsRNA twice, injured 5 days after the initiation of RNAi, and fixed 5 days after injury (S4G Fig). For the homeostatic Erk inhibition experiment, animals were fed β-catenin-1 dsRNA twice, incubated in PD immediately following the first feeding, and fixed 10 days after the initiation of RNAi (S4E Fig). bmp4; smad1 RNAi animals were fed four times in 2 weeks. Animals were amputated 7 days after the last feeding and fragments were fixed 7 days after amputation (Fig 5C). For timecourse experiments, homeostatic β-catenin-1 animals were fed two (7 days) to three (14 days, 21 days) times and fixed 5−14 days after the last feeding (Figs 6A and S6A). FoxD RNAi animals were fed six times in 3 weeks. β-catenin-1 dsRNA was then fed two times in 1 week and animals were fixed after 21 days (Figs 6B, S6B, and S6C). Transplantation procedures Animals were soaked in 0.2% chloretone for 2 min and placed on moist WhatmanTM filter paper (GE Healthcare, Life Sciences) on a cold block to limit movement. A ~250 µm piece of tissue at the midline posterior to the pharynx was excised from a recipient animal, and the animal was allowed to recover in Holtfreter’s solution for 2 min. For D-V transplants, after a second incubation in chloretone, donor tissue (from a matched AP and ML location) was transplanted into the recipient with reversed DV polarity. For DVB transplants, a ~250 µm piece of tissue from the donor animal lateral edge (at a matched AP location) was transplanted into the midline posterior to the pharynx of a recipient. The recipient animal was then covered with cigarette rolling paper soaked in Holtfreter’s solution and placed at 10°C in the dark. After 24 hours, animals were placed in planarian water and returned to 20°C. Erk signaling inhibition PD0325901 (in short, PD) was dissolved in DMSO, used at 10 μM, and replaced daily. Animals were fed β-catenin-1 dsRNA, incubated in PD immediately following feeding, and amputated 3–5 days later. Animals were fixed for further analysis 5–7 days after parasagittal amputation. Fluorescence in situ hybridizations RNA probes were synthesized in vitro and whole-mount FISH was performed. Briefly, animals were killed in 5% NAC and treated with proteinase K (2 μg/ml). Following overnight hybridizations, samples were washed twice in pre-hybridization buffer, 1:1 pre-hybridization-2 × SSC, 2 × SSC, 0.2 × SSC, PBS with Triton-X (PBST). Subsequently, blocking was performed in a 10% Western Blocking Reagent (Roche, 11921673001) PBST solution for DIG and DNP probes, or in a 5% Western Blocking Reagent and 5% Horse serum PBST solution for FITC probes. Samples were incubated overnight at 4°C in antibody. Antibody washes were then performed for one hour, followed by tyramide development. Peroxidase inactivation with 1% sodium azide was done for 90 min at room temperature. Microscopy and image analysis Brightfield images were taken with a Zeiss Discovery Microscope. Fluorescent images were taken with a Leica SP8 or Leica Stellaris confocal microscope. Fiji/ImageJ and Adobe Photoshop were used to perform brightness and contrast adjustments. All FISH images shown are representative of all images taken in each condition and are maximum intensity projections. PCG domain lengths were quantified in a condition-blind manner using FIJI/ImageJ. A line was drawn at the posterior or anterior boundary of prominent PCG signal and the distance from this line to the head or tail tip was measured and normalized to animal length. To generate line plots of signal intensity, fluorescence intensity was measured along a line of width 100 pixels drawn through the midline from the dorsal to ventral side of the animal using FIJI/ImageJ. Raw data was smoothed using a Gaussian filter. Supporting information S1 Fig. Validation of β-catenin-1 and bmp4 RNAi phenotypes. A) Posterior-lateral expression of anterior PCGs (sFRP-1, ndl-4) following β-catenin-1 RNAi (18–21d). Foci containing multiple anterior PCG+ cells were apparent at the DVB (foci typically had >10 cells). Scattered anterior PCG+ cells were also present in the animal. B) Ventrally biased (admp, nlg-7) PCG expression at the DVB after β-catenin-1 RNAi (7d). C) Dorsally biased (bmp4, nlg-8) and ventrally biased (admp, nlg-7) PCG expression in tails after β-catenin-1 RNAi (21d). D) Dorsal expression of the ventral neural marker eye53-1 following bmp4 RNAi (45d). E, F) PCG expression domains in bmp4 RNAi animals (45d). Blue arrow denotes anterior (posterior PCG) or posterior (anterior PCG) boundary of PCG expression domain. G) The posterior expansion of notum signal on the dorsal surface after bmp4 RNAi is because of expansion of the anterior pole (notum+ChAT−). H) Fluorescence intensity of the PCGs bmp4 and admp and the epidermal DVB marker laminB along a line is plotted. Smoothed data is represented with a thick stroke. Individual data points are listed in S1 Data. I) Anterior and posterior PCG expression is concentrated near the DVB during regeneration (96 hpa). Single-channel images of the merged images in Fig 1C are presented. J, K) Fluorescence intensity of anterior and posterior PCGs and the epidermal DVB marker laminB along a line is plotted. Smoothed data is represented with a thick stroke. Individual data points are listed in S1 Data. L) Ectopic anterior PCG (ndl-4) expression near the DVB after β-catenin-1 RNAi (14d). D, dorsal; V, ventral. Fluorescence intensity along a line is plotted and individual data points are listed in S1 Data. M) Ectopic anterior PCG (sFRP-1, ndl-4) expression after β-catenin-1 RNAi (18d) is detected in longitudinal (myoD+), circular (nkx1-1+), and lateral DV (nk4+) muscle at the lateral animal edge. Single-channel images of the merged images in Fig 1F are presented. N) Co-expression of anterior PCGs (sFRP-1, ndl-4) and the secreted molecules lactadherin and netrin-1 in muscle at the DVB after β-catenin-1 RNAi (18d). Colored boxes, area depicted in photos. Scale bars, 100 μm (A, C–G), 50 μm (B, I, L–N), 25 μm (G, inset). https://doi.org/10.1371/journal.pbio.3003482.s001 (PDF) S2 Fig. Identification of novel DVB marker genes. A) Cartoon shows region of the animal isolated for scRNA-seq. Violin plots show the number of UMIs (left) and genes (center) per cell. Individual data points are listed in S1 Data. Table (right) shows the total number of cells after quality control (QC), mean reads, and mean genes detected per cell. B) Violin plot shows the percentage of reads mapped to mitochondrial genes. Individual data points are listed in S1 Data. C) Subclustering of muscle cells, labeled by colF-2, reveals the muscle subtype composition at the DVB. UMAP plots show the overlapping expression of DVB marker genes in longitudinal muscle cells, labeled by myoD. Novel markers of the DVB are labeled in black. Genes with DVB expression verified by FISH are separated by dotted lines. D) UMAP plots show the expression of DVB marker genes in circular muscle cells. E) Heatmap shows the expression of FISH-screened subcluster 1-enriched genes across myoD+ longitudinal muscle subclusters. F) UMAP plots show DVB-enriched gene expression. G) Table of 32 longitudinal muscle subcluster 1-enriched genes categorized by the type of protein each gene encodes. Genes labeled in bold font have DVB expression patterns verified by FISH. H) Table of 11 unnamed DVB-enriched genes listed by dd_ID. The PFAM domains comprising proteins encoded by these genes are listed. I) Protein domain structure of novel DVB markers. J) Whole-body expression of novel DVB marker genes by FISH. K) Plot shows the Pearson correlation coefficients for pairwise comparisons between DVB-enriched genes in longitudinal muscle subcluster 1. Scale bars, 100 μm (J). https://doi.org/10.1371/journal.pbio.3003482.s002 (PDF) S3 Fig. Controls for ectopic DVB experiments. A) D-V transplant animals one day post-transplant (dpt). B) D-V transplant outgrowths show muscle DVB gene expression (lactadherin, frem-1). C) Post-pharyngeal D-V transplant outgrowths express wntP-2, indicative of posterior identity. D) Ventral outgrowths that result from D-V transplant are anteriorized (sFRP-1 ndl-4+) after β-catenin-1 RNAi (10–14d). E) smad1 RNAi animals retain their original DVB (laminB), which is anteriorized (sFRP-1 ndl-4+) after β-catenin-1 RNAi (10d). Yellow arrowhead, anterior PCG focus. Scale bars, 250 μm (A), 100 μm (B–E). https://doi.org/10.1371/journal.pbio.3003482.s003 (PDF) S4 Fig. Controls for DVB injury and removal experiments. A) Lateral wedge and internal hole punch animals one day post-injury (dpi). B) β-catenin-1 RNAi animals after internal hole punch. In rare cases (n = 7/59), a focus of anterior PCG (sFRP-1, ndl-4) expression develops at the wound. β-catenin-1 RNAi, hole punch animals that do not ectopically express anterior PCGs at the wound (n = 51/59) are shown for comparison and the image from Fig 5A is presented. C) β-catenin-1 RNAi animals incubated in PD one dpa. D) PD treatment blocks regeneration of the DVB (laminB) following parasagittal amputation (n = 18/21). In 3/21 animals, a patch of laminB+ DVB was present after parasagittal amputation and PD treatment, indicating incomplete DVB removal or partial DVB regeneration. E) PD treatment does not prevent homeostatic outgrowth associated with ectopic head formation after β-catenin-1 RNAi (10d). F) β-catenin-1 RNAi animals incubated in PD after parasagittal amputation. In some animals (n = 9/22), anterior PCG (sFRP-1, ndl-4) expression is detected at the wound, though these animals lack PCG foci formation and apparent homeostatic outgrowth. β-catenin-1 RNAi, PD-treated animals that lack anterior PCG expression at the wound (n = 12/22) are shown for comparison and the image from Fig 5B is presented. One animal had a patch of laminB+ DVB present at the wound (n = 1/22), indicating incomplete DVB removal or partial DVB regeneration. G) PD treatment after DVB removal blocks regeneration of muscle DVB (lactadherin, frem-1) cells. Despite no DVB regeneration, low levels of anterior PCG (sFRP-1, ndl-4) expression can be detected at the wound (n = 9/10). Colored boxes, area depicted in photos. Scale bars, 250 μm (A, C, E), 100 μm (B, D, F, G), 50 μm (F, inset), 25 μm (G, inset). https://doi.org/10.1371/journal.pbio.3003482.s004 (PDF) S5 Fig. Muscle DVB gene expression regeneration. A) Regeneration of anterior PCG (sFRP-1, ndl-4) and muscle DVB (lactadherin, frem-1, dd_30210, egf-6) gene expression domains 0–72 hpa. Patterned muscle DVB gene expression is detected in the blastema by 36 hpa, along with a focus of sFRP-1 ndl-4 expression. Colored box, area depicted in photos. Scale bars, 50 μm (A). https://doi.org/10.1371/journal.pbio.3003482.s005 (PDF) S6 Fig. FoxD RNAi blocks anterior pole formation. A) Detection of anterior pole transcripts (FoxD, notum) after β-catenin-1 RNAi (7–21d). B) FoxD expression in the anterior pole is significantly reduced (n = 6/12) or absent (n = 6/12) after FoxD; control RNAi and absent (n = 12/12) after FoxD; β-catenin-1 RNAi. C) Anterior PCG (sFRP-1, ndl-4, ndl-5) gradients are not grossly affected by FoxD RNAi. Colored boxes, area depicted in photos. Scale bars, 50 μm (A, B), 100 μm (C). https://doi.org/10.1371/journal.pbio.3003482.s006 (PDF) S1 Table. Subcluster-enriched genes identified by DVB scRNA-seq data. https://doi.org/10.1371/journal.pbio.3003482.s007 (XLSX) S2 Table. Contig annotation of genes used in figures. https://doi.org/10.1371/journal.pbio.3003482.s008 (XLSX) S1 Data. Data used to generate plots. https://doi.org/10.1371/journal.pbio.3003482.s009 (XLSX) Acknowledgments The authors thank members of the Reddien lab for helpful comments and discussion. We thank Conor McMann for measuring PCG gradient lengths in bmp4 RNAi animals and Patrick Aoude for 10× mapping and computational support.

Li, Huamei;Liu, Jingchao;Wang, Guige;Liu, Zhenyu;Cao, Meng;Sun, Lingyun;Peng, Cheng;Liu, Yiyao;Ma, Liang;Xiong, Qing

doi: 10.1371/journal.pbio.3003477pmid: 41248160

Introduction Single-cell (SC) sequencing has evolved into an indispensable tool for conducting comprehensive investigations into gene expression and functional attributes within individual cells, thereby significantly advancing our understanding of the intrinsic heterogeneity among cell types and their pivotal roles in developmental and disease processes [1,2]. However, the process of SC sequencing inherently sacrifices spatial information during sample processing, a critical component essential for gaining insights into the dynamic cellular microenvironment [3]. Spatial transcriptomics (ST) technologies, like 10× Visium, preserve cell locations within tissue sections [4,5]. Nonetheless, ST technologies often grapple with challenges related to spatial resolution and the provision of comprehensive SC gene expression data. The integration of ST with SC data represents a highly effective approach for expediting the analysis of cell distribution within tissues, revealing intercellular relationships, and gaining a profound understanding of the structural and functional intricacies of tissues [6–9]. Regrettably, this integration encounters complexities stemming from the high-dimensional nature of the data. To address this challenge, the widespread adoption of spatial deconvolution strategies has emerged. These strategies leverage SC reference data to deconvolute ST data and estimate the cellular composition within each spatial spot, as demonstrated by methods like Cell2location [10], CARD [11], RCTD [12], SpatialDWLS [13], STRIDE [14], DestVI [15], and Stereoscope [16]. However, these approaches are still constrained by resolution and typically provide information on the relative proportions of cell types, lacking detailed insights at the SC level, which hinders the in-depth exploration of cell states, interaction patterns, and neighboring cell populations. Establishing a channel that maps individual cells to specific spatial spots, enabling segmentation at SC granularity and potentially inferring unmeasured gene expressions, offers a broader perspective for the integrated analysis of ST and SC data [17]. Several methods have been publicly disclosed. One common approach, like Seurat [18], employs a label transfer method that maps SC labels onto spots. This method is based on Seurat’s anchor-based data integration framework, which identifies shared features between the reference SC dataset and the query ST dataset, thereby enabling the transfer of cell type labels from SC to ST spots. Tangram [19] employs a deep learning framework to reveal intricate spatial structures at the SC level, aiding in the detection of complex cell distributions and relationships. On the other hand, CellTrek [3] utilizes a random forest model trained on ST data to predict spatial coordinates, benefiting from shared dimension reduction features with SC data, enhancing its efficiency with large-scale ST datasets. CytoSPACE [20] employs a correlation-driven cost function based on convex optimization to allocate individual cells to precise spatial spots, ensuring effective and reliable cell positioning. Despite these advances, current mapping tools face notable limitations. Label transfer-based strategies, such as Seurat’s, struggle under certain conditions. For high-resolution (SC or subcellular resolution; e.g., image-based platforms) ST data, they can annotate cell types per spot while preserving the original gene expression profiles, which benefits mechanistic analyses. However, in platforms such as Xenium, the low transcript capture rate leads to extremely sparse expression matrices, limiting transcriptome coverage and reducing analytical depth. For low-resolution (multiple cells per spot; e.g., spatial barcoding-based platforms) ST data, these tools typically assume each spot originates from a single cell type, overlooking potential cell mixtures within spots and thereby introducing bias. Beyond label transfer, SC mapping strategies also encounter two major challenges. First, aligning unpaired SC and ST datasets is difficult, particularly in minimizing mismatches and missing mappings. Second, in low-resolution ST data, spots frequently contain multiple cells, complicating estimates of cell counts and the inference of spatial relationships within heterogeneous neighborhoods. Different tools address these challenges with varying trade-offs. CellTrek [3] performs well in mapping unpaired SC and ST datasets, assigning cell types to corresponding ST regions, but it has limited ability to resolve cellular composition or fine-grained spatial relationships within spots. Conversely, CytoSPACE [20], by integrating the RCTD framework [12], provides more accurate estimates of cellular composition and spatial distribution, yet it assigns cells at the tissue-wide level and does not resolve the issue of unpaired SC and ST mapping. Together, these limitations underscore the need for new methods that can simultaneously handle unpaired SC–ST data integration and accurately infer spatial relationships within heterogeneous spots—an essential step toward advancing integrative SC and ST research. Herein, we present Cell2Spatial, a unified framework for reconstructing tissue architecture at SC resolution. Cell2Spatial integrates a spatially weighted maximum likelihood model with spatial hotspot detection to robustly quantify single-cell spot similarity and reduce mismatches between SC and ST datasets. For low-resolution platforms, it employs a corrected saturation model to estimate spot-level cell counts by modeling the relationship between library size and gene counts. To enable efficient large-scale mapping, a feedforward neural network (FNN) is combined with a linear sum assignment algorithm, ensuring accurate and scalable cell-to-spot allocation. Compared with existing approaches, Cell2Spatial offers a practical improvement in reconstructing tissue structure and cellular composition, seamlessly integrates unpaired SC–ST data, and reliably captures the spatiotemporal dynamics of cell states. Results Overview of Cell2Spatial The Cell2Spatial framework aims to map single cells to spatial spots for reconstructing tissue architecture (Fig 1). The process begins with normalizing SC and ST data, ideally from the same tissue source, using Seurat’s SCTransform to establish a common basis for gene expression comparison. To quantify similarities between single cells and spatial spots, maximum likelihood model is employed. Efficiency is enhanced by identifying cell-type-specific genes through a modified entropy-based method (Fig 1, Step 1; see “Materials and methods”). These gene sets, along with their expression profiles, form probability matrices that construct the maximum likelihood model and generate a likelihood matrix. Spatial context is incorporated through Seurat’s canonical correlation analysis (CCA) [18] for SC–ST integration, followed by principal component analysis (PCA) for dimensional reduction. A Euclidean distance matrix is derived from the Uniform Manifold Approximation and Projection (UMAP) embedding, constructed on the PCA space. This distance matrix is applied as weights (scaled to 1) to the likelihood matrix, resulting in a weighted likelihood matrix that balances gene expression similarity with spatial proximity (Fig 1, Step 2; see “Materials and methods”). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Overview of the Cell2Spatial algorithm. The Cell2Spatial framework comprises seven fundamental components: data preprocessing, marker gene selection, similarity matrix construction, hotspot detection, deconvolution and cell count estimation, single-cell (SC) pre-assignment, and final mapping to spots (see “Materials and methods”). SC and spatial transcriptomics (ST) data were standardized using SCTransform in Seurat, with cell-type-specific genes identified through a modified entropy-based method (Step 1). A spatially weighted similarity matrix was created using a maximum likelihood model, with Seurat’s canonical correlation analysis (CCA) connecting single cells to spatial spots to yield spatial weights (Step 2). Cell type hotspots were identified using the Getis-Ord G* index combined with normality testing, while robust linear regression was applied for deconvolution of cellular composition, and corrected saturation model was used to estimate cell counts in low-resolution ST data (Steps 3-5). Feedforward neural network (FNN) was employed for cell pre-assignment to ST clusters (Step 6), and the Jonker-Volgenant algorithm iteratively mapped individual cells to specific spatial spots based on similarities and estimated cell counts (Step 7). https://doi.org/10.1371/journal.pbio.3003477.g001 To improve robustness, particularly when SC and ST data may not perfectly align in terms of cell types, the weighted likelihood matrix undergoes refinement. Spots that do not correspond to any pre-defined cell types in the SC dataset are filtered out determined by the Getis-Ord G* index coupled with a normal test based on signature scores from cell-type-specific genes, ensuring focus on relevant spatial regions (hotspot regions) while minimizing mapping noise (Fig 1, Step 3; see “Materials and methods”). Understanding the cellular compositions within ST sections is critical for recovering tissue spatial organization. This is achieved through a deconvolution model that applies robust linear regression model (RLM) in conjunction with base matrix constructed from specific gene sets (Fig 1, Step 4; see “Materials and methods”) [21,22]. The corrected saturation model, was designed to estimate cell counts per spot in low-resolution ST data by linking the library size and gene count, through a saturation-based relationship (Fig 1, Step 5; see “Materials and methods”). For high-resolution ST data, where spots typically capture single cells, it assigns a cell count of 1 per spot. Therefore, the total number of cells to be aligned is calculated as the sum of cell counts across the remaining spots, with the number of cells for each type sampled from the SC dataset based on these totals and the estimated cellular proportions, yielding candidate mapping SC data. To assign candidate single cells to spatial spots, the Jonker-Volgenant algorithm [23] is employed, utilizing the refined weighted likelihood matrix and the estimated cell counts per spot. In large-scale applications, however, this algorithm imposes substantial memory and processing demands. To mitigate these challenges, a FNN framework is implemented. The FNN is trained to pre-allocate cells to ST clusters based on the co-embedding of SC and ST data (Fig 1, Step 7; see “Materials and methods”). This pre-allocation streamlines the execution of the Jonker-Volgenant algorithm, allowing it to be applied iteratively within each ST cluster, ensuring efficient and scalable SC-to-spot assignments (Fig 1, Step 7; see “Materials and methods”). Evaluation of the performance of Cell2Spatial using mouse brain transcriptome data To evaluate the performance of Cell2Spatial, ST data from the 10× Visium platform for the mouse brain was obtained, comprising 2,696 spots. Among these, around 66.53% of genes exhibited no expression, with the highest UMI count recorded at 25,580 and a median count of 24,332 UMIs (S1A Fig and S1 Table). Additionally, we utilized a reference dataset (SC) from the Allen Institute, which comprises approximately 14,000 adult mouse cortical cells, covering 23 distinct cell types (S1B Fig). As the Fig 2A shown, the application of Cell2Spatial enabled us to map the single cells to precise spatial spots, and the findings unveiled the presence of distinct anatomical distributions of various cell types within the mouse brain (Fig 2A and 2B). Notably, our approach capitalizes on a modified specificity entropy strategy for the identification of cell-type-specific genes, showcasing noteworthy discriminatory power and demonstrating its effectiveness (S1C Fig). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Performance assessment of Cell2Spatial on real and simulated mouse brain spatial transcriptomics (ST) data. (A) Spatial architecture of mouse brain reconstructed using Cell2Spatial. Cell types are marked by color codes, and each dot represents an individual cell. (B) Distribution of cell types Astro, L2/3 IT, L4, L5 IT, and L6 CT on mouse brain ST slice. (C) Spatial architecture of the mouse brain reconstructed using CytoSPACE, CellTrek, Tangram, and Seurat, respectively. Cell types are color-coded. Each dot represents an individual cell. (D) Spatial k-distance (k = 10) of L2/3 IT, L4, L5 IT, L5 PT, NP, L6 IT, L6 CT, L6b, Oligo to Astro cells. Violin boxplots illustrate the median, and interquartile ranges (25%–75%), with whiskers extending up to 1.5 times the interquartile range beyond the box. (E) Bar plot showing the conservation of highly variable genes (HVGs) between reconstructed spatial expression data from mapping tools and the original spatial expression data (see “Materials and methods”). (F) Spatial heat maps showing the performance of Cell2Spatial for aligning scRNA-seq data (with 5% added noise) to spatial spots in ST datasets simulated with five cells on average (see “Materials and methods”). Only cell types with prominent spatial structures have been presented for the sake of clarity. The color intensity of each spot corresponds to the number of mapped single cells. (G) Box plots showing the performance across various methods, and noise levels (0, 5%, 10%, 15%, and 20%) in accurately assigning individual cells to their respective locations in simulated ST datasets. Each data point represents a distinct cell type. The central lines within the boxes, the box edges, and the whiskers correspond to the medians, the first and third quartiles, and the minimum and maximum values, respectively, within 1.5 times the interquartile range of the box boundaries. P-values were obtained by Kruskal–Wallis test, as it is appropriate for comparing more than three groups of non-normally distributed data without assuming equal variances. (H) Boxplot showing the end-to-end Jaccard index at the single-cell level, measuring the consistency between the coordinates of single cells in the reconstructed data and their coordinates in synthetic ST data. Perturbations were introduced by randomly modifying gene expression in 0%, 5%, 10%, 15%, or 20% of genes. For each perturbation level, three independent replicates were generated to ensure robustness of the evaluation. The data underlying this figure can be found at https://doi.org/10.5281/zenodo.17212677. https://doi.org/10.1371/journal.pbio.3003477.g002 Next, the performance of Cell2Spatial with other mapping tools capable of segmenting spatial spots into SC granularity was compared, including CytoSPACE [20], CellTrek [3], Tangram [19], and Seurat [18]. The findings indicated that, the performance of the other four mapping tools within the ST sections was significantly weaker than that achieved by Cell2Spatial (Figs 2C and S1D–S1G). Importantly, Tangram’s mapping did not reveal a distinct alignment between cell types and anatomical structures, and Seurat directly assigned spots as individual cell types. Furthermore, as shown in Fig 2D, Cell2Spatial generated consistent spatial relationships between specific cell types (L2/3 IT, L4, L5 IT, L5 PT, NP, L6 IT, L6 CT, L6b, Oligo, to Astro) based on k-distances, which were in agreement with previous studies [3,19] (Fig 2D and S2 Table). CellTrek was also efficient, whereas Tangram exhibited less favorable performance. Additionally, we reconstructed spot-level expression profiles using the mapped SC resolution ST dataset and assessed highly variable genes (HVGs) conservation, with Cell2Spatial demonstrating the highest conservation compared to unprocessed ST data (Fig 2E; Cell2Spatial: 0.357, CytoSPACE: 0.166, CellTrek: 0.221, Tangram: 0.145; see “Materials and methods”). This indicates its effectiveness in preserving the intrinsic gene characteristics of spatial spots within tissue sections. Notably, since Seurat’s label transfer function was used, the spot-level expression in the ST data remained unchanged post-transfer and was excluded from this analysis. Nevertheless, the limited information regarding cell composition and location within the real ST data of mouse brain mitigate our evaluation of Cell2Spatial’s performance. To overcome this challenge, simulated ST datasets of the mouse brain were generated with accurately known cell type localization and composition at various noise levels (see “Materials and methods”). Applied Cell2Spatial, CytoSPACE, CellTrek, Tangram, and Seurat to map the SC data of the mouse brain to the simulated ST datasets, our results demonstrated that Cell2Spatial adeptly restored the spatial positions of cell types, and exhibited a strong alignment with actual cell counts of spots, as reflected by the root mean square error (RMSE) between predicted and expected counts (Cell2Spatial: 2.21; CytoSPACE: 3.29; CellTrek: 4.10; Tangram: 8.52; Seurat: 6.38) (Figs 2F and S2A–2C). We simulated spots with varying cell counts to further evaluate Cell2Spatial’s ability to estimate cell numbers per spot (S2D Fig). Across different values of Lambda (Lambda = 1:20; representing the Poisson distribution parameter used to generate random cell counts), Cell2Spatial consistently achieved Pearson correlation coefficients (PCCs) above 0.89 (median: 0.941; maximum: 0.952), whereas CytoSPACE achieved coefficients above 0.61 (median: 0.693; maximum: 0.783), demonstrating a significant performance difference (Two-sided Wilcoxon test; p-value < 0.001) (S2E–S2G Fig and S3 Table). Consistently, RMSEs were also significantly lower for Cell2Spatial (median: 1.15) compared to CytoSPACE (median: 2.76) (Two-sided Wilcoxon test; p-value < 0.001) (S2G Fig). To assess performance more effectively, we simulated ST datasets with known cell type composition and spatial distributions under different perturbation levels (see “Materials and methods”). Based on accuracy indices, Cell2Spatial consistently outperformed other tools at the cell type level across all noise conditions (Fig 2G and S4 Table). Additionally, we investigated a crucial aspect of the analysis: mapping SC datasets designed for simulating ST data to their corresponding synthetic ST data to assess the accuracy of cell localization—i.e., end-to-end mapping. To minimize the influence of stochastic processes on performance evaluation, we performed three replicates under each perturbation level. The results indicated that CellTrek achieved the highest Jaccard index (mean index: 0.193), with Cell2Spatial performing next (mean index: 0.167), outperforming both CytoSPACE (mean index: 0.106) and Tangram (mean index: 0.160) in overall accuracy (Fig 2H). The relatively strong accuracy of CellTrek may stem from its Random Forests framework, which can better capture complex, non-linear relationships in ST data and provide greater robustness to noise compared with the linear assignment algorithm used by Cell2Spatial. Notably, all tools showed low Jaccard indices (<0.20), reflecting a potential limitation: current methods mainly distinguish between cell types but lack power to resolve cells within the same type, making them better at reconstructing cell-type distributions than pinpointing individual cell locations. Taken together, Cell2Spatial can accurately infer the composition of cell types within spatial structures, and effectively reconstruct spatial architectures for tissues. Assessing Cell2Spatial’s effectiveness across diverse metrics and conditions ST datasets with varying spot resolutions (Lambda = 5, 10, 15, 20) and cellular compositions were simulated, where Lambda served as the Poisson distribution parameter for randomly generating cell numbers per synthetic spot. We initially focused on Cell2Spatial’s ability to predict cellular proportions within these simulated ST sections. To facilitate comparison, we incorporated 10 additionally spatial deconvolution tools: Cell2location [10], SpatialDWLS [13], RCTD [12], Stereoscope [16], DestVI [15], SpaOTsc [24], novoSpaRc [25], SPOTlight [26], CARD [11], and DSTG [27], using SC data from the mouse brain as the reference to estimate cellular proportions in each spot from the synthetic ST datasets. For mapping tools, including Cell2Spatial, CytoSPACE, CellTrek, and Tangram, we aligned single cells from mouse brain to the synthetic ST datasets, and then calculated the cellular proportions. In the case of Seurat’s label-transfer method, we utilized the probability matrix obtained through the “TransferData” function to represent cellular proportions in each spot. As shown in Fig 3A, across various spot resolutions (Lambda = 5, 10, 15, and 20), Cell2Spatial (PCCs: 0.969–0.976; RMSEs: 0.0135–0.0148) outperformed similar tools such as CytoSPACE (PCCs: 0.910–0.974; RMSEs: 0.0134–0.0271), CellTrek (PCCs: 0.777–0.833; RMSEs: 0.0332–0.0402), and Tangram (PCCs: 0.246–0.354; RMSEs: 0.0541–0.0615) in predicting cellular proportions (Fig 3A; see “Materials and methods”). Additionally, when compared to the spatial deconvolution tool CARD, Cell2Spatial demonstrated comparable efficacy in predicting cellular compositions, achieving PCCs greater than 0.97, and RMSEs below 0.015 across all resolutions (Fig 3A; bottom panel). We further benchmarked Cell2Spatial using 32 published SC and simulated ST datasets from Li and colleagues [28], which cover human brain, liver, lung, kidney, and pancreas, as well as mouse pancreas and trachea. These datasets contain a median of eight cell types (range: 6–128), with eight datasets including more than 10 cell types. Each simulated dataset comprises 1,000 spots with known cellular compositions. On these benchmarks, Cell2Spatial demonstrated strong performance in reconstructing spot-level cell type compositions, achieving an average PCC of 0.784 (range: 0.427–0.964; highest in dataset30, lowest in dataset1), ranking second among all tools, just behind Cell2location (range: 0.619–0.942) (Fig 3B). For RMSE, Cell2Spatial achieved an average of 0.114 (range: 0.027–0.220), ranking fifth overall. While Cell2location (range: 0.031–0.145) showed the best overall accuracy, Cell2Spatial maintained competitive performance in estimating cell type compositions (Fig 3B). These benchmarking results on both simulated and published datasets confirm that Cell2Spatial achieves high accuracy and robustness across diverse conditions. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Evaluation of spatial transcriptomics (ST) tools across various metrics and conditions. (A) Bar plots showing Pearson correlation coefficients (PCCs) and Root Mean Square Errors (RMSEs) between predicted and true proportions in synthetic ST datasets under different spot resolution levels (Lambda = 5, 10, 15, and 20). The Lambda value represents the expected cell counts in spots, modeled by a Poisson distribution. Error bars indicate variability, determined through a permutation strategy repeated 100 times, where 2,000 spots were down-sampled, and PCC and RMSE were calculated for each iteration. (B) Violin combined with box plots showing the PCCs and RMSEs between estimated and actual cellular proportions. Black lines within the boxes indicate medians. Thirty-two simulated ST datasets, together with their corresponding single-cell (SC) datasets, were retrieved from the study by Li and colleagues [28]. (C) Kullback–Leibler (KL) divergence metrics for spatial cell charting methods, comparing each cell type against synthetic ST dataset references. (D) Grouped bar plots showing cosine similarities between recovered spot expressions post-SC mapping and the original synthetic ST dataset expressions. Error bars represent values obtained via permutation strategy, with 2,000 spots down-sampled and cosine similarities calculated across 100 iterations. (E) Grouped bar plots showing the conservation of highly variable genes (HVGs) between recovered spot expressions and synthetic ST dataset expressions following SC mapping. (F) Bar plot showing the average Jaccard index for cell type consistency in spots, comparing predictions from our score-guided mapping accuracy (SGMA; see “Materials and methods”) strategy against true cell types in synthetic ST datasets. (G) Box plot displaying SGMA indexes for cell type consistency in spots, comparing predictions using mapping tools (Cell2Spatial, CytoSPACE, CellTrek, and Tangram) with true cell types in synthetic ST datasets, based on the SGMA strategy (see “Materials and methods”). (H) Scatter plot mapping a subset of mouse brain SC data to ST spots under unmatched conditions (where cell types from SC data represent only a subset of those in ST datasets). Each dot corresponds to an individual cell, with cell types color-coded. (I) Violin combined with box plots showing the PCC and RMSE between estimated and actual cell counts in spots. Simulated ST datasets, together with their corresponding SC datasets, were retrieved from the study by Li and colleagues [28]. (J) Spatial feature plots illustrating inferred cell counts in spots using the DAPI channel in fluorescence images (counted via Squidpy [29]), Cell2Spatial, and CytoSPACE. The top panel represents a coronal section of the mouse brain, while the bottom panel represents the whole brain. Color shading indicates different cell counts in spots. (K, L) Density plots combined with fitted lines, showing the consistency between cell counts estimated by Cell2Spatial or CytoSPACE and those counted using DAPI via the Squidpy tool [29]. “R” represents the Spearman correlation, with p-values obtained from the two-sided t-tests. (M) Scatter plot with line showing the influence of marker size on Cell2Spatial performance for different cell types. The marker size range is 10 to 200, with a step size of 10. (N, O) Scatter plots with line showing peak memory usage and elapsed time for different mapping tools under varying spot sizes, ranging from 1,000 to 10,000, with a step size of 1,000. The data underlying this figure can be found at https://doi.org/10.5281/zenodo.17212677. https://doi.org/10.1371/journal.pbio.3003477.g003 Next, to assess the spatial density of cell charting results against the actuary spatial distribution for various cell types in the synthetic ST datasets (Fig 3A), Kullback–Leibler (KL)-divergence was used. The analysis showed that Cell2Spatial had consistently lower KL-divergence values than CytoSPACE, CellTrek, and Tangram, indicating that Cell2Spatial more accurately recovered spatial cellular structures (Fig 3C; see “Materials and methods”). We also evaluated gene expression recovery by comparing the cosine similarity between the recovered spot expressions and the original synthetic datasets. Cell2Spatial consistently achieved the highest similarity (approximately 0.8) across all datasets, highlighting its effectiveness in accurately reconstructing gene expression patterns (Fig 3D; Cell2Spatial: 0.787–0.798; CytoSPACE: 0.779–0.789; CellTrek: 0.753–0.778; Tangram: 0.712–0.736). Furthermore, we assessed HVGs conservation, which serves as a proxy for the preservation of the biological signal in the ST expression profile before and after mapping, and found that Cell2Spatial performed well in retaining HVGs across all spot resolutions, demonstrating its ability to preserve key gene expression features effectively (Fig 3E; Average HVGs conserved scores: Cell2Spatial (0.556); CytoSPACE (0.411); CellTrek (0.361); Tangram (0.310)). Moreover, we proposed a metric, the Score-Guided Mapping Accuracy (SGMA) index. This method scores spots using marker genes for specific cell types, capturing the prominent spatial regions of these cell types and comparing the consistency of these regions in the ST spatial data before and after mapping (see “Materials and methods”). Using marker genes to score synthetic ST spots, we identified high-confidence spots for each cell type. The results, as shown in Fig 3F, demonstrated that the cell type distributions inferred by SGMA strategy were highly consistent with the true distributions, with Jaccard indexes exceeding 0.75 (Fig 3F; range: 0.764–0.853), demonstrating SGMA’s effectiveness in evaluating mapping performance. Applying SGMA index to various mapping tools across multiple synthetic ST datasets, we found that Cell2Spatial achieved the highest median Jaccard index for recovering ST cell type distributions, reinforcing its strong performance in cell type identification (Fig 3G; Cell2Spatial: 0.503–0.560; CytoSPACE: 0.289–0.497; CellTrek: 0.304–0.428; Tangram: 0.248–0.315). Notably, SGMA index is not only valuable for synthetic ST datasets but also serves as an effective benchmark for evaluating mapping tools on real experimental ST data. In particularly, to investigate the performance of mapping tools under mismatched conditions, where the cell types in the SC data represent only a subset of those in the ST dataset, we focused on three cell types—Astro, L2/3 IT, and L4—from the mouse brain SC dataset. We used four mapping tools to project single cells from these cell types to ST spots. The results revealed that Cell2Spatial effectively reconstructed the spatial distribution of these three cell types (Fig 3H). In contrast, CytoSPACE, which employs a full spatial slice mapping approach, struggled to accurately reconstruct the tissue’s spatial structure under these mismatched conditions. CellTrek partially captured the spatial distribution of the three cell types but was less precise for Astro cells, while Tangram did not clearly reveal the spatial distribution of these cell types. These findings highlight Cell2Spatial’s capability to handle complex and mismatched datasets effectively. To validate the accuracy of spot-level cell count estimation by Cell2Spatial, we applied both Cell2Spatial and CytoSPACE to 32 simulated ST datasets generated by Li and colleagues [28], each containing spots with known cell numbers. Cell2Spatial achieved higher correlations with true counts (PCC range: 0.507–0.903; mean = 0.725) compared to CytoSPACE (PCC range: 0.480–0.898; mean = 0.700) (Fig 3I). RMSE values further confirmed its advantage, with Cell2Spatial showing significantly lower errors (range: 1.643–3.846; mean = 2.390) relative to CytoSPACE (range: 1.935–3.834; mean = 2.683) (Fig 3I). We next applied Cell2Spatial to two mouse brain ST datasets (10× Visium) with available DAPI channel images. Cell counts were quantified from DAPI staining using Squidpy [29] and compared with estimates from Cell2Spatial and CytoSPACE. Cell2Spatial’s estimates more closely matched the DAPI-based counts, achieving correlation coefficients of 0.62 and 0.49, outperforming CytoSPACE (coefficients: 0.52 and 0.35) (Fig 3J–3L; see “Materials and methods”). To assess the effect of predefined maximum cell numbers (nmax), we varied nmax between 4 and 24 in a synthetic mouse brain ST dataset (true nmax = 14). As nmax deviated from the true value, RMSEs increased (S2H Fig; left panel). Nonetheless, PCCs consistently exceeded 0.8, indicating that despite differences in nmax, the overall trend of spot-level cell distribution was preserved (S2H Fig; right panel). Considering the central role of cell type-specific genes in Cell2Spatial, we investigated how varying the number of selected specific genes per cell type affected Cell2Spatial’s performance. The results revealed that while increasing the number of specific genes enhanced the conservation of HVGs in the reconstructed spatial expression profiles, Cell2Spatial’s performance remained relatively stable within a certain range (HVGs conservation: 0.47 ± 0.15) (Fig 3M). This stability suggests that Cell2Spatial is resilient to changes in gene selection. Finally, we evaluated the computational efficiency of the mapping tools. Cell2Spatial demonstrated relatively high memory usage, though it stayed within acceptable limits, and consistently achieved relatively short runtimes even as the spot size increased (Fig 3N and 3O; see “Materials and methods”). In contrast, CytoSPACE and Tangram showed a linear increase in computational time, highlighting Cell2Spatial’s good efficiency and practical advantages. Overall, our evaluation shows that Cell2Spatial performs well across various metrics and conditions. Its effectiveness in estimating cellular compositions, recovering gene expression, and maintaining computational efficiency, along with its adaptability to different input parameters, makes Cell2Spatial a valuable tool for ST analysis. Spatial profiling of single-cells in various tissue types with Cell2Spatial To assess the ability of Cell2Spatial in analyzing ST data from diverse tissue sources, ST data [30] along with a reference scRNA-Seq dataset of human thymus tissue [31] were obtained. The thymus ST dataset consisted of 1,628 spots, with a maximum of 38,161 unique molecular identifiers (UMIs) and a median of 91,66 UMIs, while the SC dataset encompassed 40 distinct cell types. Through the process of assigning cells to ST spots, Cell2Spatial successfully mapped 39 cell types to the thymus section, revealing notable differences in cell type distribution between the medulla and cortical regions (Figs 4A and S3A–S3D and S1 Table). As known, the human thymus is the site of T-cell development, with double-negative (DN) and double-positive (DP) T cells residing in the cortical region, while single-positive (SP) cells are mainly localized to the medullary region (Fig 4B). We tested the ability of various mapping tools to capture the thymic T cell development process, with focus on the DN, DP, ABT.entry (transition from DP to SP states), CD8+ T, CD4+ T cells, and cortical thymic epithelial cells (cTECs) and medullary thymic epithelial cells (mTECs) that are involved in T cell maturation. Cell2Spatial revealed distinct spatial patterns, with DN and DP cells predominantly inhabiting the thymic cortex, and ABT.entry cells strategically positioned in the cortical region near the corticomedullary junction (Fig 4C). On the other hand, CD8+ T and CD4+ T cells, representing mature SP cells, were predominantly found in the thymic medulla (Fig 4C). Furthermore, cTECs were primarily concentrated in the cortical region, while mTECs were detected in the medullary region, consistent with previous researches [31–33]. Interestingly, the presence of a small subset of cTECs in the medullary region hints at the existence of certain cTEC subtypes exhibiting characteristics of mTECs [31,34]. These suggest that the thymus architecture restored by Cell2Spatial aligned with the developmental pathway of T-cells. While CytoSPACE, CellTrek, and Tangram also partially captured the relationship between critical T-cell subpopulations and thymus structures, their performance slightly lags behind Cell2Spatial (Fig 4C). Notably, Seurat’s performance was sub-optimal due to its interpretation of spots at the SC level. These findings highlight the effectiveness and reliable of Cell2Spatial in deciphering the essential spatial structures of anatomical tissues. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. Performance assessment of Cell2Spatial in real ST data from various tissue types. (A) Spatial architecture of the human thymus (Suo and colleagues) [30] reconstructed using Cell2Spatial. Cell types are color-coded. Each dot represents an individual cell. (B) Thymus anatomy and the T cell development process. (Left) Hematoxylin and eosin (H&E) staining images depicting the thymus structure. The white line marks the medullary-cortical junction, with colored regions indicating the medullary areas and uncolored regions representing the cortical areas. (Right) T cell development stages, encompassing the double-negative (DN), double-positive (DP), and single-positive (SP) stages. DN and DP stages were primarily located in the cortical regions, while the SP stage is concentrated in the medullary regions. (C) Distribution of selected cell types on human thymus ST slices. From left to right, cell types include DN, DP, ABT.entry, CD8+ T cells, CD4+ T cells, cTECs, and mTECs. (D) Scatter plots of various mapping tools showing the concordance between the relative Euclidean distance ranking of T cell subtypes at different stages and their developmental sequence determined by pseudo-time analysis. Euclidean distances of individual cells were calculated relative to a reference point at the center of the medullary region (regions 3, 7, and 8 in S3A Fig, left panel). For each subtype, the mean distance from the reference point was used to establish a spatial ranking, which was then compared with the developmental order. The strength of this association was quantified using Spearman correlation analysis. The blue line represents a linear fit, and the shaded area denotes 95% confidence interval. Different colored dots represent distinct T cell subgroups. “R” stands for the Pearson correlation coefficient, and the p-values were obtained from two-sided t test, which follows a t-distribution to account for smaller sample sizes. (E) Gene set enrichment analysis (GSEA) performed on single cells mapped to both the medullary and cortical regions using Cell2Spatial. (F) Left: Epithelial cell transcriptomes from a mouse kidney single-cell atlas were mapped onto spatial spots of a normal mouse kidney using Cell2Spatial, displayed with jitter within their assigned spots. Right: The same representation, with cells colored based on their known distance to the inner medulla. (G) Concordance between predicted and known distances of each epithelial state to the base of the inner medulla (illustrated in the left-top panel). “R” stands for the Spearman correlation coefficient, and the p-values were obtained from two-sided t-tests. (H) (i) Anatomical structures within the dorsolateral prefrontal cortex (DLPFC), (ii) Spatial architecture determined using Cell2Spatial, (iii) CytoSPACE, and (iv) Heat map showing the distribution of cell types in anatomical structures of the DLPFC. The intensity of the color indicates the cellular fraction within the specific anatomical region. (top) Cell2Spatial; (bottom) CytoSPACE. (I) Bar plots showing score-guided mapping accuracy (SGMA) metrics of various mapping tools across multiple spatial tissue samples, including human thymus, mouse kidney, DLPFC, human lung, intestinal, and breast tissues. The data underlying this figure can be found at https://doi.org/10.5281/zenodo.17212677. https://doi.org/10.1371/journal.pbio.3003477.g004 The correlation between the average spatial distances of different T-cell subtypes and their developmental pseudo-time were also analyzed. As shown in Fig 4D, the results of Cell2Spatial were highly consistent (R = 0.81), with later stages of development situated closer to the medullary center (Fig 4D and S5 Table). Additionally, we carried out a gene set enrichment analysis (GSEA) to compare single cells that mapped to the cortical and medullary regions. As shown in Fig 4E, cells mapped by Cell2Spatial to the medullary region exhibited significant enrichment in processes related to T-cell activation and adaptive immunity. Conversely, cells mapped to the cortical region displayed a strong association with cell proliferation and positive selection (Fig 4E). Thus, Cell2Spatial exhibited good ability in restoring the developmental trajectories of different cell types within a spatial context compared to similar methods. Next, we examined whether Cell2Spatial could accurately replicate known spatial organization patterns in the mouse kidney [35,36], using both full-spectrum SC data and ST data. In the mouse kidney ST data, 1,835 spots were surveyed, with a median of 34,439 UMIs per spot and a maximum value of 74,120 UMIs. Notably, 82% of genes within these spots exhibited zero-expression values. Furthermore, the corresponding SC data (12,987 cells) encompassed 32 distinct epithelial differentiation states. As the results shown, Cell2Spatial not only reconstructed known regional structures, but also arranged almost 30 epithelial states in the established locations (R = 0.83) of the renal unit epithelium and collecting duct system (Figs 4F, 4G, and S4A–S4H and S1 and S6 Tables). The results were consistent with the actual anatomical structures of mouse kidney, and demonstrated Cell2Spatial’s better performance compared to outcomes obtained with other tools. Finally, we scrutinized the ability of Cell2Spatial to restore the spatial architecture of the dorsolateral prefrontal cortex (DLPFC) [37] (S5A and S5B Fig and S1 Table). In the DLPFC ST dataset, 3,639 spots are present, with a median of 4,120 UMIs per spot and a maximum of 17,436 UMIs. Notably, 93.42% of genes exhibit no expression, indicating high sparsity. Additionally, the corresponding SC dataset (snRNA-seq), as reported by Mathy and colleagues [38], consists of 33,914 cells representing 17 distinct cell types. Cell2Spatial effectively mapped the major cell types to spatial spots, aligning with previous studies [37,38], and surpassing other tools in accurately reproducing the distribution of critical cell types within the prefrontal cortex (Figs 4H and S5C–S5F). However, it is crucial to acknowledge the absence of precise cell localization in these real ST datasets, making it difficult to quantify the performance of different tools regarding prior anatomical structures, development, and functional inferences. To address this limitation, we proposed a strategy based on hypothesis testing and Jaccard Index framework to provide a unified assessment of various performance dimensions: SGMA (Fig 3F and 3G; see “Materials and methods”). According to the SGMA metrics calculated by our method, Cell2Spatial consistently outperformed other methods across multiple datasets, including thymus, mouse kidney, DLPFC, human lung [39,40], intestine [9], and BRAC [8] (Figs 4I, S6A–S6E, and S7A–S7I and S1 and S7 Tables). In summary, based on a comprehensive assessment of seven ST datasets spanning six tissue types, Cell2Spatial stands out as a powerful tool for reconstructing cellular developmental pathways, understanding biological functions, and decoding structural localization compared to similar tools. Comparative evaluation of Cell2Spatial across multiple high-resolution spatial platforms Given the limited resolution of spots containing 1–10 cells in the 10× Visium platform, a comparative evaluation of Cell2Spatial’s performance in high-resolution spatial techniques was conducted using the Xenium In Situ platform, capable of mapping hundreds of transcripts at subcellular resolution [41]. For details, mouse brain data from Xenium platform was acquired, comprising 496 genes and 36,553 spots, with median UMI counts per location averaging around 207 and a minimum of 133. The Allen database’s Mouse brain SC dataset, consisting of 23 cell types and 14,249 cells, was spatially mapped (S1 Table). Our analysis revealed that Cell2Spatial closely matched spatial-based RCTD [12] annotations (RCTD commonly used for accurate annotations of high-resolution ST data), accurately depicting critical brain regions such as VLMC, L2/L3 IT, L4, L5 PT, L6CT, L6b, and Oligo (Fig 5A and 5B). While CytoSPACE demonstrated good reconstruction capabilities, Tangram exhibited some inaccuracies, particularly in capturing the distribution of Oligo cells in the mouse brain (Fig 5A and 5B). Notably, CellTrek failed to handle high-resolution Xenium data. Additionally, Seurat’s labeling strategy, which transforms cell type labels onto spatial spots rather than assigning individuals cells to spatial positions, was also excluded from our comparative analysis. Quantitative comparisons of Cell2Spatial, CytoSPACE, and Tangram demonstrated that Cell2Spatial achieved higher accuracy in mapping each cell type to spatial positions (Fig 5C). Although it effectively reconstructed high-resolution spatial structures, further validation is needed to confirm the accuracy of its spatial gene expression patterns. By investigated mouse brain data from Xenium, we selected the top 10 representative genes for each cell type using the FindAllMarkers function from Seurat and calculated their signature scores in spots (Fig 5D). The results indicated that Cell2Spatial’s reconstructed gene expression patterns closely matched the real data (R > 0.5), surpassing the performance of CytoSPACE and Tangram (Fig 5E). To further evaluate the end-to-end performance of Cell2Spatial on high-resolution data, we used mouse brain Visium HD data comprising 492,460 spots at 8 µm resolution, representing SC-scale granularity (S1 Table). For computational efficiency and memory optimization, the dataset was down-sampled to 50,000 spots. A synthetic spatial dataset (ST) was constructed from the expression matrix and spatial coordinates of these spots, while a corresponding SC dataset was generated from the same expression matrix with cluster identities assigned through Seurat-based clustering. The SC dataset was mapped back to the synthetic ST dataset to assess mapping accuracy, defined as the proportion of cells in each cluster correctly positioned at their original spatial locations. The results revealed that Cell2Spatial effectively reconstructed the spatial distribution of cells, closely matching the ground truth (Fig 5F). Comparative analysis with CytoSPACE and Tangram demonstrated that Cell2Spatial achieved higher mapping accuracy on the Visium HD platform (Fig 5G; Average accuracies: Cell2Spatial (0.49), CytoSPACE (0.30), Tangram (0.19)) under end-to-end mapping conditions. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. Application of Cell2Spatial to 10× Genomics Xenium mouse brain data. (A) Spatial architecture of the mouse brain reconstructed using Cell2Spatial, CytoSPACE, and Tangram. Ground Truth (derived by RCTD [12] tool) depicts the distribution of different cell types in the reference space. Each point represents an individual cell, with distinct colors indicating various cell types. (B) Distributions of VLMC, L2/L3 IT, L4, L5 PT, L6 CT, L6b, and Oligo cells from the outer boundary to the inner region of the mouse brain for Ground Truth, Cell2Spatial, CytoSPACE, and Tangram, respectively. Each point denotes a single cell. (C) Grouped bar plot illustrating the accuracy of different cell type distributions in the Mouse Brain reconstructed by Cell2Spatial, CytoSPACE, and Tangram. Accuracy was calculated as the number of correctly mapped cells for a specific cell type divided by the total number of that cell type in the Ground Truth. (D) Heatmaps showing the expression distribution of the top 10 over-expressed genes for VLMC, L2/L3 IT, L4, L5 PT, L6 CT, L6b, and Oligo cell types in the 10× Genomics Xenium spatial data. Colors in the heatmap represent the expression levels. (E) Scatter plots demonstrate the consistency of signature scores for different cell types between the spatial architectures reconstructed by Cell2Spatial, CytoSPACE, and Tangram, compared to the Ground Truth. Each point represents a single spot. R denotes the Pearson correlation coefficient, and p-values were derived from two-sided t-tests. (F) Evaluation of Cell2Spatial end-to-end performance using Visium HD spatial data (S1 Table) from the mouse brain. The left panel shows the true spatial distribution of cells, while the right panel presents the reconstructed distribution by Cell2Spatial. Panels C3 and C4 illustrate the spatial distribution of two selected single-cell clusters. Each point represents an individual cell. (G) Boxplot showing the accuracy of various mapping tools in reconstructing spatial organization using mouse brain Visium HD data. Accuracy is measured as the proportion of cells within each single-cell cluster accurately mapped to their original spatial positions. P-values were derived from two-sided t-tests. The data underlying this figure can be found at https://doi.org/10.5281/zenodo.17212677. https://doi.org/10.1371/journal.pbio.3003477.g005 Slide-seqV2 achieves close to SC resolution, albeit with lower transcript capture rates. Utilizing an existing mouse hippocampus SC RNA-seq dataset generated by Saunders and colleagues [42], encompassing 16 cell types with 52,846 cells, we reconstructed the spatial architecture of the Slide-seq V2 mouse hippocampus dataset, which consist of 53,173 spots and 23,264 genes (S1 Table). Within this dataset, where 98.19% of gene expression matrix entries are 0 and the median UMI counts per location hover around 302, with a minimum of only 10, Cell2Spatial has effectively reconstructed the spatial organization of the mouse hippocampus, accurately delineating regions like the entorhinal cortex, CA principal cells, dentate principal cells, oligodendrocytes, astrocytes, and ependymal areas, aligning with spatial-based RCTD annotations (i.e., ground truth) (Fig 6A and 6B). While CytoSPACE also displayed promising reconstruction abilities, it notably misallocated numerous entorhinal cortex cells to the CA principal cells region. Conversely, Tangram exhibited a disorderly distribution pattern (Fig 6A and 6B). Further quantitative analysis confirmed Cell2Spatial performed better overall compared to CytoSPACE and Tangram (Fig 6C). Additionally, in another Slide-seq V2 dataset comprising 11,626 positions of mouse cerebellum (S1 Table), with 98.55% of gene expression matrix entries being zero and median UMI counts per location approximately 30,014, Cell2Spatial accurately assigned cell type labels and captured the multi-layered structure of the cerebellum, such as Bergmann, Granule, and Purkinje, consistent with RCTD annotations (Fig 6D–6F). The analysis of cellular compositions in reconstructed spatial structures demonstrated Cell2Spatial’s high consistency with RCTD-derived cellular compositions (R > 0.97), while CytoSPACE and Tangram lagged significantly behind, underscoring their inadequacy for low-capture-rate spatial data (Fig 6G and 6H). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 6. Application of Cell2Spatial to Slide-seq V2 hippocampus and cerebellum data. (A) Spatial architecture of the mouse hippocampus reconstructed using Cell2Spatial, CytoSPACE, and Tangram. Ground Truth (derived by RCTD [12] tool) represents the distribution of different cell types in the reference space. Each point represents a single cell, with different colors indicating different cell types. (B) Distributions of Entorhinal cortex, CA Principal cells, Dentate Principal cells, Oligodendrocytes, Astrocytes, and Ependymal cells for Ground Truth, Cell2Spatial, CytoSPACE, and Tangram. Each point represents a single cell. (C) Grouped bar plot showing the accuracy of cell type distributions in the hippocampus reconstructed by Cell2Spatial, CytoSPACE, and Tangram. Accuracy was calculated as the number of correctly mapped cells for a specific cell type divided by the total number of that cell type in the Ground Truth. (D) Spatial architectures of the cerebellum reconstructed using Cell2Spatial, CytoSPACE, and Tangram. Ground Truth represents the distribution of different cell types in the reference space. Each point represents a single cell, with distinct colors indicating different cell types. (E) Distributions of Bergmann cells, Granule cells, Purkinje cells, Oligodendrocytes, Astrocytes, and Fibroblasts for Ground Truth, Cell2Spatial, CytoSPACE, and Tangram. Each point represents a single cell. (F) Grouped bar plot showing the accuracy of different cell type distributions in the cerebellum reconstructed by Cell2Spatial, CytoSPACE, and Tangram. Accuracy was calculated as the number of correctly mapped cells for a specific cell type divided by the total number of that cell type in Ground Truth. (G, H) Scatter plots showing the consistency between the spatial cellular compositions recovered by Cell2Spatial, CytoSPACE, and Tangram, compared to Ground Truth for both the hippocampus (G) and cerebellum (H). Each point represents a cell type. R indicating the Pearson correlation coefficient (PCC), with p-values obtained from two-sided t-tests, following a t-distribution to account for smaller sample sizes. The data underlying this figure can be found at https://doi.org/10.5281/zenodo.17212677. https://doi.org/10.1371/journal.pbio.3003477.g006 Overall, these findings strongly indicate that Cell2Spatial remains robust in its performance even when processing high-resolution spatial data, even with lower transcription capture rates. Diverse applications of Cell2Spatial To investigate the potential applications of Cell2Spatial, we utilized it to analyze ST data (comprising 1,438 spots) of mouse kidney tissue, aiming to explore the spatial co-existences among various cell types and the temporal differentiation of crucial cell types within the spatial framework. By assigning single cells from the mouse kidney (12,987 cells encompassing 13 cell types) reported by Ransick and colleagues [36] to spots, as illustrated in Fig 7A, distinct spatial distributions among different cell types were observed (Figs 7A, S8A, and S8B and S1 Table). To quantify these relationships, we introduced a co-existence index, and the results revealed robust spatial co-localization links between T cells, fibroblasts, and natural killer (NK) cells (Fig 7B and S8 Table; see “Materials and methods”). To further verify these findings, we highlighted the expression of their respective marker genes (Cd3d, Col3a1, and Nkg7) in the ST data of mouse kidney, which showed a significant overlap in the spatial distribution of these three cell types, confirming the results of Cell2Spatial (S8C Fig). Additionally, we elucidated the differentiation timelines of proximal tubular and distal tubular cells through pseudo-time analysis using Monocle2 [43–45]. As shown in Fig 7C and 7D, proximal tubules exhibited an outward-to-inward differentiation trend, while distal tubules displayed an inward-to-outward progression (Fig 7C and 7D). The opposing timelines align with the findings of Wei and colleagues [3]. Thus, Cell2Spatial can accurately track the timelines of cell differentiation. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 7. Cell2Spatial enables versatile applications across various analytical domains. (A) Left: Spatial architecture of the mouse kidney depicted by Cell2Spatial. Each dot represents an individual cell, with cell types are color-coded. Right: Uniform Manifold Approximation and Projection (UMAP) plot showing the single-cell atlas of mouse kidney. (B) Heat map showing the coexistence of mapped cell types in spatial spots. The Pearson correlation coefficient (PCC) is obtained based on coexistence index (see “Materials and methods”). (C) Trajectory analysis of proximal tubule cells (left), with spatial mapping of the pseudo-time values in the tissue section (right). (D) Trajectory analysis of distal tubule cells (left), with spatial mapping of the pseudo-time values in the tissue section (right). (E) Left: Renal cell cancer (RCC) tissue section displaying spot clustering determined using the Seurat common processing strategy. Right: Pre-defined spatial coordinates of the tertiary lymphoid structure (TLS) for reference. (F) Heat map showing the relationship of inclusion between TLS and spatial clusters. The intensity of color indicates the number of TLS spots located in the cluster. (G) Left: Spatial architecture of the RCC tissue section reconstructed using Cell2Spatial. Each dot represents an individual cell and cell types were color-coded. Right: Enlarged cell atlas located within the TLS region. (H) Overview of cell types co-existence in the RCC tissue section. Edge thickness reflects the strength of the co-existence index, with edges below a specified cutoff (0.05) excluded. (I) Circle plots showing the interactions between B cells and other cell types within an independent spot cluster. The thickness of the edges indicates the strength of interactions. Left: Cluster 3; Right: Cluster 4. (J) Circle plots depicting the interactions between B cells and other cell types within an independent spot cluster. Left: Cluster 2; Right: Cluster 6. The data underlying this figure can be found at https://doi.org/10.5281/zenodo.17212677. https://doi.org/10.1371/journal.pbio.3003477.g007 To determine whether Cell2Spatial can detect critical structural regions within tissues, ST data with 3,206 spots concerning tertiary lymphoid structures (TLS) associated with renal cell cancer (RCC) from the study published by Meylan and colleagues [46] were acquired (Fig 7E and S1 Table). TLSs are closely associated with tumor development and metastasis, carrying diagnostic and therapeutic significance in cancer. In this dataset, the precise coordinates of TLS are well-documented, with a notable concentration in Cluster 3 and Cluster 4 (Fig 7E and 7F). By superimposing an immune SC atlas (down-sampled to 24,834 cells covering 25 major immune cell types) [47] onto the ST section of RCC, a diverse array of immune cell types within the TLS were observed (Fig 7G). Spatial cell co-existence analysis revealed that B cells were the major interacting cell type (Fig 7H and S9 Table), and had the most substantial and abundant interactions with other immune cells in Cluster 4, closely followed by Cluster 3 (Fig 7I and 7J). These findings suggest the presence of a functionally significant structure (Clusters 3 and 4) in RCC tissue, in alignment with previously documented structures. Taken together, Cell2Spatial can used to explore functional and structural regions within tissues, and the intricate network of interactions between diverse cell types and their spatial arrangement. Overview of Cell2Spatial The Cell2Spatial framework aims to map single cells to spatial spots for reconstructing tissue architecture (Fig 1). The process begins with normalizing SC and ST data, ideally from the same tissue source, using Seurat’s SCTransform to establish a common basis for gene expression comparison. To quantify similarities between single cells and spatial spots, maximum likelihood model is employed. Efficiency is enhanced by identifying cell-type-specific genes through a modified entropy-based method (Fig 1, Step 1; see “Materials and methods”). These gene sets, along with their expression profiles, form probability matrices that construct the maximum likelihood model and generate a likelihood matrix. Spatial context is incorporated through Seurat’s canonical correlation analysis (CCA) [18] for SC–ST integration, followed by principal component analysis (PCA) for dimensional reduction. A Euclidean distance matrix is derived from the Uniform Manifold Approximation and Projection (UMAP) embedding, constructed on the PCA space. This distance matrix is applied as weights (scaled to 1) to the likelihood matrix, resulting in a weighted likelihood matrix that balances gene expression similarity with spatial proximity (Fig 1, Step 2; see “Materials and methods”). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Overview of the Cell2Spatial algorithm. The Cell2Spatial framework comprises seven fundamental components: data preprocessing, marker gene selection, similarity matrix construction, hotspot detection, deconvolution and cell count estimation, single-cell (SC) pre-assignment, and final mapping to spots (see “Materials and methods”). SC and spatial transcriptomics (ST) data were standardized using SCTransform in Seurat, with cell-type-specific genes identified through a modified entropy-based method (Step 1). A spatially weighted similarity matrix was created using a maximum likelihood model, with Seurat’s canonical correlation analysis (CCA) connecting single cells to spatial spots to yield spatial weights (Step 2). Cell type hotspots were identified using the Getis-Ord G* index combined with normality testing, while robust linear regression was applied for deconvolution of cellular composition, and corrected saturation model was used to estimate cell counts in low-resolution ST data (Steps 3-5). Feedforward neural network (FNN) was employed for cell pre-assignment to ST clusters (Step 6), and the Jonker-Volgenant algorithm iteratively mapped individual cells to specific spatial spots based on similarities and estimated cell counts (Step 7). https://doi.org/10.1371/journal.pbio.3003477.g001 To improve robustness, particularly when SC and ST data may not perfectly align in terms of cell types, the weighted likelihood matrix undergoes refinement. Spots that do not correspond to any pre-defined cell types in the SC dataset are filtered out determined by the Getis-Ord G* index coupled with a normal test based on signature scores from cell-type-specific genes, ensuring focus on relevant spatial regions (hotspot regions) while minimizing mapping noise (Fig 1, Step 3; see “Materials and methods”). Understanding the cellular compositions within ST sections is critical for recovering tissue spatial organization. This is achieved through a deconvolution model that applies robust linear regression model (RLM) in conjunction with base matrix constructed from specific gene sets (Fig 1, Step 4; see “Materials and methods”) [21,22]. The corrected saturation model, was designed to estimate cell counts per spot in low-resolution ST data by linking the library size and gene count, through a saturation-based relationship (Fig 1, Step 5; see “Materials and methods”). For high-resolution ST data, where spots typically capture single cells, it assigns a cell count of 1 per spot. Therefore, the total number of cells to be aligned is calculated as the sum of cell counts across the remaining spots, with the number of cells for each type sampled from the SC dataset based on these totals and the estimated cellular proportions, yielding candidate mapping SC data. To assign candidate single cells to spatial spots, the Jonker-Volgenant algorithm [23] is employed, utilizing the refined weighted likelihood matrix and the estimated cell counts per spot. In large-scale applications, however, this algorithm imposes substantial memory and processing demands. To mitigate these challenges, a FNN framework is implemented. The FNN is trained to pre-allocate cells to ST clusters based on the co-embedding of SC and ST data (Fig 1, Step 7; see “Materials and methods”). This pre-allocation streamlines the execution of the Jonker-Volgenant algorithm, allowing it to be applied iteratively within each ST cluster, ensuring efficient and scalable SC-to-spot assignments (Fig 1, Step 7; see “Materials and methods”). Evaluation of the performance of Cell2Spatial using mouse brain transcriptome data To evaluate the performance of Cell2Spatial, ST data from the 10× Visium platform for the mouse brain was obtained, comprising 2,696 spots. Among these, around 66.53% of genes exhibited no expression, with the highest UMI count recorded at 25,580 and a median count of 24,332 UMIs (S1A Fig and S1 Table). Additionally, we utilized a reference dataset (SC) from the Allen Institute, which comprises approximately 14,000 adult mouse cortical cells, covering 23 distinct cell types (S1B Fig). As the Fig 2A shown, the application of Cell2Spatial enabled us to map the single cells to precise spatial spots, and the findings unveiled the presence of distinct anatomical distributions of various cell types within the mouse brain (Fig 2A and 2B). Notably, our approach capitalizes on a modified specificity entropy strategy for the identification of cell-type-specific genes, showcasing noteworthy discriminatory power and demonstrating its effectiveness (S1C Fig). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Performance assessment of Cell2Spatial on real and simulated mouse brain spatial transcriptomics (ST) data. (A) Spatial architecture of mouse brain reconstructed using Cell2Spatial. Cell types are marked by color codes, and each dot represents an individual cell. (B) Distribution of cell types Astro, L2/3 IT, L4, L5 IT, and L6 CT on mouse brain ST slice. (C) Spatial architecture of the mouse brain reconstructed using CytoSPACE, CellTrek, Tangram, and Seurat, respectively. Cell types are color-coded. Each dot represents an individual cell. (D) Spatial k-distance (k = 10) of L2/3 IT, L4, L5 IT, L5 PT, NP, L6 IT, L6 CT, L6b, Oligo to Astro cells. Violin boxplots illustrate the median, and interquartile ranges (25%–75%), with whiskers extending up to 1.5 times the interquartile range beyond the box. (E) Bar plot showing the conservation of highly variable genes (HVGs) between reconstructed spatial expression data from mapping tools and the original spatial expression data (see “Materials and methods”). (F) Spatial heat maps showing the performance of Cell2Spatial for aligning scRNA-seq data (with 5% added noise) to spatial spots in ST datasets simulated with five cells on average (see “Materials and methods”). Only cell types with prominent spatial structures have been presented for the sake of clarity. The color intensity of each spot corresponds to the number of mapped single cells. (G) Box plots showing the performance across various methods, and noise levels (0, 5%, 10%, 15%, and 20%) in accurately assigning individual cells to their respective locations in simulated ST datasets. Each data point represents a distinct cell type. The central lines within the boxes, the box edges, and the whiskers correspond to the medians, the first and third quartiles, and the minimum and maximum values, respectively, within 1.5 times the interquartile range of the box boundaries. P-values were obtained by Kruskal–Wallis test, as it is appropriate for comparing more than three groups of non-normally distributed data without assuming equal variances. (H) Boxplot showing the end-to-end Jaccard index at the single-cell level, measuring the consistency between the coordinates of single cells in the reconstructed data and their coordinates in synthetic ST data. Perturbations were introduced by randomly modifying gene expression in 0%, 5%, 10%, 15%, or 20% of genes. For each perturbation level, three independent replicates were generated to ensure robustness of the evaluation. The data underlying this figure can be found at https://doi.org/10.5281/zenodo.17212677. https://doi.org/10.1371/journal.pbio.3003477.g002 Next, the performance of Cell2Spatial with other mapping tools capable of segmenting spatial spots into SC granularity was compared, including CytoSPACE [20], CellTrek [3], Tangram [19], and Seurat [18]. The findings indicated that, the performance of the other four mapping tools within the ST sections was significantly weaker than that achieved by Cell2Spatial (Figs 2C and S1D–S1G). Importantly, Tangram’s mapping did not reveal a distinct alignment between cell types and anatomical structures, and Seurat directly assigned spots as individual cell types. Furthermore, as shown in Fig 2D, Cell2Spatial generated consistent spatial relationships between specific cell types (L2/3 IT, L4, L5 IT, L5 PT, NP, L6 IT, L6 CT, L6b, Oligo, to Astro) based on k-distances, which were in agreement with previous studies [3,19] (Fig 2D and S2 Table). CellTrek was also efficient, whereas Tangram exhibited less favorable performance. Additionally, we reconstructed spot-level expression profiles using the mapped SC resolution ST dataset and assessed highly variable genes (HVGs) conservation, with Cell2Spatial demonstrating the highest conservation compared to unprocessed ST data (Fig 2E; Cell2Spatial: 0.357, CytoSPACE: 0.166, CellTrek: 0.221, Tangram: 0.145; see “Materials and methods”). This indicates its effectiveness in preserving the intrinsic gene characteristics of spatial spots within tissue sections. Notably, since Seurat’s label transfer function was used, the spot-level expression in the ST data remained unchanged post-transfer and was excluded from this analysis. Nevertheless, the limited information regarding cell composition and location within the real ST data of mouse brain mitigate our evaluation of Cell2Spatial’s performance. To overcome this challenge, simulated ST datasets of the mouse brain were generated with accurately known cell type localization and composition at various noise levels (see “Materials and methods”). Applied Cell2Spatial, CytoSPACE, CellTrek, Tangram, and Seurat to map the SC data of the mouse brain to the simulated ST datasets, our results demonstrated that Cell2Spatial adeptly restored the spatial positions of cell types, and exhibited a strong alignment with actual cell counts of spots, as reflected by the root mean square error (RMSE) between predicted and expected counts (Cell2Spatial: 2.21; CytoSPACE: 3.29; CellTrek: 4.10; Tangram: 8.52; Seurat: 6.38) (Figs 2F and S2A–2C). We simulated spots with varying cell counts to further evaluate Cell2Spatial’s ability to estimate cell numbers per spot (S2D Fig). Across different values of Lambda (Lambda = 1:20; representing the Poisson distribution parameter used to generate random cell counts), Cell2Spatial consistently achieved Pearson correlation coefficients (PCCs) above 0.89 (median: 0.941; maximum: 0.952), whereas CytoSPACE achieved coefficients above 0.61 (median: 0.693; maximum: 0.783), demonstrating a significant performance difference (Two-sided Wilcoxon test; p-value < 0.001) (S2E–S2G Fig and S3 Table). Consistently, RMSEs were also significantly lower for Cell2Spatial (median: 1.15) compared to CytoSPACE (median: 2.76) (Two-sided Wilcoxon test; p-value < 0.001) (S2G Fig). To assess performance more effectively, we simulated ST datasets with known cell type composition and spatial distributions under different perturbation levels (see “Materials and methods”). Based on accuracy indices, Cell2Spatial consistently outperformed other tools at the cell type level across all noise conditions (Fig 2G and S4 Table). Additionally, we investigated a crucial aspect of the analysis: mapping SC datasets designed for simulating ST data to their corresponding synthetic ST data to assess the accuracy of cell localization—i.e., end-to-end mapping. To minimize the influence of stochastic processes on performance evaluation, we performed three replicates under each perturbation level. The results indicated that CellTrek achieved the highest Jaccard index (mean index: 0.193), with Cell2Spatial performing next (mean index: 0.167), outperforming both CytoSPACE (mean index: 0.106) and Tangram (mean index: 0.160) in overall accuracy (Fig 2H). The relatively strong accuracy of CellTrek may stem from its Random Forests framework, which can better capture complex, non-linear relationships in ST data and provide greater robustness to noise compared with the linear assignment algorithm used by Cell2Spatial. Notably, all tools showed low Jaccard indices (<0.20), reflecting a potential limitation: current methods mainly distinguish between cell types but lack power to resolve cells within the same type, making them better at reconstructing cell-type distributions than pinpointing individual cell locations. Taken together, Cell2Spatial can accurately infer the composition of cell types within spatial structures, and effectively reconstruct spatial architectures for tissues. Assessing Cell2Spatial’s effectiveness across diverse metrics and conditions ST datasets with varying spot resolutions (Lambda = 5, 10, 15, 20) and cellular compositions were simulated, where Lambda served as the Poisson distribution parameter for randomly generating cell numbers per synthetic spot. We initially focused on Cell2Spatial’s ability to predict cellular proportions within these simulated ST sections. To facilitate comparison, we incorporated 10 additionally spatial deconvolution tools: Cell2location [10], SpatialDWLS [13], RCTD [12], Stereoscope [16], DestVI [15], SpaOTsc [24], novoSpaRc [25], SPOTlight [26], CARD [11], and DSTG [27], using SC data from the mouse brain as the reference to estimate cellular proportions in each spot from the synthetic ST datasets. For mapping tools, including Cell2Spatial, CytoSPACE, CellTrek, and Tangram, we aligned single cells from mouse brain to the synthetic ST datasets, and then calculated the cellular proportions. In the case of Seurat’s label-transfer method, we utilized the probability matrix obtained through the “TransferData” function to represent cellular proportions in each spot. As shown in Fig 3A, across various spot resolutions (Lambda = 5, 10, 15, and 20), Cell2Spatial (PCCs: 0.969–0.976; RMSEs: 0.0135–0.0148) outperformed similar tools such as CytoSPACE (PCCs: 0.910–0.974; RMSEs: 0.0134–0.0271), CellTrek (PCCs: 0.777–0.833; RMSEs: 0.0332–0.0402), and Tangram (PCCs: 0.246–0.354; RMSEs: 0.0541–0.0615) in predicting cellular proportions (Fig 3A; see “Materials and methods”). Additionally, when compared to the spatial deconvolution tool CARD, Cell2Spatial demonstrated comparable efficacy in predicting cellular compositions, achieving PCCs greater than 0.97, and RMSEs below 0.015 across all resolutions (Fig 3A; bottom panel). We further benchmarked Cell2Spatial using 32 published SC and simulated ST datasets from Li and colleagues [28], which cover human brain, liver, lung, kidney, and pancreas, as well as mouse pancreas and trachea. These datasets contain a median of eight cell types (range: 6–128), with eight datasets including more than 10 cell types. Each simulated dataset comprises 1,000 spots with known cellular compositions. On these benchmarks, Cell2Spatial demonstrated strong performance in reconstructing spot-level cell type compositions, achieving an average PCC of 0.784 (range: 0.427–0.964; highest in dataset30, lowest in dataset1), ranking second among all tools, just behind Cell2location (range: 0.619–0.942) (Fig 3B). For RMSE, Cell2Spatial achieved an average of 0.114 (range: 0.027–0.220), ranking fifth overall. While Cell2location (range: 0.031–0.145) showed the best overall accuracy, Cell2Spatial maintained competitive performance in estimating cell type compositions (Fig 3B). These benchmarking results on both simulated and published datasets confirm that Cell2Spatial achieves high accuracy and robustness across diverse conditions. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Evaluation of spatial transcriptomics (ST) tools across various metrics and conditions. (A) Bar plots showing Pearson correlation coefficients (PCCs) and Root Mean Square Errors (RMSEs) between predicted and true proportions in synthetic ST datasets under different spot resolution levels (Lambda = 5, 10, 15, and 20). The Lambda value represents the expected cell counts in spots, modeled by a Poisson distribution. Error bars indicate variability, determined through a permutation strategy repeated 100 times, where 2,000 spots were down-sampled, and PCC and RMSE were calculated for each iteration. (B) Violin combined with box plots showing the PCCs and RMSEs between estimated and actual cellular proportions. Black lines within the boxes indicate medians. Thirty-two simulated ST datasets, together with their corresponding single-cell (SC) datasets, were retrieved from the study by Li and colleagues [28]. (C) Kullback–Leibler (KL) divergence metrics for spatial cell charting methods, comparing each cell type against synthetic ST dataset references. (D) Grouped bar plots showing cosine similarities between recovered spot expressions post-SC mapping and the original synthetic ST dataset expressions. Error bars represent values obtained via permutation strategy, with 2,000 spots down-sampled and cosine similarities calculated across 100 iterations. (E) Grouped bar plots showing the conservation of highly variable genes (HVGs) between recovered spot expressions and synthetic ST dataset expressions following SC mapping. (F) Bar plot showing the average Jaccard index for cell type consistency in spots, comparing predictions from our score-guided mapping accuracy (SGMA; see “Materials and methods”) strategy against true cell types in synthetic ST datasets. (G) Box plot displaying SGMA indexes for cell type consistency in spots, comparing predictions using mapping tools (Cell2Spatial, CytoSPACE, CellTrek, and Tangram) with true cell types in synthetic ST datasets, based on the SGMA strategy (see “Materials and methods”). (H) Scatter plot mapping a subset of mouse brain SC data to ST spots under unmatched conditions (where cell types from SC data represent only a subset of those in ST datasets). Each dot corresponds to an individual cell, with cell types color-coded. (I) Violin combined with box plots showing the PCC and RMSE between estimated and actual cell counts in spots. Simulated ST datasets, together with their corresponding SC datasets, were retrieved from the study by Li and colleagues [28]. (J) Spatial feature plots illustrating inferred cell counts in spots using the DAPI channel in fluorescence images (counted via Squidpy [29]), Cell2Spatial, and CytoSPACE. The top panel represents a coronal section of the mouse brain, while the bottom panel represents the whole brain. Color shading indicates different cell counts in spots. (K, L) Density plots combined with fitted lines, showing the consistency between cell counts estimated by Cell2Spatial or CytoSPACE and those counted using DAPI via the Squidpy tool [29]. “R” represents the Spearman correlation, with p-values obtained from the two-sided t-tests. (M) Scatter plot with line showing the influence of marker size on Cell2Spatial performance for different cell types. The marker size range is 10 to 200, with a step size of 10. (N, O) Scatter plots with line showing peak memory usage and elapsed time for different mapping tools under varying spot sizes, ranging from 1,000 to 10,000, with a step size of 1,000. The data underlying this figure can be found at https://doi.org/10.5281/zenodo.17212677. https://doi.org/10.1371/journal.pbio.3003477.g003 Next, to assess the spatial density of cell charting results against the actuary spatial distribution for various cell types in the synthetic ST datasets (Fig 3A), Kullback–Leibler (KL)-divergence was used. The analysis showed that Cell2Spatial had consistently lower KL-divergence values than CytoSPACE, CellTrek, and Tangram, indicating that Cell2Spatial more accurately recovered spatial cellular structures (Fig 3C; see “Materials and methods”). We also evaluated gene expression recovery by comparing the cosine similarity between the recovered spot expressions and the original synthetic datasets. Cell2Spatial consistently achieved the highest similarity (approximately 0.8) across all datasets, highlighting its effectiveness in accurately reconstructing gene expression patterns (Fig 3D; Cell2Spatial: 0.787–0.798; CytoSPACE: 0.779–0.789; CellTrek: 0.753–0.778; Tangram: 0.712–0.736). Furthermore, we assessed HVGs conservation, which serves as a proxy for the preservation of the biological signal in the ST expression profile before and after mapping, and found that Cell2Spatial performed well in retaining HVGs across all spot resolutions, demonstrating its ability to preserve key gene expression features effectively (Fig 3E; Average HVGs conserved scores: Cell2Spatial (0.556); CytoSPACE (0.411); CellTrek (0.361); Tangram (0.310)). Moreover, we proposed a metric, the Score-Guided Mapping Accuracy (SGMA) index. This method scores spots using marker genes for specific cell types, capturing the prominent spatial regions of these cell types and comparing the consistency of these regions in the ST spatial data before and after mapping (see “Materials and methods”). Using marker genes to score synthetic ST spots, we identified high-confidence spots for each cell type. The results, as shown in Fig 3F, demonstrated that the cell type distributions inferred by SGMA strategy were highly consistent with the true distributions, with Jaccard indexes exceeding 0.75 (Fig 3F; range: 0.764–0.853), demonstrating SGMA’s effectiveness in evaluating mapping performance. Applying SGMA index to various mapping tools across multiple synthetic ST datasets, we found that Cell2Spatial achieved the highest median Jaccard index for recovering ST cell type distributions, reinforcing its strong performance in cell type identification (Fig 3G; Cell2Spatial: 0.503–0.560; CytoSPACE: 0.289–0.497; CellTrek: 0.304–0.428; Tangram: 0.248–0.315). Notably, SGMA index is not only valuable for synthetic ST datasets but also serves as an effective benchmark for evaluating mapping tools on real experimental ST data. In particularly, to investigate the performance of mapping tools under mismatched conditions, where the cell types in the SC data represent only a subset of those in the ST dataset, we focused on three cell types—Astro, L2/3 IT, and L4—from the mouse brain SC dataset. We used four mapping tools to project single cells from these cell types to ST spots. The results revealed that Cell2Spatial effectively reconstructed the spatial distribution of these three cell types (Fig 3H). In contrast, CytoSPACE, which employs a full spatial slice mapping approach, struggled to accurately reconstruct the tissue’s spatial structure under these mismatched conditions. CellTrek partially captured the spatial distribution of the three cell types but was less precise for Astro cells, while Tangram did not clearly reveal the spatial distribution of these cell types. These findings highlight Cell2Spatial’s capability to handle complex and mismatched datasets effectively. To validate the accuracy of spot-level cell count estimation by Cell2Spatial, we applied both Cell2Spatial and CytoSPACE to 32 simulated ST datasets generated by Li and colleagues [28], each containing spots with known cell numbers. Cell2Spatial achieved higher correlations with true counts (PCC range: 0.507–0.903; mean = 0.725) compared to CytoSPACE (PCC range: 0.480–0.898; mean = 0.700) (Fig 3I). RMSE values further confirmed its advantage, with Cell2Spatial showing significantly lower errors (range: 1.643–3.846; mean = 2.390) relative to CytoSPACE (range: 1.935–3.834; mean = 2.683) (Fig 3I). We next applied Cell2Spatial to two mouse brain ST datasets (10× Visium) with available DAPI channel images. Cell counts were quantified from DAPI staining using Squidpy [29] and compared with estimates from Cell2Spatial and CytoSPACE. Cell2Spatial’s estimates more closely matched the DAPI-based counts, achieving correlation coefficients of 0.62 and 0.49, outperforming CytoSPACE (coefficients: 0.52 and 0.35) (Fig 3J–3L; see “Materials and methods”). To assess the effect of predefined maximum cell numbers (nmax), we varied nmax between 4 and 24 in a synthetic mouse brain ST dataset (true nmax = 14). As nmax deviated from the true value, RMSEs increased (S2H Fig; left panel). Nonetheless, PCCs consistently exceeded 0.8, indicating that despite differences in nmax, the overall trend of spot-level cell distribution was preserved (S2H Fig; right panel). Considering the central role of cell type-specific genes in Cell2Spatial, we investigated how varying the number of selected specific genes per cell type affected Cell2Spatial’s performance. The results revealed that while increasing the number of specific genes enhanced the conservation of HVGs in the reconstructed spatial expression profiles, Cell2Spatial’s performance remained relatively stable within a certain range (HVGs conservation: 0.47 ± 0.15) (Fig 3M). This stability suggests that Cell2Spatial is resilient to changes in gene selection. Finally, we evaluated the computational efficiency of the mapping tools. Cell2Spatial demonstrated relatively high memory usage, though it stayed within acceptable limits, and consistently achieved relatively short runtimes even as the spot size increased (Fig 3N and 3O; see “Materials and methods”). In contrast, CytoSPACE and Tangram showed a linear increase in computational time, highlighting Cell2Spatial’s good efficiency and practical advantages. Overall, our evaluation shows that Cell2Spatial performs well across various metrics and conditions. Its effectiveness in estimating cellular compositions, recovering gene expression, and maintaining computational efficiency, along with its adaptability to different input parameters, makes Cell2Spatial a valuable tool for ST analysis. Spatial profiling of single-cells in various tissue types with Cell2Spatial To assess the ability of Cell2Spatial in analyzing ST data from diverse tissue sources, ST data [30] along with a reference scRNA-Seq dataset of human thymus tissue [31] were obtained. The thymus ST dataset consisted of 1,628 spots, with a maximum of 38,161 unique molecular identifiers (UMIs) and a median of 91,66 UMIs, while the SC dataset encompassed 40 distinct cell types. Through the process of assigning cells to ST spots, Cell2Spatial successfully mapped 39 cell types to the thymus section, revealing notable differences in cell type distribution between the medulla and cortical regions (Figs 4A and S3A–S3D and S1 Table). As known, the human thymus is the site of T-cell development, with double-negative (DN) and double-positive (DP) T cells residing in the cortical region, while single-positive (SP) cells are mainly localized to the medullary region (Fig 4B). We tested the ability of various mapping tools to capture the thymic T cell development process, with focus on the DN, DP, ABT.entry (transition from DP to SP states), CD8+ T, CD4+ T cells, and cortical thymic epithelial cells (cTECs) and medullary thymic epithelial cells (mTECs) that are involved in T cell maturation. Cell2Spatial revealed distinct spatial patterns, with DN and DP cells predominantly inhabiting the thymic cortex, and ABT.entry cells strategically positioned in the cortical region near the corticomedullary junction (Fig 4C). On the other hand, CD8+ T and CD4+ T cells, representing mature SP cells, were predominantly found in the thymic medulla (Fig 4C). Furthermore, cTECs were primarily concentrated in the cortical region, while mTECs were detected in the medullary region, consistent with previous researches [31–33]. Interestingly, the presence of a small subset of cTECs in the medullary region hints at the existence of certain cTEC subtypes exhibiting characteristics of mTECs [31,34]. These suggest that the thymus architecture restored by Cell2Spatial aligned with the developmental pathway of T-cells. While CytoSPACE, CellTrek, and Tangram also partially captured the relationship between critical T-cell subpopulations and thymus structures, their performance slightly lags behind Cell2Spatial (Fig 4C). Notably, Seurat’s performance was sub-optimal due to its interpretation of spots at the SC level. These findings highlight the effectiveness and reliable of Cell2Spatial in deciphering the essential spatial structures of anatomical tissues. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. Performance assessment of Cell2Spatial in real ST data from various tissue types. (A) Spatial architecture of the human thymus (Suo and colleagues) [30] reconstructed using Cell2Spatial. Cell types are color-coded. Each dot represents an individual cell. (B) Thymus anatomy and the T cell development process. (Left) Hematoxylin and eosin (H&E) staining images depicting the thymus structure. The white line marks the medullary-cortical junction, with colored regions indicating the medullary areas and uncolored regions representing the cortical areas. (Right) T cell development stages, encompassing the double-negative (DN), double-positive (DP), and single-positive (SP) stages. DN and DP stages were primarily located in the cortical regions, while the SP stage is concentrated in the medullary regions. (C) Distribution of selected cell types on human thymus ST slices. From left to right, cell types include DN, DP, ABT.entry, CD8+ T cells, CD4+ T cells, cTECs, and mTECs. (D) Scatter plots of various mapping tools showing the concordance between the relative Euclidean distance ranking of T cell subtypes at different stages and their developmental sequence determined by pseudo-time analysis. Euclidean distances of individual cells were calculated relative to a reference point at the center of the medullary region (regions 3, 7, and 8 in S3A Fig, left panel). For each subtype, the mean distance from the reference point was used to establish a spatial ranking, which was then compared with the developmental order. The strength of this association was quantified using Spearman correlation analysis. The blue line represents a linear fit, and the shaded area denotes 95% confidence interval. Different colored dots represent distinct T cell subgroups. “R” stands for the Pearson correlation coefficient, and the p-values were obtained from two-sided t test, which follows a t-distribution to account for smaller sample sizes. (E) Gene set enrichment analysis (GSEA) performed on single cells mapped to both the medullary and cortical regions using Cell2Spatial. (F) Left: Epithelial cell transcriptomes from a mouse kidney single-cell atlas were mapped onto spatial spots of a normal mouse kidney using Cell2Spatial, displayed with jitter within their assigned spots. Right: The same representation, with cells colored based on their known distance to the inner medulla. (G) Concordance between predicted and known distances of each epithelial state to the base of the inner medulla (illustrated in the left-top panel). “R” stands for the Spearman correlation coefficient, and the p-values were obtained from two-sided t-tests. (H) (i) Anatomical structures within the dorsolateral prefrontal cortex (DLPFC), (ii) Spatial architecture determined using Cell2Spatial, (iii) CytoSPACE, and (iv) Heat map showing the distribution of cell types in anatomical structures of the DLPFC. The intensity of the color indicates the cellular fraction within the specific anatomical region. (top) Cell2Spatial; (bottom) CytoSPACE. (I) Bar plots showing score-guided mapping accuracy (SGMA) metrics of various mapping tools across multiple spatial tissue samples, including human thymus, mouse kidney, DLPFC, human lung, intestinal, and breast tissues. The data underlying this figure can be found at https://doi.org/10.5281/zenodo.17212677. https://doi.org/10.1371/journal.pbio.3003477.g004 The correlation between the average spatial distances of different T-cell subtypes and their developmental pseudo-time were also analyzed. As shown in Fig 4D, the results of Cell2Spatial were highly consistent (R = 0.81), with later stages of development situated closer to the medullary center (Fig 4D and S5 Table). Additionally, we carried out a gene set enrichment analysis (GSEA) to compare single cells that mapped to the cortical and medullary regions. As shown in Fig 4E, cells mapped by Cell2Spatial to the medullary region exhibited significant enrichment in processes related to T-cell activation and adaptive immunity. Conversely, cells mapped to the cortical region displayed a strong association with cell proliferation and positive selection (Fig 4E). Thus, Cell2Spatial exhibited good ability in restoring the developmental trajectories of different cell types within a spatial context compared to similar methods. Next, we examined whether Cell2Spatial could accurately replicate known spatial organization patterns in the mouse kidney [35,36], using both full-spectrum SC data and ST data. In the mouse kidney ST data, 1,835 spots were surveyed, with a median of 34,439 UMIs per spot and a maximum value of 74,120 UMIs. Notably, 82% of genes within these spots exhibited zero-expression values. Furthermore, the corresponding SC data (12,987 cells) encompassed 32 distinct epithelial differentiation states. As the results shown, Cell2Spatial not only reconstructed known regional structures, but also arranged almost 30 epithelial states in the established locations (R = 0.83) of the renal unit epithelium and collecting duct system (Figs 4F, 4G, and S4A–S4H and S1 and S6 Tables). The results were consistent with the actual anatomical structures of mouse kidney, and demonstrated Cell2Spatial’s better performance compared to outcomes obtained with other tools. Finally, we scrutinized the ability of Cell2Spatial to restore the spatial architecture of the dorsolateral prefrontal cortex (DLPFC) [37] (S5A and S5B Fig and S1 Table). In the DLPFC ST dataset, 3,639 spots are present, with a median of 4,120 UMIs per spot and a maximum of 17,436 UMIs. Notably, 93.42% of genes exhibit no expression, indicating high sparsity. Additionally, the corresponding SC dataset (snRNA-seq), as reported by Mathy and colleagues [38], consists of 33,914 cells representing 17 distinct cell types. Cell2Spatial effectively mapped the major cell types to spatial spots, aligning with previous studies [37,38], and surpassing other tools in accurately reproducing the distribution of critical cell types within the prefrontal cortex (Figs 4H and S5C–S5F). However, it is crucial to acknowledge the absence of precise cell localization in these real ST datasets, making it difficult to quantify the performance of different tools regarding prior anatomical structures, development, and functional inferences. To address this limitation, we proposed a strategy based on hypothesis testing and Jaccard Index framework to provide a unified assessment of various performance dimensions: SGMA (Fig 3F and 3G; see “Materials and methods”). According to the SGMA metrics calculated by our method, Cell2Spatial consistently outperformed other methods across multiple datasets, including thymus, mouse kidney, DLPFC, human lung [39,40], intestine [9], and BRAC [8] (Figs 4I, S6A–S6E, and S7A–S7I and S1 and S7 Tables). In summary, based on a comprehensive assessment of seven ST datasets spanning six tissue types, Cell2Spatial stands out as a powerful tool for reconstructing cellular developmental pathways, understanding biological functions, and decoding structural localization compared to similar tools. Comparative evaluation of Cell2Spatial across multiple high-resolution spatial platforms Given the limited resolution of spots containing 1–10 cells in the 10× Visium platform, a comparative evaluation of Cell2Spatial’s performance in high-resolution spatial techniques was conducted using the Xenium In Situ platform, capable of mapping hundreds of transcripts at subcellular resolution [41]. For details, mouse brain data from Xenium platform was acquired, comprising 496 genes and 36,553 spots, with median UMI counts per location averaging around 207 and a minimum of 133. The Allen database’s Mouse brain SC dataset, consisting of 23 cell types and 14,249 cells, was spatially mapped (S1 Table). Our analysis revealed that Cell2Spatial closely matched spatial-based RCTD [12] annotations (RCTD commonly used for accurate annotations of high-resolution ST data), accurately depicting critical brain regions such as VLMC, L2/L3 IT, L4, L5 PT, L6CT, L6b, and Oligo (Fig 5A and 5B). While CytoSPACE demonstrated good reconstruction capabilities, Tangram exhibited some inaccuracies, particularly in capturing the distribution of Oligo cells in the mouse brain (Fig 5A and 5B). Notably, CellTrek failed to handle high-resolution Xenium data. Additionally, Seurat’s labeling strategy, which transforms cell type labels onto spatial spots rather than assigning individuals cells to spatial positions, was also excluded from our comparative analysis. Quantitative comparisons of Cell2Spatial, CytoSPACE, and Tangram demonstrated that Cell2Spatial achieved higher accuracy in mapping each cell type to spatial positions (Fig 5C). Although it effectively reconstructed high-resolution spatial structures, further validation is needed to confirm the accuracy of its spatial gene expression patterns. By investigated mouse brain data from Xenium, we selected the top 10 representative genes for each cell type using the FindAllMarkers function from Seurat and calculated their signature scores in spots (Fig 5D). The results indicated that Cell2Spatial’s reconstructed gene expression patterns closely matched the real data (R > 0.5), surpassing the performance of CytoSPACE and Tangram (Fig 5E). To further evaluate the end-to-end performance of Cell2Spatial on high-resolution data, we used mouse brain Visium HD data comprising 492,460 spots at 8 µm resolution, representing SC-scale granularity (S1 Table). For computational efficiency and memory optimization, the dataset was down-sampled to 50,000 spots. A synthetic spatial dataset (ST) was constructed from the expression matrix and spatial coordinates of these spots, while a corresponding SC dataset was generated from the same expression matrix with cluster identities assigned through Seurat-based clustering. The SC dataset was mapped back to the synthetic ST dataset to assess mapping accuracy, defined as the proportion of cells in each cluster correctly positioned at their original spatial locations. The results revealed that Cell2Spatial effectively reconstructed the spatial distribution of cells, closely matching the ground truth (Fig 5F). Comparative analysis with CytoSPACE and Tangram demonstrated that Cell2Spatial achieved higher mapping accuracy on the Visium HD platform (Fig 5G; Average accuracies: Cell2Spatial (0.49), CytoSPACE (0.30), Tangram (0.19)) under end-to-end mapping conditions. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. Application of Cell2Spatial to 10× Genomics Xenium mouse brain data. (A) Spatial architecture of the mouse brain reconstructed using Cell2Spatial, CytoSPACE, and Tangram. Ground Truth (derived by RCTD [12] tool) depicts the distribution of different cell types in the reference space. Each point represents an individual cell, with distinct colors indicating various cell types. (B) Distributions of VLMC, L2/L3 IT, L4, L5 PT, L6 CT, L6b, and Oligo cells from the outer boundary to the inner region of the mouse brain for Ground Truth, Cell2Spatial, CytoSPACE, and Tangram, respectively. Each point denotes a single cell. (C) Grouped bar plot illustrating the accuracy of different cell type distributions in the Mouse Brain reconstructed by Cell2Spatial, CytoSPACE, and Tangram. Accuracy was calculated as the number of correctly mapped cells for a specific cell type divided by the total number of that cell type in the Ground Truth. (D) Heatmaps showing the expression distribution of the top 10 over-expressed genes for VLMC, L2/L3 IT, L4, L5 PT, L6 CT, L6b, and Oligo cell types in the 10× Genomics Xenium spatial data. Colors in the heatmap represent the expression levels. (E) Scatter plots demonstrate the consistency of signature scores for different cell types between the spatial architectures reconstructed by Cell2Spatial, CytoSPACE, and Tangram, compared to the Ground Truth. Each point represents a single spot. R denotes the Pearson correlation coefficient, and p-values were derived from two-sided t-tests. (F) Evaluation of Cell2Spatial end-to-end performance using Visium HD spatial data (S1 Table) from the mouse brain. The left panel shows the true spatial distribution of cells, while the right panel presents the reconstructed distribution by Cell2Spatial. Panels C3 and C4 illustrate the spatial distribution of two selected single-cell clusters. Each point represents an individual cell. (G) Boxplot showing the accuracy of various mapping tools in reconstructing spatial organization using mouse brain Visium HD data. Accuracy is measured as the proportion of cells within each single-cell cluster accurately mapped to their original spatial positions. P-values were derived from two-sided t-tests. The data underlying this figure can be found at https://doi.org/10.5281/zenodo.17212677. https://doi.org/10.1371/journal.pbio.3003477.g005 Slide-seqV2 achieves close to SC resolution, albeit with lower transcript capture rates. Utilizing an existing mouse hippocampus SC RNA-seq dataset generated by Saunders and colleagues [42], encompassing 16 cell types with 52,846 cells, we reconstructed the spatial architecture of the Slide-seq V2 mouse hippocampus dataset, which consist of 53,173 spots and 23,264 genes (S1 Table). Within this dataset, where 98.19% of gene expression matrix entries are 0 and the median UMI counts per location hover around 302, with a minimum of only 10, Cell2Spatial has effectively reconstructed the spatial organization of the mouse hippocampus, accurately delineating regions like the entorhinal cortex, CA principal cells, dentate principal cells, oligodendrocytes, astrocytes, and ependymal areas, aligning with spatial-based RCTD annotations (i.e., ground truth) (Fig 6A and 6B). While CytoSPACE also displayed promising reconstruction abilities, it notably misallocated numerous entorhinal cortex cells to the CA principal cells region. Conversely, Tangram exhibited a disorderly distribution pattern (Fig 6A and 6B). Further quantitative analysis confirmed Cell2Spatial performed better overall compared to CytoSPACE and Tangram (Fig 6C). Additionally, in another Slide-seq V2 dataset comprising 11,626 positions of mouse cerebellum (S1 Table), with 98.55% of gene expression matrix entries being zero and median UMI counts per location approximately 30,014, Cell2Spatial accurately assigned cell type labels and captured the multi-layered structure of the cerebellum, such as Bergmann, Granule, and Purkinje, consistent with RCTD annotations (Fig 6D–6F). The analysis of cellular compositions in reconstructed spatial structures demonstrated Cell2Spatial’s high consistency with RCTD-derived cellular compositions (R > 0.97), while CytoSPACE and Tangram lagged significantly behind, underscoring their inadequacy for low-capture-rate spatial data (Fig 6G and 6H). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 6. Application of Cell2Spatial to Slide-seq V2 hippocampus and cerebellum data. (A) Spatial architecture of the mouse hippocampus reconstructed using Cell2Spatial, CytoSPACE, and Tangram. Ground Truth (derived by RCTD [12] tool) represents the distribution of different cell types in the reference space. Each point represents a single cell, with different colors indicating different cell types. (B) Distributions of Entorhinal cortex, CA Principal cells, Dentate Principal cells, Oligodendrocytes, Astrocytes, and Ependymal cells for Ground Truth, Cell2Spatial, CytoSPACE, and Tangram. Each point represents a single cell. (C) Grouped bar plot showing the accuracy of cell type distributions in the hippocampus reconstructed by Cell2Spatial, CytoSPACE, and Tangram. Accuracy was calculated as the number of correctly mapped cells for a specific cell type divided by the total number of that cell type in the Ground Truth. (D) Spatial architectures of the cerebellum reconstructed using Cell2Spatial, CytoSPACE, and Tangram. Ground Truth represents the distribution of different cell types in the reference space. Each point represents a single cell, with distinct colors indicating different cell types. (E) Distributions of Bergmann cells, Granule cells, Purkinje cells, Oligodendrocytes, Astrocytes, and Fibroblasts for Ground Truth, Cell2Spatial, CytoSPACE, and Tangram. Each point represents a single cell. (F) Grouped bar plot showing the accuracy of different cell type distributions in the cerebellum reconstructed by Cell2Spatial, CytoSPACE, and Tangram. Accuracy was calculated as the number of correctly mapped cells for a specific cell type divided by the total number of that cell type in Ground Truth. (G, H) Scatter plots showing the consistency between the spatial cellular compositions recovered by Cell2Spatial, CytoSPACE, and Tangram, compared to Ground Truth for both the hippocampus (G) and cerebellum (H). Each point represents a cell type. R indicating the Pearson correlation coefficient (PCC), with p-values obtained from two-sided t-tests, following a t-distribution to account for smaller sample sizes. The data underlying this figure can be found at https://doi.org/10.5281/zenodo.17212677. https://doi.org/10.1371/journal.pbio.3003477.g006 Overall, these findings strongly indicate that Cell2Spatial remains robust in its performance even when processing high-resolution spatial data, even with lower transcription capture rates. Diverse applications of Cell2Spatial To investigate the potential applications of Cell2Spatial, we utilized it to analyze ST data (comprising 1,438 spots) of mouse kidney tissue, aiming to explore the spatial co-existences among various cell types and the temporal differentiation of crucial cell types within the spatial framework. By assigning single cells from the mouse kidney (12,987 cells encompassing 13 cell types) reported by Ransick and colleagues [36] to spots, as illustrated in Fig 7A, distinct spatial distributions among different cell types were observed (Figs 7A, S8A, and S8B and S1 Table). To quantify these relationships, we introduced a co-existence index, and the results revealed robust spatial co-localization links between T cells, fibroblasts, and natural killer (NK) cells (Fig 7B and S8 Table; see “Materials and methods”). To further verify these findings, we highlighted the expression of their respective marker genes (Cd3d, Col3a1, and Nkg7) in the ST data of mouse kidney, which showed a significant overlap in the spatial distribution of these three cell types, confirming the results of Cell2Spatial (S8C Fig). Additionally, we elucidated the differentiation timelines of proximal tubular and distal tubular cells through pseudo-time analysis using Monocle2 [43–45]. As shown in Fig 7C and 7D, proximal tubules exhibited an outward-to-inward differentiation trend, while distal tubules displayed an inward-to-outward progression (Fig 7C and 7D). The opposing timelines align with the findings of Wei and colleagues [3]. Thus, Cell2Spatial can accurately track the timelines of cell differentiation. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 7. Cell2Spatial enables versatile applications across various analytical domains. (A) Left: Spatial architecture of the mouse kidney depicted by Cell2Spatial. Each dot represents an individual cell, with cell types are color-coded. Right: Uniform Manifold Approximation and Projection (UMAP) plot showing the single-cell atlas of mouse kidney. (B) Heat map showing the coexistence of mapped cell types in spatial spots. The Pearson correlation coefficient (PCC) is obtained based on coexistence index (see “Materials and methods”). (C) Trajectory analysis of proximal tubule cells (left), with spatial mapping of the pseudo-time values in the tissue section (right). (D) Trajectory analysis of distal tubule cells (left), with spatial mapping of the pseudo-time values in the tissue section (right). (E) Left: Renal cell cancer (RCC) tissue section displaying spot clustering determined using the Seurat common processing strategy. Right: Pre-defined spatial coordinates of the tertiary lymphoid structure (TLS) for reference. (F) Heat map showing the relationship of inclusion between TLS and spatial clusters. The intensity of color indicates the number of TLS spots located in the cluster. (G) Left: Spatial architecture of the RCC tissue section reconstructed using Cell2Spatial. Each dot represents an individual cell and cell types were color-coded. Right: Enlarged cell atlas located within the TLS region. (H) Overview of cell types co-existence in the RCC tissue section. Edge thickness reflects the strength of the co-existence index, with edges below a specified cutoff (0.05) excluded. (I) Circle plots showing the interactions between B cells and other cell types within an independent spot cluster. The thickness of the edges indicates the strength of interactions. Left: Cluster 3; Right: Cluster 4. (J) Circle plots depicting the interactions between B cells and other cell types within an independent spot cluster. Left: Cluster 2; Right: Cluster 6. The data underlying this figure can be found at https://doi.org/10.5281/zenodo.17212677. https://doi.org/10.1371/journal.pbio.3003477.g007 To determine whether Cell2Spatial can detect critical structural regions within tissues, ST data with 3,206 spots concerning tertiary lymphoid structures (TLS) associated with renal cell cancer (RCC) from the study published by Meylan and colleagues [46] were acquired (Fig 7E and S1 Table). TLSs are closely associated with tumor development and metastasis, carrying diagnostic and therapeutic significance in cancer. In this dataset, the precise coordinates of TLS are well-documented, with a notable concentration in Cluster 3 and Cluster 4 (Fig 7E and 7F). By superimposing an immune SC atlas (down-sampled to 24,834 cells covering 25 major immune cell types) [47] onto the ST section of RCC, a diverse array of immune cell types within the TLS were observed (Fig 7G). Spatial cell co-existence analysis revealed that B cells were the major interacting cell type (Fig 7H and S9 Table), and had the most substantial and abundant interactions with other immune cells in Cluster 4, closely followed by Cluster 3 (Fig 7I and 7J). These findings suggest the presence of a functionally significant structure (Clusters 3 and 4) in RCC tissue, in alignment with previously documented structures. Taken together, Cell2Spatial can used to explore functional and structural regions within tissues, and the intricate network of interactions between diverse cell types and their spatial arrangement. Discussion This study underscores the utility of Cell2Spatial as a powerful tool for mapping individual cells from SC datasets to spatial spots within ST data, simplifying the generation of spatial cellular maps. In contrast to traditional ST deconvolution methods [10,12,15], Cell2Spatial enables a detailed investigation of each cell’s specific characteristics in relation to spatial spots. While Cell2Spatial shares similarities with established spatial mapping methods like CytoSPACE [20], CellTrek [3], Tangram [19], and Seurat [18], we emphasize its unique attributes. Cell2Spatial utilizes a spatially weighted maximum likelihood model to assess the similarity between cells and spatial spots, ensuring accurate representation of spatial relationships within tissues and enhancing the detection of cellular heterogeneity. When cell types in SC and ST datasets are not fully congruent, Cell2Spatial employs a hotspot detection strategy to detect the spatial distribution of specific cell types. This approach facilitates the exclusion of single cells corresponding to absent cell types in the ST dataset, thereby improving the reliability of spatial mapping. By selecting “cell-count-regulated genes,” Cell2Spatial accurately estimates cell numbers in each spot, enabling precise reconstruction of cell type composition within spatial contexts. Additionally, Cell2Spatial employs a FNN to preassign cells to spatial clusters based on a joint embedding of SC and ST data. This methodology optimizes computational efficiency and memory usage, significantly enhancing overall performance. Cell2Spatial thus effectively reconstructs the spatial architecture of tissue slices from SC data, and addresses the complexities of spatial mapping between SC and ST datasets when cell types are not perfectly aligned. According to performance evaluations that encompassing a range of tissue ST data types from the 10× Visium platform with relatively low resolution in spot, Cell2Spatial demonstrated good performance in accurately recovering spatial architectures, gene expression, and cellular compositions, outperforming the capabilities of existing tools. Notably, for thymus spatial data, Cell2Spatial effectively reconstructed developmental pathways, elucidated biological functions, and provided insights into structural positioning. Moreover, when utilizing high-resolution ST data (approaching SC or subcellular levels), including 10× Xenium and Slide-seq V2, Cell2Spatial exhibited outstanding capabilities in reconstructing spatial architectural structures. Even in ST data with lower transcript capture rates, Cell2Spatial maintains robustness and accuracy, outperforming CytoSPACE and Tangram. Particularly, its performance in analyzing mouse kidney tissue emphasizing its strong capacity to track the temporal evolution of key cell types. By overlaying an immune SC atlas with RCC ST data, Cell2Spatial unveiled intricate interactions among distinct cell types and identified disease-related regions, i.e., TLS. Overall, through reconstructing tissue spatial architectures at SC resolution, Cell2Spatial provides enhanced biological insights, such as differential expression analysis, developmental trajectories, cell–cell interactions, and complex spatial gene expression patterns. Although our study primarily concentrated on the utilization of SC data to investigate potential spatial architectural structures within 10× Visium, Xenium, HD Visium, and Slide-seq V2 ST data, Cell2Spatial is a reference-based method that can be applied to diverse spatial data from various sources, such as MERSCOPE [48], SMART-Seq2 [49], and Drop-Seq [50]. It’s essential to note that while using Cell2Spatial, its performance may be influenced by several underlying factors. Firstly, Cell2Spatial relies on specificity entropy to identify representative gene sets specific to cell types, which requires well-annotated or clustered SC data for accurate identification. Secondly, measuring similarity between individual cells and spatial spots may introduce biases, especially for cell types with similar evolutionary lineages, leading to discrepancies between their mapped spatial positions and true distributions. Moreover, RNA spread in ST sequencing may cause uncorrected quantitative biases in the ST data. Specifically, Cell2Spatial shows diminished performance in end-to-end mapping, likely due to insufficient cell type-specific markers to effectively differentiate variations among cells of the same type. Additionally, the limitations of the Linear Assignment Algorithm in capturing non-linear relationships between single cells and spatial spots contribute to this performance drop. For sub-cellular resolution datasets, integrating high-resolution tissue imaging with DAPI staining for accurate segmentation could improve mapping precision. Moreover, developing algorithms that exploit sub-cellular resolution transcriptomic signals would enable a more detailed exploration of spatial heterogeneity within individual cells, further enhancing Cell2Spatial’s capabilities. These limitations highlight an area for improvement in future research. Despite these challenges, Cell2Spatial has the potential to become a valuable tool for biological research, advancing the study of cellular and tissue spatial structures in the context of human diseases. Materials and methods 1. Cell2Spatial analytical framework Data collection and normalization. The SC reference expression profile is assumed to be an N × C matrix, with N representing genes and C representing cells. The dataset is well-annotated (or clustered) into k distinct cell types (or clusters). Any cell type with fewer than a predefined threshold of cells (default: 5 cells) was excluded from the SC dataset. The spatial transcriptome (ST) expression profile is an M × S matrix, where M represents genes and S represents spots. A total of G genes is shared between the SC and ST. Data normalization was achieved by applying Seurat’s SCTransform method to both the SC (X → G × C) and ST data (Q → G × S). Cell-type-specific genes. Given that most genes cannot effectively distinguish among cell types, specific genes corresponding to distinct cell types were identified with computational efficiency using a specificity metric inspired by Shannon entropy structure (S9A Fig). Initially, the SC profile X was used to generate the average expression matrix X* (X* → G × K), with rows representing genes and columns representing cell types. Each entry in represents the mean expressions level of gene i in cell type j. Gene specificity scores were then computed using the following formula: (1) Here, represents the value characterizing the cell-type specificity of gene i (higher values indicate greater gene specificity), represents the mean expression of gene i across all cell types. The summation yields zero if the gene i is uniformly expressed across cell types, and reaches a maximum value of when it is exclusively expressed in a single cell type. The weight for gene i, was calculated as: (2) Thus, the gene weights are directly proportional to the expression level and serve as a key factor in determining the specificity values. Here, denotes the maximum average expression of gene i across all cell types, represents the smallest of these maximum average expression levels across all genes, and is overall maximum element in the matrix , encompassing all genes and all cell types. Each gene was initially assigned to the cell type in which it showed the highest average expression. For each cell type, we then calculated the expression variance of all genes across individual cells. Among the genes assigned to that cell type, those with variances exceeding the mean variance (calculated across all genes) by more than three standard deviations were excluded, to eliminate genes with unusually high within-type variability. For each cell type, the remaining assigned genes that passed the variance filter were ranked based on their specificity scores, and the top 100 genes were selected by default. This resulted in a gene set , where represents the set of specific genes for cell type j. The total number of selected genes, denoted as m (, as some cell types may contain fewer than 100 qualified genes after filtering). Measuring similarity between single cells and spots. To gauge the similarity between single cells and spatial spots, the specific gene list was initially converted into a vector and duplicates were removed, resulting in a length of m as previously defined. Then, the probability of specific gene i for each cell (i.e., ), was estimated using after applying Laplace Smoothing (by adding 1e-6 to account for zero values). The formula used was: , ensuring that . Based on the multinomial assumption, the likelihood function for a spot t with profile to cell was calculated as , and can be further described as follows: (3) Given that the geometric spatial distance can provide varying viewpoints to reflect the association between single cells and spots, Seurat’s CCA strategy was used for the integration of SC and ST datasets. Subsequently, UMAP was applied to obtain a two-dimensional projection of the integrated data. Although UMAP’s embeddings may distort global distances to enhance visualization, its local structure preservation ensures a meaningful representation of the relationships between a spatial spot and its neighboring cells, helping to characterize tissue architecture and cellular distributions [51] (S9B Fig; performance assessment of distance weighting versus no weighting for Cell2Spatial). Thus, the Euclidean distance between each spot and single cell was calculated, resulting in a distance matrix composed of S spots and C cells. A scaled distance-based weight matrix was then derived using the following formula: (4) where is a vector representing the distances between spot t and all single cells in UMAP coordinates, denotes the quantile value of the distance vector at the specifies probability level (with probs constrained between 0.01 and 1, defaulting to 0.3) (S9C Fig). Notably, if spot t is closer to cell in the UMAP projection, the weight is higher. If is below 0, it is re-set to 0. To combine these distance weights further, the likelihood similarity incorporating spatial distance-based weights was calculated as follows: (5) Here, is the original likelihood matrix representing the similarity between spots and cells, and is the final similarity matrix after incorporating spatial information. is the vector of observed likelihoods for spot t across all cells. is the corresponding normalized spatial distance weight matrix of the same dimensions, and “” indicates element-wise multiplication. Determination of spatial hotspot regions of cell types. To detect spatial hotspots for each cell type, the Getis-Ord G* index was employed, involving several steps: Initially, spatial coordinates of spots within the tissue sample were extracted from the ST Seurat object. Then, a neighborhood structure was established using k-nearest neighbors, where each spot’s nearest neighbors (default: 5 spots) were determined to form its neighborhood set. Next, the AddModuleScore function from Seurat was used to calculate signature scores for spots based on the gene sets specific to that cell type in . Finally, the statistic G* was calculated for each spot using the following formula: (6) where is the G* index of spot t in cell type j, represents the vector of signature scores for cell type j across all the spots, denotes the score for spot , is the global mean score of the ST section. is a binary weight indicating whether spot is a neighbor of spot t (1 if true, 0 otherwise), is the standard deviation of the signature scores in , and S represents the total number of spots as previously defined. represents the difference between the signature scores of neighboring spots and the global mean . Denominator normalizes the numerator by adjusting for the scale of the weighted differences. The standard G* is essentially a z-score that measures how the neighborhoods of spot t deviate from the context mean score [52]. Therefore, the p-value for each statistic of cell type j was computed as: (7) where is the cumulative distribution function of the standard normal distribution (S9D Fig). Spot t was classified as a hotspot for cell type j if (default: ). The processes mentioned above were applied to all spots, resulting in a Boolean matrix with dimensions S (spots) by K (cell types). Each entry in this matrix indicates whether a given spot is a hotspot (1) or not (0) for a specific cell type. This yields a list of detected hotspots for each cell type, denoted as , with a unique length of . represents the set of detected hotspots belonging to cell type j. Clustering spots of ST data. Following the standard Seurat [53] workflow for clustering ST data, the SCTransform function was used for normalization and variance stabilization. SCTransform computes Pearson residuals for each gene, reflecting deviations from expected expression based on a regularized negative binomial model. Genes with high variance in these residuals are considered highly variable. By default, the top 3,000 HVGs were used for clustering, which was performed with the resolution parameter set to 0.2 (as default). The obtained ST clusters were defined as a list of set , where represents spots belonging to the -th cluster. Estimation of cellular compositions in spatial section. To examine the cellular composition within a spatial section, SC expression profile was used as a reference, with m cell-type-specific genes in serving as distinctive features. This process generated a base matrix Bm×K, where rows represent genes, columns represent cell types, and each entry represents the average expression of a specific gene in the corresponding cell type. A reference-based deconvolution strategy, inspired by the LinDeconSeq tool [21,54], was established using a RLM to predict the relative prevalence of distinct cell types within each spot. To enhance estimation accuracy across the entire ST section, the expression vector of spot t was smoothed by the averaged expression of the six (as default) nearest neighbors within the same ST-defined cluster, including spot t itself, resulting in an updated vector . This smoothing strategy leverages the spatial coherence of gene expression and cell type composition within local regions, while mitigating estimation bias caused by uneven cell numbers across spots [55,56] (S9E Fig). Then, incorporating RLM to link and Bm×K allows for the estimation of cellular proportions of spot t by minimizing the squared discrepancies between observed data. (8) where is the non-negative cellular proportion vector to be estimated for spot t, with dimension , and . To estimate the overall cell type composition of the ST tissue section, cellular proportions are typically aggregated across all spatial spots. Since spots within the same spatial cluster tend to share similar cellular profiles, reference-based deconvolution methods may assign low, non-zero proportions to cell types that are genuinely absent in certain clusters. While these small residual estimates are individually minor, their cumulative effect across multiple spots can lead to systematic biases in the global cell type composition inference. To address this, a correction factor based on the hotspot matrix was introduced, calculated as follows: (9) Here, represents the r-th cluster to which spot i belongs. The expression is a vector of length K that reflects the number of times each cell type significantly appears in the cluster . This vector then was rescaled to a range of 0–1 using the rescale function (i.e., Min-max normalization) from “scales” R package (version 1.3.0). Each element of V is then binarized using a predefined threshold τ (default 0.01) as follows: (10) The resulting matrix V is a wrapped correction factor matrix with dimensions ( represents the number of spot clusters as previously defined). Therefore, the cellular composition can be adjusted using the following formula: (11) represents the correction factors of cluster . denotes the transpose of the r-th row vector of matrix V, allowing for a valid dot production. The cellular proportions of the spots in (excluding spots not classified as hotspot for any cell type) are organized into a matrix , where each row represents the adjusted compositions of one spot. Finally, the cellular compositions for the entire ST slice were derived by following formula. (12) is the final estimated cellular proportion vector of length K for the entire ST section (S9F Fig). Estimation of cell counts per spot for low-resolution spatial data using a corrected saturation model. To infer the number of cells per spot in low-resolution ST data, we applied a corrected saturation model that accounts for spot-level library size and gene count, assuming that the number of detected genes per spot follows a saturation curve as cell numbers increase. Specifically, the spatial expression matrix records UMI counts for G genes across spots, where represents the UMI count of gene i in spot t. To reduce noise, genes with low expression were excluded by applying a spot-level threshold (e.g., UMI ≥ 2), retaining only those with reliable signals. For each spot t, the library size is defined as , and the gene count is calculated as , where is an indicator function. Assuming the maximum number of cells among spots is (e.g., 10 for 10× Visium), we modeled the relationship among the gene count , cell number , and library size using a corrected saturation model: . is the maximum observed gene count across all spots, and represents the average number of genes per cell. To estimate , we initially computed a composite score for each spot to balance library size and gene count: (13) where and are the weights for library size and gene count (default: 0.5). The reference spot was estimated from the mean of spots exceeding the quantile cutoff of the composite score (90% by default) and was assumed to contain cells. The corresponding reference library size and gene count were then used to compute g: i.e., . Subsequently, the cell number for spot t was obtained by solving: (14) To ensure consistency, we incorporated the library size constraint , where . Finally, we calculated the harmonic mean of the and , and rounded to the nearest integer. i.e., . is a vector of cell counts across all spots. To maintain biological plausibility and adhere to platform resolution, the was constrained to the range [1, ]. Notably, for high-resolution spatial data at the SC level, set “max.cells.in.spot = 1” (or “platform.res = High”) to ensure one cell per spot. For subcellular-resolution platforms, such as STOMICS and Visium HD, adjust the bin size to approximate the dimensions of a single cell (e.g., for Visium HD, set bin.size = 8 µm). Assigning single cells to spatial spots. Mapping single cells to spatial spots enables the reconstruction of spatial architecture at SC resolution for low-resolution data and high gene capture expression patterns for high-resolution data. The mapping process involves the following steps: (1) Update SC Seurat object: Using the previously mentioned cellular fractions and spot cell , the number of cells in each cell type was calculated as . For the j-th cell type, cells were down-sampled from the corresponding cell type in the SC data X. If the cell counts of the j-the cell type in X was less than , sampling with replacement was applied; otherwise, sampling without replacement. Gaussian noise (mean = 0, sd = 1) was added to the gene expression of duplicated cells to avoid a singular matrix. The updated SC data denoted as X’ → G × C*, where . (2) Neural network design for pre-assigned individual cells to ST clusters: Recognizing the high algorithmic complexity involved in large-scale SC allocation, global cell-to-spot mapping becomes computationally intensive and less scalable. To address this, we introduced the FNN framework to preassign single cells mentioned in step (1) to ST clusters, effectively decomposing the original global mapping task into multiple subspace mapping problems. Specifically, reference-based coembedding approach from the Seurat package was used with SC data ( → G × C*) as the reference and the ST data (→ G × S’) as the query. From the coembedded data, a four-layer neural network was trained. The input layer comprised principal components (PCs)-embedded ST data (default 30 PCs), followed by a fully connected hidden layer with 64 neurons and a rectified linear unit (ReLU) activation function. The subsequent two hidden layers consisted of 128 units each. To address overfitting and enhance computational efficiency, dropout (rate = 0.2), batch normalization, and L2 regularization techniques were applied at each hidden layer. The output layer was based on the softmax function and the number of neurons in the aligned with ST clusters. The FNN’s learning rate was set to 0.000001. The involved major processes as follows: (15) represents the input layer of PCs of ST data. Notably, at the start of training, cells were randomly divided into a training set (95%) and a test set (5%). The test set was utilized at each epoch to evaluate the model’s performance and generalization capability. The training procedure adopted a 5-fold cross-validation approach, aiming to minimized the cross-entropy loss, which evaluates the similarity between predicted and actual categories. (16) is a binary indicator if ST cluster r is correct label for spot t, is the predicted probability of observation spot t for cluster r. is the regularization parameter controlling the impact of the regularization term on the loss function, with a default value of 0.01. After completing the training of the FNN model, the co-embedded SC data was input into the model to determine the probability of each cell belonging to specific ST cluster. The probability matrix is denoted as → C* × R. Given the pre-estimated cell counts within each spot, the cell allocation requirement for a specific cluster is determined by summing the estimated cells across all spots within that cluster. However, the FNN’s allocation of single cells to ST clusters may lead to imbalances, resulting in predicted cell counts that deviate from the expected values. To address this issue, an iterative optimization strategy based on the probability matrix is introduced, as illustrated in S9G Fig. This results in the creation of a list container : (17) is a set comprising the cells that pre-assigned to ST cluster . (3) Assigning single cells to spots using linear assignment strategy: Based on the cell counts within each spot, container , and the weighted likelihood similarity matrix , which is a subset of and the spots in rows consisted only of highlighted spots in . The columns in represent the cells source from cells , with each column’s values taken from corresponding to the same cell names. To minimize the spread of mapped single cells from specific cell type to regions they do not belong to, was adjusted as follows: (18) is a binary (0–1) matrix, where each column represents a binary vector of length . For each cell associated with cell type , is defined as a subset of values extracted from the corresponding column in the binary matrix , retaining only the entries corresponding to the hotspots in . The Jonker-Volgenant algorithm [23] was then employed to allocate single cells to specific spots to identify the ideal cell-to-spot assignment that maximize the following linear cost function (S9H Fig). For cluster : (19) where is a subset of , with rows corresponding to spots from , columns represent cells from . A is a Boolean matrix, where if single cell assigned to spot t. The constraint mentioned above guarantees that each cell is assigned only to one spot. The optimization is performed through the utilization of Python’s lapjv package (version 1.3.24) [20,57]. These processes are iteratively applied to other clusters. (4) Generating spatial coordinates for mapped single cells: To represent SC coordinates within spatial slices, the minimum Euclidean distance between adjacent spots was calculated. Center coordinates for each spot were extracted, and random coordinates for assigned single cells were generated using the spatstat.random::runifdisc function, with the radius set to half of this minimum distance. This method established the spatial positions of all single cells. The mapping results were subsequently integrated into a “Seurat” object and provided to the user. Notably, for small-scale datasets, the FNN pre-assignment strategy can be omitted, with a linear assignment algorithm applied directly. 2. Benchmarking and simulations Simulation of spatial transcriptome data. To assess the precision and robustness of Cell2Spatial, simulated ST datasets were generated using the Allen mouse brain SC dataset with brain ST data as the reference template. For training, 50% of cells from each type were randomly selected, and gene expression perturbations (0%, 5%, 10%, 15%, and 20%) were introduced by shuffling expression values across cells to simulate noise. The top 2,000 HVGs were identified using the FindVariableFeatures function in Seurat, and PCCs were calculated to construct a cell–spot correlation matrix. Random spot-level cell counts were drawn from a Poisson distribution (Lambda = 5), and cells were assigned to spots according to the correlation matrix and sampled counts. The expression profile of each spot was then obtained by summing the expression values of its assigned single cells. For each simulated dataset, the number of cells, cell types, and spatial coordinates per spot were recorded for downstream evaluation. To further examine deconvolution performance and gene expression recovery, additional synthetic ST datasets were created with Poisson-distributed cell counts using Lambda values of 5, 10, 15, and 20. Benchmark metrics. Accuracy index: Various spatial mapping tools (Cell2Spatial, CytoSPACE, CellTrek, Tangram, and Seurat) were used to project the remaining single cells of Allen mouse brain onto simulated ST sections with different levels of noise (Fig 2G). The performance of these tools was evaluated using the following formula. (20) [range: 0–1] represents the accuracy metric of cell type j for effectively mapping to the simulated spots, which integrates both the relative abundance of cells in spatial spots and the RMSE between predicted and true counts. denotes the true count for cell type j at spatial spot t, while represents the predicted count for the same type at spot t. S refers to the total number of spots in the ST slice. is the RMSE between the predicted and true counts for cell type j across all spots. Highly variable genes (HVGs) conservation: The HVGs conservation score is a critical metric for assessing the preservation of biological signals. To compute this score, Seurat’s FindVariableFeatures function was utilized to identify HVGs in each ST dataset, both before and after mapping. For each batch, the target was to identify 500 HVGs. When evaluating the impact of the “group.size” parameter on mapping performance, the number of HVGs was set to 2,000. The HVG conservation score is defined as: (21) Jaccard index: To evaluate the consistency between the mapped ST data and the synthetic ST data, the Jaccard index was used. The index, calculated at the cell type or SC level for each spot, is defined as: (22) Here, denotes the cell types or cells within the mapped spot, refers to those within the corresponding spot in the synthetic dataset. Pearson correlation coefficients (PCCs) and root mean square errors (RMSEs): Spatial deconvolution performance was compared using PCC and RMSE metrics. (1) For spatial mapping tools, cell type proportions in each spot were calculated based on the spatial distribution of assigned single cells. (2) For spatial deconvolution tools, proportions were inferred from the ST data using the corresponding SC dataset as a reference. Overall cell type composition in the ST slices was estimated by randomly selecting 2,000 spots, and the composition was computed as follows: (23) where k denotes the number of cell types, t represents the index of spot. PCC and RMSE were then used to evaluate the concordance between synthetic and predicted compositions at the overall cellular composition level of the ST section. This procedure was repeated 100 times to generate error bars. Cosine similarity of gene expression: Expression consistency between the mapped ST data and the synthetic ST data was evaluated by aggregating the profiles for each spot in the mapped ST data. This was done by summing the expression values of each gene across the single cells assigned to that spot, resulting in the recovered ST data. Consistency was then assessed by calculating the cosine similarity using 2,000 HVGs between the recovered and synthetic ST spots. Kullback–Leibler (KL) divergence: To assess the global similarity of spatial structures, a KL-divergence-based approach from the CellTrek (version 0.0.94) package was utilized. The CellTrek::SColoc function computed a 2D grid kernel density for each cell type using kde2d from the “MASS” package (version 7.3–58.3), with a default bandwidth (h = 100) corresponding to the spatial distance between adjacent ST spots and a grid size (n) of 25. The KL-divergence was then calculated for the 2D density distributions between each pair of cell types from the mapped and synthetic data, with the synthetic data utilizing advanced SC information to achieve SC resolution. 3. Comparison of Cell2Spatial with publicly available spatial tools Cell2Spatial. Cell2Spatial primarily employs default parameters for spot segmentation of both real and simulated 10× Visium ST data. Key parameters used include: normalize.method = ‘SCTransform’, fix.cells.in.spot = TRUE, knn.spots = 5, and marker.selection = ‘shannon’. For 10× Xenium, Visium HD, and Slide-seq V2 high-resolution ST data, the maximum cell count per spot was set to 1. CytoSPACE. CytoSPACE [20] assigns individual cells to precise spatial spots by optimizing a correlation-based cost function, taking into account the estimated cell count per spot. This optimization is carried out using a shortest augmenting path algorithm. The “duplicated cells” option and “lapjv” solver were used during the CytoSPACE execution, and a standardized mean cell count of 5 per spot was maintained across all 10× Visium samples. For 10× Xenium, Visium HD, and Slide-seq V2 high-resolution ST data, the mean cell count was set to 2 for successful running. CellTrek. CellTrek utilizes ST data to train a random forest model for predicting spatial coordinates, leveraging dimension reduction features shared with SC data [3]. The traint function from the “CellTrek” R package (version 0.0.94) was to obtain the co-embedding of ST and SC data for Visium samples. Subsequently, the celltrek function with default parameters (“reduction = ‘pca’, intp = T, intp_pnt = 10,000, intp_lin = F, nPCs = 30, ntree = 1,000, dist_thresh = 0.4, top_spot = 10, spot_n = 10, repel_r = 5, repel_iter = 10, keep_model = T) was used to project individual cells onto the ST coordinates. Cells were then assigned to their nearest spots based on the Euclidean distance. Notably, CellTrek failed to execute successfully for 10× Xenium, Visium HD, and Slide-seq V2 high-resolution data, thus it was excluded from the assessment and comparison of high-resolution ST data. Tangram. Tangram [19] is a specialized deep-learning framework designed to reveal intricate spatial structures at the SC level. The “Seurat” object was converted into an “h5ad” file for all Visium samples, and Tangram was applied with default settings. The individual cells were mapped to ST spots by utilizing all accessible genes for each cell. To visualize these single cells on ST sections, random coordinates were generated for single cells based on their association with the respective spots. Seurat. Seurat [18] employs a label transfer approach to map ST spots onto individual cell type, and considers each spot as a unit at the SC granularity level. The FindTransferAnchors function in Seurat was used in this study, with ST data as the query and the corresponding SC data as the reference. Subsequently, the TransferData function was utilized to assign SC labels (i.e., cell types) to the ST spots. Default parameters were used as specified in the official documentation. Cell2location. The guidelines on the Cell2location website (https://cell2location.readthedocs.io/en/latest/notebooks/cell2location_tutorial.html) were followed. The SC regression model was trained with the parameters “max_epochs = 100” and “lr = 0.002”. The Cell2location model was then obtained using “max_epochs = 10,000”. SpatialDWLS. The guidelines on the SpatialDWLS website (https://rubd.github.io/Giotto_site/articles/tut7_giotto_enrichment.html) were followed, with the parameter set to “n_cell = 20”. RCTD. The guidelines on the RCTD GitHub repository (https://raw.githack.com/dmcable/spacexr/master/vignettes/spatial-transcriptomics.html) were followed, with the parameter set to “doublet_mode = full”. Stereoscope. The guidelines on the website (https://docs.scvi-tools.org/en/stable/user_guide/models/stereoscope.html) were followed. The SC model was trained with parameters “max_epochs = 50” and the spatial model was trained with parameters “max_epochs = 100”. DestVI. The guidelines on the DestVI website (https://docs.scvi-tools.org/en/stable/tutorials/notebooks/DestVI_tutorial.html) were followed. The SC model was trained with parameters “max_epochs = 50, lr = 0.001, number of training genes =2,000”. The spatial model was trained with parameters “max_epochs = 100”. SpaOTsc. The guidlines on the SpaOTsc GitHub repository (https://github.com/zcang/SpaOTsc) were followed. The spatial distribution of genes was obtained using the function ‘issc.transport_plan’ with parameters “alpha=0, rho=1.0, epsilon=0.1, scaling=False”. novoSpaRc. The guidelines on the GitHub repository of novoSpaRc (https://github.com/rajewsky-lab/novosparc/blob/master/reconstruct_drosophila_embryo_tutorial.ipynb) were followed, with the parameters set to “alpha_linear = 0.5, loss_fun = square_loss, epsilon = 5 × 10−3”. SPOTlight. The guidelines on the SPOTlight GitHub repository (https://marcelosua.github.io/SPOTlight/) were followed, with the parameter set to “transf = uv, method = nsNMF”. DSTG. The guidelines on the DSTG GitHub repository (https://github.com/Su-informatics-lab/DSTG) were followed. CARD. The guidelines on the CARD GitHub repository (https://yma-lab.github.io/CARD/documentation/04_CARD_Example.html) were followed, with the parameters set to “minCountGene = 100, minCountSpot = 5”. 4. Score-Guided Mapping Accuracy (SGMA) Given the lack of information on precise cell location in real ST data, a strategy called SGMA was devised to gauge the performance of the mapping tools. Briefly, Seurat’s FindAllMarker function (logfc.threshold = 0.25, min.pct = 0.25) was used to identify marker genes for each reference cell type within the corresponding SC dataset, and the top 20 overexpressed genes for each cell type were screened. The AddModuleScore function from Seurat was then applied in conjunction with the previously identified marker gene list to score the spots in the ST data. Thereafter, the scores of each cell type in spots were fitted to a Gaussian distribution, and the p-value for each score was calculated. Candidate spots potentially harboring a specific cell type were identified based on p-value ≤0.01 or 0.05. Finally, the performance of the mapping tool was quantified using the formula below: (24) represents the set of spots that the mapping tool assigns to cell type j, Sj represents the actual spots (determined by the Gaussian distribution mentioned above) where cell type j is located, signifies the count of shared spots between and , represents the total number of spots in the combined set of and , and k is the number of cell types. For a more intuitive comparison, accuracy metrics of various ST datasets were scaled to 0–1. 5. Cell counting in low-resolution spatial spots using DAPI from Visium fluorescence Cell counts for the spots were obtained using the DAPI channel of the Visium fluorescence image, the spatial dataset was processed using “Squidpy” python package (version 1.6.0) [29]. The image was first smoothed using the smooth method, followed by segmentation with the watershed algorithm. Segmentation features were then calculated and integrated into the associated “AnnData” object. The cell counts for each spot were derived by extracting the “segmentation_label” information from the “AnnData” object. 6. Memory usage and time consumption analysis To evaluate the memory usage and time consumption of Cell2Spatial, CytoSPACE, CellTrek, Tangram, and Seurat, 10 simulated ST datasets with 1,000–10,000 spots were generated using mouse brain data. The mapping programs were then executed with the corresponding SC data. Memory usage and time consumption were assessed using the “peakRAM” R package (version 1.0.2) [58]. 7. Gene set enrichment analysis (GSEA) GSEA was performed to determine whether single cells mapped onto thymic spatial sections by Cell2Spatial can provide insights into critical functional regions. Firstly, the single cells in the medullary region were compared with those in other areas using Seurat’s FindAllMarkers function (Wilcoxon rank sum test). The functional gene sets were obtained from the C5 ontology gene sets within the MsigDB database. Subsequently, the “clusterProfiler” package (version 4.0.5) [44] and bitr function were used to convert gene symbols into Entrez IDs. The genes were sorted in descending order based on the fold change in expression. GSEA function was utilized with the “TERM2GENE = C5” parameter for enrichment analysis, and terms with adjusted p-values below 0.01 were considered significantly enriched. Finally, the gseaplot2 function from the “enrichplot” package (version 1.18.4) was used to visualize the selected terms. 8. Spatial co-existence analysis of cell types in low-resolution spatial data using Cell2Spatial The spatial co-existence of the different cell types was determined by projecting individual cells onto ST sections. Following the mapping of single cells to their respective spatial spots, the spatial co-localization among the various cell types was quantified using the formula as follows: (25) |Si| and |Sj| represent the number of spots containing cell types i and j, respectively, |Sij| denotes the number of spots containing both cell type i and j, and Eij is the coexistence index for cell types i and j (ranging from 0 to 1). Eij = 1 signifies complete co-localization of the cell types in the ST sections, and Eij = 0 indicates a lack of co-localization. 9. Statistical analysis Pearson and Spearman correlation analysis, Student t test, Wilcoxon rank sum test, and Kruskal–Wallis test were performed as appropriate. P-value ≤0.05 was considered statistically significant. All statistical analyses were performed using R version 4.1. 10. Availability of datasets The spatial data sources used in this study can be accessed through following links: (1) 10× Visium data of the mouse brain, available at [https://satijalab.org/seurat/articles/spatial_vignette.html], along with the corresponding SC atlas [https://www.dropbox.com/s/cuowvm4vrf65pvq/allen_cortex.rds?dl=1] [59]; (2) 10× Visium data of the human thymus, accessible at [https://developmental.cellatlas.io/fetal-immune] [30], and the corresponding SC atlas of the thymus [DOI: https://doi.org/10.5281/zenodo.3572422] [31]; (3) 10× Visium data of the mouse kidney at https://www.10xgenomics.com/resources/datasets?query=&page=1&configure%5BhitsPerPage%5D=50&configure%5BmaxValuesPerFacet%5D=1000 and the GEO dataset GSE171406 [35], and the correspondence SC atlas (GSE129798) [11]; (4) 10× Visium data of the human DLPFC at [http://spatial.libd.org/spatialLIBD/] [37], and the matched SC atlas [https://cells.ucsc.edu/] [60]; (5) 10× Visium data of the human lung in GSE178361 [40], and the corresponding SC atlas [https://singlecell.broadinstitute.org/single_cell/study/SCP1219] [39]; (6) 10× Visium data of the human intestine in GSE158328 [9], and the SC atlas in GSE158702 [9]; (7) 10× Visium data of the human breast [https://doi.org/10.5281/zenodo.4739739], along with the SC atlas [GEO: GSE176078] [8]; (8) 10× Visium data of the renal cell carcinoma with TLS in GSE175540 [46], and the SC atlas [https://zenodo.org/records/4263972] [47]. For high-resolution ST data: (1) 10× Xenium dataset of the mouse brain is available from https://cf.10xgenomics.com/samples/xenium/1.0.2/Xenium_V1_FF_Mouse_Brain_Coronal_Subset_CTX_HP/Xenium_V1_FF_Mouse_Brain_Coronal_Subset_CTX_HP_outs.zip; (2) 10× Visium HD mouse brain dataset can be found at https://www.10xgenomics.com/datasets/visium-hd-cytassist-gene-expression-libraries-of-mouse-brain-he; (3) For the Slide-seq V2 platform, both the Mouse hippocampus dataset and the paired scRNA-seq reference were downloaded from https://satijalab.org/seurat/articles/spatial_vignette. Similarly, both the Mouse cerebellum dataset and the paired scRNA-seq reference were downloaded from the single cell portal project (https://singlecell.broadinstitute.org/single_cell/study/SCP948). Additionally, two mouse brain ST datasets with DAPI channels in fluorescence images for cell counting in spots were obtained from [https://support.10xgenomics.com/spatial-gene-expression/datasets/1.1.0/V1_Adult_Mouse_Brain_Coronal_Section_2] and [https://cf.10xgenomics.com/samples/spatial-exp/1.3.0/Visium_FFPE_Mouse_Brain_IF/Visium_FFPE_Mouse_Brain_IF_web_summary.html]. Finally, 32 simulated ST datasets with corresponding SC references were retrieved from https://drive.google.com/drive/folders/1pHmE9cg_tMcouV1LFJFtbyBJNp7oQo9J?usp=sharing. More details are provided in S1 Table. 1. Cell2Spatial analytical framework Data collection and normalization. The SC reference expression profile is assumed to be an N × C matrix, with N representing genes and C representing cells. The dataset is well-annotated (or clustered) into k distinct cell types (or clusters). Any cell type with fewer than a predefined threshold of cells (default: 5 cells) was excluded from the SC dataset. The spatial transcriptome (ST) expression profile is an M × S matrix, where M represents genes and S represents spots. A total of G genes is shared between the SC and ST. Data normalization was achieved by applying Seurat’s SCTransform method to both the SC (X → G × C) and ST data (Q → G × S). Cell-type-specific genes. Given that most genes cannot effectively distinguish among cell types, specific genes corresponding to distinct cell types were identified with computational efficiency using a specificity metric inspired by Shannon entropy structure (S9A Fig). Initially, the SC profile X was used to generate the average expression matrix X* (X* → G × K), with rows representing genes and columns representing cell types. Each entry in represents the mean expressions level of gene i in cell type j. Gene specificity scores were then computed using the following formula: (1) Here, represents the value characterizing the cell-type specificity of gene i (higher values indicate greater gene specificity), represents the mean expression of gene i across all cell types. The summation yields zero if the gene i is uniformly expressed across cell types, and reaches a maximum value of when it is exclusively expressed in a single cell type. The weight for gene i, was calculated as: (2) Thus, the gene weights are directly proportional to the expression level and serve as a key factor in determining the specificity values. Here, denotes the maximum average expression of gene i across all cell types, represents the smallest of these maximum average expression levels across all genes, and is overall maximum element in the matrix , encompassing all genes and all cell types. Each gene was initially assigned to the cell type in which it showed the highest average expression. For each cell type, we then calculated the expression variance of all genes across individual cells. Among the genes assigned to that cell type, those with variances exceeding the mean variance (calculated across all genes) by more than three standard deviations were excluded, to eliminate genes with unusually high within-type variability. For each cell type, the remaining assigned genes that passed the variance filter were ranked based on their specificity scores, and the top 100 genes were selected by default. This resulted in a gene set , where represents the set of specific genes for cell type j. The total number of selected genes, denoted as m (, as some cell types may contain fewer than 100 qualified genes after filtering). Measuring similarity between single cells and spots. To gauge the similarity between single cells and spatial spots, the specific gene list was initially converted into a vector and duplicates were removed, resulting in a length of m as previously defined. Then, the probability of specific gene i for each cell (i.e., ), was estimated using after applying Laplace Smoothing (by adding 1e-6 to account for zero values). The formula used was: , ensuring that . Based on the multinomial assumption, the likelihood function for a spot t with profile to cell was calculated as , and can be further described as follows: (3) Given that the geometric spatial distance can provide varying viewpoints to reflect the association between single cells and spots, Seurat’s CCA strategy was used for the integration of SC and ST datasets. Subsequently, UMAP was applied to obtain a two-dimensional projection of the integrated data. Although UMAP’s embeddings may distort global distances to enhance visualization, its local structure preservation ensures a meaningful representation of the relationships between a spatial spot and its neighboring cells, helping to characterize tissue architecture and cellular distributions [51] (S9B Fig; performance assessment of distance weighting versus no weighting for Cell2Spatial). Thus, the Euclidean distance between each spot and single cell was calculated, resulting in a distance matrix composed of S spots and C cells. A scaled distance-based weight matrix was then derived using the following formula: (4) where is a vector representing the distances between spot t and all single cells in UMAP coordinates, denotes the quantile value of the distance vector at the specifies probability level (with probs constrained between 0.01 and 1, defaulting to 0.3) (S9C Fig). Notably, if spot t is closer to cell in the UMAP projection, the weight is higher. If is below 0, it is re-set to 0. To combine these distance weights further, the likelihood similarity incorporating spatial distance-based weights was calculated as follows: (5) Here, is the original likelihood matrix representing the similarity between spots and cells, and is the final similarity matrix after incorporating spatial information. is the vector of observed likelihoods for spot t across all cells. is the corresponding normalized spatial distance weight matrix of the same dimensions, and “” indicates element-wise multiplication. Determination of spatial hotspot regions of cell types. To detect spatial hotspots for each cell type, the Getis-Ord G* index was employed, involving several steps: Initially, spatial coordinates of spots within the tissue sample were extracted from the ST Seurat object. Then, a neighborhood structure was established using k-nearest neighbors, where each spot’s nearest neighbors (default: 5 spots) were determined to form its neighborhood set. Next, the AddModuleScore function from Seurat was used to calculate signature scores for spots based on the gene sets specific to that cell type in . Finally, the statistic G* was calculated for each spot using the following formula: (6) where is the G* index of spot t in cell type j, represents the vector of signature scores for cell type j across all the spots, denotes the score for spot , is the global mean score of the ST section. is a binary weight indicating whether spot is a neighbor of spot t (1 if true, 0 otherwise), is the standard deviation of the signature scores in , and S represents the total number of spots as previously defined. represents the difference between the signature scores of neighboring spots and the global mean . Denominator normalizes the numerator by adjusting for the scale of the weighted differences. The standard G* is essentially a z-score that measures how the neighborhoods of spot t deviate from the context mean score [52]. Therefore, the p-value for each statistic of cell type j was computed as: (7) where is the cumulative distribution function of the standard normal distribution (S9D Fig). Spot t was classified as a hotspot for cell type j if (default: ). The processes mentioned above were applied to all spots, resulting in a Boolean matrix with dimensions S (spots) by K (cell types). Each entry in this matrix indicates whether a given spot is a hotspot (1) or not (0) for a specific cell type. This yields a list of detected hotspots for each cell type, denoted as , with a unique length of . represents the set of detected hotspots belonging to cell type j. Clustering spots of ST data. Following the standard Seurat [53] workflow for clustering ST data, the SCTransform function was used for normalization and variance stabilization. SCTransform computes Pearson residuals for each gene, reflecting deviations from expected expression based on a regularized negative binomial model. Genes with high variance in these residuals are considered highly variable. By default, the top 3,000 HVGs were used for clustering, which was performed with the resolution parameter set to 0.2 (as default). The obtained ST clusters were defined as a list of set , where represents spots belonging to the -th cluster. Estimation of cellular compositions in spatial section. To examine the cellular composition within a spatial section, SC expression profile was used as a reference, with m cell-type-specific genes in serving as distinctive features. This process generated a base matrix Bm×K, where rows represent genes, columns represent cell types, and each entry represents the average expression of a specific gene in the corresponding cell type. A reference-based deconvolution strategy, inspired by the LinDeconSeq tool [21,54], was established using a RLM to predict the relative prevalence of distinct cell types within each spot. To enhance estimation accuracy across the entire ST section, the expression vector of spot t was smoothed by the averaged expression of the six (as default) nearest neighbors within the same ST-defined cluster, including spot t itself, resulting in an updated vector . This smoothing strategy leverages the spatial coherence of gene expression and cell type composition within local regions, while mitigating estimation bias caused by uneven cell numbers across spots [55,56] (S9E Fig). Then, incorporating RLM to link and Bm×K allows for the estimation of cellular proportions of spot t by minimizing the squared discrepancies between observed data. (8) where is the non-negative cellular proportion vector to be estimated for spot t, with dimension , and . To estimate the overall cell type composition of the ST tissue section, cellular proportions are typically aggregated across all spatial spots. Since spots within the same spatial cluster tend to share similar cellular profiles, reference-based deconvolution methods may assign low, non-zero proportions to cell types that are genuinely absent in certain clusters. While these small residual estimates are individually minor, their cumulative effect across multiple spots can lead to systematic biases in the global cell type composition inference. To address this, a correction factor based on the hotspot matrix was introduced, calculated as follows: (9) Here, represents the r-th cluster to which spot i belongs. The expression is a vector of length K that reflects the number of times each cell type significantly appears in the cluster . This vector then was rescaled to a range of 0–1 using the rescale function (i.e., Min-max normalization) from “scales” R package (version 1.3.0). Each element of V is then binarized using a predefined threshold τ (default 0.01) as follows: (10) The resulting matrix V is a wrapped correction factor matrix with dimensions ( represents the number of spot clusters as previously defined). Therefore, the cellular composition can be adjusted using the following formula: (11) represents the correction factors of cluster . denotes the transpose of the r-th row vector of matrix V, allowing for a valid dot production. The cellular proportions of the spots in (excluding spots not classified as hotspot for any cell type) are organized into a matrix , where each row represents the adjusted compositions of one spot. Finally, the cellular compositions for the entire ST slice were derived by following formula. (12) is the final estimated cellular proportion vector of length K for the entire ST section (S9F Fig). Estimation of cell counts per spot for low-resolution spatial data using a corrected saturation model. To infer the number of cells per spot in low-resolution ST data, we applied a corrected saturation model that accounts for spot-level library size and gene count, assuming that the number of detected genes per spot follows a saturation curve as cell numbers increase. Specifically, the spatial expression matrix records UMI counts for G genes across spots, where represents the UMI count of gene i in spot t. To reduce noise, genes with low expression were excluded by applying a spot-level threshold (e.g., UMI ≥ 2), retaining only those with reliable signals. For each spot t, the library size is defined as , and the gene count is calculated as , where is an indicator function. Assuming the maximum number of cells among spots is (e.g., 10 for 10× Visium), we modeled the relationship among the gene count , cell number , and library size using a corrected saturation model: . is the maximum observed gene count across all spots, and represents the average number of genes per cell. To estimate , we initially computed a composite score for each spot to balance library size and gene count: (13) where and are the weights for library size and gene count (default: 0.5). The reference spot was estimated from the mean of spots exceeding the quantile cutoff of the composite score (90% by default) and was assumed to contain cells. The corresponding reference library size and gene count were then used to compute g: i.e., . Subsequently, the cell number for spot t was obtained by solving: (14) To ensure consistency, we incorporated the library size constraint , where . Finally, we calculated the harmonic mean of the and , and rounded to the nearest integer. i.e., . is a vector of cell counts across all spots. To maintain biological plausibility and adhere to platform resolution, the was constrained to the range [1, ]. Notably, for high-resolution spatial data at the SC level, set “max.cells.in.spot = 1” (or “platform.res = High”) to ensure one cell per spot. For subcellular-resolution platforms, such as STOMICS and Visium HD, adjust the bin size to approximate the dimensions of a single cell (e.g., for Visium HD, set bin.size = 8 µm). Assigning single cells to spatial spots. Mapping single cells to spatial spots enables the reconstruction of spatial architecture at SC resolution for low-resolution data and high gene capture expression patterns for high-resolution data. The mapping process involves the following steps: (1) Update SC Seurat object: Using the previously mentioned cellular fractions and spot cell , the number of cells in each cell type was calculated as . For the j-th cell type, cells were down-sampled from the corresponding cell type in the SC data X. If the cell counts of the j-the cell type in X was less than , sampling with replacement was applied; otherwise, sampling without replacement. Gaussian noise (mean = 0, sd = 1) was added to the gene expression of duplicated cells to avoid a singular matrix. The updated SC data denoted as X’ → G × C*, where . (2) Neural network design for pre-assigned individual cells to ST clusters: Recognizing the high algorithmic complexity involved in large-scale SC allocation, global cell-to-spot mapping becomes computationally intensive and less scalable. To address this, we introduced the FNN framework to preassign single cells mentioned in step (1) to ST clusters, effectively decomposing the original global mapping task into multiple subspace mapping problems. Specifically, reference-based coembedding approach from the Seurat package was used with SC data ( → G × C*) as the reference and the ST data (→ G × S’) as the query. From the coembedded data, a four-layer neural network was trained. The input layer comprised principal components (PCs)-embedded ST data (default 30 PCs), followed by a fully connected hidden layer with 64 neurons and a rectified linear unit (ReLU) activation function. The subsequent two hidden layers consisted of 128 units each. To address overfitting and enhance computational efficiency, dropout (rate = 0.2), batch normalization, and L2 regularization techniques were applied at each hidden layer. The output layer was based on the softmax function and the number of neurons in the aligned with ST clusters. The FNN’s learning rate was set to 0.000001. The involved major processes as follows: (15) represents the input layer of PCs of ST data. Notably, at the start of training, cells were randomly divided into a training set (95%) and a test set (5%). The test set was utilized at each epoch to evaluate the model’s performance and generalization capability. The training procedure adopted a 5-fold cross-validation approach, aiming to minimized the cross-entropy loss, which evaluates the similarity between predicted and actual categories. (16) is a binary indicator if ST cluster r is correct label for spot t, is the predicted probability of observation spot t for cluster r. is the regularization parameter controlling the impact of the regularization term on the loss function, with a default value of 0.01. After completing the training of the FNN model, the co-embedded SC data was input into the model to determine the probability of each cell belonging to specific ST cluster. The probability matrix is denoted as → C* × R. Given the pre-estimated cell counts within each spot, the cell allocation requirement for a specific cluster is determined by summing the estimated cells across all spots within that cluster. However, the FNN’s allocation of single cells to ST clusters may lead to imbalances, resulting in predicted cell counts that deviate from the expected values. To address this issue, an iterative optimization strategy based on the probability matrix is introduced, as illustrated in S9G Fig. This results in the creation of a list container : (17) is a set comprising the cells that pre-assigned to ST cluster . (3) Assigning single cells to spots using linear assignment strategy: Based on the cell counts within each spot, container , and the weighted likelihood similarity matrix , which is a subset of and the spots in rows consisted only of highlighted spots in . The columns in represent the cells source from cells , with each column’s values taken from corresponding to the same cell names. To minimize the spread of mapped single cells from specific cell type to regions they do not belong to, was adjusted as follows: (18) is a binary (0–1) matrix, where each column represents a binary vector of length . For each cell associated with cell type , is defined as a subset of values extracted from the corresponding column in the binary matrix , retaining only the entries corresponding to the hotspots in . The Jonker-Volgenant algorithm [23] was then employed to allocate single cells to specific spots to identify the ideal cell-to-spot assignment that maximize the following linear cost function (S9H Fig). For cluster : (19) where is a subset of , with rows corresponding to spots from , columns represent cells from . A is a Boolean matrix, where if single cell assigned to spot t. The constraint mentioned above guarantees that each cell is assigned only to one spot. The optimization is performed through the utilization of Python’s lapjv package (version 1.3.24) [20,57]. These processes are iteratively applied to other clusters. (4) Generating spatial coordinates for mapped single cells: To represent SC coordinates within spatial slices, the minimum Euclidean distance between adjacent spots was calculated. Center coordinates for each spot were extracted, and random coordinates for assigned single cells were generated using the spatstat.random::runifdisc function, with the radius set to half of this minimum distance. This method established the spatial positions of all single cells. The mapping results were subsequently integrated into a “Seurat” object and provided to the user. Notably, for small-scale datasets, the FNN pre-assignment strategy can be omitted, with a linear assignment algorithm applied directly. Data collection and normalization. The SC reference expression profile is assumed to be an N × C matrix, with N representing genes and C representing cells. The dataset is well-annotated (or clustered) into k distinct cell types (or clusters). Any cell type with fewer than a predefined threshold of cells (default: 5 cells) was excluded from the SC dataset. The spatial transcriptome (ST) expression profile is an M × S matrix, where M represents genes and S represents spots. A total of G genes is shared between the SC and ST. Data normalization was achieved by applying Seurat’s SCTransform method to both the SC (X → G × C) and ST data (Q → G × S). Cell-type-specific genes. Given that most genes cannot effectively distinguish among cell types, specific genes corresponding to distinct cell types were identified with computational efficiency using a specificity metric inspired by Shannon entropy structure (S9A Fig). Initially, the SC profile X was used to generate the average expression matrix X* (X* → G × K), with rows representing genes and columns representing cell types. Each entry in represents the mean expressions level of gene i in cell type j. Gene specificity scores were then computed using the following formula: (1) Here, represents the value characterizing the cell-type specificity of gene i (higher values indicate greater gene specificity), represents the mean expression of gene i across all cell types. The summation yields zero if the gene i is uniformly expressed across cell types, and reaches a maximum value of when it is exclusively expressed in a single cell type. The weight for gene i, was calculated as: (2) Thus, the gene weights are directly proportional to the expression level and serve as a key factor in determining the specificity values. Here, denotes the maximum average expression of gene i across all cell types, represents the smallest of these maximum average expression levels across all genes, and is overall maximum element in the matrix , encompassing all genes and all cell types. Each gene was initially assigned to the cell type in which it showed the highest average expression. For each cell type, we then calculated the expression variance of all genes across individual cells. Among the genes assigned to that cell type, those with variances exceeding the mean variance (calculated across all genes) by more than three standard deviations were excluded, to eliminate genes with unusually high within-type variability. For each cell type, the remaining assigned genes that passed the variance filter were ranked based on their specificity scores, and the top 100 genes were selected by default. This resulted in a gene set , where represents the set of specific genes for cell type j. The total number of selected genes, denoted as m (, as some cell types may contain fewer than 100 qualified genes after filtering). Measuring similarity between single cells and spots. To gauge the similarity between single cells and spatial spots, the specific gene list was initially converted into a vector and duplicates were removed, resulting in a length of m as previously defined. Then, the probability of specific gene i for each cell (i.e., ), was estimated using after applying Laplace Smoothing (by adding 1e-6 to account for zero values). The formula used was: , ensuring that . Based on the multinomial assumption, the likelihood function for a spot t with profile to cell was calculated as , and can be further described as follows: (3) Given that the geometric spatial distance can provide varying viewpoints to reflect the association between single cells and spots, Seurat’s CCA strategy was used for the integration of SC and ST datasets. Subsequently, UMAP was applied to obtain a two-dimensional projection of the integrated data. Although UMAP’s embeddings may distort global distances to enhance visualization, its local structure preservation ensures a meaningful representation of the relationships between a spatial spot and its neighboring cells, helping to characterize tissue architecture and cellular distributions [51] (S9B Fig; performance assessment of distance weighting versus no weighting for Cell2Spatial). Thus, the Euclidean distance between each spot and single cell was calculated, resulting in a distance matrix composed of S spots and C cells. A scaled distance-based weight matrix was then derived using the following formula: (4) where is a vector representing the distances between spot t and all single cells in UMAP coordinates, denotes the quantile value of the distance vector at the specifies probability level (with probs constrained between 0.01 and 1, defaulting to 0.3) (S9C Fig). Notably, if spot t is closer to cell in the UMAP projection, the weight is higher. If is below 0, it is re-set to 0. To combine these distance weights further, the likelihood similarity incorporating spatial distance-based weights was calculated as follows: (5) Here, is the original likelihood matrix representing the similarity between spots and cells, and is the final similarity matrix after incorporating spatial information. is the vector of observed likelihoods for spot t across all cells. is the corresponding normalized spatial distance weight matrix of the same dimensions, and “” indicates element-wise multiplication. Determination of spatial hotspot regions of cell types. To detect spatial hotspots for each cell type, the Getis-Ord G* index was employed, involving several steps: Initially, spatial coordinates of spots within the tissue sample were extracted from the ST Seurat object. Then, a neighborhood structure was established using k-nearest neighbors, where each spot’s nearest neighbors (default: 5 spots) were determined to form its neighborhood set. Next, the AddModuleScore function from Seurat was used to calculate signature scores for spots based on the gene sets specific to that cell type in . Finally, the statistic G* was calculated for each spot using the following formula: (6) where is the G* index of spot t in cell type j, represents the vector of signature scores for cell type j across all the spots, denotes the score for spot , is the global mean score of the ST section. is a binary weight indicating whether spot is a neighbor of spot t (1 if true, 0 otherwise), is the standard deviation of the signature scores in , and S represents the total number of spots as previously defined. represents the difference between the signature scores of neighboring spots and the global mean . Denominator normalizes the numerator by adjusting for the scale of the weighted differences. The standard G* is essentially a z-score that measures how the neighborhoods of spot t deviate from the context mean score [52]. Therefore, the p-value for each statistic of cell type j was computed as: (7) where is the cumulative distribution function of the standard normal distribution (S9D Fig). Spot t was classified as a hotspot for cell type j if (default: ). The processes mentioned above were applied to all spots, resulting in a Boolean matrix with dimensions S (spots) by K (cell types). Each entry in this matrix indicates whether a given spot is a hotspot (1) or not (0) for a specific cell type. This yields a list of detected hotspots for each cell type, denoted as , with a unique length of . represents the set of detected hotspots belonging to cell type j. Clustering spots of ST data. Following the standard Seurat [53] workflow for clustering ST data, the SCTransform function was used for normalization and variance stabilization. SCTransform computes Pearson residuals for each gene, reflecting deviations from expected expression based on a regularized negative binomial model. Genes with high variance in these residuals are considered highly variable. By default, the top 3,000 HVGs were used for clustering, which was performed with the resolution parameter set to 0.2 (as default). The obtained ST clusters were defined as a list of set , where represents spots belonging to the -th cluster. Estimation of cellular compositions in spatial section. To examine the cellular composition within a spatial section, SC expression profile was used as a reference, with m cell-type-specific genes in serving as distinctive features. This process generated a base matrix Bm×K, where rows represent genes, columns represent cell types, and each entry represents the average expression of a specific gene in the corresponding cell type. A reference-based deconvolution strategy, inspired by the LinDeconSeq tool [21,54], was established using a RLM to predict the relative prevalence of distinct cell types within each spot. To enhance estimation accuracy across the entire ST section, the expression vector of spot t was smoothed by the averaged expression of the six (as default) nearest neighbors within the same ST-defined cluster, including spot t itself, resulting in an updated vector . This smoothing strategy leverages the spatial coherence of gene expression and cell type composition within local regions, while mitigating estimation bias caused by uneven cell numbers across spots [55,56] (S9E Fig). Then, incorporating RLM to link and Bm×K allows for the estimation of cellular proportions of spot t by minimizing the squared discrepancies between observed data. (8) where is the non-negative cellular proportion vector to be estimated for spot t, with dimension , and . To estimate the overall cell type composition of the ST tissue section, cellular proportions are typically aggregated across all spatial spots. Since spots within the same spatial cluster tend to share similar cellular profiles, reference-based deconvolution methods may assign low, non-zero proportions to cell types that are genuinely absent in certain clusters. While these small residual estimates are individually minor, their cumulative effect across multiple spots can lead to systematic biases in the global cell type composition inference. To address this, a correction factor based on the hotspot matrix was introduced, calculated as follows: (9) Here, represents the r-th cluster to which spot i belongs. The expression is a vector of length K that reflects the number of times each cell type significantly appears in the cluster . This vector then was rescaled to a range of 0–1 using the rescale function (i.e., Min-max normalization) from “scales” R package (version 1.3.0). Each element of V is then binarized using a predefined threshold τ (default 0.01) as follows: (10) The resulting matrix V is a wrapped correction factor matrix with dimensions ( represents the number of spot clusters as previously defined). Therefore, the cellular composition can be adjusted using the following formula: (11) represents the correction factors of cluster . denotes the transpose of the r-th row vector of matrix V, allowing for a valid dot production. The cellular proportions of the spots in (excluding spots not classified as hotspot for any cell type) are organized into a matrix , where each row represents the adjusted compositions of one spot. Finally, the cellular compositions for the entire ST slice were derived by following formula. (12) is the final estimated cellular proportion vector of length K for the entire ST section (S9F Fig). Estimation of cell counts per spot for low-resolution spatial data using a corrected saturation model. To infer the number of cells per spot in low-resolution ST data, we applied a corrected saturation model that accounts for spot-level library size and gene count, assuming that the number of detected genes per spot follows a saturation curve as cell numbers increase. Specifically, the spatial expression matrix records UMI counts for G genes across spots, where represents the UMI count of gene i in spot t. To reduce noise, genes with low expression were excluded by applying a spot-level threshold (e.g., UMI ≥ 2), retaining only those with reliable signals. For each spot t, the library size is defined as , and the gene count is calculated as , where is an indicator function. Assuming the maximum number of cells among spots is (e.g., 10 for 10× Visium), we modeled the relationship among the gene count , cell number , and library size using a corrected saturation model: . is the maximum observed gene count across all spots, and represents the average number of genes per cell. To estimate , we initially computed a composite score for each spot to balance library size and gene count: (13) where and are the weights for library size and gene count (default: 0.5). The reference spot was estimated from the mean of spots exceeding the quantile cutoff of the composite score (90% by default) and was assumed to contain cells. The corresponding reference library size and gene count were then used to compute g: i.e., . Subsequently, the cell number for spot t was obtained by solving: (14) To ensure consistency, we incorporated the library size constraint , where . Finally, we calculated the harmonic mean of the and , and rounded to the nearest integer. i.e., . is a vector of cell counts across all spots. To maintain biological plausibility and adhere to platform resolution, the was constrained to the range [1, ]. Notably, for high-resolution spatial data at the SC level, set “max.cells.in.spot = 1” (or “platform.res = High”) to ensure one cell per spot. For subcellular-resolution platforms, such as STOMICS and Visium HD, adjust the bin size to approximate the dimensions of a single cell (e.g., for Visium HD, set bin.size = 8 µm). Assigning single cells to spatial spots. Mapping single cells to spatial spots enables the reconstruction of spatial architecture at SC resolution for low-resolution data and high gene capture expression patterns for high-resolution data. The mapping process involves the following steps: (1) Update SC Seurat object: Using the previously mentioned cellular fractions and spot cell , the number of cells in each cell type was calculated as . For the j-th cell type, cells were down-sampled from the corresponding cell type in the SC data X. If the cell counts of the j-the cell type in X was less than , sampling with replacement was applied; otherwise, sampling without replacement. Gaussian noise (mean = 0, sd = 1) was added to the gene expression of duplicated cells to avoid a singular matrix. The updated SC data denoted as X’ → G × C*, where . (2) Neural network design for pre-assigned individual cells to ST clusters: Recognizing the high algorithmic complexity involved in large-scale SC allocation, global cell-to-spot mapping becomes computationally intensive and less scalable. To address this, we introduced the FNN framework to preassign single cells mentioned in step (1) to ST clusters, effectively decomposing the original global mapping task into multiple subspace mapping problems. Specifically, reference-based coembedding approach from the Seurat package was used with SC data ( → G × C*) as the reference and the ST data (→ G × S’) as the query. From the coembedded data, a four-layer neural network was trained. The input layer comprised principal components (PCs)-embedded ST data (default 30 PCs), followed by a fully connected hidden layer with 64 neurons and a rectified linear unit (ReLU) activation function. The subsequent two hidden layers consisted of 128 units each. To address overfitting and enhance computational efficiency, dropout (rate = 0.2), batch normalization, and L2 regularization techniques were applied at each hidden layer. The output layer was based on the softmax function and the number of neurons in the aligned with ST clusters. The FNN’s learning rate was set to 0.000001. The involved major processes as follows: (15) represents the input layer of PCs of ST data. Notably, at the start of training, cells were randomly divided into a training set (95%) and a test set (5%). The test set was utilized at each epoch to evaluate the model’s performance and generalization capability. The training procedure adopted a 5-fold cross-validation approach, aiming to minimized the cross-entropy loss, which evaluates the similarity between predicted and actual categories. (16) is a binary indicator if ST cluster r is correct label for spot t, is the predicted probability of observation spot t for cluster r. is the regularization parameter controlling the impact of the regularization term on the loss function, with a default value of 0.01. After completing the training of the FNN model, the co-embedded SC data was input into the model to determine the probability of each cell belonging to specific ST cluster. The probability matrix is denoted as → C* × R. Given the pre-estimated cell counts within each spot, the cell allocation requirement for a specific cluster is determined by summing the estimated cells across all spots within that cluster. However, the FNN’s allocation of single cells to ST clusters may lead to imbalances, resulting in predicted cell counts that deviate from the expected values. To address this issue, an iterative optimization strategy based on the probability matrix is introduced, as illustrated in S9G Fig. This results in the creation of a list container : (17) is a set comprising the cells that pre-assigned to ST cluster . (3) Assigning single cells to spots using linear assignment strategy: Based on the cell counts within each spot, container , and the weighted likelihood similarity matrix , which is a subset of and the spots in rows consisted only of highlighted spots in . The columns in represent the cells source from cells , with each column’s values taken from corresponding to the same cell names. To minimize the spread of mapped single cells from specific cell type to regions they do not belong to, was adjusted as follows: (18) is a binary (0–1) matrix, where each column represents a binary vector of length . For each cell associated with cell type , is defined as a subset of values extracted from the corresponding column in the binary matrix , retaining only the entries corresponding to the hotspots in . The Jonker-Volgenant algorithm [23] was then employed to allocate single cells to specific spots to identify the ideal cell-to-spot assignment that maximize the following linear cost function (S9H Fig). For cluster : (19) where is a subset of , with rows corresponding to spots from , columns represent cells from . A is a Boolean matrix, where if single cell assigned to spot t. The constraint mentioned above guarantees that each cell is assigned only to one spot. The optimization is performed through the utilization of Python’s lapjv package (version 1.3.24) [20,57]. These processes are iteratively applied to other clusters. (4) Generating spatial coordinates for mapped single cells: To represent SC coordinates within spatial slices, the minimum Euclidean distance between adjacent spots was calculated. Center coordinates for each spot were extracted, and random coordinates for assigned single cells were generated using the spatstat.random::runifdisc function, with the radius set to half of this minimum distance. This method established the spatial positions of all single cells. The mapping results were subsequently integrated into a “Seurat” object and provided to the user. Notably, for small-scale datasets, the FNN pre-assignment strategy can be omitted, with a linear assignment algorithm applied directly. 2. Benchmarking and simulations Simulation of spatial transcriptome data. To assess the precision and robustness of Cell2Spatial, simulated ST datasets were generated using the Allen mouse brain SC dataset with brain ST data as the reference template. For training, 50% of cells from each type were randomly selected, and gene expression perturbations (0%, 5%, 10%, 15%, and 20%) were introduced by shuffling expression values across cells to simulate noise. The top 2,000 HVGs were identified using the FindVariableFeatures function in Seurat, and PCCs were calculated to construct a cell–spot correlation matrix. Random spot-level cell counts were drawn from a Poisson distribution (Lambda = 5), and cells were assigned to spots according to the correlation matrix and sampled counts. The expression profile of each spot was then obtained by summing the expression values of its assigned single cells. For each simulated dataset, the number of cells, cell types, and spatial coordinates per spot were recorded for downstream evaluation. To further examine deconvolution performance and gene expression recovery, additional synthetic ST datasets were created with Poisson-distributed cell counts using Lambda values of 5, 10, 15, and 20. Benchmark metrics. Accuracy index: Various spatial mapping tools (Cell2Spatial, CytoSPACE, CellTrek, Tangram, and Seurat) were used to project the remaining single cells of Allen mouse brain onto simulated ST sections with different levels of noise (Fig 2G). The performance of these tools was evaluated using the following formula. (20) [range: 0–1] represents the accuracy metric of cell type j for effectively mapping to the simulated spots, which integrates both the relative abundance of cells in spatial spots and the RMSE between predicted and true counts. denotes the true count for cell type j at spatial spot t, while represents the predicted count for the same type at spot t. S refers to the total number of spots in the ST slice. is the RMSE between the predicted and true counts for cell type j across all spots. Highly variable genes (HVGs) conservation: The HVGs conservation score is a critical metric for assessing the preservation of biological signals. To compute this score, Seurat’s FindVariableFeatures function was utilized to identify HVGs in each ST dataset, both before and after mapping. For each batch, the target was to identify 500 HVGs. When evaluating the impact of the “group.size” parameter on mapping performance, the number of HVGs was set to 2,000. The HVG conservation score is defined as: (21) Jaccard index: To evaluate the consistency between the mapped ST data and the synthetic ST data, the Jaccard index was used. The index, calculated at the cell type or SC level for each spot, is defined as: (22) Here, denotes the cell types or cells within the mapped spot, refers to those within the corresponding spot in the synthetic dataset. Pearson correlation coefficients (PCCs) and root mean square errors (RMSEs): Spatial deconvolution performance was compared using PCC and RMSE metrics. (1) For spatial mapping tools, cell type proportions in each spot were calculated based on the spatial distribution of assigned single cells. (2) For spatial deconvolution tools, proportions were inferred from the ST data using the corresponding SC dataset as a reference. Overall cell type composition in the ST slices was estimated by randomly selecting 2,000 spots, and the composition was computed as follows: (23) where k denotes the number of cell types, t represents the index of spot. PCC and RMSE were then used to evaluate the concordance between synthetic and predicted compositions at the overall cellular composition level of the ST section. This procedure was repeated 100 times to generate error bars. Cosine similarity of gene expression: Expression consistency between the mapped ST data and the synthetic ST data was evaluated by aggregating the profiles for each spot in the mapped ST data. This was done by summing the expression values of each gene across the single cells assigned to that spot, resulting in the recovered ST data. Consistency was then assessed by calculating the cosine similarity using 2,000 HVGs between the recovered and synthetic ST spots. Kullback–Leibler (KL) divergence: To assess the global similarity of spatial structures, a KL-divergence-based approach from the CellTrek (version 0.0.94) package was utilized. The CellTrek::SColoc function computed a 2D grid kernel density for each cell type using kde2d from the “MASS” package (version 7.3–58.3), with a default bandwidth (h = 100) corresponding to the spatial distance between adjacent ST spots and a grid size (n) of 25. The KL-divergence was then calculated for the 2D density distributions between each pair of cell types from the mapped and synthetic data, with the synthetic data utilizing advanced SC information to achieve SC resolution. Simulation of spatial transcriptome data. To assess the precision and robustness of Cell2Spatial, simulated ST datasets were generated using the Allen mouse brain SC dataset with brain ST data as the reference template. For training, 50% of cells from each type were randomly selected, and gene expression perturbations (0%, 5%, 10%, 15%, and 20%) were introduced by shuffling expression values across cells to simulate noise. The top 2,000 HVGs were identified using the FindVariableFeatures function in Seurat, and PCCs were calculated to construct a cell–spot correlation matrix. Random spot-level cell counts were drawn from a Poisson distribution (Lambda = 5), and cells were assigned to spots according to the correlation matrix and sampled counts. The expression profile of each spot was then obtained by summing the expression values of its assigned single cells. For each simulated dataset, the number of cells, cell types, and spatial coordinates per spot were recorded for downstream evaluation. To further examine deconvolution performance and gene expression recovery, additional synthetic ST datasets were created with Poisson-distributed cell counts using Lambda values of 5, 10, 15, and 20. Benchmark metrics. Accuracy index: Various spatial mapping tools (Cell2Spatial, CytoSPACE, CellTrek, Tangram, and Seurat) were used to project the remaining single cells of Allen mouse brain onto simulated ST sections with different levels of noise (Fig 2G). The performance of these tools was evaluated using the following formula. (20) [range: 0–1] represents the accuracy metric of cell type j for effectively mapping to the simulated spots, which integrates both the relative abundance of cells in spatial spots and the RMSE between predicted and true counts. denotes the true count for cell type j at spatial spot t, while represents the predicted count for the same type at spot t. S refers to the total number of spots in the ST slice. is the RMSE between the predicted and true counts for cell type j across all spots. Highly variable genes (HVGs) conservation: The HVGs conservation score is a critical metric for assessing the preservation of biological signals. To compute this score, Seurat’s FindVariableFeatures function was utilized to identify HVGs in each ST dataset, both before and after mapping. For each batch, the target was to identify 500 HVGs. When evaluating the impact of the “group.size” parameter on mapping performance, the number of HVGs was set to 2,000. The HVG conservation score is defined as: (21) Jaccard index: To evaluate the consistency between the mapped ST data and the synthetic ST data, the Jaccard index was used. The index, calculated at the cell type or SC level for each spot, is defined as: (22) Here, denotes the cell types or cells within the mapped spot, refers to those within the corresponding spot in the synthetic dataset. Pearson correlation coefficients (PCCs) and root mean square errors (RMSEs): Spatial deconvolution performance was compared using PCC and RMSE metrics. (1) For spatial mapping tools, cell type proportions in each spot were calculated based on the spatial distribution of assigned single cells. (2) For spatial deconvolution tools, proportions were inferred from the ST data using the corresponding SC dataset as a reference. Overall cell type composition in the ST slices was estimated by randomly selecting 2,000 spots, and the composition was computed as follows: (23) where k denotes the number of cell types, t represents the index of spot. PCC and RMSE were then used to evaluate the concordance between synthetic and predicted compositions at the overall cellular composition level of the ST section. This procedure was repeated 100 times to generate error bars. Cosine similarity of gene expression: Expression consistency between the mapped ST data and the synthetic ST data was evaluated by aggregating the profiles for each spot in the mapped ST data. This was done by summing the expression values of each gene across the single cells assigned to that spot, resulting in the recovered ST data. Consistency was then assessed by calculating the cosine similarity using 2,000 HVGs between the recovered and synthetic ST spots. Kullback–Leibler (KL) divergence: To assess the global similarity of spatial structures, a KL-divergence-based approach from the CellTrek (version 0.0.94) package was utilized. The CellTrek::SColoc function computed a 2D grid kernel density for each cell type using kde2d from the “MASS” package (version 7.3–58.3), with a default bandwidth (h = 100) corresponding to the spatial distance between adjacent ST spots and a grid size (n) of 25. The KL-divergence was then calculated for the 2D density distributions between each pair of cell types from the mapped and synthetic data, with the synthetic data utilizing advanced SC information to achieve SC resolution. 3. Comparison of Cell2Spatial with publicly available spatial tools Cell2Spatial. Cell2Spatial primarily employs default parameters for spot segmentation of both real and simulated 10× Visium ST data. Key parameters used include: normalize.method = ‘SCTransform’, fix.cells.in.spot = TRUE, knn.spots = 5, and marker.selection = ‘shannon’. For 10× Xenium, Visium HD, and Slide-seq V2 high-resolution ST data, the maximum cell count per spot was set to 1. CytoSPACE. CytoSPACE [20] assigns individual cells to precise spatial spots by optimizing a correlation-based cost function, taking into account the estimated cell count per spot. This optimization is carried out using a shortest augmenting path algorithm. The “duplicated cells” option and “lapjv” solver were used during the CytoSPACE execution, and a standardized mean cell count of 5 per spot was maintained across all 10× Visium samples. For 10× Xenium, Visium HD, and Slide-seq V2 high-resolution ST data, the mean cell count was set to 2 for successful running. CellTrek. CellTrek utilizes ST data to train a random forest model for predicting spatial coordinates, leveraging dimension reduction features shared with SC data [3]. The traint function from the “CellTrek” R package (version 0.0.94) was to obtain the co-embedding of ST and SC data for Visium samples. Subsequently, the celltrek function with default parameters (“reduction = ‘pca’, intp = T, intp_pnt = 10,000, intp_lin = F, nPCs = 30, ntree = 1,000, dist_thresh = 0.4, top_spot = 10, spot_n = 10, repel_r = 5, repel_iter = 10, keep_model = T) was used to project individual cells onto the ST coordinates. Cells were then assigned to their nearest spots based on the Euclidean distance. Notably, CellTrek failed to execute successfully for 10× Xenium, Visium HD, and Slide-seq V2 high-resolution data, thus it was excluded from the assessment and comparison of high-resolution ST data. Tangram. Tangram [19] is a specialized deep-learning framework designed to reveal intricate spatial structures at the SC level. The “Seurat” object was converted into an “h5ad” file for all Visium samples, and Tangram was applied with default settings. The individual cells were mapped to ST spots by utilizing all accessible genes for each cell. To visualize these single cells on ST sections, random coordinates were generated for single cells based on their association with the respective spots. Seurat. Seurat [18] employs a label transfer approach to map ST spots onto individual cell type, and considers each spot as a unit at the SC granularity level. The FindTransferAnchors function in Seurat was used in this study, with ST data as the query and the corresponding SC data as the reference. Subsequently, the TransferData function was utilized to assign SC labels (i.e., cell types) to the ST spots. Default parameters were used as specified in the official documentation. Cell2location. The guidelines on the Cell2location website (https://cell2location.readthedocs.io/en/latest/notebooks/cell2location_tutorial.html) were followed. The SC regression model was trained with the parameters “max_epochs = 100” and “lr = 0.002”. The Cell2location model was then obtained using “max_epochs = 10,000”. SpatialDWLS. The guidelines on the SpatialDWLS website (https://rubd.github.io/Giotto_site/articles/tut7_giotto_enrichment.html) were followed, with the parameter set to “n_cell = 20”. RCTD. The guidelines on the RCTD GitHub repository (https://raw.githack.com/dmcable/spacexr/master/vignettes/spatial-transcriptomics.html) were followed, with the parameter set to “doublet_mode = full”. Stereoscope. The guidelines on the website (https://docs.scvi-tools.org/en/stable/user_guide/models/stereoscope.html) were followed. The SC model was trained with parameters “max_epochs = 50” and the spatial model was trained with parameters “max_epochs = 100”. DestVI. The guidelines on the DestVI website (https://docs.scvi-tools.org/en/stable/tutorials/notebooks/DestVI_tutorial.html) were followed. The SC model was trained with parameters “max_epochs = 50, lr = 0.001, number of training genes =2,000”. The spatial model was trained with parameters “max_epochs = 100”. SpaOTsc. The guidlines on the SpaOTsc GitHub repository (https://github.com/zcang/SpaOTsc) were followed. The spatial distribution of genes was obtained using the function ‘issc.transport_plan’ with parameters “alpha=0, rho=1.0, epsilon=0.1, scaling=False”. novoSpaRc. The guidelines on the GitHub repository of novoSpaRc (https://github.com/rajewsky-lab/novosparc/blob/master/reconstruct_drosophila_embryo_tutorial.ipynb) were followed, with the parameters set to “alpha_linear = 0.5, loss_fun = square_loss, epsilon = 5 × 10−3”. SPOTlight. The guidelines on the SPOTlight GitHub repository (https://marcelosua.github.io/SPOTlight/) were followed, with the parameter set to “transf = uv, method = nsNMF”. DSTG. The guidelines on the DSTG GitHub repository (https://github.com/Su-informatics-lab/DSTG) were followed. CARD. The guidelines on the CARD GitHub repository (https://yma-lab.github.io/CARD/documentation/04_CARD_Example.html) were followed, with the parameters set to “minCountGene = 100, minCountSpot = 5”. Cell2Spatial. Cell2Spatial primarily employs default parameters for spot segmentation of both real and simulated 10× Visium ST data. Key parameters used include: normalize.method = ‘SCTransform’, fix.cells.in.spot = TRUE, knn.spots = 5, and marker.selection = ‘shannon’. For 10× Xenium, Visium HD, and Slide-seq V2 high-resolution ST data, the maximum cell count per spot was set to 1. CytoSPACE. CytoSPACE [20] assigns individual cells to precise spatial spots by optimizing a correlation-based cost function, taking into account the estimated cell count per spot. This optimization is carried out using a shortest augmenting path algorithm. The “duplicated cells” option and “lapjv” solver were used during the CytoSPACE execution, and a standardized mean cell count of 5 per spot was maintained across all 10× Visium samples. For 10× Xenium, Visium HD, and Slide-seq V2 high-resolution ST data, the mean cell count was set to 2 for successful running. CellTrek. CellTrek utilizes ST data to train a random forest model for predicting spatial coordinates, leveraging dimension reduction features shared with SC data [3]. The traint function from the “CellTrek” R package (version 0.0.94) was to obtain the co-embedding of ST and SC data for Visium samples. Subsequently, the celltrek function with default parameters (“reduction = ‘pca’, intp = T, intp_pnt = 10,000, intp_lin = F, nPCs = 30, ntree = 1,000, dist_thresh = 0.4, top_spot = 10, spot_n = 10, repel_r = 5, repel_iter = 10, keep_model = T) was used to project individual cells onto the ST coordinates. Cells were then assigned to their nearest spots based on the Euclidean distance. Notably, CellTrek failed to execute successfully for 10× Xenium, Visium HD, and Slide-seq V2 high-resolution data, thus it was excluded from the assessment and comparison of high-resolution ST data. Tangram. Tangram [19] is a specialized deep-learning framework designed to reveal intricate spatial structures at the SC level. The “Seurat” object was converted into an “h5ad” file for all Visium samples, and Tangram was applied with default settings. The individual cells were mapped to ST spots by utilizing all accessible genes for each cell. To visualize these single cells on ST sections, random coordinates were generated for single cells based on their association with the respective spots. Seurat. Seurat [18] employs a label transfer approach to map ST spots onto individual cell type, and considers each spot as a unit at the SC granularity level. The FindTransferAnchors function in Seurat was used in this study, with ST data as the query and the corresponding SC data as the reference. Subsequently, the TransferData function was utilized to assign SC labels (i.e., cell types) to the ST spots. Default parameters were used as specified in the official documentation. Cell2location. The guidelines on the Cell2location website (https://cell2location.readthedocs.io/en/latest/notebooks/cell2location_tutorial.html) were followed. The SC regression model was trained with the parameters “max_epochs = 100” and “lr = 0.002”. The Cell2location model was then obtained using “max_epochs = 10,000”. SpatialDWLS. The guidelines on the SpatialDWLS website (https://rubd.github.io/Giotto_site/articles/tut7_giotto_enrichment.html) were followed, with the parameter set to “n_cell = 20”. RCTD. The guidelines on the RCTD GitHub repository (https://raw.githack.com/dmcable/spacexr/master/vignettes/spatial-transcriptomics.html) were followed, with the parameter set to “doublet_mode = full”. Stereoscope. The guidelines on the website (https://docs.scvi-tools.org/en/stable/user_guide/models/stereoscope.html) were followed. The SC model was trained with parameters “max_epochs = 50” and the spatial model was trained with parameters “max_epochs = 100”. DestVI. The guidelines on the DestVI website (https://docs.scvi-tools.org/en/stable/tutorials/notebooks/DestVI_tutorial.html) were followed. The SC model was trained with parameters “max_epochs = 50, lr = 0.001, number of training genes =2,000”. The spatial model was trained with parameters “max_epochs = 100”. SpaOTsc. The guidlines on the SpaOTsc GitHub repository (https://github.com/zcang/SpaOTsc) were followed. The spatial distribution of genes was obtained using the function ‘issc.transport_plan’ with parameters “alpha=0, rho=1.0, epsilon=0.1, scaling=False”. novoSpaRc. The guidelines on the GitHub repository of novoSpaRc (https://github.com/rajewsky-lab/novosparc/blob/master/reconstruct_drosophila_embryo_tutorial.ipynb) were followed, with the parameters set to “alpha_linear = 0.5, loss_fun = square_loss, epsilon = 5 × 10−3”. SPOTlight. The guidelines on the SPOTlight GitHub repository (https://marcelosua.github.io/SPOTlight/) were followed, with the parameter set to “transf = uv, method = nsNMF”. DSTG. The guidelines on the DSTG GitHub repository (https://github.com/Su-informatics-lab/DSTG) were followed. CARD. The guidelines on the CARD GitHub repository (https://yma-lab.github.io/CARD/documentation/04_CARD_Example.html) were followed, with the parameters set to “minCountGene = 100, minCountSpot = 5”. 4. Score-Guided Mapping Accuracy (SGMA) Given the lack of information on precise cell location in real ST data, a strategy called SGMA was devised to gauge the performance of the mapping tools. Briefly, Seurat’s FindAllMarker function (logfc.threshold = 0.25, min.pct = 0.25) was used to identify marker genes for each reference cell type within the corresponding SC dataset, and the top 20 overexpressed genes for each cell type were screened. The AddModuleScore function from Seurat was then applied in conjunction with the previously identified marker gene list to score the spots in the ST data. Thereafter, the scores of each cell type in spots were fitted to a Gaussian distribution, and the p-value for each score was calculated. Candidate spots potentially harboring a specific cell type were identified based on p-value ≤0.01 or 0.05. Finally, the performance of the mapping tool was quantified using the formula below: (24) represents the set of spots that the mapping tool assigns to cell type j, Sj represents the actual spots (determined by the Gaussian distribution mentioned above) where cell type j is located, signifies the count of shared spots between and , represents the total number of spots in the combined set of and , and k is the number of cell types. For a more intuitive comparison, accuracy metrics of various ST datasets were scaled to 0–1. 5. Cell counting in low-resolution spatial spots using DAPI from Visium fluorescence Cell counts for the spots were obtained using the DAPI channel of the Visium fluorescence image, the spatial dataset was processed using “Squidpy” python package (version 1.6.0) [29]. The image was first smoothed using the smooth method, followed by segmentation with the watershed algorithm. Segmentation features were then calculated and integrated into the associated “AnnData” object. The cell counts for each spot were derived by extracting the “segmentation_label” information from the “AnnData” object. 6. Memory usage and time consumption analysis To evaluate the memory usage and time consumption of Cell2Spatial, CytoSPACE, CellTrek, Tangram, and Seurat, 10 simulated ST datasets with 1,000–10,000 spots were generated using mouse brain data. The mapping programs were then executed with the corresponding SC data. Memory usage and time consumption were assessed using the “peakRAM” R package (version 1.0.2) [58]. 7. Gene set enrichment analysis (GSEA) GSEA was performed to determine whether single cells mapped onto thymic spatial sections by Cell2Spatial can provide insights into critical functional regions. Firstly, the single cells in the medullary region were compared with those in other areas using Seurat’s FindAllMarkers function (Wilcoxon rank sum test). The functional gene sets were obtained from the C5 ontology gene sets within the MsigDB database. Subsequently, the “clusterProfiler” package (version 4.0.5) [44] and bitr function were used to convert gene symbols into Entrez IDs. The genes were sorted in descending order based on the fold change in expression. GSEA function was utilized with the “TERM2GENE = C5” parameter for enrichment analysis, and terms with adjusted p-values below 0.01 were considered significantly enriched. Finally, the gseaplot2 function from the “enrichplot” package (version 1.18.4) was used to visualize the selected terms. 8. Spatial co-existence analysis of cell types in low-resolution spatial data using Cell2Spatial The spatial co-existence of the different cell types was determined by projecting individual cells onto ST sections. Following the mapping of single cells to their respective spatial spots, the spatial co-localization among the various cell types was quantified using the formula as follows: (25) |Si| and |Sj| represent the number of spots containing cell types i and j, respectively, |Sij| denotes the number of spots containing both cell type i and j, and Eij is the coexistence index for cell types i and j (ranging from 0 to 1). Eij = 1 signifies complete co-localization of the cell types in the ST sections, and Eij = 0 indicates a lack of co-localization. 9. Statistical analysis Pearson and Spearman correlation analysis, Student t test, Wilcoxon rank sum test, and Kruskal–Wallis test were performed as appropriate. P-value ≤0.05 was considered statistically significant. All statistical analyses were performed using R version 4.1. 10. Availability of datasets The spatial data sources used in this study can be accessed through following links: (1) 10× Visium data of the mouse brain, available at [https://satijalab.org/seurat/articles/spatial_vignette.html], along with the corresponding SC atlas [https://www.dropbox.com/s/cuowvm4vrf65pvq/allen_cortex.rds?dl=1] [59]; (2) 10× Visium data of the human thymus, accessible at [https://developmental.cellatlas.io/fetal-immune] [30], and the corresponding SC atlas of the thymus [DOI: https://doi.org/10.5281/zenodo.3572422] [31]; (3) 10× Visium data of the mouse kidney at https://www.10xgenomics.com/resources/datasets?query=&page=1&configure%5BhitsPerPage%5D=50&configure%5BmaxValuesPerFacet%5D=1000 and the GEO dataset GSE171406 [35], and the correspondence SC atlas (GSE129798) [11]; (4) 10× Visium data of the human DLPFC at [http://spatial.libd.org/spatialLIBD/] [37], and the matched SC atlas [https://cells.ucsc.edu/] [60]; (5) 10× Visium data of the human lung in GSE178361 [40], and the corresponding SC atlas [https://singlecell.broadinstitute.org/single_cell/study/SCP1219] [39]; (6) 10× Visium data of the human intestine in GSE158328 [9], and the SC atlas in GSE158702 [9]; (7) 10× Visium data of the human breast [https://doi.org/10.5281/zenodo.4739739], along with the SC atlas [GEO: GSE176078] [8]; (8) 10× Visium data of the renal cell carcinoma with TLS in GSE175540 [46], and the SC atlas [https://zenodo.org/records/4263972] [47]. For high-resolution ST data: (1) 10× Xenium dataset of the mouse brain is available from https://cf.10xgenomics.com/samples/xenium/1.0.2/Xenium_V1_FF_Mouse_Brain_Coronal_Subset_CTX_HP/Xenium_V1_FF_Mouse_Brain_Coronal_Subset_CTX_HP_outs.zip; (2) 10× Visium HD mouse brain dataset can be found at https://www.10xgenomics.com/datasets/visium-hd-cytassist-gene-expression-libraries-of-mouse-brain-he; (3) For the Slide-seq V2 platform, both the Mouse hippocampus dataset and the paired scRNA-seq reference were downloaded from https://satijalab.org/seurat/articles/spatial_vignette. Similarly, both the Mouse cerebellum dataset and the paired scRNA-seq reference were downloaded from the single cell portal project (https://singlecell.broadinstitute.org/single_cell/study/SCP948). Additionally, two mouse brain ST datasets with DAPI channels in fluorescence images for cell counting in spots were obtained from [https://support.10xgenomics.com/spatial-gene-expression/datasets/1.1.0/V1_Adult_Mouse_Brain_Coronal_Section_2] and [https://cf.10xgenomics.com/samples/spatial-exp/1.3.0/Visium_FFPE_Mouse_Brain_IF/Visium_FFPE_Mouse_Brain_IF_web_summary.html]. Finally, 32 simulated ST datasets with corresponding SC references were retrieved from https://drive.google.com/drive/folders/1pHmE9cg_tMcouV1LFJFtbyBJNp7oQo9J?usp=sharing. More details are provided in S1 Table. Supporting information S1 Fig. Spatial characterization and cell-type-specific genes in mouse brain tissue. (A) (i) Clustering of spots from mouse brain spatial transcriptomics (ST) data, with clusters color-coded. (ii) Hematoxylin and eosin (H&E) staining image of mouse brain. (iii) Uniform Manifold Approximation and Projection (UMAP) showing clusters of spots inferred using Seurat [18] common processes. (B) UMAP plot showing the single-cell atlas of mouse brain. Each dot represents an individual cell. Cell types are marked by color codes. (C) Bubble plot showing the expression of five representative cell-type–specific marker genes selected from the top 30 overexpressed genes, as inferred by Cell2Spatial. Color intensity reflects the expression level of each gene in each cell type, while dot size corresponds to the proportion of cells expressing the gene. (D–G) Distribution of selected cell types with distinct spatial locations in mouse brain tissue, depicted using various mapping tools: (D) CytoSPACE, (E) CellTrek, (F) Tangram, and (G) Seurat. Each dot represents an individual cell. The data underlying this figure can be found at https://doi.org/10.5281/zenodo.17212677. https://doi.org/10.1371/journal.pbio.3003477.s001 (TIF) S2 Fig. Synthetic data generation and performance evaluation for spatial transcriptomics (ST) mapping tools. (A) Synthetic data generation strategy: This strategy comprises three main steps: (i) Randomly sampling cell counts for each spot based on a Poisson distribution, (ii) Assessing single-cell (SC) and spot similarity, and (iii) Aggregating the expression of nearby single cells based on the cell counts determined in step (i). To introduce variability, k% of genes are randomly perturbed with noise, resulting in synthetic ST data with diverse noise profiles (see “Materials and methods”). (B) Clustering of synthetic spots with varying noise levels on mouse brain tissue section. (C) Spatial heat maps showing the performance of publicly available mapping tools (CytoSPACE, CellTrek, Tangram, and Seurat) for aligning SC data (with 5% added noise) to spatial spots in ST datasets simulated with five cells on average (see “Materials and methods”). To enhance clarity, we have presented only cell types with prominent spatial structures. The color intensity of each spot corresponds to the number of single cells assigned. (D) Strategy for estimating the number of cells in spots (see “Materials and methods”). (E) Concordance between the predicted and expected number of cells based on the synthetic mouse brain ST data, with dot size indicating the density of spots sharing the same predicted cell count. (i) Lambda = 5; (ii) Lambda = 20. Cell2Spatial (left panel); CytoSPACE (right panel). (F) Heat map showing the Pearson correlation coefficients (PCCs) between predicted and simulated benchmarking data for both Cell2Spatial and CytoSPACE. Lambda values ranging from 1 to 20 were used for random sampling the number of cells following a Poisson distribution. (G) Box and violin plots displaying the distributions of PCCs (i) and root mean square errors (RMSEs) (ii) between predicted and actual counts in spots for Cell2Spatial and CytoSPACE. P-values were determined using the Wilcoxon tests. (H) Bar plot showing the influence of the preset maximum cell count per spot on prediction accuracy in low-resolution ST data. (left) RMSEs quantifies the deviation between predicted and actual cell counts in the synthetic ST dataset, while (right) the PCCs assesses their concordance. The underlying data for this figure can be found at https://zenodo.org/records/17212677. https://doi.org/10.1371/journal.pbio.3003477.s002 (TIF) S3 Fig. Exploring the spatial architecture of the human thymus. (A) (i) Clustering of spots from human thymus spatial transcriptomics (ST) data, with clusters color-coded. (ii) Hematoxylin and eosin (H&E) staining image of the thymus. (iii) Spatial location of each cluster on thymus ST section, highlighted by red color. (B) Uniform Manifold Approximation and Projection (UMAP) plot showing the single-cell atlas of human thymus. Each dot represents an individual cell, and cell types are marked by colors. (C) Spatial architectures of human thymus reconstructed using CytoSPACE (i), CellTrek (ii), Tangram (iii), and Seurat (iv). Each dot represents an individual cell. Cell types are marked by color codes. (D) Bar plot showing the number of cell types effectively mapped to spatial locations. “Reference” represents the total number of cell types in the single-cell atlas of the human thymus. The underlying data for this figure can be found at https://zenodo.org/records/17212677. https://doi.org/10.1371/journal.pbio.3003477.s003 (TIF) S4 Fig. Investigating the spatial architecture of the mouse kidney. (A) Clustering spots of mouse kidney spatial transcriptomics (ST) data. Clusters are marked by color codes (left). Hematoxylin and eosin (H&E) staining of the corresponding kidney section (right). (B) Spots within each cluster on the mouse kidney ST section highlighted in red. (C–F) Mapping of epithelial cell transcriptomes from the mouse kidney single-cell atlas onto spatial spots of the corresponding ST section. Left: reconstructed spatial architectures with cells displayed using jitter within assigned spots. Right: the same representations with cells colored according to their known distance from the inner medulla. (C) CytoSPACE; (D) CellTrek; (E) Tangram; (F) Seurat. (G) Scatterplot showing the consistency between the cellular proportions in spatial architectures reconstructed with various mapping tools and cellular compositions predicted by the CARD spatial deconvolution tool [11]. The blue line denotes the linear fit, and the shaded area represents the 95% confidence interval. Different colors of points indicate distinct cell types. “R” represents the Pearson correlation coefficient (PCC). P-values were obtained using two-sided t-tests. (H) Bar plot showing the number of cell types effectively mapped to spatial locations. “Reference” represents the total number of cell types in the single-cell atlas of the mouse kidney. The underlying data for this figure can be found at https://zenodo.org/records/17212677. https://doi.org/10.1371/journal.pbio.3003477.s004 (TIF) S5 Fig. Exploring the spatial architecture of the human dorsolateral prefrontal cortex (DPLFC). (A) Clustering spots of DPLFC spatial transcriptomics (ST) data. Clusters are marked by color codes (top). Hematoxylin and eosin (H&E) staining image of DLPFC tissue (bottom). (B) Uniform Manifold Approximation and Projection (UMAP) plot showing the single-cell atlas of human prefrontal cortex. Each dot represents an individual cell. Cell types are marked by color codes. (C) Spatial architectures of DLPFC tissue reconstructed using CellTrek (i), Tangram (ii), and Seurat (iii), respectively. Each dot represents an individual cell. Cell types are marked by color codes. (D) Heat map showing the distribution of cell types in anatomical structures of DLPFC. The intensity of the color indicates the cellular fraction within the specific anatomical region. CellTrek (left); Tangram (middle); Seurat (right). (E) Distribution of selected cell types in DLPFC tissue, reconstructed by different mapping tools. Each dot represents one cell. (F) Bar plot showing the number of cell types effectively mapped to spatial locations. “Reference” represents the total number of cell types in the single-cell atlas of the human prefrontal cortex. The underlying data for this figure can be found at https://zenodo.org/records/17212677. https://doi.org/10.1371/journal.pbio.3003477.s005 (TIF) S6 Fig. Spatial transcriptomics (ST) and single-cell atlases of multiple human tissues. (A) (i) Clustering spots of human lung ST data. (ii) Hematoxylin and eosin (H&E) staining image of human lung tissue. (iii) Uniform Manifold Approximation and Projection (UMAP) showing the single-cell atlas of the human lung under COVID-19 state. (B) (i) Clustering spots of human intestinal ST data. (ii) H&E image of human intestinal tissue. (iii) UMAP plot showing the single-cell atlas of human intestine, annotated based on the marker genes provided by Fawkner-Corbett and colleagues [9]. (iv) Heat map showing pair-wise correlations of annotated cell types. (C, D) Clustering spots of human breast ST data (i); H&E staining image of human breast tissue (ii) [8]. (C) BRCA.1; (D) BRCA.2. (E) UMAP showing the single-cell atlas of human breast cancer patients [8]. Each dot represents an individual cell, and cell types are marked by color codes. The underlying data for this figure can be found at https://zenodo.org/records/17212677. https://doi.org/10.1371/journal.pbio.3003477.s006 (TIF) S7 Fig. Comparative analysis of spatial architectures and cellular proportions in Lung, Breast, and Intestinal tissues recovered by different mapping tools. (A–D) Spatial architectures of human Lung (A), Breast (B and C), and Intestinal (D) tissues reconstructed by Cell2Spatial, CytoSPACE, CellTrek, Tangram, and Seurat. Each dot represents an individual cell and cell types are marked by color codes. (E–H) Scatter plots showing the consistency between the cellular proportions in spatial architectures reconstructed with various mapping tools and the cellular compositions predicted by CARD spatial deconvolution tool [11]. The blue line denotes the linear fit, and the shaded area represents the 95% confidence interval. Different colors of points indicate distinct cell types. “R” represents the Pearson correlation coefficient (PCC). P-values were obtained by two-sided t-tests. (E) Lung; (F) BRCA.1; (G) BRCA.2; (H) Intestinal. (I) Table summarizing the number of cell types effectively mapped to spatial locations by each tool. “Reference” represents the total number of cell types in the single-cell atlas of the human tissues. The underlying data for this figure can be found at https://zenodo.org/records/17212677. https://doi.org/10.1371/journal.pbio.3003477.s007 (TIF) S8 Fig. Characterization of cell distribution in mouse kidney and renal cell carcinoma (RCC) tissues using Cell2Spatial. (A) Hematoxylin and eosin (H&E) staining image of mouse kidney tissue. (B) Spatial positioning of immune cell types in RCC tissue sections inferred by Cell2Spatial. The panel with the red border shows the corresponding H&E image of the RCC tissue. (C) Spatial plot showing the expression of selected marker genes in mouse kidney spatial transcriptomics (ST) data. T cell (left); Fibroblasts (Fb, middle); and Natural Killer (NK) cells (right). https://doi.org/10.1371/journal.pbio.3003477.s008 (TIF) S9 Fig. Assessment of multiple parameters for Cell2Spatial performance and assignment strategy. (A) Boxplot showing the distribution of elapsed time for marker selection using Cell2Spatial’s strategy compared with the Wilcoxon test implemented in the Seurat framework. Each point represents one dataset. (B) Boxplot comparing end-to-end mapping accuracy of Cell2Spatial with and without spatial weighting, using mouse brain Visium HD data (8 µm, down-sampled to 50,000 spots). p-value was calculated by the two-sided Wilcoxon test. (C) Bar plot summarizing the frequency with which different quantile cutoffs produced the highest overlap of highly variable genes (HVGs), indicating the most stable cutoff across conditions. To generate these results, Cell2Spatial was applied to map single cells onto mouse kidney spatial transcriptomics (ST) data and reconstruct spatial expression profiles. HVGs were then computed from both reconstructed and original data across gene set sizes ranging from 50 to 3,000, and the consistency of HVG selection was quantified using the Jaccard index. This analysis was repeated across quantile cutoffs from 0.05 to 1 (step = 0.05). (D) Distribution of Getis-Ord G* indices for spatial spots corresponding to each cell type, derived from mouse kidney 10× Visium ST data. The black curve represents the observed Getis-Ord G* distribution, while the red curve shows the fitted normal distribution. (E) Scatter plots showing the predicted cellular compositions of spots by Cell2Spatial. The left panel displays results without smoothing using adjacent spots (k = 0), while the right panel incorporates smoothing with the five nearest neighboring spots (k = 5). True proportions are derived from a simulated ST dataset of the mouse brain. (F) Violin plots combined with boxplots showing the consistency between the estimated overall cellular compositions by Cell2Spatial and the simulated ground truth under different settings. The left panel shows root mean square error (RMSE) values, while the right panel shows Pearson correlation coefficients (PCCs). For each of 100 iterations, 1,000 spots were randomly sampled from the ST slice, and their aggregated cellular compositions were compared with the true proportions derived from the simulated dataset. (G) Schematic of the iterative strategy for preassigning single cells to ST clusters, utilizing a feedforward neural network (FNN) model. (H) Pseudocode illustrating the cell-to-spot assignment process. The underlying data for this figure can be found at https://zenodo.org/records/17212677. https://doi.org/10.1371/journal.pbio.3003477.s009 (TIF) S1 Table. Summarize the data information utilized in this study. https://doi.org/10.1371/journal.pbio.3003477.s010 (XLSX) S2 Table. Median k-distances to the Astro cell type depicted by various mapping tools. https://doi.org/10.1371/journal.pbio.3003477.s011 (XLSX) S3 Table. Assessing the predictive performance of cell counts in spots (Cell2Spatial vs. CytoSPACE). https://doi.org/10.1371/journal.pbio.3003477.s012 (XLSX) S4 Table. Accuracy metrics of mapping tools based on synthetic datasets. https://doi.org/10.1371/journal.pbio.3003477.s013 (XLSX) S5 Table. Distance of T cell subsets at different developmental stages relative to the thymic medullary center. https://doi.org/10.1371/journal.pbio.3003477.s014 (XLSX) S6 Table. Performance metrics for mapping tools derived from mouse kidney data. https://doi.org/10.1371/journal.pbio.3003477.s015 (XLSX) S7 Table. Accuracy metrics of mapping tools on spatial transcriptomics (ST) data from multiple tissues. https://doi.org/10.1371/journal.pbio.3003477.s016 (XLSX) S8 Table. Co-existence atlas of cell types in mouse kidney. https://doi.org/10.1371/journal.pbio.3003477.s017 (XLSX) S9 Table. Co-existence atlas of immune cell types in tertiary lymphoid structures (TLSs) region of renal cell carcinoma (RCC) patients. https://doi.org/10.1371/journal.pbio.3003477.s018 (XLSX) Acknowledgments We are grateful to Prof. Chuanle Xiao for providing valuable suggestions that helped improve this manuscript.

Hodgson, Evie R.;Hayward, Jenni A.;Leonard, Rachel A.;Makota, Fadzai Victor;van Dooren, Giel G.

doi: 10.1371/journal.pbio.3003520pmid: 41289326

Introduction Iron–sulfur (FeS) clusters are among nature’s most ancient yet versatile cofactors [1]. They serve as prosthetic groups for proteins involved in a myriad of fundamental cellular processes, such as in electron transport chains, ribosome assembly, DNA repair and synthesis, and iron homeostasis [1,2]. FeS clusters cannot be scavenged from the environment and, instead, require dedicated machinery to facilitate their biosynthesis. In eukaryotes, distinct biosynthetic pathways are expressed in the specific subcellular compartments that harbor FeS cluster-requiring proteins [3]. Mitochondrial FeS proteins obtain clusters synthesized by the so-called Iron Sulfur Cluster (ISC) pathway. The Sulfur Mobilization pathway, a similar but functionally distinct pathway, supplies clusters to plastid-localized FeS proteins. Finally, the so-called Cytosolic Iron–sulfur Assembly (CIA) pathway generates [4Fe-4S] clusters in the cytosol and distributes them to both cytosolic and nuclear FeS proteins. The synthesis and maturation of cytosolic FeS clusters depend on the provision of a sulfur- (and possibly iron-) containing product from the mitochondrial ISC pathway [4,5]. Electrons required for the assembly of cytosolic FeS clusters are derived from NADPH via an electron transport chain consisting of the diflavin reductase, Tah18 (also referred to as NDOR1 or ATR3), and the FeS protein, Dre2 (also referred to as CIAPIN1; [5]). In most eukaryotes, the integration of iron, sulfur, and electrons into cytosolic [4Fe-4S] clusters occur on a scaffold complex consisting of the P-loop NTPases NBP35 and Cfd1 [6,7]. Once assembled, [4Fe-4S] clusters are thought to be trafficked by the Nar1 protein (also referred to as IOP1, NARFL or CIAO3) to the so-called CIA Targeting Complex (CTC; [8–11]). The CTC is comprised of the proteins CIA1 (also referred to as CIAO1), CIA2 (also referred to as CIAO2B, MIP18 or AE7), and MMS19 (also referred to as Met18) and facilitates the incorporation of FeS clusters into cytosolic and nuclear apo-FeS proteins, thereby serving as a bridge between the FeS cluster carrier Nar1 and recipient FeS proteins [6,10,12]. Apicomplexans constitute a large phylum of protozoan parasites that impart considerable health and socio-economic burdens globally. Apicomplexans are a part of the larger Alveolate superphylum of eukaryotes, which also include ciliates (e.g., Paramecium tetraurelia and Tetrahymena thermophilus), dinozoans (e.g., the photosynthetic coral symbiont Symbiodinium microadriaticum and the oyster parasite Perkinsus marinus), and a recently identified group known as the chrompodellids (e.g., Chromera velia and Vitrella brassicaformis). Within the alveolates, the apicomplexans, chrompodellids and dinozoans form a clade known as the myzozoans [13,14]. Key apicomplexan parasites include those of the Plasmodium genus, the causative agents of malaria in humans, and Toxoplasma gondii, the causative agent of toxoplasmosis. The core proteins involved in the CIA pathway, including all three proteins of the CTC, are conserved in apicomplexans [15]. We demonstrated recently that the T. gondii homolog of the CIA scaffold protein NBP35 (TgNBP35) was essential for parasite survival [15]. In contrast to its cytosolic localization in most other eukaryotes, we found that TgNBP35 was anchored to the outer face of the outer mitochondrial membrane courtesy of an N-terminal transmembrane domain (TMD). The N-terminal TMD of TgNBP35 is conserved throughout myzozoan NBP35 homologs, but absent from ciliates, where the NBP35 homolog localizes to the cytosol [15,16]. These findings suggest that the assembly of cytosolic [4Fe-4S] clusters occurs on the cytosolic face of the mitochondrion of myzozoans. The subcellular localization and importance of the remaining CIA pathway proteins has not been studied previously in myzozoans, and we set out to test this using T. gondii as our experimental model. We found that most candidate proteins of the T. gondii CIA pathway localize to the parasite mitochondrion and are critical for parasite proliferation. Curiously, we found that the CTC proteins TgCIA1, TgCIA2, and TgMMS19 exhibit a dual localization to the mitochondrion and cytosol, with mitochondrial localization of this complex mediated by a loop on the TgCIA1 protein that is conserved throughout myzozoans, and which is critical for its role in parasite biology. Taken together, our study reveals that the mitochondrion acts as a hub for the CIA pathway in Toxoplasma and related organisms. Results Cytosolic FeS cluster assembly occurs at the mitochondrion of Toxoplasma gondii We set out to determine the subcellular localization of the previously uncharacterized candidate CIA pathway proteins in T. gondii. Using CRISPR/Cas9 genome editing, we introduced a 3× hemagglutinin (HA) epitope tag into the 3′ end of the coding sequences of the proposed electron transfer chain proteins TgTah18 (www.ToxoDB.org gene identifier TGGT1_249320; [17]) and TgDre2 (TGGT1_216900), and all three of the proposed CTC proteins, TgCIA1 (TGGT1_313280), TgCIA2 (TGGT1_306590) and TgMMS19 (TGGT1_222230), in an anhydrotetracycline (ATc)-regulatable TgNBP35-cMyc strain that we had generated previously [15]. We validated correct integration of the tag in each parasite line by PCR analysis (S1A–S1E Fig). We were unsuccessful in integrating a HA tag into the 3′ end of the coding sequence for the proposed FeS transfer protein TgNar1 (TGGT1_242580). Instead, we introduced a HA-mini-Auxin Inducible Degron (HA-mAID) tag at the 5′ end of the coding region of TgNar1 in a tdTomato and Tir1-expressing parasite strain (RH∆ku80/Tir1-FLAG/tdTomato; [18]; S1F Fig). To test whether each protein was expressed in the disease-causing tachyzoite stage of the parasite life cycle, we extracted proteins from the TgTah18-HA, TgDre2-HA, HA-mAID-TgNar1, TgCIA1-HA, TgCIA2-HA, and TgMMS19-HA lines, separated them by Sodium Dodecyl Sulfate-Polyacrylamide Gel Electrophoresis (SDS-PAGE), and performed western blotting with anti-HA antibodies. All six proteins were expressed and were of approximately the expected masses for the epitope-tagged proteins (Fig 1A–1F). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. The CIA pathway localizes to the mitochondrion of Toxoplasma gondii. (A–F) Western blots of proteins extracted from (A) TgTah18-HA, (B) TgDre2-HA, (C) HA-mAID-TgNar1, (D) TgCIA1-HA, (E) TgCIA2-HA, or (F) TgMMS19-HA expressing parasites, separated by SDS-PAGE and probed with anti-HA antibodies. TgMMS19-HA appears as two separate bands, the lower of which may represent a degradation product. (G–L) Immunofluorescence assays of (G) TgTah18-HA, (H) TgDre2-HA, (I) HA-mAID-TgNar1, (J) TgCIA1-HA, (K) TgCIA2-HA, or (L) TgMMS19-HA expressing parasites, probed with anti-HA antibodies (green) to label the proteins-of-interest and anti-TgTom40 antibodies (magenta) to label the mitochondrion. Scale bars are 2 µm. DIC, differential interference contrast transmission image. Right, corresponding fluorescence profiles depicting the intensity of anti-HA (green) and anti-TgTom40 (magenta) labeling along the yellow line on merged images. The numerical data underlying this Figure can be found in S1 Data. (M) Model for the CIA pathway in T. gondii parasites, with Tah18, NBP35 and Nar1 positioned on the outer face of the mitochondrion, Dre2 in the cytosol, and the CIA Targeting Complex (CTC; consisting of CIA1, CIA2 and MMS19), exhibiting dual cytosolic and mitochondrial localization. Electrons (e−) are sourced from NADPH via an electron transfer chain consisting of Tah18 and Dre2. Sulfur (yellow) and possibly iron (red) are sourced from the iron–sulfur cluster (ISC) synthesis pathway in the mitochondrion, assembled as [4Fe-4S] clusters on the NBP35 scaffold, then trafficked via Nar1 to the CTC. The CTC donates FeS clusters to client FeS proteins. https://doi.org/10.1371/journal.pbio.3003520.g001 To determine the subcellular localization of each protein, we performed immunofluorescence assays. We found that TgTah18-HA and HA-mAID-TgNar1 both co-localized with the mitochondrial marker TgTom40, whereas TgDre2-HA localized throughout the parasite cytosol (Fig 1G–1I). Curiously, all three proteins of the putative CTC exhibited a dual localization, overlapping with TgTom40 in the mitochondrion and also exhibiting diffuse labeling through the cytosol (Fig 1J–1L). Taken together, these data indicate that all the putative CIA pathway proteins in T. gondii, with the exception of TgDre2 but including the previously characterized scaffold protein TgNBP35 [15,16], localize to the mitochondrion. We showed previously that TgNBP35 localizes to the cytosolic face of the outer mitochondrial membrane [15], and our data are therefore consistent with the CIA pathway occurring on the cytosolic face of the outer mitochondrial membrane in T. gondii (Fig 1M). The CIA pathway is critical for parasite proliferation and protein synthesis in Toxoplasma gondii We set out to test the importance of each of the candidate CIA pathway proteins for parasite proliferation. Using CRISPR/Cas9 genome editing, we introduced mAID-HA epitope tags at the 3′ ends of the coding regions for TgTah18, TgDre2, TgCIA1, TgCIA2 and TgMMS19 of tdTomato/Tir1 parasites (S2 Fig). The mAID tagging system facilitates proteasomal degradation of mAID-fused proteins upon the addition of the small molecule 3-indoleacetic acid (IAA; [19]). Since the abundance of the resulting proteins can be conditionally regulated, we termed the resulting lines expressing mAID-tagged CIA pathway proteins regulatable (r)-TgTah18-mAID-HA, rTgDre2-mAID-HA, rHA-mAID-TgNar1, rTgCIA1-mAID-HA, rTgCIA2-mAID-HA, and rTgMMS19-mAID-HA. To determine whether the mAID-tagged proteins could be regulated upon the addition of IAA, we cultured each parasite line in the absence or presence of IAA for 24 h. We extracted the resulting proteins, separated them by SDS-PAGE and performed western blotting. We found that the abundances of all six candidate CIA pathway proteins were substantially depleted upon IAA addition (Fig 2A–2F). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Candidate CIA pathway proteins are important for parasite proliferation. (A–G) Fluorescence proliferation assays and western blotting of (A) rTgTah18-mAID-HA, (B) rTgDre2-mAID-HA, (C) rHA-mAID-TgNar1, (D) rTgCIA1-mAID-HA, (E) rTgCIA2-mAID-HA, (F) rTgMMS19-mAID-HA, and (G) WT (RH∆ku80/Tir1-FLAG/tdTomato) parasites cultured in the absence (black) or presence (red) of IAA. Parasite proliferation is expressed as a percentage of the final fluorescence measurement in the -IAA condition for each line. Individual data points and error bars represent the mean ± SD of three technical replicates. Data are representative of three independent experiments. Error bars not visible are smaller than the symbol. The numerical data underlying this Figure can be found in S1 Data. For the western blotting, parasites were cultured in the absence (−) or presence (+) of IAA for 24 h, with protein samples separated by SDS-PAGE then subjected to western blotting with anti-HA antibodies to detect the mAID-HA-tagged protein of interest and anti-TgTom40 antibodies as a loading control (A–F, left). https://doi.org/10.1371/journal.pbio.3003520.g002 To explore the contribution of each CIA pathway protein to parasite proliferation, we performed fluorescence-based proliferation assays, measuring tdTomato fluorescence in parasites across time as a function of parasite proliferation. We found that in the absence of IAA, all parasite lines proliferated normally, exhibiting a typical sigmoidal growth curve (Fig 2). The parental RH∆ku80/Tir1-FLAG/tdTomato (WT) line proliferated normally in the presence of IAA, indicating that IAA is not toxic to parasites (Fig 2G). By contrast, the rTgTah18-mAID-HA, rTgDre2-mAID-HA, rHA-mAID-TgNar1, rTgCIA1-mAID-HA, rTgCIA2-mAID-HA, and rTgMMS19-mAID-HA lines all exhibited a severe proliferation defect when cultured in the presence of IAA (Fig 2A–2F), with the phenotype upon the depletion of TgNar1 consistent with that reported in another recent study [20]. These data indicate that TgTah18, TgDre2, TgNar1, TgCIA1, TgCIA2 and TgMMS19 are all critical for parasite proliferation. Next, we wanted to explore whether the candidate proteins function in the CIA pathway. Given its curious dual localization to the mitochondrion and cytosol, we initially focused on the CTC protein TgCIA1. Previous studies have found that impairment of the cytosolic FeS assembly pathway in T. gondii leads to a depletion in the abundances of some cytosolic and nuclear FeS cluster-containing proteins, including the ribosome recycling protein TgABCE1 [15,20–22]. We integrated a 3× cMyc tag at the 3′ end of the TgABCE1 open reading frame in the rTgCIA1-mAID-HA line using CRISPR/Cas9 genome editing (S3A and S3B Fig). We cultured these parasites for 0, 24, or 48 hours on IAA, separated proteins by SDS-PAGE, and performed western blotting to detect the TgABCE1-cMyc protein. We found a small but significant decrease in TgABCE1-cMyc abundance after 24 h of IAA incubation and a strong ~75% decrease after 48 h (Fig 3A and 3B). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Characterizing the effects of depleting candidate CIA pathway proteins in Toxoplasma gondii parasites. (A) Western blots of proteins extracted from rTgCIA1-mAID-HA/TgABCE1-cMyc parasites cultured for 0, 24 or 48 h in IAA and separated by SDS-PAGE. Samples were probed with anti-HA, anti-cMyc, and anti-TgTom40 antibodies. (B) Relative abundance of the TgABCE1-cMyc protein in the western blot was determined as a percentage of the 0 h control, with abundances normalized using the TgTom40 loading control. Data points represent the mean ± SD of five independent experiments for the 0 and 24 h conditions and three independent experiments for the 48 h condition. Data were analyzed using a one-way ANOVA followed by Tukey’s multiple comparisons test, with p values shown. (C) Western blots measuring the incorporation of puromycin into proteins from WT (RH∆ku80-Tir1-FLAG/tdTomato) or rTgCIA1-mAID-HA parasites cultured in the absence (−) or presence (+) of IAA for 24 h, separated by SDS-PAGE and probed with anti-puromycin, anti-HA, and anti-TgTom40 antibodies. (D) Relative abundance of puromycin incorporation was determined as a percentage of the -IAA control for each parasite line, with abundances normalized using the TgTom40 loading control. Data points represent the mean ± SD of three independent experiments. Data were analyzed using a one-way ANOVA followed by Tukey’s multiple comparisons test, with relevant p values shown. (E) Multiple sequence alignment of a region of the T. gondii CIA2 protein (TgCIA2) with CIA2 homologs in Saccharomyces cerevisiae (ScCIA2), Homo sapiens (HsCIA2), and Drosophila melanogaster (DmCIA2). The cysteine residue proposed to function in FeS cluster binding is highlighted by a red box. (F) Western blot of proteins extracted from rTgCIA2-mAID-HA/cTgCIA2WT-Ty1 or rTgCIA2-mAID-HA/cTgCIA2C524A-Ty1 parasites, separated by SDS-PAGE and probed with anti-Ty1 antibodies to detect the cTgCIA2WT-Ty1 and cTgCIA2C524A-Ty1 proteins, and anti-TgTom40 antibodies as a loading control. (G) Fluorescence proliferation assays of rTgCIA2-mAID-HA, rTgCIA2-mAID-HA/cTgCIA2WT-Ty1, or rTgCIA2-mAID-HA/cTgCIA2C524A-Ty1 parasites cultured in the absence (black) or presence (red) of IAA. Parasite proliferation is expressed as a percentage of the -IAA condition on the final day of the experiment for each parasite line, with values depicting the mean ± SD of three technical replicates. Error bars not visible are smaller than the symbol. Data are representative of three independent experiments. The numerical data underlying this Figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003520.g003 ABCE1 and other cytosolic FeS proteins are important for protein translation from cytosolic ribosomes. Previous studies have found that depletion of proteins that contribute to the CIA pathway result in a global impairment of protein synthesis in T. gondii [20,21]. We cultured WT and rTgCIA1-mAID-HA parasites in the absence or presence of IAA for 24 h, then treated parasites with the protein synthesis inhibitor puromycin for 15 min. Puromycin becomes incorporated into nascent polypeptides during translation and can be detected by anti-puromycin antibodies, thus providing a measure for newly synthesized proteins. We separated proteins from puromycin-treated parasites using SDS-PAGE and performed western blotting. We observed a significant ~80% depletion in puromycin-labeled proteins in IAA-treated rTgCIA1-mAID-HA parasites, whereas puromycin labeling was unchanged in IAA-treated WT parasites (Fig 3C and 3D). We next performed puromycin incorporation assays in rTgTah18-mAID-HA, rTgDre2-mAID-HA, rHA-mAID-TgNar1, rTgCIA2-mAID-HA and rTgMMS19-mAID-HA parasites cultured for 24 h in the absence or presence of IAA. We observed significantly less puromycin incorporation upon the depletion of TgTah18-mAID-HA, HA-mAID-TgNar1, TgCIA2-mAID-HA and TgMMS19-mAID-HA but not upon the depletion of TgDre2-mAID-HA (S4A and S4B Fig). To test whether TgDre2 knockdown leads to the depletion of TgABCE1, we integrated a cMyc epitope tag into the TgABCE1 locus of rTgDre2-mAID-HA parasites (S3C Fig). We cultured these parasites for 0, 24, or 48 hours on IAA, separated proteins by SDS-PAGE, and performed western blotting to detect the TgABCE1-cMyc protein. We found a ~40% decrease in TgABCE1-cMyc abundance after 24 h of IAA incubation and a ~60% decrease after 48 h (S4C Fig). Loss of TgDre2, therefore, does impact the abundance of the cytosolic Fe–S protein TgABCE1, leaving open the possibility that TgDre2 has a role in cytosolic Fe–S synthesis. Taken together, our data indicate that the putative CIA pathway proteins of T. gondii are essential for parasite proliferation, and that most of the candidate CIA pathway proteins in T. gondii (with the possible exception of TgDre2) are critical for protein translation, a process that relies on cytosolic FeS proteins like ABCE1. The CIA2 protein in animals and fungi contains a reactive cysteine residue that is thought to function in FeS cluster binding and facilitating the transfer of FeS clusters to client proteins [10,23,24]. This cysteine residue has been shown to be critical for CIA pathway function and cell survival in these eukaryotes [23,24]. Alignments of yeast and animal CIA2 proteins with TgCIA2 revealed that the reactive cysteine residue is conserved as residue 524 in the TgCIA2 protein (Figs 3E and S5). We hypothesized that if TgCIA2 functions like its counterpart in other eukaryotes, mutating the TgCIA2C524 residue would ablate CIA pathway function and, subsequently, lead to impaired parasite proliferation. We constitutively expressed either Ty1 epitope-tagged WT TgCIA2 (constitutive (c)TgCIA2WT-Ty1) or a Ty1-tagged TgCIA2 mutant wherein the cysteine at residue 524 was mutated to Ala (cTgCIA2C524A-Ty1) in rTgCIA2-mAID-HA parasites. Western blotting revealed that both Ty1-tagged TgCIA2 proteins were expressed at similar levels (Fig 3F) and fluorescence proliferation assays revealed that expression of cTgCIA2WT-Ty1 fully rescued parasite proliferation upon knockdown of the TgCIA2-mAID-HA protein (Fig 3G). By contrast, we observed minimal proliferation of rTgCIA2-mAID-HA/cTgCIA2C524A-Ty1 parasites cultured in the presence of IAA (Fig 3G). These data indicate that the proposed FeS cluster binding Cys-524 residue of TgCIA2 is critical for its function, consistent with the hypothesis that the TgCIA2 protein of T. gondii functions in a similar manner to the CIA2 protein of other eukaryotes in coordinating FeS cluster binding on the CTC [10,23,24]. TgCIA1, TgCIA2, and TgMMS19 are components of the same protein complex We set out to explore the candidate CTC proteins of T. gondii in more detail. First, we asked whether the TgCIA1, TgCIA2 and TgMMS19 proteins exist in a protein complex. We extracted proteins from TgCIA1-HA, TgCIA2-HA and TgMMS19-HA expressing parasites, separated them by blue native (BN)-PAGE, and performed western blotting with anti-HA antibodies. All three proteins were present in a complex of slightly larger than 720 kDa (Fig 4A, black arrowhead). The TgCIA1-HA protein, but not the others, was also observed in smaller protein complexes of ~400 and ~240 kDa (Fig 4A, red and blue arrowheads). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. Characterizing the protein composition of the CIA Targeting Complex (CTC) in Toxoplasma gondii parasites. (A) Western blot of proteins extracted from TgCIA1-HA, TgCIA2-HA and TgMMS19-HA expressing parasites, separated by BN-PAGE and probed with anti-HA antibodies or anti-TgTom40 as a loading control. (B, D) Western blots of proteins extracted from rTgCIA1-mAID-cMyc/TgCIA2-HA and rTgCIA1-mAID-cMyc/TgMMS19-HA parasites cultured in the absence (-) or presence (+) of IAA for 24 h, separated by either SDS-PAGE (B) or BN-PAGE (D) and probed with anti-HA, anti-cMyc or anti-TgTom40 antibodies. Data are representative of three independent experiments. (C) Quantification of the western blots depicted in (B), depicting relative abundance of the TgCIA2-HA or TgMMS19-HA proteins determined as a percentage of the -IAA control, with abundances normalized using the TgTom40 loading control. Data points represent the mean ± SD of three independent experiments. Data were analyzed using a paired t test with p values shown. The numerical data underlying this Figure can be found in S1 Data. (E, F) Western blot of proteins extracted from rTgCIA2-mAID-HA/TgCIA1-smFP-cMyc and rTgMMS19-mAID-HA/TgCIA1-smFP-cMyc parasites, cultured for 24 h in the absence (−) or presence (+) of IAA, separated by either (E) SDS-PAGE or (F) BN-PAGE and probed with anti-cMyc, anti-HA, or anti-TgTom40 antibodies. Data are representative of three independent experiments. In the BN-PAGE western blots (A, D, and F), the black arrowheads indicate the >720 kDa candidate CIA Targeting Complex; red and blue arrowheads indicate the lower mass complexes containing the TgCIA1 protein. https://doi.org/10.1371/journal.pbio.3003520.g004 To test whether the other candidate CIA pathway proteins were present in protein complexes, we also undertook BN-PAGE western blotting on proteins extracted from the rTgTah18-mAID-HA, rTgDre2-mAID-HA and rHA-mAID-TgNar1 lines. We observed TgTah18-mAID-HA in a protein complex of ~330 kDa, but were unable to clearly detect the TgDre2-mAID-HA or HA-mAID-TgNar1 proteins in complexes (S6A Fig). These data suggest that TgTah18, TgDre2, and TgNar1 are not components of the >720 kDa CTC. If the TgCIA1, TgCIA2, and TgMMS19 proteins exist in the same complex, we reasoned that depleting one would result in alterations to the protein complex in which the other two proteins reside. We initially introduced Ty1-epitope tags into the native TgCIA2 and TgMMS19 loci of the rTgCIA1-mAID-HA line, but were subsequently unable to detect robust TgCIA2-Ty1 and TgMMS19-Ty1 expression, possibly because the antibody we have to the Ty1 epitope is not of sufficiently high affinity. Since we knew we could detect the HA-tagged versions of these proteins, we generated a new IAA-regulatable TgCIA1-expressing line by fusing a mAID-cMyc tag into the 3′ region of the TgCIA1 open reading frame (S7A Fig). We then introduced a HA tag into the 3′ region of the TgCIA2 or TgMMS19 open reading frames of this parasite line (S7B and S7C Fig). We cultured the resulting rTgCIA1-mAID-cMyc/TgCIA2-HA and rTgCIA1-mAID-cMyc/TgMMS19-HA parasites in the absence or presence of IAA for 24 h, separated proteins by SDS-PAGE, and performed western blotting. We observed robust depletion of the TgCIA1-mAID-cMyc protein upon the addition of IAA (Fig 4B). We also a observed an ~80% depletion of the TgCIA2-HA protein, and a small but significant ~20% depletion of the TgMMS19-HA protein (Fig 4B and 4C). To determine the timing of the observed TgCIA2-HA depletion upon TgCIA1 knockdown, we cultured rTgCIA1-mAID-cMyc/TgCIA2-HA parasites in IAA for a range of times between 0 and 12 h. We observed a depletion of the TgCIA1-mAID-cMyc protein as soon as 3 h after IAA addition, and a concomitant and significant decrease in the abundance of the TgCIA2-HA protein (S6B Fig). These data indicate that TgCIA1 knockdown impacts the stability of the TgCIA2 and, to a lesser extent, the TgMMS19 proteins. We next extracted proteins from rTgCIA1-mAID-cMyc/TgCIA2-HA and rTgCIA1-mAID-cMyc/TgMMS19-HA parasites cultured in the absence or presence of IAA for 24 h, separated proteins by BN-PAGE, and performed western blotting. TgCIA1-mAID-cMyc knockdown resulted in a depletion and reduced molecular mass in both the TgCIA2-HA- and TgMMS19-HA-containing protein complexes (Fig 4D), with the depletion of the TgCIA2-HA complex occurring concomitantly with TgCIA1-mAID-cMyc knockdown (S6C Fig). We next undertook the reverse experiment, asking whether depletion of TgCIA2 or TgMMS19 resulted in changes in the abundance of TgCIA1 and impairment of the TgCIA1-containing protein complexes. We integrated a spaghetti monster-fluorescent protein-cMyc (smFP-cMyc) tag into the TgCIA1 locus of rTgCIA2-mAID-HA and rTgMMS19-mAID-HA parasites (S7D Fig). We cultured the rTgCIA2-mAID-HA/TgCIA1-smFP-cMyc and rTgMMS19-mAID-HA/TgCIA1-smFP-cMyc lines in the absence or presence of IAA for 24 h and performed both SDS-PAGE and BN-PAGE western blotting. We observed no changes to TgCIA1-smFP-cMyc protein abundance upon the knockdown of either TgCIA2-mAID-HA or TgMMS19-mAID-HA (Fig 4E). However, we observed a complete loss of the >720 kDa TgCIA1-smFP-cMyc-containing complex upon the knockdown of both TgCIA2-mAID-HA and TgMMS19-mAID-HA (Fig 4F, black arrowhead). The two smaller TgCIA1-containing complexes were unaffected by TgCIA2 or TgMMS19 depletion (Fig 4F, red and blue arrowheads). As a direct test for whether TgCIA1 and TgCIA2 interact, we performed co-immunoprecipitation experiments on the rTgCIA2-mAID-HA/TgCIA1-smFP-cMyc parasite line. Immunoprecipitation of TgCIA1-smFP-cMyc with anti-cMyc beads co-purified TgCIA2-mAID-HA but not the mitochondrial outer membrane protein TgTom40, and immunoprecipitation of TgCIA2-mAID-HA with anti-HA beads co-purified TgCIA1-smFP-cMyc but not TgTom40 (S6D Fig), providing further evidence that TgCIA1 and TgCIA2 are components of the same protein complex. Taken together, our data are consistent with TgCIA1, TgCIA2, and TgMMS19 existing in the same complex of >720 kDa. The CTC of animals exists as a heteromeric complex consisting of two copies of each of the three proteins [10]. The combined mass of a similarly arranged complex in T. gondii would be ~850 kDa, which is conceivably the mass we observe for the TgCIA1, TgCIA2, and TgMMS19 complexes in BN-PAGE. Our data also indicate that TgCIA1 exists in smaller protein complexes that do not include TgCIA2 or TgMMS19. A novel loop in the TgCIA1 protein facilitates the dual cytosolic and mitochondrial localization of the CIA targeting complex in T. gondii The candidate CTC proteins of T. gondii exhibit an interesting dual localization to the mitochondrion and cytosol that is suggestive of a dynamic role for this complex in parasite biology. We set out to characterize the mitochondrial targeting of the CTC in more detail. We first asked whether TgCIA1 was required for the mitochondrial targeting of TgCIA2 and TgMMS19. We performed immunofluorescence assays on rTgCIA1-mAID-cMyc/TgCIA2-HA or rTgCIA1-mAID-cMyc/TgMMS19-HA parasites cultured for 3 h in the absence or presence of IAA, probing for the HA-tagged TgCIA2 and TgMMS19 proteins. As expected, both TgCIA2-HA and TgMMS19-HA exhibited dual mitochondrial and cytosolic localization in parasites cultured in the absence of IAA (Fig 5A and 5B). Notably, however, the mitochondrial localization of both TgCIA2-HA and TgMMS19-HA was significantly reduced upon TgCIA1-mAID-cMyc depletion (Fig 5A and 5B). This indicates that both TgCIA2 and TgMMS19 depend on TgCIA1 for their mitochondrial targeting. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. Localization of the CTC to the mitochondrion of Toxoplasma gondii is mediated by TgCIA1. (A–D) Immunofluorescence assays of (A) rTgCIA1-mAID-cMyc/TgCIA2-HA, (B) rTgCIA1-mAID-cMyc/TgMMS19-HA, (C) rTgCIA2-mAID-HA/TgCIA1-smFP-cMyc, and (D) rTgMMS19-mAID-HA/TgCIA1-smFP-cMyc parasites, cultured in the absence (top) or presence (bottom) of IAA for either 3 h (A, B) or 24 h (C, D). Samples were probed with anti-HA (A, B) or anti-cMyc (C, D) antibodies to detect the protein-of-interest (green) and anti-TgTom40 antibodies to detect the mitochondrion (magenta). Scale bars are 2 µm. DIC, differential interference contrast. Middle, corresponding fluorescence profiles depicting the intensity of anti-HA or anti-cMyc labeling (green) and the anti-TgTom40 labeling (magenta) along the yellow line on merged images. Right, the correlation between the protein of interest and TgTom40 was quantified using the Pearson correlation coefficient (r) and analyzed using an unpaired t test, with the p values shown. The numerical data underlying this Figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003520.g005 We next asked whether the mitochondrial localization of TgCIA1 is dependent on either TgCIA2 or TgMMS19. We cultured rTgCIA2-mAID-HA/TgCIA1-smFP-cMyc or rTgMMS19-mAID-HA/TgCIA1-smFP-cMyc parasites in the absence or presence of IAA for 24 h and performed immunofluorescence assays. Neither the depletion of TgCIA2-mAID-HA nor TgMMS19-mAID-HA significantly altered the localization of TgCIA1-smFP-cMyc (Fig 5C and 5D), indicating that the mitochondrial localization of TgCIA1 is not dependent on either TgCIA2 or TgMMS19. Taken together, these data indicate that targeting of the CTC to the mitochondrion of T. gondii is mediated by TgCIA1. Next, we asked whether TgCIA1 targeting to the mitochondrion is the result of interactions with other mitochondrially-localized components of the CIA pathway. We showed previously that TgNBP35, the proposed scaffold of the CIA pathway, is anchored to the mitochondrial outer membrane courtesy of an N-terminal TMD [15]. We cultured rTgNBP35-cMyc/TgCIA1-HA parasites in the absence or presence of ATc for 2 days to deplete TgNBP35-cMyc abundance, then performed immunofluorescence assays with anti-HA antibodies to determine TgCIA1-HA localization. We found that TgCIA1-HA continued to localize to the mitochondrion following TgNBP35-cMyc depletion (S8A Fig). We next tested whether TgTah18, another mitochondrially-localized component of the CIA pathway in T. gondii (Fig 1G), is important for the mitochondrial localization of TgCIA1. We integrated a smFP-cMyc tag into the TgCIA1 locus of rTgTah18-mAID-HA parasites (S8B Fig). We cultured rTgTah18-mAID-HA/TgCIA1-smFP-cMyc parasites in the absence or presence of IAA for 24 h and performed immunofluorescence assays. We found that the localization of TgCIA1-smFP-cMyc was unchanged upon TgTah18-mAID-HA depletion (S8C Fig). Taken together, these data indicate that the mitochondrial targeting of TgCIA1 is independent of TgNBP35 and TgTah18. A recent study found that Nar1 homologs from other eukaryotes harbor a C-terminal amino acid motif that facilitates Nar1 interaction with the CTC [25]. This motif, which includes a C-terminal tryptophan residue, appears to be conserved in TgNar1 (S8D Fig), which could explain why we were unable to C-terminally tag this protein. To test whether mitochondrially-localized TgNar1 facilitates the mitochondrial localization of TgCIA1, we introduced a smFP-cMyc tag into the TgCIA1 locus of rHA-mAID-TgNar1 parasites (S8B Fig). We cultured the resulting rHA-mAID-TgNar1/TgCIA1-smFP-cMyc parasites in the absence or presence of IAA for 24 h and performed both SDS-PAGE and BN-PAGE western blotting. Depleting HA-mAID-TgNar1 did not affect the abundance of the TgCIA1-smFP-cMyc protein or the formation of the TgCIA1-smFP-cMyc-containing protein complexes (S8E and S8F Fig). We were also unable to detect TgCIA1-smFP-cMyc protein upon immunoprecipitation of HA-mAID-TgNar1 protein (S8G Fig), suggesting that these proteins do not stably interact. This is consistent with our BN-PAGE data that TgNar1 is not part of the CTC (S6A Fig), although our data do not rule out that transient interactions occur between these proteins. Finally, we cultured rHA-mAID-TgNar1/TgCIA1-smFP-cMyc parasites in the absence or presence of IAA for 24 h and performed immunofluorescence assays to detect the TgCIA1-smFP-cMyc protein. We found that the dual mitochondrial/cytosolic localization of TgCIA1-smFP-cMyc was unchanged upon HA-mAID-TgNar1 depletion (S8H Fig). Taken together, these data indicate that the mitochondrial localization of TgCIA1 is not due to interactions with mitochondrially localized TgNar1. We looked to the AlphaFold2 predicted structure of TgCIA1 for clues into what could be mediating the targeting of TgCIA1 (and by extension, the CTC) to the mitochondrion. CIA1 belongs to the WD40 protein family, which commonly mediate protein-protein interactions [26,27]. The CIA1 proteins from yeast and animals contain characteristic WD40 repeat domains folded into a 7-bladed β-propeller arranged around a central axis (Fig 6A; [10,28]). Each WD40 domain (or “blade”) is comprised of four antiparallel β-strands (denoted A–D in Fig 6A, pink) with small loops of 5–12 amino acids connecting the C and D strands (CD loops; Fig 6A; [10,28]). In the AlphaFold2 predicted structure of TgCIA1, the 7-bladed β-propeller is structurally conserved. However, the TgCIA1 protein contains three large insertions compared to the yeast and animal proteins (Figs 6A and S9; [10,28]). These insertions are located in the CD loops of the first, third, and fifth β-propeller blades, and we termed these the CD1, CD3, and CD5 loops, respectively (Fig 6A). The CD1 loop is 122 amino acids long in TgCIA1, the CD3 loop is 82 amino acids long, and the CD5 loop is 289 amino acids (S9 Fig). Extended CD loops are also present in the CIA1 homologs of other myzozoans (S9 Fig), a eukaryotic taxon that includes apicomplexans and their closest free-living relatives such as chrompodellids and dinozoans [13,14]. This is most apparent in the CD5 loop of myzozoan CIA1 proteins, which ranges in length from 45 amino acids in the oyster parasite P. marinus to 313 amino acids in the malaria-causing parasite P. falciparum (S9 Fig). This is in contrast to the much shorter, six-amino acid-long CD5 loop of CIA1 in the ciliate Paramecium tetraurelia, a sister taxon to the myzozoans. The extended CD loops in the TgCIA1 structure are predicted with poor confidence by AlphaFold2, and for the most part lack clear structural elements (Fig 6A). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 6. An elongated CD5 loop is necessary for targeting TgCIA1 to the mitochondrion of Toxoplasma gondii. (A) Structure of the Saccharomyces cerevisiae protein ScCIA1 (PDB: 2HES) and the predicted AlphaFold2 structure of TgCIA1(UniProt ID: A0A125YRQ0), with the A, B, C, and D strands of the third propellor blade in the ScCIA1 protein labeled in pink. The loops connecting the C and D strands of blade 1 (CD1 loop; red), blade 3 (CD3 loop, yellow), and blade 5 (CD5 loop; green) in each structure were colored using EzMol software [63]. (B, E) Western blots of proteins extracted from rTgCIA1-mAID-HA/cTgCIA1WT-Ty1 (WT-Ty1), rTgCIA1-mAID-HA/cTgCIA1ScCD1-Ty1 (ScCD1-Ty1), rTgCIA1-mAID-HA/cTgCIA1ScCD3-Ty1 (ScCD3-Ty1), and rTgCIA1-mAID-HA/cTgCIA1ScCD5-Ty1 (ScCD5-Ty1) parasites cultured in the absence or presence of IAA for 24 h, separated by (B) SDS-PAGE or (E) BN-PAGE, and probed with anti-Ty1, anti-HA, or anti-TgTom40 antibodies. The masses of the predicted CTC (black arrowhead) and smaller TgCIA1-containing complexes (red and blue arrowheads) are depicted in the BN-PAGE data. (C) Immunofluorescence assays of rTgCIA1-mAID-HA/cTgCIA1WT-Ty1 (WT-Ty1), rTgCIA1-mAID-HA/cTgCIA1ScCD1-Ty1 (ScCD1-Ty1), rTgCIA1-mAID-HA/cTgCIA1ScCD3-Ty1 (ScCD3-Ty1), and rTgCIA1-mAID-HA/cTgCIA1ScCD5-Ty1 (ScCD5-Ty1) parasites, probed with anti-Ty1 (green) and anti-TgTom40 (magenta) antibodies. Scale bars are 2 µm. DIC, differential inference contrast. Right, corresponding fluorescence profiles depicting the intensity of anti-Ty1 (green) and anti-TgTom40 (magenta) labeling along the yellow line on merged images. (D) The correlation between Ty1-tagged proteins and TgTom40 was quantified using the Pearson correlation coefficient (r). The data were analyzed using a one-way ANOVA followed by Tukey’s multiple comparisons test with p values shown. (F) Fluorescence proliferation assays of rTgCIA1-mAID-HA, rTgCIA1-mAID-HA/cTgCIA1WT-Ty1, rTgCIA1-mAID-HA/cTgCIA1ScCD1-Ty1, rTgCIA1-mAID-HA/cTgCIA1ScCD3-Ty1, and rTgCIA1-mAID-HA/cTgCIA1ScCD5-Ty1 parasites cultured in the absence (black) or presence (red) of IAA. Parasite proliferation is expressed as a percentage of the final fluorescence measurement in the -IAA condition for each line. Individual data points and error bars represent the mean ± SD of three technical replicates. Error bars not visible are smaller than the symbol. Data are representative of three independent experiments. The numerical data underlying this Figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003520.g006 Notably, the CD loops are located on the opposite side of the CIA1 protein from that which interacts with CIA2 [10]. This places the CD loops in position to be involved in interactions external to the CTC. We therefore tested whether the extended CD1, CD3, or CD5 loops of TgCIA1 could have a role in mitochondrial targeting. We generated TgCIA1 transgenes in which the endogenous CD1, CD3, or CD5 loops of TgCIA1 were substituted for the equivalent, but substantially shorter, loops from the cytosolic CIA1 protein of yeast. We fused the resulting transgenes to a Ty1 epitope tag to enable their detection, and overexpressed them from the constitutive α-tubulin promoter in the rTgCIA1-mAID-HA line. We termed the resulting proteins cTgCIA1ScCD1-Ty1, cTgCIA1ScCD3-Ty1, and cTgCIA1ScCD5-Ty1, respectively, and also included a cTgCIA1WT-Ty1 control in our analyses. SDS-PAGE western blot analysis confirmed that each protein was expressed in T. gondii and were of the expected masses (Fig 6B). We observed some differences in expression levels between the constitutively-expressed proteins, with the cTgCIA1WT-Ty1 and cTgCIA1ScCD1-Ty1 proteins more abundant than the cTgCIA1ScCD3-Ty1 and cTgCIA1ScCD5-Ty1 proteins (S10A Fig). To test the localization of the modified proteins, we performed immunofluorescence assays. As expected, the cTgCIA1WT-Ty1 protein exhibited dual localization to the cytosol and mitochondrion (Fig 6C). Both cTgCIA1ScCD1-Ty1 and cTgCIA1ScCD3-Ty1 proteins also localized dually to the mitochondrion and cytosol (Fig 6C). This infers that the native CD1 and CD3 loops of TgCIA1 are not important for mitochondrial targeting, although we observed that rTgCIA1-mAID-HA/cTgCIA1ScCD1-Ty1 parasites exhibited aberrant mitochondrial morphology, with ~75% of parasites containing “tadpole”-like mitochondria instead of the typical “lasso” and ‘branched’ structures that mitochondria adopt in intracellular parasites (S10B Fig; [29]). In contrast to the other CD loop mutants, the mitochondrial localization of the cTgCIA1ScCD5-Ty1 protein was significantly reduced, with the protein localized predominantly to the cytosol (Fig 6C and 6D). This indicates that the extended CD5 loop of TgCIA1 is necessary for mitochondrial targeting. We next explored the importance of the extended CD loops for TgCIA1 function. We first performed BN-PAGE western blotting. Each of the CD loop mutants associated with a complex of ~720 kDa, which likely represents the CTC (Fig 6E; black arrowhead). This implies that the CD1, CD3, and CD5 loops are not required for TgCIA1 to assemble into the CTC. The cTgCIA1WT-Ty1, cTgCIA1ScCD1-Ty1, and cTgCIA1ScCD3-Ty1 proteins were all found in the smaller TgCIA1 complexes (Fig 6E; red and blue arrowheads), similar in mass to those we had observed previously with the natively tagged TgCIA1-HA protein (Fig 4A). The smaller TgCIA1 complexes in the cTgCIA1WT-Ty1 and cTgCIA1ScCD1-Ty1 lines were, in proportion to the ~720 kDa complex, more abundant than for the natively tagged TgCIA1-HA protein (compare Figs 4A to 6E), possibly an artifact of protein overexpression from the non-native α-tubulin promoter. Curiously, the cTgCIA1ScCD5-Ty1 protein did not appear to be present in the smaller mass complexes (Fig 6E), suggesting a role for the CD5 loop in assembly of these complexes. Next, we tested whether the CD1, CD3, or CD5 loops of TgCIA1 are important for parasite proliferation. We conducted fluorescence proliferation and plaque assays on rTgCIA1-mAID-HA parasites constitutively expressing WT TgCIA1 or the CD loop mutants in the absence or presence of IAA. As expected, the severe proliferation defect observed upon TgCIA1-mAID-HA depletion was rescued by constitutive expression of cTgCIA1WT-Ty1 (Figs 6F and S10C). cTgCIA1ScCD1-Ty1-expressing parasites exhibited a moderate proliferation defect upon TgCIA1-mAID-HA depletion, resulting in fewer (rather than smaller) plaques (Figs 6F and S10C). This suggests a possible role for the CD1 loop in processes that affect the viability of extracellular parasites or the ability of parasites to invade host cells. Parasites expressing the cTgCIA1ScCD3-Ty1 protein exhibited only minor defects in proliferation when the TgCIA1-mAID-HA protein was depleted (Figs 6F and S10C). Notably, cTgCIA1ScCD5-Ty1-expressing parasites exhibited a severe impairment of parasite proliferation when TgCIA1-mAID-HA was depleted, indistinguishable from the proliferation defect observed in non-complemented rTgCIA1-mAID-HA parasites (Figs 6F and S10C). This indicates that the CD5 loop is critical for TgCIA1 protein function. Having demonstrated that the CD5 loop of TgCIA1 is necessary for mitochondrial targeting, we wondered whether the CD5 loop alone could mediate mitochondrial protein targeting. We inserted the CD5 loop of TgCIA1 into a green fluorescent protein (GFP)-Ty1 reporter at an internal site in GFP shown previously to tolerate insertions (between the eighth and ninth β-strands of GFP; [30]). We expressed GFPTgCD5-Ty1 in T. gondii parasites and attempted to select parasites stably expressing the transgene. We found that, following selection, very few parasites expressed the GFPTgCD5-Ty1 protein. In those that did, the GFPTgCD5-Ty1 protein exhibited a dual localization to both the mitochondrion and cytosol (S11 Fig). However, we noticed that the mitochondrial morphology in these parasites appeared aberrant, suggesting a potential toxic effect of GFPTgCD5-Ty1 overexpression that complicates our interpretation of the data. We next generated a transgene encoding the structurally characterized Drosophila melanogaster CIA1 protein (DmCIA1; [10]) in which we replaced the native CD5 loop of DmCIA1 with the TgCIA1 CD5 loop. We expressed cDmCIA1TgCD5-Ty1 or a corresponding cDmCIA1WT-Ty1 protein in rTgCIA1-mAID-HA parasites and performed immunofluorescence assays to determine protein localization. As expected, cDmCIA1WT-Ty1 localized in the cytosol and did not overlap with the mitochondrion (Fig 7A). By contrast, the cDmCIA1TgCD5-Ty1 protein co-localized with the mitochondrial marker (Fig 7A). Taken together, these data indicate that the CD5 loop of TgCIA1 is sufficient to mediate mitochondrial localization. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 7. The CD5 loop of TgCIA1 is sufficient for mitochondrial targeting, with the positioning of the loop important for TgCIA1 function. (A) Immunofluorescence assays of parasites constitutively expressing cDmCIA1WT-Ty1 (top) or cDmCIA1TgCD5-Ty1, (bottom) probed with anti-Ty1 (green) and anti-TgTom40 (magenta; mitochondrion) antibodies. Scale bars are 2 µm. DIC, differential interference contrast. Right, corresponding fluorescence plots depicting the intensity of anti-Ty1 (green) and anti-TgTom40 (magenta) labeling along the yellow line in merged images. (B) Constitutive expression of a cTgCIA1CD5-to-CD1-Ty1 protein variant in rTgCIA1-mAID-HA parasites, shown in schematic (top), and probed in an immunofluorescence assay with anti-Ty1 (green) and anti-TgTom40 (magenta; mitochondrion) antibodies (bottom). Scale bars is 2 µm. DIC, differential interference contrast. Right, corresponding fluorescence plot depicting the intensity of anti-Ty1 (green) and anti-TgTom40 (magenta) labeling along the yellow line in merged image. Left, western blot of proteins extracted from rTgCIA1-mAID-HA/cTgCIA1CD5-to-CD1-Ty1 parasites, separated by SDS-PAGE and probed with anti-Ty1 antibodies. (C) Fluorescence proliferation assays of rTgCIA1-mAID-HA, rTgCIA1-mAID-HA/cTgCIA1WT-Ty1, and rTgCIA1-mAID-HA/cTgCIA1CD5-to-CD1-Ty1 parasites, cultured in the absence (black) or presence (red) of IAA. Parasite proliferation is expressed as a percentage of the fluorescence measurement in the -IAA condition on the final day of the assay for each line. Individual data points and error bars represent the mean ± SD of three technical replicates. Error bars not visible are smaller than the symbol. Data are representative of three independent experiments. The numerical data underlying this Figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003520.g007 We next asked whether the position of the CD5 loop in the TgCIA1 protein was important for mitochondrial localization or protein function. We engineered a TgCIA1 construct in which the amino acid sequence encoding the CD5 loop of TgCIA1 was located in the CD1 loop of the protein, with the native CD5 loop replaced by the equivalent CD5 loop of yeast (Fig 7B). We constitutively expressed the resulting protein, which we termed cTgCIA1CD5-to-CD1-Ty1, in rTgCIA1-mAID-HA parasites, validated expression by western blotting (Fig 7B), and performed an immunofluorescence assay to determine localization of the protein. Interestingly, we observed that the cTgCIA1CD5-to-CD1-Ty1 protein localized exclusively to the mitochondrion, no longer exhibiting the dual localization we observed in the wild type TgCIA1 protein (Fig 7B). We also found that constitutive expression of the cTgCIA1CD5-to-CD1-Ty1 protein was unable to rescue the proliferation defect observed upon rTgCIA1-mAID-HA knockdown (Figs 7C and S10D). Taken together, our data indicate that the mitochondrial targeting of the CTC is mediated by TgCIA1. Specifically, the CD5 loop of TgCIA1 is both necessary and sufficient for mitochondrial targeting, and this targeting is independent of the other mitochondrially-localized CIA pathway proteins TgNBP35, TgTah18, and TgNar1. Our data also indicate that, while the position of the CD5 loop in the protein is not critical for mitochondrial targeting, it is critical for facilitating the dual localization of TgCIA1 to the cytosol and mitochondrion. Finally, we have shown that the CD5 loop of TgCIA1, and its positioning within the protein, is critical for TgCIA1 to carry out its functions in parasites. A myzozoan-specific amino acid motif in the CD5 loop mediates the mitochondrial localization of TgCIA1 We next set out to uncover the features of the CD5 loop of TgCIA1 that facilitate mitochondrial targeting. Alignments of the CIA1 protein from a range of eukaryotes identified the presence of a short, conserved motif consisting of three aromatic and one positively charged amino acid in the CD5 loop of the CIA1 protein in T. gondii and other myzozoans (Figs 8A and S9). This motif (and the extended CD5 loop generally) was not present in other eukaryotic clades, including ciliates such as P. tetraurelia, which are the nearest relatives of the myzozoans [14]. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 8. An aromatic amino acid motif in the CD5 loop of TgCIA1 facilitates mitochondrial targeting. (A) Left, a region of a multiple sequence alignment of CIA1 homologs highlighting a motif of amino acid residues within the CD5 loop of the protein that is conserved in myzozoans. Asterisks denote key residues of the motif, with green denoting aromatic residues and orange denoting a positively charged residue. Right, a schematic of the motif observed in the dinozoans Perkinsus marinus and Symbiodinium microadriaticum, the apicomplexans Plasmodium falciparum, Cryptosporidium parvum, and Toxoplasma gondii, and the chrompodellids Vitrella brassicaformis and Chromera velia. Green = aromatic, orange = positively charged, gray = not conserved. Amino acid length of the CD5 loop in each species is listed on the right. (B, C) Western blots of proteins extracted from TgCIA1WT-HA (WT-HA), TgCIA1W526A-HA (W526A-HA), TgCIA1Y527A-HA (Y527A-HA), and TgCIA1R533A-HA (R533A-HA) expressing parasites, separated by (B) SDS-PAGE or (C) BN-PAGE, and probed with anti-HA or anti-TgTom40 antibodies. The black arrowhead indicates the >720 kDa CIA Targeting Complex; red and blue arrowheads indicate the lower mass complexes containing the TgCIA1 protein. (D) Immunofluorescence assays of TgCIA1WT-HA (WT-HA), TgCIA1W526A-HA (W526A-HA), TgCIA1Y527A-HA (Y527A-HA), and TgCIA1R533A-HA (R533A-HA) parasites. The proteins of interest (green) and the mitochondrion (magenta) were labeled with anti-HA and anti-TgTom40 antibodies, respectively. Schematics depicting the modified amino acid sequence in the CD5 motif of the proteins from each panel are included next to images (left). Scale bars are 2 µm. DIC, differential interference contrast. Right, corresponding fluorescence profile depicting intensity of anti-HA (green) and anti-TgTom40 (magenta) labeling along the yellow lines of the merged images. (E) The correlation between HA-tagged proteins of interest and TgTom40 was quantified using the Pearson correlation coefficient (r) and the data were analyzed using a one-way ANOVA followed by Tukey’s multiple comparisons test. (F) Plaque assays of TgCIA1WT-HA, TgCIA1W526A-HA, TgCIA1Y527A-HA, and TgCIA1R533A-HA parasites. Parasites were cultured in the absence (top) or presence (bottom) of IAA for 8 days and are representative of three independent experiments. (G) Immunofluorescence assays of rTgCIA1-mAID-HA/cTgCIA1WT-Ty1 and rTgCIA1-mAID-HA/cTgCIA1F532A-Ty1 parasites. The cTgCIA1WT-Ty1 protein (green) and the mitochondrion (magenta) were labeled with anti-Ty1 and anti-TgTom40 antibodies, respectively. Schematics depicting the amino acid sequence in the CD5 motif of the proteins from each panel are included next to images (left). Scale bars are 2 µm. DIC, differential interference contrast. Right, corresponding fluorescence profile depicting intensity of anti-Ty1 (green) and anti-TgTom40 (magenta) labeling along the yellow lines of the merged images. (H) The correlation between Ty1-tagged proteins and TgTom40 was quantified using the Pearson correlation coefficient (r). The data were analyzed using a one-way ANOVA (alongside the r values depicted in S12F Fig) followed by Tukey’s multiple comparisons test, with the p value shown. (I) Fluorescence proliferation assays of rTgCIA1-mAID-HA, rTgCIA1-mAID-HA/cTgCIA1WT-Ty1, and rTgCIA1-mAID-HA/cTgCIA1F532A-Ty1 parasites, grown in the absence (black) or presence (red) of IAA. Parasite proliferation is expressed as a percentage of the fluorescence measurement in the -IAA condition on the final day of the assay for each line. Individual data points and error bars represent the mean ± SD of three technical replicates. Error bars not visible are smaller than the symbol. Data are representative of three independent experiments. The numerical data underlying this Figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003520.g008 We hypothesized that the conserved motif of the CD5 loop could facilitate the mitochondrial targeting of TgCIA1. To test this, we used CRISPR/Cas9-based genome editing to individually substitute each residue in the motif for alanine in the native TgCIA1 locus of the TgCIA1-HA/rTgNBP35-cMyc parasite line. We were successful in generating mutants in the W526, Y527, and R533 residues (but not the F532 residue), terming the resulting proteins TgCIA1W526A-HA, TgCIA1Y527A-HA, and TgCIA1R533A-HA (S12A Fig). SDS-PAGE western blot analyses confirmed that the mutated proteins were expressed at similar abundances to a TgCIA1WT-HA control (Figs 8B and S12B). We next performed BN-PAGE western blotting and found that all mutated proteins were present in the >720 kDa CTC as well as in the smaller TgCIA1-containing complexes we had observed previously (Fig 8C). To test whether the motif is important for mitochondrial targeting, we performed immunofluorescence assays. Notably, all mutations in the CD5 loop motif resulted in the TgCIA1 protein localizing predominantly to the cytosol (Fig 8D), although quantifications revealed that the TgCIA1R533A-HA protein exhibited significantly greater mitochondrial co-localization than the TgCIA1W526A-HA and TgCIA1Y527A-HA proteins (Fig 8E). Next, we investigated whether the W526, Y527, or R533 residues of the conserved CD5 loop motif contribute to the role of TgCIA1 in parasite proliferation. We performed plaque assays comparing the proliferation of parasites expressing TgCIA1W526A-HA, TgCIA1Y527A-HA, and TgCIA1R533A-HA to parasites expressing TgCIA1WT-HA. We found that TgCIA1W526A-HA and TgCIA1Y527A-HA expressing parasites exhibited severe proliferation defects, indicating that these residues are critical for TgCIA1 function (Fig 8F). By contrast, the proliferation of parasites expressing TgCIA1R533A-HA was indistinguishable from those of parasites expressing TgCIA1WT-HA, suggesting the R533 residue is largely dispensable for TgCIA1 function (Fig 8F). We were unable to modify the TgCIA1 locus to express a F532A mutation using the genome editing approach. As an alternative, we constitutively expressed TgCIA1F532A-Ty1 from the α-tubulin promoter in rTgCIA1-mAID-HA parasites, generating a line we termed rTgCIA1-mAID-HA/cTgCIA1F532A-Ty1. We found that the cTgCIA1F532A-Ty1 protein was expressed at a similar abundance, and in protein complexes of similar masses, to the cTgCIA1WT-Ty1 protein (S12C and S12D Fig). Notably, we found that the cTgCIA1F532A-Ty1 protein localized predominantly to the cytosol, exhibiting significantly less mitochondrial co-localization than the cTgCIA1WT-Ty1 protein (Fig 8G and 8H). We also constitutively-expressed TgCIA1W526A-Ty1, TgCIA1Y527A-Ty1 and cTgCIA1R533A-Ty1 isoforms in rTgCIA1-mAID-HA parasites, and found that the localization of these matched what we observed in the genome-edited point mutants, with all TgCIA1 variants localizing predominantly to the cytosol, although the cTgCIA1R533A-Ty1 again exhibited greater mitochondrial co-localization than the other variants (S12E and S12F Fig). Finally, we undertook fluorescence proliferation assays and plaque assays to test whether the F532 residue of the CD5 loop is important for TgCIA1 function. We compared proliferation of rTgCIA1-mAID-HA/cTgCIA1F532A-Ty1 parasites to rTgCIA1-mAID-HA and rTgCIA1-mAID-HA/cTgCIA1WT-Ty1 parasite lines cultured in the absence or presence of IAA. We observed that proliferation of rTgCIA1-mAID-HA/cTgCIA1F532A parasites was severely impaired when TgCIA1-mAID-HA was knocked down (Figs 8I and S10E), indicating that the F532 residue is essential for TgCIA1 protein function. We also tested the proliferation of rTgCIA1-mAID-HA/TgCIA1W526A-Ty1, rTgCIA1-mAID-HA/TgCIA1Y527A-Ty1, and rTgCIA1-mAID-HA/cTgCIA1R533A-Ty1 parasites in the absence or presence of IAA. This revealed that rTgCIA1-mAID-HA/cTgCIA1Y527A-Ty1 and rTgCIA1-mAID-HA/cTgCIA1R533A-Ty1 parasites proliferated normally upon knockdown of the TgCIA1-mAID-HA protein, whereas proliferation of rTgCIA1-mAID-HA/cTgCIA1W526A-Ty1 parasites was substantially reduced upon TgCIA1-mAID-HA depletion, although not to the same extent as observed in rTgCIA1-mAID-HA/cTgCIA1F532A-Ty1 parasites (S12G and S10E Figs). These findings suggest that constitutive overexpression of the TgCIA1W526A and TgCIA1Y527A mutant isoforms can partially or fully rescue the severe proliferation defects observed in the genome-edited point mutants. Taken together, these data indicate that the W526, Y527, and F532 residues of the aromatic amino acid motif in the CD5 loop of TgCIA1 are critical for both mitochondrial localization of the TgCIA1 protein and for the functional role of TgCIA1 in parasite proliferation. The R533 residue contributes to the mitochondrial localization of TgCIA1, although not to the same extent as the other residues of this motif that we tested. Surprisingly, despite its role in mitochondrial targeting, the R533 residue of TgCIA1 appears to be dispensable for TgCIA1 function. Given the conservation of the CD5 loop motif in myzozoans, we asked whether CD5 loops from other myzozoans could complement the function of T. gondii CD5 loop. We replaced the CD5 loop of TgCIA1 with the equivalent CD5 loop from the chrompodellid V. brassicaformis or the dinozoan S. microadriaticum, generating proteins we called cTgCIA1VbCD5-Ty1 or cTgCIA1SmCD5-Ty1. The V. brassicaformis CD5 loop contains the same residues in the conserved aromatic motif as the TgCIA1 protein (Fig 8A), but is considerably shorter than the T. gondii CD5 loop (87 amino acids versus 249 amino acids). The S. microadriaticum CD5 loop only encodes two amino acids of the motif (the phenylalanine and arginine residues; Fig 8A), and is also considerably shorter than the T. gondii CD5 loop (97 amino acids). We expressed the cTgCIA1VbCD5-Ty1 and cTgCIA1SmCD5-Ty1 proteins in rTgCIA1-mAID-HA parasites, validated expression by SDS-PAGE western blotting (Fig 9A and 9B), and performed immunofluorescence assays to determine protein localization. Both the cTgCIA1VbCD5-Ty1 and cTgCIA1SmCD5-Ty1 proteins localized predominantly to the cytosol (Fig 9A–9C), although the cTgCIA1SmCD5-Ty1 protein also exhibited some observable co-localization with the mitochondrion (Fig 9B, arrowheads). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 9. The CD5 loop of the CIA1 protein is functionally conserved in myzozoans. (A, B) Immunofluorescence assay of (A) rTgCIA1-mAID-HA/cTgCIA1VbCD5-Ty1 and (B) rTgCIA1-mAID-HA/cTgCIA1SmCD5-Ty1 parasites, probed with anti-Ty1 (green) and anti-TgTom40 (magenta; mitochondrion) antibodies. Schematics depicting the amino acid sequence in the CD5 motif of the proteins from each panel are included below the Ty1 labeled images. Scale bars are 2 µm. DIC, differential interference contrast. Arrowheads highlight regions where the cTgCIA1SmCD5-Ty1 protein exhibits visible mitochondrial localization. Right, corresponding fluorescence profile depicting intensity of anti-Ty1 (green) and anti-TgTom40 (magenta) labeling along the yellow line on merged image. Left, western blots of the proteins of interest, separated by SDS-PAGE and probed with anti-Ty1 antibodies. (C) The correlation between the Ty1-tagged proteins of interest and TgTom40 was quantified using the Pearson correlation coefficient (r). The data were analyzed using a one-way ANOVA followed by Tukey’s multiple comparisons test with p values shown. (D) Fluorescence proliferation assays of rTgCIA1-mAID-HA, rTgCIA1-mAID-HA/cTgCIA1WT-Ty1, rTgCIA1-mAID-HA/cTgCIA1VbCD5-Ty1, and rTgCIA1-mAID-HA/cTgCIA1SmCD5-Ty1 parasites, grown in the absence (black) or presence (red) of IAA. Parasite proliferation is expressed as a percentage of the fluorescence measurement in the -IAA condition on the final day of the assay for each line. Individual data points and error bars represent the mean ± SD of three technical replicates. Error bars not visible are smaller than the symbol. Results are representative of three independent experiments. The data for the rTgCIA1mAID-HA and rTgCIA1-mAID-HA/cTgCIA1WT-Ty1 lines are identical to those depicted in Fig 7C, the experiments for which were performed simultaneously. The numerical data underlying this Figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003520.g009 Next, we investigated whether the CD5 loops of the SmCIA1 and VbCIA1 homologs were functionally equivalent to CD5 loop of TgCIA1. We cultured rTgCIA1-mAID-HA/cTgCIA1VbCD5-Ty1 and rTgCIA1-mAID-HA/cTgCIA1SmCD5-Ty1 parasites in the absence or presence of IAA and performed fluorescence proliferation and plaque assays. These revealed that both the cTgCIA1SmCD5-Ty1- and cTgCIA1VbCD5-Ty1-complemented lines proliferated normally when cultured in the presence of IAA (Figs 9D and S10D), indicating that the CD5 loops from the S. microadriaticum and V. brassicaformis CIA1 proteins can functionally replace the equivalent loop in the TgCIA1 protein of T. gondii. Taken together, our data demonstrate that the CD5 loop of TgCIA1 contains a motif that is conserved throughout the myzozoans (Fig 8A), and which contributes to targeting the CIA1 protein to the mitochondrion. Numerous residues of this motif, including a phenylalanine residue from this motif that is found in all the analyzed myzozoan CIA1 sequences except Cryptosporidium parvum (Fig 8A), are critical for TgCIA1 to carry out its biological role. The CD5 loop of TgCIA1 is not required for cytosolic protein synthesis We have shown that the CTC of T. gondii exhibits dual localization to the mitochondrion and cytosol, courtesy of a myzozoan-specific CD5 loop in the CIA1 protein of the complex (Fig 10A). It is conceivable that the mitochondrial localization of TgCIA1 is important for the transfer of [4Fe-4S] clusters from mitochondrially-localized Nar1 to the CTC. TgCIA1 localizes to the mitochondrion independently of its association with TgNar1 (S8H Fig), but it is possible that the CD5 loop of TgCIA1 interacts with mitochondrial outer membrane lipids or an accessory protein of the mitochondrial outer membrane (Fig 10Ai–10Aii). This could place the CTC in position on the outer membrane to interact with TgNar1 and enable [4Fe-4S] cluster transfer to occur. In these scenarios, the CD5 loop of TgCIA1 functions as a mitochondrial targeting signal to facilitate FeS cluster transfer from TgNar1. Notably, T. gondii expresses the FeS proteins TgELP3 and TgRlmN on the outer face of outer mitochondrial membrane [31]. An alternative possibility is, therefore, that the CD5 loop instead functions in enabling TgCIA1 and the CTC complex to interact with these client FeS proteins on the outer membrane (Fig 10Aiii). In this scenario, the CD5 loop functions not in enabling [4Fe-4S] cluster transfer from TgNar1, but instead to enable [4Fe-4S] cluster transfer from the CTC to client mitochondrial outer membrane proteins. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 10. A model for the spatial organization and evolution of the CIA pathway in Toxoplasma gondii and related eukaryotes. (A) Models for the spatial organization of the CIA pathway in T. gondii and related organisms. A sulfur-containing product of the mitochondrial ISC pathway (possibly a [2Fe-2S] cluster) is exported from the mitochondrion. [4Fe-4S] clusters assemble on the NBP35 scaffold (N35) at the mitochondrial outer membrane, with electrons for this process donated from NADPH via mitochondrial Tah18 (T18) and possibly Dre2 (D2). [4Fe-4S] clusters are transferred from the NBP35 scaffold to Nar1 (N1), and further to the CIA targeting complex (CTC) comprised of CIA1 (C1), CIA2 (C2) and MMS19 (M19). CIA1 contains an extended CD5 loop (green) that mediates localization of the entire CTC to the mitochondrion. The CD5 loop may bind to (i) outer mitochondrial membrane lipids or (ii) outer mitochondrial membrane accessory proteins, and place the CTC in position to receive FeS clusters from Nar1. Alternatively, (iii) the CD5 loop may mediate interactions with FeS client proteins that are anchored on the mitochondrial outer membrane, thus facilitating FeS transfer from the CTC to these proteins. Elements of the diagram were created in BioRender. Hodgson, E. (2025) https://BioRender.com/ecxz792. (B) Western blots measuring the incorporation of puromycin into proteins from rTgCIA1-mAID-HA/cTgCIA1ScCD5-Ty1 (ScCD5) or rTgCIA1-mAID-HA/cTgCIA1CD5-to-CD1-Ty1 (CD5-to-CD1) parasites cultured in the absence (−) or presence (+) of IAA for 24 h, separated by SDS-PAGE and probed with anti-puromycin and anti-TgTom40 antibodies. (C) Relative abundance of puromycin incorporation from (B) was determined as a percentage of the -IAA control for each parasite line, with abundances normalized using the TgTom40 loading control. Data points represent the mean ± SD of three independent experiments. Data were analyzed using a one-way ANOVA followed by Tukey’s multiple comparisons test, with relevant p values shown. The numerical data underlying this Figure can be found in S1 Data. (D) An illustrative phylogenetic tree adapted from currently accepted models of eukaryotic evolution [63–65]. Mitochondrial targeting domains for key components of the CIA machinery, including the extended CD5 loop of CIA1, the N-terminal (N-term) transmembrane domain (TMD) of NBP35 and the C-terminal (C-term) TMDs of the FeS proteins Elp3 and RlmN, all evolved early in myzozoan evolution, subsequent to their divergence from the ciliate lineage. The endosymbiotic acquisition of a chloroplast also occurred early in myzozoan evolution. https://doi.org/10.1371/journal.pbio.3003520.g010 If the mitochondrial localization of TgCIA1 is critical for FeS cluster transfer from TgNar1, we predicted that impairing the mitochondrial localization of TgCIA1 will impair all downstream processes that require the CTC, such as cytosolic protein translation. We therefore measured protein translation in a parasite strain expressing a TgCIA1 variant that is unable to target to the mitochondrion (cTgCIA1ScCD5-Ty1, which lacks the mitochondrial targeting CD5 loop and localizes exclusively to the cytosol; Fig 6C). We compared this to protein translation in a parasite strain expressing a TgCIA1 variant that is targeted exclusively to the mitochondrion (cTgCIA1CD5-to-CD1-Ty1; Fig 7B). Remarkably, we still observed robust cytosolic protein translation in parasites expressing only the cytosolically localized cTgCIA1ScCD5-Ty1 protein (Fig 10B and 10C). By contrast, we observed a significant depletion of cytosolic protein translation when parasites expressed only the mitochondrially-localized cTgCIA1CD5-to-CD1-Ty1 protein (Fig 10B and 10C), similar to the defect we observed when TgCIA1 is depleted (Fig 3C and 3D). These data indicate that CD5 loop-dependent mitochondrial localization of TgCIA1 is not required for the FeS cluster-dependent process of protein translation in the cytosol. These data are, therefore, inconsistent with the hypothesis that the CD5 loop of TgCIA1 is important for FeS cluster transfer from Nar1 to the CTC at the mitochondrial outer membrane (Fig 10Ai–10Aii). Instead, our data indicate that some functions facilitated by TgCIA1 (such as cytosolic protein translation) are independent of the CD5 loop. Our data also suggest that the cytosolic localization of TgCIA1 is required to enable cytosolic protein translation, since we observed a strong translation defect in parasites expressing only the mitochondrially-localized cTgCIA1CD5-to-CD1-Ty1 protein, although we cannot rule out this defect is due to the aberrant positioning of the CD5 loop in this protein interfering with CTC function. Cytosolic FeS cluster assembly occurs at the mitochondrion of Toxoplasma gondii We set out to determine the subcellular localization of the previously uncharacterized candidate CIA pathway proteins in T. gondii. Using CRISPR/Cas9 genome editing, we introduced a 3× hemagglutinin (HA) epitope tag into the 3′ end of the coding sequences of the proposed electron transfer chain proteins TgTah18 (www.ToxoDB.org gene identifier TGGT1_249320; [17]) and TgDre2 (TGGT1_216900), and all three of the proposed CTC proteins, TgCIA1 (TGGT1_313280), TgCIA2 (TGGT1_306590) and TgMMS19 (TGGT1_222230), in an anhydrotetracycline (ATc)-regulatable TgNBP35-cMyc strain that we had generated previously [15]. We validated correct integration of the tag in each parasite line by PCR analysis (S1A–S1E Fig). We were unsuccessful in integrating a HA tag into the 3′ end of the coding sequence for the proposed FeS transfer protein TgNar1 (TGGT1_242580). Instead, we introduced a HA-mini-Auxin Inducible Degron (HA-mAID) tag at the 5′ end of the coding region of TgNar1 in a tdTomato and Tir1-expressing parasite strain (RH∆ku80/Tir1-FLAG/tdTomato; [18]; S1F Fig). To test whether each protein was expressed in the disease-causing tachyzoite stage of the parasite life cycle, we extracted proteins from the TgTah18-HA, TgDre2-HA, HA-mAID-TgNar1, TgCIA1-HA, TgCIA2-HA, and TgMMS19-HA lines, separated them by Sodium Dodecyl Sulfate-Polyacrylamide Gel Electrophoresis (SDS-PAGE), and performed western blotting with anti-HA antibodies. All six proteins were expressed and were of approximately the expected masses for the epitope-tagged proteins (Fig 1A–1F). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. The CIA pathway localizes to the mitochondrion of Toxoplasma gondii. (A–F) Western blots of proteins extracted from (A) TgTah18-HA, (B) TgDre2-HA, (C) HA-mAID-TgNar1, (D) TgCIA1-HA, (E) TgCIA2-HA, or (F) TgMMS19-HA expressing parasites, separated by SDS-PAGE and probed with anti-HA antibodies. TgMMS19-HA appears as two separate bands, the lower of which may represent a degradation product. (G–L) Immunofluorescence assays of (G) TgTah18-HA, (H) TgDre2-HA, (I) HA-mAID-TgNar1, (J) TgCIA1-HA, (K) TgCIA2-HA, or (L) TgMMS19-HA expressing parasites, probed with anti-HA antibodies (green) to label the proteins-of-interest and anti-TgTom40 antibodies (magenta) to label the mitochondrion. Scale bars are 2 µm. DIC, differential interference contrast transmission image. Right, corresponding fluorescence profiles depicting the intensity of anti-HA (green) and anti-TgTom40 (magenta) labeling along the yellow line on merged images. The numerical data underlying this Figure can be found in S1 Data. (M) Model for the CIA pathway in T. gondii parasites, with Tah18, NBP35 and Nar1 positioned on the outer face of the mitochondrion, Dre2 in the cytosol, and the CIA Targeting Complex (CTC; consisting of CIA1, CIA2 and MMS19), exhibiting dual cytosolic and mitochondrial localization. Electrons (e−) are sourced from NADPH via an electron transfer chain consisting of Tah18 and Dre2. Sulfur (yellow) and possibly iron (red) are sourced from the iron–sulfur cluster (ISC) synthesis pathway in the mitochondrion, assembled as [4Fe-4S] clusters on the NBP35 scaffold, then trafficked via Nar1 to the CTC. The CTC donates FeS clusters to client FeS proteins. https://doi.org/10.1371/journal.pbio.3003520.g001 To determine the subcellular localization of each protein, we performed immunofluorescence assays. We found that TgTah18-HA and HA-mAID-TgNar1 both co-localized with the mitochondrial marker TgTom40, whereas TgDre2-HA localized throughout the parasite cytosol (Fig 1G–1I). Curiously, all three proteins of the putative CTC exhibited a dual localization, overlapping with TgTom40 in the mitochondrion and also exhibiting diffuse labeling through the cytosol (Fig 1J–1L). Taken together, these data indicate that all the putative CIA pathway proteins in T. gondii, with the exception of TgDre2 but including the previously characterized scaffold protein TgNBP35 [15,16], localize to the mitochondrion. We showed previously that TgNBP35 localizes to the cytosolic face of the outer mitochondrial membrane [15], and our data are therefore consistent with the CIA pathway occurring on the cytosolic face of the outer mitochondrial membrane in T. gondii (Fig 1M). The CIA pathway is critical for parasite proliferation and protein synthesis in Toxoplasma gondii We set out to test the importance of each of the candidate CIA pathway proteins for parasite proliferation. Using CRISPR/Cas9 genome editing, we introduced mAID-HA epitope tags at the 3′ ends of the coding regions for TgTah18, TgDre2, TgCIA1, TgCIA2 and TgMMS19 of tdTomato/Tir1 parasites (S2 Fig). The mAID tagging system facilitates proteasomal degradation of mAID-fused proteins upon the addition of the small molecule 3-indoleacetic acid (IAA; [19]). Since the abundance of the resulting proteins can be conditionally regulated, we termed the resulting lines expressing mAID-tagged CIA pathway proteins regulatable (r)-TgTah18-mAID-HA, rTgDre2-mAID-HA, rHA-mAID-TgNar1, rTgCIA1-mAID-HA, rTgCIA2-mAID-HA, and rTgMMS19-mAID-HA. To determine whether the mAID-tagged proteins could be regulated upon the addition of IAA, we cultured each parasite line in the absence or presence of IAA for 24 h. We extracted the resulting proteins, separated them by SDS-PAGE and performed western blotting. We found that the abundances of all six candidate CIA pathway proteins were substantially depleted upon IAA addition (Fig 2A–2F). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Candidate CIA pathway proteins are important for parasite proliferation. (A–G) Fluorescence proliferation assays and western blotting of (A) rTgTah18-mAID-HA, (B) rTgDre2-mAID-HA, (C) rHA-mAID-TgNar1, (D) rTgCIA1-mAID-HA, (E) rTgCIA2-mAID-HA, (F) rTgMMS19-mAID-HA, and (G) WT (RH∆ku80/Tir1-FLAG/tdTomato) parasites cultured in the absence (black) or presence (red) of IAA. Parasite proliferation is expressed as a percentage of the final fluorescence measurement in the -IAA condition for each line. Individual data points and error bars represent the mean ± SD of three technical replicates. Data are representative of three independent experiments. Error bars not visible are smaller than the symbol. The numerical data underlying this Figure can be found in S1 Data. For the western blotting, parasites were cultured in the absence (−) or presence (+) of IAA for 24 h, with protein samples separated by SDS-PAGE then subjected to western blotting with anti-HA antibodies to detect the mAID-HA-tagged protein of interest and anti-TgTom40 antibodies as a loading control (A–F, left). https://doi.org/10.1371/journal.pbio.3003520.g002 To explore the contribution of each CIA pathway protein to parasite proliferation, we performed fluorescence-based proliferation assays, measuring tdTomato fluorescence in parasites across time as a function of parasite proliferation. We found that in the absence of IAA, all parasite lines proliferated normally, exhibiting a typical sigmoidal growth curve (Fig 2). The parental RH∆ku80/Tir1-FLAG/tdTomato (WT) line proliferated normally in the presence of IAA, indicating that IAA is not toxic to parasites (Fig 2G). By contrast, the rTgTah18-mAID-HA, rTgDre2-mAID-HA, rHA-mAID-TgNar1, rTgCIA1-mAID-HA, rTgCIA2-mAID-HA, and rTgMMS19-mAID-HA lines all exhibited a severe proliferation defect when cultured in the presence of IAA (Fig 2A–2F), with the phenotype upon the depletion of TgNar1 consistent with that reported in another recent study [20]. These data indicate that TgTah18, TgDre2, TgNar1, TgCIA1, TgCIA2 and TgMMS19 are all critical for parasite proliferation. Next, we wanted to explore whether the candidate proteins function in the CIA pathway. Given its curious dual localization to the mitochondrion and cytosol, we initially focused on the CTC protein TgCIA1. Previous studies have found that impairment of the cytosolic FeS assembly pathway in T. gondii leads to a depletion in the abundances of some cytosolic and nuclear FeS cluster-containing proteins, including the ribosome recycling protein TgABCE1 [15,20–22]. We integrated a 3× cMyc tag at the 3′ end of the TgABCE1 open reading frame in the rTgCIA1-mAID-HA line using CRISPR/Cas9 genome editing (S3A and S3B Fig). We cultured these parasites for 0, 24, or 48 hours on IAA, separated proteins by SDS-PAGE, and performed western blotting to detect the TgABCE1-cMyc protein. We found a small but significant decrease in TgABCE1-cMyc abundance after 24 h of IAA incubation and a strong ~75% decrease after 48 h (Fig 3A and 3B). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Characterizing the effects of depleting candidate CIA pathway proteins in Toxoplasma gondii parasites. (A) Western blots of proteins extracted from rTgCIA1-mAID-HA/TgABCE1-cMyc parasites cultured for 0, 24 or 48 h in IAA and separated by SDS-PAGE. Samples were probed with anti-HA, anti-cMyc, and anti-TgTom40 antibodies. (B) Relative abundance of the TgABCE1-cMyc protein in the western blot was determined as a percentage of the 0 h control, with abundances normalized using the TgTom40 loading control. Data points represent the mean ± SD of five independent experiments for the 0 and 24 h conditions and three independent experiments for the 48 h condition. Data were analyzed using a one-way ANOVA followed by Tukey’s multiple comparisons test, with p values shown. (C) Western blots measuring the incorporation of puromycin into proteins from WT (RH∆ku80-Tir1-FLAG/tdTomato) or rTgCIA1-mAID-HA parasites cultured in the absence (−) or presence (+) of IAA for 24 h, separated by SDS-PAGE and probed with anti-puromycin, anti-HA, and anti-TgTom40 antibodies. (D) Relative abundance of puromycin incorporation was determined as a percentage of the -IAA control for each parasite line, with abundances normalized using the TgTom40 loading control. Data points represent the mean ± SD of three independent experiments. Data were analyzed using a one-way ANOVA followed by Tukey’s multiple comparisons test, with relevant p values shown. (E) Multiple sequence alignment of a region of the T. gondii CIA2 protein (TgCIA2) with CIA2 homologs in Saccharomyces cerevisiae (ScCIA2), Homo sapiens (HsCIA2), and Drosophila melanogaster (DmCIA2). The cysteine residue proposed to function in FeS cluster binding is highlighted by a red box. (F) Western blot of proteins extracted from rTgCIA2-mAID-HA/cTgCIA2WT-Ty1 or rTgCIA2-mAID-HA/cTgCIA2C524A-Ty1 parasites, separated by SDS-PAGE and probed with anti-Ty1 antibodies to detect the cTgCIA2WT-Ty1 and cTgCIA2C524A-Ty1 proteins, and anti-TgTom40 antibodies as a loading control. (G) Fluorescence proliferation assays of rTgCIA2-mAID-HA, rTgCIA2-mAID-HA/cTgCIA2WT-Ty1, or rTgCIA2-mAID-HA/cTgCIA2C524A-Ty1 parasites cultured in the absence (black) or presence (red) of IAA. Parasite proliferation is expressed as a percentage of the -IAA condition on the final day of the experiment for each parasite line, with values depicting the mean ± SD of three technical replicates. Error bars not visible are smaller than the symbol. Data are representative of three independent experiments. The numerical data underlying this Figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003520.g003 ABCE1 and other cytosolic FeS proteins are important for protein translation from cytosolic ribosomes. Previous studies have found that depletion of proteins that contribute to the CIA pathway result in a global impairment of protein synthesis in T. gondii [20,21]. We cultured WT and rTgCIA1-mAID-HA parasites in the absence or presence of IAA for 24 h, then treated parasites with the protein synthesis inhibitor puromycin for 15 min. Puromycin becomes incorporated into nascent polypeptides during translation and can be detected by anti-puromycin antibodies, thus providing a measure for newly synthesized proteins. We separated proteins from puromycin-treated parasites using SDS-PAGE and performed western blotting. We observed a significant ~80% depletion in puromycin-labeled proteins in IAA-treated rTgCIA1-mAID-HA parasites, whereas puromycin labeling was unchanged in IAA-treated WT parasites (Fig 3C and 3D). We next performed puromycin incorporation assays in rTgTah18-mAID-HA, rTgDre2-mAID-HA, rHA-mAID-TgNar1, rTgCIA2-mAID-HA and rTgMMS19-mAID-HA parasites cultured for 24 h in the absence or presence of IAA. We observed significantly less puromycin incorporation upon the depletion of TgTah18-mAID-HA, HA-mAID-TgNar1, TgCIA2-mAID-HA and TgMMS19-mAID-HA but not upon the depletion of TgDre2-mAID-HA (S4A and S4B Fig). To test whether TgDre2 knockdown leads to the depletion of TgABCE1, we integrated a cMyc epitope tag into the TgABCE1 locus of rTgDre2-mAID-HA parasites (S3C Fig). We cultured these parasites for 0, 24, or 48 hours on IAA, separated proteins by SDS-PAGE, and performed western blotting to detect the TgABCE1-cMyc protein. We found a ~40% decrease in TgABCE1-cMyc abundance after 24 h of IAA incubation and a ~60% decrease after 48 h (S4C Fig). Loss of TgDre2, therefore, does impact the abundance of the cytosolic Fe–S protein TgABCE1, leaving open the possibility that TgDre2 has a role in cytosolic Fe–S synthesis. Taken together, our data indicate that the putative CIA pathway proteins of T. gondii are essential for parasite proliferation, and that most of the candidate CIA pathway proteins in T. gondii (with the possible exception of TgDre2) are critical for protein translation, a process that relies on cytosolic FeS proteins like ABCE1. The CIA2 protein in animals and fungi contains a reactive cysteine residue that is thought to function in FeS cluster binding and facilitating the transfer of FeS clusters to client proteins [10,23,24]. This cysteine residue has been shown to be critical for CIA pathway function and cell survival in these eukaryotes [23,24]. Alignments of yeast and animal CIA2 proteins with TgCIA2 revealed that the reactive cysteine residue is conserved as residue 524 in the TgCIA2 protein (Figs 3E and S5). We hypothesized that if TgCIA2 functions like its counterpart in other eukaryotes, mutating the TgCIA2C524 residue would ablate CIA pathway function and, subsequently, lead to impaired parasite proliferation. We constitutively expressed either Ty1 epitope-tagged WT TgCIA2 (constitutive (c)TgCIA2WT-Ty1) or a Ty1-tagged TgCIA2 mutant wherein the cysteine at residue 524 was mutated to Ala (cTgCIA2C524A-Ty1) in rTgCIA2-mAID-HA parasites. Western blotting revealed that both Ty1-tagged TgCIA2 proteins were expressed at similar levels (Fig 3F) and fluorescence proliferation assays revealed that expression of cTgCIA2WT-Ty1 fully rescued parasite proliferation upon knockdown of the TgCIA2-mAID-HA protein (Fig 3G). By contrast, we observed minimal proliferation of rTgCIA2-mAID-HA/cTgCIA2C524A-Ty1 parasites cultured in the presence of IAA (Fig 3G). These data indicate that the proposed FeS cluster binding Cys-524 residue of TgCIA2 is critical for its function, consistent with the hypothesis that the TgCIA2 protein of T. gondii functions in a similar manner to the CIA2 protein of other eukaryotes in coordinating FeS cluster binding on the CTC [10,23,24]. TgCIA1, TgCIA2, and TgMMS19 are components of the same protein complex We set out to explore the candidate CTC proteins of T. gondii in more detail. First, we asked whether the TgCIA1, TgCIA2 and TgMMS19 proteins exist in a protein complex. We extracted proteins from TgCIA1-HA, TgCIA2-HA and TgMMS19-HA expressing parasites, separated them by blue native (BN)-PAGE, and performed western blotting with anti-HA antibodies. All three proteins were present in a complex of slightly larger than 720 kDa (Fig 4A, black arrowhead). The TgCIA1-HA protein, but not the others, was also observed in smaller protein complexes of ~400 and ~240 kDa (Fig 4A, red and blue arrowheads). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. Characterizing the protein composition of the CIA Targeting Complex (CTC) in Toxoplasma gondii parasites. (A) Western blot of proteins extracted from TgCIA1-HA, TgCIA2-HA and TgMMS19-HA expressing parasites, separated by BN-PAGE and probed with anti-HA antibodies or anti-TgTom40 as a loading control. (B, D) Western blots of proteins extracted from rTgCIA1-mAID-cMyc/TgCIA2-HA and rTgCIA1-mAID-cMyc/TgMMS19-HA parasites cultured in the absence (-) or presence (+) of IAA for 24 h, separated by either SDS-PAGE (B) or BN-PAGE (D) and probed with anti-HA, anti-cMyc or anti-TgTom40 antibodies. Data are representative of three independent experiments. (C) Quantification of the western blots depicted in (B), depicting relative abundance of the TgCIA2-HA or TgMMS19-HA proteins determined as a percentage of the -IAA control, with abundances normalized using the TgTom40 loading control. Data points represent the mean ± SD of three independent experiments. Data were analyzed using a paired t test with p values shown. The numerical data underlying this Figure can be found in S1 Data. (E, F) Western blot of proteins extracted from rTgCIA2-mAID-HA/TgCIA1-smFP-cMyc and rTgMMS19-mAID-HA/TgCIA1-smFP-cMyc parasites, cultured for 24 h in the absence (−) or presence (+) of IAA, separated by either (E) SDS-PAGE or (F) BN-PAGE and probed with anti-cMyc, anti-HA, or anti-TgTom40 antibodies. Data are representative of three independent experiments. In the BN-PAGE western blots (A, D, and F), the black arrowheads indicate the >720 kDa candidate CIA Targeting Complex; red and blue arrowheads indicate the lower mass complexes containing the TgCIA1 protein. https://doi.org/10.1371/journal.pbio.3003520.g004 To test whether the other candidate CIA pathway proteins were present in protein complexes, we also undertook BN-PAGE western blotting on proteins extracted from the rTgTah18-mAID-HA, rTgDre2-mAID-HA and rHA-mAID-TgNar1 lines. We observed TgTah18-mAID-HA in a protein complex of ~330 kDa, but were unable to clearly detect the TgDre2-mAID-HA or HA-mAID-TgNar1 proteins in complexes (S6A Fig). These data suggest that TgTah18, TgDre2, and TgNar1 are not components of the >720 kDa CTC. If the TgCIA1, TgCIA2, and TgMMS19 proteins exist in the same complex, we reasoned that depleting one would result in alterations to the protein complex in which the other two proteins reside. We initially introduced Ty1-epitope tags into the native TgCIA2 and TgMMS19 loci of the rTgCIA1-mAID-HA line, but were subsequently unable to detect robust TgCIA2-Ty1 and TgMMS19-Ty1 expression, possibly because the antibody we have to the Ty1 epitope is not of sufficiently high affinity. Since we knew we could detect the HA-tagged versions of these proteins, we generated a new IAA-regulatable TgCIA1-expressing line by fusing a mAID-cMyc tag into the 3′ region of the TgCIA1 open reading frame (S7A Fig). We then introduced a HA tag into the 3′ region of the TgCIA2 or TgMMS19 open reading frames of this parasite line (S7B and S7C Fig). We cultured the resulting rTgCIA1-mAID-cMyc/TgCIA2-HA and rTgCIA1-mAID-cMyc/TgMMS19-HA parasites in the absence or presence of IAA for 24 h, separated proteins by SDS-PAGE, and performed western blotting. We observed robust depletion of the TgCIA1-mAID-cMyc protein upon the addition of IAA (Fig 4B). We also a observed an ~80% depletion of the TgCIA2-HA protein, and a small but significant ~20% depletion of the TgMMS19-HA protein (Fig 4B and 4C). To determine the timing of the observed TgCIA2-HA depletion upon TgCIA1 knockdown, we cultured rTgCIA1-mAID-cMyc/TgCIA2-HA parasites in IAA for a range of times between 0 and 12 h. We observed a depletion of the TgCIA1-mAID-cMyc protein as soon as 3 h after IAA addition, and a concomitant and significant decrease in the abundance of the TgCIA2-HA protein (S6B Fig). These data indicate that TgCIA1 knockdown impacts the stability of the TgCIA2 and, to a lesser extent, the TgMMS19 proteins. We next extracted proteins from rTgCIA1-mAID-cMyc/TgCIA2-HA and rTgCIA1-mAID-cMyc/TgMMS19-HA parasites cultured in the absence or presence of IAA for 24 h, separated proteins by BN-PAGE, and performed western blotting. TgCIA1-mAID-cMyc knockdown resulted in a depletion and reduced molecular mass in both the TgCIA2-HA- and TgMMS19-HA-containing protein complexes (Fig 4D), with the depletion of the TgCIA2-HA complex occurring concomitantly with TgCIA1-mAID-cMyc knockdown (S6C Fig). We next undertook the reverse experiment, asking whether depletion of TgCIA2 or TgMMS19 resulted in changes in the abundance of TgCIA1 and impairment of the TgCIA1-containing protein complexes. We integrated a spaghetti monster-fluorescent protein-cMyc (smFP-cMyc) tag into the TgCIA1 locus of rTgCIA2-mAID-HA and rTgMMS19-mAID-HA parasites (S7D Fig). We cultured the rTgCIA2-mAID-HA/TgCIA1-smFP-cMyc and rTgMMS19-mAID-HA/TgCIA1-smFP-cMyc lines in the absence or presence of IAA for 24 h and performed both SDS-PAGE and BN-PAGE western blotting. We observed no changes to TgCIA1-smFP-cMyc protein abundance upon the knockdown of either TgCIA2-mAID-HA or TgMMS19-mAID-HA (Fig 4E). However, we observed a complete loss of the >720 kDa TgCIA1-smFP-cMyc-containing complex upon the knockdown of both TgCIA2-mAID-HA and TgMMS19-mAID-HA (Fig 4F, black arrowhead). The two smaller TgCIA1-containing complexes were unaffected by TgCIA2 or TgMMS19 depletion (Fig 4F, red and blue arrowheads). As a direct test for whether TgCIA1 and TgCIA2 interact, we performed co-immunoprecipitation experiments on the rTgCIA2-mAID-HA/TgCIA1-smFP-cMyc parasite line. Immunoprecipitation of TgCIA1-smFP-cMyc with anti-cMyc beads co-purified TgCIA2-mAID-HA but not the mitochondrial outer membrane protein TgTom40, and immunoprecipitation of TgCIA2-mAID-HA with anti-HA beads co-purified TgCIA1-smFP-cMyc but not TgTom40 (S6D Fig), providing further evidence that TgCIA1 and TgCIA2 are components of the same protein complex. Taken together, our data are consistent with TgCIA1, TgCIA2, and TgMMS19 existing in the same complex of >720 kDa. The CTC of animals exists as a heteromeric complex consisting of two copies of each of the three proteins [10]. The combined mass of a similarly arranged complex in T. gondii would be ~850 kDa, which is conceivably the mass we observe for the TgCIA1, TgCIA2, and TgMMS19 complexes in BN-PAGE. Our data also indicate that TgCIA1 exists in smaller protein complexes that do not include TgCIA2 or TgMMS19. A novel loop in the TgCIA1 protein facilitates the dual cytosolic and mitochondrial localization of the CIA targeting complex in T. gondii The candidate CTC proteins of T. gondii exhibit an interesting dual localization to the mitochondrion and cytosol that is suggestive of a dynamic role for this complex in parasite biology. We set out to characterize the mitochondrial targeting of the CTC in more detail. We first asked whether TgCIA1 was required for the mitochondrial targeting of TgCIA2 and TgMMS19. We performed immunofluorescence assays on rTgCIA1-mAID-cMyc/TgCIA2-HA or rTgCIA1-mAID-cMyc/TgMMS19-HA parasites cultured for 3 h in the absence or presence of IAA, probing for the HA-tagged TgCIA2 and TgMMS19 proteins. As expected, both TgCIA2-HA and TgMMS19-HA exhibited dual mitochondrial and cytosolic localization in parasites cultured in the absence of IAA (Fig 5A and 5B). Notably, however, the mitochondrial localization of both TgCIA2-HA and TgMMS19-HA was significantly reduced upon TgCIA1-mAID-cMyc depletion (Fig 5A and 5B). This indicates that both TgCIA2 and TgMMS19 depend on TgCIA1 for their mitochondrial targeting. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. Localization of the CTC to the mitochondrion of Toxoplasma gondii is mediated by TgCIA1. (A–D) Immunofluorescence assays of (A) rTgCIA1-mAID-cMyc/TgCIA2-HA, (B) rTgCIA1-mAID-cMyc/TgMMS19-HA, (C) rTgCIA2-mAID-HA/TgCIA1-smFP-cMyc, and (D) rTgMMS19-mAID-HA/TgCIA1-smFP-cMyc parasites, cultured in the absence (top) or presence (bottom) of IAA for either 3 h (A, B) or 24 h (C, D). Samples were probed with anti-HA (A, B) or anti-cMyc (C, D) antibodies to detect the protein-of-interest (green) and anti-TgTom40 antibodies to detect the mitochondrion (magenta). Scale bars are 2 µm. DIC, differential interference contrast. Middle, corresponding fluorescence profiles depicting the intensity of anti-HA or anti-cMyc labeling (green) and the anti-TgTom40 labeling (magenta) along the yellow line on merged images. Right, the correlation between the protein of interest and TgTom40 was quantified using the Pearson correlation coefficient (r) and analyzed using an unpaired t test, with the p values shown. The numerical data underlying this Figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003520.g005 We next asked whether the mitochondrial localization of TgCIA1 is dependent on either TgCIA2 or TgMMS19. We cultured rTgCIA2-mAID-HA/TgCIA1-smFP-cMyc or rTgMMS19-mAID-HA/TgCIA1-smFP-cMyc parasites in the absence or presence of IAA for 24 h and performed immunofluorescence assays. Neither the depletion of TgCIA2-mAID-HA nor TgMMS19-mAID-HA significantly altered the localization of TgCIA1-smFP-cMyc (Fig 5C and 5D), indicating that the mitochondrial localization of TgCIA1 is not dependent on either TgCIA2 or TgMMS19. Taken together, these data indicate that targeting of the CTC to the mitochondrion of T. gondii is mediated by TgCIA1. Next, we asked whether TgCIA1 targeting to the mitochondrion is the result of interactions with other mitochondrially-localized components of the CIA pathway. We showed previously that TgNBP35, the proposed scaffold of the CIA pathway, is anchored to the mitochondrial outer membrane courtesy of an N-terminal TMD [15]. We cultured rTgNBP35-cMyc/TgCIA1-HA parasites in the absence or presence of ATc for 2 days to deplete TgNBP35-cMyc abundance, then performed immunofluorescence assays with anti-HA antibodies to determine TgCIA1-HA localization. We found that TgCIA1-HA continued to localize to the mitochondrion following TgNBP35-cMyc depletion (S8A Fig). We next tested whether TgTah18, another mitochondrially-localized component of the CIA pathway in T. gondii (Fig 1G), is important for the mitochondrial localization of TgCIA1. We integrated a smFP-cMyc tag into the TgCIA1 locus of rTgTah18-mAID-HA parasites (S8B Fig). We cultured rTgTah18-mAID-HA/TgCIA1-smFP-cMyc parasites in the absence or presence of IAA for 24 h and performed immunofluorescence assays. We found that the localization of TgCIA1-smFP-cMyc was unchanged upon TgTah18-mAID-HA depletion (S8C Fig). Taken together, these data indicate that the mitochondrial targeting of TgCIA1 is independent of TgNBP35 and TgTah18. A recent study found that Nar1 homologs from other eukaryotes harbor a C-terminal amino acid motif that facilitates Nar1 interaction with the CTC [25]. This motif, which includes a C-terminal tryptophan residue, appears to be conserved in TgNar1 (S8D Fig), which could explain why we were unable to C-terminally tag this protein. To test whether mitochondrially-localized TgNar1 facilitates the mitochondrial localization of TgCIA1, we introduced a smFP-cMyc tag into the TgCIA1 locus of rHA-mAID-TgNar1 parasites (S8B Fig). We cultured the resulting rHA-mAID-TgNar1/TgCIA1-smFP-cMyc parasites in the absence or presence of IAA for 24 h and performed both SDS-PAGE and BN-PAGE western blotting. Depleting HA-mAID-TgNar1 did not affect the abundance of the TgCIA1-smFP-cMyc protein or the formation of the TgCIA1-smFP-cMyc-containing protein complexes (S8E and S8F Fig). We were also unable to detect TgCIA1-smFP-cMyc protein upon immunoprecipitation of HA-mAID-TgNar1 protein (S8G Fig), suggesting that these proteins do not stably interact. This is consistent with our BN-PAGE data that TgNar1 is not part of the CTC (S6A Fig), although our data do not rule out that transient interactions occur between these proteins. Finally, we cultured rHA-mAID-TgNar1/TgCIA1-smFP-cMyc parasites in the absence or presence of IAA for 24 h and performed immunofluorescence assays to detect the TgCIA1-smFP-cMyc protein. We found that the dual mitochondrial/cytosolic localization of TgCIA1-smFP-cMyc was unchanged upon HA-mAID-TgNar1 depletion (S8H Fig). Taken together, these data indicate that the mitochondrial localization of TgCIA1 is not due to interactions with mitochondrially localized TgNar1. We looked to the AlphaFold2 predicted structure of TgCIA1 for clues into what could be mediating the targeting of TgCIA1 (and by extension, the CTC) to the mitochondrion. CIA1 belongs to the WD40 protein family, which commonly mediate protein-protein interactions [26,27]. The CIA1 proteins from yeast and animals contain characteristic WD40 repeat domains folded into a 7-bladed β-propeller arranged around a central axis (Fig 6A; [10,28]). Each WD40 domain (or “blade”) is comprised of four antiparallel β-strands (denoted A–D in Fig 6A, pink) with small loops of 5–12 amino acids connecting the C and D strands (CD loops; Fig 6A; [10,28]). In the AlphaFold2 predicted structure of TgCIA1, the 7-bladed β-propeller is structurally conserved. However, the TgCIA1 protein contains three large insertions compared to the yeast and animal proteins (Figs 6A and S9; [10,28]). These insertions are located in the CD loops of the first, third, and fifth β-propeller blades, and we termed these the CD1, CD3, and CD5 loops, respectively (Fig 6A). The CD1 loop is 122 amino acids long in TgCIA1, the CD3 loop is 82 amino acids long, and the CD5 loop is 289 amino acids (S9 Fig). Extended CD loops are also present in the CIA1 homologs of other myzozoans (S9 Fig), a eukaryotic taxon that includes apicomplexans and their closest free-living relatives such as chrompodellids and dinozoans [13,14]. This is most apparent in the CD5 loop of myzozoan CIA1 proteins, which ranges in length from 45 amino acids in the oyster parasite P. marinus to 313 amino acids in the malaria-causing parasite P. falciparum (S9 Fig). This is in contrast to the much shorter, six-amino acid-long CD5 loop of CIA1 in the ciliate Paramecium tetraurelia, a sister taxon to the myzozoans. The extended CD loops in the TgCIA1 structure are predicted with poor confidence by AlphaFold2, and for the most part lack clear structural elements (Fig 6A). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 6. An elongated CD5 loop is necessary for targeting TgCIA1 to the mitochondrion of Toxoplasma gondii. (A) Structure of the Saccharomyces cerevisiae protein ScCIA1 (PDB: 2HES) and the predicted AlphaFold2 structure of TgCIA1(UniProt ID: A0A125YRQ0), with the A, B, C, and D strands of the third propellor blade in the ScCIA1 protein labeled in pink. The loops connecting the C and D strands of blade 1 (CD1 loop; red), blade 3 (CD3 loop, yellow), and blade 5 (CD5 loop; green) in each structure were colored using EzMol software [63]. (B, E) Western blots of proteins extracted from rTgCIA1-mAID-HA/cTgCIA1WT-Ty1 (WT-Ty1), rTgCIA1-mAID-HA/cTgCIA1ScCD1-Ty1 (ScCD1-Ty1), rTgCIA1-mAID-HA/cTgCIA1ScCD3-Ty1 (ScCD3-Ty1), and rTgCIA1-mAID-HA/cTgCIA1ScCD5-Ty1 (ScCD5-Ty1) parasites cultured in the absence or presence of IAA for 24 h, separated by (B) SDS-PAGE or (E) BN-PAGE, and probed with anti-Ty1, anti-HA, or anti-TgTom40 antibodies. The masses of the predicted CTC (black arrowhead) and smaller TgCIA1-containing complexes (red and blue arrowheads) are depicted in the BN-PAGE data. (C) Immunofluorescence assays of rTgCIA1-mAID-HA/cTgCIA1WT-Ty1 (WT-Ty1), rTgCIA1-mAID-HA/cTgCIA1ScCD1-Ty1 (ScCD1-Ty1), rTgCIA1-mAID-HA/cTgCIA1ScCD3-Ty1 (ScCD3-Ty1), and rTgCIA1-mAID-HA/cTgCIA1ScCD5-Ty1 (ScCD5-Ty1) parasites, probed with anti-Ty1 (green) and anti-TgTom40 (magenta) antibodies. Scale bars are 2 µm. DIC, differential inference contrast. Right, corresponding fluorescence profiles depicting the intensity of anti-Ty1 (green) and anti-TgTom40 (magenta) labeling along the yellow line on merged images. (D) The correlation between Ty1-tagged proteins and TgTom40 was quantified using the Pearson correlation coefficient (r). The data were analyzed using a one-way ANOVA followed by Tukey’s multiple comparisons test with p values shown. (F) Fluorescence proliferation assays of rTgCIA1-mAID-HA, rTgCIA1-mAID-HA/cTgCIA1WT-Ty1, rTgCIA1-mAID-HA/cTgCIA1ScCD1-Ty1, rTgCIA1-mAID-HA/cTgCIA1ScCD3-Ty1, and rTgCIA1-mAID-HA/cTgCIA1ScCD5-Ty1 parasites cultured in the absence (black) or presence (red) of IAA. Parasite proliferation is expressed as a percentage of the final fluorescence measurement in the -IAA condition for each line. Individual data points and error bars represent the mean ± SD of three technical replicates. Error bars not visible are smaller than the symbol. Data are representative of three independent experiments. The numerical data underlying this Figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003520.g006 Notably, the CD loops are located on the opposite side of the CIA1 protein from that which interacts with CIA2 [10]. This places the CD loops in position to be involved in interactions external to the CTC. We therefore tested whether the extended CD1, CD3, or CD5 loops of TgCIA1 could have a role in mitochondrial targeting. We generated TgCIA1 transgenes in which the endogenous CD1, CD3, or CD5 loops of TgCIA1 were substituted for the equivalent, but substantially shorter, loops from the cytosolic CIA1 protein of yeast. We fused the resulting transgenes to a Ty1 epitope tag to enable their detection, and overexpressed them from the constitutive α-tubulin promoter in the rTgCIA1-mAID-HA line. We termed the resulting proteins cTgCIA1ScCD1-Ty1, cTgCIA1ScCD3-Ty1, and cTgCIA1ScCD5-Ty1, respectively, and also included a cTgCIA1WT-Ty1 control in our analyses. SDS-PAGE western blot analysis confirmed that each protein was expressed in T. gondii and were of the expected masses (Fig 6B). We observed some differences in expression levels between the constitutively-expressed proteins, with the cTgCIA1WT-Ty1 and cTgCIA1ScCD1-Ty1 proteins more abundant than the cTgCIA1ScCD3-Ty1 and cTgCIA1ScCD5-Ty1 proteins (S10A Fig). To test the localization of the modified proteins, we performed immunofluorescence assays. As expected, the cTgCIA1WT-Ty1 protein exhibited dual localization to the cytosol and mitochondrion (Fig 6C). Both cTgCIA1ScCD1-Ty1 and cTgCIA1ScCD3-Ty1 proteins also localized dually to the mitochondrion and cytosol (Fig 6C). This infers that the native CD1 and CD3 loops of TgCIA1 are not important for mitochondrial targeting, although we observed that rTgCIA1-mAID-HA/cTgCIA1ScCD1-Ty1 parasites exhibited aberrant mitochondrial morphology, with ~75% of parasites containing “tadpole”-like mitochondria instead of the typical “lasso” and ‘branched’ structures that mitochondria adopt in intracellular parasites (S10B Fig; [29]). In contrast to the other CD loop mutants, the mitochondrial localization of the cTgCIA1ScCD5-Ty1 protein was significantly reduced, with the protein localized predominantly to the cytosol (Fig 6C and 6D). This indicates that the extended CD5 loop of TgCIA1 is necessary for mitochondrial targeting. We next explored the importance of the extended CD loops for TgCIA1 function. We first performed BN-PAGE western blotting. Each of the CD loop mutants associated with a complex of ~720 kDa, which likely represents the CTC (Fig 6E; black arrowhead). This implies that the CD1, CD3, and CD5 loops are not required for TgCIA1 to assemble into the CTC. The cTgCIA1WT-Ty1, cTgCIA1ScCD1-Ty1, and cTgCIA1ScCD3-Ty1 proteins were all found in the smaller TgCIA1 complexes (Fig 6E; red and blue arrowheads), similar in mass to those we had observed previously with the natively tagged TgCIA1-HA protein (Fig 4A). The smaller TgCIA1 complexes in the cTgCIA1WT-Ty1 and cTgCIA1ScCD1-Ty1 lines were, in proportion to the ~720 kDa complex, more abundant than for the natively tagged TgCIA1-HA protein (compare Figs 4A to 6E), possibly an artifact of protein overexpression from the non-native α-tubulin promoter. Curiously, the cTgCIA1ScCD5-Ty1 protein did not appear to be present in the smaller mass complexes (Fig 6E), suggesting a role for the CD5 loop in assembly of these complexes. Next, we tested whether the CD1, CD3, or CD5 loops of TgCIA1 are important for parasite proliferation. We conducted fluorescence proliferation and plaque assays on rTgCIA1-mAID-HA parasites constitutively expressing WT TgCIA1 or the CD loop mutants in the absence or presence of IAA. As expected, the severe proliferation defect observed upon TgCIA1-mAID-HA depletion was rescued by constitutive expression of cTgCIA1WT-Ty1 (Figs 6F and S10C). cTgCIA1ScCD1-Ty1-expressing parasites exhibited a moderate proliferation defect upon TgCIA1-mAID-HA depletion, resulting in fewer (rather than smaller) plaques (Figs 6F and S10C). This suggests a possible role for the CD1 loop in processes that affect the viability of extracellular parasites or the ability of parasites to invade host cells. Parasites expressing the cTgCIA1ScCD3-Ty1 protein exhibited only minor defects in proliferation when the TgCIA1-mAID-HA protein was depleted (Figs 6F and S10C). Notably, cTgCIA1ScCD5-Ty1-expressing parasites exhibited a severe impairment of parasite proliferation when TgCIA1-mAID-HA was depleted, indistinguishable from the proliferation defect observed in non-complemented rTgCIA1-mAID-HA parasites (Figs 6F and S10C). This indicates that the CD5 loop is critical for TgCIA1 protein function. Having demonstrated that the CD5 loop of TgCIA1 is necessary for mitochondrial targeting, we wondered whether the CD5 loop alone could mediate mitochondrial protein targeting. We inserted the CD5 loop of TgCIA1 into a green fluorescent protein (GFP)-Ty1 reporter at an internal site in GFP shown previously to tolerate insertions (between the eighth and ninth β-strands of GFP; [30]). We expressed GFPTgCD5-Ty1 in T. gondii parasites and attempted to select parasites stably expressing the transgene. We found that, following selection, very few parasites expressed the GFPTgCD5-Ty1 protein. In those that did, the GFPTgCD5-Ty1 protein exhibited a dual localization to both the mitochondrion and cytosol (S11 Fig). However, we noticed that the mitochondrial morphology in these parasites appeared aberrant, suggesting a potential toxic effect of GFPTgCD5-Ty1 overexpression that complicates our interpretation of the data. We next generated a transgene encoding the structurally characterized Drosophila melanogaster CIA1 protein (DmCIA1; [10]) in which we replaced the native CD5 loop of DmCIA1 with the TgCIA1 CD5 loop. We expressed cDmCIA1TgCD5-Ty1 or a corresponding cDmCIA1WT-Ty1 protein in rTgCIA1-mAID-HA parasites and performed immunofluorescence assays to determine protein localization. As expected, cDmCIA1WT-Ty1 localized in the cytosol and did not overlap with the mitochondrion (Fig 7A). By contrast, the cDmCIA1TgCD5-Ty1 protein co-localized with the mitochondrial marker (Fig 7A). Taken together, these data indicate that the CD5 loop of TgCIA1 is sufficient to mediate mitochondrial localization. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 7. The CD5 loop of TgCIA1 is sufficient for mitochondrial targeting, with the positioning of the loop important for TgCIA1 function. (A) Immunofluorescence assays of parasites constitutively expressing cDmCIA1WT-Ty1 (top) or cDmCIA1TgCD5-Ty1, (bottom) probed with anti-Ty1 (green) and anti-TgTom40 (magenta; mitochondrion) antibodies. Scale bars are 2 µm. DIC, differential interference contrast. Right, corresponding fluorescence plots depicting the intensity of anti-Ty1 (green) and anti-TgTom40 (magenta) labeling along the yellow line in merged images. (B) Constitutive expression of a cTgCIA1CD5-to-CD1-Ty1 protein variant in rTgCIA1-mAID-HA parasites, shown in schematic (top), and probed in an immunofluorescence assay with anti-Ty1 (green) and anti-TgTom40 (magenta; mitochondrion) antibodies (bottom). Scale bars is 2 µm. DIC, differential interference contrast. Right, corresponding fluorescence plot depicting the intensity of anti-Ty1 (green) and anti-TgTom40 (magenta) labeling along the yellow line in merged image. Left, western blot of proteins extracted from rTgCIA1-mAID-HA/cTgCIA1CD5-to-CD1-Ty1 parasites, separated by SDS-PAGE and probed with anti-Ty1 antibodies. (C) Fluorescence proliferation assays of rTgCIA1-mAID-HA, rTgCIA1-mAID-HA/cTgCIA1WT-Ty1, and rTgCIA1-mAID-HA/cTgCIA1CD5-to-CD1-Ty1 parasites, cultured in the absence (black) or presence (red) of IAA. Parasite proliferation is expressed as a percentage of the fluorescence measurement in the -IAA condition on the final day of the assay for each line. Individual data points and error bars represent the mean ± SD of three technical replicates. Error bars not visible are smaller than the symbol. Data are representative of three independent experiments. The numerical data underlying this Figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003520.g007 We next asked whether the position of the CD5 loop in the TgCIA1 protein was important for mitochondrial localization or protein function. We engineered a TgCIA1 construct in which the amino acid sequence encoding the CD5 loop of TgCIA1 was located in the CD1 loop of the protein, with the native CD5 loop replaced by the equivalent CD5 loop of yeast (Fig 7B). We constitutively expressed the resulting protein, which we termed cTgCIA1CD5-to-CD1-Ty1, in rTgCIA1-mAID-HA parasites, validated expression by western blotting (Fig 7B), and performed an immunofluorescence assay to determine localization of the protein. Interestingly, we observed that the cTgCIA1CD5-to-CD1-Ty1 protein localized exclusively to the mitochondrion, no longer exhibiting the dual localization we observed in the wild type TgCIA1 protein (Fig 7B). We also found that constitutive expression of the cTgCIA1CD5-to-CD1-Ty1 protein was unable to rescue the proliferation defect observed upon rTgCIA1-mAID-HA knockdown (Figs 7C and S10D). Taken together, our data indicate that the mitochondrial targeting of the CTC is mediated by TgCIA1. Specifically, the CD5 loop of TgCIA1 is both necessary and sufficient for mitochondrial targeting, and this targeting is independent of the other mitochondrially-localized CIA pathway proteins TgNBP35, TgTah18, and TgNar1. Our data also indicate that, while the position of the CD5 loop in the protein is not critical for mitochondrial targeting, it is critical for facilitating the dual localization of TgCIA1 to the cytosol and mitochondrion. Finally, we have shown that the CD5 loop of TgCIA1, and its positioning within the protein, is critical for TgCIA1 to carry out its functions in parasites. A myzozoan-specific amino acid motif in the CD5 loop mediates the mitochondrial localization of TgCIA1 We next set out to uncover the features of the CD5 loop of TgCIA1 that facilitate mitochondrial targeting. Alignments of the CIA1 protein from a range of eukaryotes identified the presence of a short, conserved motif consisting of three aromatic and one positively charged amino acid in the CD5 loop of the CIA1 protein in T. gondii and other myzozoans (Figs 8A and S9). This motif (and the extended CD5 loop generally) was not present in other eukaryotic clades, including ciliates such as P. tetraurelia, which are the nearest relatives of the myzozoans [14]. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 8. An aromatic amino acid motif in the CD5 loop of TgCIA1 facilitates mitochondrial targeting. (A) Left, a region of a multiple sequence alignment of CIA1 homologs highlighting a motif of amino acid residues within the CD5 loop of the protein that is conserved in myzozoans. Asterisks denote key residues of the motif, with green denoting aromatic residues and orange denoting a positively charged residue. Right, a schematic of the motif observed in the dinozoans Perkinsus marinus and Symbiodinium microadriaticum, the apicomplexans Plasmodium falciparum, Cryptosporidium parvum, and Toxoplasma gondii, and the chrompodellids Vitrella brassicaformis and Chromera velia. Green = aromatic, orange = positively charged, gray = not conserved. Amino acid length of the CD5 loop in each species is listed on the right. (B, C) Western blots of proteins extracted from TgCIA1WT-HA (WT-HA), TgCIA1W526A-HA (W526A-HA), TgCIA1Y527A-HA (Y527A-HA), and TgCIA1R533A-HA (R533A-HA) expressing parasites, separated by (B) SDS-PAGE or (C) BN-PAGE, and probed with anti-HA or anti-TgTom40 antibodies. The black arrowhead indicates the >720 kDa CIA Targeting Complex; red and blue arrowheads indicate the lower mass complexes containing the TgCIA1 protein. (D) Immunofluorescence assays of TgCIA1WT-HA (WT-HA), TgCIA1W526A-HA (W526A-HA), TgCIA1Y527A-HA (Y527A-HA), and TgCIA1R533A-HA (R533A-HA) parasites. The proteins of interest (green) and the mitochondrion (magenta) were labeled with anti-HA and anti-TgTom40 antibodies, respectively. Schematics depicting the modified amino acid sequence in the CD5 motif of the proteins from each panel are included next to images (left). Scale bars are 2 µm. DIC, differential interference contrast. Right, corresponding fluorescence profile depicting intensity of anti-HA (green) and anti-TgTom40 (magenta) labeling along the yellow lines of the merged images. (E) The correlation between HA-tagged proteins of interest and TgTom40 was quantified using the Pearson correlation coefficient (r) and the data were analyzed using a one-way ANOVA followed by Tukey’s multiple comparisons test. (F) Plaque assays of TgCIA1WT-HA, TgCIA1W526A-HA, TgCIA1Y527A-HA, and TgCIA1R533A-HA parasites. Parasites were cultured in the absence (top) or presence (bottom) of IAA for 8 days and are representative of three independent experiments. (G) Immunofluorescence assays of rTgCIA1-mAID-HA/cTgCIA1WT-Ty1 and rTgCIA1-mAID-HA/cTgCIA1F532A-Ty1 parasites. The cTgCIA1WT-Ty1 protein (green) and the mitochondrion (magenta) were labeled with anti-Ty1 and anti-TgTom40 antibodies, respectively. Schematics depicting the amino acid sequence in the CD5 motif of the proteins from each panel are included next to images (left). Scale bars are 2 µm. DIC, differential interference contrast. Right, corresponding fluorescence profile depicting intensity of anti-Ty1 (green) and anti-TgTom40 (magenta) labeling along the yellow lines of the merged images. (H) The correlation between Ty1-tagged proteins and TgTom40 was quantified using the Pearson correlation coefficient (r). The data were analyzed using a one-way ANOVA (alongside the r values depicted in S12F Fig) followed by Tukey’s multiple comparisons test, with the p value shown. (I) Fluorescence proliferation assays of rTgCIA1-mAID-HA, rTgCIA1-mAID-HA/cTgCIA1WT-Ty1, and rTgCIA1-mAID-HA/cTgCIA1F532A-Ty1 parasites, grown in the absence (black) or presence (red) of IAA. Parasite proliferation is expressed as a percentage of the fluorescence measurement in the -IAA condition on the final day of the assay for each line. Individual data points and error bars represent the mean ± SD of three technical replicates. Error bars not visible are smaller than the symbol. Data are representative of three independent experiments. The numerical data underlying this Figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003520.g008 We hypothesized that the conserved motif of the CD5 loop could facilitate the mitochondrial targeting of TgCIA1. To test this, we used CRISPR/Cas9-based genome editing to individually substitute each residue in the motif for alanine in the native TgCIA1 locus of the TgCIA1-HA/rTgNBP35-cMyc parasite line. We were successful in generating mutants in the W526, Y527, and R533 residues (but not the F532 residue), terming the resulting proteins TgCIA1W526A-HA, TgCIA1Y527A-HA, and TgCIA1R533A-HA (S12A Fig). SDS-PAGE western blot analyses confirmed that the mutated proteins were expressed at similar abundances to a TgCIA1WT-HA control (Figs 8B and S12B). We next performed BN-PAGE western blotting and found that all mutated proteins were present in the >720 kDa CTC as well as in the smaller TgCIA1-containing complexes we had observed previously (Fig 8C). To test whether the motif is important for mitochondrial targeting, we performed immunofluorescence assays. Notably, all mutations in the CD5 loop motif resulted in the TgCIA1 protein localizing predominantly to the cytosol (Fig 8D), although quantifications revealed that the TgCIA1R533A-HA protein exhibited significantly greater mitochondrial co-localization than the TgCIA1W526A-HA and TgCIA1Y527A-HA proteins (Fig 8E). Next, we investigated whether the W526, Y527, or R533 residues of the conserved CD5 loop motif contribute to the role of TgCIA1 in parasite proliferation. We performed plaque assays comparing the proliferation of parasites expressing TgCIA1W526A-HA, TgCIA1Y527A-HA, and TgCIA1R533A-HA to parasites expressing TgCIA1WT-HA. We found that TgCIA1W526A-HA and TgCIA1Y527A-HA expressing parasites exhibited severe proliferation defects, indicating that these residues are critical for TgCIA1 function (Fig 8F). By contrast, the proliferation of parasites expressing TgCIA1R533A-HA was indistinguishable from those of parasites expressing TgCIA1WT-HA, suggesting the R533 residue is largely dispensable for TgCIA1 function (Fig 8F). We were unable to modify the TgCIA1 locus to express a F532A mutation using the genome editing approach. As an alternative, we constitutively expressed TgCIA1F532A-Ty1 from the α-tubulin promoter in rTgCIA1-mAID-HA parasites, generating a line we termed rTgCIA1-mAID-HA/cTgCIA1F532A-Ty1. We found that the cTgCIA1F532A-Ty1 protein was expressed at a similar abundance, and in protein complexes of similar masses, to the cTgCIA1WT-Ty1 protein (S12C and S12D Fig). Notably, we found that the cTgCIA1F532A-Ty1 protein localized predominantly to the cytosol, exhibiting significantly less mitochondrial co-localization than the cTgCIA1WT-Ty1 protein (Fig 8G and 8H). We also constitutively-expressed TgCIA1W526A-Ty1, TgCIA1Y527A-Ty1 and cTgCIA1R533A-Ty1 isoforms in rTgCIA1-mAID-HA parasites, and found that the localization of these matched what we observed in the genome-edited point mutants, with all TgCIA1 variants localizing predominantly to the cytosol, although the cTgCIA1R533A-Ty1 again exhibited greater mitochondrial co-localization than the other variants (S12E and S12F Fig). Finally, we undertook fluorescence proliferation assays and plaque assays to test whether the F532 residue of the CD5 loop is important for TgCIA1 function. We compared proliferation of rTgCIA1-mAID-HA/cTgCIA1F532A-Ty1 parasites to rTgCIA1-mAID-HA and rTgCIA1-mAID-HA/cTgCIA1WT-Ty1 parasite lines cultured in the absence or presence of IAA. We observed that proliferation of rTgCIA1-mAID-HA/cTgCIA1F532A parasites was severely impaired when TgCIA1-mAID-HA was knocked down (Figs 8I and S10E), indicating that the F532 residue is essential for TgCIA1 protein function. We also tested the proliferation of rTgCIA1-mAID-HA/TgCIA1W526A-Ty1, rTgCIA1-mAID-HA/TgCIA1Y527A-Ty1, and rTgCIA1-mAID-HA/cTgCIA1R533A-Ty1 parasites in the absence or presence of IAA. This revealed that rTgCIA1-mAID-HA/cTgCIA1Y527A-Ty1 and rTgCIA1-mAID-HA/cTgCIA1R533A-Ty1 parasites proliferated normally upon knockdown of the TgCIA1-mAID-HA protein, whereas proliferation of rTgCIA1-mAID-HA/cTgCIA1W526A-Ty1 parasites was substantially reduced upon TgCIA1-mAID-HA depletion, although not to the same extent as observed in rTgCIA1-mAID-HA/cTgCIA1F532A-Ty1 parasites (S12G and S10E Figs). These findings suggest that constitutive overexpression of the TgCIA1W526A and TgCIA1Y527A mutant isoforms can partially or fully rescue the severe proliferation defects observed in the genome-edited point mutants. Taken together, these data indicate that the W526, Y527, and F532 residues of the aromatic amino acid motif in the CD5 loop of TgCIA1 are critical for both mitochondrial localization of the TgCIA1 protein and for the functional role of TgCIA1 in parasite proliferation. The R533 residue contributes to the mitochondrial localization of TgCIA1, although not to the same extent as the other residues of this motif that we tested. Surprisingly, despite its role in mitochondrial targeting, the R533 residue of TgCIA1 appears to be dispensable for TgCIA1 function. Given the conservation of the CD5 loop motif in myzozoans, we asked whether CD5 loops from other myzozoans could complement the function of T. gondii CD5 loop. We replaced the CD5 loop of TgCIA1 with the equivalent CD5 loop from the chrompodellid V. brassicaformis or the dinozoan S. microadriaticum, generating proteins we called cTgCIA1VbCD5-Ty1 or cTgCIA1SmCD5-Ty1. The V. brassicaformis CD5 loop contains the same residues in the conserved aromatic motif as the TgCIA1 protein (Fig 8A), but is considerably shorter than the T. gondii CD5 loop (87 amino acids versus 249 amino acids). The S. microadriaticum CD5 loop only encodes two amino acids of the motif (the phenylalanine and arginine residues; Fig 8A), and is also considerably shorter than the T. gondii CD5 loop (97 amino acids). We expressed the cTgCIA1VbCD5-Ty1 and cTgCIA1SmCD5-Ty1 proteins in rTgCIA1-mAID-HA parasites, validated expression by SDS-PAGE western blotting (Fig 9A and 9B), and performed immunofluorescence assays to determine protein localization. Both the cTgCIA1VbCD5-Ty1 and cTgCIA1SmCD5-Ty1 proteins localized predominantly to the cytosol (Fig 9A–9C), although the cTgCIA1SmCD5-Ty1 protein also exhibited some observable co-localization with the mitochondrion (Fig 9B, arrowheads). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 9. The CD5 loop of the CIA1 protein is functionally conserved in myzozoans. (A, B) Immunofluorescence assay of (A) rTgCIA1-mAID-HA/cTgCIA1VbCD5-Ty1 and (B) rTgCIA1-mAID-HA/cTgCIA1SmCD5-Ty1 parasites, probed with anti-Ty1 (green) and anti-TgTom40 (magenta; mitochondrion) antibodies. Schematics depicting the amino acid sequence in the CD5 motif of the proteins from each panel are included below the Ty1 labeled images. Scale bars are 2 µm. DIC, differential interference contrast. Arrowheads highlight regions where the cTgCIA1SmCD5-Ty1 protein exhibits visible mitochondrial localization. Right, corresponding fluorescence profile depicting intensity of anti-Ty1 (green) and anti-TgTom40 (magenta) labeling along the yellow line on merged image. Left, western blots of the proteins of interest, separated by SDS-PAGE and probed with anti-Ty1 antibodies. (C) The correlation between the Ty1-tagged proteins of interest and TgTom40 was quantified using the Pearson correlation coefficient (r). The data were analyzed using a one-way ANOVA followed by Tukey’s multiple comparisons test with p values shown. (D) Fluorescence proliferation assays of rTgCIA1-mAID-HA, rTgCIA1-mAID-HA/cTgCIA1WT-Ty1, rTgCIA1-mAID-HA/cTgCIA1VbCD5-Ty1, and rTgCIA1-mAID-HA/cTgCIA1SmCD5-Ty1 parasites, grown in the absence (black) or presence (red) of IAA. Parasite proliferation is expressed as a percentage of the fluorescence measurement in the -IAA condition on the final day of the assay for each line. Individual data points and error bars represent the mean ± SD of three technical replicates. Error bars not visible are smaller than the symbol. Results are representative of three independent experiments. The data for the rTgCIA1mAID-HA and rTgCIA1-mAID-HA/cTgCIA1WT-Ty1 lines are identical to those depicted in Fig 7C, the experiments for which were performed simultaneously. The numerical data underlying this Figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003520.g009 Next, we investigated whether the CD5 loops of the SmCIA1 and VbCIA1 homologs were functionally equivalent to CD5 loop of TgCIA1. We cultured rTgCIA1-mAID-HA/cTgCIA1VbCD5-Ty1 and rTgCIA1-mAID-HA/cTgCIA1SmCD5-Ty1 parasites in the absence or presence of IAA and performed fluorescence proliferation and plaque assays. These revealed that both the cTgCIA1SmCD5-Ty1- and cTgCIA1VbCD5-Ty1-complemented lines proliferated normally when cultured in the presence of IAA (Figs 9D and S10D), indicating that the CD5 loops from the S. microadriaticum and V. brassicaformis CIA1 proteins can functionally replace the equivalent loop in the TgCIA1 protein of T. gondii. Taken together, our data demonstrate that the CD5 loop of TgCIA1 contains a motif that is conserved throughout the myzozoans (Fig 8A), and which contributes to targeting the CIA1 protein to the mitochondrion. Numerous residues of this motif, including a phenylalanine residue from this motif that is found in all the analyzed myzozoan CIA1 sequences except Cryptosporidium parvum (Fig 8A), are critical for TgCIA1 to carry out its biological role. The CD5 loop of TgCIA1 is not required for cytosolic protein synthesis We have shown that the CTC of T. gondii exhibits dual localization to the mitochondrion and cytosol, courtesy of a myzozoan-specific CD5 loop in the CIA1 protein of the complex (Fig 10A). It is conceivable that the mitochondrial localization of TgCIA1 is important for the transfer of [4Fe-4S] clusters from mitochondrially-localized Nar1 to the CTC. TgCIA1 localizes to the mitochondrion independently of its association with TgNar1 (S8H Fig), but it is possible that the CD5 loop of TgCIA1 interacts with mitochondrial outer membrane lipids or an accessory protein of the mitochondrial outer membrane (Fig 10Ai–10Aii). This could place the CTC in position on the outer membrane to interact with TgNar1 and enable [4Fe-4S] cluster transfer to occur. In these scenarios, the CD5 loop of TgCIA1 functions as a mitochondrial targeting signal to facilitate FeS cluster transfer from TgNar1. Notably, T. gondii expresses the FeS proteins TgELP3 and TgRlmN on the outer face of outer mitochondrial membrane [31]. An alternative possibility is, therefore, that the CD5 loop instead functions in enabling TgCIA1 and the CTC complex to interact with these client FeS proteins on the outer membrane (Fig 10Aiii). In this scenario, the CD5 loop functions not in enabling [4Fe-4S] cluster transfer from TgNar1, but instead to enable [4Fe-4S] cluster transfer from the CTC to client mitochondrial outer membrane proteins. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 10. A model for the spatial organization and evolution of the CIA pathway in Toxoplasma gondii and related eukaryotes. (A) Models for the spatial organization of the CIA pathway in T. gondii and related organisms. A sulfur-containing product of the mitochondrial ISC pathway (possibly a [2Fe-2S] cluster) is exported from the mitochondrion. [4Fe-4S] clusters assemble on the NBP35 scaffold (N35) at the mitochondrial outer membrane, with electrons for this process donated from NADPH via mitochondrial Tah18 (T18) and possibly Dre2 (D2). [4Fe-4S] clusters are transferred from the NBP35 scaffold to Nar1 (N1), and further to the CIA targeting complex (CTC) comprised of CIA1 (C1), CIA2 (C2) and MMS19 (M19). CIA1 contains an extended CD5 loop (green) that mediates localization of the entire CTC to the mitochondrion. The CD5 loop may bind to (i) outer mitochondrial membrane lipids or (ii) outer mitochondrial membrane accessory proteins, and place the CTC in position to receive FeS clusters from Nar1. Alternatively, (iii) the CD5 loop may mediate interactions with FeS client proteins that are anchored on the mitochondrial outer membrane, thus facilitating FeS transfer from the CTC to these proteins. Elements of the diagram were created in BioRender. Hodgson, E. (2025) https://BioRender.com/ecxz792. (B) Western blots measuring the incorporation of puromycin into proteins from rTgCIA1-mAID-HA/cTgCIA1ScCD5-Ty1 (ScCD5) or rTgCIA1-mAID-HA/cTgCIA1CD5-to-CD1-Ty1 (CD5-to-CD1) parasites cultured in the absence (−) or presence (+) of IAA for 24 h, separated by SDS-PAGE and probed with anti-puromycin and anti-TgTom40 antibodies. (C) Relative abundance of puromycin incorporation from (B) was determined as a percentage of the -IAA control for each parasite line, with abundances normalized using the TgTom40 loading control. Data points represent the mean ± SD of three independent experiments. Data were analyzed using a one-way ANOVA followed by Tukey’s multiple comparisons test, with relevant p values shown. The numerical data underlying this Figure can be found in S1 Data. (D) An illustrative phylogenetic tree adapted from currently accepted models of eukaryotic evolution [63–65]. Mitochondrial targeting domains for key components of the CIA machinery, including the extended CD5 loop of CIA1, the N-terminal (N-term) transmembrane domain (TMD) of NBP35 and the C-terminal (C-term) TMDs of the FeS proteins Elp3 and RlmN, all evolved early in myzozoan evolution, subsequent to their divergence from the ciliate lineage. The endosymbiotic acquisition of a chloroplast also occurred early in myzozoan evolution. https://doi.org/10.1371/journal.pbio.3003520.g010 If the mitochondrial localization of TgCIA1 is critical for FeS cluster transfer from TgNar1, we predicted that impairing the mitochondrial localization of TgCIA1 will impair all downstream processes that require the CTC, such as cytosolic protein translation. We therefore measured protein translation in a parasite strain expressing a TgCIA1 variant that is unable to target to the mitochondrion (cTgCIA1ScCD5-Ty1, which lacks the mitochondrial targeting CD5 loop and localizes exclusively to the cytosol; Fig 6C). We compared this to protein translation in a parasite strain expressing a TgCIA1 variant that is targeted exclusively to the mitochondrion (cTgCIA1CD5-to-CD1-Ty1; Fig 7B). Remarkably, we still observed robust cytosolic protein translation in parasites expressing only the cytosolically localized cTgCIA1ScCD5-Ty1 protein (Fig 10B and 10C). By contrast, we observed a significant depletion of cytosolic protein translation when parasites expressed only the mitochondrially-localized cTgCIA1CD5-to-CD1-Ty1 protein (Fig 10B and 10C), similar to the defect we observed when TgCIA1 is depleted (Fig 3C and 3D). These data indicate that CD5 loop-dependent mitochondrial localization of TgCIA1 is not required for the FeS cluster-dependent process of protein translation in the cytosol. These data are, therefore, inconsistent with the hypothesis that the CD5 loop of TgCIA1 is important for FeS cluster transfer from Nar1 to the CTC at the mitochondrial outer membrane (Fig 10Ai–10Aii). Instead, our data indicate that some functions facilitated by TgCIA1 (such as cytosolic protein translation) are independent of the CD5 loop. Our data also suggest that the cytosolic localization of TgCIA1 is required to enable cytosolic protein translation, since we observed a strong translation defect in parasites expressing only the mitochondrially-localized cTgCIA1CD5-to-CD1-Ty1 protein, although we cannot rule out this defect is due to the aberrant positioning of the CD5 loop in this protein interfering with CTC function. Discussion In this study, we have examined the localization and importance of the cytosolic FeS cluster assembly (CIA) pathway of T. gondii parasites. Our data support a model wherein the CIA pathway of T. gondii occurs on the cytosolic face of the mitochondrion (Fig 10A). We propose that, like in other eukaryotes [4,5], a sulfur-containing product (perhaps a [2Fe-2S] cluster) is exported from the mitochondrion. This transport, like in other eukaryotes, involves the mitochondrial inner membrane ABC transporter TgABCB7L/TgATM1 [21,22], and could involve outer membrane porins and/or mitoNEET proteins, as proposed recently for animal cells [32]. Homologs of both porins and mitoNEET proteins are encoded in the T. gondii genome and localize to the parasite mitochondrion [33,34]. A [4Fe-4S] cluster is then assembled on a NBP35 scaffold, which localizes to the outer face of the outer mitochondrial membrane of T. gondii [15]. In other eukaryotes, [4Fe-4S] cluster assembly on the NBP35 scaffold relies on electrons donated from NAPDH via an electron transfer chain consisting of Tah18 and Dre2 [35,36]. We found that TgTah18 localizes to the mitochondrion, and that its depletion leads to a defect in cytosolic protein translation, which is also observed upon the depletion of most other candidate CIA pathway proteins in T. gondii (Figs 3 and S4; [20]). This is consistent with TgTah18 being a component of the CIA pathway in T. gondii (Fig 10A). By contrast, TgDre2 localizes to the cytosol and is not required for cytosolic protein synthesis, although loss of TgDre2 does impact the abundance of the cytosolic FeS protein TgABCE1 (S4 Fig). This suggests that, although TgDre2 is critical for parasite proliferation, it may not function in the parasite CIA pathway in the same way as the other CIA pathway proteins (Fig 10A). A limitation to this conclusion is that the assays that we have used in this study to measure CIA pathway function, and which have been published in other recent studies [15,20–22], provide only indirect measures for CIA pathway activity (i.e., measure the stability of cytosolic FeS proteins or measure processes that rely on the incorporation of FeS clusters into cytosolic proteins, such as protein translation and lipid droplet biology). A priority for future research in this area will be to develop assays that directly measure the incorporation of FeS clusters into client cytosolic proteins via the CIA pathway in these parasites. In other eukaryotes, Nar1 is proposed to facilitate the transfer of a [4Fe-4S] cluster from the NBP35 scaffold to the CTC [6,8,26,37]. We found that TgNar1 is critical for parasite proliferation and cytosolic protein translation, consistent with TgNar1 functioning in the CIA pathway. This aligns with data from a recent study by Renaud and colleagues, who showed that depletion of TgNar1 resulted in the formation of lipid droplets, a phenotype proposed to result from an impaired CIA pathway [20]. We did not observe a stable interaction between TgNar1 and the CTC (S8 Fig), although it is likely that FeS cluster transfer between Nar1 and the CTC is a transient process. TgNar1 contains a conserved C-terminal tryptophan residue that was shown to facilitate interactions between Nar1 and the CTC in mammalian cells (S8 Fig; [25]). Our data indicate that TgNar1 localizes exclusively to the mitochondrion, suggesting that the transfer of [4Fe-4S] clusters from the TgNBP35 scaffold to the CTC likely takes places on the mitochondrial outer membrane in T. gondii (Fig 10A). The CTC of T. gondii parasites exhibits a curious dual localization to both the mitochondrion and cytosol. We show that the mitochondrial association of the CTC is dependent on the TgCIA1 protein of the complex, and that the CD5 loop of TgCIA1 is both necessary and sufficient for its mitochondrial localization. Notably, we identified a conserved aromatic amino acid motif in the loop that is criticial for mitochondrial association. Aromatic amino acid residue-containing motifs have been implicated in both protein–protein and protein–lipid interactions [38,39]. We propose that the CD5 motif facilitates the mitochondrial localization of TgCIA1 by associating with either lipids or proteins on the mitochondrial outer membrane (Fig 10Ai–10Aiii). We found that deleting the CD5 loop or modifying key residues in the aromatic amino acid motif impairs both mitochondrial targeting and TgCIA1 function. This indicates that the CD5 loop, and the aromatic motif therein, plays a critical role in TgCIA1 (and, by extension, CTC) function. One interpretation of these data is that the mitochondrial localization of TgCIA1 is essential for its function. Curiously, however, mutating the arginine residue at position 533 in the aromatic amino acid motif to alanine impairs mitochondrial targeting, but does not ablate protein function (Fig 8 and S12 Fig). It is possible that the aromatic amino acid motif in the CD5 loop plays two independent roles in TgCIA1 biology: a non-essential role in mitochondrial targeting and an essential role in protein function. A limitation of our localization experiments is that they cannot rule out that some mitochondrial association remains in the TgCIA1R533A mutant protein, and indeed quantifications of the extent of overlap between the various point mutants we generated in this study consistently suggest a greater degree of overlap between the TgCIA1R533A protein and the mitochondrion than the other point mutants in the aromatic motif (Figs 8 and S12). An important remaining question is therefore what role(s) the CD5 loop and aromatic amino acid motif play in TgCIA1 protein function. We considered the possibility that the mitochondrial association of TgCIA1 that is facilitated by the CD5 loop positions the CTC to receive FeS clusters from TgNar1 at the mitochondrial outer membrane (Fig 10Ai–10Aii). However, we found that parasites expressing a TgCIA1 variant that lacked the CD5 loop (and is impaired in mitochondrial targeting) is still capable of cytosolic protein synthesis (Fig 10B and 10C), a process that relies on a functional CIA pathway [20,21]. This indicates that [4Fe-4S] cluster transfer to cytosolic proteins can still occur in the absence of the CD5 loop of CIA1, suggesting that the CD5 loop of TgCIA1 (and, by extension, the mitochondrial localization of the CTC) is not required for the maturation of all cytosolic FeS proteins. However, as noted above, our localization studies lack the resolution to definitively conclude that no mitochondrial localization occurs in the CD5 loop mutant. An alternative possibility is that the CD5 loop of TgCIA1 has a critical role in the transfer of FeS clusters from the CTC to client FeS proteins that reside on the mitochondrial outer membrane (Fig 10Aiii). In this scenario, the CD5 loop functions not as a mitochondrial targeting signal, but in facilitating interactions between the CTC and mitochondrial client proteins. T. gondii contains at least two FeS proteins, TgElp3 and TgRlmN, that localize to the mitochondrial outer membrane courtesy of C-terminal TMDs [31,40]. Both TgElp3 and TgRlmN are predicted to be important for T. gondii proliferation [40–42], so their interactions with the CTC are likely important for parasite survival. In other eukaryotes, [4Fe-4S] cluster transfer from the CTC to Elp3 is proposed to be facilitated by the adaptor protein Elp4 [25,43]. The T. gondii genome lacks an Elp4 homolog, and it is possible that the CD5 loop of TgCIA1 instead serves this adaptor role. A recent study identified TgHCF101 as a cytosolic adaptor protein that facilitates FeS cluster transfer from the CTC to TgABCE1 [20], and our study indicates that the CD5 loop of TgCIA1 is not required for cytosolic protein translation (which is dependent on TgABCE1), suggesting that different processes may mediate FeS cluster transfer from the CTC to cytosolic and mitochondrial client proteins. Future studies that examine the role of the CD5 loop in TgCIA1 function, and whether it functions in interactions with mitochondrial outer membrane proteins such as TgElp3 and TgRlmN, will be of particular interest. The extended CD5 loop of TgCIA is functionally conserved throughout the myzozoan lineage, but absent from ciliates, the closest relatives of the myzozoans (Figs 8, 9, and S9). Intriguingly, the mitochondrial targeting N-terminal TMD of TgNBP35, as well as the C-terminal mitochondrial targeting domains of TgElp3 and TgRlmN, also appear to have been acquired early in myzozoan evolution (Fig 10D; [15,16,40,41]). Taken together, these observations indicate that the evolution of the myzozoan lineage coincided with migration of the CIA pathway to the mitochondrial outer membrane, and the targeting of some [4Fe-4S] proteins to the mitochondrial outer membrane (Fig 10D). The selective forces that shaped this reorganization of the cytosolic FeS cluster assembly pathway remain unclear. It is conceivable that the relocation of FeS proteins like Elp3 and RlmN to the outer mitochondrial membrane necessitated the relocation of the CIA pathway to the mitochondrion. Interestingly, myzozoan evolution also coincided with the endosymbiotic acquisition of a chloroplast, and the ancestral myzozoan likely relied on photosynthesis for its survival [14]. Iron is an important component of the light harvesting electron transport chains of photosynthesis, and plastids require iron for a range of other important enzymatic processes [44]. The marine environments in which myzozoans evolved is poor in available iron, with iron a major limiting nutrient in the primary productivity of oceans [45,46]. This iron limitation, coupled with the acquisition of an iron-hungry plastid organelle, may have generated a selective pressure to optimize iron utilization in early myzozoans. Locating the CIA pathway to the outer surface of the mitochondrion places it in close proximity to ISC pathway in the mitochondrion, which provides the building blocks for the CIA pathway [4,47]. The mass migration of the CIA pathway to the mitochondrial surface may therefore have evolved to increase the efficiency of iron usage in myzozoans, although the extent to which the localization of the CIA pathway to the mitochondrion increases the efficiency of iron usage is unclear. Taken together, our study provides a comprehensive analysis of CIA pathway of the apicomplexan parasite T. gondii, showing that this pathway is critical for parasite survival. Our data indicate that, in contrast to the CIA pathway of animals, plants and fungi, the mitochondrion of T. gondii acts as a hub for the CIA pathway. Our study reveals the evolution of a conserved loop in the CIA1 protein of T. gondii and related eukaryotes that facilitates the mitochondrial assoication of the CTC, and which is critical for CIA1 function. The novel features of the CIA pathway that we and others have identified indicate that a ubiquitious and essential metabolic pathway can nevertheless vary considerably between different eukaryotic lineages [15,20]. Materials and methods Parasite culturing Tachyzoite-stage T. gondii parasites were cultured in human foreskin fibroblasts (HFFs) using Dulbecco’s Modified Eagle Medium (DMEM) supplemented with 1% (v/v) fetal bovine serum, 2 g/L NaHCO3, 0.2 mM L-glutamine, 50 units/mL penicillin, 50 µg/mL streptomycin, 10 µg/mL gentamicin, and 0.25 µg/mL amphotericin b. Cultures were kept in humidified incubators at 37°C and 5% CO2. Where appropriate, 0.1 mM IAA (Merck catalog number I2886), 0.5 µg/mL ATc (Merck catalog number 37919), or an equivalent volume of 100% ethanol (vehicle control) was added. Generation of genetically modified parasites We used CRISPR/Cas9 genome editing to introduce 3× HA, 3′ mini-auxin-inducible degron-3× HA (mAID-HA), 3′ mAID-3× cMyc (mAID-cMyc), 5′ HA-mAID, 3′ Ty1, or 3′ spaghetti monster fluorescent protein (smFP)-cMyc epitope tags into the 3′ or 5′ regions of the open reading frames of TgTah18 (www.ToxoDB.org identifier TGGT1_249320; [17]), TgDre2 (TGGT1_216900), TgNar1 (TGGT1_242580) TgCIA1 (TGGT1_313280), TgCIA2 (TGGT1_306590), TgMMS19 (TGGT1_222230), or TgABCE1 (TGGT1_216790). To enable, these modifications, we introduced DNA encoding single guide (sg)RNAs targeting the 3′ or 5′ regions of these genes into the vector pSAG1::Cas9-U6-UPRT (Addgene plasmid 54467, [48]) using Q5-site directed mutagenesis (New England Biolabs). We performed the Q5 reaction using the gene-specific “3′ or 5′ CRISPR fwd” primer and a “universal CRISPR rvs” primer (S1 Table). We amplified a donor DNA sequence encoding the epitope tag plus 50 bp flanks of the target gene using gene specific “tag fwd” and “tag rvs” primers and HA, C-terminal mAID-HA, N-terminal HA-mAID, mAID-cMyc, cMyc, Ty1, or smFP-cMyc gBlocks (Integrated DNA Technologies, IDT) as templates (S1 Table). We transfected the sgRNA-expressing plasmids and HA-containing donor DNAs targeting the 3′ region of the TgTah18, TgDre2, TgCIA1, TgCIA2, or TgMMS19 into an ATc regulatable TgNBP35-cMyc strain described previously [15]. This generated TgTah18-HA, TgDre2-HA, TgCIA1-HA, TgCIA2-HA, and TgMMS19-HA expressing parasites. We transfected the sgRNA-expressing plasmids targeting TgTah18, TgDre2, TgNar1, TgCIA1, TgCIA2, or TgMMS19 and C-terminal mAID-HA, N-terminal HA-mAID, or mAID-cMyc-containing donor DNAs into RH∆ku80-Tir1-FLAG parasites expressing tdTomato red fluorescent protein (described previously, [18]). This generated the IAA-regulatable rTgTah18-mAID-HA, rTgDre2-mAID-HA, rHA-mAID-TgNar1, rTgCIA1-mAID-HA, rTgCIA2-mAID-HA, rTgMMS19-mAID-HA, and rTgCIA1-mAID-cMyc parasite lines. We transfected the sgRNA-expressing plasmid targeting TgABCE1 and the corresponding cMyc donor DNA into the rTgCIA1-mAID-HA and rTgDre2-mAID-HA parasite lines to generated the rTgCIA1-mAID-HA/TgABCE1-cMyc and rTgDre2-mAID-HA/TgABCE1-cMyc lines, respectively. We transfected the sgRNA-expressing plasmids targeting TgCIA2 and TgMMS19 and the corresponding HA donor DNA into the rTgCIA1-mAID-cMyc parasite line to generate the rTgCIA1-mAID-cMyc/TgCIA1-HA and rTgCIA1-mAID-cMyc/TgMMS19-HA lines. We transfected the sgRNA-expressing plasmid targeting TgCIA1 and the corresponding smFP-cMyc donor DNA into the rTgCIA2-mAID-HA, rTgMMS19-mAID-HA, rTgTah18-mAID-HA, and rHA-mAID-TgNar1 parasite lines to generate the rTgCIA2-mAID-HA/TgCIA1-smFP-cMyc, rTgMMS19-mAID-HA/TgCIA1-smFP-cMyc, rTgTah18-mAID-HA/TgCIA1-smFP-cMyc, and rHA-mAID-TgNar1/TgCIA1-smFP-cMyc lines, respectively. Two days after each of the transfections described above, GFP-positive parasites (which express Cas9-GFP encoded on the modified pSAG1::Cas9-U6 vector) were sorted into wells of a 96-well plate using a FACSMelody cell sorter (BD Biosciences). After approximately one week, we identified wells containing clonal parasites (i.e., wells with single plaques), extracted genomic DNA from parasites in these wells, and performed PCR screening analyses to identify genetically modified parasites using gene-specific “3′ or 5′ scrn fwd” and “3′ or 5′ scrn rvs” primers (S1 Table). To complement the rTgCIA2-mAID-HA line with a constitutively-expressed copy of wild type TgCIA2, we synthesized the TgCIA2WT open reading frame and an in-frame 3′ Ty1 epitope tag as a gBlock (S1 Table). We PCR amplified the TgCIA2WT-Ty1 open reading frame with the primers CIA2 comp fwd and Ty1 rvs, digested the resulting PCR product with BglII and XmaI and ligated this into the equivalent restriction enzyme cut sites of pUDTG [49]. The resulting vector, which we termed pUDTTy(TgCIA2WT), expresses TgCIA2WT-Ty1 from the constitutive α-tubulin promoter, in a vector that contains a pyrimethamine-resistant TgDHFR selectable marker and a flanking sequence of the non-essential TgUPRT gene for genomic integration. To complement the rTgCIA2-mAID-HA line with a constitutively-expressed copy of TgCIA2 in which the putative FeS cluster-binding residue Cys-524 was mutated to Ala, we synthesized the TgCIA2C524A open reading frame and an in-frame 3′ Ty1 tag as a gBlock (S1 Table). We PCR amplified the TgCIA2C524A-Ty1 open reading frame with the primers CIA2 comp fwd and Ty1 rvs, digested the resulting PCR product with BglII and AvrII (excising the TgCIA2C524A open reading frame) and ligated this into the equivalent sites of pUDTTy(TgCIA2WT) generating a plasmid we termed pUDTTy(TgCIA2C524A). We linearized the pUDTTy(TgCIA2WT) and pUDTTy(TgCIA2C524A) vectors with MfeI, transfected into rTgCIA2-mAID-HA parasites, selected on 1 µM pyrimethamine and cloned parasites by limiting dilution. This generated the rTgCIA2-mAID-HA/cTgCIA2WT-Ty1 and rTgCIA2-mAID-HA/cTgCIA2C524A-Ty1 parasite lines. To complement the rTgCIA1-mAID-HA line with constitutively-expressed copies of wild type TgCIA1 (TgCIA1WT), TgCIA1 with the CD1 loop replaced with the CD1 loop from yeast (TgCIA1ScCD1), TgCIA1 with the CD5 loop replaced with the CD5 loop from yeast (TgCIA1ScCD5), TgCIA1 with the CD1 loop replaced with the T. gondii CD5 loop and the native T. gondii CD5 loop replaced with the CD5 loop from yeast (TgCIA1CD5-to-CD1), TgCIA1 with the CD5 loop replaced with the CD5 loop from V. brassicaformis (TgCIA1VbCD5), or TgCIA1 with the CD5 loop replaced with the CD5 loop from S. minutum (TgCIA1SmCD5), we synthesized the open reading frames of the desired proteins as gBlocks (S1 Table). We PCR amplified the TgCIA1 variants with the primers CIA1 comp fwd and Ty1 rvs, digested the resulting PCR product with BglII and AvrII, and ligated this into the equivalent restriction enzyme sites of the pUDTTy(TgCIA2WT) vector. This generated vectors we termed pUDTTy(TgCIA1variant). We linearized the resulting vectors with MfeI, transfected into rTgCIA1-mAID-HA parasites, selected on 1 µM pyrimethamine, and cloned parasites by limiting dilution. This generated the rTgCIA1-mAID-HA/cTgCIA1CD variant-Ty1 cell lines. To complement the rTgCIA1-mAID-HA line with constitutively-expressed copies of TgCIA1 with the CD3 loop replaced with the CD3 loop from yeast (TgCIA1ScCD3), or where residues in the CD5 aromatic amino acid motif were modified to alanines (TgCIA1W526A, TgCIA1Y527A, TgCIA1F532A, or TgCIA1R533A), we undertook a site-directed mutagenesis approach. We initially digested the pUDTTy(TgCIA1WT) vector with SpeI and NsiI, and religated the vector using the compatible sticky ends generated in the digest. This removed a large portion of the vector (the DHFR selectable marker and some of the UPRT flank), and made the resulting TgCIA1-encoding vector smaller and, consequently, easier to use for site-directed mutagenesis. We performed Q5 site-directed mutagenesis with the primers CIA1ScCD3, CIA1W526A, CIA1Y527A, CIA1F532A, or CIA1R533A fwd and rvs as per the manufacturer’s instructions (New England Biolabs). After verifying successful mutagenesis by Sanger sequencing, we excised the regions of the resulting vectors encoding the TgCIA1 variant with BglII and AvrII, and ligated these into the equivalent sites of the pUDTTy(TgCIA2WT) vector. This generated vectors we termed pUDTTy(TgCIA1variant). We linearized the resulting vectors with MfeI, transfected into rTgCIA1-mAID-HA parasites, selected on 1 µM pyrimethamine, and cloned parasites by limiting dilution. This generated the rTgCIA1-mAID-HA/cTgCIA1ScCD3-Ty1 and rTgCIA1-mAID-HA/cTgCIA1aromatic motif mutant-Ty1 cell lines. To constitutively express GFP containing the CD5 loop of TgCIA1 (GFPTgCD5), we synthesized a GFP-TgCD5 gBlock and PCR amplified the GFP-TgCD5 open reading frame with the primers GFP fwd and rvs (S1 Table). We digested the resulting PCR product with BglII and AvrII, and ligated this into the equivalent restriction enzyme sites of the pUDTTy(TgCIA2WT) vector. This generated vectors we termed pUDTTy(GFPTgCD5). We linearized the resulting vector with MfeI, transfected into RH∆ku80-Tir1-FLAG/tdTomato parasites, and selected on 1 µM pyrimethamine. To constitutively-express wild type D. melanogaster CIA1 (DmCIA1WT) or DmCIA1 with the CD5 loop replaced with the CD5 loop from T. gondii (DmCIA1TgCD5), we synthesized both variants as gBlocks, then PCR amplified them using the primers DmCIA1 fwd and Ty1 rvs. We digested the resulting PCR products with BglII and AvrII, and ligated them into the equivalent restriction enzyme sites of the pUDTTy(TgCIA2WT) vector. This generated vectors we termed pUDTTy(DmCIA1WT/TgCD5). We linearized the resulting vectors with MfeI, transfected into rTgCIA1-mAID-HA parasites, selected on 1 µM pyrimethamine, and cloned parasites by limiting dilution. This generated the rTgCIA1-mAID-HA/cDmCIA1WT/TgCD5-Ty1 cell lines. To engineer point mutations in the aromatic amino acid motif of the CD5 loop of the native TgCIA1 locus of T. gondii, we adopted a CRISPR/Cas9-based genome editing approach. We introduced DNA encoding sgRNAs targeting the genome near the region encoding the TgCIA1W526 and TgCIA1Y527 residues or near the region encoding the TgCIA1F532 and TgCIA1R533 residues into the pSAG1::Cas9-U6-UPRT vector using Q5-site directed mutagenesis (New England Biolabs). We performed the Q5 reaction using the gene specific “CIA1-W526/Y527 CRISPR fwd” or “CIA1-F532/R533 CRISPR fwd” primers and a “universal CRISPR rvs” primer (S1 Table). To generate repair templates, we annealed complementary oligonucleotides termed W526A fwd and rvs, Y527A fwd and rvs or R533A fwd and rvs (S1 Table) by mixing 2 nmol of the fwd and rvs oligos then heating to 98°C for 3 min, before allowing the samples to cool to room temperature. We co-transfected the annealed W526A or Y527A oligos with the plasmid encoding the CIA1-W526/Y527-targeting sgRNA, and the annealed R533A oligos with the plasmid encoding the CIA1-F532/R533 sgRNA, into rTgNBP35-cMyc/TgCIA1-HA parasites, then sorted and cloned GFP-positive parasites 2 days post-transfection. We extracted genomic DNA from resulting clones and amplified the region encoding the TgCIA1 aromatic amino acid motif with the primers CIA1 motif seq fwd and rvs (S1 Table), and identified clones harboring successful genetic modifications by Sanger sequencing of the resulting PCR products. Transfections were performed as described previously [50], using 2 mm gap cuvettes and a single 1.5 kV pulse at 25 µF capacitance and 50 Ω resistance using a Bio-Rad Gene Pulser II electroporator. Immunofluorescence assays Confluent HFFs growing on glass coverslips were inoculated with freshly egressed parasites and cultured for ~24 hours at 5% CO2 and 37°C. Infected host cells were then fixed in 3% (v/v) paraformaldehyde in phosphate-buffered saline (PBS; 137 mM NaCl, 2.7 mM KCl, 10 mM Na2HPO4, 1.8 mM KH2PO4, pH 7.4) for 15–20 min and permeabilized in 0.25% (v/v) Triton X-100 (TX-100) in PBS for 10 min. Samples were blocked in 2% or 4% (w/v) bovine serum albumin (BSA) in PBS overnight at 4°C. Samples were probed with either rat anti-HA (1:200 dilution; clone 3F10 Sigma-Aldrich, catalog number 11867423001), mouse anti-cMyc (1:100 or 1:200 dilution; clone 9E10 Santa Cruz Biotechnology, catalog number SC-40) or mouse anti-Ty1 (1:200 dilution; clone BB2, [51]) primary antibodies, together with rabbit anti-TgTom40 (1: 2,000 dilution; [52]) primary antibody. Samples were subsequently probed with donkey anti-rat AlexaFluor 488 (1:500 dilution; Thermo Fisher Scientific, catalog number A21208), goat anti-mouse AlexaFluor 488 (1:250 or 1:500 dilution; Thermo Fisher Scientific, catalog number A11029), or goat anti-mouse AlexaFluor 546 (1:250 dilution; Thermo Fisher Scientific, catalog number A11030) secondary antibodies together with goat anti-rabbit AlexaFluor 546 (1:500 dilution; Thermo Fisher Scientific, catalog number A11035) or goat anti-rabbit AlexaFluor 647 (1:500 dilution; Thermo Fisher Scientific, catalog number A21245) secondary antibodies. Images were acquired on a DeltaVision Elite deconvolution microscope (GE Healthcare) using a Photometrics CoolSNAP HQ2 camera. Brightness and contrast were adjusted linearly using SoftWoRx suit 2.0 software. Images were artificially colored and the fluorescence intensity profiles, measuring pixel intensity along a line of interest, were generated using FIJI software [53]. The fluorescence intensity values were normalized to the maximum value of either channel using the equation x normalized = (x − x minimum)/(x maximum − x minimum) and plotted using GraphPad Prism 9 software. To analyze the degree of signal overlap between two channels within the boundary of a parasite vacuole Pearson’s correlation coefficients were determined. Pearson’s r values were generated using the plugin Coloc2 with Costes regression using FIJI software [53] and plotted using GraphPad Prism 10 software. To test the extent of non-specific antibody labeling, we performed immunofluorescence assays on TATi∆ku80 parasites lacking epitope tags. These data revealed minimal levels of non-specific labeling with the epitope tag antibodies at the antibody concentrations and imaging conditions that we used in the study (S13 Fig). SDS-PAGE, BN-PAGE, and western blotting For SDS-PAGE, parasite proteins were solubilized in LDS sample buffer (Thermo Fisher Scientific) containing 2.5% v/v β-mercaptoethanol to a concentration of 2.5 × 105 parasites/µL. Proteins from 2.5 × 106 parasite equivalents were separated on a pre-cast NuPage 12% Bis-Tris polyacrylamide gel (Thermo Fisher Scientific) and transferred onto a 0.45 µm pore-sized nitrocellulose membrane. For BN-PAGE, parasite proteins were solubilized in BN-PAGE lysis buffer (Thermo Fisher Scientific NativePAGE sample buffer containing 1% v/v TX-100, 2 mM EDTA, and 1× cOmplete protease inhibitor cocktail, Merck, catalog number 11873580001) to a concentration of 2.5 × 105 parasites/µl. Proteins from 2.5 × 106 parasite equivalents were separated on a pre-cast NativePAGE 4%–16% Bis-Tris polyacrylamide gel (Thermo Fisher Scientific), transferred onto a polyvinylidene difluoride (PVDF) membrane, fixed in 10% (v/v) acetic acid, and de-stained in methanol. Nitrocellulose and PVDF membranes were blocked overnight in “Blotto,” a Tris-buffered saline (TBS; 137 mM NaCl, 2.7 mM KCl, 25 mM Tris-HCl, pH 7.4) solution containing 4% (w/v) skim milk powder. Membranes were subsequently probed with rat anti-HA (1:200–1:400 dilution; clone 3F10 Sigma-Aldrich, catalog number 11867423001), mouse anti-cMyc (1:100 or 1:200 dilution; clone 9E10 Santa Cruz Biotechnology, catalog number SC-40), mouse anti-Ty1 (1:200 or 1:400 dilution; clone BB2 [51]), mouse-anti-puromycin (1:3,000 dilution; clone 12D10, Merck catalog number MABE343), or rabbit anti-TgTom40 (1:2,000 dilution [52]) primary antibodies diluted in Blotto. Secondary antibodies used were horseradish peroxidase-conjugated goat anti-rat IgG (1:5,000 or 1:10,000 dilution; Abcam, catalog number ab97057), goat anti-mouse IgG (1:2,500, 1:5,000, or 1:10,000 dilution; Abcam, catalog number ab6789), or goat anti-rabbit IgG (1:5,000 or 1:10,000 dilution; Abcam, catalog number ab97051). Antibody-labeled membranes were incubated in enhanced chemiluminescence solution (0.04% w/v luminol, 0.007% w/v coumaric acid, 0.01% H2O2, 100 mM Tris pH 9.35) and imaged using a ChemiDoc MP imaging system (Bio-Rad). In western blotting experiments comparing the expression of different proteins in a particular parasite line (e.g., where band intensities were quantified), the same membrane was probed multiple times with different antibodies, with bound antibodies removing by stripping between probings. For stripping, membranes were treated twice in stripping buffer (200 mM glycine, 3.5 mM SDS, 1% v/v Tween-20, pH 2) for 15 min, twice in PBS for 10 min, twice in TBS containing 0.05% (v/v) Tween-20 for 5 min, and then blocked in Blotto for at least 1 h. Where relevant, band intensities were quantified using ImageJ (1.53k; [54]). Puromycin incorporation assays Puromycin incorporation assays were adapted from a previously described methodology [21]. Parasites were cultured in the absence or presence of IAA for 24 h. Intracellular parasites were mechanically egressed from host cells by passage through a 26-gauge needle, and host cell debris removed by filtering samples through a 5 µm PVDF filter. Parasites were pelleted by centrifugation at 1,500g and resuspended in pre-warmed complete culture medium (±IAA) to a final concentration of approximately 1.5 × 107 parasites/mL. Puromycin (Merck, catalog number P8833) was added to the parasite samples to a final concentration of 10 µg/mL, and parasites were incubated in a humidified 37°C incubator at 5% CO2 for 15 min with loosened lids to allow gas exchange. Puromycin-labeled parasites were pelleted by centrifugation at 12,000g for 1 min, washed in room temperature PBS, re-pelleted by centrifugation at 12,000g for 1 min, and resuspended in reducing LDS sample buffer (Thermo Fisher Scientific) to a final concentration of 2.5 × 105 parasites/µl. Proteins were separated by SDS-PAGE and detected by western blotting as described above. Co-immunoprecipitations Immunoprecipitations were performed as described previously [49,52]. Briefly, parasites were solubilized on ice for at least 30 min in a lysis buffer containing 1% (v/v) TX-100, 150 mM NaCl, 2 mM EDTA, 1× cOmplete protease inhibitor cocktail, and 50 mM Tris-HCl (pH 7.4). Insoluble proteins were removed by centrifugation at 21,000g and the clarified supernatant was incubated overnight with anti-HA affinity matrix (Sigma-Aldrich, catalog number 11815016001) or Myc-Trap anti-cMyc agarose beads (Chromotek, catalog reference yta) at 4°C on a spinning wheel. Unbound proteins were precipitated with trichloroacetic acid. Beads containing bound proteins were washed in a wash buffer containing 0.01% or 1% v/v TX-100, 150 mM NaCl, 2 mM EDTA, 50 mM Tris-HCl pH 7.4, and proteins were eluted from the beads using LDS sample buffer (Thermo Fisher Scientific) containing 2.5% v/v β-mercaptoethanol. The unbound and bound fractions were resuspended in reducing LDS sample buffer, with protein samples subsequently separated by SDS-PAGE and detected by western blotting as described above. Fluorescence proliferation assays Fluorescence proliferation assays were performed as described previously [55]. Optical bottom 96-well plates (Corning or Greiner) containing confluent HFFs were inoculated with 2,000 tdTomato-expressing parasites, and cultured in the absence or presence of IAA in phenol-red-free DMEM. Each condition was performed in triplicate. Well fluorescence was analyzed using a FLUOstar Optima plate reader (BMG Labtech) once daily, or twice daily during the expected exponential growth phase, over a 5–6 day period. Fluorescence values (with average background fluorescence subtracted) were expressed as a percentage of the average final fluorescence measurement in the -IAA condition for each cell line and were plotted using GraphPad Prism 9 or 10, with a sigmoidal curve fitted to the data using the equation: . Plaque assays Confluent HFFs in 25 cm2 tissue culture flasks were inoculated with 500 parasites and cultured in the absence or presence of IAA for 6 to 8 days at 37°C and 5% CO2. Infected HFFs were stained with Gram’s crystal violet solution (Merck, catalog number 94448) for 1–3 hours and rinsed with PBS. Flasks were air-dried before imaging with a CanoScan 9000F scanner (Canon). Multiple sequence alignment and structural analyses CIA1 homologs were identified by screening the CIA1 ortholog group (OG6_102166) on OrthoMCL (https://orthomcl.org/orthomcl; [56]). If a CIA homolog sequence was not found in an organism-of-interest on OrthoMCL, additional BLAST searches were conducted on NCBI (https://blast.ncbi.nlm.nih.gov/Blast.cgi; [57]) or UniProt (https://www.uniprot.org; [58]). We performed reciprocal BLAST searches with each candidate hit on ToxoDB (www.toxodb.org; [17]) and excluded sequences where TgCIA1 was not recognized as the top hit. Clustal Omega (https://www.ebi.ac.uk/Tools/msa/clustalo/; [59]) was used to generate a multiple sequence alignment. Sequences that exhibited significant misalignment were excluded. The alignments were plotted using pyBoxshade (https://github.com/mdbaron42/pyBoxshade). Multiple sequence alignments of TgCIA2 and TgNar1 with homologs from other organisms were also performed using Clustal Omega, and plotted using pyBoxshade, with protein sequences acquired from VEuPathDB (www.veupathdb.org; [60]), ToxoDB and UniProt. The CIA2 alignment was manually adjusted using Jalview (version 2.11.4.1; [61]). The structure of S. cerevisiae CIA1 was generated previously [28] and was obtained from the Protein Data Bank (PDB: 2HES). The predicted structure of TgCIA1 was obtained from AlphaFold2 (UniProt ID A0A125YRQ0; [62]). The structural analyses of CIA1 homologs were viewed and colored using EzMol software (http://www.sbg.bio.ic.ac.uk/ezmol/; [63]). Data analysis and availability All statistical tests were performed in GraphPad Prism 9 or 10 software and are described in the figure legends. Most fluorescence proliferation assays, plaque assays and western blots depicted in the figures are representative of at least three independent experiments (as indicated in the figure legends). These replicate proliferation, plaque assay and western blotting data, as well as additional representative immunofluorescence images from the microscopy experiments, are included in S2 Data. Source images for all western blots, PCRs, and plaque assays are included in a S1 Raw Images, and the numerical values (both raw and normalized) for the graphed data are included in the S1 Data. Parasite culturing Tachyzoite-stage T. gondii parasites were cultured in human foreskin fibroblasts (HFFs) using Dulbecco’s Modified Eagle Medium (DMEM) supplemented with 1% (v/v) fetal bovine serum, 2 g/L NaHCO3, 0.2 mM L-glutamine, 50 units/mL penicillin, 50 µg/mL streptomycin, 10 µg/mL gentamicin, and 0.25 µg/mL amphotericin b. Cultures were kept in humidified incubators at 37°C and 5% CO2. Where appropriate, 0.1 mM IAA (Merck catalog number I2886), 0.5 µg/mL ATc (Merck catalog number 37919), or an equivalent volume of 100% ethanol (vehicle control) was added. Generation of genetically modified parasites We used CRISPR/Cas9 genome editing to introduce 3× HA, 3′ mini-auxin-inducible degron-3× HA (mAID-HA), 3′ mAID-3× cMyc (mAID-cMyc), 5′ HA-mAID, 3′ Ty1, or 3′ spaghetti monster fluorescent protein (smFP)-cMyc epitope tags into the 3′ or 5′ regions of the open reading frames of TgTah18 (www.ToxoDB.org identifier TGGT1_249320; [17]), TgDre2 (TGGT1_216900), TgNar1 (TGGT1_242580) TgCIA1 (TGGT1_313280), TgCIA2 (TGGT1_306590), TgMMS19 (TGGT1_222230), or TgABCE1 (TGGT1_216790). To enable, these modifications, we introduced DNA encoding single guide (sg)RNAs targeting the 3′ or 5′ regions of these genes into the vector pSAG1::Cas9-U6-UPRT (Addgene plasmid 54467, [48]) using Q5-site directed mutagenesis (New England Biolabs). We performed the Q5 reaction using the gene-specific “3′ or 5′ CRISPR fwd” primer and a “universal CRISPR rvs” primer (S1 Table). We amplified a donor DNA sequence encoding the epitope tag plus 50 bp flanks of the target gene using gene specific “tag fwd” and “tag rvs” primers and HA, C-terminal mAID-HA, N-terminal HA-mAID, mAID-cMyc, cMyc, Ty1, or smFP-cMyc gBlocks (Integrated DNA Technologies, IDT) as templates (S1 Table). We transfected the sgRNA-expressing plasmids and HA-containing donor DNAs targeting the 3′ region of the TgTah18, TgDre2, TgCIA1, TgCIA2, or TgMMS19 into an ATc regulatable TgNBP35-cMyc strain described previously [15]. This generated TgTah18-HA, TgDre2-HA, TgCIA1-HA, TgCIA2-HA, and TgMMS19-HA expressing parasites. We transfected the sgRNA-expressing plasmids targeting TgTah18, TgDre2, TgNar1, TgCIA1, TgCIA2, or TgMMS19 and C-terminal mAID-HA, N-terminal HA-mAID, or mAID-cMyc-containing donor DNAs into RH∆ku80-Tir1-FLAG parasites expressing tdTomato red fluorescent protein (described previously, [18]). This generated the IAA-regulatable rTgTah18-mAID-HA, rTgDre2-mAID-HA, rHA-mAID-TgNar1, rTgCIA1-mAID-HA, rTgCIA2-mAID-HA, rTgMMS19-mAID-HA, and rTgCIA1-mAID-cMyc parasite lines. We transfected the sgRNA-expressing plasmid targeting TgABCE1 and the corresponding cMyc donor DNA into the rTgCIA1-mAID-HA and rTgDre2-mAID-HA parasite lines to generated the rTgCIA1-mAID-HA/TgABCE1-cMyc and rTgDre2-mAID-HA/TgABCE1-cMyc lines, respectively. We transfected the sgRNA-expressing plasmids targeting TgCIA2 and TgMMS19 and the corresponding HA donor DNA into the rTgCIA1-mAID-cMyc parasite line to generate the rTgCIA1-mAID-cMyc/TgCIA1-HA and rTgCIA1-mAID-cMyc/TgMMS19-HA lines. We transfected the sgRNA-expressing plasmid targeting TgCIA1 and the corresponding smFP-cMyc donor DNA into the rTgCIA2-mAID-HA, rTgMMS19-mAID-HA, rTgTah18-mAID-HA, and rHA-mAID-TgNar1 parasite lines to generate the rTgCIA2-mAID-HA/TgCIA1-smFP-cMyc, rTgMMS19-mAID-HA/TgCIA1-smFP-cMyc, rTgTah18-mAID-HA/TgCIA1-smFP-cMyc, and rHA-mAID-TgNar1/TgCIA1-smFP-cMyc lines, respectively. Two days after each of the transfections described above, GFP-positive parasites (which express Cas9-GFP encoded on the modified pSAG1::Cas9-U6 vector) were sorted into wells of a 96-well plate using a FACSMelody cell sorter (BD Biosciences). After approximately one week, we identified wells containing clonal parasites (i.e., wells with single plaques), extracted genomic DNA from parasites in these wells, and performed PCR screening analyses to identify genetically modified parasites using gene-specific “3′ or 5′ scrn fwd” and “3′ or 5′ scrn rvs” primers (S1 Table). To complement the rTgCIA2-mAID-HA line with a constitutively-expressed copy of wild type TgCIA2, we synthesized the TgCIA2WT open reading frame and an in-frame 3′ Ty1 epitope tag as a gBlock (S1 Table). We PCR amplified the TgCIA2WT-Ty1 open reading frame with the primers CIA2 comp fwd and Ty1 rvs, digested the resulting PCR product with BglII and XmaI and ligated this into the equivalent restriction enzyme cut sites of pUDTG [49]. The resulting vector, which we termed pUDTTy(TgCIA2WT), expresses TgCIA2WT-Ty1 from the constitutive α-tubulin promoter, in a vector that contains a pyrimethamine-resistant TgDHFR selectable marker and a flanking sequence of the non-essential TgUPRT gene for genomic integration. To complement the rTgCIA2-mAID-HA line with a constitutively-expressed copy of TgCIA2 in which the putative FeS cluster-binding residue Cys-524 was mutated to Ala, we synthesized the TgCIA2C524A open reading frame and an in-frame 3′ Ty1 tag as a gBlock (S1 Table). We PCR amplified the TgCIA2C524A-Ty1 open reading frame with the primers CIA2 comp fwd and Ty1 rvs, digested the resulting PCR product with BglII and AvrII (excising the TgCIA2C524A open reading frame) and ligated this into the equivalent sites of pUDTTy(TgCIA2WT) generating a plasmid we termed pUDTTy(TgCIA2C524A). We linearized the pUDTTy(TgCIA2WT) and pUDTTy(TgCIA2C524A) vectors with MfeI, transfected into rTgCIA2-mAID-HA parasites, selected on 1 µM pyrimethamine and cloned parasites by limiting dilution. This generated the rTgCIA2-mAID-HA/cTgCIA2WT-Ty1 and rTgCIA2-mAID-HA/cTgCIA2C524A-Ty1 parasite lines. To complement the rTgCIA1-mAID-HA line with constitutively-expressed copies of wild type TgCIA1 (TgCIA1WT), TgCIA1 with the CD1 loop replaced with the CD1 loop from yeast (TgCIA1ScCD1), TgCIA1 with the CD5 loop replaced with the CD5 loop from yeast (TgCIA1ScCD5), TgCIA1 with the CD1 loop replaced with the T. gondii CD5 loop and the native T. gondii CD5 loop replaced with the CD5 loop from yeast (TgCIA1CD5-to-CD1), TgCIA1 with the CD5 loop replaced with the CD5 loop from V. brassicaformis (TgCIA1VbCD5), or TgCIA1 with the CD5 loop replaced with the CD5 loop from S. minutum (TgCIA1SmCD5), we synthesized the open reading frames of the desired proteins as gBlocks (S1 Table). We PCR amplified the TgCIA1 variants with the primers CIA1 comp fwd and Ty1 rvs, digested the resulting PCR product with BglII and AvrII, and ligated this into the equivalent restriction enzyme sites of the pUDTTy(TgCIA2WT) vector. This generated vectors we termed pUDTTy(TgCIA1variant). We linearized the resulting vectors with MfeI, transfected into rTgCIA1-mAID-HA parasites, selected on 1 µM pyrimethamine, and cloned parasites by limiting dilution. This generated the rTgCIA1-mAID-HA/cTgCIA1CD variant-Ty1 cell lines. To complement the rTgCIA1-mAID-HA line with constitutively-expressed copies of TgCIA1 with the CD3 loop replaced with the CD3 loop from yeast (TgCIA1ScCD3), or where residues in the CD5 aromatic amino acid motif were modified to alanines (TgCIA1W526A, TgCIA1Y527A, TgCIA1F532A, or TgCIA1R533A), we undertook a site-directed mutagenesis approach. We initially digested the pUDTTy(TgCIA1WT) vector with SpeI and NsiI, and religated the vector using the compatible sticky ends generated in the digest. This removed a large portion of the vector (the DHFR selectable marker and some of the UPRT flank), and made the resulting TgCIA1-encoding vector smaller and, consequently, easier to use for site-directed mutagenesis. We performed Q5 site-directed mutagenesis with the primers CIA1ScCD3, CIA1W526A, CIA1Y527A, CIA1F532A, or CIA1R533A fwd and rvs as per the manufacturer’s instructions (New England Biolabs). After verifying successful mutagenesis by Sanger sequencing, we excised the regions of the resulting vectors encoding the TgCIA1 variant with BglII and AvrII, and ligated these into the equivalent sites of the pUDTTy(TgCIA2WT) vector. This generated vectors we termed pUDTTy(TgCIA1variant). We linearized the resulting vectors with MfeI, transfected into rTgCIA1-mAID-HA parasites, selected on 1 µM pyrimethamine, and cloned parasites by limiting dilution. This generated the rTgCIA1-mAID-HA/cTgCIA1ScCD3-Ty1 and rTgCIA1-mAID-HA/cTgCIA1aromatic motif mutant-Ty1 cell lines. To constitutively express GFP containing the CD5 loop of TgCIA1 (GFPTgCD5), we synthesized a GFP-TgCD5 gBlock and PCR amplified the GFP-TgCD5 open reading frame with the primers GFP fwd and rvs (S1 Table). We digested the resulting PCR product with BglII and AvrII, and ligated this into the equivalent restriction enzyme sites of the pUDTTy(TgCIA2WT) vector. This generated vectors we termed pUDTTy(GFPTgCD5). We linearized the resulting vector with MfeI, transfected into RH∆ku80-Tir1-FLAG/tdTomato parasites, and selected on 1 µM pyrimethamine. To constitutively-express wild type D. melanogaster CIA1 (DmCIA1WT) or DmCIA1 with the CD5 loop replaced with the CD5 loop from T. gondii (DmCIA1TgCD5), we synthesized both variants as gBlocks, then PCR amplified them using the primers DmCIA1 fwd and Ty1 rvs. We digested the resulting PCR products with BglII and AvrII, and ligated them into the equivalent restriction enzyme sites of the pUDTTy(TgCIA2WT) vector. This generated vectors we termed pUDTTy(DmCIA1WT/TgCD5). We linearized the resulting vectors with MfeI, transfected into rTgCIA1-mAID-HA parasites, selected on 1 µM pyrimethamine, and cloned parasites by limiting dilution. This generated the rTgCIA1-mAID-HA/cDmCIA1WT/TgCD5-Ty1 cell lines. To engineer point mutations in the aromatic amino acid motif of the CD5 loop of the native TgCIA1 locus of T. gondii, we adopted a CRISPR/Cas9-based genome editing approach. We introduced DNA encoding sgRNAs targeting the genome near the region encoding the TgCIA1W526 and TgCIA1Y527 residues or near the region encoding the TgCIA1F532 and TgCIA1R533 residues into the pSAG1::Cas9-U6-UPRT vector using Q5-site directed mutagenesis (New England Biolabs). We performed the Q5 reaction using the gene specific “CIA1-W526/Y527 CRISPR fwd” or “CIA1-F532/R533 CRISPR fwd” primers and a “universal CRISPR rvs” primer (S1 Table). To generate repair templates, we annealed complementary oligonucleotides termed W526A fwd and rvs, Y527A fwd and rvs or R533A fwd and rvs (S1 Table) by mixing 2 nmol of the fwd and rvs oligos then heating to 98°C for 3 min, before allowing the samples to cool to room temperature. We co-transfected the annealed W526A or Y527A oligos with the plasmid encoding the CIA1-W526/Y527-targeting sgRNA, and the annealed R533A oligos with the plasmid encoding the CIA1-F532/R533 sgRNA, into rTgNBP35-cMyc/TgCIA1-HA parasites, then sorted and cloned GFP-positive parasites 2 days post-transfection. We extracted genomic DNA from resulting clones and amplified the region encoding the TgCIA1 aromatic amino acid motif with the primers CIA1 motif seq fwd and rvs (S1 Table), and identified clones harboring successful genetic modifications by Sanger sequencing of the resulting PCR products. Transfections were performed as described previously [50], using 2 mm gap cuvettes and a single 1.5 kV pulse at 25 µF capacitance and 50 Ω resistance using a Bio-Rad Gene Pulser II electroporator. Immunofluorescence assays Confluent HFFs growing on glass coverslips were inoculated with freshly egressed parasites and cultured for ~24 hours at 5% CO2 and 37°C. Infected host cells were then fixed in 3% (v/v) paraformaldehyde in phosphate-buffered saline (PBS; 137 mM NaCl, 2.7 mM KCl, 10 mM Na2HPO4, 1.8 mM KH2PO4, pH 7.4) for 15–20 min and permeabilized in 0.25% (v/v) Triton X-100 (TX-100) in PBS for 10 min. Samples were blocked in 2% or 4% (w/v) bovine serum albumin (BSA) in PBS overnight at 4°C. Samples were probed with either rat anti-HA (1:200 dilution; clone 3F10 Sigma-Aldrich, catalog number 11867423001), mouse anti-cMyc (1:100 or 1:200 dilution; clone 9E10 Santa Cruz Biotechnology, catalog number SC-40) or mouse anti-Ty1 (1:200 dilution; clone BB2, [51]) primary antibodies, together with rabbit anti-TgTom40 (1: 2,000 dilution; [52]) primary antibody. Samples were subsequently probed with donkey anti-rat AlexaFluor 488 (1:500 dilution; Thermo Fisher Scientific, catalog number A21208), goat anti-mouse AlexaFluor 488 (1:250 or 1:500 dilution; Thermo Fisher Scientific, catalog number A11029), or goat anti-mouse AlexaFluor 546 (1:250 dilution; Thermo Fisher Scientific, catalog number A11030) secondary antibodies together with goat anti-rabbit AlexaFluor 546 (1:500 dilution; Thermo Fisher Scientific, catalog number A11035) or goat anti-rabbit AlexaFluor 647 (1:500 dilution; Thermo Fisher Scientific, catalog number A21245) secondary antibodies. Images were acquired on a DeltaVision Elite deconvolution microscope (GE Healthcare) using a Photometrics CoolSNAP HQ2 camera. Brightness and contrast were adjusted linearly using SoftWoRx suit 2.0 software. Images were artificially colored and the fluorescence intensity profiles, measuring pixel intensity along a line of interest, were generated using FIJI software [53]. The fluorescence intensity values were normalized to the maximum value of either channel using the equation x normalized = (x − x minimum)/(x maximum − x minimum) and plotted using GraphPad Prism 9 software. To analyze the degree of signal overlap between two channels within the boundary of a parasite vacuole Pearson’s correlation coefficients were determined. Pearson’s r values were generated using the plugin Coloc2 with Costes regression using FIJI software [53] and plotted using GraphPad Prism 10 software. To test the extent of non-specific antibody labeling, we performed immunofluorescence assays on TATi∆ku80 parasites lacking epitope tags. These data revealed minimal levels of non-specific labeling with the epitope tag antibodies at the antibody concentrations and imaging conditions that we used in the study (S13 Fig). SDS-PAGE, BN-PAGE, and western blotting For SDS-PAGE, parasite proteins were solubilized in LDS sample buffer (Thermo Fisher Scientific) containing 2.5% v/v β-mercaptoethanol to a concentration of 2.5 × 105 parasites/µL. Proteins from 2.5 × 106 parasite equivalents were separated on a pre-cast NuPage 12% Bis-Tris polyacrylamide gel (Thermo Fisher Scientific) and transferred onto a 0.45 µm pore-sized nitrocellulose membrane. For BN-PAGE, parasite proteins were solubilized in BN-PAGE lysis buffer (Thermo Fisher Scientific NativePAGE sample buffer containing 1% v/v TX-100, 2 mM EDTA, and 1× cOmplete protease inhibitor cocktail, Merck, catalog number 11873580001) to a concentration of 2.5 × 105 parasites/µl. Proteins from 2.5 × 106 parasite equivalents were separated on a pre-cast NativePAGE 4%–16% Bis-Tris polyacrylamide gel (Thermo Fisher Scientific), transferred onto a polyvinylidene difluoride (PVDF) membrane, fixed in 10% (v/v) acetic acid, and de-stained in methanol. Nitrocellulose and PVDF membranes were blocked overnight in “Blotto,” a Tris-buffered saline (TBS; 137 mM NaCl, 2.7 mM KCl, 25 mM Tris-HCl, pH 7.4) solution containing 4% (w/v) skim milk powder. Membranes were subsequently probed with rat anti-HA (1:200–1:400 dilution; clone 3F10 Sigma-Aldrich, catalog number 11867423001), mouse anti-cMyc (1:100 or 1:200 dilution; clone 9E10 Santa Cruz Biotechnology, catalog number SC-40), mouse anti-Ty1 (1:200 or 1:400 dilution; clone BB2 [51]), mouse-anti-puromycin (1:3,000 dilution; clone 12D10, Merck catalog number MABE343), or rabbit anti-TgTom40 (1:2,000 dilution [52]) primary antibodies diluted in Blotto. Secondary antibodies used were horseradish peroxidase-conjugated goat anti-rat IgG (1:5,000 or 1:10,000 dilution; Abcam, catalog number ab97057), goat anti-mouse IgG (1:2,500, 1:5,000, or 1:10,000 dilution; Abcam, catalog number ab6789), or goat anti-rabbit IgG (1:5,000 or 1:10,000 dilution; Abcam, catalog number ab97051). Antibody-labeled membranes were incubated in enhanced chemiluminescence solution (0.04% w/v luminol, 0.007% w/v coumaric acid, 0.01% H2O2, 100 mM Tris pH 9.35) and imaged using a ChemiDoc MP imaging system (Bio-Rad). In western blotting experiments comparing the expression of different proteins in a particular parasite line (e.g., where band intensities were quantified), the same membrane was probed multiple times with different antibodies, with bound antibodies removing by stripping between probings. For stripping, membranes were treated twice in stripping buffer (200 mM glycine, 3.5 mM SDS, 1% v/v Tween-20, pH 2) for 15 min, twice in PBS for 10 min, twice in TBS containing 0.05% (v/v) Tween-20 for 5 min, and then blocked in Blotto for at least 1 h. Where relevant, band intensities were quantified using ImageJ (1.53k; [54]). Puromycin incorporation assays Puromycin incorporation assays were adapted from a previously described methodology [21]. Parasites were cultured in the absence or presence of IAA for 24 h. Intracellular parasites were mechanically egressed from host cells by passage through a 26-gauge needle, and host cell debris removed by filtering samples through a 5 µm PVDF filter. Parasites were pelleted by centrifugation at 1,500g and resuspended in pre-warmed complete culture medium (±IAA) to a final concentration of approximately 1.5 × 107 parasites/mL. Puromycin (Merck, catalog number P8833) was added to the parasite samples to a final concentration of 10 µg/mL, and parasites were incubated in a humidified 37°C incubator at 5% CO2 for 15 min with loosened lids to allow gas exchange. Puromycin-labeled parasites were pelleted by centrifugation at 12,000g for 1 min, washed in room temperature PBS, re-pelleted by centrifugation at 12,000g for 1 min, and resuspended in reducing LDS sample buffer (Thermo Fisher Scientific) to a final concentration of 2.5 × 105 parasites/µl. Proteins were separated by SDS-PAGE and detected by western blotting as described above. Co-immunoprecipitations Immunoprecipitations were performed as described previously [49,52]. Briefly, parasites were solubilized on ice for at least 30 min in a lysis buffer containing 1% (v/v) TX-100, 150 mM NaCl, 2 mM EDTA, 1× cOmplete protease inhibitor cocktail, and 50 mM Tris-HCl (pH 7.4). Insoluble proteins were removed by centrifugation at 21,000g and the clarified supernatant was incubated overnight with anti-HA affinity matrix (Sigma-Aldrich, catalog number 11815016001) or Myc-Trap anti-cMyc agarose beads (Chromotek, catalog reference yta) at 4°C on a spinning wheel. Unbound proteins were precipitated with trichloroacetic acid. Beads containing bound proteins were washed in a wash buffer containing 0.01% or 1% v/v TX-100, 150 mM NaCl, 2 mM EDTA, 50 mM Tris-HCl pH 7.4, and proteins were eluted from the beads using LDS sample buffer (Thermo Fisher Scientific) containing 2.5% v/v β-mercaptoethanol. The unbound and bound fractions were resuspended in reducing LDS sample buffer, with protein samples subsequently separated by SDS-PAGE and detected by western blotting as described above. Fluorescence proliferation assays Fluorescence proliferation assays were performed as described previously [55]. Optical bottom 96-well plates (Corning or Greiner) containing confluent HFFs were inoculated with 2,000 tdTomato-expressing parasites, and cultured in the absence or presence of IAA in phenol-red-free DMEM. Each condition was performed in triplicate. Well fluorescence was analyzed using a FLUOstar Optima plate reader (BMG Labtech) once daily, or twice daily during the expected exponential growth phase, over a 5–6 day period. Fluorescence values (with average background fluorescence subtracted) were expressed as a percentage of the average final fluorescence measurement in the -IAA condition for each cell line and were plotted using GraphPad Prism 9 or 10, with a sigmoidal curve fitted to the data using the equation: . Plaque assays Confluent HFFs in 25 cm2 tissue culture flasks were inoculated with 500 parasites and cultured in the absence or presence of IAA for 6 to 8 days at 37°C and 5% CO2. Infected HFFs were stained with Gram’s crystal violet solution (Merck, catalog number 94448) for 1–3 hours and rinsed with PBS. Flasks were air-dried before imaging with a CanoScan 9000F scanner (Canon). Multiple sequence alignment and structural analyses CIA1 homologs were identified by screening the CIA1 ortholog group (OG6_102166) on OrthoMCL (https://orthomcl.org/orthomcl; [56]). If a CIA homolog sequence was not found in an organism-of-interest on OrthoMCL, additional BLAST searches were conducted on NCBI (https://blast.ncbi.nlm.nih.gov/Blast.cgi; [57]) or UniProt (https://www.uniprot.org; [58]). We performed reciprocal BLAST searches with each candidate hit on ToxoDB (www.toxodb.org; [17]) and excluded sequences where TgCIA1 was not recognized as the top hit. Clustal Omega (https://www.ebi.ac.uk/Tools/msa/clustalo/; [59]) was used to generate a multiple sequence alignment. Sequences that exhibited significant misalignment were excluded. The alignments were plotted using pyBoxshade (https://github.com/mdbaron42/pyBoxshade). Multiple sequence alignments of TgCIA2 and TgNar1 with homologs from other organisms were also performed using Clustal Omega, and plotted using pyBoxshade, with protein sequences acquired from VEuPathDB (www.veupathdb.org; [60]), ToxoDB and UniProt. The CIA2 alignment was manually adjusted using Jalview (version 2.11.4.1; [61]). The structure of S. cerevisiae CIA1 was generated previously [28] and was obtained from the Protein Data Bank (PDB: 2HES). The predicted structure of TgCIA1 was obtained from AlphaFold2 (UniProt ID A0A125YRQ0; [62]). The structural analyses of CIA1 homologs were viewed and colored using EzMol software (http://www.sbg.bio.ic.ac.uk/ezmol/; [63]). Data analysis and availability All statistical tests were performed in GraphPad Prism 9 or 10 software and are described in the figure legends. Most fluorescence proliferation assays, plaque assays and western blots depicted in the figures are representative of at least three independent experiments (as indicated in the figure legends). These replicate proliferation, plaque assay and western blotting data, as well as additional representative immunofluorescence images from the microscopy experiments, are included in S2 Data. Source images for all western blots, PCRs, and plaque assays are included in a S1 Raw Images, and the numerical values (both raw and normalized) for the graphed data are included in the S1 Data. Supporting information S1 Fig. Epitope tagging of candidate CIA pathway proteins in Toxoplasma gondii. (A–E) 3× hemagglutinin epitope tags (HA) were integrated into the 3′ regions of the open reading frames of the genes encoding (A) TgTah18, (B) TgDre2, (C) TgCIA1, (D) TgCIA2, or (E) TgMMS19 in ATc-regulatable rTgNBP35-cMyc parasites. (F) A 3× hemagglutinin-mini-auxin inducible degron (HA-mAID) epitope tag was integrated into the 5′ region of the TgNar1 open reading frame in RH∆ku80/Tir1-FLAG/tdTomato parasites, generating the rHA-mAID-TgNar1 parasite line. A schematic depicting the target locus before and after modification, the approximate position of the forward and reverse primers used in the PCR analysis, and the expected sizes of the PCR products in the native and modified genomic loci, are shown at the top of each panel. The PCR screens testing for genetic modifications are shown at the bottom of each panel. PCRs were performed using forward and reverse primers specific to the target site of each gene, and using genomic DNA extracted from clonal parasite lines. Genomic DNA from a wild type (WT) parasite line was used as a control for the expected size of the native locus in each screen. https://doi.org/10.1371/journal.pbio.3003520.s001 (TIF) S2 Fig. Generating IAA-regulatable strains of candidate CIA pathway proteins in Toxoplasma gondii. (A–E) Mini-auxin inducible degron-3× hemagglutinin (mAID-HA) epitope tags were integrated into the 3′ regions of open reading frames of the genes encoding (A) TgTah18, (B) TgDre2, (C) TgCIA1, (D) TgCIA2, or (E) TgMMS19 in RH∆ku80/Tir1-FLAG/tdTomato parasites generating the rTgTah18-mAID-HA, rTgDre2-mAID-HA, rTgCIA1-mAID-HA, rTgCIA2-mAID-HA, and rTgMMS19-mAID-HA parasite lines. A schematic depicting the target locus before and after modification, the approximate position of the forward and reverse primers used in the PCR analysis, and the expected sizes of the PCR products in the native and modified genomic loci, are shown at the top of each panel. The PCR screens testing for genetic modifications are shown at the bottom of each panel. PCRs were performed using forward and reverse primers specific to the target site of each gene, and using genomic DNA extracted from clonal parasite lines. Genomic DNA from a WT parasite line was used as a control for the expected size of the native locus in each screen. https://doi.org/10.1371/journal.pbio.3003520.s002 (TIF) S3 Fig. Epitope tagging TgABCE1 in the rTgCIA1-mAID-HA and rTgDre2-mAID-HA lines. (A) A schematic depicting the TgABCE1 genomic locus before and after introduction of a 3× cMyc epitope tag into the 3′ region of the open reading frame of the gene, the approximate position of the forward and reverse primers used in the PCR analyses, and the expected sizes of the PCR products in the native and modified TgABCE1 loci. (B, C) PCR analyses using the forward and reverse primers and template genomic DNA extracted from (B) rTgCIA1-mAID-HA/TgABCE1-cMyc parasite clones and (C) a rTgDre2-mAID-HA/TgABCE1-cMyc parasite clone. Genomic DNA from a WT parasite line was used as a control for the expected size of the native locus. https://doi.org/10.1371/journal.pbio.3003520.s003 (TIF) S4 Fig. Depletion of most candidate CIA pathway proteins leads to a decrease in protein translation in Toxoplasma gondii parasites. (A) Western blots measuring the incorporation of puromycin into proteins from rTgTah18-mAID-HA, rTgDre2-mAID-HA, rHA-mAID-TgNar1, rTgCIA2-mAID-HA, and rTgMMS19-mAID-HA parasites cultured in the absence or presence of IAA for 24 h, probed with anti-puromycin, anti-HA or anti-TgTom40 antibodies. The membrane was also stained following transfer with the protein-binding dye Ponceau S. (B) Relative abundance of puromycin incorporation into each parasite line was determined as a percentage of the -IAA control, with abundances normalized using the TgTom40 loading control. Data points represent the mean ± SD of three independent experiments. Data were analyzed using a one-way ANOVA followed by Tukey’s multiple comparisons test with relevant p values shown. (C) Western blots of proteins extracted from rTgDre2-mAID-HA/TgABCE1-cMyc parasites cultured for 0, 24, or 48 h in IAA and separated by SDS-PAGE. Samples were probed with anti-HA, anti-cMyc, and anti-TgTom40 antibodies. The relative abundance of the TgABCE1-cMyc protein in the western blot was determined as a percentage of the 0 h control, with abundances normalized using the TgTom40 loading control. Data points represent the mean ± SD of three independent experiments. Data were analyzed using a one-way ANOVA followed by Tukey’s multiple comparisons test, with p values shown. The numerical data underlying this Figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003520.s004 (TIF) S5 Fig. Multiple sequence alignment of the TgCIA2 protein with homologs from other eukaryotes. A multiple sequence alignment of the TgCIA2 protein with homologs from the yeast Saccharomyces cerevisiae (ScCIA2; UniProt accession number P38829), Homo sapiens (HsCIA2; UniProt Q9Y3D0), and the fruit fly Drosophila melanogaster (DmCIA2; UniProt Q9VTC4). The reactive cysteine residue of CIA2 that is proposed to function in FeS cluster binding is highlighted by a red box. https://doi.org/10.1371/journal.pbio.3003520.s005 (TIF) S6 Fig. Characterizing the protein composition of the CIA Targeting Complex (CTC) in Toxoplasma gondii parasites. (A) Western blot of proteins extracted from rTgDre2-mAID-HA, rTgTah18-mAID-HA, and rHA-mAID-TgNar1/TgCIA1-smFP-cMyc parasites, separated by BN-PAGE and probed with anti-HA antibodies to detect the TgDre2-mAID-HA, TgTah18-mAID-HA, and HA-mAID-TgNar1 proteins. (B, C) Western blots of proteins extracted from rTgCIA1-mAID-cMyc/TgCIA2-HA parasites cultured for 0–12 h in IAA, separated by (B) SDS-PAGE or (C) BN-PAGE and probed with anti-HA antibodies to detect the TgCIA2-HA protein, anti-cMyc antibodies to detect the TgCIA1-mAID-cMyc protein, and anti-TgTom40 antibodies as a loading control. For the SDS-PAGE western blots (B), the relative abundance of the TgCIA2-HA protein was determined as a percentage of the 0 h IAA control, with abundances normalized using the TgTom40 loading control. Data points represent the mean ± SD of three independent experiments. Data were analyzed using a one-way ANOVA followed by Tukey’s multiple comparisons test with relevant p values shown. The numerical data underlying this Figure can be found in S1 Data. In the BN-PAGE western blot (C), the black arrowhead indicates the >720 kDa candidate CIA Targeting Complex. (D) Western blot of proteins extracted from rTgCIA2-mAID-HA/TgCIA1-smFP-cMyc parasites and immunoprecipitated using anti-HA- or anti-cMyc-conjugated agarose beads. Extracted fractions include total protein prior to immunoprecipitation (T), unbound proteins (U), and antibody-bound proteins (B) in the indicated proportions. Protein fractions were separated by SDS-PAGE and probed with anti-cMyc, anti-HA, or anti-TgTom40 antibodies. Data are representative of three independent experiments for the anti-HA immunoprecipitation and two independent experiments for the anti-cMyc immunoprecipitation. Asterisks depict likely degradation products of the TgCIA1-smFP-cMyc protein. https://doi.org/10.1371/journal.pbio.3003520.s006 (TIF) S7 Fig. Epitope tagging candidate CIA pathway proteins in Toxoplasma gondii. (A) A mAID-cMyc epitope tag was integrated into the 3′ region of the open reading frames of TgCIA1 in RH∆ku80/Tir1-FLAG/tdTomato parasites, generating the rTgCIA1-mAID-cMyc parasite line. (B, C) HA epitope tags were integrated into the (B) TgCIA2 or (C) TgMMS19 loci of the rTgCIA1-mAID-cMyc parasite line, generating the rTgCIA1-mAID-cMyc/TgCIA2-HA and rTgCIA1-mAID-cMyc/TgMMS19-HA parasite lines. (D) A spaghetti monster fluorescent protein-cMyc (smFP-cMyc) epitope tag was integrated into the 3′ region of the open reading frame of TgCIA1 in IAA-regulatable rTgCIA2-mAID-HA (top) or rTgMMS19-mAID-HA (bottom) parasites, generating the rTgCIA2-mAID-HA/TgCIA1-smFP-cMyc and rTgMMS19-mAID-HA/TgCIA1-smFP-cMyc parasite lines. A schematic depicting the target locus before and after modification, the approximate position of the forward and reverse primers used in the PCR analysis, and the expected sizes of the PCR products in the native and modified genomic loci, are shown at the top of each panel. The PCR analyses testing for genomic modifications are shown at the bottom of each panel. PCRs were performed using forward and reverse primers specific to the target site of each gene, and using genomic DNA extracted from clonal parasite lines. Genomic DNA from a WT parasite line was used as a control for the expected size of the native locus in each screen. https://doi.org/10.1371/journal.pbio.3003520.s007 (TIF) S8 Fig. The mitochondrial localization of TgCIA1 is not dependent on other candidate CIA pathway proteins. (A) Immunofluorescence assays of rTgNBP35-cMyc/TgCIA1-HA parasites, cultured in the absence (top) or presence (bottom) of ATc for two days. Samples were probed with anti-HA to detect TgCIA1-HA (green) and anti-TgTom40 antibodies to detect the mitochondrion (magenta). (B) A spaghetti monster fluorescent protein-cMyc (smFP-cMyc) epitope tag was integrated into the 3′ region of the open reading frame of TgCIA1 in IAA-regulatable rTgTah18-mAID-HA (top) or rHA-mAID-TgNar1 (bottom) parasites, generating the rTgTah18-mAID-HA/TgCIA1-smFP-cMyc and rHA-mAID-TgNar1/TgCIA1-smFP-cMyc parasite lines. A schematic depicting the target locus before and after modification, the approximate position of the forward and reverse primers used in the PCR analysis, and the expected sizes of the PCR products in the native and modified genomic loci, are shown at the top of the panel. The PCR analyses testing for genomic modifications are shown at the bottom of the panel. Note that the PCR screens for the candidate rTgTah18-mAID-HA/TgCIA1-smFP-cMyc clones was performed on the same gel as the rTgCIA2-mAID-HA/TgCIA1-smFP-cMyc clone (S7D Fig) and therefore has the same ladder and WT control. (C) Immunofluorescence assays of rTgTah18-mAID-HA/TgCIA1-smFP-cMyc parasites cultured in the absence (top) or presence (bottom) of IAA for 24 h. Samples were probed with anti-cMyc to detect TgCIA1-smFP-cMyc (green) and anti-TgTom40 antibodies to detect the mitochondrion (magenta). (D) Multiple sequence alignment of the C-terminal region of the Toxoplasma gondii Nar1 protein (TgNar1) with Nar1 homologs from the chrompodellid Vitrella brassicaformis (VbNar1; www.veupathdb.org accession number Vbra_21454; [60]), the ciliate Tetrahymena thermophila (TtNar1; UniProt accession number Q22NP0), the amoebozoan Dictyostelium discoideum (DdNar1; UniProt Q54F30), the plant Arabidopsis thaliana (AtNar1; UniProt Q94CL6), and the animals D. melanogaster (DmNar1; UniProt Q8SYS7) and H. sapiens (HsNar1; NCBI accession number NP071938.1). (E, F) Western blot of proteins extracted from rHA-mAID-TgNar1/TgCIA1-smFP-cMyc parasites cultured in the absence or presence of IAA for 24 h. Proteins were separated by SDS-PAGE (E) or BN-PAGE (F) and were probed with anti-cMyc, anti-HA, or anti-TgTom40 antibodies. The mean relative abundance of the TgCIA1-smFP-cMyc protein for the SDS-PAGE western blot was quantified a percentage of the no IAA control following normalization using the TgTom40 loading control (E, right). Data points represent the mean ± SD of three independent experiments and were analyzed using a paired t test with the p value shown. The numerical data underlying this Figure can be found in S1 Data. The black arrowhead indicates the >720 kDa CTC and the red and blue arrowheads indicate the lower mass TgCIA1-containing complexes observed in the BN-PAGE western blot (F). (G) Western blot of proteins extracted from rHA-mAID-TgNar1/TgCIA1-smFP-cMyc parasites and immunoprecipitated using anti-HA-conjugated agarose beads. Extracted fractions include total protein prior to immunoprecipitation (T), unbound proteins (U), and antibody-bound proteins (B). Protein fractions were separated by SDS-PAGE and probed with anti-cMyc, anti-HA, or anti-TgTom40 antibodies. Data are representative of three independent experiments. (H) Immunofluorescence assays of rHA-mAID-TgNar1/TgCIA1-smFP-cMyc parasites, cultured in the absence (top) or presence (bottom) of IAA for 24 h. Samples were probed with anti-cMyc to detect TgCIA1-smFP-cMyc (green) and anti-TgTom40 antibodies (magenta) to detect the mitochondrion. For all immunofluorescence assays depicted in this figure, the scale bars are 2 µm and DIC denotes the differential interference contrast transmission image. https://doi.org/10.1371/journal.pbio.3003520.s008 (TIF) S9 Fig. Multiple sequence alignment of TgCIA1 with homologs from other eukaryotes. A multiple sequence alignment of the TgCIA1 protein with homologs from the dinozoans Perkinsus marinus (NCBI accession number: XP_002785992.1) and Symbiodinium microadriaticum (NCBI CAE7872903.1), the apicomplexans Plasmodium falciparum (www.veupathdb.org accession number PF3D7_1209400; [60]) and Cryptosporidium parvum (VEuPathDB cgd1_2230), the chrompodellids Vitrella brassicaformis (VEuPathDB Vbra_18342), and Chromera velia (VEuPathDB Cvel_5245), the yeast Saccharomyces cerevisiae (UniProt accession number Q05583), the ciliate Paramecium tetraurelia (UniProt A0BLF7), the plant Arabidopsis thaliana (UniProt F4JVW1), the amoebozoan Dictyostelium discoideum (UniProt Q55DA2), the animal Drosophila melanogaster (UniProt Q7K1Y4) and the oomycete Phytophthora infestans (UniProt D0NMX9). The positions of each of the seven blades of the β-propellor structure of CIA1 are indicated. Sequences corresponding to the CD loops of interest are highlighted in pink (CD1 loop), yellow (CD3 loop), and green (CD5 loop), with the remaining CD loops and the AB loop of blade 7 (which is slightly extended in TgCIA1) highlighted in blue. The alignment was generated in Clustal Omega and plotted using pyBoxshade. https://doi.org/10.1371/journal.pbio.3003520.s009 (PDF) S10 Fig. The effects of mutating the CD loops of TgCIA1 on mitochondrial morphology and parasite proliferation. (A) Relative abundances of proteins depicted in the Fig 6B western blots were determined as a percentage of the -IAA condition for the cTgCIA1WT-Ty1 protein and normalized using the TgTom40 loading control. Data points represent the mean ± SD of three independent experiments. Data were analyzed using a one-way ANOVA followed by Tukey’s multiple comparisons test, with significant p values (<0.05) with respect to the -IAA condition of the cTgCIA1WT-Ty1-IAA protein shown. (B) Quantification of mitochondrial morphology in rTgCIA1-mAID-HA/cTgCIA1WT-Ty1 (WT-Ty1), rTgCIA1-mAID-HA/cTgCIA1ScCD1-Ty1 (ScCD1-Ty1), rTgCIA1-mAID-HA/cTgCIA1ScCD3-Ty1 (ScCD3-Ty1), and rTgCIA1-mAID-HA/cTgCIA1ScCD5-Ty1 (ScCD5-Ty1) parasites. Mitochondria were observed by immunofluorescence assays using anti-TgTom40 antibodies, and were classified as lasso, branched, linear, or tadpole shaped, with representative images of each category shown above. The morphologies of mitochondria in 150 vacuoles containing 4–16 intracellular parasites were determined in each parasite line across three independent experiments, with the observer blinded to the identities of the samples being examined. (C–E) Plaque assays of rTgCIA1-mAID-HA parasites and rTgCIA1-mAID-HA parasites constitutively expressing TgCIA1 variants, including (C) cTgCIA1WT-Ty1 (WT-Ty1), cTgCIA1ScCD1-Ty1 (ScCD1-Ty1), cTgCIA1ScCD3-Ty1 (ScCD3-Ty1) and cTgCIA1ScCD5-Ty1 (ScCD5-Ty1), (D) cTgCIA1WT-Ty1 (WT-Ty1), cTgCIA1CD5-to-CD1-Ty1 (CD5-to-CD1-Ty1), cTgCIA1VbCD5-Ty1 (VbCD5-Ty1), and cTgCIA1SmCD5-Ty1 (SmCD5-Ty1), and (E) cTgCIA1WT-Ty1 (WT-Ty1), cTgCIA1W526A-Ty1 (W526A-Ty1) cTgCIA1Y527A-Ty1 (Y527A-Ty1), cTgCIA1F532A-Ty1 (F532A-Ty1), and cTgCIA1R533A-Ty1 (R533A). Parasites were cultured in the absence (top) or presence (bottom) of IAA for six days and are representative of three independent experiments. Each plaque assay was set up simultaneously with the fluorescence proliferation assays depicted in Figs 6F, 7C, 8I, 9D, and S12G. The numerical data underlying this Figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003520.s010 (TIF) S11 Fig. The CD5 loop of TgCIA1 is sufficient to target GFP to the mitochondrion. Immunofluorescence assay of parasites constitutively expressing a GFP-Ty1 variant containing the CD5 loop of TgCIA1 between the eighth and ninth β-strands of GFP (GFPTgCD5-Ty1), probed with anti-Ty1 antibodies to detect the GFPTgCD5-Ty1 protein (green) and anti-TgTom40 antibodies to detect the mitochondrion (magenta). Scale bars are 2 µm. DIC, differential interference contrast. Right, corresponding fluorescence plots depicting the intensity of anti-Ty1 (green) and anti-TgTom40 (magenta) labeling along the yellow line in merged images. The numerical data underlying this Figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003520.s011 (TIF) S12 Fig. An aromatic amino acid motif in the CD5 loop of TgCIA1 facilitates mitochondrial targeting. (A) Sanger DNA sequencing chromatograms depicting the nucleotides modified in the TgCIA1 gene to generate substitutions of the W526, Y527, and R533 residues of the protein to alanine. Mutated codons for each line are highlighted with a magenta box. (B) The relative abundance of proteins depicted in Fig 8B was determined as a percentage of the WT-HA protein and normalized using the TgTom40 loading control. Data points represent the mean ± SD of three independent experiments. Data were analyzed using a one-way ANOVA followed by Tukey’s multiple comparisons test with p values shown. (C, D) Western blots of proteins extracted from rTgCIA1-mAID-HA/cTgCIA1WT-Ty1 (WT-Ty1), rTgCIA1-mAID-HA/cTgCIA1W526A-Ty1 (W526A-Ty1), rTgCIA1-mAID-HA/cTgCIA1Y527A-Ty1 (Y527A-Ty1), rTgCIA1-mAID-HA/cTgCIA1F532A-Ty1 (F532A-Ty1), and rTgCIA1-mAID-HA/cTgCIA1R533A-Ty1 (R533A-Ty1) parasites, separated by (C) SDS-PAGE or (D) BN-PAGE, and probed with anti-Ty1 or anti-TgTom40 antibodies. The black arrowhead indicates the >720 kDa CIA Targeting Complex; red and blue arrowheads indicate the lower mass complexes containing the TgCIA1 protein. (E) Immunofluorescence assays of rTgCIA1-mAID-HA/cTgCIA1WT-Ty1 (WT-Ty1; also shown in Fig 8G), rTgCIA1-mAID-HA/cTgCIA1W526A-Ty1 (W526A-Ty1), rTgCIA1-mAID-HA/cTgCIA1Y527A-Ty1 (Y527A-Ty1), and rTgCIA1-mAID-HA/cTgCIA1R533A-Ty1 (R533A-Ty1) parasites. The complemented proteins of interest (green) and the mitochondrion (magenta) were labeled with anti-Ty1 and anti-TgTom40 antibodies, respectively. Schematics depicting the modified amino acid sequence in the CD5 motif of the proteins from each panel are included next to images (left). Scale bars are 2 µm. DIC, differential interference contrast. Right, corresponding fluorescence profile depicting intensity of anti-Ty1 (green) and anti-TgTom40 (magenta) labeling along the yellow lines of the merged images. (F) The correlation between Ty1-tagged proteins and TgTom40 was quantified using the Pearson correlation coefficient (r) and the data were analyzed using a one-way ANOVA followed by Tukey’s multiple comparisons test with p values shown. (G) Fluorescence proliferation assays of rTgCIA1-mAID-HA and rTgCIA1-mAID-HA/cTgCIA1WT-Ty1 (also shown in Fig 8I), rTgCIA1-mAID-HA/cTgCIA1W526A-Ty1, rTgCIA1-mAID-HA/cTgCIA1Y527A-Ty1, and rTgCIA1-mAID-HA/cTgCIA1R533A-Ty1 parasites, grown in the absence (black) or presence (red) of IAA. Parasite proliferation is expressed as a percentage of the fluorescence measurement in the -IAA condition on the final day of the assay for each line. Individual data points and error bars represent the mean ± SD of three technical replicates. Error bars not visible are smaller than the symbol. Data are representative of three independent experiments. The numerical data underlying this Figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003520.s012 (TIF) S13 Fig. Immunofluorescence assay controls on untagged parasites. (A–C) Immunofluorescence assays of TATi∆ku80 parasites lacking epitope tags (A–C, top) and epitope-tagged positive control parasites (A, B bottom, TgCIA2-mAID-HA/TgCIA1-smFP-cMyc parasites; C bottom, TgCIA1-mAID-HA/cTgCIA1WT-Ty1 parasites) were performed to test for non-specific antibody labeling in untagged parasites. Parasite samples were probed on the same day with (A) anti-HA (green) and anti-TgTom40 (magenta) antibodies, (B) anti-cMyc (green), and anti-TgTom40 (magenta) antibodies, and (C) anti-Ty1 (green) and anti-TgTom40 (magenta) antibodies using the same antibody dilutions for the negative and positive control samples. For image processing, the contrast and brightness of a positive control image of each sample was adjusted linearly, and the same adjustments (i.e., the same minimum and maximum pixel intensities) were applied to TATi∆ku80 parasites. Scale bars are 2 µm. DIC, differential interference contrast transmission image. https://doi.org/10.1371/journal.pbio.3003520.s013 (TIF) S1 Table. Oligonucleotides and gBlocks used in this study (all oligonucleotides are listed in a 5′ to 3′ orientation; sgRNA-coding sequences are underlined). https://doi.org/10.1371/journal.pbio.3003520.s014 (PDF) S1 Raw images. The source images of all the western blots, PCR gels, and plaque assays that were cropped in generating figures in the manuscript. Refer to the Figure legends in the manuscript for details on each figure. https://doi.org/10.1371/journal.pbio.3003520.s015 (PDF) S1 Data. The raw and normalized numerical data for the graphical figures depicted in the manuscript. Refer to the Figure legends in the manuscript for details on each figure. https://doi.org/10.1371/journal.pbio.3003520.s016 (XLSX) S2 Data. Replicate data of representative experiments included in the manuscript, including immunofluorescence assays, SDS-PAGE and BN-PAGE western blots, fluorescence proliferation assays, and plaque assays. Refer to the Figure legends in the manuscript for details on each figure. https://doi.org/10.1371/journal.pbio.3003520.s017 (PDF) Acknowledgments We thank Mick Devoy, Fei-Ju Li, and Harpreet Vohra for assistance with flow cytometry, and the 2020 ANU Cell Biology course for undertaking the initial studies on the subcellular localizations of the various CIA pathway proteins. We are grateful to VEuPathDB for providing numerous data sets and bioinformatic search tools that were integral to the research undertaken in this study.

Editors, The PLOS Biology

doi: 10.1371/journal.pbio.3003507pmid: 41264553

Following the publication of the Expression of Concern on this article [1,2], concerns were raised about regions of similarity within the first two lanes in the left panel of Fig S2A of [1], and that [1] states that “thymocyte viability was not significantly reduced even without using the OP9-DL1 co-culture system (Figure S5A)” without viability percentages or statistical test results in the text or figure legend. As noted in [2], the images provided in S1 File in [2] to support the left western blot panel of Figure S2A in [1] do not appear to match the published figure panel and did not clarify the concerns about discontinuities or similarities in the published figure. The published figure has a stronger and clearer interaction signal than the underlying images; however, the underlying images – while different from the published figure – show a band in the position that would indicate an interaction, and support the conclusion. The corresponding author responded to state the following: In Figure S5A in [1], the data represent the percentage viability of thymocytes co-cultured with OP9 or OP9-DL1 cells under control, Dkk1, and Wnt 3A treatment conditions. The statement “thymocyte viability was not significantly reduced even without using the OP9-DL1 co-culture system” refers to a separate cell death assay performed on thymocyte cultures (presented in Figure S5B in [1]). The MTT assay readings (Figure S5A in [1]) were obtained using a spectrophotometer, and cell death measurements (Figure S5B in [1]) were performed using a hemocytometer under a phase contrast microscope. All measurements in Figure S5 in [1] were manually recorded and the data underlying Figure S5 are no longer available. The viability percentages are presented on the y-axes of Figures S5A and B in [1] and statistical significance for these observations was assessed using one-way ANOVA, with a p-value threshold of 0.01 considered significant for viability assays.The Figure S5 assays in [1] were conducted to confirm that no substantial differences in cell viability existed between cultures under the experimental conditions used. The PLOS Biology Editors issue this Editorial Note to provide an update to the Expression of Concern [2] previously issued on this article [1] and to provide readers with the additional information provided about Figure S5.

Tripodi, Matteo;Asari, Hiroki

doi: 10.1371/journal.pbio.3003406pmid: 41187108

Introduction In the mammalian brain, histaminergic neurons are exclusively located in the tuberomammillary nucleus (TMN) and surrounding areas of the posterior hypothalamus [1]. They express histidine decarboxylase (HDC) and innervate widely across the brain [2], including the retina [3–6]. These neurons remain silent during sleep, but increase their firing during wakefulness and reach their maximal activity at the state of high vigilance, much as other monoaminergic systems do, such as the serotonergic neurons in the dorsal raphe nucleus (DRN) of the brain stem [2,7,8]. Brain histamine exerts its effects on neurons via three types of receptors: the H1 and H2 receptors mediate excitation of postsynaptic cells, while the H3 receptors serve as an auto-receptor to presynaptically inhibit histamine release [1]. The role of the H4 receptors in the central nervous system remains unknown [9]. The central histaminergic system has been implicated in regulating an animal’s internal state and a broad range of behavior, including the sleep–wake cycle [10,11], arousal [12,13], suppression of food intake [14–16], and learning [17–20]. Histaminergic neurons, however, corelease γ-aminobutyric acid (GABA) [21] and interact with other neuromodulatory systems, such as the cholinergic and serotonergic systems [2,22,23]. This makes it difficult to dissect the effects of histamine by itself on neural circuit functions. In particular, histaminergic modulation of sensory processing remains largely elusive [24,25]. The retina performs the first stage of visual processing, and conveys visual information to the brain via spike trains of retinal ganglion cells (RGCs). Growing evidence supports that the activity of RGCs is correlated with various behavioral measures, such as locomotion and pupil size [26–29], pointing to a more dynamic role of the retina than is assumed for a deterministic processor. The vertebrate retina is part of the central nervous system equipped with both intrinsic and extrinsic neuromodulation mechanisms, much as other brain regions are [24,25]. For instance, dopamine is synthesized and released by a group of retinal amacrine cells under the control of circadian rhythm, and contributes to adjusting the retinal circuitry for rod or cone vision by modulating gap-junction couplings [30]. Such retinal intrinsic mechanisms of neuromodulation are, however, much slower than needed to modulate retinal function to follow rapid behavioral changes [31,32]. To this end, the retina seems to exploit the centrifugal system to receive inputs from a variety of brain regions in a species-specific manner [4]. For example, histaminergic and serotonergic projections from TMN and DRN, respectively, have been anatomically described in the mammalian retina [3,5,6,33]. However, compared to the isthmo-optic pathway in birds [34] or the olfacto-retinal circuit in fish [35], little is known about the function of the mammalian centrifugal system [36,37]. Combining in vivo single-unit extracellular recordings from either the lateral geniculate nucleus (LGN) or the optic tract (OT) with pharmacological and chemogenetic tools, here we explored how histamine affects the early visual processing in awake head-fixed mice. During the recording sessions, we simultaneously monitored the subject animal’s behavior, such as pupil dynamics and locomotion, to better correlate the effects of histamine on the visual response properties with those on the behavioral measures. We also took a computational approach to clarify the net effect of histamine on the visual responses from functional viewpoints. Because visual processing in the brain fully relies on the signals from the retina, characterization of the retinal operation in vivo is indispensable to better understand how the visual system works. Results Histamine leads to weaker and slower visual responses in the mouse lateral geniculate nucleus To investigate how histamine influences the mouse visual system, we first performed in vivo single-unit extracellular recordings from the LGN in awake head-fixed mice (Figs 1A, 1B, and S1), and compared the visual response properties of the recorded cells before and after perturbing the central histaminergic system (Figs 1 and 2; see Materials and methods for details). Besides directly examining the responses to repeatedly presented visual stimuli, we employed stimulus ensemble statistical techniques (“reverse correlation”) to systematically characterize the visual response properties of the recorded cells in the framework of a linear-nonlinear cascade model [29]. In particular, using full-field white-noise stimuli, we analyzed histaminergic effects on (1) the linear temporal filter, estimated by the spike-triggered average (STA) stimulus—i.e., the mean stimulus that triggered spiking responses—under a given behavioral state; and (2) the static nonlinear gain function, i.e., an instantaneous mapping of the STA output to the neural responses (Fig 2B). We also functionally mapped the receptive field (RF) of the cells by calculating the linear spatial filter from their responses to random binary “checkerboard” stimuli, where the contrast of each spatial location was randomly and independently inverted over time. To manipulate the histaminergic system, we used two different approaches (S2 Fig). The first was the chemogenetic activation of the central histaminergic system [38]. We injected Cre-dependent adeno-associated viruses (AAVs) carrying pharmacologically selective actuator module (PSAM)-5HT3HC channels into the TMN of the posterior hypothalamus in HDC-Cre transgenic mice, and then achieved selective activation of the target neurons by systemic administration of the pharmacologically selective effector molecule (PSEM; 5 mg/kg). The second approach was the use of pharmacological tools in wild-type mice (Fig 1C) [8]. Specifically, we used chlorphenamine (5 mg/kg) [39,40], an antagonist of a postsynaptic histamine receptor H1; and ciproxifan (12 mg/kg) [41,42], an antagonist of the H3 auto-receptor that presynaptically inhibits histamine release. These drugs effectively down- and up-regulate the action of histamine in the nervous system, respectively. Thus, resulting changes in visual processing, if any, were expected to be in opposite directions. While we let the animals move freely under the head-fixed condition, the reverse correlation analysis was done on the periods when the animals stayed stationary (<2 cm/s) to minimize the side effect of locomotion behavior (S3 Fig) [26,43]. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Histamine reduced LGN visual responses in awake head-fixed mice. (A,B) Schematic diagram of LGN recordings (A) and a representative histological image showing the electrode location (B; DiI stain, red). (C) Schematic diagram of histamine receptor localizations and their cellular function. Chlorphenamine blocks the postsynaptic receptor H1 (orange) that mediates the action of histamine postsynaptically, while ciproxifan blocks the autoreceptor H3 (green) that inhibits presynaptic histamine release. (D) Visual responses of a representative LGN cell to full-field contrast-inverting stimuli (2 s intervals) before and after chemogenetically activating HDC+ cells in TMN: top, spike raster across trials; bottom, zoom-in of the spike raster around stimulus onset (−50 to 250 ms, red shade on top) and peri-stimulus time histogram, showing a delay of the onset response after the chemogenetic treatment. (E,F) Pairwise comparison of the LGN population responses before and after the chemogenetic treatment (E, peak latency: untreated, 39 ± 17 ms, median ± median absolute deviation; treated, 64 ± 20 ms; p < 0.001, Wilcoxon signed-rank test with Bonferroni correction; F, peak firing rate: untreated, 82 ± 33 Hz; treated, 57 ± 26 Hz; p = 0.04, n = 26 LGN cells from 3 animals): circle, ON peak; cross, OFF peak; gray, inter-quartile range. (G–I) Corresponding data before and after ciproxifan administration (G, representative cell’s response, with a zoom-in around stimulus offset; H, peak latency: 37 ± 22 ms vs. 50 ± 26 ms, p < 0.001; I, peak firing rate: 95 ± 35 Hz vs. 71 ± 31; p < 0.001, n = 60 LGN cells from 4 animals). (J–L) Corresponding data before and after chlorphenamine administration (J, representative cell’s response; K, peak latency: 42 ± 20 ms vs. 38 ± 16 ms, p = 0.3; L, peak firing rate: 81 ± 43 Hz vs. 76 ± 33; p = 0.5, n = 56 LGN cells from 3 animals). (M–O) Corresponding data before and after saline administration (M, representative cell’s response; N, peak latency: 35 ± 22 ms vs. 33 ± 21 ms, p = 0.2; O, peak firing rate: 97 ± 48 Hz vs. 89 ± 34; p = 0.5, n = 53 LGN cells from 3 animals). (P,Q) Cumulative distributions of modulation indices before and after each treatment (in corresponding colors): P, peak latencies; Q, peak firing rate; * p < 0.05; ** p < 0.01; *** p < 0.001 from the post-hoc test against the saline (control) condition on the average group ranks (Kruskal–Wallis test). Data and code underlying this figure are available at https://doi.org/10.5281/zenodo.17016431. https://doi.org/10.1371/journal.pbio.3003406.g001 Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Histamine made the visual responses of LGN weaker and slower in awake head-fixed mice. (A,B) Schematic diagram of LGN recordings (A) and linear-nonlinear cascade model (B). Reverse-correlation was used to estimate the linear filter and static nonlinearity that represent response kinetics and strength, respectively. (C) Linear filters of two representative LGN cells before (gray) and after chemogenetic activation of HDC+ cells in TMN (red). We focused only on those data when the animal stayed stationary (<2 cm/s) to minimize the side effect of locomotion behavior (S3 Fig). (D–F) Representative LGN linear filters before (gray) and after pharmacological treatment (D, 12 mg/kg ciproxifan, green; E, 5 mg/kg chlorphenamine, orange) or saline injection (F). (G–J) Static nonlinearities averaged across LGN population before and after treatment in corresponding colors (G, chemogenetics, n = 52 from 3 animals; H, ciproxifan, n = 126 from 4 animals; I, chlorphenamine, n = 110 from 3 animals; J, saline, n = 92 from 3 animals): line, median; gray shade, 25 and 75 percentiles. (K–N) Cumulative distributions of the modulation index before and after each treatment (in corresponding color): peak latency (K) and frequency (L) of the linear temporal filters, mean evoked firing rate during stimulation (M), and static nonlinearity (i.e., response gain; N): * p < 0.05; ** p < 0.01; *** p < 0.001 from the post-hoc test against the saline control on the average group ranks (Kruskal–Wallis test). Data and code underlying this figure are available at https://doi.org/10.5281/zenodo.17016431. https://doi.org/10.1371/journal.pbio.3003406.g002 Upon chemogenetically activating HDC+ neurons in TMN, we found that LGN neurons generally showed weaker and slower visual responses. The peak responses to stimulus onset and offset were both significantly reduced in amplitude and prolonged in latency (Fig 1D–1F). Moreover, the mean firing rate during the white-noise visual stimulation became significantly lower (median, −4.7 Hz; interquartile range (IQR), 7.9 Hz; p < 0.001, Wilcoxon signed-rank test here and thereafter unless otherwise noted; n = 52 from 3 animals); and the response gain also significantly decreased (median, −33%; IQR, 55%; p < 0.001) that we computed as the ratio of the estimated static nonlinearity functions before and after PSEM injection (Fig 2C and 2G). The temporal filter became elongated regardless of their functional cell types (S4 Fig), leading to significantly longer peak latencies (median, +31 ms; IQR, 33 ms; p < 0.001) and lower peak frequencies (median, −16%; IQR, 47%; p < 0.001). Consistent results were obtained when histamine release was pharmacologically up-regulated. Specifically, after the application of ciproxifan (H3 antagonist), both the onset and offset peak responses became weaker and longer (Fig 1G–1I). Furthermore, during white-noise stimulation, we found significantly reduced mean firing rates (median, −3.0 Hz; IQR, 5.5 Hz; p < 0.001; n = 126 cells from 4 animals) and response gain (median, −23%; IQR, 65%; p = 0.002); and the linear temporal filters showed significantly longer peak latency (median, +16 ms; IQR, 11 ms; p < 0.001) and lower peak frequency (median, −39%; IQR, 20%; p = 0.08; Fig 2D and 2H). These effects of the pharmacological treatment were less prominent than those of the chemogenetic manipulation (Figs 1P, 1Q, and 2K–2N). None of these treatments significantly affected the estimated RF size at the population level (S5 Fig). In contrast, chlorphenamine (H1 antagonist) led to marginal changes in the visual response properties of LGN neurons in the opposite direction. The onset/offset stimulus responses were faster in some cells, but the changes were not statistically significant at the population level (Fig 1J–1L). Likewise, we observed a tendency towards an increased mean evoked firing rate (median, +1.1 Hz; IQR, 6.8 Hz; p = 0.09; n = 110 cells from 3 animals) and response gain (median, +14%; IQR, 94%; p = 0.015), and facilitated response kinetics with significantly higher peak frequency of the linear temporal filter (median, +4%; IQR, 15%; p = 0.02) but no significant changes in its peak latency (median, 0 ms; IQR, 5 ms; p = 0.7) or the RF size (S5 Fig). Taken together, these results indicate that histamine reduces the response gain and slows down the response kinetics in the mouse LGN. Histaminergic modulation starts from the mouse retina Histaminergic neurons are located solely in the hypothalamus, most densely in TMN, and project their axons widely across the brain [2], including the mouse retina [3,4,6]. To test if the observed histaminergic modulation of the visual processing in LGN originates in the retina, we next performed in vivo OT recordings to monitor the retinal output responses in awake head-fixed mice (Fig 3A and 3B) [29] before and after manipulating the central histaminergic system (Figs 3 and S6–S9). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Histamine primarily affected the visual response kinetics of RGCs. (A,B) Schematic diagram of in vivo optic tract (OT) recordings (A) and a representative histological image showing the electrode location (B; DiI stain, red). (C–F) Linear filters of two representative RGCs before (gray) and after chemogenetic (A, red) or pharmacological treatment (B, 12 mg/kg ciproxifan, green; C, 5 mg/kg chlorphenamine, orange; D, saline, blue). (G–J) Static nonlinearities averaged across RGC population before and after treatment in corresponding colors (G, chemogenetics, n = 26 cells from 4 animals; H, ciproxifan, n = 36 from 4 animals; I, chlorphenamine, n = 33 from 3 animals; J, saline, n = 46 cells): line, median; gray shade, 25 and 75 percentiles. (K–N) Cumulative distributions of the modulation index before and after each treatment (in corresponding color; n = 23 from 3 animals for chemogenetic treatment of non-TMN HDC+ cells): peak latency (K) and frequency (L) of the linear temporal filters, mean evoked firing rate during stimulation (M), and static nonlinearity gain function (N); * p < 0.05; ** p < 0.01; *** p < 0.001 from the post-hoc test against the saline control condition in the average group ranks (Kruskal–Wallis test). Data and code underlying this figure are available at https://doi.org/10.5281/zenodo.17016431. https://doi.org/10.1371/journal.pbio.3003406.g003 Consistent with the results in LGN, chemogenetic activation of HDC+ neurons in TMN led to reduced firing of RGCs (median, −9.8 Hz; IQR, 13.1 Hz; p = 0.01; n = 26 from 4 animals) with lower response gain (median, −28%; IQR, 57%; p = 0.01); and slower response kinetics with elongated linear temporal filters (Fig 3C and 3G; peak latency change, median, 10 ms; IQR, 33 ms; p < 0.001; peak frequency change, median, −17%; IQR, 34%, p = 0.003). We also found that the latencies of the peak responses to stimulus onset and offset were both significantly longer, with a reduced peak firing (S6A and S6F Fig). In contrast, no change in these response parameters was observed when targeting HDC+ populations in the anterior hypothalamus outside TMN for the PSAM/PSEM activation (firing rate change, median, 1.7 Hz, IQR, 16.5 Hz, p = 0.6; gain change, median, 8%, IQR, 99%, p = 0.5; latency change, median 0 ms, IQR, 12 ms, p = 0.6; frequency change, median −6%, IQR, 17%, p = 0.09; n = 23 from 3 animals). Neither the onset nor the offset stimulus response was affected (S6B and S6G Fig). Anterograde tracing of HDC+ cells—via injection of Cre-dependent AAVs encoding fluorescent marker proteins into the TMN of HDC-Cre animals—showed labeled axons in the optic chiasm and/or the optic nerve (3 out of 9 animals; S2C and S2D Fig). Unlike previous reports [3,4,6], however, we did not identify any labeled axons in the retinal tissues (e.g., S2E Fig), likely because these projections are sparse and located far from the cell bodies, making the signals too weak. Nonetheless, our data collectively support that the histaminergic modulation of the early visual system is mediated by the centrifugal projections specifically from TMN (Figs 3G–3J and S2). The effects of the pharmacological treatments were overall similar between LGN and RGCs, though certain differences existed. First, pharmacologically blocking the H3 receptors with ciproxifan led to slower response kinetics in both LGN and RGCs (latency change, median, 10 ms; IQR, 14 ms; p < 0.001; frequency change, median, −7%; IQR, 25%; p = 0.12; n = 36 from 4 animals), but the firing rates and the response gain remained largely unaffected in RGCs (Fig 3D and 3H; firing rate change, median −0.1 Hz; IQR, 12.7 Hz; p = 0.3; gain change, median, −7%; IQR, 69%, p = 0.5). This was also the case with the peak onset/offset stimulus responses (S6C and S6H Fig). Second, while chlorphenamine (H1 antagonist) drove faster kinetics with marginally higher firing in both LGN and RGCs, the effects on the temporal filters were primarily on the peak frequencies in LGN (Fig 2E) but on the peak latencies in RGCs (Fig 3E; latency change, median, −5 ms; IQR, 8 ms; p = 0.006; frequency change, median, +4%; IQR, 25%; p = 0.6; firing rate change, median, +2 Hz, IQR, 11 Hz; p = 0.09; gain change, median, 4%, IQR, 90%, p = 0.3; n = 33 RGCs from 3 animals). Likewise, a shorter peak latency was observed in the onset/offset stimulus responses of RGCs (S6D and S6I Fig). Finally, RGC RFs did not show significant changes in size, either (S8 Fig). Although diverse effects of histamine across retinal cell types have been reported in previous ex vivo studies [6,44–49], the histaminergic effects observed in this study were rather uniform across RGC or LGN populations. First, the effects on the peak responses to the contrast-inverting stimuli did not significantly differ between the stimulus onset and offset (Figs 1 and S6). Second, the effects on the temporal filter and nonlinearity were not substantially different among the six functional groups that we identified for RGCs and LGN cells, respectively (S4 and S7 Figs). Our datasets are, however, not large enough to identify all the ~ 40 physiological cell-types reported thus far [50,51]; and we could not target a specific cell-type or tailor visual stimuli for each cell during recordings due to technical limitations. Lastly, we identified direction-selective (DS) and orientation-selective (OS) cells using the responses to moving grating stimuli in eight different directions (see Materials and methods for details), but we found no differences in the DS or OS index values—defined by the normalized population vector length—before and after each treatment (S9 Fig). Taken all together, our data indicate that (1) the mouse early visual processing is subject to histaminergic modulation controlled by TMN, starting from the retina; (2) histamine generally suppresses the firing and slows the response kinetics; and (3) response latency and frequency tuning are likely modulated by different mechanisms. Pupil dynamics and locomotion are irrelevant to histaminergic modulation of the early visual processing in mice Histamine is a potent neuromodulator implicated in various brain functions and behaviors, such as sleep–wake cycles and arousal [1,8]. Thus, it is possible that the observed histaminergic effects on the visual response properties in the mouse early visual system are a result of correlated behavioral modulation. For example, a larger pupil size leads to stronger responses because more photons reach the retina [29]. To address this issue, we examined the relationship between the histaminergic effects on the RGC/LGN visual responses and simultaneously monitored behavioral measures (Figs 4 and S10). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. Histaminergic modulation of pupil dynamics was not causally linked with that of RGC/LGN visual responses. (A) Representative time series of the pupil size (light yellow, top 33 percentile; dark yellow, bottom 33 percentile) and its time derivative (light blue, positive; dark blue, negative) during randomly flickering full-field stimulus presentation, along with those of the firing rate dynamics of two example RGCs and locomotion (from top to bottom). (B) Estimated linear filter and nonlinearity for a representative RGC (B, cell 1) using constricted or dilated pupil periods (top) or using constricting or dilating pupil periods (bottom). Note a temporal shift of the linear filter in the former case, while a shift in the filter strength in the latter. (C) Corresponding data for another RGC (cell 2). (D–F) Comparison of the RGC population response properties between constricted and dilated pupil periods (n = 152 RGCs from 16 animals): D, peak latency, 65 ± 19 ms vs. 60 ± 17 ms, median ± median absolute deviation, p < 0.001, Wilcoxon signed-rank test; E, peak frequency, 8.2 ± 2.6 Hz vs. 8.2 ± 2.7 Hz, p = 0.6; F, mean firing rate, 26 ± 19 Hz vs. 28 ± 18 Hz, p = 0.6. (G–J) Comparison of the RGC population response properties between constricting and dilating pupil periods: G, ON-OFF polarity index of the temporal filters, −0.02 ± 0.33 vs. −0.25 ± 0.33, p < 0.001; H, peak latency, 62 ± 18 ms vs. 63 ± 18 ms, p = 0.9; I, peak frequency, 8.3 ± 2.8 Hz vs. 8.3 ± 3.0 Hz, p = 0.9; J, mean firing rate, 26 ± 18 Hz vs. 28 ± 19 Hz, p = 0.7. (K) Probability distributions of relative pupil size before and after treatment: from left to right, PSAM/PSEM for TMN HDC+ cells (red) or non-TMN cells (purple), ciproxifan (green), chlorphenamine (orange), and saline (blue). (L) Comparison of peak latencies before treatment with constricted pupil vs. after treatment with dilated pupil. Histamine nevertheless significantly slowed the RGC population responses (arrow heads): * p < 0.05; ** p < 0.01; *** p < 0.001, Wilcoxon signed rank test. (M) Comparison of peak latencies before treatment with dilated pupil vs. after treatment with constricted pupil. Effects of H1 blocker could not be distinguished from those of pupil dilation (arrow head, p = 0.5). Data and code underlying this figure are available at https://doi.org/10.5281/zenodo.17016431. https://doi.org/10.1371/journal.pbio.3003406.g004 We first assessed the effects of the pupil size on the RGC/LGN visual responses. In particular, we performed the reverse correlation analysis using subsampled data with a constricted or dilated pupil for each recording (below 33 or above 66 percentiles of the pupil size, respectively; e.g., Fig 4A–4C). Under the control condition, we found that the larger the pupil size, the faster the response kinetics for both RGCs (Fig 4D and 4E) and LGN cells (S10A and S10B Fig). We also found that the LGN responses were generally stronger with a dilated pupil than with a constricted pupil (S10C Fig). In contrast, variations in the mean evoked firing rate of RGCs between constricted and dilated pupil periods were not uniform across the population, but dependent on their response polarity (R = 0.20, p = 0.015, Pearson’s correlation with the temporal filter ON–OFF polarity index, defined by the difference between the peak and valley, normalized by the sum of the two; n = 152 RGCs from 16 animals). How does this pupil size effect interrelate with the histaminergic effect? As reported before [52], chlorphenamine (H1 antagonist) led to a substantial increase in the baseline pupil size due to its anticholinergic action (Fig 4K; 67 ± 34%, mean ± standard deviation (SD) of the median pupil size, n = 6 animals). We compared the RGC kinetics after blocking H1 with a constricted pupil (hence relatively slower) to those before the treatment with a dilated pupil (hence relatively faster), and found no significant difference (Fig 4M). Thus, we cannot exclude a possible contribution of the baseline pupil dilation to the response facilitation effect of chlorphenamine. In contrast, chemogenetic activation of TMN HDC+ cells resulted in a slight increase in the baseline pupil size (2 ± 23%, n = 6 animals), while ciproxifan (H3 antagonist) treatment led to a slight decrease (Fig 4K; −7 ± 18%, n = 8 animals), though they both resulted in slower RGC/LGN response dynamics (Figs 3 and S6). Such discrepancies are likely due to side effects of the manipulations beyond histamine: e.g., ciproxifan has modest affinity to adrenergic receptors [53] and inhibits monoamine oxidase A and B [54], while some HDC+ cells co-release histamine and GABA [21]. Here, we found that the response kinetics of RGC/LGN after these manipulations with a dilated pupil (hence relatively faster) were nevertheless slower than those under the control condition with a constricted pupil (hence relatively slower; Fig 4L). Therefore, the resulting slow response kinetics after up-regulating histamine cannot be attributable to the pupil size effects. We next analyzed the effect of pupil dynamics. In general, changes in pupil size occur much more slowly than the response dynamics of the early visual system: e.g., pupillary light reflex has a latency of hundreds of milliseconds and a time-constant of about a second [55–57]. Nevertheless, when the pupil is dilating, presented stimuli on the screen should appear brighter on the retina than intended, allowing RGCs to respond to physically darker stimuli and thereby biasing the response polarity balance. Conversely, one would expect the opposite effect when the pupil is constricting. We indeed identified the corresponding changes in the ON/OFF balance of both RGC and LGN responses (Figs 4G and S10D), using the temporal filter ON-OFF polarity index. An increase of the mean evoked firing rate was also observed in LGN, but not in RGCs, during pupil dilation (Figs 4J and S10G). However, no change in the response kinetics were observed in both RGCs and LGN between dilating and constricting pupil periods (Figs 4H, 4I, S10E, and S10F). The frequencies of saccades or eye blinks did not change, either, after the pharmacological or chemogenetic treatments (S5A and S5B Fig). We thus conclude that the pupil dynamics should have little to do with the observed histaminergic effects on the RGC/LGN visual responses. Locomotion has been reported to profoundly affect information processing throughout the mouse visual system [26,43,58–60]. When we compared the temporal filter characteristics during running and stationary periods (at a threshold of 2 cm/s), we indeed found that locomotion facilitated visual responses of RGCs and LGN cells (S3D and S3E Fig). Importantly, we excluded these running periods from the analysis of the histaminergic effects described above. Furthermore, the mice remained stationary most of the time under the head-fixed condition (S3A Fig), and neither the chemogenetic nor the pharmacological treatments resulted in a significant change in either the fraction of time spent running (S3B Fig) or the median running speed (S3C Fig). Thus, locomotion and its effects on RGC/LGN responses are irrelevant to the observed histaminergic modulation, and likely involve different mechanisms. Taken together, neither pupil dynamics nor locomotion behavior was causally linked with the histaminergic modulation of the mouse early visual system. We conclude that the central histaminergic system modulates the mouse vision, effectively leading to slower and weaker responses. Response gain modulation is more strongly correlated with peak latency modulation than with peak frequency modulation What are the mechanisms underlying the histaminergic modulation of the early visual processing? To address this question, we next took a computational approach to examine the relationship between different aspects of the visual response properties that we identified to be affected by histamine (Fig 5). In particular, we simulated the activity of an integrate-and-fire neuron with Poisson spiking and refractoriness in response to white-noise input stimuli (Fig 5A), and processed the output spike trains in the framework of the linear-nonlinear cascade model to characterize the simulated response properties in the same way as the experimental data (Fig 2B). Here, we varied two model parameters, gain and resting baseline, to examine the effects of their changes on the following four aspects of the simulated responses: spike counts, the peak latency and frequency of the estimated linear temporal filter, and the slope of the estimated static nonlinearity (expressed as half-wave rectification). The rest model parameters were fixed in all simulations, including the underlying linear filter given by a trigonometric function with logarithmic scale in time, and the refractory period filter given by an exponential function. The spike threshold was drawn from a random uniform distribution ranging from 0 to 1 for each time point to approximate the Poisson spiking properties. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. Gain modulation accompanies the peak latency modulation but not the peak frequency modulation via histaminergic effects. (A–E) Using a computational model outlined in A, here we examined the consequences of varying the baseline and gain levels on the response features, such as spike counts (B), peak latency (C), and frequency (D) of the estimated linear filter from the simulated responses, and slope of static nonlinear gain function of the model neuron (E). See Materials and methods for details. (F,G) Relationship between the magnitudes of the gain modulation and response kinetics modulation by perturbing the histaminergic system for LGN neurons (chemogenetics, red; ciproxifan, green; chlorphenamine, orange; n = 300 from 10 animals). Cell types are indicated by different markers (S4 and S7 Figs). A stronger correlation was found for the peak latencies (F, R = −0.35, p < 0.001) than for the peak frequencies (G, Pearson’s R = 0.12, p = 0.18). Here, we combined all data across cell types and experimental conditions to assess the relationship between the two dependent variables as a result of histamine level manipulations. (H,I) Corresponding figure panels for RGC responses (n = 101 from 11 animals). Significant correlation to the gain modulation was found for the peak latency modulation (R = −0.26, p = 0.03), but not for the peak frequency modulation (R = 0.22, p = 0.11). Data and code underlying this figure are available at https://doi.org/10.5281/zenodo.17016431. https://doi.org/10.1371/journal.pbio.3003406.g005 When the model neuron had a higher gain or higher resting baseline, subthreshold signal reached the spike threshold faster and more frequently; hence, total spike counts became higher (Fig 5B) and the peak latency of the estimated linear temporal filter appeared shorter (Fig 5C). In contrast, the apparent peak frequency of the simulated responses was dependent more strongly on the resting baseline than on the gain (Fig 5D). These changes in the estimated linear temporal filter properties were consistent with the experimental observation, where the magnitude of the response gain modulation by histaminergic perturbation was significantly correlated with that of the peak latency modulation (Fig 5F and 5H), but much less with that of the peak frequency modulation (Fig 5G and 5I). This agreement between the observed and simulated data indicates that, despite the simplicity, our model captures aspects of neuronal response modulation to a first approximation. The model analysis supports that the histaminergic effects on RGCs should involve at least gain and baseline modulations. First, the H1 receptors likely mediate the gain modulation in RGCs, given that chlorphenamine (H1 antagonist) led to changes in the peak latency but not the peak frequency, along with changes in the estimated static nonlinearity largely by scaling (Fig 3). Second, the peak frequency was modulated as well as the peak latency of RGCs upon chemogenetic activation of HDC+ neurons in TMN, accompanied by a shift of the estimated static nonlinearity. This implies changes in the baseline resting potential of the cells. Blocking the H3 receptors, however, led to changes in the RGC response kinetics without much affecting the mean firing rates. Furthermore, the net effects of histamine perturbations on LGN responses were more diverse (Figs 1 and 2) and not fully explainable by the simple model (Fig 5). Thus, we expect that mechanisms besides gain and baseline modulations are also involved, such as neural circuit effects through retinal amacrine cells for example [48]. Histamine leads to weaker and slower visual responses in the mouse lateral geniculate nucleus To investigate how histamine influences the mouse visual system, we first performed in vivo single-unit extracellular recordings from the LGN in awake head-fixed mice (Figs 1A, 1B, and S1), and compared the visual response properties of the recorded cells before and after perturbing the central histaminergic system (Figs 1 and 2; see Materials and methods for details). Besides directly examining the responses to repeatedly presented visual stimuli, we employed stimulus ensemble statistical techniques (“reverse correlation”) to systematically characterize the visual response properties of the recorded cells in the framework of a linear-nonlinear cascade model [29]. In particular, using full-field white-noise stimuli, we analyzed histaminergic effects on (1) the linear temporal filter, estimated by the spike-triggered average (STA) stimulus—i.e., the mean stimulus that triggered spiking responses—under a given behavioral state; and (2) the static nonlinear gain function, i.e., an instantaneous mapping of the STA output to the neural responses (Fig 2B). We also functionally mapped the receptive field (RF) of the cells by calculating the linear spatial filter from their responses to random binary “checkerboard” stimuli, where the contrast of each spatial location was randomly and independently inverted over time. To manipulate the histaminergic system, we used two different approaches (S2 Fig). The first was the chemogenetic activation of the central histaminergic system [38]. We injected Cre-dependent adeno-associated viruses (AAVs) carrying pharmacologically selective actuator module (PSAM)-5HT3HC channels into the TMN of the posterior hypothalamus in HDC-Cre transgenic mice, and then achieved selective activation of the target neurons by systemic administration of the pharmacologically selective effector molecule (PSEM; 5 mg/kg). The second approach was the use of pharmacological tools in wild-type mice (Fig 1C) [8]. Specifically, we used chlorphenamine (5 mg/kg) [39,40], an antagonist of a postsynaptic histamine receptor H1; and ciproxifan (12 mg/kg) [41,42], an antagonist of the H3 auto-receptor that presynaptically inhibits histamine release. These drugs effectively down- and up-regulate the action of histamine in the nervous system, respectively. Thus, resulting changes in visual processing, if any, were expected to be in opposite directions. While we let the animals move freely under the head-fixed condition, the reverse correlation analysis was done on the periods when the animals stayed stationary (<2 cm/s) to minimize the side effect of locomotion behavior (S3 Fig) [26,43]. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Histamine reduced LGN visual responses in awake head-fixed mice. (A,B) Schematic diagram of LGN recordings (A) and a representative histological image showing the electrode location (B; DiI stain, red). (C) Schematic diagram of histamine receptor localizations and their cellular function. Chlorphenamine blocks the postsynaptic receptor H1 (orange) that mediates the action of histamine postsynaptically, while ciproxifan blocks the autoreceptor H3 (green) that inhibits presynaptic histamine release. (D) Visual responses of a representative LGN cell to full-field contrast-inverting stimuli (2 s intervals) before and after chemogenetically activating HDC+ cells in TMN: top, spike raster across trials; bottom, zoom-in of the spike raster around stimulus onset (−50 to 250 ms, red shade on top) and peri-stimulus time histogram, showing a delay of the onset response after the chemogenetic treatment. (E,F) Pairwise comparison of the LGN population responses before and after the chemogenetic treatment (E, peak latency: untreated, 39 ± 17 ms, median ± median absolute deviation; treated, 64 ± 20 ms; p < 0.001, Wilcoxon signed-rank test with Bonferroni correction; F, peak firing rate: untreated, 82 ± 33 Hz; treated, 57 ± 26 Hz; p = 0.04, n = 26 LGN cells from 3 animals): circle, ON peak; cross, OFF peak; gray, inter-quartile range. (G–I) Corresponding data before and after ciproxifan administration (G, representative cell’s response, with a zoom-in around stimulus offset; H, peak latency: 37 ± 22 ms vs. 50 ± 26 ms, p < 0.001; I, peak firing rate: 95 ± 35 Hz vs. 71 ± 31; p < 0.001, n = 60 LGN cells from 4 animals). (J–L) Corresponding data before and after chlorphenamine administration (J, representative cell’s response; K, peak latency: 42 ± 20 ms vs. 38 ± 16 ms, p = 0.3; L, peak firing rate: 81 ± 43 Hz vs. 76 ± 33; p = 0.5, n = 56 LGN cells from 3 animals). (M–O) Corresponding data before and after saline administration (M, representative cell’s response; N, peak latency: 35 ± 22 ms vs. 33 ± 21 ms, p = 0.2; O, peak firing rate: 97 ± 48 Hz vs. 89 ± 34; p = 0.5, n = 53 LGN cells from 3 animals). (P,Q) Cumulative distributions of modulation indices before and after each treatment (in corresponding colors): P, peak latencies; Q, peak firing rate; * p < 0.05; ** p < 0.01; *** p < 0.001 from the post-hoc test against the saline (control) condition on the average group ranks (Kruskal–Wallis test). Data and code underlying this figure are available at https://doi.org/10.5281/zenodo.17016431. https://doi.org/10.1371/journal.pbio.3003406.g001 Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Histamine made the visual responses of LGN weaker and slower in awake head-fixed mice. (A,B) Schematic diagram of LGN recordings (A) and linear-nonlinear cascade model (B). Reverse-correlation was used to estimate the linear filter and static nonlinearity that represent response kinetics and strength, respectively. (C) Linear filters of two representative LGN cells before (gray) and after chemogenetic activation of HDC+ cells in TMN (red). We focused only on those data when the animal stayed stationary (<2 cm/s) to minimize the side effect of locomotion behavior (S3 Fig). (D–F) Representative LGN linear filters before (gray) and after pharmacological treatment (D, 12 mg/kg ciproxifan, green; E, 5 mg/kg chlorphenamine, orange) or saline injection (F). (G–J) Static nonlinearities averaged across LGN population before and after treatment in corresponding colors (G, chemogenetics, n = 52 from 3 animals; H, ciproxifan, n = 126 from 4 animals; I, chlorphenamine, n = 110 from 3 animals; J, saline, n = 92 from 3 animals): line, median; gray shade, 25 and 75 percentiles. (K–N) Cumulative distributions of the modulation index before and after each treatment (in corresponding color): peak latency (K) and frequency (L) of the linear temporal filters, mean evoked firing rate during stimulation (M), and static nonlinearity (i.e., response gain; N): * p < 0.05; ** p < 0.01; *** p < 0.001 from the post-hoc test against the saline control on the average group ranks (Kruskal–Wallis test). Data and code underlying this figure are available at https://doi.org/10.5281/zenodo.17016431. https://doi.org/10.1371/journal.pbio.3003406.g002 Upon chemogenetically activating HDC+ neurons in TMN, we found that LGN neurons generally showed weaker and slower visual responses. The peak responses to stimulus onset and offset were both significantly reduced in amplitude and prolonged in latency (Fig 1D–1F). Moreover, the mean firing rate during the white-noise visual stimulation became significantly lower (median, −4.7 Hz; interquartile range (IQR), 7.9 Hz; p < 0.001, Wilcoxon signed-rank test here and thereafter unless otherwise noted; n = 52 from 3 animals); and the response gain also significantly decreased (median, −33%; IQR, 55%; p < 0.001) that we computed as the ratio of the estimated static nonlinearity functions before and after PSEM injection (Fig 2C and 2G). The temporal filter became elongated regardless of their functional cell types (S4 Fig), leading to significantly longer peak latencies (median, +31 ms; IQR, 33 ms; p < 0.001) and lower peak frequencies (median, −16%; IQR, 47%; p < 0.001). Consistent results were obtained when histamine release was pharmacologically up-regulated. Specifically, after the application of ciproxifan (H3 antagonist), both the onset and offset peak responses became weaker and longer (Fig 1G–1I). Furthermore, during white-noise stimulation, we found significantly reduced mean firing rates (median, −3.0 Hz; IQR, 5.5 Hz; p < 0.001; n = 126 cells from 4 animals) and response gain (median, −23%; IQR, 65%; p = 0.002); and the linear temporal filters showed significantly longer peak latency (median, +16 ms; IQR, 11 ms; p < 0.001) and lower peak frequency (median, −39%; IQR, 20%; p = 0.08; Fig 2D and 2H). These effects of the pharmacological treatment were less prominent than those of the chemogenetic manipulation (Figs 1P, 1Q, and 2K–2N). None of these treatments significantly affected the estimated RF size at the population level (S5 Fig). In contrast, chlorphenamine (H1 antagonist) led to marginal changes in the visual response properties of LGN neurons in the opposite direction. The onset/offset stimulus responses were faster in some cells, but the changes were not statistically significant at the population level (Fig 1J–1L). Likewise, we observed a tendency towards an increased mean evoked firing rate (median, +1.1 Hz; IQR, 6.8 Hz; p = 0.09; n = 110 cells from 3 animals) and response gain (median, +14%; IQR, 94%; p = 0.015), and facilitated response kinetics with significantly higher peak frequency of the linear temporal filter (median, +4%; IQR, 15%; p = 0.02) but no significant changes in its peak latency (median, 0 ms; IQR, 5 ms; p = 0.7) or the RF size (S5 Fig). Taken together, these results indicate that histamine reduces the response gain and slows down the response kinetics in the mouse LGN. Histaminergic modulation starts from the mouse retina Histaminergic neurons are located solely in the hypothalamus, most densely in TMN, and project their axons widely across the brain [2], including the mouse retina [3,4,6]. To test if the observed histaminergic modulation of the visual processing in LGN originates in the retina, we next performed in vivo OT recordings to monitor the retinal output responses in awake head-fixed mice (Fig 3A and 3B) [29] before and after manipulating the central histaminergic system (Figs 3 and S6–S9). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Histamine primarily affected the visual response kinetics of RGCs. (A,B) Schematic diagram of in vivo optic tract (OT) recordings (A) and a representative histological image showing the electrode location (B; DiI stain, red). (C–F) Linear filters of two representative RGCs before (gray) and after chemogenetic (A, red) or pharmacological treatment (B, 12 mg/kg ciproxifan, green; C, 5 mg/kg chlorphenamine, orange; D, saline, blue). (G–J) Static nonlinearities averaged across RGC population before and after treatment in corresponding colors (G, chemogenetics, n = 26 cells from 4 animals; H, ciproxifan, n = 36 from 4 animals; I, chlorphenamine, n = 33 from 3 animals; J, saline, n = 46 cells): line, median; gray shade, 25 and 75 percentiles. (K–N) Cumulative distributions of the modulation index before and after each treatment (in corresponding color; n = 23 from 3 animals for chemogenetic treatment of non-TMN HDC+ cells): peak latency (K) and frequency (L) of the linear temporal filters, mean evoked firing rate during stimulation (M), and static nonlinearity gain function (N); * p < 0.05; ** p < 0.01; *** p < 0.001 from the post-hoc test against the saline control condition in the average group ranks (Kruskal–Wallis test). Data and code underlying this figure are available at https://doi.org/10.5281/zenodo.17016431. https://doi.org/10.1371/journal.pbio.3003406.g003 Consistent with the results in LGN, chemogenetic activation of HDC+ neurons in TMN led to reduced firing of RGCs (median, −9.8 Hz; IQR, 13.1 Hz; p = 0.01; n = 26 from 4 animals) with lower response gain (median, −28%; IQR, 57%; p = 0.01); and slower response kinetics with elongated linear temporal filters (Fig 3C and 3G; peak latency change, median, 10 ms; IQR, 33 ms; p < 0.001; peak frequency change, median, −17%; IQR, 34%, p = 0.003). We also found that the latencies of the peak responses to stimulus onset and offset were both significantly longer, with a reduced peak firing (S6A and S6F Fig). In contrast, no change in these response parameters was observed when targeting HDC+ populations in the anterior hypothalamus outside TMN for the PSAM/PSEM activation (firing rate change, median, 1.7 Hz, IQR, 16.5 Hz, p = 0.6; gain change, median, 8%, IQR, 99%, p = 0.5; latency change, median 0 ms, IQR, 12 ms, p = 0.6; frequency change, median −6%, IQR, 17%, p = 0.09; n = 23 from 3 animals). Neither the onset nor the offset stimulus response was affected (S6B and S6G Fig). Anterograde tracing of HDC+ cells—via injection of Cre-dependent AAVs encoding fluorescent marker proteins into the TMN of HDC-Cre animals—showed labeled axons in the optic chiasm and/or the optic nerve (3 out of 9 animals; S2C and S2D Fig). Unlike previous reports [3,4,6], however, we did not identify any labeled axons in the retinal tissues (e.g., S2E Fig), likely because these projections are sparse and located far from the cell bodies, making the signals too weak. Nonetheless, our data collectively support that the histaminergic modulation of the early visual system is mediated by the centrifugal projections specifically from TMN (Figs 3G–3J and S2). The effects of the pharmacological treatments were overall similar between LGN and RGCs, though certain differences existed. First, pharmacologically blocking the H3 receptors with ciproxifan led to slower response kinetics in both LGN and RGCs (latency change, median, 10 ms; IQR, 14 ms; p < 0.001; frequency change, median, −7%; IQR, 25%; p = 0.12; n = 36 from 4 animals), but the firing rates and the response gain remained largely unaffected in RGCs (Fig 3D and 3H; firing rate change, median −0.1 Hz; IQR, 12.7 Hz; p = 0.3; gain change, median, −7%; IQR, 69%, p = 0.5). This was also the case with the peak onset/offset stimulus responses (S6C and S6H Fig). Second, while chlorphenamine (H1 antagonist) drove faster kinetics with marginally higher firing in both LGN and RGCs, the effects on the temporal filters were primarily on the peak frequencies in LGN (Fig 2E) but on the peak latencies in RGCs (Fig 3E; latency change, median, −5 ms; IQR, 8 ms; p = 0.006; frequency change, median, +4%; IQR, 25%; p = 0.6; firing rate change, median, +2 Hz, IQR, 11 Hz; p = 0.09; gain change, median, 4%, IQR, 90%, p = 0.3; n = 33 RGCs from 3 animals). Likewise, a shorter peak latency was observed in the onset/offset stimulus responses of RGCs (S6D and S6I Fig). Finally, RGC RFs did not show significant changes in size, either (S8 Fig). Although diverse effects of histamine across retinal cell types have been reported in previous ex vivo studies [6,44–49], the histaminergic effects observed in this study were rather uniform across RGC or LGN populations. First, the effects on the peak responses to the contrast-inverting stimuli did not significantly differ between the stimulus onset and offset (Figs 1 and S6). Second, the effects on the temporal filter and nonlinearity were not substantially different among the six functional groups that we identified for RGCs and LGN cells, respectively (S4 and S7 Figs). Our datasets are, however, not large enough to identify all the ~ 40 physiological cell-types reported thus far [50,51]; and we could not target a specific cell-type or tailor visual stimuli for each cell during recordings due to technical limitations. Lastly, we identified direction-selective (DS) and orientation-selective (OS) cells using the responses to moving grating stimuli in eight different directions (see Materials and methods for details), but we found no differences in the DS or OS index values—defined by the normalized population vector length—before and after each treatment (S9 Fig). Taken all together, our data indicate that (1) the mouse early visual processing is subject to histaminergic modulation controlled by TMN, starting from the retina; (2) histamine generally suppresses the firing and slows the response kinetics; and (3) response latency and frequency tuning are likely modulated by different mechanisms. Pupil dynamics and locomotion are irrelevant to histaminergic modulation of the early visual processing in mice Histamine is a potent neuromodulator implicated in various brain functions and behaviors, such as sleep–wake cycles and arousal [1,8]. Thus, it is possible that the observed histaminergic effects on the visual response properties in the mouse early visual system are a result of correlated behavioral modulation. For example, a larger pupil size leads to stronger responses because more photons reach the retina [29]. To address this issue, we examined the relationship between the histaminergic effects on the RGC/LGN visual responses and simultaneously monitored behavioral measures (Figs 4 and S10). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. Histaminergic modulation of pupil dynamics was not causally linked with that of RGC/LGN visual responses. (A) Representative time series of the pupil size (light yellow, top 33 percentile; dark yellow, bottom 33 percentile) and its time derivative (light blue, positive; dark blue, negative) during randomly flickering full-field stimulus presentation, along with those of the firing rate dynamics of two example RGCs and locomotion (from top to bottom). (B) Estimated linear filter and nonlinearity for a representative RGC (B, cell 1) using constricted or dilated pupil periods (top) or using constricting or dilating pupil periods (bottom). Note a temporal shift of the linear filter in the former case, while a shift in the filter strength in the latter. (C) Corresponding data for another RGC (cell 2). (D–F) Comparison of the RGC population response properties between constricted and dilated pupil periods (n = 152 RGCs from 16 animals): D, peak latency, 65 ± 19 ms vs. 60 ± 17 ms, median ± median absolute deviation, p < 0.001, Wilcoxon signed-rank test; E, peak frequency, 8.2 ± 2.6 Hz vs. 8.2 ± 2.7 Hz, p = 0.6; F, mean firing rate, 26 ± 19 Hz vs. 28 ± 18 Hz, p = 0.6. (G–J) Comparison of the RGC population response properties between constricting and dilating pupil periods: G, ON-OFF polarity index of the temporal filters, −0.02 ± 0.33 vs. −0.25 ± 0.33, p < 0.001; H, peak latency, 62 ± 18 ms vs. 63 ± 18 ms, p = 0.9; I, peak frequency, 8.3 ± 2.8 Hz vs. 8.3 ± 3.0 Hz, p = 0.9; J, mean firing rate, 26 ± 18 Hz vs. 28 ± 19 Hz, p = 0.7. (K) Probability distributions of relative pupil size before and after treatment: from left to right, PSAM/PSEM for TMN HDC+ cells (red) or non-TMN cells (purple), ciproxifan (green), chlorphenamine (orange), and saline (blue). (L) Comparison of peak latencies before treatment with constricted pupil vs. after treatment with dilated pupil. Histamine nevertheless significantly slowed the RGC population responses (arrow heads): * p < 0.05; ** p < 0.01; *** p < 0.001, Wilcoxon signed rank test. (M) Comparison of peak latencies before treatment with dilated pupil vs. after treatment with constricted pupil. Effects of H1 blocker could not be distinguished from those of pupil dilation (arrow head, p = 0.5). Data and code underlying this figure are available at https://doi.org/10.5281/zenodo.17016431. https://doi.org/10.1371/journal.pbio.3003406.g004 We first assessed the effects of the pupil size on the RGC/LGN visual responses. In particular, we performed the reverse correlation analysis using subsampled data with a constricted or dilated pupil for each recording (below 33 or above 66 percentiles of the pupil size, respectively; e.g., Fig 4A–4C). Under the control condition, we found that the larger the pupil size, the faster the response kinetics for both RGCs (Fig 4D and 4E) and LGN cells (S10A and S10B Fig). We also found that the LGN responses were generally stronger with a dilated pupil than with a constricted pupil (S10C Fig). In contrast, variations in the mean evoked firing rate of RGCs between constricted and dilated pupil periods were not uniform across the population, but dependent on their response polarity (R = 0.20, p = 0.015, Pearson’s correlation with the temporal filter ON–OFF polarity index, defined by the difference between the peak and valley, normalized by the sum of the two; n = 152 RGCs from 16 animals). How does this pupil size effect interrelate with the histaminergic effect? As reported before [52], chlorphenamine (H1 antagonist) led to a substantial increase in the baseline pupil size due to its anticholinergic action (Fig 4K; 67 ± 34%, mean ± standard deviation (SD) of the median pupil size, n = 6 animals). We compared the RGC kinetics after blocking H1 with a constricted pupil (hence relatively slower) to those before the treatment with a dilated pupil (hence relatively faster), and found no significant difference (Fig 4M). Thus, we cannot exclude a possible contribution of the baseline pupil dilation to the response facilitation effect of chlorphenamine. In contrast, chemogenetic activation of TMN HDC+ cells resulted in a slight increase in the baseline pupil size (2 ± 23%, n = 6 animals), while ciproxifan (H3 antagonist) treatment led to a slight decrease (Fig 4K; −7 ± 18%, n = 8 animals), though they both resulted in slower RGC/LGN response dynamics (Figs 3 and S6). Such discrepancies are likely due to side effects of the manipulations beyond histamine: e.g., ciproxifan has modest affinity to adrenergic receptors [53] and inhibits monoamine oxidase A and B [54], while some HDC+ cells co-release histamine and GABA [21]. Here, we found that the response kinetics of RGC/LGN after these manipulations with a dilated pupil (hence relatively faster) were nevertheless slower than those under the control condition with a constricted pupil (hence relatively slower; Fig 4L). Therefore, the resulting slow response kinetics after up-regulating histamine cannot be attributable to the pupil size effects. We next analyzed the effect of pupil dynamics. In general, changes in pupil size occur much more slowly than the response dynamics of the early visual system: e.g., pupillary light reflex has a latency of hundreds of milliseconds and a time-constant of about a second [55–57]. Nevertheless, when the pupil is dilating, presented stimuli on the screen should appear brighter on the retina than intended, allowing RGCs to respond to physically darker stimuli and thereby biasing the response polarity balance. Conversely, one would expect the opposite effect when the pupil is constricting. We indeed identified the corresponding changes in the ON/OFF balance of both RGC and LGN responses (Figs 4G and S10D), using the temporal filter ON-OFF polarity index. An increase of the mean evoked firing rate was also observed in LGN, but not in RGCs, during pupil dilation (Figs 4J and S10G). However, no change in the response kinetics were observed in both RGCs and LGN between dilating and constricting pupil periods (Figs 4H, 4I, S10E, and S10F). The frequencies of saccades or eye blinks did not change, either, after the pharmacological or chemogenetic treatments (S5A and S5B Fig). We thus conclude that the pupil dynamics should have little to do with the observed histaminergic effects on the RGC/LGN visual responses. Locomotion has been reported to profoundly affect information processing throughout the mouse visual system [26,43,58–60]. When we compared the temporal filter characteristics during running and stationary periods (at a threshold of 2 cm/s), we indeed found that locomotion facilitated visual responses of RGCs and LGN cells (S3D and S3E Fig). Importantly, we excluded these running periods from the analysis of the histaminergic effects described above. Furthermore, the mice remained stationary most of the time under the head-fixed condition (S3A Fig), and neither the chemogenetic nor the pharmacological treatments resulted in a significant change in either the fraction of time spent running (S3B Fig) or the median running speed (S3C Fig). Thus, locomotion and its effects on RGC/LGN responses are irrelevant to the observed histaminergic modulation, and likely involve different mechanisms. Taken together, neither pupil dynamics nor locomotion behavior was causally linked with the histaminergic modulation of the mouse early visual system. We conclude that the central histaminergic system modulates the mouse vision, effectively leading to slower and weaker responses. Response gain modulation is more strongly correlated with peak latency modulation than with peak frequency modulation What are the mechanisms underlying the histaminergic modulation of the early visual processing? To address this question, we next took a computational approach to examine the relationship between different aspects of the visual response properties that we identified to be affected by histamine (Fig 5). In particular, we simulated the activity of an integrate-and-fire neuron with Poisson spiking and refractoriness in response to white-noise input stimuli (Fig 5A), and processed the output spike trains in the framework of the linear-nonlinear cascade model to characterize the simulated response properties in the same way as the experimental data (Fig 2B). Here, we varied two model parameters, gain and resting baseline, to examine the effects of their changes on the following four aspects of the simulated responses: spike counts, the peak latency and frequency of the estimated linear temporal filter, and the slope of the estimated static nonlinearity (expressed as half-wave rectification). The rest model parameters were fixed in all simulations, including the underlying linear filter given by a trigonometric function with logarithmic scale in time, and the refractory period filter given by an exponential function. The spike threshold was drawn from a random uniform distribution ranging from 0 to 1 for each time point to approximate the Poisson spiking properties. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. Gain modulation accompanies the peak latency modulation but not the peak frequency modulation via histaminergic effects. (A–E) Using a computational model outlined in A, here we examined the consequences of varying the baseline and gain levels on the response features, such as spike counts (B), peak latency (C), and frequency (D) of the estimated linear filter from the simulated responses, and slope of static nonlinear gain function of the model neuron (E). See Materials and methods for details. (F,G) Relationship between the magnitudes of the gain modulation and response kinetics modulation by perturbing the histaminergic system for LGN neurons (chemogenetics, red; ciproxifan, green; chlorphenamine, orange; n = 300 from 10 animals). Cell types are indicated by different markers (S4 and S7 Figs). A stronger correlation was found for the peak latencies (F, R = −0.35, p < 0.001) than for the peak frequencies (G, Pearson’s R = 0.12, p = 0.18). Here, we combined all data across cell types and experimental conditions to assess the relationship between the two dependent variables as a result of histamine level manipulations. (H,I) Corresponding figure panels for RGC responses (n = 101 from 11 animals). Significant correlation to the gain modulation was found for the peak latency modulation (R = −0.26, p = 0.03), but not for the peak frequency modulation (R = 0.22, p = 0.11). Data and code underlying this figure are available at https://doi.org/10.5281/zenodo.17016431. https://doi.org/10.1371/journal.pbio.3003406.g005 When the model neuron had a higher gain or higher resting baseline, subthreshold signal reached the spike threshold faster and more frequently; hence, total spike counts became higher (Fig 5B) and the peak latency of the estimated linear temporal filter appeared shorter (Fig 5C). In contrast, the apparent peak frequency of the simulated responses was dependent more strongly on the resting baseline than on the gain (Fig 5D). These changes in the estimated linear temporal filter properties were consistent with the experimental observation, where the magnitude of the response gain modulation by histaminergic perturbation was significantly correlated with that of the peak latency modulation (Fig 5F and 5H), but much less with that of the peak frequency modulation (Fig 5G and 5I). This agreement between the observed and simulated data indicates that, despite the simplicity, our model captures aspects of neuronal response modulation to a first approximation. The model analysis supports that the histaminergic effects on RGCs should involve at least gain and baseline modulations. First, the H1 receptors likely mediate the gain modulation in RGCs, given that chlorphenamine (H1 antagonist) led to changes in the peak latency but not the peak frequency, along with changes in the estimated static nonlinearity largely by scaling (Fig 3). Second, the peak frequency was modulated as well as the peak latency of RGCs upon chemogenetic activation of HDC+ neurons in TMN, accompanied by a shift of the estimated static nonlinearity. This implies changes in the baseline resting potential of the cells. Blocking the H3 receptors, however, led to changes in the RGC response kinetics without much affecting the mean firing rates. Furthermore, the net effects of histamine perturbations on LGN responses were more diverse (Figs 1 and 2) and not fully explainable by the simple model (Fig 5). Thus, we expect that mechanisms besides gain and baseline modulations are also involved, such as neural circuit effects through retinal amacrine cells for example [48]. Discussion Here we provide in vivo electrophysiological evidence for histaminergic modulation of the mouse early visual system: the more the histamine released, the slower and weaker the LGN and RGC visual responses in awake head-fixed mice (Figs 1–3). Anterograde tracing of TMN HDC+ neurons confirmed the projection of their axons into the optic chiasm and optic nerve, but not in the retinal tissue (S2C–S2E Fig). Nonetheless, here we used the same mouse lines as previous studies that demonstrated the presence of histaminergic HDC+ axons in the retina [5,6]. Moreover, we observed modulation at the level of RGC axons in the OT in vivo (Figs 3 and S6). It is intriguing that punctate signals of presynaptic markers expressed in TMN HDC+ cells were identified in the optic chiasm (S2C Fig), raising a potential role for axonal modulation. However, its contribution should be minimal, if any, because RGC axon potentials should reach the recording site in the OT within 1 ms after exiting the eye cup (optic nerve length in mice, <10 mm; conduction velocity, around 10 m/s) [61]. While axonal conduction velocity depends on various factors, including myelination and channel density, its modulation should then occur only on a much shorter time scale than the observed changes in RGC/LGN response kinetics (up to tens of milliseconds). Taken together, we suggest that the histaminergic modulation likely arises in the retina. Previous ex vivo studies have reported diverse histaminergic effects in all five major cell types in the retina across species [6,44–49]. Primate RGCs, for instance, either increase or decrease their baseline activity, while their visual responses typically become weaker and slower after histamine application ex vivo [44,46]. In contrast, rodent RGCs mostly show an increase in the baseline activity, with various changes in their visual response properties in a histamine dose-dependent manner [6,44]. Some of those changes are interrelated: e.g., a broadening of direction-selectivity (DS) tuning can be explained as a natural consequence of the increased baseline activity because a decrease of the DS index (DSI) should follow by definition. Indeed, a reduction of DSI was reported at RGC axon terminals, following an increased baseline activity via arousal or locomotion [26,27]. However, the mechanisms underlying the histaminergic modulation of the retina remain mostly unclear. We took a computational modeling approach to better understand the net effect of histamine from functional viewpoints (Fig 5), rather than pursuing the underlying biophysical mechanisms. For example, we suggest that the H1 receptors mediate gain modulation in RGCs: chlorphenamine (H1 antagonist) affected the response latency but not the frequency tuning (Fig 3), a distinctive feature of gain modulation identified by our computational modeling analysis (Fig 5). This action of histamine does not have to be directly on RGCs expressing the H1 receptors, but can be mediated indirectly by circuit mechanisms. Indeed, histamine affects both inward and outward currents of amacrine cells in the mouse retina [48]. The effects of histamine can then be more profound than gain modulation, much as we observed diverse outcomes in RGCs and LGN when different pharmacological and chemogenetic perturbation tools were employed (see also [6]). Correlation between histamine levels and various behavioral measures has been widely reported [62,63]. For instance, the activity of central histaminergic neurons generally follows the circadian rhythm, showing high levels of activity during the active period (i.e., night time for mice as nocturnal animals) and low levels during the sleep period [7,10]. Here we performed all our experiments during daytime (ZT4-ZT9). Moreover, the mice remained calm under head fixation, as evidenced by their low spontaneous running behavior (S3 Fig), following a week of habituation sessions before recordings. The retinal histamine level of the subject mice was then expected to be relatively low during the recordings, given that the mammalian retina contains a comparable level of histamine with other brain areas (around 20–140 ng/g wet tissue, i.e., about 0.1–1 μM) [64]. This will explain why we found stronger effects by chemogenetically activating HDC+ neurons in TMN or pharmacologically blocking H3 receptors (mimicking an increase of histamine) than by administering H1 antagonists (mimicking a decrease of histamine). However, regulation of the histamine release and metabolism is suggested to be both tissue- and species-specific [64]: e.g., light stimulation of dark-adapted rabbits at night leads to a substantial reduction of histamine in the retina and the optic nerve (by about 40%–50%), but not in other parts of the central nervous system. It is a future challenge to determine the exact level of retinal histamine and its fluctuation under more precisely controlled behavioral contexts. Our observations on the histaminergic effects have certain discrepancies from those reported in previous studies. For example, it has been reported that histamine increased the baseline activity of RGCs in the mouse retinal explants [6]. This effect was dose-dependent and saturated at 5 μM histamine with an increased baseline firing by ~4 Hz on average. We instead found a reduced firing upon H3 antagonist application or chemogenetic activation of TMN HDC+ cells, while an enhanced firing after blocking H1 receptors (Figs 3 and S6). This might be because the baseline activity of RGCs in awake mice is already high, by ~20 Hz compared to the ex vivo condition [29]. Alternatively, the enhancement of the baseline activity in retinal explants could be an artifact of histamine overdose. An isolated retinal tissue should have no baseline histamine because the retina itself does not produce histamine [3,5]. However, bath-application of histamine at the physiological range (0.1–1 μM) [64] or that of H3 antagonists had no effect on the baseline firing ex vivo [6]. Moreover, chemogenetic activation of HDC+ axon terminals in retinal explants led to a net reduction of firing, with ~10% of RGCs showing a significant decrease of the baseline firing while only ~1% showing an increase [65]. The retinal environment clearly differs between ex vivo and in vivo conditions. One should then be cautious as findings from ex vivo studies may not be directly applicable to the awake, more natural state. What is the biological relevance of histaminergic modulation of the early visual system, starting at the retina? We speculate that species-specific ethological factors play a role. For example, some species may sleep with their eyes open [66–69], including mice [70–73]. Furthermore, as nocturnal animals, mice are less active during daytime. Facilitating visual responses at low histamine level may then be ethologically beneficial for mice to respond faster to visual threats during daytime, especially when mice are less active. In contrast, when animals are more active, other mechanisms, such as arousal, may compensate for the histaminergic effects [26,27]. Such arguments also hold for those species without eyelids, including some amphibians, reptiles, and fish. To our knowledge, we are the first to report histaminergic effects on LGN in awake mice. Recordings from LGN in cats and guinea pigs (under the anesthetized condition or brain slices [74,75]) suggest that histamine leads to a slow depolarization of the relay cells, enhancing the baseline activity but introducing a lag in their visual responses. However, substantial differences exist in the early visual processing between awake and anesthetized conditions [29,76]. Further studies are needed to address how histamine affects visual processing and perception in awake animals in tissue- and species-specific manners. Growing evidence supports significant correlations between awake retinal responses and various behavioral measures, such as pupil dynamics and locomotion [26–29]. Such correlations were also found in our data sets; however, we demonstrated that an animal’s pupil dynamics and locomotion behavior have little to do with the observed histaminergic effects on RGC/LGN responses (Figs 4, S3, and S10). This indicates that mechanisms other than histamine are involved in the retinal modulation by arousal and locomotion. The retina exploits not just histamine but also many other neuromodulators, synthesized inside [77] or outside the retina [36,37], such as the centrifugal serotonergic system [78–80]. It is a future challenge to identify what behavior is controlled by histamine and vice versa, and characterize the anatomical and physiological features of those HDC+ cells projecting to the retina to clarify behavioral relevance of the histaminergic modulation of the early visual system. Materials and methods No statistical method was used to predetermine the sample size. The significance level was 0.05 (with multiple comparison correction where appropriate) in all analyses unless noted otherwise. All experiments were performed under the license 233/2017-PR and 220/2024-PR from the Italian Ministry of Health, following protocols approved by the Institutional Animal Care and Use Committee at European Molecular Biology Laboratory. The data analyses were done in Python and Matlab. Animals Animals were housed on a 12 h light-dark cycle, with ad libitum access to water and food. All experiments were performed on female mice. A total of 22 wild-type (C57BL6/J; RRID:IMSR_JAX:000664) and 12 hemizygous HDC-Cre (Hdctm1.1(icre)Wwis/J; RRID:IMSR_JAX:021198) mice were used for pharmacological and chemogenetic manipulations, respectively. Additional 9 hemizygous HDC-Cre mice were used for anterograde tracing of HDC+ cells in TMN. Surgical procedures All surgical procedures were performed in animals from 4 to 19 weeks old (9.2 weeks median age). Before the surgery, animals were injected with Carprofen (5 mg/kg) and then anaesthetized with isoflurane (4% for induction, 1% for maintenance in O2). Throughout the surgery, the temperature of the animals was kept stable at 37 °C using a heating pad (Supertech Physiological). Ointment (VitA-Pos, Ursapharm) was applied on both eyes to prevent them from drying during the surgery, and the animals were placed in a stereotaxic frame (Stoelting 51625). A portion of the scalp was removed to expose the skull, and the periosteum was scraped away with a round scalpel to increase adherence of the dental cement. For those mice belonging to the chemogenetic manipulation group, the skull was aligned to level bregma and lambda on the same horizontal plane, and a 100 µm craniotomy was performed using a motorized drill attached to the stereotaxic arm. Viral solution (250 nL; AAV9::FLEX-PSAM Y115F,L141F:5HT3 HC; Vector Biolabs, VB773) was injected with an automatic injector pump at 1 nL/sec, at [−2.6, ±0.8, −5.3] or [−1.7, ±0.5, −5.3] in [Anterior–Posterior (AP), Medial–Lateral (ML), Dorsal–Ventral (DV)] coordinates for targeting TMN and non-TMN areas of the hypothalamus, respectively. All animals were then implanted with a custom-made titanium head plate, cemented on the skull using a mixture of cyanoacrylate (Loctite 401, Henkel) and dental cement (Paladur, Kulzer). The head plate featured a hole (8 mm diameter), allowing for clear observation of both bregma and lambda. The skull surface was then glazed with a thick layer of cyanoacrylate to support the skull with mechanical, atmospheric, and biological protection, while still allowing for visual identification of reference points. After the surgery animals were allowed to recover on a heating pad and then housed in individual cages. During the following seven days, the mice were administered with analgesia (Carprofen; 50 mg/mL) diluted in the drinking water. Anterograde tracing For anterograde tracing of HDC+ cells, we injected recombinant AAVs carrying Cre-dependent fluorescent markers (100 nL; rAAV9::CAG-FLEX-Synaptophysin-GFP or AAV2/1::CAG-FLEX-axon-GFP) into the TMN of HDC-Cre mice (n = 9) as described above. At least 10 days after the injection, the animals were anesthetized (2.5% Avertin, 16 μL/g, intraperitoneal injection) and perfused with paraformaldehyde (PFA; 4% in phosphate buffer solution). The brain tissues including the optic chiasms and optic nerves were dissected and post-fixed overnight in 4% PFA at 4 °C. Coronal sections of the brain tissue (thickness, 80 μm) were then examined under a laser scanning confocal microscope (Leica TCS SP5; S2 Fig). In vivo electrophysiology After recovery from the surgery, animals were habituated to head fixation by placing them in the experimental apparatus over the course of one week, twice a day for up to two hours. On the day of the recording, we first placed the subject animal in the recording set up with its head fixed, and determined the electrode penetration path to the target area (OT, [−1.34, +1.87, +4.74], [−1.70, +1.87, +4.74], or [−1.82, +2.35, +4.07] in [AP, ML, DV] coordinates; LGN, [−2.3, +2.3, +2.8]) using the robotic stereotaxic system (StereoDrive, NeuroStar). The animal was then briefly anesthetized (with isoflurane for about 5 min) and a hole was drilled around the electrode entry point on the skull. After the removal of the anesthesia, an acute silicone probe (Buzsaki32L, Neuronexus; or P2, Cambridge Neurotech) coated with a fluorescent dye (DiI stain, Invitrogen, D282) was lowered at 5 µm/s using the robotic arm until visual responses were found in the target area. All recordings were done during the daytime (between 11 am/ZT4 and 4 pm/ZT9; 1 pm/ZT6 ±1 h, mean ± SD; 96% of the paired recordings were conducted within 2 h of each other). After presenting a battery of visual stimuli, we briefly anesthetized the animal with isoflurane and intraperitoneally injected 0.9% saline solution alone or that containing chlorphenamine (H1 antagonist, 5 mg/kg) [39,40], ciproxifan (H3 antagonist, 12 mg/kg) [41,42], or PSEM (5 mg/kg) [38]. About 20 min after the injection and removal of the anesthesia, we presented the same battery of visual stimuli to the animal for the second recording session. Here, we did not target the H2 or H4 receptors because blocking the H2 receptors has little effect on retinal physiology ex vivo [6], and the functional role of the H4 receptors in the central nervous system remains unclear [9]. In our electrophysiology experiments, we could not keep the same cells long enough to test if the effects of these manipulations were reversible or not. After all the recording sessions, the electrode position was verified histologically (e.g., Figs 1B, LGN and 3B, OT). After retracting the silicon probe, the mice were anesthetized (2.5% Avertin, 16 μL/g, intraperitoneal injection) and perfused with 4% PFA. The brain tissue was harvested and post-fixed overnight in 4% PFA at 4 °C. Coronal sections of the brain tissue (thickness, 100–150 μm) were then examined under a fluorescence microscope (Leica, LMD7000 with N2.1 filter cube) to visualize the trace left by the DiI stain on the probe. To examine the PSAM expression, staining of the slices was performed (S2A, and S2B Fig): primary staining, 1-h incubation with an alpha-bungarotoxin/biotin-XX conjugate (ThemoFisher, B1196) diluted with PBS with 0.1% Tween (1/500) at room temperature; secondary staining, streptavidin and alexa-fluorophore, 1/500 in PBS(T) 0.1% for 2 hours at room temperature. Visual stimulation Visual stimuli were presented as described before [29]. Briefly, visual stimuli were projected onto a spherical screen (20 cm in radius) painted with ultra-violet (UV) reflective white paint, placed 20 cm from the mouse’s left eye. A gamma-corrected digital light processing device (DLP, Texas Instruments, DLPDLCR3010EVM-LC) was used as a light source after the green and the red light-emitting diodes (LEDs) were replaced with UV (365nm, LZ1-00UV00, LED Engine) and infrared (IR; 950nm, SFH 4725S, Osram) LEDs, respectively. The IR light was used as a synchronization signal and recorded with a photodiode (PDA100A2, Thorlabs). Stimulation was conducted at 60 frames per second (FPS), and covered 73° in azimuth and 44° in altitude from the mouse eye position. The maximum light intensity at the eye position was 31 mW/m2 (15.4 mW/m2 for UV LED and 15.9 mW/m2 for blue LED; measured with S121C sensor, Thorlabs), leading to mesopic to photopic conditions. We presented the following stimuli using QDSpy: a randomly flickering full-field stimulus (5 min), consisting of a random sequence of black and white frames at 60 FPS; a black-and-white binary dense noise stimulus (15 min), consisting of a 32-by-18 pixels checkerboard, where each pixel followed an independent random sequence at 60 FPS but the overall luminance of the frame was kept constant at the mean value; a moving grating stimulus in eight different directions (spatial frequencies of square waves, 3° or 20°; moving speed, 7.5°/s or 15°/s); full-field contrast-inverting stimuli in an “OFF-ON-OFF” sequence at maximum contrast (2 s each), followed by a sinusoid (1.5 Hz) with a linearly increasing amplitude from 0% to 100% contrast over 10 s (10 trials) or a sequence of sinusoids with different temporal frequencies (1.875, 3.75 and 7.5 Hz, each for 2 s; 15 and 30 Hz for 1 s) at maximum contrast (10 trials), with an inter-trial interval of 1 s (gray screen). Electrophysiology data analysis We adapted previously established methods of spiking sorting and OT data analysis [29]. In brief, we first concatenated the raw data from the same animal before and after treatment into a single binary file, and used Kilosort 2.0 to sort spikes with a set of default parameters, except for the spike detection threshold to be 6 during optimization. Single units were then identified by clustering in principal component space using Phy for visualization and manual data curation. Only those units that maintained the average spike waveforms and autocorrelograms with a minimal refractory period of 1 ms were kept for subsequent analyses (S1 Fig). Specifically, we selected RGCs/LGN cells as those with robust visual responses: i.e., , where is the response during the “ON-OFF” part of the OFF-ON-OFF stimulus sequence, indicates the mean over trials, and the variance over time (bin size, 1/60 ms), respectively; and the estimated temporal filter should have in at least one time bin within 200 ms from the spike onset (see below for details). Single-units with little or no visual responses were excluded from the analysis as nontarget cells, such as the axons from the parabigeminal nucleus in the OT [81,82]. In total, we obtained 337 RGCs from 20 animals and 557 cells in LGN from 14 mice. These RGCs typically had a triphasic spike waveform as expected for axonal signals [83]. To characterize the peak latency and firing rate of the responses to full-field contrast-inverting stimuli (Figs 1 and S6), we first identified the cell’s response preference to stimulus polarity using an ON–OFF index defined as , where and are the mean firing rate during the ON and the second OFF periods of the OFF–ON–OFF stimulus sequence. We then computed the peri-stimulus time histogram (bin size, 1 ms), smoothed it with a Gaussian filter (kernel width, 6 SD), and identified the first peak upon stimulus onset or offset for those cells with positive or negative ON–OFF index values, respectively. We then performed pairwise comparisons on the peak latencies and firing rates before and after each treatment (e.g., Fig 1F–1J). To compare the changes in the detected peak latency and firing rates across different conditions (e.g., Fig 1K and 1L), we performed Kruskal–Wallis test on a modulation index defined as , where is either the detected peak latency or peak firing rate before and after treatment. For systematically characterizing the visual response properties, we used stimulus ensemble statistical techniques (reverse correlation methods; 500 ms window; Δt = 1/60 s bin size) to calculate the linear filter and static nonlinear gain function of the recorded cells in response to white-noise stimuli as described before (Fig 2B) [29]. Briefly, we first obtained the linear filter of each cell by calculating a STA of the stimulus with ±1 being “white” and “black,” respectively, under a given behavioral state (e.g., when the animal stayed stationary below 2 cm/s). As a quality measure, p-value was computed for each time bin against a null hypothesis that the STA follows a normal distribution with a mean of zero and a variance of , where is the total number of spikes. As a measure of the cell’s response kinetics, we then estimated the peak latency by fitting a difference-of-Gaussian curve to the linear filter; and the spectral peak frequency by the Fourier analysis on the linear filter. The ON–OFF polarity index of a temporal filter was defined as the difference of its peak and valley, divided by the sum of the two (Figs 4G and S10D). Spatial response properties were examined for STAs from “checkerboard” stimulation, using the time window where the eye position remained within 1 inter-quartile range in both horizontal and vertical coordinates for at least 2 s (S5C and S8A Figs) to minimize the effects of eye movements while retaining enough data for RF estimation. In particular, we fitted a two-dimensional Gaussian envelope to the spatial filter at the peak latency (e.g., S5D Fig), and the RF size was estimated as twice the mean SD of the long and short axes. Modulation index was used to characterize the change in a response feature before and after treatment, such as the peak latency and frequency, the mean firing rate, and the RF size (Figs 2K–2N and 3K–3N). Here we focused only on those data when the mouse stayed stationary (<2 cm/s) to minimize the side effects of locomotion behavior (S3 Fig) [26,43]. The linear temporal filters from full-field white-noise stimuli were used for broadly classifying the response types of RGCs and LGN cells, respectively (S4A, S4B, S7A, and S7B Figs). Specifically, we first used t-Distributed Stochastic Neighbor Embedding to map the STAs onto two-dimensional space, and then used K-means++ algorithm for heuristically categorizing the responses into the following six types: fast ON, ON, slow ON, fast OFF, OFF, and slow OFF. Our data sets were not large enough to physiologically classify and identify all the cell types reported thus far [50,51]. Direction-selectivity (DS) and orientation-selectivity (OS) indices were calculated by projecting the responses to the moving grating stimuli in eight different directions onto a complex exponential: , where and are the angle of the -th direction and the cell’s corresponding responses, respectively; and and for the DS and OS indices, respectively (S9 Fig). Static nonlinear gain function of each cell (bin size, 0.1) was computed as: P(response | stimulus) = N(stimulus | response)/N(stimulus)/Δt, where N(stimulus | response) and N(stimulus) are the distributions of spike-triggered stimulus ensembles projected onto the L2-normalized STA and the entire stimulus ensembles, respectively. The ratio of P(response | stimulus) before and after the treatment was used as a measure of response gain modulation for each cell. Behavioral data analysis For all experiments, the animal’s left eye—i.e., the side presented with visual stimuli—was recorded using an IR camera (The Imaging Source, DMK23UV024) to monitor its motion and pupil dynamics at 30–60 Hz. Pupil detection was done using a Mask-Region based Convolutional Neural Network, trained as described previously [29]. A two-dimensional ellipse was fit to the pupil, and the large axis diameter was used as a measure of the pupil size, normalized by the median value before treatment for each animal (Fig 4A). To examine the pupil size effects on the response dynamics, a threshold at 33 and 66 percentiles was used to define constricted, neutral, and dilated pupil period, respectively. To examine the effects of the pupil dynamics, we took the time derivative of the pupil size dynamics, and set a threshold at ±0.01 point per second to define constricting, stable, and dilating pupil periods, respectively. We also monitored the running speed of an animal based on the turning speed of the custom-made rotary treadmill throughout the recordings. We set a threshold at 2 cm/s to detect running behavior (S3A Fig). For quantification, we measured the fraction of the running period (S3B Fig), and the median running speed when mice moved at >2 cm/s before and after each treatment (S3C Fig). Model analysis To explore how different aspects of neuronal intrinsic properties and output responses are interrelated between each other, we simulated a model neuron’s responses to white-noise stimuli using different model parameter values (Fig 5A), and characterized these simulated responses using a linear-nonlinear cascade model as we did for the experimental data. Specifically, we employed an integrate-and-fire neuron model with Poisson spiking and refractoriness. A model neuron’s discrete linear filter (of length ) was given by a trigonometric function with logarithmic scale in time: i.e., , where is a temporal spacing parameter and for . The spike-generating function was then modeled by first projecting the white-noise Gaussian stimulus (of length 1,000,000; normalized to have a maximum value of 1) onto , followed by the addition of baseline and the multiplication by gain : i.e., . For each simulation, we chose the value of from to in steps of ; and that of from with . The spiking response was then generated by thresholding (at a random threshold for every time point derived from a uniform distribution from 0 to 1) and refractoriness (given by an exponential filter: for with ). The output spike trains were then subject to the reverse-correlation analysis as described above to characterize the apparent linear filter and static nonlinearity of the model neuron, much as we did for the experimentally recorded RGCs and LGN cells. Specifically, we focused on the peak latency and frequency of the estimated linear filters and the slope of the estimated static nonlinear gain function, and examined how these features depended on the baseline and the intrinsic gain of the model neuron to provide a minimal phenomenological explanation to our observations (Fig 5B–5E). Animals Animals were housed on a 12 h light-dark cycle, with ad libitum access to water and food. All experiments were performed on female mice. A total of 22 wild-type (C57BL6/J; RRID:IMSR_JAX:000664) and 12 hemizygous HDC-Cre (Hdctm1.1(icre)Wwis/J; RRID:IMSR_JAX:021198) mice were used for pharmacological and chemogenetic manipulations, respectively. Additional 9 hemizygous HDC-Cre mice were used for anterograde tracing of HDC+ cells in TMN. Surgical procedures All surgical procedures were performed in animals from 4 to 19 weeks old (9.2 weeks median age). Before the surgery, animals were injected with Carprofen (5 mg/kg) and then anaesthetized with isoflurane (4% for induction, 1% for maintenance in O2). Throughout the surgery, the temperature of the animals was kept stable at 37 °C using a heating pad (Supertech Physiological). Ointment (VitA-Pos, Ursapharm) was applied on both eyes to prevent them from drying during the surgery, and the animals were placed in a stereotaxic frame (Stoelting 51625). A portion of the scalp was removed to expose the skull, and the periosteum was scraped away with a round scalpel to increase adherence of the dental cement. For those mice belonging to the chemogenetic manipulation group, the skull was aligned to level bregma and lambda on the same horizontal plane, and a 100 µm craniotomy was performed using a motorized drill attached to the stereotaxic arm. Viral solution (250 nL; AAV9::FLEX-PSAM Y115F,L141F:5HT3 HC; Vector Biolabs, VB773) was injected with an automatic injector pump at 1 nL/sec, at [−2.6, ±0.8, −5.3] or [−1.7, ±0.5, −5.3] in [Anterior–Posterior (AP), Medial–Lateral (ML), Dorsal–Ventral (DV)] coordinates for targeting TMN and non-TMN areas of the hypothalamus, respectively. All animals were then implanted with a custom-made titanium head plate, cemented on the skull using a mixture of cyanoacrylate (Loctite 401, Henkel) and dental cement (Paladur, Kulzer). The head plate featured a hole (8 mm diameter), allowing for clear observation of both bregma and lambda. The skull surface was then glazed with a thick layer of cyanoacrylate to support the skull with mechanical, atmospheric, and biological protection, while still allowing for visual identification of reference points. After the surgery animals were allowed to recover on a heating pad and then housed in individual cages. During the following seven days, the mice were administered with analgesia (Carprofen; 50 mg/mL) diluted in the drinking water. Anterograde tracing For anterograde tracing of HDC+ cells, we injected recombinant AAVs carrying Cre-dependent fluorescent markers (100 nL; rAAV9::CAG-FLEX-Synaptophysin-GFP or AAV2/1::CAG-FLEX-axon-GFP) into the TMN of HDC-Cre mice (n = 9) as described above. At least 10 days after the injection, the animals were anesthetized (2.5% Avertin, 16 μL/g, intraperitoneal injection) and perfused with paraformaldehyde (PFA; 4% in phosphate buffer solution). The brain tissues including the optic chiasms and optic nerves were dissected and post-fixed overnight in 4% PFA at 4 °C. Coronal sections of the brain tissue (thickness, 80 μm) were then examined under a laser scanning confocal microscope (Leica TCS SP5; S2 Fig). In vivo electrophysiology After recovery from the surgery, animals were habituated to head fixation by placing them in the experimental apparatus over the course of one week, twice a day for up to two hours. On the day of the recording, we first placed the subject animal in the recording set up with its head fixed, and determined the electrode penetration path to the target area (OT, [−1.34, +1.87, +4.74], [−1.70, +1.87, +4.74], or [−1.82, +2.35, +4.07] in [AP, ML, DV] coordinates; LGN, [−2.3, +2.3, +2.8]) using the robotic stereotaxic system (StereoDrive, NeuroStar). The animal was then briefly anesthetized (with isoflurane for about 5 min) and a hole was drilled around the electrode entry point on the skull. After the removal of the anesthesia, an acute silicone probe (Buzsaki32L, Neuronexus; or P2, Cambridge Neurotech) coated with a fluorescent dye (DiI stain, Invitrogen, D282) was lowered at 5 µm/s using the robotic arm until visual responses were found in the target area. All recordings were done during the daytime (between 11 am/ZT4 and 4 pm/ZT9; 1 pm/ZT6 ±1 h, mean ± SD; 96% of the paired recordings were conducted within 2 h of each other). After presenting a battery of visual stimuli, we briefly anesthetized the animal with isoflurane and intraperitoneally injected 0.9% saline solution alone or that containing chlorphenamine (H1 antagonist, 5 mg/kg) [39,40], ciproxifan (H3 antagonist, 12 mg/kg) [41,42], or PSEM (5 mg/kg) [38]. About 20 min after the injection and removal of the anesthesia, we presented the same battery of visual stimuli to the animal for the second recording session. Here, we did not target the H2 or H4 receptors because blocking the H2 receptors has little effect on retinal physiology ex vivo [6], and the functional role of the H4 receptors in the central nervous system remains unclear [9]. In our electrophysiology experiments, we could not keep the same cells long enough to test if the effects of these manipulations were reversible or not. After all the recording sessions, the electrode position was verified histologically (e.g., Figs 1B, LGN and 3B, OT). After retracting the silicon probe, the mice were anesthetized (2.5% Avertin, 16 μL/g, intraperitoneal injection) and perfused with 4% PFA. The brain tissue was harvested and post-fixed overnight in 4% PFA at 4 °C. Coronal sections of the brain tissue (thickness, 100–150 μm) were then examined under a fluorescence microscope (Leica, LMD7000 with N2.1 filter cube) to visualize the trace left by the DiI stain on the probe. To examine the PSAM expression, staining of the slices was performed (S2A, and S2B Fig): primary staining, 1-h incubation with an alpha-bungarotoxin/biotin-XX conjugate (ThemoFisher, B1196) diluted with PBS with 0.1% Tween (1/500) at room temperature; secondary staining, streptavidin and alexa-fluorophore, 1/500 in PBS(T) 0.1% for 2 hours at room temperature. Visual stimulation Visual stimuli were presented as described before [29]. Briefly, visual stimuli were projected onto a spherical screen (20 cm in radius) painted with ultra-violet (UV) reflective white paint, placed 20 cm from the mouse’s left eye. A gamma-corrected digital light processing device (DLP, Texas Instruments, DLPDLCR3010EVM-LC) was used as a light source after the green and the red light-emitting diodes (LEDs) were replaced with UV (365nm, LZ1-00UV00, LED Engine) and infrared (IR; 950nm, SFH 4725S, Osram) LEDs, respectively. The IR light was used as a synchronization signal and recorded with a photodiode (PDA100A2, Thorlabs). Stimulation was conducted at 60 frames per second (FPS), and covered 73° in azimuth and 44° in altitude from the mouse eye position. The maximum light intensity at the eye position was 31 mW/m2 (15.4 mW/m2 for UV LED and 15.9 mW/m2 for blue LED; measured with S121C sensor, Thorlabs), leading to mesopic to photopic conditions. We presented the following stimuli using QDSpy: a randomly flickering full-field stimulus (5 min), consisting of a random sequence of black and white frames at 60 FPS; a black-and-white binary dense noise stimulus (15 min), consisting of a 32-by-18 pixels checkerboard, where each pixel followed an independent random sequence at 60 FPS but the overall luminance of the frame was kept constant at the mean value; a moving grating stimulus in eight different directions (spatial frequencies of square waves, 3° or 20°; moving speed, 7.5°/s or 15°/s); full-field contrast-inverting stimuli in an “OFF-ON-OFF” sequence at maximum contrast (2 s each), followed by a sinusoid (1.5 Hz) with a linearly increasing amplitude from 0% to 100% contrast over 10 s (10 trials) or a sequence of sinusoids with different temporal frequencies (1.875, 3.75 and 7.5 Hz, each for 2 s; 15 and 30 Hz for 1 s) at maximum contrast (10 trials), with an inter-trial interval of 1 s (gray screen). Electrophysiology data analysis We adapted previously established methods of spiking sorting and OT data analysis [29]. In brief, we first concatenated the raw data from the same animal before and after treatment into a single binary file, and used Kilosort 2.0 to sort spikes with a set of default parameters, except for the spike detection threshold to be 6 during optimization. Single units were then identified by clustering in principal component space using Phy for visualization and manual data curation. Only those units that maintained the average spike waveforms and autocorrelograms with a minimal refractory period of 1 ms were kept for subsequent analyses (S1 Fig). Specifically, we selected RGCs/LGN cells as those with robust visual responses: i.e., , where is the response during the “ON-OFF” part of the OFF-ON-OFF stimulus sequence, indicates the mean over trials, and the variance over time (bin size, 1/60 ms), respectively; and the estimated temporal filter should have in at least one time bin within 200 ms from the spike onset (see below for details). Single-units with little or no visual responses were excluded from the analysis as nontarget cells, such as the axons from the parabigeminal nucleus in the OT [81,82]. In total, we obtained 337 RGCs from 20 animals and 557 cells in LGN from 14 mice. These RGCs typically had a triphasic spike waveform as expected for axonal signals [83]. To characterize the peak latency and firing rate of the responses to full-field contrast-inverting stimuli (Figs 1 and S6), we first identified the cell’s response preference to stimulus polarity using an ON–OFF index defined as , where and are the mean firing rate during the ON and the second OFF periods of the OFF–ON–OFF stimulus sequence. We then computed the peri-stimulus time histogram (bin size, 1 ms), smoothed it with a Gaussian filter (kernel width, 6 SD), and identified the first peak upon stimulus onset or offset for those cells with positive or negative ON–OFF index values, respectively. We then performed pairwise comparisons on the peak latencies and firing rates before and after each treatment (e.g., Fig 1F–1J). To compare the changes in the detected peak latency and firing rates across different conditions (e.g., Fig 1K and 1L), we performed Kruskal–Wallis test on a modulation index defined as , where is either the detected peak latency or peak firing rate before and after treatment. For systematically characterizing the visual response properties, we used stimulus ensemble statistical techniques (reverse correlation methods; 500 ms window; Δt = 1/60 s bin size) to calculate the linear filter and static nonlinear gain function of the recorded cells in response to white-noise stimuli as described before (Fig 2B) [29]. Briefly, we first obtained the linear filter of each cell by calculating a STA of the stimulus with ±1 being “white” and “black,” respectively, under a given behavioral state (e.g., when the animal stayed stationary below 2 cm/s). As a quality measure, p-value was computed for each time bin against a null hypothesis that the STA follows a normal distribution with a mean of zero and a variance of , where is the total number of spikes. As a measure of the cell’s response kinetics, we then estimated the peak latency by fitting a difference-of-Gaussian curve to the linear filter; and the spectral peak frequency by the Fourier analysis on the linear filter. The ON–OFF polarity index of a temporal filter was defined as the difference of its peak and valley, divided by the sum of the two (Figs 4G and S10D). Spatial response properties were examined for STAs from “checkerboard” stimulation, using the time window where the eye position remained within 1 inter-quartile range in both horizontal and vertical coordinates for at least 2 s (S5C and S8A Figs) to minimize the effects of eye movements while retaining enough data for RF estimation. In particular, we fitted a two-dimensional Gaussian envelope to the spatial filter at the peak latency (e.g., S5D Fig), and the RF size was estimated as twice the mean SD of the long and short axes. Modulation index was used to characterize the change in a response feature before and after treatment, such as the peak latency and frequency, the mean firing rate, and the RF size (Figs 2K–2N and 3K–3N). Here we focused only on those data when the mouse stayed stationary (<2 cm/s) to minimize the side effects of locomotion behavior (S3 Fig) [26,43]. The linear temporal filters from full-field white-noise stimuli were used for broadly classifying the response types of RGCs and LGN cells, respectively (S4A, S4B, S7A, and S7B Figs). Specifically, we first used t-Distributed Stochastic Neighbor Embedding to map the STAs onto two-dimensional space, and then used K-means++ algorithm for heuristically categorizing the responses into the following six types: fast ON, ON, slow ON, fast OFF, OFF, and slow OFF. Our data sets were not large enough to physiologically classify and identify all the cell types reported thus far [50,51]. Direction-selectivity (DS) and orientation-selectivity (OS) indices were calculated by projecting the responses to the moving grating stimuli in eight different directions onto a complex exponential: , where and are the angle of the -th direction and the cell’s corresponding responses, respectively; and and for the DS and OS indices, respectively (S9 Fig). Static nonlinear gain function of each cell (bin size, 0.1) was computed as: P(response | stimulus) = N(stimulus | response)/N(stimulus)/Δt, where N(stimulus | response) and N(stimulus) are the distributions of spike-triggered stimulus ensembles projected onto the L2-normalized STA and the entire stimulus ensembles, respectively. The ratio of P(response | stimulus) before and after the treatment was used as a measure of response gain modulation for each cell. Behavioral data analysis For all experiments, the animal’s left eye—i.e., the side presented with visual stimuli—was recorded using an IR camera (The Imaging Source, DMK23UV024) to monitor its motion and pupil dynamics at 30–60 Hz. Pupil detection was done using a Mask-Region based Convolutional Neural Network, trained as described previously [29]. A two-dimensional ellipse was fit to the pupil, and the large axis diameter was used as a measure of the pupil size, normalized by the median value before treatment for each animal (Fig 4A). To examine the pupil size effects on the response dynamics, a threshold at 33 and 66 percentiles was used to define constricted, neutral, and dilated pupil period, respectively. To examine the effects of the pupil dynamics, we took the time derivative of the pupil size dynamics, and set a threshold at ±0.01 point per second to define constricting, stable, and dilating pupil periods, respectively. We also monitored the running speed of an animal based on the turning speed of the custom-made rotary treadmill throughout the recordings. We set a threshold at 2 cm/s to detect running behavior (S3A Fig). For quantification, we measured the fraction of the running period (S3B Fig), and the median running speed when mice moved at >2 cm/s before and after each treatment (S3C Fig). Model analysis To explore how different aspects of neuronal intrinsic properties and output responses are interrelated between each other, we simulated a model neuron’s responses to white-noise stimuli using different model parameter values (Fig 5A), and characterized these simulated responses using a linear-nonlinear cascade model as we did for the experimental data. Specifically, we employed an integrate-and-fire neuron model with Poisson spiking and refractoriness. A model neuron’s discrete linear filter (of length ) was given by a trigonometric function with logarithmic scale in time: i.e., , where is a temporal spacing parameter and for . The spike-generating function was then modeled by first projecting the white-noise Gaussian stimulus (of length 1,000,000; normalized to have a maximum value of 1) onto , followed by the addition of baseline and the multiplication by gain : i.e., . For each simulation, we chose the value of from to in steps of ; and that of from with . The spiking response was then generated by thresholding (at a random threshold for every time point derived from a uniform distribution from 0 to 1) and refractoriness (given by an exponential filter: for with ). The output spike trains were then subject to the reverse-correlation analysis as described above to characterize the apparent linear filter and static nonlinearity of the model neuron, much as we did for the experimentally recorded RGCs and LGN cells. Specifically, we focused on the peak latency and frequency of the estimated linear filters and the slope of the estimated static nonlinear gain function, and examined how these features depended on the baseline and the intrinsic gain of the model neuron to provide a minimal phenomenological explanation to our observations (Fig 5B–5E). Supporting information S1 Fig. Spike waveform and autocorrelation of spike trains of representative cells across different conditions. (A) Average spike waveform (top) and auto-correlogram (bottom) of two representative LGN cells (left and right) before (black) and after (red) chemogenetic activation of TMN HDC+ cells. (B–D) Corresponding data for ciproxifan (B, green), chlorphenamine (C, orange), and saline administration (D, blue), each with two representative LGN cells. (E–I) corresponding data for representative RGCs from optic tract recordings (E, chemogenetic activation of HDC+ cells in TMN; F, chemogenetic activation of HDC+ cells outside TMN; G, ciproxifan; H, chlorphenamine; I, saline). Data and code underlying this figure are available at https://doi.org/10.5281/zenodo.17016431. https://doi.org/10.1371/journal.pbio.3003406.s001 (TIF) S2 Fig. Pharmacological and chemogenetic approaches to manipulate the central histaminergic system in the mouse brain. (A,B) Viral delivery of PSAM 5HT3HC channel to HDC+ cells in the TMN (marked with dotted lines) of the posterior hypothalamus (A) or those in the anterior hypothalamus (B; control). (C–E) Histological image examples of anterograde tracing of HDC+ cells in TMN. Labeled axons were found in the optic chiasm (C) or the optic nerve (D) after injecting rAAV9::CAG-FLEX-Synaptophysin-GFP or AAV2/1::CAG-FLEX-axon-GFP, respectively, in the TMN of HDC-Cre mice. However, no visible signal was detected in the isolated retinal tissue of all animals examined (n = 9; see panel E for example). Data and code underlying this figure are available at https://doi.org/10.5281/zenodo.17016431. https://doi.org/10.1371/journal.pbio.3003406.s002 (TIF) S3 Fig. Locomotion of an animal facilitates the visual responses of LGN and RGCs. (A) Probability distribution of an animal’s locomotion speed during recordings before (gray) and after treatment: from left to right, chlorphenamine (CP, orange), ciproxifan (CXN, green), PSAM/PSEM for HDC+ cells in non-TMN (purple) or TMN (red), and saline (blue). (B) The fraction of running period (>2 cm/s) before and after each treatment. No significant change was observed (p = 0.07, Kruskal-Wallis test). (C) The median locomotion speed (during which animals moved at >2 cm/s) before and after each treatment. No significant change was observed (p = 0.3). (D) Comparison of the LGN population response properties between stationary and running periods (n = 104 from 5 animals with a running period ranging between 20% and 80%): from left to right, peak latency (50 ± 19 ms versus 48 ± 19 ms, median ± median absolute deviation; p < 0.001, Wilcoxon signed-rank test), peak frequency (8.2 ± 1.9 Hz versus 8.2 ± 3.0 Hz; p = 0.4), mean firing rate (11 ± 6 Hz versus 20 ± 8 Hz; p < 0.001). (E) Corresponding data for RGCs (n = 44 from 5 animals with a running period ranging between 20% and 80%): from left to right, peak latency (58 ± 19 ms versus 49 ± 18 ms; p < 0.001), peak frequency (8.2 ± 2.8 Hz versus 9.2 ± 2.7 Hz; p = 0.6), mean firing rate (24 ± 22 Hz versus 23 ± 22 Hz; p = 0.5). Data and code underlying this figure are available at https://doi.org/10.5281/zenodo.17016431. https://doi.org/10.1371/journal.pbio.3003406.s003 (TIF) S4 Fig. Functional classification of LGN visual response types and batch effect analysis. (A) t-Distributed Stochastic Neighbor Embedding (t-SNE) embedding of the STAs of LGN cells. Different markers and shadings are used for distinct response categories (shown in B). (B) Each panel represents one of the six response categories: slow off, off, fast off, fast on, on, and slow on. In each panel, each row represents a cell’s STA (color-coded with red and blue hue, indicating positive and negative filter values, respectively); and the overlaid gray line shows the average STAs in each response type. (C–F) Modulation indices on LGN response characteristics across individual animals: from left to right, peak latency (C), peak frequency (D), mean evoked firing rate (E), and nonlinearity (F). The effects of histamine were largely consistent across animals. Data and code underlying this figure are available at https://doi.org/10.5281/zenodo.17016431. https://doi.org/10.1371/journal.pbio.3003406.s004 (TIF) S5 Fig. Histamine did not affect LGN RF size. (A, B) Blink (A) and saccade (B) frequencies before and after different treatments: from left to right, chlorphenamine (CP), ciproxifan (CPN), PSAM/PSEM for non-TMN HDC+ cells and TMN HDC+ cells, and saline. None of these changes were statistically significant: p = 0.10, Kruskal–Wallis test on blink frequency changes; p = 0.15, saccade frequency changes. (C) Representative time series of the eye position (X- and Y-coordinates of the pupil center extracted from eye-tracking camera images) during white-noise “checkerboard” stimulation (black/red, centered and stable eye position used for reverse correlation; gray, non-centered or non-stable period excluded from the analysis; probability distribution shown on the right), before (left) and after (right) chemogenetic activation of TMN HDC+ cells. (D) Estimated spatial filter (receptive field; RF) of two example LGN cells before (top) and after (bottom) chemogenetic activation of HDC+ cells in TMN. Note a shift of RF position of all recorded cells due to a shift of the animal’s resting eye position after the treatment (see panel C). (E–H) Cumulative distribution of the modulation index of LGN cells before and after treatment (E, receptive field size; F, peak latency; G, peak frequency; H, mean evoked firing rate): orange, chlorphenamine, n = 32 from 3 animals; green, ciproxifan, n = 61 from 4 animals; red, PSAM/PSEM for HDC+ cells in TMN, n = 20 cells from 3 animals. Post-hoc test against saline control (n = 47 from 3 animals) after Kruskal–Wallis test: * p < 0.05; ** p < 0.01; *** p < 0.001. Data and code underlying this figure are available at https://doi.org/10.5281/zenodo.17016431. https://doi.org/10.1371/journal.pbio.3003406.s005 (TIF) S6 Fig. Histamine primarily modulated the visual response kinetics of RGCs. (A–E) Visual responses of a representative RGC to full-field contrast-inverting stimuli (2 s intervals) before and after chemogenetic (A, HDC+ cells in TMN; B, HDC+ cells in non-TMN) or pharmacological treatment (C, ciproxifan; D, chlorphenamine; E, saline); top, spike raster across trials; bottom, zoom-in of the spike raster around stimulus onset or offset (−50 to 250 ms, red shade on top) and peri-stimulus time histogram. (F–J) Pairwise comparison of the RGC population responses before and after treatment (top, peak latency; bottom, peak firing rate): * p < 0.05; ** p < 0.01, Wilcoxon signed-rank test with Bonferroni correction. F, PSAM/PSEM for TMN HDC+ cells: 57 ± 15 ms versus 66 ± 17 ms peak latency, p = 0.005; 122 ± 54 Hz versus 112 ± 66 Hz peak frequency, p = 0.08; median ± median absolute deviation, n = 16 RGCs from 4 animals. G, PSAM/PSEM for non-TMN HDC+ cells: 42 ± 17 ms versus 43 ± 14 ms peak latency, p = 0.3; 46 ± 30 Hz versus 59 ± 40 Hz peak frequency, p = 0.3; n = 24 RGCs from 3 animals. H, ciproxifan: 42 ± 31 ms versus 53 ± 31 ms peak latency, p = 0.002; 72 ± 80 Hz versus 78 ± 78 Hz peak frequency, p = 0.3; n = 32 RGCs from 4 animals. I, chlorphenamine: 61 ± 16 ms versus 47 ± 18 ms peak latency, p = 0.026; 55 ± 60 Hz versus 102 ± 88 Hz peak frequency, p = 0.05; n = 22 RGCs from 3 animals. J, saline: 73 ± 30 ms versus 73 ± 31 ms peak latency, p = 0.5; 92 ± 50 Hz versus 92 ± 56 Hz peak frequency, p = 1; n = 37 RGCs from 3 animals. (K,L) Cumulative distributions of the modulation index before and after each treatment (in corresponding colors): K, peak latencies; L, peak firing rate; * p < 0.05 from the post-hoc test against the saline (control) condition on the average group ranks (Kruskal–Wallis test). Data and code underlying this figure are available at https://doi.org/10.5281/zenodo.17016431. https://doi.org/10.1371/journal.pbio.3003406.s006 (TIF) S7 Fig. Functional classification of RGC visual response types and batch effect analysis. (A) t-SNE embedding of the RGC STAs. Different markers and shadings are used for distinct response categories (shown in B). (B) Each panel represents one of the six response categories: slow off, off, fast off, fast on, on, and slow on. In each panel, each row represents a cell’s STA (color-coded with red and blue hue, indicating positive and negative filter values, respectively); and the overlaid gray line shows the average STAs in each response type. (C–F) Modulation indices on RGC response characteristics across animals: from left to right, peak latency (C), peak frequency (D), mean firing rate (E), and nonlinearity (F). The effects of histamine were generally consistent across animals, and no substantial batch effect was observed. Data and code underlying this figure are available at https://doi.org/10.5281/zenodo.17016431. https://doi.org/10.1371/journal.pbio.3003406.s007 (TIF) S8 Fig. Histamine did not affect RGC RF size. (A) Representative time series of the eye position (X- and Y-coordinate of the pupil center extracted from eye-tracking camera images) during white-noise “checkerboard” stimulus presentation (black/green, centered and stable eye position used for reverse correlation analysis; gray, non-centered or non-stable period excluded from the analysis; probability distribution shown on the right), along with those of firing rate dynamics of two example RGCs and locomotion (from top to bottom), before (left) and after (right) ciproxifan administration. (B) Estimated spatial filter (receptive field) of the two example RGCs (top and middle) before (left) and after (right) ciproxifan administration. Note a shift of RF position of all recorded cells due to a shift of resting eye position after the treatment (see panel A). (C–F) Cumulative distribution of the modulation index of RGCs before and after treatment (C, receptive field size; D, peak latency; E, peak frequency, F, mean firing rate): orange, chlorphenamine, n = 6 cells from 3 animals; green, ciproxifan, n = 25 cells from 4 animals; purple, PSAM/PSEM for HDC+ cells in non-TMN, n = 9 cells from 3 animals; red, PSAM/PSEM for HDC+ cells in TMN, n = 8 cells from 4 animals. Post-hoc test against saline control (n = 17 from 3 animals) after Kruskal–Wallis test: * p < 0.05; ** p < 0.01. Data and code underlying this figure are available at https://doi.org/10.5281/zenodo.17016431. https://doi.org/10.1371/journal.pbio.3003406.s008 (TIF) S9 Fig. Histamine did not affect direction- or orientation-selectivity of RGCs or LGN cells. (A) The average firing rate of a representative direction-selective (DS) RGC in response to moving gratings in eight different directions before (gray) and after treatment (color-coded): from left to right, PSAM/PSEM for TMN HDC+ cells (red) or non-TMN cells (purple), ciproxifan (green), chlorphenamine (orange), and saline (blue). (B) DS indices of RGC populations before and after treatment. Those with DS index >0.15 in either condition were highlighted in the corresponding color (from left to right, n = 11, 16, 31, 14, and 30 RGCs, respectively). No significant change was observed in the DS index values: p > 0.2 in all cases (Wilcoxon signed-rank test). (C) Corresponding data for the OS indices of RGCs. No significant change was observed (p > 0.7 in all cases). (D,E) Corresponding population data for the DS and OS indices of LGN cells (from left to right, n = 42, 76, 27, 58 LGN cells; p > 0.06 and 0.3 in all cases, respectively). Data and code underlying this figure are available at https://doi.org/10.5281/zenodo.17016431. https://doi.org/10.1371/journal.pbio.3003406.s009 (TIF) S10 Fig. Histaminergic effects on pupil dynamics are irrelevant to those on LGN visual responses. (A–C) Comparison of the LGN population response properties between constricted and dilated pupil periods (n = 350 RGCs from 13 animals): A, peak latency, 51 ± 20 ms versus 48 ± 19 ms, median ± median absolute deviation, p < 0.001, Wilcoxon signed-rank test; B, peak frequency, 7.9 ± 2.1 Hz versus 8.1 ± 2.4 Hz, p = 0.004; C, mean firing rate, 11 ± 7 versus 15 ± 7 Hz, p < 0.001. (D–G) Comparison of the LGN population response properties between constricting and dilating pupil periods: D, ON-OFF polarity index of the temporal filters, 0.07 ± 0.22 versus −0.06 ± 0.27, p < 0.001; E, peak latency, 49 ± 19 ms versus 49 ± 19 ms, p = 0.8; F, peak frequency, 7.9 ± 2.0 Hz versus 8.1 ± 2.1 Hz, p = 0.3; G, mean firing rate, 11 ± 7 Hz versus 14 ± 7 Hz, p = 0.004. (H) Comparison of peak latencies before treatment with constricted pupil versus after treatment with dilated pupil: from left to right, PSAM/PSEM for TMN HDC+ cells (red), ciproxifan (green), chlorphenamine (orange), and saline (blue): ** p < 0.01; *** p < 0.001, Wilcoxon signed rank test. (I) Comparison of peak latencies before treatment with dilated pupil versus after treatment with constricted pupil. Data and code underlying this figure are available at https://doi.org/10.5281/zenodo.17016431. https://doi.org/10.1371/journal.pbio.3003406.s010 (TIF) Acknowledgments The EMBL Light Imaging Facility is acknowledged for support in histological image acquisition and analyses; EMBL Gene Editing and Virus Facility for virus production; EMBL IT Support for provision of computer and data storage servers; and the LAR facility for taking care of animals. We thank Dmitry Molotkov, Tom Boissonnet, Vanesa Pelcastre, Sahana Rao, Shabnam Chandel, Lily Knowles for their help in experiments, and all the Asari lab members for many useful discussions.

Ji, Miao-Jin;Wu, Tong-Xuan;Tian, Chenhao;Cao, Xiang;Wei, Ruyuan;Yang, Yin-Yin;Meng, Xinran;Tang, Huanyao;Cui, Tiantao;Yang, Jiao;Tang, Xin;Liu, Chao

doi: 10.1371/journal.pbio.3003513pmid: 41284713

Introduction Amyloid precursor protein (APP) is a type I transmembrane protein processed by secretases to generate β-amyloid peptides and other fragments. While APP is best known for its central role in Alzheimer’s disease pathogenesis, its physiological functions remain incompletely understood. Recent studies have demonstrated that APP plays critical roles in neuronal development [1,2], synaptic function [3], and intracellular signaling [4]. These processes influence a wide range of central nervous system-regulated functions, including learning, memory [5–7], and motor functions [8–10]. The APP-null mice exhibited impaired motor performance, learning capacity, and poor memory [5–7]. While the role of APP in cognitive impairment has been extensively studied, its involvement in motor control remains less explored. Interestingly, in contrast to other movement disorder models, APP-null mice specifically display deficits in locomotor activity and grip strength [8,9]. Initial hypotheses suggested that APP loss disrupts neuromuscular junctions. However, APP single-knockout mice exhibited no significant alterations in neuromuscular transmission or synaptic morphology at the neuromuscular junctions [11], suggesting that APP may instead regulate higher-order motor control circuits rather than the function of neuromuscular junctions. In mammals, voluntary movement is orchestrated by the corticospinal tract, a motor pathway connecting the motor cortex to lower motor neurons and involving multiple regions, including the motor cortex, subcortical nuclei, cerebellum, brainstem, and spinal cord [12]. Conditional APP knockout under the Nex promoter—driving Cre expression in corticospinal tract-associated regions, including the motor cortex, superior and inferior colliculi, medulla oblongata, pons, cerebellar granule cell layer, and deep cerebellar nuclei (DCNs) [13], does not impair grip strength or locomotor activity [7], suggesting that APP in these brain regions is unlikely to account for its role in motor function regulation. The cerebellum is a crucial subcortical region involved in motor control, comprising the cerebellar cortex, DCNs, and medulla. Purkinje cells (PCs) in the cerebellar cortex receive input from mossy fibers and climbing fibers, integrating these signals and projecting to DCN neurons. As the sole output neurons of the cerebellar cortex, PCs are critical for encoding motor function. In humans, cerebellar activity increases linearly with increasing grip force in the ipsilateral limb [14]. Patients with cerebellar lesions display muscle force production deficits [15]. The action potential (AP) firing of PCs is spontaneous and of high frequency, dependent on voltage-gated sodium channels. In the cerebellar cortex, Nav1.6 and Nav1.1 are predominant in PCs, while Nav1.2 is primarily expressed in cerebellar granule cells [16], as demonstrated by morphological [17] and electrophysiological studies [18]. Nav1.6 accounts for ~70% of persistent sodium currents and ~82%–92% of resurgent sodium currents in PCs [18]. Three types of sodium currents are typically studied in PCs: resurgent currents, responsible for the generation of spontaneous firing; transient sodium currents, contributing to the upstroke of APs, and persistent sodium currents, which regulate plateau potentials and amplify dendritic input [19]. APP is abundantly expressed in cerebellar PCs [20]. Our prior study showed that APP increases Nav1.6 currents by enhancing its cell surface localization in vitro [21]. However, whether APP regulates the physiological function of cerebellar PCs and participates in motor control remains unclear. In this study, we investigated the role of APP in cerebellar PCs by analyzing the motor function deficits, PC firing patterns, and cerebellar circuit dysfunctions in APP-null mice. We found that APP in cerebellar PCs is critical for PC firing through the regulation of Nav1.6 and plays an essential role in PC-mediated motor control. Results Cerebellar PC APP mediates motor functions in mice Consistent with previous study [8,9], we observed that APP-null (App−/−) mice exhibited reduced locomotor activity (S1A Fig) and weaker forelimb grip strength (S1B Fig) compared with wild-type (WT, App+/+) littermates. Further, animals were tested for their capacity for motor learning and coordination using a rotarod and footprint test. APP-null mice showed normal motor learning over 3 days of training and their latency to fall was similar to that of WT mice in the test day (S1C Fig). During the footprint test, there was no differences between APP-null and WT mice in sway length and the overlap, while the stride length and the stance length were all significantly lower in APP-null group compared with that of WT (App+/+) group (S1D and S1E Fig). The reduced stride length in footprint test (S1E Fig) likely reflects smaller body size in APP-null mice (S1F Fig). To control for this confounder, we performed the balance beam assay. No significant differences were observed in traversal time or number of errors on 12- or 6-mm beams (S1G Fig), indicating intact gross motor coordination in APP-null mice. Thus, APP deletion specifically impairs open field locomotion and forelimb grip strength, independent of anthropometric differences. APP is abundantly expressed in PCs of adult WT mice (Fig 1A). Given that cerebellar PCs play a critical role in motor function, we hypothesized that the deletion of PC-specific APP could be the primary cause of the motor dysfunction observed in APP-null mice. To specifically evaluate the effect of PC APP on motor performance, we injected a lentivirus encoding human APP695 with mCherry (LV-CAMK2A-SP-Flag-APP695-IRES2-mCherry) into the cerebellar cortex (Fig 1B). The CAMK2A promoter, which has been shown to drive gene expression specifically in PCs [22], was used in this experiment. Expression of mCherry was observed exclusively in the Calbindin-positive PCs (Fig 1B). Laser capture microdissection (LCM) coupled with Reverse Transcription Quantitative PCR (RT-qPCR) analysis further confirmed a significant increase of the APP expression in PCs (Fig 1C). PC-specific expression of APP in APP-null mice significantly improved their performance in locomotor activity (Fig 1D) and forelimb grip strength test (Fig 1E), but didn’t affect their body size (Fig 1F). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Cerebellar Purkinje cell (PC)-specific amyloid precursor protein (APP) expression modulates motor function. (A) APP (red) and calbindin (green) immunofluorescence in 20-µm-thick cerebellar sagittal sections. The APP immunofluorescence in wild-type (App+/+) or APP-null (App−/−) mice were acquired under identical imaging settings. Insets: magnified views. Scale bars: 100 µm (main), 20 µm (insets). The PC contours are indicated by dashed lines. (B) Lentivirus-mediated overexpression of APP695 in the cerebellar cortex of App−/− mice. Representative images showing co-expression of APP/mCherry (red) and the PC marker calbindin (green) in a sagittal section of cerebellum. Scale bars: 100 µm (main); 20 µm (insets). (C) Left: representative fluorescence images showing before or after laser capture microdissection (LCM) of LV-transduced PCs (mcherry+) from fresh-frozen coronal sections. Right: App mRNA levels in dissected cells detected by qPCR (Data presentation: median with interquartile range. App+/+ + vec, n = 8; App−/− + vec, n = 4; App−/− + APP, n = 7 samples; Kruskal–Wallis test). Scale bar: 40 µm. (D–F) Motor phenotyping of App−/− mice with Purkinje APP rescue: (D) Open field total distance (10-min test). (E) Forelimb grip strength. (F) Body weight of all groups of mice. App+/+ + vec, n = 12; App−/− + vec, n = 13; App−/− + APP, n = 12. (G) Conditional knockdown of App in App+/+ mice using a mixed AAV injection strategy. Representative images show the CRE (mCherry+) and APP-shRNA (EGFP+) expression in PCs in a coronal section of the cerebellum. Scale bar: 100 µm (main), 20 µm (insets). The PC contours are indicated by dashed lines. (H) Left: representative fluorescence images before or after LCM of AAV-transduced PCs (mcherry+) from fresh-frozen coronal sections of App+/+ mice. Right: App mRNA levels in dissected cells detected by qPCR (control, n = 5, App kd, n = 10 samples). (I, J) Motor deficits following Purkinje APP knockdown: (I) Open field mobility. (J) Forelimb grip strength decline. (Control, n = 10; App kd, n = 8 mice). Data presentation: mean with SEM. Statistics: (D–F) one-way ANOVA; (H–J) Unpaired t test; * P < 0.05; ** P < 0.01; *** P < 0.001. The data underlying this Figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003513.g001 To further investigate the contribution of APP to motor control, we utilized mice with conditional APP knockdown restricted to PCs. We achieved PC-specific App knockdown by co-injecting WT mouse cerebellar cortex with AAV2/9-L7-Cre [23] and AAV2/9-DIO-(U6-EGFP)-shApp [24], whereby Cre recombinase activates the expression of shRNA targeting App specifically in L7-positive cells. Immunofluorescence confirmed efficient PC-specific transduction mediated by AAV under the L7 promoter, as shown by its colocalization with calbindin (Figs 1G and S2). LCM coupled with RT-qPCR verified dramatic APP downregulation in transduced neurons (Fig 1H). Subsequent behavioral analyses revealed significant motor deficits in conditional knockdown mice, including reduced ambulation in open field tests (Fig 1I) and impaired forelimb grip strength (Fig 1J). Taken together, these results showed that PC-specific APP expression mediates cerebellar-dependent motor performance. Full-length APP is necessary for rescue of motor function deficits in APP-null mice Proteolytic processing of full-length APP generates bioactive fragments including sAPPα (sufficient for spatial learning and long-term potentiation rescue in APP-null mice) [9] and APP intracellular domain (AICD) (with transcriptional activity). To determine whether the intact holoprotein is required for motor functions, we delivered holo-APP, sAPPα, or AICD to APP-null mice via cerebellar lentivirus injection. Fluorescence imaging confirmed targeted expression in PCs (Fig 2A). LCM-coupled RT-qPCR verified successful APP reconstitution in transduced neurons (Fig 2B). Behavioral analyses demonstrated that only full-length APP (holo-APP) significantly rescued motor deficits, exhibiting significant improvement in open field ambulation (Fig 2C) and grip strength (Fig 2D) in APP-null mice. sAPPα and AICD provided no significant rescue (Fig 2C and 2D). These results establish that full-length APP is necessary for cerebellum-mediated motor function. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Full-length APP is required for the rescue of motor deficit in APP-null mice. (A) Lentivirus-mediated overexpression of full-length APP (holo-APP), sAPPα, or AICD in the cerebellar cortex of App−/− mice. Representative images showing co-expression of APP fragments/mCherry (red) and the PC marker calbindin (green) from coronal sections of the cerebellum. Scale bars: 100 µm (main); 40 µm (insets). (B) QPCR analysis of LCMed PCs (mcherry+) from fresh-frozen sections. Top: Primer specificity schematic for APP fragments. Bottom: APP fragments mRNA level normalized to Gapdh. Data presentation: median with interquartile range. App−/− + vec, n = 4; App−/− + holo-APP, n = 5; App−/− + sAPPα, n = 4, App−/− + AICD, n = 4 samples. Kruskal–Wallis test. (C, D) Motor phenotyping of App−/− mice with Purkinje APP fragments rescue (n = 9 mice/group): (C) Open field total distance (10-min test). (D) Forelimb grip strength. N = 9 for each group. Data presentation: mean with SEM. Statistics: one-way ANOVA. ** P < 0.01; ns, non-significant. The data underlying this Figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003513.g002 APP deficiency impairs PC electrophysiological activity APP plays a crucial role in neurodevelopment [1,2], so we examined whether APP deficiency impaired the development of PCs. Golgi staining and Sholl analysis revealed that the dendritic branches and spine density were similar between 2-month-old APP-null and WT littermates (S3A–S3C Fig). Further, we compared the morphology of PC in 12-month-old mice. No significant difference in the dendritic branches and spine density was observed between APP-null and WT mice (S3D–S3F Fig). These results indicated that APP deficiency affects neither PC development nor provokes their degeneration. Next, we investigated whether loss of APP altered the firing patterns of PCs. As established, PCs display three characteristic firing states: silent, simple spikes (SSs), and complex spikes (CSs) [18,25,26]. Most of 22 PCs (19/22, 86.4%) recorded from WT mice exhibited spontaneous firing, while 13.6% (3/22) of PCs maintained silent within 1 h of recording. Complex spikes recorded in 27.3% (6/22) of PCs from WT mice (Fig 3A). The percentage of silent PCs increased to 42.1% (8/19, 42.11%), and no CSs was recorded in 19 PCs from APP-null mice (Fig 3A). We further analyzed the spontaneous firing rate and various AP characteristics, including threshold, peak amplitude, half-width, and afterhyperpolarization potential (AHP) of PCs with SSs in brain slices recorded from APP-null and WT mice. The spontaneous firing rate of PCs in APP-null mice was significantly lower than that in WT mice (Fig 3B). The rheobase of AP in APP-null PC was higher than that in WT mice (Fig 3C). In addition, lower peak amplitude, higher threshold, longer half-width, and bigger AHP (Fig 3D and 3E) were observed in the spontaneous AP spikes of APP-null than those of WT PCs. These results indicate that APP-deficiency significantly impairs the firing properties of cerebellar PCs. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. APP deficiency impairs the normal firing patterns of PCs. (A) Representative spontaneous firing patterns recorded in PCs of wild-type (App+/+) mice. Pie charts in the right panel showing the percentage of PC firing pattern. App+/+, n = 22; App−/−, n = 19 cells. (B) Representative spontaneous simple spikes recorded in PCs. Quantification of firing rates analyzed from 1-min recordings during stable whole-cell access (≥10 min after breakthrough). App+/+, n = 9; App−/−, n = 8. (C) Left: Representative voltage responses to depolarizing ramp current injections (−50 to +500 pA over 1 s) recorded PCs under whole-cell current-clamp. Right: Rheobase quantification (minimal current eliciting first action potential). n = 8 cells/group. (D) Representative spontaneous action potentials and the corresponding dV/dt curve. (E) Statistics of peak amplitude, threshold, half-width, and afterhyperpolarization potential (AHP) of action potentials shown in (D), App+/+, n = 10; App−/−, n = 9. Scale bars are indicated in the images. Data are represented as the mean with SEM. Student t test: * P < 0.05; *** P < 0.001. The data underlying this Figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003513.g003 Reduced Nav1.6, but not Nav1.1-mediated sodium current was detected in APP-null PCs Consistent with APP enhancing Nav1.6 currents in heterologous systems [21], whole-cell recordings in cerebellar slices revealed altered sodium currents in APP-null PCs. Transient (INaT), persistent (INaP), and resurgent (INaR) currents were evoked using established protocols (S4 Fig) [18]. Given that Nav1.6 and Nav1.1 are the predominant voltage-gated sodium channels in mouse PCs [16], we next sought to determine which of these isoforms was affected by APP deficiency. To this end, we applied subtype-specific blockers/inhibitors: 4,9-anhydrotetrodotoxin (4,9-ah-TTX; selective Nav1.6 blocker) and ICA-121431 (Nav1.1 inhibitor) [19]. Voltage steps (−70 to +10 mV) from a holding potential of −90 mV elicited INaT, mediated by Nav1.6 and Nav1.1 (TTX/4,9-ahTTX + ICA-sensitive, S4A and S4D Fig). APP-null PCs showed reduced peak INaT but unchanged voltage-dependence (Fig 4A and 4B). Pharmacological dissection revealed selective reduction of Nav1.6-mediated INaT (Fig 4C), while Nav1.1 currents remained unaffected (Fig 4D). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. APP deficiency selectively reduces Nav1.6-mediated Na+ currents in cerebellar PCs. (A) Representative traces for whole-cell recording of transient Na+ currents (INa) in PCs of App+/+ or App−/− mice. (B) Whole cell I–V curve, peak currents, and activation curves of transient INa as shown in (A). App+/+, n = 8; App−/−, n = 14 cells. (C, D) Representative traces and statistical bar-charts showing amplitude of transient INa (INaT) before or during bath application of (C) Nav1.6-specific blocker 4,9-anhydrotetrodotoxin (4,9-ahTTX, 200 nM) or (D) Nav1.1-specific inhibitor ICA-121431 (ICA, 350 nM). N = 6 cells for each group. The INa1.6 or INa1.1 were determined by subtracting the sodium currents recorded during bath application of specific blockers/inhibitors from that before drug application. (E) Representative persistent Na+ currents recorded in PCs. The right panels show statistics of the peak persistent currents and voltage potentials eliciting peak persistent Na+ currents. App+/+, n = 6; App−/−, n = 12 cells. (F) Representative traces (the left panel) and statistics for whole-cell recording of resurgent INa in PCs. The middle panel: Whole cell I–V curve of resurgent INa; the right panel: peak resurgent currents. App+/+, n = 9; App−/−, n = 11 cells. Student t test; * P < 0.05; ** P < 0.01; ns, nonsignificant. Scale bars are indicated in the images. (G) APP co-immunoprecipitated with Nav1.6 but not Nav1.1. HEK293 cells co-expressing Flag-tagged APP and either Nav1.6 or Nav1.1-EGFP were lysed and immunoprecipitated (IP) with anti-Flag affinity beads (Smart-Lifesciences, #SA042001). Immunoblots (IB) were probed for: Nav1.6 (Alomone Labs #ASC-009), EGFP (Roche Applied Science, #11814460001) or Flag (AlpVHHs, #016-303-005). (H) Nav1.6 interacts with full-length APP (holo-APP), but not soluble APPα (sAPPα) and APP intracellular domain (AICD). HEK293 cells co-expressing Nav1.6 and Flag-tagged holo-APP, sAPPα, or AICD were lysed and IPed with anti-Flag affinity beads. IB were probed for Nav1.6 or Flag. Mock transfected controls are shown in the first lane. Input controls (7.5% lysate) shown below/beside corresponding lanes. Data representative of 3 biological replicates. The data underlying this Figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003513.g004 Slow ramps (−90 to +30 mV, 0.12 mV/ms) evoked INaP, mediated by Nav1.6 in PCs (blocked by TTX/4,9-ahTTX; S4B, S4E, and S5A Figs). APP-null PCs exhibited diminished INaP amplitude without activating voltage shift (Fig 4E). Similarly, INaR (mainly mediated by Nav1.6, evoked by +30 mV and followed by a repolarization to voltage steps between −60 and −10 mV, S4C, S4F, and S5B Figs) was reduced in APP-null PCs with unaltered voltage dependence (Fig 4F). However, Nav1.6 mRNA (Scn8a) levels in PCs did not differ between WT and APP-null mice (S6 Fig), suggesting that APP may not influence Scn8a gene expression. Co-immunoprecipitation (co-IP) assays in transfected HEK293 cells (S7 Fig) confirmed a specific physical interaction between APP and Nav1.6 (Fig 4G). This binding was dependent on holo-APP, as neither the soluble ectodomain sAPPα nor the AICD co-precipitated with Nav1.6 (Fig 4H), consistent with the requirement of holo-APP for functional rescue observed in vivo (Fig 2). Taken together, APP deficiency selectively impairs Nav1.6 but not Nav1.1-mediated sodium currents in PCs through molecular interactions. Nav1.6 mediates PC firing and motor deficits in APP-null mice Our findings align with prior evidence that persistent/resurgent INa primarily depends on Nav1.6, while transient INa involves Nav1.1-Nav1.6 synergy [18,27]. Nav1.6 is critical for PC repetitive firing [18]. Accordingly, 4,9-ahTTX abolished repetitive firing in WT PCs (Fig 5A). Conversely, the Nav1.6 positive allosteric modulator poneratoxin (PoTX, inhibits Nav1.6 inactivation, S8 Fig) [28], restored firing in APP-null PCs (Fig 5B). ICA-121431 showed no effect in WT (Fig 5C) or APP-null PCs (Fig 5D). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. Nav1.6 mediated the deficits of PC firing and motor function in APP-null mice. (A–D) Electrophysiological characterization of PC spontaneous firing in acute cerebellar coronal slices (300-µm-thick). (A) Representative traces and quantification of spontaneous firing rate and spike amplitude before/during application of 200 nM 4,9-ahTTX in App+/+ mice. n = 7 cells. (B) Representative traces and quantification of spontaneous firing rate and spike amplitude before/during application of 30 nM PoTX (Nav1.6 positive allosteric modulator) in App−/− mice. n = 5 cells/group. (C, D) Representative traces and quantification of spontaneous firing rate and spike amplitude before/during application of 350 nM ICA (Nav1.1 inhibitor) in App+/+ (C) or App−/− (D) mice. Statistics: Paired t test; * P < 0.05; ** P < 0.01; *** P < 0.001 vs. baseline; ns, non-significant. n = 4 cells/group. Scale bars are indicated in the images. (E) Morphology verification of cerebellar cannula placement in a coronal section of mice cerebellum. (F) Open field locomotion: Total distance traveled (m) in App+/+ (4,9-ahTTX/vehicle) and App−/− (PoTX/vehicle) mice. Mice received bilateral microinjections of saline (vehicle), 200 nM 4,9-ahTTX, or 20 pM PoTX into the cerebellar cortex 20 min prior to testing. (G) Forelimb grip strength: maximal force/body weight in same cohorts as (F). Statistics (F, G): App+/+ + vehicle, n = 10; App+/+ + 4,9-ahTTX, n = 8; App−/− + vehicle, n = 8; App−/− + PoTX, n = 9; One-way ANOVA with Tukey post-hoc; * P < 0.05; ** P < 0.01, *** P < 0.001; ns, non-significant. ctrl, control; 4,9-ahTTX, 4,9-anhydrotetrodotoxin; ICA, ICA-121431; PoTX, poneratoxin. The data underlying this Figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003513.g005 Intracerebellar infusion studies further implicated the essential role of Nav1.6 in APP-mediated motor function. Histology confirmed cannula placement in cerebellar cortex (Fig 5E). Nav1.6 blocker infusion impaired WT mouse performance in open field and grip-strength tests (Fig 5F and 5G), whereas PoTX rescued motor deficits in APP-null mice (Fig 5F and 5G). These data provide evidence that APP influences PC firing and Motor function through Nav1.6. Exogenous APP expression restored Nav1.6 currents and firing properties of PCs in APP-null mice We next investigated whether the restoration of the behavioral phenotypes in APP-null mice via exogenous APP expression was linked to the improvement in Nav1.6 function in PCs. Whole-cell patch clamp recordings revealed that transient, persistent, and resurgent INa were all rescued by exogenous APP expression in APP-null PCs. The peak transient INa was restored to levels comparable to WT mice (Fig 6A and 6B). APP expression in PCs did not affect the activation curve (Fig 6B). Decreased persistent and resurgent current were also restored by APP expression in PCs (Fig 6C and 6D). These results indicated that exogenous expression of APP in PCs restore the Nav1.6 currents. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 6. Exogenous APP expression restored the Nav1.6 currents and ameliorated the abnormalities in the firing patterns of PCs of APP-null mice. (A, B) Representative traces for whole-cell recording of (A) transient Na+ current and (B) statistics of whole-cell I–V curve, peak transient currents, and activation curves in PCs of App+/+ + vec, App−/− + vec, and App−/− + APP mice. N = 8 cells for each group. (C) Representative persistent Na+ current recorded in PCs. The right panels show statistics of the peak persistent currents and the voltage eliciting the peak currents. App+/+ + vec, n = 6; App−/− + vec, n = 8; App−/− + APP, n = 6 cells. (D) Representative resurgent sodium currents and I–V curves recorded in PCs. App+/+ + vec, n = 8; App−/− + vec, n = 8; App−/− + APP, n = 8 cells. (E) Pie chart showing the percentage of PC firing patterns in App+/+ + vec, App−/− + vec, and App−/− + APP mice. (F) Representative spontaneous simple spikes recorded in PCs. Quantification of firing rates analyzed from 1-min recordings during stable whole-cell access (≥10 min after breakthrough). App+/+ + vec, n = 12; App−/− + vec, n = 9; App−/− + APP, n = 13 cells. (G) Representative membrane potential responses to depolarizing ramp current injections (−50 to +500 pA over 1 s) recorded from PCs under whole-cell current-clamp. Right: Rheobase quantification. N = 9 cells/group. (H) Representative images of spontaneous action potentials and the dV/dt curves. (I) Statistics of peak amplitude, threshold, half-width, and afterhyperpolarization potential (AHP) of spontaneous action potentials shown in (H), App+/+ + vec, n = 10; App−/− + vec, n = 8; App−/−- + APP, n = 8 cells. Scale bars are indicated in the images. One-way ANOVA and post-hoc Turkey’s test. * P < 0.05; ** P < 0.01; *** P < 0.001. The data underlying this Figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003513.g006 In the analysis of firing properties, we found that exogenous APP expression in APP-null PCs restored spontaneous firing patterns nearly to the WT level. Complex spikes were observed in the APP-expressing group (2/21, 9.5%), but not in the control group (Fig 6E). The spontaneous firing frequency of the APP-expressing group was almost restored to that of WT mice (Fig 6F). Additionally, exogenous APP-expression in APP-null PCs restored the rheobase (Fig 6G). Moreover, the spontaneous AP spikes from exogenous APP-expressing PCs displayed higher peak amplitude, lower threshold, shorter half-width, and smaller AHP (Fig 6H and 6I). These data indicate that exogenous APP expression in PCs restored Nav1.6 currents and firing properties in APP-null mice. Exogenous APP expression in PCs restored Nav1.6-dependent GABAergic inputs to DCN neurons in APP-null mice PC projections are the sole output of the cerebellar cortex and represent the major input to DCN neurons [29], forming approximately 70%–80% of all synaptic connections onto DCN cells [30,31]. To explore whether inhibitory transmission from PCs to DCN neurons was affected by APP deficiency, we performed whole-cell patch clamp recordings in cerebellar slices. DCN neurons innervated by PCs were labeled using trans-synaptic labeling, achieved by injecting AAV2/1-hSyn-CRE into the cerebellar cortex and AAV2/9-hSyn-DIO-EGFP in the DCN (Fig 7A). After 3 weeks of viral expression, EGFP-positive DCN neurons were recorded in brain slice (Fig 7B). No significant difference was observed in the number of EGFP-labeled DCN neurons between the different virus-injected groups. The amplitude and the frequency of spontaneous inhibitory post-synaptic currents (sIPSCs) in APP-null DCN neurons were lower than those in WT DCN neurons, and the abnormal GABAergic synaptic transmission APP-null mice was restored by APP expression in PCs (Fig 7C). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 7. Exogenous amyloid precursor protein (APP) expression restored the defects of the sIPSCs evoked by Nav1.6-mediated Purkenji cell inputs in APP-null mice. (A) Schematic of the viral injection strategy to achieve APP overexpression in PCs and trans-synaptic tracing of downstream neurons in the DCN. (B) Representative example of fluorescence-assisted selection of cells for patch-clamp recordings. Images from an acute cerebellar slice observed in differential interference contrast (DIC, left), or fluorescence mode (EGFP, right). (C) Left panel: Representative traces of sIPSCs in the PC-projected DCN neurons held at 0 mV. Brain slices of App+/+ + vec, App−/− + vec and App−/− + APP mice were recorded; GABAB-receptor antagonist picrotoxin (PTX, 100 μM) was perfused at the end of the recordings to validate IPSCs. Right panel: statistics of IPSCs frequency and amplitude, n = 10 cells for each group. (D) IPSCs recorded from the PC-projected DCN neurons with 4,9-ahTTX (200 nM), n = 8 cells for each group. (E) Miniature IPSCs (mIPSCs) recorded from PC-projected DCN neurons in the presence of TTX (1 μM), n = 8 for each group. Scale bars as presented in the images. A 5-min recording segment from a stable period, defined as commencing at least 10 min after obtaining whole-cell access, was used for statistical analysis. Data are represented as the mean with SEM (C, D IPSCs amplitude) or the median with interquartile range (D IPSCs frequency, E). One-way ANOVA and Kruskal–Wallis test. * P < 0.05; ** P < 0.01; *** P < 0.001; ns, non-significant. The data underlying this Figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003513.g007 To further examine the role of Nav1.6 in AP transduction, we performed sIPSCs recordings during perfusion with 4,9 ah-TTX. This Nav1.6 blocker reduced the amplitude and the frequency of sIPSCs in all groups (Fig 7D). Importantly, application of 4,9 ah-TTX almost completely abolished the rescue effect of APP expression in GABAergic transmission of PC-DCN neurons (Fig 7D). These findings indicate that APP plays a crucial role in maintaining the inhibitory synaptic transmission from PCs to DCN neurons, and this function was Nav1.6-dependent. In addition, mIPSCs were also analyzed to identify whether APP deficiency influenced the AP firing-independent synaptic transmission. No significant difference in the amplitude or the frequency of miniature inhibitory post-synaptic currents (mIPSC) was observed among WT, APP-null + vec, and APP-null + APP groups (Fig 7E). Taken together, our data provide strong evidence that APP was essential for sustaining the repetitive firing activity of PCs and for maintaining inhibitory synaptic transmission from PCs to DCN neurons via Nav1.6, a process critical for cerebellar-dependent motor performance. Cerebellar PC APP mediates motor functions in mice Consistent with previous study [8,9], we observed that APP-null (App−/−) mice exhibited reduced locomotor activity (S1A Fig) and weaker forelimb grip strength (S1B Fig) compared with wild-type (WT, App+/+) littermates. Further, animals were tested for their capacity for motor learning and coordination using a rotarod and footprint test. APP-null mice showed normal motor learning over 3 days of training and their latency to fall was similar to that of WT mice in the test day (S1C Fig). During the footprint test, there was no differences between APP-null and WT mice in sway length and the overlap, while the stride length and the stance length were all significantly lower in APP-null group compared with that of WT (App+/+) group (S1D and S1E Fig). The reduced stride length in footprint test (S1E Fig) likely reflects smaller body size in APP-null mice (S1F Fig). To control for this confounder, we performed the balance beam assay. No significant differences were observed in traversal time or number of errors on 12- or 6-mm beams (S1G Fig), indicating intact gross motor coordination in APP-null mice. Thus, APP deletion specifically impairs open field locomotion and forelimb grip strength, independent of anthropometric differences. APP is abundantly expressed in PCs of adult WT mice (Fig 1A). Given that cerebellar PCs play a critical role in motor function, we hypothesized that the deletion of PC-specific APP could be the primary cause of the motor dysfunction observed in APP-null mice. To specifically evaluate the effect of PC APP on motor performance, we injected a lentivirus encoding human APP695 with mCherry (LV-CAMK2A-SP-Flag-APP695-IRES2-mCherry) into the cerebellar cortex (Fig 1B). The CAMK2A promoter, which has been shown to drive gene expression specifically in PCs [22], was used in this experiment. Expression of mCherry was observed exclusively in the Calbindin-positive PCs (Fig 1B). Laser capture microdissection (LCM) coupled with Reverse Transcription Quantitative PCR (RT-qPCR) analysis further confirmed a significant increase of the APP expression in PCs (Fig 1C). PC-specific expression of APP in APP-null mice significantly improved their performance in locomotor activity (Fig 1D) and forelimb grip strength test (Fig 1E), but didn’t affect their body size (Fig 1F). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Cerebellar Purkinje cell (PC)-specific amyloid precursor protein (APP) expression modulates motor function. (A) APP (red) and calbindin (green) immunofluorescence in 20-µm-thick cerebellar sagittal sections. The APP immunofluorescence in wild-type (App+/+) or APP-null (App−/−) mice were acquired under identical imaging settings. Insets: magnified views. Scale bars: 100 µm (main), 20 µm (insets). The PC contours are indicated by dashed lines. (B) Lentivirus-mediated overexpression of APP695 in the cerebellar cortex of App−/− mice. Representative images showing co-expression of APP/mCherry (red) and the PC marker calbindin (green) in a sagittal section of cerebellum. Scale bars: 100 µm (main); 20 µm (insets). (C) Left: representative fluorescence images showing before or after laser capture microdissection (LCM) of LV-transduced PCs (mcherry+) from fresh-frozen coronal sections. Right: App mRNA levels in dissected cells detected by qPCR (Data presentation: median with interquartile range. App+/+ + vec, n = 8; App−/− + vec, n = 4; App−/− + APP, n = 7 samples; Kruskal–Wallis test). Scale bar: 40 µm. (D–F) Motor phenotyping of App−/− mice with Purkinje APP rescue: (D) Open field total distance (10-min test). (E) Forelimb grip strength. (F) Body weight of all groups of mice. App+/+ + vec, n = 12; App−/− + vec, n = 13; App−/− + APP, n = 12. (G) Conditional knockdown of App in App+/+ mice using a mixed AAV injection strategy. Representative images show the CRE (mCherry+) and APP-shRNA (EGFP+) expression in PCs in a coronal section of the cerebellum. Scale bar: 100 µm (main), 20 µm (insets). The PC contours are indicated by dashed lines. (H) Left: representative fluorescence images before or after LCM of AAV-transduced PCs (mcherry+) from fresh-frozen coronal sections of App+/+ mice. Right: App mRNA levels in dissected cells detected by qPCR (control, n = 5, App kd, n = 10 samples). (I, J) Motor deficits following Purkinje APP knockdown: (I) Open field mobility. (J) Forelimb grip strength decline. (Control, n = 10; App kd, n = 8 mice). Data presentation: mean with SEM. Statistics: (D–F) one-way ANOVA; (H–J) Unpaired t test; * P < 0.05; ** P < 0.01; *** P < 0.001. The data underlying this Figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003513.g001 To further investigate the contribution of APP to motor control, we utilized mice with conditional APP knockdown restricted to PCs. We achieved PC-specific App knockdown by co-injecting WT mouse cerebellar cortex with AAV2/9-L7-Cre [23] and AAV2/9-DIO-(U6-EGFP)-shApp [24], whereby Cre recombinase activates the expression of shRNA targeting App specifically in L7-positive cells. Immunofluorescence confirmed efficient PC-specific transduction mediated by AAV under the L7 promoter, as shown by its colocalization with calbindin (Figs 1G and S2). LCM coupled with RT-qPCR verified dramatic APP downregulation in transduced neurons (Fig 1H). Subsequent behavioral analyses revealed significant motor deficits in conditional knockdown mice, including reduced ambulation in open field tests (Fig 1I) and impaired forelimb grip strength (Fig 1J). Taken together, these results showed that PC-specific APP expression mediates cerebellar-dependent motor performance. Full-length APP is necessary for rescue of motor function deficits in APP-null mice Proteolytic processing of full-length APP generates bioactive fragments including sAPPα (sufficient for spatial learning and long-term potentiation rescue in APP-null mice) [9] and APP intracellular domain (AICD) (with transcriptional activity). To determine whether the intact holoprotein is required for motor functions, we delivered holo-APP, sAPPα, or AICD to APP-null mice via cerebellar lentivirus injection. Fluorescence imaging confirmed targeted expression in PCs (Fig 2A). LCM-coupled RT-qPCR verified successful APP reconstitution in transduced neurons (Fig 2B). Behavioral analyses demonstrated that only full-length APP (holo-APP) significantly rescued motor deficits, exhibiting significant improvement in open field ambulation (Fig 2C) and grip strength (Fig 2D) in APP-null mice. sAPPα and AICD provided no significant rescue (Fig 2C and 2D). These results establish that full-length APP is necessary for cerebellum-mediated motor function. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Full-length APP is required for the rescue of motor deficit in APP-null mice. (A) Lentivirus-mediated overexpression of full-length APP (holo-APP), sAPPα, or AICD in the cerebellar cortex of App−/− mice. Representative images showing co-expression of APP fragments/mCherry (red) and the PC marker calbindin (green) from coronal sections of the cerebellum. Scale bars: 100 µm (main); 40 µm (insets). (B) QPCR analysis of LCMed PCs (mcherry+) from fresh-frozen sections. Top: Primer specificity schematic for APP fragments. Bottom: APP fragments mRNA level normalized to Gapdh. Data presentation: median with interquartile range. App−/− + vec, n = 4; App−/− + holo-APP, n = 5; App−/− + sAPPα, n = 4, App−/− + AICD, n = 4 samples. Kruskal–Wallis test. (C, D) Motor phenotyping of App−/− mice with Purkinje APP fragments rescue (n = 9 mice/group): (C) Open field total distance (10-min test). (D) Forelimb grip strength. N = 9 for each group. Data presentation: mean with SEM. Statistics: one-way ANOVA. ** P < 0.01; ns, non-significant. The data underlying this Figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003513.g002 APP deficiency impairs PC electrophysiological activity APP plays a crucial role in neurodevelopment [1,2], so we examined whether APP deficiency impaired the development of PCs. Golgi staining and Sholl analysis revealed that the dendritic branches and spine density were similar between 2-month-old APP-null and WT littermates (S3A–S3C Fig). Further, we compared the morphology of PC in 12-month-old mice. No significant difference in the dendritic branches and spine density was observed between APP-null and WT mice (S3D–S3F Fig). These results indicated that APP deficiency affects neither PC development nor provokes their degeneration. Next, we investigated whether loss of APP altered the firing patterns of PCs. As established, PCs display three characteristic firing states: silent, simple spikes (SSs), and complex spikes (CSs) [18,25,26]. Most of 22 PCs (19/22, 86.4%) recorded from WT mice exhibited spontaneous firing, while 13.6% (3/22) of PCs maintained silent within 1 h of recording. Complex spikes recorded in 27.3% (6/22) of PCs from WT mice (Fig 3A). The percentage of silent PCs increased to 42.1% (8/19, 42.11%), and no CSs was recorded in 19 PCs from APP-null mice (Fig 3A). We further analyzed the spontaneous firing rate and various AP characteristics, including threshold, peak amplitude, half-width, and afterhyperpolarization potential (AHP) of PCs with SSs in brain slices recorded from APP-null and WT mice. The spontaneous firing rate of PCs in APP-null mice was significantly lower than that in WT mice (Fig 3B). The rheobase of AP in APP-null PC was higher than that in WT mice (Fig 3C). In addition, lower peak amplitude, higher threshold, longer half-width, and bigger AHP (Fig 3D and 3E) were observed in the spontaneous AP spikes of APP-null than those of WT PCs. These results indicate that APP-deficiency significantly impairs the firing properties of cerebellar PCs. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. APP deficiency impairs the normal firing patterns of PCs. (A) Representative spontaneous firing patterns recorded in PCs of wild-type (App+/+) mice. Pie charts in the right panel showing the percentage of PC firing pattern. App+/+, n = 22; App−/−, n = 19 cells. (B) Representative spontaneous simple spikes recorded in PCs. Quantification of firing rates analyzed from 1-min recordings during stable whole-cell access (≥10 min after breakthrough). App+/+, n = 9; App−/−, n = 8. (C) Left: Representative voltage responses to depolarizing ramp current injections (−50 to +500 pA over 1 s) recorded PCs under whole-cell current-clamp. Right: Rheobase quantification (minimal current eliciting first action potential). n = 8 cells/group. (D) Representative spontaneous action potentials and the corresponding dV/dt curve. (E) Statistics of peak amplitude, threshold, half-width, and afterhyperpolarization potential (AHP) of action potentials shown in (D), App+/+, n = 10; App−/−, n = 9. Scale bars are indicated in the images. Data are represented as the mean with SEM. Student t test: * P < 0.05; *** P < 0.001. The data underlying this Figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003513.g003 Reduced Nav1.6, but not Nav1.1-mediated sodium current was detected in APP-null PCs Consistent with APP enhancing Nav1.6 currents in heterologous systems [21], whole-cell recordings in cerebellar slices revealed altered sodium currents in APP-null PCs. Transient (INaT), persistent (INaP), and resurgent (INaR) currents were evoked using established protocols (S4 Fig) [18]. Given that Nav1.6 and Nav1.1 are the predominant voltage-gated sodium channels in mouse PCs [16], we next sought to determine which of these isoforms was affected by APP deficiency. To this end, we applied subtype-specific blockers/inhibitors: 4,9-anhydrotetrodotoxin (4,9-ah-TTX; selective Nav1.6 blocker) and ICA-121431 (Nav1.1 inhibitor) [19]. Voltage steps (−70 to +10 mV) from a holding potential of −90 mV elicited INaT, mediated by Nav1.6 and Nav1.1 (TTX/4,9-ahTTX + ICA-sensitive, S4A and S4D Fig). APP-null PCs showed reduced peak INaT but unchanged voltage-dependence (Fig 4A and 4B). Pharmacological dissection revealed selective reduction of Nav1.6-mediated INaT (Fig 4C), while Nav1.1 currents remained unaffected (Fig 4D). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. APP deficiency selectively reduces Nav1.6-mediated Na+ currents in cerebellar PCs. (A) Representative traces for whole-cell recording of transient Na+ currents (INa) in PCs of App+/+ or App−/− mice. (B) Whole cell I–V curve, peak currents, and activation curves of transient INa as shown in (A). App+/+, n = 8; App−/−, n = 14 cells. (C, D) Representative traces and statistical bar-charts showing amplitude of transient INa (INaT) before or during bath application of (C) Nav1.6-specific blocker 4,9-anhydrotetrodotoxin (4,9-ahTTX, 200 nM) or (D) Nav1.1-specific inhibitor ICA-121431 (ICA, 350 nM). N = 6 cells for each group. The INa1.6 or INa1.1 were determined by subtracting the sodium currents recorded during bath application of specific blockers/inhibitors from that before drug application. (E) Representative persistent Na+ currents recorded in PCs. The right panels show statistics of the peak persistent currents and voltage potentials eliciting peak persistent Na+ currents. App+/+, n = 6; App−/−, n = 12 cells. (F) Representative traces (the left panel) and statistics for whole-cell recording of resurgent INa in PCs. The middle panel: Whole cell I–V curve of resurgent INa; the right panel: peak resurgent currents. App+/+, n = 9; App−/−, n = 11 cells. Student t test; * P < 0.05; ** P < 0.01; ns, nonsignificant. Scale bars are indicated in the images. (G) APP co-immunoprecipitated with Nav1.6 but not Nav1.1. HEK293 cells co-expressing Flag-tagged APP and either Nav1.6 or Nav1.1-EGFP were lysed and immunoprecipitated (IP) with anti-Flag affinity beads (Smart-Lifesciences, #SA042001). Immunoblots (IB) were probed for: Nav1.6 (Alomone Labs #ASC-009), EGFP (Roche Applied Science, #11814460001) or Flag (AlpVHHs, #016-303-005). (H) Nav1.6 interacts with full-length APP (holo-APP), but not soluble APPα (sAPPα) and APP intracellular domain (AICD). HEK293 cells co-expressing Nav1.6 and Flag-tagged holo-APP, sAPPα, or AICD were lysed and IPed with anti-Flag affinity beads. IB were probed for Nav1.6 or Flag. Mock transfected controls are shown in the first lane. Input controls (7.5% lysate) shown below/beside corresponding lanes. Data representative of 3 biological replicates. The data underlying this Figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003513.g004 Slow ramps (−90 to +30 mV, 0.12 mV/ms) evoked INaP, mediated by Nav1.6 in PCs (blocked by TTX/4,9-ahTTX; S4B, S4E, and S5A Figs). APP-null PCs exhibited diminished INaP amplitude without activating voltage shift (Fig 4E). Similarly, INaR (mainly mediated by Nav1.6, evoked by +30 mV and followed by a repolarization to voltage steps between −60 and −10 mV, S4C, S4F, and S5B Figs) was reduced in APP-null PCs with unaltered voltage dependence (Fig 4F). However, Nav1.6 mRNA (Scn8a) levels in PCs did not differ between WT and APP-null mice (S6 Fig), suggesting that APP may not influence Scn8a gene expression. Co-immunoprecipitation (co-IP) assays in transfected HEK293 cells (S7 Fig) confirmed a specific physical interaction between APP and Nav1.6 (Fig 4G). This binding was dependent on holo-APP, as neither the soluble ectodomain sAPPα nor the AICD co-precipitated with Nav1.6 (Fig 4H), consistent with the requirement of holo-APP for functional rescue observed in vivo (Fig 2). Taken together, APP deficiency selectively impairs Nav1.6 but not Nav1.1-mediated sodium currents in PCs through molecular interactions. Nav1.6 mediates PC firing and motor deficits in APP-null mice Our findings align with prior evidence that persistent/resurgent INa primarily depends on Nav1.6, while transient INa involves Nav1.1-Nav1.6 synergy [18,27]. Nav1.6 is critical for PC repetitive firing [18]. Accordingly, 4,9-ahTTX abolished repetitive firing in WT PCs (Fig 5A). Conversely, the Nav1.6 positive allosteric modulator poneratoxin (PoTX, inhibits Nav1.6 inactivation, S8 Fig) [28], restored firing in APP-null PCs (Fig 5B). ICA-121431 showed no effect in WT (Fig 5C) or APP-null PCs (Fig 5D). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. Nav1.6 mediated the deficits of PC firing and motor function in APP-null mice. (A–D) Electrophysiological characterization of PC spontaneous firing in acute cerebellar coronal slices (300-µm-thick). (A) Representative traces and quantification of spontaneous firing rate and spike amplitude before/during application of 200 nM 4,9-ahTTX in App+/+ mice. n = 7 cells. (B) Representative traces and quantification of spontaneous firing rate and spike amplitude before/during application of 30 nM PoTX (Nav1.6 positive allosteric modulator) in App−/− mice. n = 5 cells/group. (C, D) Representative traces and quantification of spontaneous firing rate and spike amplitude before/during application of 350 nM ICA (Nav1.1 inhibitor) in App+/+ (C) or App−/− (D) mice. Statistics: Paired t test; * P < 0.05; ** P < 0.01; *** P < 0.001 vs. baseline; ns, non-significant. n = 4 cells/group. Scale bars are indicated in the images. (E) Morphology verification of cerebellar cannula placement in a coronal section of mice cerebellum. (F) Open field locomotion: Total distance traveled (m) in App+/+ (4,9-ahTTX/vehicle) and App−/− (PoTX/vehicle) mice. Mice received bilateral microinjections of saline (vehicle), 200 nM 4,9-ahTTX, or 20 pM PoTX into the cerebellar cortex 20 min prior to testing. (G) Forelimb grip strength: maximal force/body weight in same cohorts as (F). Statistics (F, G): App+/+ + vehicle, n = 10; App+/+ + 4,9-ahTTX, n = 8; App−/− + vehicle, n = 8; App−/− + PoTX, n = 9; One-way ANOVA with Tukey post-hoc; * P < 0.05; ** P < 0.01, *** P < 0.001; ns, non-significant. ctrl, control; 4,9-ahTTX, 4,9-anhydrotetrodotoxin; ICA, ICA-121431; PoTX, poneratoxin. The data underlying this Figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003513.g005 Intracerebellar infusion studies further implicated the essential role of Nav1.6 in APP-mediated motor function. Histology confirmed cannula placement in cerebellar cortex (Fig 5E). Nav1.6 blocker infusion impaired WT mouse performance in open field and grip-strength tests (Fig 5F and 5G), whereas PoTX rescued motor deficits in APP-null mice (Fig 5F and 5G). These data provide evidence that APP influences PC firing and Motor function through Nav1.6. Exogenous APP expression restored Nav1.6 currents and firing properties of PCs in APP-null mice We next investigated whether the restoration of the behavioral phenotypes in APP-null mice via exogenous APP expression was linked to the improvement in Nav1.6 function in PCs. Whole-cell patch clamp recordings revealed that transient, persistent, and resurgent INa were all rescued by exogenous APP expression in APP-null PCs. The peak transient INa was restored to levels comparable to WT mice (Fig 6A and 6B). APP expression in PCs did not affect the activation curve (Fig 6B). Decreased persistent and resurgent current were also restored by APP expression in PCs (Fig 6C and 6D). These results indicated that exogenous expression of APP in PCs restore the Nav1.6 currents. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 6. Exogenous APP expression restored the Nav1.6 currents and ameliorated the abnormalities in the firing patterns of PCs of APP-null mice. (A, B) Representative traces for whole-cell recording of (A) transient Na+ current and (B) statistics of whole-cell I–V curve, peak transient currents, and activation curves in PCs of App+/+ + vec, App−/− + vec, and App−/− + APP mice. N = 8 cells for each group. (C) Representative persistent Na+ current recorded in PCs. The right panels show statistics of the peak persistent currents and the voltage eliciting the peak currents. App+/+ + vec, n = 6; App−/− + vec, n = 8; App−/− + APP, n = 6 cells. (D) Representative resurgent sodium currents and I–V curves recorded in PCs. App+/+ + vec, n = 8; App−/− + vec, n = 8; App−/− + APP, n = 8 cells. (E) Pie chart showing the percentage of PC firing patterns in App+/+ + vec, App−/− + vec, and App−/− + APP mice. (F) Representative spontaneous simple spikes recorded in PCs. Quantification of firing rates analyzed from 1-min recordings during stable whole-cell access (≥10 min after breakthrough). App+/+ + vec, n = 12; App−/− + vec, n = 9; App−/− + APP, n = 13 cells. (G) Representative membrane potential responses to depolarizing ramp current injections (−50 to +500 pA over 1 s) recorded from PCs under whole-cell current-clamp. Right: Rheobase quantification. N = 9 cells/group. (H) Representative images of spontaneous action potentials and the dV/dt curves. (I) Statistics of peak amplitude, threshold, half-width, and afterhyperpolarization potential (AHP) of spontaneous action potentials shown in (H), App+/+ + vec, n = 10; App−/− + vec, n = 8; App−/−- + APP, n = 8 cells. Scale bars are indicated in the images. One-way ANOVA and post-hoc Turkey’s test. * P < 0.05; ** P < 0.01; *** P < 0.001. The data underlying this Figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003513.g006 In the analysis of firing properties, we found that exogenous APP expression in APP-null PCs restored spontaneous firing patterns nearly to the WT level. Complex spikes were observed in the APP-expressing group (2/21, 9.5%), but not in the control group (Fig 6E). The spontaneous firing frequency of the APP-expressing group was almost restored to that of WT mice (Fig 6F). Additionally, exogenous APP-expression in APP-null PCs restored the rheobase (Fig 6G). Moreover, the spontaneous AP spikes from exogenous APP-expressing PCs displayed higher peak amplitude, lower threshold, shorter half-width, and smaller AHP (Fig 6H and 6I). These data indicate that exogenous APP expression in PCs restored Nav1.6 currents and firing properties in APP-null mice. Exogenous APP expression in PCs restored Nav1.6-dependent GABAergic inputs to DCN neurons in APP-null mice PC projections are the sole output of the cerebellar cortex and represent the major input to DCN neurons [29], forming approximately 70%–80% of all synaptic connections onto DCN cells [30,31]. To explore whether inhibitory transmission from PCs to DCN neurons was affected by APP deficiency, we performed whole-cell patch clamp recordings in cerebellar slices. DCN neurons innervated by PCs were labeled using trans-synaptic labeling, achieved by injecting AAV2/1-hSyn-CRE into the cerebellar cortex and AAV2/9-hSyn-DIO-EGFP in the DCN (Fig 7A). After 3 weeks of viral expression, EGFP-positive DCN neurons were recorded in brain slice (Fig 7B). No significant difference was observed in the number of EGFP-labeled DCN neurons between the different virus-injected groups. The amplitude and the frequency of spontaneous inhibitory post-synaptic currents (sIPSCs) in APP-null DCN neurons were lower than those in WT DCN neurons, and the abnormal GABAergic synaptic transmission APP-null mice was restored by APP expression in PCs (Fig 7C). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 7. Exogenous amyloid precursor protein (APP) expression restored the defects of the sIPSCs evoked by Nav1.6-mediated Purkenji cell inputs in APP-null mice. (A) Schematic of the viral injection strategy to achieve APP overexpression in PCs and trans-synaptic tracing of downstream neurons in the DCN. (B) Representative example of fluorescence-assisted selection of cells for patch-clamp recordings. Images from an acute cerebellar slice observed in differential interference contrast (DIC, left), or fluorescence mode (EGFP, right). (C) Left panel: Representative traces of sIPSCs in the PC-projected DCN neurons held at 0 mV. Brain slices of App+/+ + vec, App−/− + vec and App−/− + APP mice were recorded; GABAB-receptor antagonist picrotoxin (PTX, 100 μM) was perfused at the end of the recordings to validate IPSCs. Right panel: statistics of IPSCs frequency and amplitude, n = 10 cells for each group. (D) IPSCs recorded from the PC-projected DCN neurons with 4,9-ahTTX (200 nM), n = 8 cells for each group. (E) Miniature IPSCs (mIPSCs) recorded from PC-projected DCN neurons in the presence of TTX (1 μM), n = 8 for each group. Scale bars as presented in the images. A 5-min recording segment from a stable period, defined as commencing at least 10 min after obtaining whole-cell access, was used for statistical analysis. Data are represented as the mean with SEM (C, D IPSCs amplitude) or the median with interquartile range (D IPSCs frequency, E). One-way ANOVA and Kruskal–Wallis test. * P < 0.05; ** P < 0.01; *** P < 0.001; ns, non-significant. The data underlying this Figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003513.g007 To further examine the role of Nav1.6 in AP transduction, we performed sIPSCs recordings during perfusion with 4,9 ah-TTX. This Nav1.6 blocker reduced the amplitude and the frequency of sIPSCs in all groups (Fig 7D). Importantly, application of 4,9 ah-TTX almost completely abolished the rescue effect of APP expression in GABAergic transmission of PC-DCN neurons (Fig 7D). These findings indicate that APP plays a crucial role in maintaining the inhibitory synaptic transmission from PCs to DCN neurons, and this function was Nav1.6-dependent. In addition, mIPSCs were also analyzed to identify whether APP deficiency influenced the AP firing-independent synaptic transmission. No significant difference in the amplitude or the frequency of miniature inhibitory post-synaptic currents (mIPSC) was observed among WT, APP-null + vec, and APP-null + APP groups (Fig 7E). Taken together, our data provide strong evidence that APP was essential for sustaining the repetitive firing activity of PCs and for maintaining inhibitory synaptic transmission from PCs to DCN neurons via Nav1.6, a process critical for cerebellar-dependent motor performance. Discussion Our results demonstrate that the APP deficiency leads to aberrant sodium currents, altered firing patterns in PCs, reduced inhibitory synaptic transmission to DCN, and impaired motor function mediated by PCs. Specifically, we show that the reduction in Nav1.6 sodium currents is a key factor contributing to the motor deficits observed in APP-null mice. These defects were recapitulated by PC-specific APP knockdown and rescued upon exogenous expression of APP in PCs. Taken together, our findings reveal a novel role for APP in motor control and provide new insights into the molecular mechanisms underlying this process. Motor deficits in grip strength and locomotion in APP-deficient mice have been reported by several groups in the past [8,32]. However, the underlying mechanisms have remained poorly understood. The absence of prominent learning and memory phenotypes in single APP knockout mice has been attributed to functional compensation by APP family members, APLP1 and APLP2, with more pronounced deficits observed in APP and APLP double knockout mice [10]. Interestingly, APLPs do not compensate for certain phenotypes seen in APP knockout mice, particularly those related to motor function and body size, suggesting that APP plays a unique role in the regulation of motor functions within the APP family. A key question remains: in which brain regions or cell types does APP mediate its effects on motor control? Within the motor system, APP knockout mice exhibit normal neuromuscular junction morphology [11]. Furthermore, mice with conditional APP knockout targeted to excitatory neurons via the Nex promoter show intact motor functions [7]. These observations guided our focus on cerebellar PCs to investigate APP’s role in motor control. Our exogenous rescue and conditional knockdown experiments demonstrate that APP in PCs is critical for motor function, appropriate expression of APP in PCs is essential for normal motor performance, particularly specific behaviors such as locomotion and forelimb grasping ability. However, it should be noted that this study focuses solely on cerebellar PCs; we cannot exclude the possibility that APP also modulates motor function via other regions of the cerebrospinal circuit, such as spinal motor neurons. Nav1.6-mediated persistent sodium current (INaP) contributes to postural tone and amplifies locomotor outputs in spinal cord bistable motor neurons [33]. Given that APP is also expressed in spinal motor neurons [34], it is likely that spinal APP may similarly participate in the regulation of motor function. As a key component in the formation of internal models of the cerebellum, PCs have the capacity to process a vast array of signals [35]. The output information from PCs is encoded by two distinct types of APs: high-frequency SSs and low-frequency CSs. A single PC receives excitatory input from between 100,000 and 200,000 parallel fibers, which modulate the intrinsically driven high-frequency SS discharge [18,36]. In monkeys, increased SS discharge during grasping and lifting movements was observed in 56% of PCs [37], supporting the hypothesis that the cerebellum plans or controls movements through a forward internal model. Complex spikes are characterized by a large Na+ somatic spike followed by a burst of smaller spikelets, which are generated by a massive depolarization of the entire PC induced by climbing fiber activation [35]. These CSs serve as error signals crucial for online correction and motor learning [38]. Although the precise representations of PC firing patterns remain incompletely understood, there is a clear correlation between PC firing and motor output. Therefore, aberrant firing patterns in PCs are likely to influence both grip force and locomotion. Specifically, the decline in SS firing rates and the reduction in CS generation may contribute to the motor impairments, such as the deficits in grip strength and locomotion observed in APP-null mice. We previously reported that APP interacts with Nav1.6 and regulates its function in HEK293 cells by enhancing its cell surface distribution [21]. Nav1.6 is critical to the electrophysiological activity of PCs, comprising 82%–92% of resurgent sodium currents, 69% of persistent sodium currents, and more than 37% of the transient sodium currents [18]. Previous studies have shown that Nav1.6 and Nav1.1 are two major sodium channel subtypes expressed in PCs. We employed selective blockers/inhibitors combined with eliciting protocols to isolate different sodium current components produced by Nav1.1 or Nav1.6. Our data indicate that APP regulates Nav1.6 rather than Nav1.1 currents in transient INa. APP regulates the persistent and resurgent INa, which is predominantly produced by Nav1.6. Thus, APP plays a crucial role in the AP firings of PCs by modulating Nav1.6 currents. We further demonstrated that PoTX, a positive allosteric modulator of Nav1.6 that reduces channel inactivation, consequently enhanced the persistent INa—, key determinant of repetitive AP firing. This PoTX-mediated enhancement likely compensates for the functional deficits resulting from reduced surface distribution of Nav1.6 in APP-null PCs, as evidenced by the restoration of firing properties and amelioration of motor deficits. These results further support the conclusion that Nav1.6 serves as the primary effector mediating APP-dependent regulation of cerebellar PC function. However, as shown in Fig 2, APP-null mice exhibited a higher threshold for AP generation and an enhanced AHP. These electrophysiological alterations may collectively contribute to the observed reduction in spontaneous firing frequency of PCs. In addition to regulating Nav1.6, APP may also modulate the threshold for AP generation and AHP through other mechanisms. Additionally, Nav1.6 mediates the saltatory propagation of APs, reflected by the output of PCs. APP-null mice exhibited abnormal synaptic output to post-synaptic DCN neurons, contributing to impaired motor functions. Notably, the motor deficiencies observed in APP-null mice differ from those in Nav1.6-knockout mice (such as med mice). This discrepancy may be explained by the fact that Nav1.6 is only one of the downstream targets of APP, and the partial modulation of Nav1.6 by APP is not as dramatic as the complete knockout of the Scn8a gene. In summary, this study builds on previous findings that APP-null mice exhibit motor function deficits and identifies a novel mechanism through which APP regulates motor control via Nav1.6 in PCs. Materials and methods Ethics statement All mouse maintenance and experimental procedures were approved by the Institutional Animal Care and Use Committee and the Office of Laboratory Animal Resources of Xuzhou Medical University (protocol no. 202207S009). All procedures were conducted in accordance with the Regulations for the Administration of Affairs Concerning Experimental Animals (2017) in China and the United States Public Health Service Policy on Humane Care and Use of Laboratory Animals. Animals The APP-null mice were as described previously [8]. C57BL/6J background APP-null mice were purchased from the Jackson Laboratory (stock no. 004142). C57BL/6J mice were purchased from GemPharmatech (Nanjing, China). The mice were housed in groups, with a maximum of five individuals per cage, under a 12 h light/dark cycle. Male mice were used for behavioral tests to eliminate the influence of the estrous cycle, and both sexes were used in other experiments. All mice used in the experiments were 2–3-month-old, unless otherwise specified. Mice were randomly assigned to control and treatment groups by blindly selecting from the same cage of littermates. Antibodies and plasmids The antibodies used in this study included: rabbit anti-Nav1.6 (Scn8a) (IgG, 1:300, Alomone Labs, Catalogue No. ASC-009), mouse anti-GFP (1:1,000, Roche Applied Science, Catalogue No. 11814460001), anti-flag-HRP (1:10000, AlpVHHs, Catalogue No. 016-303-005). The plasmids used in this study included: pcDNA3.1-SP-Flag-hAPP695 was a gift from Dr. Gavin S. Dawe, pcDNA3.1-SP-Flag-hAPP (18-612AA)and pcDNA3.1-hAPP-flag (637-695AA)were constructed by Genecfps (Wuxi, China). pCMV-SCN1A (human) -EGFP-Neo was purchased from MIAOLING PLASMID (P51606). Virus injection Specific SP-Flag-APP (full-length of mice APP) sequence was incorporated into a recombinant lentivirus viral vector, which features a CAMK2A promoter to drive APP expression (LV-CAMK2A-SP-Flag-APP-3’UTR-IRES2-mCherry-WPRE). The LV-CAMK2A-mCherry was used as a control. Additionally, to investigate the roles of specific APP proteolytic fragments, lentiviral vectors expressing SP-Flag-sAPPα (LV-CAMK2A-SP-Flag-APP(AA18-612)-3′UTR-IRES2-mCherry-WPRE) and AICD-Flag (LV-CAMK2A-APP(AA637-695)-Flag-3′UTR-IRES2-mCherry-WPRE) were generated. The LV-CAMK2A-MCS-IRES2-mCherry vector was used as the control. The standard titers of LVs were ≥1 × 108 TU/ml. To achieve cell-type-specific knockdown of APP in cerebellar PCs, we employed a dual recombinant adeno-associated virus (rAAV) system [23,24]. A mixture (1:1 ratio) of rAAV-L7-CRE-P2A-mCherry-WPRE-hGH (utilizing the PC-specific L7 promoter to express Cre recombinase) and rAAV-CMV-DIO-(EGFP-U6)- App shRNA (expressing APP-targeting shRNA conditionally upon Cre-mediated recombination) was injected. A mixture of rAAV-L7-CRE-P2A-mCherry-WPRE-hGH and rAAV-CMV-DIO-(EGFP-U6)-App shRNA (scrambled sequence served as control). App shRNA sequence: 5′-GCACTAACTTGCACGACTATG-3′; Scrambled App shRNA: 5′-GGTCCAAACCGTCCAGTTAAT-3′. AAV2/1-hSyn-CRE-WPREs, and AAV2/9-hSyn-DIO-EGFP were purchased from BrainVTA Biotechnology (Wuhan, China). The viral titers were 5 × 1012 TU/ml. All viruses were aliquoted and stored at −80 °C until use. The methods for stereotaxic surgery were described before [39,40]. Briefly, mice were anesthetized with ketamine (100 mg/kg, i.p., 210220BL, Hengrui, China) and xylazine (10 mg/kg, i.p., YXK18801, Yingxin LAB, China). The virus was injected using a syringe nanoliter infusion/withdraw pump (KW-ZSB, KEW BASIS) attached to a 1 μl Hamilton syringe at a rate of 0.1 μl/min. For exogenous expression in PCs, 300 nl of the respective lentivirus (holo-APP, sAPPα, AICD, or control) was injected into the cerebellar cortex (coordinates, bregma: AP = −6.3 mm; ML = ± 2.0 mm; DV = −0.35 mm). For APP knockdown in PCs, 200 nl of the virus mixture was injected into the cerebellar cortex at the same coordinates used for lentiviral APP expression (bregma: AP = −6.3 mm; ML = 0, ± 2.0 mm; DV = −0.35 mm). For the PC-DCN trans-synaptic tracing, we injected 200 nl AAV2/1-hSyn-CRE-WPREs into cerebellar cortex and 200 nl AAV2/9-hSyn-DIO-EGFP into DCN (coordinates, bregma: AP = −6.0 mm; ML = ± 2.0 mm; DV = −3.5 mm). After surgery, mice were returned to their home cages and allowed to recover for at least three weeks before further experiments. Intracerebellar drug infusion Mice were anesthetized with ketamine (100 mg/kg, i.p.) and xylazine (10 mg/kg, i.p.) prior to stereotaxic surgery. Bilateral sterile guide cannula (o.d.: 0.36 mm; i.d.: 0.3 mm, AOGUAN Biotechnology) were implanted targeting the cerebellar cortex (bregma coordinates: AP = −6.3 mm, ML = ±2.0 mm; DV = −0.35 mm) for microinjection. Following a minimum recovery period of 4 days, mice received microinjections of either saline (0.5 µl/lateral), the Nav1.6 blocker 4,9-ahTTX (200 nM, 0.5 µl/lateral; MCE), or the Nav1.6 positive allosteric modulator PoTX (20 pM; 0.5 µl/lateral, MCE, HY-P10234A) 20 min before behavioral testing. Drug infusion into the cerebellar cortex was performed over 5 min using blunted microliter syringes (Hamilton, 1 µl, 25 gauge), extending 0.2 mm beyond the guide cannula tip. The injection needles remained in place for an additional 2 min post-infusion. Laser capture microdissection (LCM) and RT-qPCR The LCM system (ZEISS, PALM MicroBeam) was preheated before use. After anesthetization with ketamine and xylazine, mice were rapidly decapitated. The extracted brains were immediately embedded in optimal cutting temperature compound (OCT, Neg-50, Epredia, USA) and sectioned into 15-µm-thick slices using a cryostat. The tissue sections were mounted onto clean glass slides and positioned within an Axio Observer inverted microscope (ZEISS). The LCM system parameters were set according to the manufacturer’s instructions: 355-nm pulse laser, 55-μJ pulse energy, and 100% cutting speed. Following identification of fluorescence-positive PCs based on characteristic morphology (large flask-shaped somata, 20–30 μm diameter), targeted cells were collected via ultraviolet laser pressure catapulting using inverted-beam geometry for anti-gravitational ejection into adhesive cap tubes (AdhesiveCap 500, Zeiss). The harvested cells were transferred to RNase-free Eppendorf tubes for mRNA extraction and quantitative PCR (qPCR) analysis. The detailed qPCR protocols and quantification methods were previously described [39]. The primer sequences used for mRNA quantification: holo-APP and sAPPα (Forward: GAAGCCATGCTCAATGACC; Reverse: ATGCTTTAGGGTGTGCTGTC); Gapdh (Forward: ATGGTGAAGGTCGGTGTGAACG; Reverse: CGCTCCTGGAAGATGGTGATGG); AICD (Forward: ATCATGGTGTGGTGGAGGTTG; Reverse: AGGTTGGATTTTCGTAGCCGT); Scn8a (Forward: AGGCCCCGACAGTTTCAAG; Reverse: GGGTGGTTTCTTGAGCTTGC). The relative quantification was calculated by 2−ΔΔct method. Whole-cell patch-clamp recording Mice were anesthetized with isoflurane (3%−4%, R510-22, RWD, China) and rapidly decapitated. The cerebellum was carefully dissected from the extracted brain using a sharp blade. Coronal cerebellar slices (300-μm-thick) were prepared using a vibratome (VT1200S, Leica Microsystems, Nussloch, Germany) in ice-cold sucrose-based artificial cerebrospinal fluid (sACSF) (sucrose 212, 3 mM KCl, 1.25 mM NaH2PO4, 26 mM NaHCO3, 10 mM glucose, 7 mM MgCl2, PH 7.3, 320 mOsm), equilibrated with 95% O2 and 5% CO2. Slices were allowed to recover in ACSF at 32 °C for 45–60 min, and then incubated at room temperature (22–24°C) in normal ACSF (124 mM NaCl, 2.5 mM KCl, 1.25 mM NaH2PO4, 1.3 mM MgSO4, 2 mM CaCl2, 26 mM NaHCO3, and 20 mM glucose; titrated to pH 7.4 with NaOH) for at least 30 min before use. PCs in brain slices were visualized under an upright microscope (Olympus BX51WI) equipped with an infrared CCD camera. Whole-cell patch-clamp recordings were performed using a MultiClamp 700B amplifier (Axon Instruments), a Digidata 1550B analog-to-digital converter (Axon Instruments), and pClamp 10.7 software (Molecular Devices, San Jose, CA). Patch electrodes had a resistance of 2–4 MΩ when filled with either a firing-recording internal solution (140 mM K-methylsulfate, 7 mM KCl, 2 mM MgCl2, 10 mM HEPES, 0.1 mM EGTA, 4 mM Na2-ATP, 0.4 mM GTP-Tris) or a low-chloride internal solution for inhibitory synaptic transmission recording (135 mM K-gluconate, 5 mM KCl, 0.2 mM EGTA, 0.5 mM CaCl2, 10 mM HEPES, 2 mM Mg-ATP, 0.1 mM GTP). The pH was adjusted to 7.2 using Tris-base, and the osmolarity was adjusted to 300 mOsm with sucrose. ICA121431 (350 nM, MCE, HY-16787) or 4,9-ahTTX (200 nM, GlpBio, GC42327) were dissolved in ACSF to selectively inhibit Nav1.1 or Nav1.6, respectively. For recording mIPSCs in PCs, 1 μM TTX (L1808N, Puhuashi Technology, China) was dissolved in ACSF. IPSCs were recorded at a holding potential of 0 mV. For whole-cell sodium current recordings, 0.2 mM CdCl2 (202908, Sigma, USA) was added to block calcium channels, and 140 mM TEACl (T2265, Sigma, USA) was included to inhibit potassium channels. Patch electrodes were filled with an intrapipette solution containing 108 mM CsF, 6 mM MgCl2, 1.8 mM EGTA, 10 mM HEPES, 4 mM Na2-ATP, and 0.3 mM Tris-GTP, pH 7.3, 280–290 mOsm. Transient sodium currents were elicited by a series of 10 mV steps from a holding potential of −90 to +60 mV. Persistent current was elicited by a slow ramp increase from −90 to +30 mV at a rate of 0.12 mV/ms. Resurgent sodium current was elicited following a step to +30 mV (from a holding potential of −90 mV) by a series of 10-mV-depolarizing voltage steps from −60 to +10 mV. Sodium currents were verified by their complete block during bath application of 1 μM TTX in ACSF. INav1.6 was isolated by subtracting the sodium currents recorded during bath application of 200 nM 4,9-ahTTX from those recorded before the blocker application. INav1.1 was isolated by subtracting the sodium currents recorded during bath application of 350 nM ICA-121431 from those recorded before the inhibitor application. All experiments were conducted to collect data during a stable period, which was defined as at least 10 min after establishing whole-cell access. Electrophysiological data were analyzed offline using Clampfit 11.2 software (Molecular Devices). To ensure high-quality intracellular recordings, only cells exhibiting a stable resting membrane potential and access resistance with no more than 20% variation were considered valid for analysis. Behavioral tests Open field test. Each mouse was placed in the center of an open field arena (a cube with 50 cm long sides and a height of 50 cm) and allowed to explore freely for 10 min. Locomotor activity was recorded using a video camera controlled by ANY-Maze 14.0 software. The total travel distance of each mouse within the arena was recorded and analyzed. Grip strength test. The grip strength of mouse forelimbs was assessed using a grid connected to a strength sensor (YLK-2N, ELECALL, China), as previously described [41]. Mice were allowed to grip a metal grid with their forelimbs, after which they were lifted by the tail and gently pulled backward until they released the grid. During the test, the hind limbs were kept away from the grid. Grip strength was measured 10 times, and the mean of the top five values was used for analysis. All grip strength values were normalized to body weight. Rotarod test. Mice were trained to run at an accelerating speed (from 4 to 40 rpm, with an acceleration of 0.1 rpm/s) on a rotarod instrument (ZH-600B, ZhenghuaBiologic, China). Animals underwent three trials of accelerating rotarod running, with a time limit of 6 min per trial. The latency to fall was recorded as an indicator of training performance. A 6-min rest period was provided between trials to minimize stress and fatigue. Footprint assay. This assay was used to assess gait abnormality. A recording paper (100 cm × 10 cm) was placed at the bottom of a clear plexiglass tunnel (100 cm × 10 cm × 10 cm), with a darkened cage at the end of the tunnel [42]. Mice with different ink-painted front and rear paws were allowed to travel through the tunnel. The following gait parameters were then measured using the footprints on the recording paper: (a) stride length, the distance between two successive rear paw prints on the right side; (b) stance length, the distance between the left and right rear paws; (c) sway length, the vertical distance between the left and right rear paws; and (d) overlap length, the distance between the center of the front and rear paw prints on the right side (S1D Fig). Balance beam test. Mice were assessed for motor coordination using a horizontal round beam (100 cm length × 6/12 mm diameter, wooden) elevated 30 cm above the floor. The beam terminated at a darkened escape platform (20 × 20 cm). After two consecutive days of training (3 trials/day per beam width; data not recorded), formal testing on day 3 was recorded: (a) traversal latency (from beam entry to platform access) and (b) total limb slips (counted by blinded observers from video recordings). Three trials per beam width were averaged for analysis. Immunofluorescence Mice were anesthetized with ketamine and xylazine (100 mg/kg and 10 mg/kg, i.p.), then subjected to transcardial perfusion with 20 ml phosphate-buffered saline (PBS), followed by 20 ml 4% paraformaldehyde (PFA). Brains were extracted and post-fixed in 4% PFA overnight at 4 °C. Following fixation, tissues were dehydrated stepwise in 15% and 30% sucrose and embedded in OCT for sectioning. Cerebellar sections (20-μm-thick) were cut using a cryostat (Leica) and mounted on gelatin-coated slides (Citoglas). Sections were rinsed with PBS and subsequently incubated overnight at 4 °C with primary antibodies diluted in PBS containing 0.1% Triton X-100, 0.05% Tween-20, and 1% goat serum. Rabbit anti-APP IgG (1:500, Abcam, Catalogue No. AB32136), mouse anti-Calbindin-D28K IgG (1:200, Proteintech, Catalogue No. AB2881769), rabbit anti-NaV1.6 (Scn8a) IgG (1:300, Alomone Labs, Catalogue No. ASC-009) were used in the study. Sections were washed three times (10 min each) with PBS, then incubated with Alexa 488- or Alexa 594-conjugated secondary antibodies (1:2,000; Jackson ImmunoResearch) for 1 h at room temperature, protected from light. Finally, images were captured using a fluorescence microscope (Olympus IX81) controlled by Cellsens Standard software (Olympus, Japan), and processed using ImageJ (NIH, Bethesda, MD, USA). Golgi staining Golgi-Cox staining was performed using the Golgi-Cox OptimStain PreKit (PK401, HiTO, USA) according to the manufacturer’s instructions. Briefly, the solution A and solution B from the kit were mixed and added to the light-protected glass bottle 24 h in advance. The following day, brains were immersed in the solution A + B solution and stored in the dark at room temperature for 2 weeks, with the impregnation solution replaced after 24 h. Afterward, the brains were transferred to solution C and stored in the dark at room temperature for 3–7 days. Brain tissue was then sectioned into 120-μm-thick slices using a vibratome and mounted onto gelatin-coated slides. Excess water was removed, and a small amount of solution C was added dropwise. After 2 min, the slide was tilted to dry. Slides were rinsed three times in ddH2O and stained for 10 min using a freshly prepared staining solution consisting of one part solution D, one part solution E, and two parts ddH2O. The slides were then washed in ddH2O for 10 min. Sections were dehydrated and cleared in xylene (10023418, Sinopharm, China) and mounted with neutral gum (G8590, Solarbio, China). Images were obtained using an Olympus BX53 microscope with Olympus CellSens Standard software and analyzed by ImageJ Fiji. Dendritic spine density analysis was conducted by counting the dendritic spines along the full length of the apical dendrite of each PC. Twelve cells were counted for statistical analysis by a researcher blinded to the treatment group. Sholl analysis was performed using ImageJ software. The number of intersections between the circles and dendrites were plotted against the radii. The diameter of the primary dendrite was measured at 10 µm from the soma. Cell culture and transfection The HEK293 cell line stably expressing Nav1.6 was described previously [21]. These cells were maintained in DMEM supplemented with 10% (v/v) fetal bovine serum (FBS) and 200 μg/ml G418 (Absin). Normal HEK293 cells were cultured in DMEM containing 10% FBS and 1% penicillin/streptomycin. Plasmid transfections used Lipofectamine 2000 (GlpBio) per manufacturer protocol. Cells were harvested 48 h post-transfection for downstream assays. Co-IP assay Transfected HEK293 cells were harvested and lysed in ice-cold lysis buffer (150 mM NaCl, 30 mM HEPES, 10 mM NaF, 1% (v/v) Triton X-100, 0.01% (w/v) SDS, and complete protease inhibitor mixtures, pH 7.5). The lysates were rotated for 2 h at 4 °C and centrifuged at 12,000 rpm for 20 min. The supernatants were collected and incubated overnight at 4 °C with Anti-Flag Affinity Beads (smart-lifesciences, Catalogue No. SA042001), then washed three times with ice-cold lysis buffer. Finally, the samples were boiled in SDS loading buffer and analyzed by Western blotting. Statistical analysis Data were analyzed using GraphPad Prism 8.0. Normality was assessed with the Shapiro–Wilk test. Normally distributed data are expressed as mean with SEM, while non-normally distributed data are presented as median with interquartile range. Outliers were identified using Dixon’s Q-test at the 95% confidence level. For comparisons between two groups, unpaired or paired Student’s t-tests were used, as appropriate. Comparisons across three or more groups were performed using one-way ANOVA (for normally distributed data with homogeneous variance) or the Kruskal–Wallis test (for non-normal distributions), followed by appropriate post hoc tests. A P-value < 0.05 was considered statistically significant. Declaration of generative AI and AI-assisted technologies in the writing process During the preparation of this work, the authors used ChatGPT and DeepSeek in order to improve language and readability. After using this tool/service, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication. Ethics statement All mouse maintenance and experimental procedures were approved by the Institutional Animal Care and Use Committee and the Office of Laboratory Animal Resources of Xuzhou Medical University (protocol no. 202207S009). All procedures were conducted in accordance with the Regulations for the Administration of Affairs Concerning Experimental Animals (2017) in China and the United States Public Health Service Policy on Humane Care and Use of Laboratory Animals. Animals The APP-null mice were as described previously [8]. C57BL/6J background APP-null mice were purchased from the Jackson Laboratory (stock no. 004142). C57BL/6J mice were purchased from GemPharmatech (Nanjing, China). The mice were housed in groups, with a maximum of five individuals per cage, under a 12 h light/dark cycle. Male mice were used for behavioral tests to eliminate the influence of the estrous cycle, and both sexes were used in other experiments. All mice used in the experiments were 2–3-month-old, unless otherwise specified. Mice were randomly assigned to control and treatment groups by blindly selecting from the same cage of littermates. Antibodies and plasmids The antibodies used in this study included: rabbit anti-Nav1.6 (Scn8a) (IgG, 1:300, Alomone Labs, Catalogue No. ASC-009), mouse anti-GFP (1:1,000, Roche Applied Science, Catalogue No. 11814460001), anti-flag-HRP (1:10000, AlpVHHs, Catalogue No. 016-303-005). The plasmids used in this study included: pcDNA3.1-SP-Flag-hAPP695 was a gift from Dr. Gavin S. Dawe, pcDNA3.1-SP-Flag-hAPP (18-612AA)and pcDNA3.1-hAPP-flag (637-695AA)were constructed by Genecfps (Wuxi, China). pCMV-SCN1A (human) -EGFP-Neo was purchased from MIAOLING PLASMID (P51606). Virus injection Specific SP-Flag-APP (full-length of mice APP) sequence was incorporated into a recombinant lentivirus viral vector, which features a CAMK2A promoter to drive APP expression (LV-CAMK2A-SP-Flag-APP-3’UTR-IRES2-mCherry-WPRE). The LV-CAMK2A-mCherry was used as a control. Additionally, to investigate the roles of specific APP proteolytic fragments, lentiviral vectors expressing SP-Flag-sAPPα (LV-CAMK2A-SP-Flag-APP(AA18-612)-3′UTR-IRES2-mCherry-WPRE) and AICD-Flag (LV-CAMK2A-APP(AA637-695)-Flag-3′UTR-IRES2-mCherry-WPRE) were generated. The LV-CAMK2A-MCS-IRES2-mCherry vector was used as the control. The standard titers of LVs were ≥1 × 108 TU/ml. To achieve cell-type-specific knockdown of APP in cerebellar PCs, we employed a dual recombinant adeno-associated virus (rAAV) system [23,24]. A mixture (1:1 ratio) of rAAV-L7-CRE-P2A-mCherry-WPRE-hGH (utilizing the PC-specific L7 promoter to express Cre recombinase) and rAAV-CMV-DIO-(EGFP-U6)- App shRNA (expressing APP-targeting shRNA conditionally upon Cre-mediated recombination) was injected. A mixture of rAAV-L7-CRE-P2A-mCherry-WPRE-hGH and rAAV-CMV-DIO-(EGFP-U6)-App shRNA (scrambled sequence served as control). App shRNA sequence: 5′-GCACTAACTTGCACGACTATG-3′; Scrambled App shRNA: 5′-GGTCCAAACCGTCCAGTTAAT-3′. AAV2/1-hSyn-CRE-WPREs, and AAV2/9-hSyn-DIO-EGFP were purchased from BrainVTA Biotechnology (Wuhan, China). The viral titers were 5 × 1012 TU/ml. All viruses were aliquoted and stored at −80 °C until use. The methods for stereotaxic surgery were described before [39,40]. Briefly, mice were anesthetized with ketamine (100 mg/kg, i.p., 210220BL, Hengrui, China) and xylazine (10 mg/kg, i.p., YXK18801, Yingxin LAB, China). The virus was injected using a syringe nanoliter infusion/withdraw pump (KW-ZSB, KEW BASIS) attached to a 1 μl Hamilton syringe at a rate of 0.1 μl/min. For exogenous expression in PCs, 300 nl of the respective lentivirus (holo-APP, sAPPα, AICD, or control) was injected into the cerebellar cortex (coordinates, bregma: AP = −6.3 mm; ML = ± 2.0 mm; DV = −0.35 mm). For APP knockdown in PCs, 200 nl of the virus mixture was injected into the cerebellar cortex at the same coordinates used for lentiviral APP expression (bregma: AP = −6.3 mm; ML = 0, ± 2.0 mm; DV = −0.35 mm). For the PC-DCN trans-synaptic tracing, we injected 200 nl AAV2/1-hSyn-CRE-WPREs into cerebellar cortex and 200 nl AAV2/9-hSyn-DIO-EGFP into DCN (coordinates, bregma: AP = −6.0 mm; ML = ± 2.0 mm; DV = −3.5 mm). After surgery, mice were returned to their home cages and allowed to recover for at least three weeks before further experiments. Intracerebellar drug infusion Mice were anesthetized with ketamine (100 mg/kg, i.p.) and xylazine (10 mg/kg, i.p.) prior to stereotaxic surgery. Bilateral sterile guide cannula (o.d.: 0.36 mm; i.d.: 0.3 mm, AOGUAN Biotechnology) were implanted targeting the cerebellar cortex (bregma coordinates: AP = −6.3 mm, ML = ±2.0 mm; DV = −0.35 mm) for microinjection. Following a minimum recovery period of 4 days, mice received microinjections of either saline (0.5 µl/lateral), the Nav1.6 blocker 4,9-ahTTX (200 nM, 0.5 µl/lateral; MCE), or the Nav1.6 positive allosteric modulator PoTX (20 pM; 0.5 µl/lateral, MCE, HY-P10234A) 20 min before behavioral testing. Drug infusion into the cerebellar cortex was performed over 5 min using blunted microliter syringes (Hamilton, 1 µl, 25 gauge), extending 0.2 mm beyond the guide cannula tip. The injection needles remained in place for an additional 2 min post-infusion. Laser capture microdissection (LCM) and RT-qPCR The LCM system (ZEISS, PALM MicroBeam) was preheated before use. After anesthetization with ketamine and xylazine, mice were rapidly decapitated. The extracted brains were immediately embedded in optimal cutting temperature compound (OCT, Neg-50, Epredia, USA) and sectioned into 15-µm-thick slices using a cryostat. The tissue sections were mounted onto clean glass slides and positioned within an Axio Observer inverted microscope (ZEISS). The LCM system parameters were set according to the manufacturer’s instructions: 355-nm pulse laser, 55-μJ pulse energy, and 100% cutting speed. Following identification of fluorescence-positive PCs based on characteristic morphology (large flask-shaped somata, 20–30 μm diameter), targeted cells were collected via ultraviolet laser pressure catapulting using inverted-beam geometry for anti-gravitational ejection into adhesive cap tubes (AdhesiveCap 500, Zeiss). The harvested cells were transferred to RNase-free Eppendorf tubes for mRNA extraction and quantitative PCR (qPCR) analysis. The detailed qPCR protocols and quantification methods were previously described [39]. The primer sequences used for mRNA quantification: holo-APP and sAPPα (Forward: GAAGCCATGCTCAATGACC; Reverse: ATGCTTTAGGGTGTGCTGTC); Gapdh (Forward: ATGGTGAAGGTCGGTGTGAACG; Reverse: CGCTCCTGGAAGATGGTGATGG); AICD (Forward: ATCATGGTGTGGTGGAGGTTG; Reverse: AGGTTGGATTTTCGTAGCCGT); Scn8a (Forward: AGGCCCCGACAGTTTCAAG; Reverse: GGGTGGTTTCTTGAGCTTGC). The relative quantification was calculated by 2−ΔΔct method. Whole-cell patch-clamp recording Mice were anesthetized with isoflurane (3%−4%, R510-22, RWD, China) and rapidly decapitated. The cerebellum was carefully dissected from the extracted brain using a sharp blade. Coronal cerebellar slices (300-μm-thick) were prepared using a vibratome (VT1200S, Leica Microsystems, Nussloch, Germany) in ice-cold sucrose-based artificial cerebrospinal fluid (sACSF) (sucrose 212, 3 mM KCl, 1.25 mM NaH2PO4, 26 mM NaHCO3, 10 mM glucose, 7 mM MgCl2, PH 7.3, 320 mOsm), equilibrated with 95% O2 and 5% CO2. Slices were allowed to recover in ACSF at 32 °C for 45–60 min, and then incubated at room temperature (22–24°C) in normal ACSF (124 mM NaCl, 2.5 mM KCl, 1.25 mM NaH2PO4, 1.3 mM MgSO4, 2 mM CaCl2, 26 mM NaHCO3, and 20 mM glucose; titrated to pH 7.4 with NaOH) for at least 30 min before use. PCs in brain slices were visualized under an upright microscope (Olympus BX51WI) equipped with an infrared CCD camera. Whole-cell patch-clamp recordings were performed using a MultiClamp 700B amplifier (Axon Instruments), a Digidata 1550B analog-to-digital converter (Axon Instruments), and pClamp 10.7 software (Molecular Devices, San Jose, CA). Patch electrodes had a resistance of 2–4 MΩ when filled with either a firing-recording internal solution (140 mM K-methylsulfate, 7 mM KCl, 2 mM MgCl2, 10 mM HEPES, 0.1 mM EGTA, 4 mM Na2-ATP, 0.4 mM GTP-Tris) or a low-chloride internal solution for inhibitory synaptic transmission recording (135 mM K-gluconate, 5 mM KCl, 0.2 mM EGTA, 0.5 mM CaCl2, 10 mM HEPES, 2 mM Mg-ATP, 0.1 mM GTP). The pH was adjusted to 7.2 using Tris-base, and the osmolarity was adjusted to 300 mOsm with sucrose. ICA121431 (350 nM, MCE, HY-16787) or 4,9-ahTTX (200 nM, GlpBio, GC42327) were dissolved in ACSF to selectively inhibit Nav1.1 or Nav1.6, respectively. For recording mIPSCs in PCs, 1 μM TTX (L1808N, Puhuashi Technology, China) was dissolved in ACSF. IPSCs were recorded at a holding potential of 0 mV. For whole-cell sodium current recordings, 0.2 mM CdCl2 (202908, Sigma, USA) was added to block calcium channels, and 140 mM TEACl (T2265, Sigma, USA) was included to inhibit potassium channels. Patch electrodes were filled with an intrapipette solution containing 108 mM CsF, 6 mM MgCl2, 1.8 mM EGTA, 10 mM HEPES, 4 mM Na2-ATP, and 0.3 mM Tris-GTP, pH 7.3, 280–290 mOsm. Transient sodium currents were elicited by a series of 10 mV steps from a holding potential of −90 to +60 mV. Persistent current was elicited by a slow ramp increase from −90 to +30 mV at a rate of 0.12 mV/ms. Resurgent sodium current was elicited following a step to +30 mV (from a holding potential of −90 mV) by a series of 10-mV-depolarizing voltage steps from −60 to +10 mV. Sodium currents were verified by their complete block during bath application of 1 μM TTX in ACSF. INav1.6 was isolated by subtracting the sodium currents recorded during bath application of 200 nM 4,9-ahTTX from those recorded before the blocker application. INav1.1 was isolated by subtracting the sodium currents recorded during bath application of 350 nM ICA-121431 from those recorded before the inhibitor application. All experiments were conducted to collect data during a stable period, which was defined as at least 10 min after establishing whole-cell access. Electrophysiological data were analyzed offline using Clampfit 11.2 software (Molecular Devices). To ensure high-quality intracellular recordings, only cells exhibiting a stable resting membrane potential and access resistance with no more than 20% variation were considered valid for analysis. Behavioral tests Open field test. Each mouse was placed in the center of an open field arena (a cube with 50 cm long sides and a height of 50 cm) and allowed to explore freely for 10 min. Locomotor activity was recorded using a video camera controlled by ANY-Maze 14.0 software. The total travel distance of each mouse within the arena was recorded and analyzed. Grip strength test. The grip strength of mouse forelimbs was assessed using a grid connected to a strength sensor (YLK-2N, ELECALL, China), as previously described [41]. Mice were allowed to grip a metal grid with their forelimbs, after which they were lifted by the tail and gently pulled backward until they released the grid. During the test, the hind limbs were kept away from the grid. Grip strength was measured 10 times, and the mean of the top five values was used for analysis. All grip strength values were normalized to body weight. Rotarod test. Mice were trained to run at an accelerating speed (from 4 to 40 rpm, with an acceleration of 0.1 rpm/s) on a rotarod instrument (ZH-600B, ZhenghuaBiologic, China). Animals underwent three trials of accelerating rotarod running, with a time limit of 6 min per trial. The latency to fall was recorded as an indicator of training performance. A 6-min rest period was provided between trials to minimize stress and fatigue. Footprint assay. This assay was used to assess gait abnormality. A recording paper (100 cm × 10 cm) was placed at the bottom of a clear plexiglass tunnel (100 cm × 10 cm × 10 cm), with a darkened cage at the end of the tunnel [42]. Mice with different ink-painted front and rear paws were allowed to travel through the tunnel. The following gait parameters were then measured using the footprints on the recording paper: (a) stride length, the distance between two successive rear paw prints on the right side; (b) stance length, the distance between the left and right rear paws; (c) sway length, the vertical distance between the left and right rear paws; and (d) overlap length, the distance between the center of the front and rear paw prints on the right side (S1D Fig). Balance beam test. Mice were assessed for motor coordination using a horizontal round beam (100 cm length × 6/12 mm diameter, wooden) elevated 30 cm above the floor. The beam terminated at a darkened escape platform (20 × 20 cm). After two consecutive days of training (3 trials/day per beam width; data not recorded), formal testing on day 3 was recorded: (a) traversal latency (from beam entry to platform access) and (b) total limb slips (counted by blinded observers from video recordings). Three trials per beam width were averaged for analysis. Open field test. Each mouse was placed in the center of an open field arena (a cube with 50 cm long sides and a height of 50 cm) and allowed to explore freely for 10 min. Locomotor activity was recorded using a video camera controlled by ANY-Maze 14.0 software. The total travel distance of each mouse within the arena was recorded and analyzed. Grip strength test. The grip strength of mouse forelimbs was assessed using a grid connected to a strength sensor (YLK-2N, ELECALL, China), as previously described [41]. Mice were allowed to grip a metal grid with their forelimbs, after which they were lifted by the tail and gently pulled backward until they released the grid. During the test, the hind limbs were kept away from the grid. Grip strength was measured 10 times, and the mean of the top five values was used for analysis. All grip strength values were normalized to body weight. Rotarod test. Mice were trained to run at an accelerating speed (from 4 to 40 rpm, with an acceleration of 0.1 rpm/s) on a rotarod instrument (ZH-600B, ZhenghuaBiologic, China). Animals underwent three trials of accelerating rotarod running, with a time limit of 6 min per trial. The latency to fall was recorded as an indicator of training performance. A 6-min rest period was provided between trials to minimize stress and fatigue. Footprint assay. This assay was used to assess gait abnormality. A recording paper (100 cm × 10 cm) was placed at the bottom of a clear plexiglass tunnel (100 cm × 10 cm × 10 cm), with a darkened cage at the end of the tunnel [42]. Mice with different ink-painted front and rear paws were allowed to travel through the tunnel. The following gait parameters were then measured using the footprints on the recording paper: (a) stride length, the distance between two successive rear paw prints on the right side; (b) stance length, the distance between the left and right rear paws; (c) sway length, the vertical distance between the left and right rear paws; and (d) overlap length, the distance between the center of the front and rear paw prints on the right side (S1D Fig). Balance beam test. Mice were assessed for motor coordination using a horizontal round beam (100 cm length × 6/12 mm diameter, wooden) elevated 30 cm above the floor. The beam terminated at a darkened escape platform (20 × 20 cm). After two consecutive days of training (3 trials/day per beam width; data not recorded), formal testing on day 3 was recorded: (a) traversal latency (from beam entry to platform access) and (b) total limb slips (counted by blinded observers from video recordings). Three trials per beam width were averaged for analysis. Immunofluorescence Mice were anesthetized with ketamine and xylazine (100 mg/kg and 10 mg/kg, i.p.), then subjected to transcardial perfusion with 20 ml phosphate-buffered saline (PBS), followed by 20 ml 4% paraformaldehyde (PFA). Brains were extracted and post-fixed in 4% PFA overnight at 4 °C. Following fixation, tissues were dehydrated stepwise in 15% and 30% sucrose and embedded in OCT for sectioning. Cerebellar sections (20-μm-thick) were cut using a cryostat (Leica) and mounted on gelatin-coated slides (Citoglas). Sections were rinsed with PBS and subsequently incubated overnight at 4 °C with primary antibodies diluted in PBS containing 0.1% Triton X-100, 0.05% Tween-20, and 1% goat serum. Rabbit anti-APP IgG (1:500, Abcam, Catalogue No. AB32136), mouse anti-Calbindin-D28K IgG (1:200, Proteintech, Catalogue No. AB2881769), rabbit anti-NaV1.6 (Scn8a) IgG (1:300, Alomone Labs, Catalogue No. ASC-009) were used in the study. Sections were washed three times (10 min each) with PBS, then incubated with Alexa 488- or Alexa 594-conjugated secondary antibodies (1:2,000; Jackson ImmunoResearch) for 1 h at room temperature, protected from light. Finally, images were captured using a fluorescence microscope (Olympus IX81) controlled by Cellsens Standard software (Olympus, Japan), and processed using ImageJ (NIH, Bethesda, MD, USA). Golgi staining Golgi-Cox staining was performed using the Golgi-Cox OptimStain PreKit (PK401, HiTO, USA) according to the manufacturer’s instructions. Briefly, the solution A and solution B from the kit were mixed and added to the light-protected glass bottle 24 h in advance. The following day, brains were immersed in the solution A + B solution and stored in the dark at room temperature for 2 weeks, with the impregnation solution replaced after 24 h. Afterward, the brains were transferred to solution C and stored in the dark at room temperature for 3–7 days. Brain tissue was then sectioned into 120-μm-thick slices using a vibratome and mounted onto gelatin-coated slides. Excess water was removed, and a small amount of solution C was added dropwise. After 2 min, the slide was tilted to dry. Slides were rinsed three times in ddH2O and stained for 10 min using a freshly prepared staining solution consisting of one part solution D, one part solution E, and two parts ddH2O. The slides were then washed in ddH2O for 10 min. Sections were dehydrated and cleared in xylene (10023418, Sinopharm, China) and mounted with neutral gum (G8590, Solarbio, China). Images were obtained using an Olympus BX53 microscope with Olympus CellSens Standard software and analyzed by ImageJ Fiji. Dendritic spine density analysis was conducted by counting the dendritic spines along the full length of the apical dendrite of each PC. Twelve cells were counted for statistical analysis by a researcher blinded to the treatment group. Sholl analysis was performed using ImageJ software. The number of intersections between the circles and dendrites were plotted against the radii. The diameter of the primary dendrite was measured at 10 µm from the soma. Cell culture and transfection The HEK293 cell line stably expressing Nav1.6 was described previously [21]. These cells were maintained in DMEM supplemented with 10% (v/v) fetal bovine serum (FBS) and 200 μg/ml G418 (Absin). Normal HEK293 cells were cultured in DMEM containing 10% FBS and 1% penicillin/streptomycin. Plasmid transfections used Lipofectamine 2000 (GlpBio) per manufacturer protocol. Cells were harvested 48 h post-transfection for downstream assays. Co-IP assay Transfected HEK293 cells were harvested and lysed in ice-cold lysis buffer (150 mM NaCl, 30 mM HEPES, 10 mM NaF, 1% (v/v) Triton X-100, 0.01% (w/v) SDS, and complete protease inhibitor mixtures, pH 7.5). The lysates were rotated for 2 h at 4 °C and centrifuged at 12,000 rpm for 20 min. The supernatants were collected and incubated overnight at 4 °C with Anti-Flag Affinity Beads (smart-lifesciences, Catalogue No. SA042001), then washed three times with ice-cold lysis buffer. Finally, the samples were boiled in SDS loading buffer and analyzed by Western blotting. Statistical analysis Data were analyzed using GraphPad Prism 8.0. Normality was assessed with the Shapiro–Wilk test. Normally distributed data are expressed as mean with SEM, while non-normally distributed data are presented as median with interquartile range. Outliers were identified using Dixon’s Q-test at the 95% confidence level. For comparisons between two groups, unpaired or paired Student’s t-tests were used, as appropriate. Comparisons across three or more groups were performed using one-way ANOVA (for normally distributed data with homogeneous variance) or the Kruskal–Wallis test (for non-normal distributions), followed by appropriate post hoc tests. A P-value < 0.05 was considered statistically significant. Declaration of generative AI and AI-assisted technologies in the writing process During the preparation of this work, the authors used ChatGPT and DeepSeek in order to improve language and readability. After using this tool/service, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication. Supporting information S1 Fig. APP-null mice exhibit motor function deficits. A battery of motor function tests was performed in App+/+ (wild-type) and App−/− mice, n = 10 for each group. (A) Open field test: representative trajectory plots and statistics of the distance traveled in the open field. (B) Statistics of grip strength test. (C) Statistics of rotarod test. (D) Representative trajectory maps and schematic measurement of footprint. (E) Statistics of sway length, stride length, stance length and overlap of the footprint assay in App+/+ and App−/− mice. (F) Statistics of mouse body weight. (G) Balance beam test. The time taken for mice to traverse beams of two different widths (6 and 12 mm) and the number of slips were recorded. Bar-charts show the quantification of traversal time and number of slips per mouse. N = 10 for each group. Scale bars are indicated in the images. Student t test: * P < 0.05; ** P < 0.01; *** P < 0.001; ns, non-significant. The data underlying this Figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003513.s001 (TIF) S2 Fig. L7 promoter-driven Cre recombinase expression is restricted to PCs in the cerebellar cortex. Representative sagittal cerebellar sections (20 μm) from wild-type mice injected with rAAV-L7-CRE-P2A-mCherry-WPRE-hGH, showing mCherry (Cre reporter, red) and immunofluorescence for the PC marker calbindin (green). Merged images reveal exclusive co-localization within PCs, confirming recombinant expression specificity. Scale bars: 100 μm (main); 20 μm (insets). https://doi.org/10.1371/journal.pbio.3003513.s002 (TIF) S3 Fig. APP deficiency does not alter PC development or induce degeneration. (A–C) Analyses in 2-month-old and (D–F) 12-month-old App−/− versus App+/+ mice. (A, D) Representative sagittal sections of the cerebellar cortex showing Golgi-stained PCs, with corresponding Sholl analysis of dendritic arborization. Concentric circles (radius increment: 25 µm) are overlaid on the binary-traced PC image to quantify intersections with the dendritic processes. App+/+ (2-month-old), n = 13; App−/− (2-month-old), n = 12; App+/+ (12-month-old), n = 10; App−/− (12-month-old), n = 10. (B, E) Measurement of primary dendrite diameter at a distance of 10 μm from the soma. App+/+ (2-month-old), n = 15; App−/− (2-month-old), n = 19; App+/+ (12-month-old), n = 12; App−/− (12-month-old), n = 12. (C, F) Quantification of dendritic spine density on distal dendrites of PCs. Red arrowheads indicate spines along a distal dendrite. App+/+ (2-month-old), n = 14; App−/− (2-month-old), n = 14; App+/+ (12-month-old), n = 15; App−/− (12-month-old), n = 15. Scale bars are indicated in the images. Data are from 4 mice per group; Student t test, ns: not significant. The data underlying this Figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003513.s003 (TIF) S4 Fig. Validation of Nav1.6 and Nav1.1 contributions to sodium currents in PCs. Whole-cell voltage-clamp recordings from wild-type (App+/+) PCs characterizing voltage-gated sodium current subtypes. (A, D) Transient current (INaT): elicited by depolarizing steps from −90 to +60 mV. (B, E) Persistent current (INaP): evoked using a 1,000-ms voltage ramp from −90 to +30 mV. (C, F) Resurgent current (INaR): induced following a +30 mV step (from Vhold = –90 mV) with repolarizing steps from −60 to +10 mV (10 mV increments). (A–C) Current traces before (black) and during application of 1 µM tetrodotoxin (TTX, pan-NaV blocker; red). (D–F) Current traces before (black) and during co-application (orange) of 200 nM 4,9-anhydrotetrodotoxin (Nav1.6 blocker) and 350 nM ICA-121431 (Nav1.1 inhibitor). Scale bars are shown in the figure. https://doi.org/10.1371/journal.pbio.3003513.s004 (TIF) S5 Fig. Nav1.6 Mediates most of the persistent and resurgent sodium currents in cerebellar PCs. (A) Persistent Na⁺ currents elicited by a 1,000-ms ramp from −90 to +30 mV. Representative traces and statistical bar-charts showing amplitude of persistent INa before or during bath application of Nav1.6 specific blocker 4,9-ahTTX (200 nM), or Nav1.1 specific inhibitor ICA-121431 (ICA, 350 nM) in wild-type (App+/+) mice. 4,9-ahTTX: n = 6 cells, ICA: n = 6 cells. (B) Resurgent Na⁺ currents elicited following a step to +30 mV (from a holding potential of −90 mV) by a series of 10 mV depolarizing voltage steps from −60 to +10 mV. Representative traces and statistical bar-charts showing amplitude of resurgent INa before or during bath application of 4,9-ahTTX or ICA in wild-type mice. 4,9-ahTTX: n = 6 cells, ICA: n = 6 cells. The Nav1.6 or Nav1.1 currents were determined by subtracting the sodium currents recorded during bath application of specific blockers/inhibitors from that before drug application. Scale bars are indicated in the figure. Statistics: Unpaired t test; *** P < 0.001. The data underlying this Figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003513.s005 (TIF) S6 Fig. APP deficiency did not affect the mRNA level of Scn8a in cerebellar PCs. QPCR analysis of LCMed cerebellar PCs in App+/+ or App−/− mice. Statistics: App+/+, n = 5; App−/−, n = 4. Unpaired t test; ns, non-significant. The data underlying this Figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003513.s006 (TIF) S7 Fig. Antibody validation for Nav1.6 and Nav1.1-EGFP in transfected HEK293 cells. Total lysates of HEK293 cells expressing either Nav1.6 or Nav1.1-GFP were immunoblotted with Nav1.6 (Alomone Labs #ASC-009) or EGFP (Roche Applied Science, #11814460001) antibody. https://doi.org/10.1371/journal.pbio.3003513.s007 (TIF) S8 Fig. PoTX potentiates persistent sodium currents in APP-null PCs. Whole-cell voltage-clamp recordings of persistent sodium currents in App−/− PCs before (black) and during bath application of 30 nM PoTX (red). Currents were elicited by a 1000-ms ramp to +30 mV from a holding potential of −90 mV. Scale bars as indicated. https://doi.org/10.1371/journal.pbio.3003513.s008 (TIF) S1 Raw Images. Raw images for blots. https://doi.org/10.1371/journal.pbio.3003513.s009 (TIF) S1 Data. Source data and statistical details for figures. https://doi.org/10.1371/journal.pbio.3003513.s010 (XLSX) Acknowledgments The authors sincerely thank Dr. Gavin S. Dawe for generously providing the hAPP695 plasmid and the Nav1.6-expressing HEK293 cell line, as well as for his expertise in APP biology. We are also grateful to Dr. Lan Bao and Dr. You-Sheng Shu for their helpful discussions and invaluable insights into voltage-gated sodium channels.

Rönneburg, Tilman;Taliadoros, Demetris;Olsson, Anna;Magnusson, Sara;Huser, Linn;Fuad, Muhammad Nafiz Ikhwan Bin Nor;Everitt, Turid;Martín-Hernández, Giselle C.;Theodorou, Panagiotis;Cederberg, Björn;Paxton, Robert J.;Seidelmann, Karsten;Webster, Matthew T.

doi: 10.1371/journal.pbio.3003458pmid: 41183133

Magalhães, Andreia D.;Emmenegger, Marc;De Cecco, Elena;Carta, Manfredi;Frontzek, Karl;Chincisan, Andra;Guo, Jingjing;Hornemann, Simone;Aguzzi, Adriano

doi: 10.1371/journal.pbio.3003488pmid: 41296806

Introduction Tau is a microtubule binding protein involved in cytoskeletal dynamics and expressed in neurons [1,2] and in extraneural tissues, including kindey [1–5]. It plays a pivotal role in a variety of neurodegenerative diseases, including Alzheimer’s disease (AD) [6], progressive supranuclear palsy, and various syndromes collectively referred to as tauopathies [7]. The presence of brain neurofibrillary tau tangles correlates with cognitive decline [8], and plasma measurements of total and phosphorylated tau have emerged as promising biomarkers for the detection and monitoring of AD progression [9–12]. Moreover, active and passive immunization with antibodies against a wide range of tau epitopes can reduce pathology and functional decline in animal models of tauopathies [13,14] and are being tested in clinical trials of neurodegenerative diseases [15]. Natural autoantibodies are immunoglobulins generated against self-antigens in the absence of external antigen stimulation [16]. They are a normal part of the immunoglobulin repertoire and have physiological roles in homeostasis and surveillance, including the clearance of cellular debris, anti-inflammatory activity, and first-line defense against pathogens [17]. However, in certain situations, natural autoantibodies can also cause pathological autoimmunity. The distinction between homeostatic and pathological autoantibodies is sometimes unclear and may depend on an individual’s physiological state [18]. The study of natural autoantibodies can therefore inform about properties of their targets, e.g., unrecognized protective or contributing roles in disease [19]. Some natural autoantibodies can cause neurological disorders, such as antibodies targeting the N-methyl-D-aspartate receptor (NMDAR) in encephalitis [20] or antibodies targeting aquaporin-4 in neuromyelitis optica [21]. Conversely, natural antibodies against amyloid-β have been suggested to protect and slow the progression of AD; aducanumab, a monoclonal antibody developed from B-cells of cognitively normal older age individuals, has been studied as a treatment for AD [22]. Anti-tau autoantibodies have been detected in plasma of patients with AD [23] and Parkinson’s disease [24], but also in non-neurodegeneration controls [23]. The effects, if any, of natural anti-tau autoantibodies in modulating the risk of developing neurodegenerative diseases are unknown. The study of individuals with anti-tau autoantibodies could clarify their potential as modifiers or biomarkers of disease. Here, we tested 40,497 plasma samples from healthy blood donors and from patients admitted to a university hospital in the frame of a two-sites cross-sectional study. We found that anti-tau autoimmunity is highly specific and surprisingly frequent, with its prevalence increasing with age. Unexpectedly, natural anti-tau autoantibodies were associated with a previously unrecognized syndrome comprising kidney and urinary disorders. Results Prevalence of naturally occurring plasma anti-tau autoantibodies We designed a cross-sectional study to investigate naturally occurring plasma IgG autoantibodies against the microtubule binding domain of tau protein (MTBD-tau) corresponding to the human truncated 4R-tau (residues 244–372 relative to 2N4R human tau). Plasma samples from 32,291 university hospital patients (age ≥18 years) and 8,206 healthy blood donors were screened by miniaturized Enzyme-Linked Immunosorbent Assay (ELISA, Fig 1A) [25–27]. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Study overview and seroprevalence of anti-MTBD-tau IgG autoantibodies. (A) Flowchart of the samples of the hospital patients’ cohort. (B, C) Distribution of −log10(EC50) values obtained from the microELISA screen of hospital (B) and blood donor (C) plasma samples. (D) Age- and sex-adjusted risk ratios and 95% confidence intervals (CI; I bars) for the detection of anti-tau autoantibodies in hospital and blood-bank plasma samples. (E) Replicability of microELISA duplicates with independent estimation of the −log10(EC50) values. Dashed lines: cut-off value of −log10(EC50) = 1.8. (F) −log10(EC50) values of samples tested by microELISA against MTBD-tau and amyloid-β pyroglutamate. (G) Same as shown in (F), but for samples tested against MTBD-tau and the cellular prion protein (PrPC). The underlying numerical data for panels (B), (C), (E), (F) and (G) can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003488.g001 We excluded 7,952 and 1,616 non-informative samples from patients and blood donors, respectively (fitting error >20% −log10(EC50) or high background), and analyzed 24,339 patient samples and 6,590 healthy blood-donor samples (Fig 1A–C). A titer of −log10(EC50) ≥ 1.8, approximately corresponding to a nominal dilution of >1/64 or an EC50 ≥ 26, was empirically selected as a cutoff to call tau-autoreactive samples (henceforth named ατ+) [25]. This threshold represents the inflection point in our serial dilution assays above background and below saturation and was chosen to best balance sensitivity and specificity. 1,169 hospital samples (4.8%) but only 104 healthy donor samples (1.6%) were ατ+ (Fig 1A–C). Hence, anti-tau immunoreactivity was more prevalent in unselected hospital patients than in healthy individuals (P < 0.001). Demographic data was available for 4,157 of the 6,590 blood donors. Median ages were 42 (IQR 29–54) and 55 years (IQR: 40–69) for healthy donors and hospital patients, respectively (P < 0.001). Of the healthy blood donors, 40.9% (n = 1,698) were women, whereas for the hospital group, 47.7% (n = 11,609) of the patients were women (P < 0.001). Multivariate log-binomial regression [28,29] adjusted for age and sex showed that hospital patients had a 2.3× higher risk than healthy donors to be ατ+ (adjusted risk ratio [aRR] 2.30, 95% confidence interval [CI] 1.83–2.92, P < 0.001, Fig 1D). The replicability of the microELISA screen was found to be high (R2 = 0.84, P < 0.001, Fig 1E). There was no cross-reactivity to two other proteins implicated in neurodegeneration, amyloid-β pyroglutamate and the cellular prion protein (PrPC) [27]. Of 12,297 patient samples, 604 samples were positive against MTBD-tau and 5 against amyloid-β pyroglutamate but none was cross-reactive against both targets (P = 1, χ2 test) (Fig 1F), Moreover, of 13,099 patient samples, 694 were reactive against MTBD-tau and 15 against PrPC, but again none was cross-reactive against both targets (P = 0.734, χ2 test) (Fig 1G). Specificity and biological activity of ατ+ samples To investigate specificity, we purified anti-tau autoantibodies by MTBD-tau affinity chromatography from four individual ατ+ patients (P1–P4) and from a pool of six ατ+ sample. Purified anti-MTBD-tau autoantibody samples had EC50 values of 0.288–14.45 µg/ml whereas the mouse anti-MTBD-tau antibody RD4 [30] had EC50 = 0.002 µg/ml (Fig 2A). In a soluble-competition immunoassay (Fig 2B), purified ατ+ autoantibodies showed a concentration-dependent binding to recombinant MTBD-tau purified by cation exchange and size exclusion chromatography and to a pool of 8 synthetic MTBD-tau peptides but not to albumin or an unrelated synthetic peptide (Fig 2C). To probe for polyreactivity, purified ατ+ autoantibodies were tested against several structurally unrelated autoantigens and bacterial proteins, including MTBD-tau, albumin, cardiolipin, DNA, insulin, and lipopolysaccharides (LPS). Anti-tau autoantibodies were reactive against MTBD-tau but not against any of the other antigens or uncoated plates (Fig 2D). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Biophysical characterization of samples from ατ+ patients. (A) Indirect ELISA of purified anti-MTBD-tau autoantibodies from four ατ+ samples and from 6 pooled ατ+ samples. All antibodies were assayed at the same concentration to compare their EC50 to that of RD4. Relatlimab (anti-LAG3) was used as negative control. EC50 values are indicated in the figure. n.d.: not determined. (B) Principle of competition ELISA. (C) Competition ELISA of purified anti-tau autoantibodies from four ατ+ patients against albumin, recombinant MTBD-tau, a pool of synthetic peptides spanning the sequence of MTBD-tau and synthetic TREM2. RD4 antibody: positive control. Mean values ± SD of two replicates. (D) Indirect ELISA to assess the reactivity of purified anti-tau autoantibodies against albumin, cardiolipin, double-stranded DNA, insulin, lipopolysaccharides (LPS), MTBD-tau and uncoated plates. Positive controls were as follows: RD4 antibody for MTBD-tau, anti-DNP antibody for cardiolipin, albumin and DNA and the IKC pool of 20 heparin plasma samples for insulin, LPS and uncoated plates. Mean values ± SD of two replicates are shown. (E) Representative immunofluorescence images of SH-SY5Y cells expressing EGFP-0N4R-tau using affinity-purified anti-tau autoantibodies. HT7 pan-tau antibody: positive control. Secondary only as negative control: omission of primary antibody. Scale bar: 60 µm. (F) Western blot of cell lysates from wild-type (SH-SY5Y WT) or tauP301L/S320F-overexpressing cells (SH-SY5Y tau) using purified anti-tau autoantibodies from four ατ+ patients. Positive control: RD4. IgG antibody only as negative control: omission of primary antibody. (G) IgG subclass typing of 13 ατ+ plasma samples. Gray and white boxes: reactive and non-reactive samples, respectively. (H) κ/λ light chain typing of the samples in G. (I) Epitope mapping of the samples in G against 25mer MTBD-tau peptides with 10 residues overlap. Vertical axis: sequences of the MTBD-tau peptides covering the sequence of MTBD-tau. The underlying numerical data for panels (A), (C), (D), (G), (H) and (I) can be found in S1 Data. Raw images of the Western blots in (F) can be found in S1 Raw Images. https://doi.org/10.1371/journal.pbio.3003488.g002 In immunofluorescent stainings, purified ατ autoantibodies co-localized to cytoplasmic EGFP-0N4RTau in SH-SY5Y cells and showed similar binding patterns to an anti-tau mouse monoclonal antibody H7, but not to non-transfected cells (Fig 2E). On western blots, purified ατ autoantibodies detected tau-specific bands in cell lysates of SH-SY5Y cells overexpressing tauP301L/S320F, but not in parental (wt) SH-SY5Y cells (Fig 2F and S1 Raw Images). Hence, the immunoreactivity of ατ+ samples was highly specific for tau. All IgG subclasses and both κ and λ light chains were present in ατ+ patients’ samples (Fig 2G and 2H). Epitope mapping and light-chain typing revealed a polytypic response in at least 8 out of the 13 ατ+ samples (Fig 2H and 2I). To investigate whether purified ατ+ autoantibodies interfere with the aggregation of MTBD-tau, samples from patients P7 and P8 were tested in an in vitro tau aggregation assay [31]. MTBD-tau aggregation was induced by heparin and shaking and monitored using Thioflavin T (ThT; Fig 3A). The presence of ατ+ autoantibodies reduced the plateau of the ThT fluorescence signal in the kinetic trace by about half, whereas antibodies purified by protein G affinity chromatography from an ατ− patient had no effect (Fig 3A). Hence, ατ+ autoantibodies were able to specifically bind to and inhibit MTBD-tau aggregation in vitro. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Natural ατ autoantibodies inhibit tau aggregation and impair tau detection. (A) Kinetic aggregation curves of MTBD-tau followed by ThT fluorescence in the absence (n = 12) or presence of purified ατ autoantibodies (patient P7 (n = 4) and P8 (n = 2)) or antibodies from an ατ-sample (n = 8). Stoichiometric ratios as indicated. Gray lines indicate individual replicates and black lines the average of the replicates. (B) Principle of the competition sandwich ELISA. (C) Competitive sandwich ELISA to assess the ability of purified anti-tau autoantibodies with matching (P4–P6) and non-matching (P2 and P3) epitopes to the commercial detection antibody, ab64193, to impair the detection of free tau441 spiked in plasma (dark blue). Serial dilutions of purified anti-tau antibodies are shown in (C) and binding at highest anti-tau antibody concentrations in (D). The underlying numerical values of panels (A), (C), and (D) can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003488.g003 Plasma tau is a promising biomarker of progression for several neurological diseases [9–12,32]. We examined the effects of ατ autoantibodies onto the performance of a plasma tau immunoassay (Fig 3B). Purified ατ autoantibodies from 5 patients (P2–P6) were added to tau441-spiked plasma. After incubation, the amount of free tau441 was analyzed by ELISA using commercial anti-tau antibodies for capture (BT2, epitope on human tau441: residues 194–198) and detection (ab64193, epitope on non-phosphorylated and phosphorylated human tau441 surrounding residue 262) [33]. Purified ατ autoantibodies P4–P6 induced a concentration-dependent impairment of detection of tau441 hampering the detection of plasma-spiked tau441 by approximately 10-fold at higher anti-MTBD-tau autoantibody concentration (Fig 3C and 3D). In contrast, P2–P3 did not show any significant impairment of the detection of plasma-spiked tau441. This variability of interference is explained by the binding epitopes on tau441. P4-P6 occupy tau residues mapping to the binding epitope of ab64193, whereas P2-P3 occupy residues mapping outside the epitope of ab64193 (Figs 3C and 3D and S1). Hence, the presence of ατ+ autoantibodies can interfere with the detection of plasma tau in immunoassays depending on the combination of epitopes of the patient samples and of commercial antibodies used in immunoassays. Demographic characteristics of tau-immunoreactive patients The age of ατ+ patients (median: 58 years; IQR: 43−71) was significantly higher than that of ατ− (median: 55; IQR: 40−68; P < 0.001, Fig 4A). The prevalence of ατ immunoreactivity increased with age, from 3.9% in patients aged <29 years to 7.6% in patients aged >90 years (P < 0.001, Fig 4B). A log-binomial regression model [28,29] estimated that the RR for the presence of anti-MTBD-tau autoantibodies was highest for patients aged 70−99 years (RR 1.26, 95% CI 1.11–1.43, P < 0.001, Fig 4C) and for women (RR 1.20, 95% CI 1.07–1.39, P = 0.002, Fig 4C). Due to the association of anti-MTBD-tau autoantibodies with increased age and female sex, all further RRs were calculated using a multivariate log-binomial regression model adjusted for age and sex. To identify potential correlations between ατ+ and specific diseases, we analyzed the admission to clinical departments. The highest RRs for ατ+ were found for angiology (aRR 1.84, 95% CI 1.21–2.63, P = 0.002) and nephrology (aRR 1.50, 95% CI 1.16–1.89, P = 0.001, Fig 4D and 4E), whereas no significant difference was found between the percentage of ατ+ and ατ− patients from the department of neurology (Fig 4D). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. Demographic characteristics of hospital patients’ cohort. (A) Age and sex pyramid of positive and negative individuals. (B) Percentage of positives among hospital patients across decadic age groups. The numbers on top of each bar correspond to the positivity rates. χ2 test for trend in proportions. (C) RR ± 95% CI for ατ+ autoantibodies according to sex and age groups. Asterisks: P < 0.05 after Bonferroni correction. (D) Breakdown of samples by clinical department. Asterisks: P < 0.05 (two-proportions z-test with Bonferroni correction). (E) aRR ± 95% CI for ατ+ samples by clinical department. Asterisks: P < 0.05 after Bonferroni correction. https://doi.org/10.1371/journal.pbio.3003488.g004 Neurological disorders and anti-tau autoimmunity We next mined pseudonymized clinical diagnoses from the clinical records of the USZ pertaining to the diagnoses of 23,375 patients as represented by ICD-10 codes (Fig 5A). Given the involvement of tau in neurodegenerative diseases [6], we focused on evaluating the association between anti-MTBD-tau autoantibodies and neurological diseases, which we categorized in 23 main groups of disorders. No associations between ατ+ and neurological diseases were identified (Fig 5A). We further performed a targeted screen using plasma samples from 47 patients with AD and 68 similarly aged non-AD patients selected from our plasma biobank (median age of AD patients 78 years, IQR 70,5–86, and median age of control patients 81 years, IQR 71–85, S1 Table). Samples were tested for anti-MTBD-tau IgG autoantibodies, as in the primary screen, and additionally for anti-MTBD-tau IgA autoantibodies and anti-full-length-tau (tau441) IgG and IgA autoantibodies. No significant difference in reactivity was observed between the plasma samples from AD patients and non-AD controls in this small convenience cohort (Fig 5B), which is in line with the previous finding that the presence of autoantibodies targeting MTBD-tau is unrelated to AD or other neurological disorders. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. aRR of MTBD-tau-autoreactivity for major groups of neurological disorders and reactivity profiles for AD screen. (A) aRR and 95% CI (I bars) for ατ+ autoantibodies according to different groups of neurological disorders. No P values remained significant after Bonferroni correction. (B) Boxplots showing the 25th, 50th (median), and 75th percentiles of the reactivity profiles for AD and control patients (Mann–Whitney U test). The underlying numerical values of panel (B) can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003488.g005 Systemic disorders and anti-tau autoimmunity We next assessed possible associations between tau autoimmunity and extraneural diseases, which we binned into 27 main groups of disorders (Fig 6A). After adjustment for multiple comparisons, ατ immunoreactivity showed significant associations with vascular disorders (aRR 1.51, 95% CI 1.28–1.77, P < 0.001), nutritional disorders (aRR 1.31, 95% CI 1.14–1.50, P < 0.001), anemia (aRR 1.49, 95% CI 1.21–1.82, P < 0.001), kidney disorders (aRR 1.27, 95% CI 1.10–1.45, P = 0.001) and urinary disorders (aRR 1.40, 95% CI 1.20–1.63, P < 0.001) (Fig 6A), whereas no association was observed among patients with autoimmune disorders (S2 Fig). There was no difference in the coincidence of all comorbidities between ατ+ samples (5, 0.46%) and ατ− samples (69, 0.33%) (P = 0.633). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 6. Association of MTBD-tau IgG autoantibodies with systemic disorders and clinical laboratory data. (A) aRR ± 95% CI (bars) for ατ+ autoantibodies grouped 27 different main systemic clinical conditions. Asterisks: P < 0.05 after Bonferroni correction. (B) −log10(P value) of the log-binomial regression for the presence of plasma MTBD-tau IgG autoantibodies for 276 disorders according to ICD-10 codes and adjusted for age and sex. Triangles pointing upwards and downwards: positive and negative coefficients, respectively; gray dashed line: P < 0.05 after Bonferroni correction (here and henceforth). P values significant after Bonferroni adjustment are labeled with the ICD codes. (C) Same as (A), but according to individual disease entities from (B). (D) Bayesian logistic regression adjusted for age and sex on ICD-10 codes significantly associated in B. and using −logit10(EC50) as continuous outcome (odds ratio ± 95% credible intervals). This analysis confirms the positive association of these ICD-10 codes with ατ reactivity (from B). (E) −log10(P value) of the log-binomial regression for the presence of plasma MTBD-tau IgG autoantibodies and clinical laboratory parameters. P < 0.05 are labeled with the clinical laboratory parameters. Gray dashed line: P < 0.05 after Bonferroni correction. eGFR CKD-EPI 2009: estimated glomerular filtration rate using the Chronic Kidney Disease Epidemiology Collaboration 2009 equation; eGFR BIS1: estimated glomerular filtration rate using the older adults Berlin Initiative Study 1 equation. INR: international normalized ratio. LDL: low-density lipoprotein. HDL: high-density lipoprotein. (F) Calculated prevalence of different ICD-10 codes in the total cohort (dark blue), ατ− (intermediate blue) and ατ+ samples (light blue). Asterisks: P < 0.05 after Bonferroni correction (two-proportions z-test). https://doi.org/10.1371/journal.pbio.3003488.g006 To address potential biases arising from grouping and selecting major categories, we explored the association between ατ+ and individual ICD-10 codes. Across 276 individual ICD-10 codes and after correction for multiple comparisons, eight exhibited significant associations with ατ+ (Fig 6B and 6C). These included “E61–deficiency of other nutrient elements” (aRR 2.22, 95% CI 1.54–3.06, P < 0.001), “I74–arterial embolism and thrombosis” (aRR 2.09, 95% CI 1.40–2.94, P < 0.001), “N30–cystitis” (aRR 1.84, 95% CI 1.32–2.47, P < 0.001), “I70–atherosclerosis” (aRR 1.57, 95% CI 1.29–1.89, P < 0.001), “D50–iron deficiency anemia” (aRR 1.50, 95% CI 1.21–1.82, P < 0.001), “N39–other urinary disorders” (aRR 1.40, 95% CI 1.19–1.64, P < 0.001), “B96–other bacterial agents as the cause of diseases” (aRR 1.40, 95% CI 1.17–1.66, P < 0.001) and “N18–chronic kidney disease” (aRR 1.38, 95% CI 1.18–1.60, P < 0.001; Fig 6C). Conversely, we found no significant negative association between tau autoreactivity and any ICD-10 codes that could suggest a decreased risk for a specific disease in ατ+ patients (Fig 6B). To verify relevant associations, we used Bayesian logistic regression using logit-transformed EC50 values (−logit[EC50]) as a continuous outcome. All the previously referred eight ICD-10 codes showed positive associations with −logit[EC50] (Fig 6D). To validate these conditions as significant associations with plasma MTBD-tau IgG autoantibodies, we examined the association between ατ+ and 106 commonly requested laboratory parameters (S2 Table) using data available for 24,177 patients, independently of any ICD-10 classifiers. After correction for multiple comparisons, the laboratory markers “Leukocytes, urine”, “Potassium,” and “Urea” were positively associated with tau autoimmunity (Fig 6E). The association with increasing levels of urea and potassium suggests a link to kidney failure, whereas leukocytes in urine are a feature of urinary tract infections. These findings support the hypothesis that anti-tau autoimmunity correlates with such disorders. Accordingly, the prevalence of chronic kidney disease was 12.4% in ατ− and 17.7% in ατ+ patients, whereas that of other urinary disorders was 10.3% in ατ− and 15.1% in ατ+ patients (Fig 6F). Prevalence of naturally occurring plasma anti-tau autoantibodies We designed a cross-sectional study to investigate naturally occurring plasma IgG autoantibodies against the microtubule binding domain of tau protein (MTBD-tau) corresponding to the human truncated 4R-tau (residues 244–372 relative to 2N4R human tau). Plasma samples from 32,291 university hospital patients (age ≥18 years) and 8,206 healthy blood donors were screened by miniaturized Enzyme-Linked Immunosorbent Assay (ELISA, Fig 1A) [25–27]. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Study overview and seroprevalence of anti-MTBD-tau IgG autoantibodies. (A) Flowchart of the samples of the hospital patients’ cohort. (B, C) Distribution of −log10(EC50) values obtained from the microELISA screen of hospital (B) and blood donor (C) plasma samples. (D) Age- and sex-adjusted risk ratios and 95% confidence intervals (CI; I bars) for the detection of anti-tau autoantibodies in hospital and blood-bank plasma samples. (E) Replicability of microELISA duplicates with independent estimation of the −log10(EC50) values. Dashed lines: cut-off value of −log10(EC50) = 1.8. (F) −log10(EC50) values of samples tested by microELISA against MTBD-tau and amyloid-β pyroglutamate. (G) Same as shown in (F), but for samples tested against MTBD-tau and the cellular prion protein (PrPC). The underlying numerical data for panels (B), (C), (E), (F) and (G) can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003488.g001 We excluded 7,952 and 1,616 non-informative samples from patients and blood donors, respectively (fitting error >20% −log10(EC50) or high background), and analyzed 24,339 patient samples and 6,590 healthy blood-donor samples (Fig 1A–C). A titer of −log10(EC50) ≥ 1.8, approximately corresponding to a nominal dilution of >1/64 or an EC50 ≥ 26, was empirically selected as a cutoff to call tau-autoreactive samples (henceforth named ατ+) [25]. This threshold represents the inflection point in our serial dilution assays above background and below saturation and was chosen to best balance sensitivity and specificity. 1,169 hospital samples (4.8%) but only 104 healthy donor samples (1.6%) were ατ+ (Fig 1A–C). Hence, anti-tau immunoreactivity was more prevalent in unselected hospital patients than in healthy individuals (P < 0.001). Demographic data was available for 4,157 of the 6,590 blood donors. Median ages were 42 (IQR 29–54) and 55 years (IQR: 40–69) for healthy donors and hospital patients, respectively (P < 0.001). Of the healthy blood donors, 40.9% (n = 1,698) were women, whereas for the hospital group, 47.7% (n = 11,609) of the patients were women (P < 0.001). Multivariate log-binomial regression [28,29] adjusted for age and sex showed that hospital patients had a 2.3× higher risk than healthy donors to be ατ+ (adjusted risk ratio [aRR] 2.30, 95% confidence interval [CI] 1.83–2.92, P < 0.001, Fig 1D). The replicability of the microELISA screen was found to be high (R2 = 0.84, P < 0.001, Fig 1E). There was no cross-reactivity to two other proteins implicated in neurodegeneration, amyloid-β pyroglutamate and the cellular prion protein (PrPC) [27]. Of 12,297 patient samples, 604 samples were positive against MTBD-tau and 5 against amyloid-β pyroglutamate but none was cross-reactive against both targets (P = 1, χ2 test) (Fig 1F), Moreover, of 13,099 patient samples, 694 were reactive against MTBD-tau and 15 against PrPC, but again none was cross-reactive against both targets (P = 0.734, χ2 test) (Fig 1G). Specificity and biological activity of ατ+ samples To investigate specificity, we purified anti-tau autoantibodies by MTBD-tau affinity chromatography from four individual ατ+ patients (P1–P4) and from a pool of six ατ+ sample. Purified anti-MTBD-tau autoantibody samples had EC50 values of 0.288–14.45 µg/ml whereas the mouse anti-MTBD-tau antibody RD4 [30] had EC50 = 0.002 µg/ml (Fig 2A). In a soluble-competition immunoassay (Fig 2B), purified ατ+ autoantibodies showed a concentration-dependent binding to recombinant MTBD-tau purified by cation exchange and size exclusion chromatography and to a pool of 8 synthetic MTBD-tau peptides but not to albumin or an unrelated synthetic peptide (Fig 2C). To probe for polyreactivity, purified ατ+ autoantibodies were tested against several structurally unrelated autoantigens and bacterial proteins, including MTBD-tau, albumin, cardiolipin, DNA, insulin, and lipopolysaccharides (LPS). Anti-tau autoantibodies were reactive against MTBD-tau but not against any of the other antigens or uncoated plates (Fig 2D). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Biophysical characterization of samples from ατ+ patients. (A) Indirect ELISA of purified anti-MTBD-tau autoantibodies from four ατ+ samples and from 6 pooled ατ+ samples. All antibodies were assayed at the same concentration to compare their EC50 to that of RD4. Relatlimab (anti-LAG3) was used as negative control. EC50 values are indicated in the figure. n.d.: not determined. (B) Principle of competition ELISA. (C) Competition ELISA of purified anti-tau autoantibodies from four ατ+ patients against albumin, recombinant MTBD-tau, a pool of synthetic peptides spanning the sequence of MTBD-tau and synthetic TREM2. RD4 antibody: positive control. Mean values ± SD of two replicates. (D) Indirect ELISA to assess the reactivity of purified anti-tau autoantibodies against albumin, cardiolipin, double-stranded DNA, insulin, lipopolysaccharides (LPS), MTBD-tau and uncoated plates. Positive controls were as follows: RD4 antibody for MTBD-tau, anti-DNP antibody for cardiolipin, albumin and DNA and the IKC pool of 20 heparin plasma samples for insulin, LPS and uncoated plates. Mean values ± SD of two replicates are shown. (E) Representative immunofluorescence images of SH-SY5Y cells expressing EGFP-0N4R-tau using affinity-purified anti-tau autoantibodies. HT7 pan-tau antibody: positive control. Secondary only as negative control: omission of primary antibody. Scale bar: 60 µm. (F) Western blot of cell lysates from wild-type (SH-SY5Y WT) or tauP301L/S320F-overexpressing cells (SH-SY5Y tau) using purified anti-tau autoantibodies from four ατ+ patients. Positive control: RD4. IgG antibody only as negative control: omission of primary antibody. (G) IgG subclass typing of 13 ατ+ plasma samples. Gray and white boxes: reactive and non-reactive samples, respectively. (H) κ/λ light chain typing of the samples in G. (I) Epitope mapping of the samples in G against 25mer MTBD-tau peptides with 10 residues overlap. Vertical axis: sequences of the MTBD-tau peptides covering the sequence of MTBD-tau. The underlying numerical data for panels (A), (C), (D), (G), (H) and (I) can be found in S1 Data. Raw images of the Western blots in (F) can be found in S1 Raw Images. https://doi.org/10.1371/journal.pbio.3003488.g002 In immunofluorescent stainings, purified ατ autoantibodies co-localized to cytoplasmic EGFP-0N4RTau in SH-SY5Y cells and showed similar binding patterns to an anti-tau mouse monoclonal antibody H7, but not to non-transfected cells (Fig 2E). On western blots, purified ατ autoantibodies detected tau-specific bands in cell lysates of SH-SY5Y cells overexpressing tauP301L/S320F, but not in parental (wt) SH-SY5Y cells (Fig 2F and S1 Raw Images). Hence, the immunoreactivity of ατ+ samples was highly specific for tau. All IgG subclasses and both κ and λ light chains were present in ατ+ patients’ samples (Fig 2G and 2H). Epitope mapping and light-chain typing revealed a polytypic response in at least 8 out of the 13 ατ+ samples (Fig 2H and 2I). To investigate whether purified ατ+ autoantibodies interfere with the aggregation of MTBD-tau, samples from patients P7 and P8 were tested in an in vitro tau aggregation assay [31]. MTBD-tau aggregation was induced by heparin and shaking and monitored using Thioflavin T (ThT; Fig 3A). The presence of ατ+ autoantibodies reduced the plateau of the ThT fluorescence signal in the kinetic trace by about half, whereas antibodies purified by protein G affinity chromatography from an ατ− patient had no effect (Fig 3A). Hence, ατ+ autoantibodies were able to specifically bind to and inhibit MTBD-tau aggregation in vitro. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Natural ατ autoantibodies inhibit tau aggregation and impair tau detection. (A) Kinetic aggregation curves of MTBD-tau followed by ThT fluorescence in the absence (n = 12) or presence of purified ατ autoantibodies (patient P7 (n = 4) and P8 (n = 2)) or antibodies from an ατ-sample (n = 8). Stoichiometric ratios as indicated. Gray lines indicate individual replicates and black lines the average of the replicates. (B) Principle of the competition sandwich ELISA. (C) Competitive sandwich ELISA to assess the ability of purified anti-tau autoantibodies with matching (P4–P6) and non-matching (P2 and P3) epitopes to the commercial detection antibody, ab64193, to impair the detection of free tau441 spiked in plasma (dark blue). Serial dilutions of purified anti-tau antibodies are shown in (C) and binding at highest anti-tau antibody concentrations in (D). The underlying numerical values of panels (A), (C), and (D) can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003488.g003 Plasma tau is a promising biomarker of progression for several neurological diseases [9–12,32]. We examined the effects of ατ autoantibodies onto the performance of a plasma tau immunoassay (Fig 3B). Purified ατ autoantibodies from 5 patients (P2–P6) were added to tau441-spiked plasma. After incubation, the amount of free tau441 was analyzed by ELISA using commercial anti-tau antibodies for capture (BT2, epitope on human tau441: residues 194–198) and detection (ab64193, epitope on non-phosphorylated and phosphorylated human tau441 surrounding residue 262) [33]. Purified ατ autoantibodies P4–P6 induced a concentration-dependent impairment of detection of tau441 hampering the detection of plasma-spiked tau441 by approximately 10-fold at higher anti-MTBD-tau autoantibody concentration (Fig 3C and 3D). In contrast, P2–P3 did not show any significant impairment of the detection of plasma-spiked tau441. This variability of interference is explained by the binding epitopes on tau441. P4-P6 occupy tau residues mapping to the binding epitope of ab64193, whereas P2-P3 occupy residues mapping outside the epitope of ab64193 (Figs 3C and 3D and S1). Hence, the presence of ατ+ autoantibodies can interfere with the detection of plasma tau in immunoassays depending on the combination of epitopes of the patient samples and of commercial antibodies used in immunoassays. Demographic characteristics of tau-immunoreactive patients The age of ατ+ patients (median: 58 years; IQR: 43−71) was significantly higher than that of ατ− (median: 55; IQR: 40−68; P < 0.001, Fig 4A). The prevalence of ατ immunoreactivity increased with age, from 3.9% in patients aged <29 years to 7.6% in patients aged >90 years (P < 0.001, Fig 4B). A log-binomial regression model [28,29] estimated that the RR for the presence of anti-MTBD-tau autoantibodies was highest for patients aged 70−99 years (RR 1.26, 95% CI 1.11–1.43, P < 0.001, Fig 4C) and for women (RR 1.20, 95% CI 1.07–1.39, P = 0.002, Fig 4C). Due to the association of anti-MTBD-tau autoantibodies with increased age and female sex, all further RRs were calculated using a multivariate log-binomial regression model adjusted for age and sex. To identify potential correlations between ατ+ and specific diseases, we analyzed the admission to clinical departments. The highest RRs for ατ+ were found for angiology (aRR 1.84, 95% CI 1.21–2.63, P = 0.002) and nephrology (aRR 1.50, 95% CI 1.16–1.89, P = 0.001, Fig 4D and 4E), whereas no significant difference was found between the percentage of ατ+ and ατ− patients from the department of neurology (Fig 4D). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. Demographic characteristics of hospital patients’ cohort. (A) Age and sex pyramid of positive and negative individuals. (B) Percentage of positives among hospital patients across decadic age groups. The numbers on top of each bar correspond to the positivity rates. χ2 test for trend in proportions. (C) RR ± 95% CI for ατ+ autoantibodies according to sex and age groups. Asterisks: P < 0.05 after Bonferroni correction. (D) Breakdown of samples by clinical department. Asterisks: P < 0.05 (two-proportions z-test with Bonferroni correction). (E) aRR ± 95% CI for ατ+ samples by clinical department. Asterisks: P < 0.05 after Bonferroni correction. https://doi.org/10.1371/journal.pbio.3003488.g004 Neurological disorders and anti-tau autoimmunity We next mined pseudonymized clinical diagnoses from the clinical records of the USZ pertaining to the diagnoses of 23,375 patients as represented by ICD-10 codes (Fig 5A). Given the involvement of tau in neurodegenerative diseases [6], we focused on evaluating the association between anti-MTBD-tau autoantibodies and neurological diseases, which we categorized in 23 main groups of disorders. No associations between ατ+ and neurological diseases were identified (Fig 5A). We further performed a targeted screen using plasma samples from 47 patients with AD and 68 similarly aged non-AD patients selected from our plasma biobank (median age of AD patients 78 years, IQR 70,5–86, and median age of control patients 81 years, IQR 71–85, S1 Table). Samples were tested for anti-MTBD-tau IgG autoantibodies, as in the primary screen, and additionally for anti-MTBD-tau IgA autoantibodies and anti-full-length-tau (tau441) IgG and IgA autoantibodies. No significant difference in reactivity was observed between the plasma samples from AD patients and non-AD controls in this small convenience cohort (Fig 5B), which is in line with the previous finding that the presence of autoantibodies targeting MTBD-tau is unrelated to AD or other neurological disorders. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. aRR of MTBD-tau-autoreactivity for major groups of neurological disorders and reactivity profiles for AD screen. (A) aRR and 95% CI (I bars) for ατ+ autoantibodies according to different groups of neurological disorders. No P values remained significant after Bonferroni correction. (B) Boxplots showing the 25th, 50th (median), and 75th percentiles of the reactivity profiles for AD and control patients (Mann–Whitney U test). The underlying numerical values of panel (B) can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3003488.g005 Systemic disorders and anti-tau autoimmunity We next assessed possible associations between tau autoimmunity and extraneural diseases, which we binned into 27 main groups of disorders (Fig 6A). After adjustment for multiple comparisons, ατ immunoreactivity showed significant associations with vascular disorders (aRR 1.51, 95% CI 1.28–1.77, P < 0.001), nutritional disorders (aRR 1.31, 95% CI 1.14–1.50, P < 0.001), anemia (aRR 1.49, 95% CI 1.21–1.82, P < 0.001), kidney disorders (aRR 1.27, 95% CI 1.10–1.45, P = 0.001) and urinary disorders (aRR 1.40, 95% CI 1.20–1.63, P < 0.001) (Fig 6A), whereas no association was observed among patients with autoimmune disorders (S2 Fig). There was no difference in the coincidence of all comorbidities between ατ+ samples (5, 0.46%) and ατ− samples (69, 0.33%) (P = 0.633). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 6. Association of MTBD-tau IgG autoantibodies with systemic disorders and clinical laboratory data. (A) aRR ± 95% CI (bars) for ατ+ autoantibodies grouped 27 different main systemic clinical conditions. Asterisks: P < 0.05 after Bonferroni correction. (B) −log10(P value) of the log-binomial regression for the presence of plasma MTBD-tau IgG autoantibodies for 276 disorders according to ICD-10 codes and adjusted for age and sex. Triangles pointing upwards and downwards: positive and negative coefficients, respectively; gray dashed line: P < 0.05 after Bonferroni correction (here and henceforth). P values significant after Bonferroni adjustment are labeled with the ICD codes. (C) Same as (A), but according to individual disease entities from (B). (D) Bayesian logistic regression adjusted for age and sex on ICD-10 codes significantly associated in B. and using −logit10(EC50) as continuous outcome (odds ratio ± 95% credible intervals). This analysis confirms the positive association of these ICD-10 codes with ατ reactivity (from B). (E) −log10(P value) of the log-binomial regression for the presence of plasma MTBD-tau IgG autoantibodies and clinical laboratory parameters. P < 0.05 are labeled with the clinical laboratory parameters. Gray dashed line: P < 0.05 after Bonferroni correction. eGFR CKD-EPI 2009: estimated glomerular filtration rate using the Chronic Kidney Disease Epidemiology Collaboration 2009 equation; eGFR BIS1: estimated glomerular filtration rate using the older adults Berlin Initiative Study 1 equation. INR: international normalized ratio. LDL: low-density lipoprotein. HDL: high-density lipoprotein. (F) Calculated prevalence of different ICD-10 codes in the total cohort (dark blue), ατ− (intermediate blue) and ατ+ samples (light blue). Asterisks: P < 0.05 after Bonferroni correction (two-proportions z-test). https://doi.org/10.1371/journal.pbio.3003488.g006 To address potential biases arising from grouping and selecting major categories, we explored the association between ατ+ and individual ICD-10 codes. Across 276 individual ICD-10 codes and after correction for multiple comparisons, eight exhibited significant associations with ατ+ (Fig 6B and 6C). These included “E61–deficiency of other nutrient elements” (aRR 2.22, 95% CI 1.54–3.06, P < 0.001), “I74–arterial embolism and thrombosis” (aRR 2.09, 95% CI 1.40–2.94, P < 0.001), “N30–cystitis” (aRR 1.84, 95% CI 1.32–2.47, P < 0.001), “I70–atherosclerosis” (aRR 1.57, 95% CI 1.29–1.89, P < 0.001), “D50–iron deficiency anemia” (aRR 1.50, 95% CI 1.21–1.82, P < 0.001), “N39–other urinary disorders” (aRR 1.40, 95% CI 1.19–1.64, P < 0.001), “B96–other bacterial agents as the cause of diseases” (aRR 1.40, 95% CI 1.17–1.66, P < 0.001) and “N18–chronic kidney disease” (aRR 1.38, 95% CI 1.18–1.60, P < 0.001; Fig 6C). Conversely, we found no significant negative association between tau autoreactivity and any ICD-10 codes that could suggest a decreased risk for a specific disease in ατ+ patients (Fig 6B). To verify relevant associations, we used Bayesian logistic regression using logit-transformed EC50 values (−logit[EC50]) as a continuous outcome. All the previously referred eight ICD-10 codes showed positive associations with −logit[EC50] (Fig 6D). To validate these conditions as significant associations with plasma MTBD-tau IgG autoantibodies, we examined the association between ατ+ and 106 commonly requested laboratory parameters (S2 Table) using data available for 24,177 patients, independently of any ICD-10 classifiers. After correction for multiple comparisons, the laboratory markers “Leukocytes, urine”, “Potassium,” and “Urea” were positively associated with tau autoimmunity (Fig 6E). The association with increasing levels of urea and potassium suggests a link to kidney failure, whereas leukocytes in urine are a feature of urinary tract infections. These findings support the hypothesis that anti-tau autoimmunity correlates with such disorders. Accordingly, the prevalence of chronic kidney disease was 12.4% in ατ− and 17.7% in ατ+ patients, whereas that of other urinary disorders was 10.3% in ατ− and 15.1% in ατ+ patients (Fig 6F). Discussion This study was designed to discover novel associations between autoreactivity to tau and human diseases. This approach of “reverse immunopathology” is hypothesis-free: it does not pose constraints on the type of diseases that might result from anti-tau immunity. While the approach is generally applicable to any autoantigen, its predictive power relies on the analysis of large patient collectives, which in turn requires the development of high-throughput methods at reasonable cost. We therefore miniaturized antibody detection to 3 µl/sample in 1,536-well microplates, allowing to run 40,000 assays/24 h with minimal hands-on time. We performed >300,000 immunoassays on >40,000 plasma samples for ατ autoreactivity. Each sample was measured at eight dilution points, enabling precise and unambiguous titer determinations [25–27]. Plasma ατ IgG were more prevalent in women and increased with age, reaching 7.6% in the cohort of 90–99-year-olds. The specificity of ατ+ plasma to MTBD-tau was high and was confirmed by multiple assays including epitope mapping. The age- and sex-adjusted ατ+ rate in hospital patients was 3-fold higher than in healthy blood donors and considerably exceeded the previously reported seroprevalences of autoantibodies targeting other intracellular neuronal targets (0.4% and 2%) [34], suggesting a potential correlation with disease states. However, we did not find any association between ατ+ and any neurological conditions including neurodegenerative diseases using both ICD-10 classification as well as a convenience AD and non-AD cohort of plasma samples, corroborating with reports of similar levels of plasma tau autoantibodies in AD patients and non-demented controls [35]. Instead, we uncovered a robust, specific association between ατ+ positivity and kidney and urinary disorders. This was independently validated by associations with higher prevalence of positive samples from the department of Nephrology, grouped kidney and urinary diseases, individual kidney and urinary diagnosis codes and related clinical laboratory biomarkers, such as high serum urea and potassium and high urine leukocytes. Seroepidemiological studies cannot establish causality: ατ seropositivity could be a cause or a consequence of the clinical syndromes associated with it [36]. Therefore, due to its design, this study cannot conclude on the homeostatic or pathological role of the anti-tau IgG autoantibodies identified in the plasma of patients and healthy blood donors. However, tau is expressed in extraneural tissues, including kidney podocytes and urinary bladder [2–5], and its ablation causes glomerular pathologies [37]. These findings raise the possibility that plasma ατ autoantibodies might drive kidney and/or urological pathologies. Alternatively, they could reflect a physiological autoimmunity with low-affinity autoantibodies playing a role in immune surveillance and the clearance of debris. Another possibility is that an underlying pathological process causing renal/bladder tissue injury and inflammation, may lead to the extracellular release of intracellular proteins, including tau. Damage to tubular epithelial cells or glomerular structures may expose intracellular tau to the immune system, potentially breaking tolerance and triggering the emergence of natural or low-affinity IgG autoantibodies against tau. Renal dysfunction may also impair clearance of circulating antibodies or immune complexes, allowing even normally subclinical reactivity to become detectable. If the association between tau autoantibodies and kidney/urinary disease is epiphenomenal rather than causal these autoantibodies may represent useful biomarkers of these diseases. Therefore, it will be important to monitor renal and urinary function in the current clinical trials of tau immunotherapy [38–40]. Recent studies have explored the use of tau measurements in plasma as biomarkers for AD diagnosis and for monitoring its progression [9–12,32,41–45]. However, we found that anti-MTBD-tau autoantibodies could hinder tau detection in plasma by binding to epitopes recognized by commercial biomarker immunoassays. The extent of this interference likely depends on both the epitopes present and the design of the diagnostic assay, varying with the specific tau analytes targeted in the assay and the particular epitopes recognized by the patients’ autoantibodies. This aligns with prior research indicating that both plasma anti-tau autoantibodies and administered anti-tau antibodies can influence the dynamics of tau levels in plasma [46]. Additionally, a recent community cohort study using an immunoassay for AD screening revealed associations between plasma tau levels and numerous comorbidities, with chronic kidney disease showing one of the strongest associations [47]. These results imply that the presence of anti-tau autoantibodies in plasma might impact the effectiveness of plasma tau as an AD biomarker. This consideration may become crucial as plasma tau levels move toward routine clinical use in AD diagnosis. Limitations Our large-scale assessment of plasma tau autoimmunity has certain limitations. Firstly, most samples analyzed in our study came from a university hospital cohort, implying a bias towards complex pathologies and polymorbidity. To account for this, we included a vast collection of samples from healthy blood donors. Secondly, our study is confined to two sites within a single region with approximately 1,500,000 inhabitants whose ethnic composition was not thoroughly examined. Additionally, a direct comparison of the prevalence of plasma anti-tau antibodies in these two cohorts may potentially suffer from variations in materials, collection methods (e.g., addition of heparin that may cause aggregation of MTBD-tau), and handling processes at the two different sites. Thirdly, our study is restricted to the analysis of specific disease groups, the absence of an independent replication cohort and the lack of a longitudinal disease design. These constraints restrict our ability to verify our findings, assess temporal dynamics and the origin of these autoantibodies, and explore potential causal relationships, including reverse causality, such as the presence of tau antibodies in patient cohorts with renal diseases. Fourthly, the detected anti-tau antibodies may crossreact with proteins expressed in the kidney or urinary tract, which could contribute to the observed association with kidney and urinary disorders. To explore this possibility, we performed an in silico analysis using BLASTP (https://blast.ncbi.nlm.nih.gov/) and compared the MTBD-tau to proteins in the human kidney and urinary proteome (UniProt/Swiss-Prot). This analysis identified residues 368–461 of nicotinamide phosphoribosyltransferase (NAMPT) as a region with notably high sequence identity to MTBD-tau (~28%; S3 Fig). NAMPT is a metabolic enzyme with broad tissue distribution but relatively low expression in the kidney and urinary tract [2]. Nevertheless, such sequence-based analysis is limited detecting linear or conformational epitopes that might mediate cross-reactivity. Furthermore, the statistical analysis of the association of MTBD-tau-autoreactivity and different disorders was based on ICD-10, a medical classification focused on reimbursement and cause-of-death statistics [48]. We therefore performed additional statistical analyses using laboratory parameters, which have provided further support for the association of tau autoimmunity and systemic disorders. Finally, we used bacterially expressed MTBD-tau as target, thereby excluding full-length, other isoforms (such as big tau, a high molecular weight (~110 kDa) tau isoform that includes exon 4a in the MAPT transcript [49]) or post-translationally modified tau epitopes, like the phosphorylated forms p-tau181 [50] and p-tau217 [51], that are increasingly used as biomarkers for AD diagnosis. As a result, our methodology may miss plasma anti-tau autoantibodies associated with AD pathology and could underestimate the prevalence of total anti-tau autoantibodies. Furthermore, this approach may introduce a bias towards detecting autoantibodies targeting total peripheral tau which is more abundant in plasma [52]. Therefore, larger studies with well-powered AD patient cohorts may be needed to more robustly assess the prevalence of autoantibodies selectively targeting brain-derived tau and to clarify their potential association with neurodegeneration. Limitations Our large-scale assessment of plasma tau autoimmunity has certain limitations. Firstly, most samples analyzed in our study came from a university hospital cohort, implying a bias towards complex pathologies and polymorbidity. To account for this, we included a vast collection of samples from healthy blood donors. Secondly, our study is confined to two sites within a single region with approximately 1,500,000 inhabitants whose ethnic composition was not thoroughly examined. Additionally, a direct comparison of the prevalence of plasma anti-tau antibodies in these two cohorts may potentially suffer from variations in materials, collection methods (e.g., addition of heparin that may cause aggregation of MTBD-tau), and handling processes at the two different sites. Thirdly, our study is restricted to the analysis of specific disease groups, the absence of an independent replication cohort and the lack of a longitudinal disease design. These constraints restrict our ability to verify our findings, assess temporal dynamics and the origin of these autoantibodies, and explore potential causal relationships, including reverse causality, such as the presence of tau antibodies in patient cohorts with renal diseases. Fourthly, the detected anti-tau antibodies may crossreact with proteins expressed in the kidney or urinary tract, which could contribute to the observed association with kidney and urinary disorders. To explore this possibility, we performed an in silico analysis using BLASTP (https://blast.ncbi.nlm.nih.gov/) and compared the MTBD-tau to proteins in the human kidney and urinary proteome (UniProt/Swiss-Prot). This analysis identified residues 368–461 of nicotinamide phosphoribosyltransferase (NAMPT) as a region with notably high sequence identity to MTBD-tau (~28%; S3 Fig). NAMPT is a metabolic enzyme with broad tissue distribution but relatively low expression in the kidney and urinary tract [2]. Nevertheless, such sequence-based analysis is limited detecting linear or conformational epitopes that might mediate cross-reactivity. Furthermore, the statistical analysis of the association of MTBD-tau-autoreactivity and different disorders was based on ICD-10, a medical classification focused on reimbursement and cause-of-death statistics [48]. We therefore performed additional statistical analyses using laboratory parameters, which have provided further support for the association of tau autoimmunity and systemic disorders. Finally, we used bacterially expressed MTBD-tau as target, thereby excluding full-length, other isoforms (such as big tau, a high molecular weight (~110 kDa) tau isoform that includes exon 4a in the MAPT transcript [49]) or post-translationally modified tau epitopes, like the phosphorylated forms p-tau181 [50] and p-tau217 [51], that are increasingly used as biomarkers for AD diagnosis. As a result, our methodology may miss plasma anti-tau autoantibodies associated with AD pathology and could underestimate the prevalence of total anti-tau autoantibodies. Furthermore, this approach may introduce a bias towards detecting autoantibodies targeting total peripheral tau which is more abundant in plasma [52]. Therefore, larger studies with well-powered AD patient cohorts may be needed to more robustly assess the prevalence of autoantibodies selectively targeting brain-derived tau and to clarify their potential association with neurodegeneration. Conclusions Our study identified a high seroprevalence of anti-MTBD-tau IgG autoantibodies in both plasma samples from university hospital patients and healthy blood donors. Tau autoimmunity is associated with female sex, older age, and previously unrecognized extraneural diseases. These findings point to unrecognized roles for tau and anti-tau autoantibodies in extraneural pathologies. Materials and methods Methods Study approval. Collection of samples and clinical data were conducted according to study protocols approved by the Cantonal Ethics Committee of the Canton of Zurich, Switzerland (KEK-ZH Nr. 2015-0561, BASEC-Nr. 2018-01042, and BASEC-Nr. 2020-01731), in accordance with the provisions of the Declaration of Helsinki and the Good Clinical Practice guidelines of the International Conference on Harmonisation. All human donors and patients included in this study provided a written general informed consent. Study design. From December 2017 until February 2020, residual heparin plasma samples were obtained from the Department of Clinical Chemistry, University Hospital of Zurich, USZ, Switzerland. Samples were collected during routine clinical care from patients admitted either as inpatients or outpatients (age ≥18 years) and were only included if basic demographic data was available, and an informed consent for research had been provided. From March until July 2020, EDTA plasma samples from blood donors were obtained from the Blood Donation Center of Zurich, Switzerland, according to standard criteria of blood donation. Exclusion criteria were as reported [25]. Plasma samples from patients of both sexes were examined in this study. Plasma samples were biobanked locally and tested in an automated indirect microELISA ([25–27] and below) for natural IgG autoantibodies against the MTBD-tau. Demographic and clinical data for the hospital cohort were obtained from clinical records of the USZ with follow-up until December 2021, while detailed clinical data for the blood donor cohort were not available for this study. ICD-10 codes [53] were used for clinical data assessment. AD patients were selected using ICD-10 code F00 or G30. Non-neurodegeneration controls were defined by the lack of any Fxx or Gxx ICD-10 codes. Automated microELISA screen. Plasma samples were tested for natural anti-MTBD-tau IgG autoantibodies in a microELISA screen [25–27]. Briefly, high-binding 1,536-well microplates (Perkin Elmer, SpectraPlate 1536 HB) were coated with 1 µg/mL of recombinant MTBD-tau (37 °C, 60 min). Plates were washed 3× with phosphate-buffered saline 0.1% Tween20 (PBST) using a Biotek El406 washer-dispenser and blocked with 5% milk (Migros)-PBST for 90 min. Plasma samples were diluted 1:20 in 1% milk-PBST and dispensed into the MTBD-tau-coated plates using ultrasound dispensing with an ECHO 555 Liquid Handler (Labcyte/Beckman Coulter). Each sample was tested at eight serial 2-fold dilutions (1:50–1:6,000) using different volumes to a final volume of 3 µL/well. Human IgG-depleted serum (MyBioSource) was used as negative and anti-tau RD4 (4-repeat isoform) mouse monoclonal antibody [30] (05–804 clone 1E1/A6 Merck Millipore) as positive control. Plates were incubated for 120 min at room temperature (RT) and washed 5x with PBST. Secondary antibody peroxidase AffiniPure goat anti-mouse IgG H + L (115-035-003, Jackson ImmunoResearch) 1:2,000 diluted in 1% milk-PBST for the RD4 positive control, peroxidase AffiniPure goat anti-human IgG Fcγ-specific (109-035-098, Jackson ImmunoResearch) 1:4,000 diluted in 1% milk-PBST for the plasma samples and the IgG-depleted serum negative control were dispensed into the plates using a Biotek MultifloFX dispenser. Plates were incubated for 60 min, at RT and washed 3× with PBST. 3,3′,5,5′-Tetramethylbenzidine (TMB) Chromogen Solution for ELISA (Invitrogen) was added as colorimetric horseradish peroxidase (HRP) substrate for 3 min at RT. Finally, 0.5 M H2SO4 was added to stop the reaction. Plates were briefly centrifuged after each dispensing step except after dispensing of TMB. Plates were read at Optical Density = 450 nm (OD450nm) in a plate reader (Perkin Elmer, Envision). For the replicability assessment, 308 samples were tested in duplicates, running the replicates on the same day but using different 1536-well assay (destination) plates, different plate coordinates for each replicate, and calculating the −log10(EC50) of each replicate independently. Non-specific cross-reactivity was assessed by testing plasma samples against MTBD-tau and amyloid-β pyroglutamate (12,297 hospital patients’ plasma samples) and the PrPC (13,099 hospital patients’ samples) using a similar protocol as described above. Production and purification of recombinant tau. The gene encoding the human truncated 4R-tau corresponding to MTBD-tau was cloned into a pRSET-A plasmid (Invitrogen) and expressed in Escherichia coli BL21(DE3) cells. Cultures were grown in Luria Broth (LB, Invitrogen) at 37 °C, induced with 1 mM isopropyl-β-D-thiogalactoside (IPTG) at an OD600 of 0.8 and grown for additional 6 h at 37 °C before harvesting by centrifugation (6,000g, 10 min, 4 °C). Pellets were suspended and sonicated (30 min, 4 °C) in 20 mM piperazine-N,N-bis(2-ethanesulfonic) acid (PIPES), pH 6.5, 1 mM ethylenediaminetetraacetic acid (EDTA) and 50 mM 2-mercaptoethanol. After addition of NaCl to a final concentration of 500 mM, samples were boiled (95 °C, 20 min) and centrifuged (9,000g, 30 min, 4 °C). Ammonium sulfate was slowly added to a final concentration of 55% m/v and the suspension was stirred (1 h, RT). Samples were centrifuged (15,000g, 10 min, 4 °C), pellets were resuspended in 20 mM 4-(2-hydroxyethyl)piperazine-1-ethanesulfonic acid (HEPES), pH 7.0, 2 mM dithiothreitol (DTT), passed through a 0.45 µm Acrodisc filter (Sigma), and loaded onto Sepharose SP Fast Flow resin (Cytiva). Tau was eluted using a linear salt gradient from 0 to 1 M NaCl in 20 mM HEPES, pH 7.0, 2 mM DTT. Fractions containing tau were concentrated using Amicon Ultra-15 centrifugal filter unit (10-kDa MWCO, Merck) and dialyzed overnight at 4 °C against phosphate-buffered saline (PBS, pH 7.4, Kantonsapotheke Zurich), 1 mM DTT. Pooled samples were passed through a HiLoad 26/60 Superdex75 (GE Healthcare) column. Protein samples were analyzed by SDS-PAGE and samples containing tau were concentrated using an Amicon Ultra-15 centrifugal filter unit (10-kDa MWCO). Samples were assessed by SDS-PAGE and electrospray ionization-mass spectrometry. Pure MTBD-tau samples were stored until further use at −80 °C. The concentration of tau was determined using a bicinchoninic acid assay (Pierce BCA Protein Assay Kit, Thermo Fisher). For the purification of full-length tau, a similar protocol was used with the following changes. The gene encoding the longest 4R isoform of human full-length tau protein, tau441, (tau/pET29b, Addgene #16316, gift from Peter Klein [54]) was cloned into a pRSET-A plasmid (Invitrogen). For protein expression, E. coli BL21(DE3)pLysS cells were transformed with the pRSET-A plasmid encoding tau441. Cells were grown in Overnight Express Instant TB media (Novagen) for 6 h at 37 °C and then for 12 h at 25 °C. Fractions containing full-length tau were concentrated using Amicon Ultra-15 centrifugal filter unit (30-kDa MWCO, Merck). Purification of anti-tau autoantibodies from patient samples. Heparin plasma (3−20 mL) was diluted 1:3.3 in PBS, and centrifuged at 6,000g for 10 min at 4 °C. The supernatant was loaded onto 3 mL of epoxy-MTBD-tau (prepared by overnight incubation of MTBD-tau and epoxy resin in 0.1 M NaH2PO4–NaOH, 1 M NaCl, pH 9.2) by repetitive loading of the plasma sample overnight at 4 °C. After washing with 50 mL of PBS, MTBD-tau autoantibodies were eluted 4× with 5 mL 0.1 M glycine–HCl, pH 2.5, and immediately neutralized to pH 7.0 with 1 M Tris-HCl, pH 8.5. Anti-MTBD-tau antibody-containing fractions were identified by indirect ELISA and concentrated stepwise using Amicon Ultra-15, Ultra-4, and Ultra-0.5 mL centrifugal filter units (50-kDa MWCO) up to a volume of 1 mL. Competitive ELISA. For the competitive sandwich ELISAs for the detection of tau, high-binding 384-well plates (Perkin Elmer, SpectraPlate 384 HB) were coated with 4 µg/mL BT2 tau monoclonal antibody (#MN1010, Thermo Fisher) in PBS. After coating, plates were washed 3× with PBST and then blocked with 5% SureBlock (Lubio) in PBS for 180 min at RT. Recombinant human tau441 (rPeptide) was diluted to a final concentration of 0.015 ng/mL and incubated with purified anti-tau autoantibodies 4-fold serially diluted (1:1.66 to 1:106.6) in 1:2 plasma under rotation at 500 rpm for 120 min at 37°C. Samples were transferred to BT2-coated plates and incubated for 45 min at RT. After washing 4× with PBST, plates were incubated with ab64193 (polyclonal IgG antibody; Abcam, 0.125 µg/mL) for 45 min at RT. Plates were washed 4x with PBST and incubated with peroxidase AffiniPure goat anti-Rabbit IgG (H + L) (111-035-045, Jackson ImmunoResearch), at a 1:2,000 dilution for 60 min at RT. Plates were washed 4× with PBST and 1-Step Ultra TMB-ELISA solution (Thermo Fisher) was added for 7 min at RT. After addition of 0.5 M H2SO4, plates were read at OD450 nm in a plate reader (Perkin Elmer, Envision). For the competitive ELISAs of MTBD-tau autoantibodies, high-binding 384-well plates were coated with 20 µL of 0.5 µg/mL of MTBD-tau overnight at 4 °C. Afterward, plates were washed 3× with PBST and blocked with 5% SureBlock (Lubio) in PBST for 120 min. Purified autoantibodies from hospital cohort patients’ plasma were diluted 1:50 in 1% SureBlock in PBST (sample buffer) and the anti-tau RD4 mouse monoclonal antibody to a final concentration of 0.4 µg/mL. Bovine serum albumin (BSA, Thermo Scientific), in-house purified recombinant MTBD-tau, a pool of eight synthetic peptides covering the sequence of MTBD-tau with 25 amino acids length and 10 amino acids of overlap (Genscript) and an unrelated 25 amino acid length synthetic TREM2 (Triggering receptor expressed on myeloid cells 2) peptide (GenScript) were used as competing antigens. Antibody samples were incubated overnight at 4 °C with serial 2-fold dilutions of antigen solutions in sample buffer, ranging from 20,000 to 2.44 nM. The antibody-antigen mixtures were then added to the plates and incubated for 45 min at RT. Plates were washed 3x with PBST, followed by the addition of secondary antibodies: peroxidase AffiniPure goat anti-Human IgG (H + L) (109-035-088, Jackson ImmunoResearch) at 1:3,000 dilution and peroxidase AffiniPure goat anti-mouse IgG (H + L) (115-035-003, Jackson ImmunoResearch) at 1:2,000 dilution. Secondary antibodies were incubated for 60 min at RT and plates were then washed 4× with PBST. TMB Chromogen Solution for ELISA (Invitrogen) was added to the plates and incubated for 7 min at RT. After addition of 0.5 M H2SO4, plates were read at OD450nm in a plate reader (SpectraMax Paradigm, Molecular Devices). Indirect ELISAs. To test for polyreactivity, purified anti-tau autoantibodies were tested by indirect ELISA against several antigens [55,56]. High-binding 384-well plates were coated overnight at 4 °C with 20 µL of 1 µg/mL of MTBD-tau, 10 µg/mL of DNA from calf-thymus (Sigma), 10 µg/mL of LPS from E. coli O111:B4 (Sigma), 5 µg/mL of human insulin (Sigma), 10 µg/mL of BSA, 2 µg/mL of cardiolipin solution from bovine heart (Sigma), or left uncoated. Plates were washed 3× with PBST and then blocked in 5% SureBlock (Lubio) in PBST for 120 min at RT. Patient purified anti-MTBD-tau autoantibodies were diluted 1:33, IgG-depleted plasma (BioSource) 1:50 diluted and used as negative control, anti-DNP (Sigma) diluted to 6 µg/mL, anti-tau RD4 to 6 µg/mL, and pooled plasma from 20 patients diluted 1:25 in 1% SureBlock in PBST as positive controls. Samples were serially diluted 12 times 1:1 with 1% SureBlock in PBST in the referred plates. Samples were incubated for 120 min at RT and washed 4× with PBST. 20 µL/well of secondary antibodies diluted in 1% SureBlock in PBST were added as follows: peroxidase AffiniPure goat anti-human IgG (H + L) (109-035-088, Jackson ImmunoResearch) diluted 1:3,000 and added to purified MTBD-tau autoantibodies and IKC pool wells, peroxidase AffiniPure goat anti-mouse IgG (H + L) (115-035-003, Jackson ImmunoResearch) at 1:2,500 dilution and added to anti-RD4 wells, and peroxidase AffiniPure goat anti-rabbit IgG (H + L) (111-035-045, Jackson ImmunoResearch) at 1:4,000 dilution and added to anti-DNP wells. Plates were then washed 4× with PBST. 20 µL of TMB Chromogen solution for ELISA (Invitrogen) was added and incubated for 7 min at RT. After addition of 0.5 M H2SO4, plates were read at OD450nm in a plate reader (SpectraMax Paradigm, Molecular Devices). For relative affinity measurements, we used an indirect ELISA in 384-well plates (Perkin Elmer, SpectraPlate 384 HB) with MTBD-tau as a coating antigen using the following parameters: The starting concentration of all antibodies, assayed in triplicates, was 10 µg/mL. They were successively diluted 1:3 to reach a concentration of 2 × 10−6 µg/mL. As reference antibody, we used purified RD4 kindly provided by Prof. Rohan de Silva (UCL Queen Square Institute of Neurology, London, UK). As additional controls, we used anti-LAG3 antibody Relatlimab [57], ατ− plasma sample, and uncoated plates. The respective EC50 values were then determined using logistic regression, as above, and the curves as well as the dots were visualized. For IgG subclassing, the following secondary antibodies were used: rabbit anti-human IgG1 (SA5-10202, Invitrogen), rabbit anti-human IgG2 (SA5-10203, Invitrogen), rabbit anti-human IgG3 (SA5-10204, Invitrogen) or rabbit anti-human IgG4 (SA5-10205, Invitrogen) at 1:1,500 dilution, peroxidase AffiniPure goat anti-rabbit IgG (H + L) antibody (111-035-045, Jackson ImmunoResearch) at 1:2,500 dilution. For immunoglobulin light chain typing, the following secondary antibodies were used: Goat anti-Human Kappa-HRP [58] and Goat anti-Human Lambda-HRP [58] at 1:4,000 dilution and peroxidase AffiniPure goat anti-mouse IgG (H + L) (115-035-003, Jackson ImmunoResearch) at 1:2,500 dilution. For the epitope mapping experiments, a similar approach was used in which high-binding 384-well plates were coated with 20 µL of 1 µg/mL of each MTBD-tau peptide (GenScript) in PBS overnight at 4°C. Plasma samples and human IgG-depleted serum (MyBioSource) used as negative control were diluted 1:50 and anti-tau RD4 mouse monoclonal antibody used as positive control (05−804 clone 1E1/A6 Merck Millipore) was diluted to a final concentration of 6 µg/mL in 1% SureBlock in PBST. The following secondary antibodies were used: peroxidase AffiniPure goat anti-human IgG (H + L) (109-035-088, Jackson ImmunoResearch) at 1:5,000 dilution and peroxidase AffiniPure goat anti-mouse IgG (H + L) (115-035-003, Jackson ImmunoResearch) at 1:2,500 dilution. For data analysis of IgG subclassing, immunoglobulin light chain typing, and epitope mapping experiments, samples were considered reactive when the OD450nm was higher than the average of all negatives OD450nm + 2× the standard deviation of the OD450nm of the negatives. Western blotting. SH-SY5Y wild-type (Sigma) and cells overexpressing double-mutated tauP301L/S320F were lysed in 0.5% Triton X-100 (Sigma Aldrich) in PBS, supplemented with cOmplete Mini EDTA-free Protease Inhibitor Cocktail (Roche) on ice and supernatant was recovered after centrifugation at 14,000g for 20 min at 4 °C. Protein concentration was determined using a bicinchoninic acid assay (Pierce BCA Protein Assay Kit, Thermo Fisher) and sample volumes were adapted to 30 µg of total protein. Samples were loaded onto NuPAGE 12% Bis-Tris gels (Invitrogen) and blotted to nitrocellulose membranes (Invitrogen) using a dry iBlot 2 Gel Transfer Device (Invitrogen, Thermo Fisher). Membranes were cut vertically along the protein ladder and blocked with 5% SureBlock in PBST at 30 min for RT and incubated overnight at 4 °C with patient-purified anti-MTBD-tau autoantibodies diluted 1:100 in 1% SureBlock, PBST. Anti-tau RD4 mouse monoclonal antibody diluted 1:4,000 was used as a positive control. Negative control membranes were not incubated with primary antibodies. Membranes were washed 3× for 5 min with PBST and incubated with the following secondary antibodies for 60 min at RT: peroxidase AffiniPure goat anti-mouse IgG (H + L) (115-035-003, Jackson ImmunoResearch) and peroxidase AffiniPure goat anti-human IgG (H + L) (109-035-088, Jackson ImmunoResearch) 1:10,000 diluted. Membranes were washed 4× for 5 min with PBST and developed using the Immobilon Crescendo HRP Substrate (Millipore). Imaging was performed with the Fusion SOLO S imaging system (Vilber). Immunofluorescence. SH-SY5Y cells were transfected with pRK5-EGFP-0N4Rtau plasmid (Addgene # 46904, a kind gift of Dr Karen Ashe [59]) using Lipofectamine 2000 according to manufacturer’s protocol. After 48 h cells were fixed with 4% paraformaldehyde for 20 min at RT, permeabilized with 0.5% BSA, 0.1% Triton X-100 in PBS for 20 min at RT, and blocked with 0.5% BSA in PBS for 60 min at RT. For the immunocytochemical assays, 1:25 purified ani-MTBD-tau autoantibodies diluted in 0.5% BSA in PBS were incubated with the cells for 60 min. Tau mouse monoclonal antibody HT7 (MN1000 Thermo Fisher Scientific) diluted 1:500 was used as a positive control. Cells were washed 3× with 0.5% BSA in PBS and incubated for 30 min at RT with: goat anti-mouse IgG (H + L) cross-adsorbed Alexa Fluor 555 (A-21422 Invitrogen) diluted 1:500 and counterstained with 4,6-diamidino-2-phenylindole (DAPI) 1 mg/mL (Thermo Fisher) diluted 1:10,000 in 0.5% BSA in PBS, and for human antibody and negative control stains with biotin AffiniPure goat anti-human IgG (H + L) (109-065-003, Jackson ImmunoResearch) diluted 1:200 in 0.5% BSA in PBS. Samples were 3x washed with 0.5% BSA in PBS. For human antibody stains, samples were further incubated for 30 min at RT with streptavidin Alexa Fluor 594 conjugate diluted 1:200 and counterstained with DAPI diluted 1:10,000 in 0.5% BSA in PBS. Cells were mounted on glass slides with Fluorescence Mounting Medium (Thermo Fisher Scientific) and imaged using a Leica TCS SP5 confocal laser scanning microscope. Imaging was performed with equipment maintained by the Center for Microscopy and Image Analysis, University of Zurich. In vitro MTBD-tau aggregation assay. MTBD-tau in vitro aggregation experiments were performed as previously described [31]. Briefly, 7 µM of in-house purified recombinant MTBD-tau, 3.5 µM of heparin (Santa Cruz Biotechnology) and 10 µM of ThT (Sigma) were diluted in PBS. The patient purified anti-tau autoantibodies and controls (plasma sample reactive against the LAG3 and IgG-purified using Protein G Sepharose (Cytiva)) were added at the indicated apparent stoichiometries in Fig 3. The mixtures with a total volume of 200 µl were added to black 96-well polystyrene microplates (Nunc, Prod. No. 265301) and ThT fluorescence (450/480 nm ex/em filters; bottom read mode) was measured at 37 °C under continuous orbital shaking (425 cpm) every 15 min for 96 h using a FLUOstar Omega microplate reader (BMG Labtech). The mean baseline fluorescence values were subtracted from the mean fluorescence values at each time point which were then normalized to maximum baseline-subtracted fluorescence values and multiplied by 100 [60]. Statistical analysis. Antibody titers were defined as the negative logarithm half-maximal responses (−log10(EC50)) obtained by fitting the OD450nm of the eight dilution points of each sample tested in the microELISA to a logistic regression fitter. We classified as positives samples with a cut-off of −log10(EC50) ≥1.8 [25], corresponding to a nominal dilution of >1/64. Non-informative samples (fitting error >20% −log10(EC50) or high background) were excluded from the analysis [25]. In cases where more than one sample was available for the same individual the most recent log10(EC50) value was used. Age is presented as median with interquartile range (IQR) and comparisons were performed using non-parametric Mann–Whitney U test. Categorized age groups and sex of positives and negatives are shown as percentages and compared using two-proportions Z-test or χ2 test for trend in proportions. Log-binomial regression models [28,61] using MTBD-tau autoreactivity were employed to estimate age- and sex-aRRs and 95% CIs, and to investigate the association between the detection of anti-MTBD-tau IgG autoantibodies and different demographic features. For the AD cohort microELISA screen, log10(EC50) values were compared using Mann–Whitney U test. Additionally, Bayesian logistic regression was conducted [25,61–63], using −logit10(EC50) values as outcome, i.e., without dichotomizing the outcome using the R package rstanarm and the following priors (prior=normal[0, 2.5, autoscale=TRUE],prior_intercept = normal[5000, 2.5, autoscale=TRUE)] prior_aux = exponential [1, autoscale =TRUE]. We aimed to confirm the positive association between ICD-10 codes previously identified using a conventional logistic regression model with high −log10(EC50) values. Each ICD-10 code was analyzed in an independent logistic regression and adjusted for age and sex. The association between MTBD-tau-autoreactivity and neurological and systemic disorders, respectively, was analyzed by applying multivariate log-binomial regression models to estimate aRRs and 95% CIs. For neurological disorders, 23 major groups of ICD-10 codes (S3 Table) corresponding to neurological disorders identified at least once in the positive samples were used. For systemic disorders, the 27 major groups of ICD-10 codes (S4 Table) or 276 individual ICD-10 codes corresponding to systemic disorders identified at least once in the positive samples and with at least 200 total counts were used to avoid overinterpretation of rare cases of disease. Individual disease entities with a P value <0.05 Bonferroni corrected for multiple comparisons were included in a multivariate log-binomial regression analysis. For the association between MTBD-tau-autoreactivity and clinical laboratory parameters, we used laboratory parameters with more than 2,000 total counts and calculated median values of the total values available for each patient in case of repetition of the clinical laboratory test. We used multivariate log-binomial regression models to estimate aRR and 95% CI using 106 clinical laboratory tests. Statistical significance was defined by two-tailed P-value ≤ 0.05. Statistical analyses and data visualization were performed using R version 4.3.2 and RStudio version 1.4.1106 [64]. Methods Study approval. Collection of samples and clinical data were conducted according to study protocols approved by the Cantonal Ethics Committee of the Canton of Zurich, Switzerland (KEK-ZH Nr. 2015-0561, BASEC-Nr. 2018-01042, and BASEC-Nr. 2020-01731), in accordance with the provisions of the Declaration of Helsinki and the Good Clinical Practice guidelines of the International Conference on Harmonisation. All human donors and patients included in this study provided a written general informed consent. Study design. From December 2017 until February 2020, residual heparin plasma samples were obtained from the Department of Clinical Chemistry, University Hospital of Zurich, USZ, Switzerland. Samples were collected during routine clinical care from patients admitted either as inpatients or outpatients (age ≥18 years) and were only included if basic demographic data was available, and an informed consent for research had been provided. From March until July 2020, EDTA plasma samples from blood donors were obtained from the Blood Donation Center of Zurich, Switzerland, according to standard criteria of blood donation. Exclusion criteria were as reported [25]. Plasma samples from patients of both sexes were examined in this study. Plasma samples were biobanked locally and tested in an automated indirect microELISA ([25–27] and below) for natural IgG autoantibodies against the MTBD-tau. Demographic and clinical data for the hospital cohort were obtained from clinical records of the USZ with follow-up until December 2021, while detailed clinical data for the blood donor cohort were not available for this study. ICD-10 codes [53] were used for clinical data assessment. AD patients were selected using ICD-10 code F00 or G30. Non-neurodegeneration controls were defined by the lack of any Fxx or Gxx ICD-10 codes. Automated microELISA screen. Plasma samples were tested for natural anti-MTBD-tau IgG autoantibodies in a microELISA screen [25–27]. Briefly, high-binding 1,536-well microplates (Perkin Elmer, SpectraPlate 1536 HB) were coated with 1 µg/mL of recombinant MTBD-tau (37 °C, 60 min). Plates were washed 3× with phosphate-buffered saline 0.1% Tween20 (PBST) using a Biotek El406 washer-dispenser and blocked with 5% milk (Migros)-PBST for 90 min. Plasma samples were diluted 1:20 in 1% milk-PBST and dispensed into the MTBD-tau-coated plates using ultrasound dispensing with an ECHO 555 Liquid Handler (Labcyte/Beckman Coulter). Each sample was tested at eight serial 2-fold dilutions (1:50–1:6,000) using different volumes to a final volume of 3 µL/well. Human IgG-depleted serum (MyBioSource) was used as negative and anti-tau RD4 (4-repeat isoform) mouse monoclonal antibody [30] (05–804 clone 1E1/A6 Merck Millipore) as positive control. Plates were incubated for 120 min at room temperature (RT) and washed 5x with PBST. Secondary antibody peroxidase AffiniPure goat anti-mouse IgG H + L (115-035-003, Jackson ImmunoResearch) 1:2,000 diluted in 1% milk-PBST for the RD4 positive control, peroxidase AffiniPure goat anti-human IgG Fcγ-specific (109-035-098, Jackson ImmunoResearch) 1:4,000 diluted in 1% milk-PBST for the plasma samples and the IgG-depleted serum negative control were dispensed into the plates using a Biotek MultifloFX dispenser. Plates were incubated for 60 min, at RT and washed 3× with PBST. 3,3′,5,5′-Tetramethylbenzidine (TMB) Chromogen Solution for ELISA (Invitrogen) was added as colorimetric horseradish peroxidase (HRP) substrate for 3 min at RT. Finally, 0.5 M H2SO4 was added to stop the reaction. Plates were briefly centrifuged after each dispensing step except after dispensing of TMB. Plates were read at Optical Density = 450 nm (OD450nm) in a plate reader (Perkin Elmer, Envision). For the replicability assessment, 308 samples were tested in duplicates, running the replicates on the same day but using different 1536-well assay (destination) plates, different plate coordinates for each replicate, and calculating the −log10(EC50) of each replicate independently. Non-specific cross-reactivity was assessed by testing plasma samples against MTBD-tau and amyloid-β pyroglutamate (12,297 hospital patients’ plasma samples) and the PrPC (13,099 hospital patients’ samples) using a similar protocol as described above. Production and purification of recombinant tau. The gene encoding the human truncated 4R-tau corresponding to MTBD-tau was cloned into a pRSET-A plasmid (Invitrogen) and expressed in Escherichia coli BL21(DE3) cells. Cultures were grown in Luria Broth (LB, Invitrogen) at 37 °C, induced with 1 mM isopropyl-β-D-thiogalactoside (IPTG) at an OD600 of 0.8 and grown for additional 6 h at 37 °C before harvesting by centrifugation (6,000g, 10 min, 4 °C). Pellets were suspended and sonicated (30 min, 4 °C) in 20 mM piperazine-N,N-bis(2-ethanesulfonic) acid (PIPES), pH 6.5, 1 mM ethylenediaminetetraacetic acid (EDTA) and 50 mM 2-mercaptoethanol. After addition of NaCl to a final concentration of 500 mM, samples were boiled (95 °C, 20 min) and centrifuged (9,000g, 30 min, 4 °C). Ammonium sulfate was slowly added to a final concentration of 55% m/v and the suspension was stirred (1 h, RT). Samples were centrifuged (15,000g, 10 min, 4 °C), pellets were resuspended in 20 mM 4-(2-hydroxyethyl)piperazine-1-ethanesulfonic acid (HEPES), pH 7.0, 2 mM dithiothreitol (DTT), passed through a 0.45 µm Acrodisc filter (Sigma), and loaded onto Sepharose SP Fast Flow resin (Cytiva). Tau was eluted using a linear salt gradient from 0 to 1 M NaCl in 20 mM HEPES, pH 7.0, 2 mM DTT. Fractions containing tau were concentrated using Amicon Ultra-15 centrifugal filter unit (10-kDa MWCO, Merck) and dialyzed overnight at 4 °C against phosphate-buffered saline (PBS, pH 7.4, Kantonsapotheke Zurich), 1 mM DTT. Pooled samples were passed through a HiLoad 26/60 Superdex75 (GE Healthcare) column. Protein samples were analyzed by SDS-PAGE and samples containing tau were concentrated using an Amicon Ultra-15 centrifugal filter unit (10-kDa MWCO). Samples were assessed by SDS-PAGE and electrospray ionization-mass spectrometry. Pure MTBD-tau samples were stored until further use at −80 °C. The concentration of tau was determined using a bicinchoninic acid assay (Pierce BCA Protein Assay Kit, Thermo Fisher). For the purification of full-length tau, a similar protocol was used with the following changes. The gene encoding the longest 4R isoform of human full-length tau protein, tau441, (tau/pET29b, Addgene #16316, gift from Peter Klein [54]) was cloned into a pRSET-A plasmid (Invitrogen). For protein expression, E. coli BL21(DE3)pLysS cells were transformed with the pRSET-A plasmid encoding tau441. Cells were grown in Overnight Express Instant TB media (Novagen) for 6 h at 37 °C and then for 12 h at 25 °C. Fractions containing full-length tau were concentrated using Amicon Ultra-15 centrifugal filter unit (30-kDa MWCO, Merck). Purification of anti-tau autoantibodies from patient samples. Heparin plasma (3−20 mL) was diluted 1:3.3 in PBS, and centrifuged at 6,000g for 10 min at 4 °C. The supernatant was loaded onto 3 mL of epoxy-MTBD-tau (prepared by overnight incubation of MTBD-tau and epoxy resin in 0.1 M NaH2PO4–NaOH, 1 M NaCl, pH 9.2) by repetitive loading of the plasma sample overnight at 4 °C. After washing with 50 mL of PBS, MTBD-tau autoantibodies were eluted 4× with 5 mL 0.1 M glycine–HCl, pH 2.5, and immediately neutralized to pH 7.0 with 1 M Tris-HCl, pH 8.5. Anti-MTBD-tau antibody-containing fractions were identified by indirect ELISA and concentrated stepwise using Amicon Ultra-15, Ultra-4, and Ultra-0.5 mL centrifugal filter units (50-kDa MWCO) up to a volume of 1 mL. Competitive ELISA. For the competitive sandwich ELISAs for the detection of tau, high-binding 384-well plates (Perkin Elmer, SpectraPlate 384 HB) were coated with 4 µg/mL BT2 tau monoclonal antibody (#MN1010, Thermo Fisher) in PBS. After coating, plates were washed 3× with PBST and then blocked with 5% SureBlock (Lubio) in PBS for 180 min at RT. Recombinant human tau441 (rPeptide) was diluted to a final concentration of 0.015 ng/mL and incubated with purified anti-tau autoantibodies 4-fold serially diluted (1:1.66 to 1:106.6) in 1:2 plasma under rotation at 500 rpm for 120 min at 37°C. Samples were transferred to BT2-coated plates and incubated for 45 min at RT. After washing 4× with PBST, plates were incubated with ab64193 (polyclonal IgG antibody; Abcam, 0.125 µg/mL) for 45 min at RT. Plates were washed 4x with PBST and incubated with peroxidase AffiniPure goat anti-Rabbit IgG (H + L) (111-035-045, Jackson ImmunoResearch), at a 1:2,000 dilution for 60 min at RT. Plates were washed 4× with PBST and 1-Step Ultra TMB-ELISA solution (Thermo Fisher) was added for 7 min at RT. After addition of 0.5 M H2SO4, plates were read at OD450 nm in a plate reader (Perkin Elmer, Envision). For the competitive ELISAs of MTBD-tau autoantibodies, high-binding 384-well plates were coated with 20 µL of 0.5 µg/mL of MTBD-tau overnight at 4 °C. Afterward, plates were washed 3× with PBST and blocked with 5% SureBlock (Lubio) in PBST for 120 min. Purified autoantibodies from hospital cohort patients’ plasma were diluted 1:50 in 1% SureBlock in PBST (sample buffer) and the anti-tau RD4 mouse monoclonal antibody to a final concentration of 0.4 µg/mL. Bovine serum albumin (BSA, Thermo Scientific), in-house purified recombinant MTBD-tau, a pool of eight synthetic peptides covering the sequence of MTBD-tau with 25 amino acids length and 10 amino acids of overlap (Genscript) and an unrelated 25 amino acid length synthetic TREM2 (Triggering receptor expressed on myeloid cells 2) peptide (GenScript) were used as competing antigens. Antibody samples were incubated overnight at 4 °C with serial 2-fold dilutions of antigen solutions in sample buffer, ranging from 20,000 to 2.44 nM. The antibody-antigen mixtures were then added to the plates and incubated for 45 min at RT. Plates were washed 3x with PBST, followed by the addition of secondary antibodies: peroxidase AffiniPure goat anti-Human IgG (H + L) (109-035-088, Jackson ImmunoResearch) at 1:3,000 dilution and peroxidase AffiniPure goat anti-mouse IgG (H + L) (115-035-003, Jackson ImmunoResearch) at 1:2,000 dilution. Secondary antibodies were incubated for 60 min at RT and plates were then washed 4× with PBST. TMB Chromogen Solution for ELISA (Invitrogen) was added to the plates and incubated for 7 min at RT. After addition of 0.5 M H2SO4, plates were read at OD450nm in a plate reader (SpectraMax Paradigm, Molecular Devices). Indirect ELISAs. To test for polyreactivity, purified anti-tau autoantibodies were tested by indirect ELISA against several antigens [55,56]. High-binding 384-well plates were coated overnight at 4 °C with 20 µL of 1 µg/mL of MTBD-tau, 10 µg/mL of DNA from calf-thymus (Sigma), 10 µg/mL of LPS from E. coli O111:B4 (Sigma), 5 µg/mL of human insulin (Sigma), 10 µg/mL of BSA, 2 µg/mL of cardiolipin solution from bovine heart (Sigma), or left uncoated. Plates were washed 3× with PBST and then blocked in 5% SureBlock (Lubio) in PBST for 120 min at RT. Patient purified anti-MTBD-tau autoantibodies were diluted 1:33, IgG-depleted plasma (BioSource) 1:50 diluted and used as negative control, anti-DNP (Sigma) diluted to 6 µg/mL, anti-tau RD4 to 6 µg/mL, and pooled plasma from 20 patients diluted 1:25 in 1% SureBlock in PBST as positive controls. Samples were serially diluted 12 times 1:1 with 1% SureBlock in PBST in the referred plates. Samples were incubated for 120 min at RT and washed 4× with PBST. 20 µL/well of secondary antibodies diluted in 1% SureBlock in PBST were added as follows: peroxidase AffiniPure goat anti-human IgG (H + L) (109-035-088, Jackson ImmunoResearch) diluted 1:3,000 and added to purified MTBD-tau autoantibodies and IKC pool wells, peroxidase AffiniPure goat anti-mouse IgG (H + L) (115-035-003, Jackson ImmunoResearch) at 1:2,500 dilution and added to anti-RD4 wells, and peroxidase AffiniPure goat anti-rabbit IgG (H + L) (111-035-045, Jackson ImmunoResearch) at 1:4,000 dilution and added to anti-DNP wells. Plates were then washed 4× with PBST. 20 µL of TMB Chromogen solution for ELISA (Invitrogen) was added and incubated for 7 min at RT. After addition of 0.5 M H2SO4, plates were read at OD450nm in a plate reader (SpectraMax Paradigm, Molecular Devices). For relative affinity measurements, we used an indirect ELISA in 384-well plates (Perkin Elmer, SpectraPlate 384 HB) with MTBD-tau as a coating antigen using the following parameters: The starting concentration of all antibodies, assayed in triplicates, was 10 µg/mL. They were successively diluted 1:3 to reach a concentration of 2 × 10−6 µg/mL. As reference antibody, we used purified RD4 kindly provided by Prof. Rohan de Silva (UCL Queen Square Institute of Neurology, London, UK). As additional controls, we used anti-LAG3 antibody Relatlimab [57], ατ− plasma sample, and uncoated plates. The respective EC50 values were then determined using logistic regression, as above, and the curves as well as the dots were visualized. For IgG subclassing, the following secondary antibodies were used: rabbit anti-human IgG1 (SA5-10202, Invitrogen), rabbit anti-human IgG2 (SA5-10203, Invitrogen), rabbit anti-human IgG3 (SA5-10204, Invitrogen) or rabbit anti-human IgG4 (SA5-10205, Invitrogen) at 1:1,500 dilution, peroxidase AffiniPure goat anti-rabbit IgG (H + L) antibody (111-035-045, Jackson ImmunoResearch) at 1:2,500 dilution. For immunoglobulin light chain typing, the following secondary antibodies were used: Goat anti-Human Kappa-HRP [58] and Goat anti-Human Lambda-HRP [58] at 1:4,000 dilution and peroxidase AffiniPure goat anti-mouse IgG (H + L) (115-035-003, Jackson ImmunoResearch) at 1:2,500 dilution. For the epitope mapping experiments, a similar approach was used in which high-binding 384-well plates were coated with 20 µL of 1 µg/mL of each MTBD-tau peptide (GenScript) in PBS overnight at 4°C. Plasma samples and human IgG-depleted serum (MyBioSource) used as negative control were diluted 1:50 and anti-tau RD4 mouse monoclonal antibody used as positive control (05−804 clone 1E1/A6 Merck Millipore) was diluted to a final concentration of 6 µg/mL in 1% SureBlock in PBST. The following secondary antibodies were used: peroxidase AffiniPure goat anti-human IgG (H + L) (109-035-088, Jackson ImmunoResearch) at 1:5,000 dilution and peroxidase AffiniPure goat anti-mouse IgG (H + L) (115-035-003, Jackson ImmunoResearch) at 1:2,500 dilution. For data analysis of IgG subclassing, immunoglobulin light chain typing, and epitope mapping experiments, samples were considered reactive when the OD450nm was higher than the average of all negatives OD450nm + 2× the standard deviation of the OD450nm of the negatives. Western blotting. SH-SY5Y wild-type (Sigma) and cells overexpressing double-mutated tauP301L/S320F were lysed in 0.5% Triton X-100 (Sigma Aldrich) in PBS, supplemented with cOmplete Mini EDTA-free Protease Inhibitor Cocktail (Roche) on ice and supernatant was recovered after centrifugation at 14,000g for 20 min at 4 °C. Protein concentration was determined using a bicinchoninic acid assay (Pierce BCA Protein Assay Kit, Thermo Fisher) and sample volumes were adapted to 30 µg of total protein. Samples were loaded onto NuPAGE 12% Bis-Tris gels (Invitrogen) and blotted to nitrocellulose membranes (Invitrogen) using a dry iBlot 2 Gel Transfer Device (Invitrogen, Thermo Fisher). Membranes were cut vertically along the protein ladder and blocked with 5% SureBlock in PBST at 30 min for RT and incubated overnight at 4 °C with patient-purified anti-MTBD-tau autoantibodies diluted 1:100 in 1% SureBlock, PBST. Anti-tau RD4 mouse monoclonal antibody diluted 1:4,000 was used as a positive control. Negative control membranes were not incubated with primary antibodies. Membranes were washed 3× for 5 min with PBST and incubated with the following secondary antibodies for 60 min at RT: peroxidase AffiniPure goat anti-mouse IgG (H + L) (115-035-003, Jackson ImmunoResearch) and peroxidase AffiniPure goat anti-human IgG (H + L) (109-035-088, Jackson ImmunoResearch) 1:10,000 diluted. Membranes were washed 4× for 5 min with PBST and developed using the Immobilon Crescendo HRP Substrate (Millipore). Imaging was performed with the Fusion SOLO S imaging system (Vilber). Immunofluorescence. SH-SY5Y cells were transfected with pRK5-EGFP-0N4Rtau plasmid (Addgene # 46904, a kind gift of Dr Karen Ashe [59]) using Lipofectamine 2000 according to manufacturer’s protocol. After 48 h cells were fixed with 4% paraformaldehyde for 20 min at RT, permeabilized with 0.5% BSA, 0.1% Triton X-100 in PBS for 20 min at RT, and blocked with 0.5% BSA in PBS for 60 min at RT. For the immunocytochemical assays, 1:25 purified ani-MTBD-tau autoantibodies diluted in 0.5% BSA in PBS were incubated with the cells for 60 min. Tau mouse monoclonal antibody HT7 (MN1000 Thermo Fisher Scientific) diluted 1:500 was used as a positive control. Cells were washed 3× with 0.5% BSA in PBS and incubated for 30 min at RT with: goat anti-mouse IgG (H + L) cross-adsorbed Alexa Fluor 555 (A-21422 Invitrogen) diluted 1:500 and counterstained with 4,6-diamidino-2-phenylindole (DAPI) 1 mg/mL (Thermo Fisher) diluted 1:10,000 in 0.5% BSA in PBS, and for human antibody and negative control stains with biotin AffiniPure goat anti-human IgG (H + L) (109-065-003, Jackson ImmunoResearch) diluted 1:200 in 0.5% BSA in PBS. Samples were 3x washed with 0.5% BSA in PBS. For human antibody stains, samples were further incubated for 30 min at RT with streptavidin Alexa Fluor 594 conjugate diluted 1:200 and counterstained with DAPI diluted 1:10,000 in 0.5% BSA in PBS. Cells were mounted on glass slides with Fluorescence Mounting Medium (Thermo Fisher Scientific) and imaged using a Leica TCS SP5 confocal laser scanning microscope. Imaging was performed with equipment maintained by the Center for Microscopy and Image Analysis, University of Zurich. In vitro MTBD-tau aggregation assay. MTBD-tau in vitro aggregation experiments were performed as previously described [31]. Briefly, 7 µM of in-house purified recombinant MTBD-tau, 3.5 µM of heparin (Santa Cruz Biotechnology) and 10 µM of ThT (Sigma) were diluted in PBS. The patient purified anti-tau autoantibodies and controls (plasma sample reactive against the LAG3 and IgG-purified using Protein G Sepharose (Cytiva)) were added at the indicated apparent stoichiometries in Fig 3. The mixtures with a total volume of 200 µl were added to black 96-well polystyrene microplates (Nunc, Prod. No. 265301) and ThT fluorescence (450/480 nm ex/em filters; bottom read mode) was measured at 37 °C under continuous orbital shaking (425 cpm) every 15 min for 96 h using a FLUOstar Omega microplate reader (BMG Labtech). The mean baseline fluorescence values were subtracted from the mean fluorescence values at each time point which were then normalized to maximum baseline-subtracted fluorescence values and multiplied by 100 [60]. Statistical analysis. Antibody titers were defined as the negative logarithm half-maximal responses (−log10(EC50)) obtained by fitting the OD450nm of the eight dilution points of each sample tested in the microELISA to a logistic regression fitter. We classified as positives samples with a cut-off of −log10(EC50) ≥1.8 [25], corresponding to a nominal dilution of >1/64. Non-informative samples (fitting error >20% −log10(EC50) or high background) were excluded from the analysis [25]. In cases where more than one sample was available for the same individual the most recent log10(EC50) value was used. Age is presented as median with interquartile range (IQR) and comparisons were performed using non-parametric Mann–Whitney U test. Categorized age groups and sex of positives and negatives are shown as percentages and compared using two-proportions Z-test or χ2 test for trend in proportions. Log-binomial regression models [28,61] using MTBD-tau autoreactivity were employed to estimate age- and sex-aRRs and 95% CIs, and to investigate the association between the detection of anti-MTBD-tau IgG autoantibodies and different demographic features. For the AD cohort microELISA screen, log10(EC50) values were compared using Mann–Whitney U test. Additionally, Bayesian logistic regression was conducted [25,61–63], using −logit10(EC50) values as outcome, i.e., without dichotomizing the outcome using the R package rstanarm and the following priors (prior=normal[0, 2.5, autoscale=TRUE],prior_intercept = normal[5000, 2.5, autoscale=TRUE)] prior_aux = exponential [1, autoscale =TRUE]. We aimed to confirm the positive association between ICD-10 codes previously identified using a conventional logistic regression model with high −log10(EC50) values. Each ICD-10 code was analyzed in an independent logistic regression and adjusted for age and sex. The association between MTBD-tau-autoreactivity and neurological and systemic disorders, respectively, was analyzed by applying multivariate log-binomial regression models to estimate aRRs and 95% CIs. For neurological disorders, 23 major groups of ICD-10 codes (S3 Table) corresponding to neurological disorders identified at least once in the positive samples were used. For systemic disorders, the 27 major groups of ICD-10 codes (S4 Table) or 276 individual ICD-10 codes corresponding to systemic disorders identified at least once in the positive samples and with at least 200 total counts were used to avoid overinterpretation of rare cases of disease. Individual disease entities with a P value <0.05 Bonferroni corrected for multiple comparisons were included in a multivariate log-binomial regression analysis. For the association between MTBD-tau-autoreactivity and clinical laboratory parameters, we used laboratory parameters with more than 2,000 total counts and calculated median values of the total values available for each patient in case of repetition of the clinical laboratory test. We used multivariate log-binomial regression models to estimate aRR and 95% CI using 106 clinical laboratory tests. Statistical significance was defined by two-tailed P-value ≤ 0.05. Statistical analyses and data visualization were performed using R version 4.3.2 and RStudio version 1.4.1106 [64]. Study approval. Collection of samples and clinical data were conducted according to study protocols approved by the Cantonal Ethics Committee of the Canton of Zurich, Switzerland (KEK-ZH Nr. 2015-0561, BASEC-Nr. 2018-01042, and BASEC-Nr. 2020-01731), in accordance with the provisions of the Declaration of Helsinki and the Good Clinical Practice guidelines of the International Conference on Harmonisation. All human donors and patients included in this study provided a written general informed consent. Study design. From December 2017 until February 2020, residual heparin plasma samples were obtained from the Department of Clinical Chemistry, University Hospital of Zurich, USZ, Switzerland. Samples were collected during routine clinical care from patients admitted either as inpatients or outpatients (age ≥18 years) and were only included if basic demographic data was available, and an informed consent for research had been provided. From March until July 2020, EDTA plasma samples from blood donors were obtained from the Blood Donation Center of Zurich, Switzerland, according to standard criteria of blood donation. Exclusion criteria were as reported [25]. Plasma samples from patients of both sexes were examined in this study. Plasma samples were biobanked locally and tested in an automated indirect microELISA ([25–27] and below) for natural IgG autoantibodies against the MTBD-tau. Demographic and clinical data for the hospital cohort were obtained from clinical records of the USZ with follow-up until December 2021, while detailed clinical data for the blood donor cohort were not available for this study. ICD-10 codes [53] were used for clinical data assessment. AD patients were selected using ICD-10 code F00 or G30. Non-neurodegeneration controls were defined by the lack of any Fxx or Gxx ICD-10 codes. Automated microELISA screen. Plasma samples were tested for natural anti-MTBD-tau IgG autoantibodies in a microELISA screen [25–27]. Briefly, high-binding 1,536-well microplates (Perkin Elmer, SpectraPlate 1536 HB) were coated with 1 µg/mL of recombinant MTBD-tau (37 °C, 60 min). Plates were washed 3× with phosphate-buffered saline 0.1% Tween20 (PBST) using a Biotek El406 washer-dispenser and blocked with 5% milk (Migros)-PBST for 90 min. Plasma samples were diluted 1:20 in 1% milk-PBST and dispensed into the MTBD-tau-coated plates using ultrasound dispensing with an ECHO 555 Liquid Handler (Labcyte/Beckman Coulter). Each sample was tested at eight serial 2-fold dilutions (1:50–1:6,000) using different volumes to a final volume of 3 µL/well. Human IgG-depleted serum (MyBioSource) was used as negative and anti-tau RD4 (4-repeat isoform) mouse monoclonal antibody [30] (05–804 clone 1E1/A6 Merck Millipore) as positive control. Plates were incubated for 120 min at room temperature (RT) and washed 5x with PBST. Secondary antibody peroxidase AffiniPure goat anti-mouse IgG H + L (115-035-003, Jackson ImmunoResearch) 1:2,000 diluted in 1% milk-PBST for the RD4 positive control, peroxidase AffiniPure goat anti-human IgG Fcγ-specific (109-035-098, Jackson ImmunoResearch) 1:4,000 diluted in 1% milk-PBST for the plasma samples and the IgG-depleted serum negative control were dispensed into the plates using a Biotek MultifloFX dispenser. Plates were incubated for 60 min, at RT and washed 3× with PBST. 3,3′,5,5′-Tetramethylbenzidine (TMB) Chromogen Solution for ELISA (Invitrogen) was added as colorimetric horseradish peroxidase (HRP) substrate for 3 min at RT. Finally, 0.5 M H2SO4 was added to stop the reaction. Plates were briefly centrifuged after each dispensing step except after dispensing of TMB. Plates were read at Optical Density = 450 nm (OD450nm) in a plate reader (Perkin Elmer, Envision). For the replicability assessment, 308 samples were tested in duplicates, running the replicates on the same day but using different 1536-well assay (destination) plates, different plate coordinates for each replicate, and calculating the −log10(EC50) of each replicate independently. Non-specific cross-reactivity was assessed by testing plasma samples against MTBD-tau and amyloid-β pyroglutamate (12,297 hospital patients’ plasma samples) and the PrPC (13,099 hospital patients’ samples) using a similar protocol as described above. Production and purification of recombinant tau. The gene encoding the human truncated 4R-tau corresponding to MTBD-tau was cloned into a pRSET-A plasmid (Invitrogen) and expressed in Escherichia coli BL21(DE3) cells. Cultures were grown in Luria Broth (LB, Invitrogen) at 37 °C, induced with 1 mM isopropyl-β-D-thiogalactoside (IPTG) at an OD600 of 0.8 and grown for additional 6 h at 37 °C before harvesting by centrifugation (6,000g, 10 min, 4 °C). Pellets were suspended and sonicated (30 min, 4 °C) in 20 mM piperazine-N,N-bis(2-ethanesulfonic) acid (PIPES), pH 6.5, 1 mM ethylenediaminetetraacetic acid (EDTA) and 50 mM 2-mercaptoethanol. After addition of NaCl to a final concentration of 500 mM, samples were boiled (95 °C, 20 min) and centrifuged (9,000g, 30 min, 4 °C). Ammonium sulfate was slowly added to a final concentration of 55% m/v and the suspension was stirred (1 h, RT). Samples were centrifuged (15,000g, 10 min, 4 °C), pellets were resuspended in 20 mM 4-(2-hydroxyethyl)piperazine-1-ethanesulfonic acid (HEPES), pH 7.0, 2 mM dithiothreitol (DTT), passed through a 0.45 µm Acrodisc filter (Sigma), and loaded onto Sepharose SP Fast Flow resin (Cytiva). Tau was eluted using a linear salt gradient from 0 to 1 M NaCl in 20 mM HEPES, pH 7.0, 2 mM DTT. Fractions containing tau were concentrated using Amicon Ultra-15 centrifugal filter unit (10-kDa MWCO, Merck) and dialyzed overnight at 4 °C against phosphate-buffered saline (PBS, pH 7.4, Kantonsapotheke Zurich), 1 mM DTT. Pooled samples were passed through a HiLoad 26/60 Superdex75 (GE Healthcare) column. Protein samples were analyzed by SDS-PAGE and samples containing tau were concentrated using an Amicon Ultra-15 centrifugal filter unit (10-kDa MWCO). Samples were assessed by SDS-PAGE and electrospray ionization-mass spectrometry. Pure MTBD-tau samples were stored until further use at −80 °C. The concentration of tau was determined using a bicinchoninic acid assay (Pierce BCA Protein Assay Kit, Thermo Fisher). For the purification of full-length tau, a similar protocol was used with the following changes. The gene encoding the longest 4R isoform of human full-length tau protein, tau441, (tau/pET29b, Addgene #16316, gift from Peter Klein [54]) was cloned into a pRSET-A plasmid (Invitrogen). For protein expression, E. coli BL21(DE3)pLysS cells were transformed with the pRSET-A plasmid encoding tau441. Cells were grown in Overnight Express Instant TB media (Novagen) for 6 h at 37 °C and then for 12 h at 25 °C. Fractions containing full-length tau were concentrated using Amicon Ultra-15 centrifugal filter unit (30-kDa MWCO, Merck). Purification of anti-tau autoantibodies from patient samples. Heparin plasma (3−20 mL) was diluted 1:3.3 in PBS, and centrifuged at 6,000g for 10 min at 4 °C. The supernatant was loaded onto 3 mL of epoxy-MTBD-tau (prepared by overnight incubation of MTBD-tau and epoxy resin in 0.1 M NaH2PO4–NaOH, 1 M NaCl, pH 9.2) by repetitive loading of the plasma sample overnight at 4 °C. After washing with 50 mL of PBS, MTBD-tau autoantibodies were eluted 4× with 5 mL 0.1 M glycine–HCl, pH 2.5, and immediately neutralized to pH 7.0 with 1 M Tris-HCl, pH 8.5. Anti-MTBD-tau antibody-containing fractions were identified by indirect ELISA and concentrated stepwise using Amicon Ultra-15, Ultra-4, and Ultra-0.5 mL centrifugal filter units (50-kDa MWCO) up to a volume of 1 mL. Competitive ELISA. For the competitive sandwich ELISAs for the detection of tau, high-binding 384-well plates (Perkin Elmer, SpectraPlate 384 HB) were coated with 4 µg/mL BT2 tau monoclonal antibody (#MN1010, Thermo Fisher) in PBS. After coating, plates were washed 3× with PBST and then blocked with 5% SureBlock (Lubio) in PBS for 180 min at RT. Recombinant human tau441 (rPeptide) was diluted to a final concentration of 0.015 ng/mL and incubated with purified anti-tau autoantibodies 4-fold serially diluted (1:1.66 to 1:106.6) in 1:2 plasma under rotation at 500 rpm for 120 min at 37°C. Samples were transferred to BT2-coated plates and incubated for 45 min at RT. After washing 4× with PBST, plates were incubated with ab64193 (polyclonal IgG antibody; Abcam, 0.125 µg/mL) for 45 min at RT. Plates were washed 4x with PBST and incubated with peroxidase AffiniPure goat anti-Rabbit IgG (H + L) (111-035-045, Jackson ImmunoResearch), at a 1:2,000 dilution for 60 min at RT. Plates were washed 4× with PBST and 1-Step Ultra TMB-ELISA solution (Thermo Fisher) was added for 7 min at RT. After addition of 0.5 M H2SO4, plates were read at OD450 nm in a plate reader (Perkin Elmer, Envision). For the competitive ELISAs of MTBD-tau autoantibodies, high-binding 384-well plates were coated with 20 µL of 0.5 µg/mL of MTBD-tau overnight at 4 °C. Afterward, plates were washed 3× with PBST and blocked with 5% SureBlock (Lubio) in PBST for 120 min. Purified autoantibodies from hospital cohort patients’ plasma were diluted 1:50 in 1% SureBlock in PBST (sample buffer) and the anti-tau RD4 mouse monoclonal antibody to a final concentration of 0.4 µg/mL. Bovine serum albumin (BSA, Thermo Scientific), in-house purified recombinant MTBD-tau, a pool of eight synthetic peptides covering the sequence of MTBD-tau with 25 amino acids length and 10 amino acids of overlap (Genscript) and an unrelated 25 amino acid length synthetic TREM2 (Triggering receptor expressed on myeloid cells 2) peptide (GenScript) were used as competing antigens. Antibody samples were incubated overnight at 4 °C with serial 2-fold dilutions of antigen solutions in sample buffer, ranging from 20,000 to 2.44 nM. The antibody-antigen mixtures were then added to the plates and incubated for 45 min at RT. Plates were washed 3x with PBST, followed by the addition of secondary antibodies: peroxidase AffiniPure goat anti-Human IgG (H + L) (109-035-088, Jackson ImmunoResearch) at 1:3,000 dilution and peroxidase AffiniPure goat anti-mouse IgG (H + L) (115-035-003, Jackson ImmunoResearch) at 1:2,000 dilution. Secondary antibodies were incubated for 60 min at RT and plates were then washed 4× with PBST. TMB Chromogen Solution for ELISA (Invitrogen) was added to the plates and incubated for 7 min at RT. After addition of 0.5 M H2SO4, plates were read at OD450nm in a plate reader (SpectraMax Paradigm, Molecular Devices). Indirect ELISAs. To test for polyreactivity, purified anti-tau autoantibodies were tested by indirect ELISA against several antigens [55,56]. High-binding 384-well plates were coated overnight at 4 °C with 20 µL of 1 µg/mL of MTBD-tau, 10 µg/mL of DNA from calf-thymus (Sigma), 10 µg/mL of LPS from E. coli O111:B4 (Sigma), 5 µg/mL of human insulin (Sigma), 10 µg/mL of BSA, 2 µg/mL of cardiolipin solution from bovine heart (Sigma), or left uncoated. Plates were washed 3× with PBST and then blocked in 5% SureBlock (Lubio) in PBST for 120 min at RT. Patient purified anti-MTBD-tau autoantibodies were diluted 1:33, IgG-depleted plasma (BioSource) 1:50 diluted and used as negative control, anti-DNP (Sigma) diluted to 6 µg/mL, anti-tau RD4 to 6 µg/mL, and pooled plasma from 20 patients diluted 1:25 in 1% SureBlock in PBST as positive controls. Samples were serially diluted 12 times 1:1 with 1% SureBlock in PBST in the referred plates. Samples were incubated for 120 min at RT and washed 4× with PBST. 20 µL/well of secondary antibodies diluted in 1% SureBlock in PBST were added as follows: peroxidase AffiniPure goat anti-human IgG (H + L) (109-035-088, Jackson ImmunoResearch) diluted 1:3,000 and added to purified MTBD-tau autoantibodies and IKC pool wells, peroxidase AffiniPure goat anti-mouse IgG (H + L) (115-035-003, Jackson ImmunoResearch) at 1:2,500 dilution and added to anti-RD4 wells, and peroxidase AffiniPure goat anti-rabbit IgG (H + L) (111-035-045, Jackson ImmunoResearch) at 1:4,000 dilution and added to anti-DNP wells. Plates were then washed 4× with PBST. 20 µL of TMB Chromogen solution for ELISA (Invitrogen) was added and incubated for 7 min at RT. After addition of 0.5 M H2SO4, plates were read at OD450nm in a plate reader (SpectraMax Paradigm, Molecular Devices). For relative affinity measurements, we used an indirect ELISA in 384-well plates (Perkin Elmer, SpectraPlate 384 HB) with MTBD-tau as a coating antigen using the following parameters: The starting concentration of all antibodies, assayed in triplicates, was 10 µg/mL. They were successively diluted 1:3 to reach a concentration of 2 × 10−6 µg/mL. As reference antibody, we used purified RD4 kindly provided by Prof. Rohan de Silva (UCL Queen Square Institute of Neurology, London, UK). As additional controls, we used anti-LAG3 antibody Relatlimab [57], ατ− plasma sample, and uncoated plates. The respective EC50 values were then determined using logistic regression, as above, and the curves as well as the dots were visualized. For IgG subclassing, the following secondary antibodies were used: rabbit anti-human IgG1 (SA5-10202, Invitrogen), rabbit anti-human IgG2 (SA5-10203, Invitrogen), rabbit anti-human IgG3 (SA5-10204, Invitrogen) or rabbit anti-human IgG4 (SA5-10205, Invitrogen) at 1:1,500 dilution, peroxidase AffiniPure goat anti-rabbit IgG (H + L) antibody (111-035-045, Jackson ImmunoResearch) at 1:2,500 dilution. For immunoglobulin light chain typing, the following secondary antibodies were used: Goat anti-Human Kappa-HRP [58] and Goat anti-Human Lambda-HRP [58] at 1:4,000 dilution and peroxidase AffiniPure goat anti-mouse IgG (H + L) (115-035-003, Jackson ImmunoResearch) at 1:2,500 dilution. For the epitope mapping experiments, a similar approach was used in which high-binding 384-well plates were coated with 20 µL of 1 µg/mL of each MTBD-tau peptide (GenScript) in PBS overnight at 4°C. Plasma samples and human IgG-depleted serum (MyBioSource) used as negative control were diluted 1:50 and anti-tau RD4 mouse monoclonal antibody used as positive control (05−804 clone 1E1/A6 Merck Millipore) was diluted to a final concentration of 6 µg/mL in 1% SureBlock in PBST. The following secondary antibodies were used: peroxidase AffiniPure goat anti-human IgG (H + L) (109-035-088, Jackson ImmunoResearch) at 1:5,000 dilution and peroxidase AffiniPure goat anti-mouse IgG (H + L) (115-035-003, Jackson ImmunoResearch) at 1:2,500 dilution. For data analysis of IgG subclassing, immunoglobulin light chain typing, and epitope mapping experiments, samples were considered reactive when the OD450nm was higher than the average of all negatives OD450nm + 2× the standard deviation of the OD450nm of the negatives. Western blotting. SH-SY5Y wild-type (Sigma) and cells overexpressing double-mutated tauP301L/S320F were lysed in 0.5% Triton X-100 (Sigma Aldrich) in PBS, supplemented with cOmplete Mini EDTA-free Protease Inhibitor Cocktail (Roche) on ice and supernatant was recovered after centrifugation at 14,000g for 20 min at 4 °C. Protein concentration was determined using a bicinchoninic acid assay (Pierce BCA Protein Assay Kit, Thermo Fisher) and sample volumes were adapted to 30 µg of total protein. Samples were loaded onto NuPAGE 12% Bis-Tris gels (Invitrogen) and blotted to nitrocellulose membranes (Invitrogen) using a dry iBlot 2 Gel Transfer Device (Invitrogen, Thermo Fisher). Membranes were cut vertically along the protein ladder and blocked with 5% SureBlock in PBST at 30 min for RT and incubated overnight at 4 °C with patient-purified anti-MTBD-tau autoantibodies diluted 1:100 in 1% SureBlock, PBST. Anti-tau RD4 mouse monoclonal antibody diluted 1:4,000 was used as a positive control. Negative control membranes were not incubated with primary antibodies. Membranes were washed 3× for 5 min with PBST and incubated with the following secondary antibodies for 60 min at RT: peroxidase AffiniPure goat anti-mouse IgG (H + L) (115-035-003, Jackson ImmunoResearch) and peroxidase AffiniPure goat anti-human IgG (H + L) (109-035-088, Jackson ImmunoResearch) 1:10,000 diluted. Membranes were washed 4× for 5 min with PBST and developed using the Immobilon Crescendo HRP Substrate (Millipore). Imaging was performed with the Fusion SOLO S imaging system (Vilber). Immunofluorescence. SH-SY5Y cells were transfected with pRK5-EGFP-0N4Rtau plasmid (Addgene # 46904, a kind gift of Dr Karen Ashe [59]) using Lipofectamine 2000 according to manufacturer’s protocol. After 48 h cells were fixed with 4% paraformaldehyde for 20 min at RT, permeabilized with 0.5% BSA, 0.1% Triton X-100 in PBS for 20 min at RT, and blocked with 0.5% BSA in PBS for 60 min at RT. For the immunocytochemical assays, 1:25 purified ani-MTBD-tau autoantibodies diluted in 0.5% BSA in PBS were incubated with the cells for 60 min. Tau mouse monoclonal antibody HT7 (MN1000 Thermo Fisher Scientific) diluted 1:500 was used as a positive control. Cells were washed 3× with 0.5% BSA in PBS and incubated for 30 min at RT with: goat anti-mouse IgG (H + L) cross-adsorbed Alexa Fluor 555 (A-21422 Invitrogen) diluted 1:500 and counterstained with 4,6-diamidino-2-phenylindole (DAPI) 1 mg/mL (Thermo Fisher) diluted 1:10,000 in 0.5% BSA in PBS, and for human antibody and negative control stains with biotin AffiniPure goat anti-human IgG (H + L) (109-065-003, Jackson ImmunoResearch) diluted 1:200 in 0.5% BSA in PBS. Samples were 3x washed with 0.5% BSA in PBS. For human antibody stains, samples were further incubated for 30 min at RT with streptavidin Alexa Fluor 594 conjugate diluted 1:200 and counterstained with DAPI diluted 1:10,000 in 0.5% BSA in PBS. Cells were mounted on glass slides with Fluorescence Mounting Medium (Thermo Fisher Scientific) and imaged using a Leica TCS SP5 confocal laser scanning microscope. Imaging was performed with equipment maintained by the Center for Microscopy and Image Analysis, University of Zurich. In vitro MTBD-tau aggregation assay. MTBD-tau in vitro aggregation experiments were performed as previously described [31]. Briefly, 7 µM of in-house purified recombinant MTBD-tau, 3.5 µM of heparin (Santa Cruz Biotechnology) and 10 µM of ThT (Sigma) were diluted in PBS. The patient purified anti-tau autoantibodies and controls (plasma sample reactive against the LAG3 and IgG-purified using Protein G Sepharose (Cytiva)) were added at the indicated apparent stoichiometries in Fig 3. The mixtures with a total volume of 200 µl were added to black 96-well polystyrene microplates (Nunc, Prod. No. 265301) and ThT fluorescence (450/480 nm ex/em filters; bottom read mode) was measured at 37 °C under continuous orbital shaking (425 cpm) every 15 min for 96 h using a FLUOstar Omega microplate reader (BMG Labtech). The mean baseline fluorescence values were subtracted from the mean fluorescence values at each time point which were then normalized to maximum baseline-subtracted fluorescence values and multiplied by 100 [60]. Statistical analysis. Antibody titers were defined as the negative logarithm half-maximal responses (−log10(EC50)) obtained by fitting the OD450nm of the eight dilution points of each sample tested in the microELISA to a logistic regression fitter. We classified as positives samples with a cut-off of −log10(EC50) ≥1.8 [25], corresponding to a nominal dilution of >1/64. Non-informative samples (fitting error >20% −log10(EC50) or high background) were excluded from the analysis [25]. In cases where more than one sample was available for the same individual the most recent log10(EC50) value was used. Age is presented as median with interquartile range (IQR) and comparisons were performed using non-parametric Mann–Whitney U test. Categorized age groups and sex of positives and negatives are shown as percentages and compared using two-proportions Z-test or χ2 test for trend in proportions. Log-binomial regression models [28,61] using MTBD-tau autoreactivity were employed to estimate age- and sex-aRRs and 95% CIs, and to investigate the association between the detection of anti-MTBD-tau IgG autoantibodies and different demographic features. For the AD cohort microELISA screen, log10(EC50) values were compared using Mann–Whitney U test. Additionally, Bayesian logistic regression was conducted [25,61–63], using −logit10(EC50) values as outcome, i.e., without dichotomizing the outcome using the R package rstanarm and the following priors (prior=normal[0, 2.5, autoscale=TRUE],prior_intercept = normal[5000, 2.5, autoscale=TRUE)] prior_aux = exponential [1, autoscale =TRUE]. We aimed to confirm the positive association between ICD-10 codes previously identified using a conventional logistic regression model with high −log10(EC50) values. Each ICD-10 code was analyzed in an independent logistic regression and adjusted for age and sex. The association between MTBD-tau-autoreactivity and neurological and systemic disorders, respectively, was analyzed by applying multivariate log-binomial regression models to estimate aRRs and 95% CIs. For neurological disorders, 23 major groups of ICD-10 codes (S3 Table) corresponding to neurological disorders identified at least once in the positive samples were used. For systemic disorders, the 27 major groups of ICD-10 codes (S4 Table) or 276 individual ICD-10 codes corresponding to systemic disorders identified at least once in the positive samples and with at least 200 total counts were used to avoid overinterpretation of rare cases of disease. Individual disease entities with a P value <0.05 Bonferroni corrected for multiple comparisons were included in a multivariate log-binomial regression analysis. For the association between MTBD-tau-autoreactivity and clinical laboratory parameters, we used laboratory parameters with more than 2,000 total counts and calculated median values of the total values available for each patient in case of repetition of the clinical laboratory test. We used multivariate log-binomial regression models to estimate aRR and 95% CI using 106 clinical laboratory tests. Statistical significance was defined by two-tailed P-value ≤ 0.05. Statistical analyses and data visualization were performed using R version 4.3.2 and RStudio version 1.4.1106 [64]. Supporting information S1 Table. Targeted AD screen samples. https://doi.org/10.1371/journal.pbio.3003488.s001 (DOCX) S2 Table. Laboratory parameters used in the statistical analysis. https://doi.org/10.1371/journal.pbio.3003488.s002 (DOCX) S3 Table. ICD-10 codes used for the grouping of neurological disorders in the statistical analysis. https://doi.org/10.1371/journal.pbio.3003488.s003 (DOCX) S4 Table. ICD-10 codes used for the grouping of systemic disorders in the statistical analysis. https://doi.org/10.1371/journal.pbio.3003488.s004 (DOCX) S1 Data. The numerical values of all replicates in Figs 1B, 1C, 1E—1G, 2A, 2C, 2D, 2G—2I, 3A, 3C, 3D, and 5B. https://doi.org/10.1371/journal.pbio.3003488.s005 (XLSX) S1 Raw Images. Raw images for the Western blots presented in Fig 2F. https://doi.org/10.1371/journal.pbio.3003488.s006 (PDF) S1 Fig. Epitope mapping of patient samples shown in Fig 3. Epitope mapping of the same samples used in the assay shown in Fig 3C and 3D against eight 25mer MTBD-tau peptides overlapping 10 residues. https://doi.org/10.1371/journal.pbio.3003488.s007 (TIFF) S2 Fig. RR for tau autoantibodies in autoimmune diseases. Forest plot showing the risk ratios and 95% CI (I bars) for the detection of tau autoantibodies in plasma samples of patients according to ICD-10 diagnosis of 6 different autoimmune diseases. aRR and 95% CI were estimated using log-binomial regression multivariate models including the respective variable, age, and sex. https://doi.org/10.1371/journal.pbio.3003488.s008 (TIFF) S3 Fig. Sequence alignment comparing the sequence of MTBD-tau (residues 13–114) to human NAMPT (residues 368–461; P43490). The alignment results from a BLASTP search with the BLOSUM62 substitution matrix (https://blast.ncbi.nlm.nih.gov/). The search was conducted against proteins present in the human kidney and urinary proteome (UniProt/Swiss-Prot). The alignment revealed 28% sequence identity. https://doi.org/10.1371/journal.pbio.3003488.s009 (PDF) Acknowledgments The authors wish to thank the hospital patients and blood donors for their generous altruistic contributions to this study. Imaging was performed with equipment maintained by the Center for Microscopy and Image Analysis, University of Zurich, Switzerland, and electrospray ionization-mass spectrometry at the Functional Genomics Center Zurich, University of Zurich, Switzerland. We thank Prof. Rohan de Silva (UCL Queen Square Institute of Neurology, London, UK) for providing the purified anti-MTBD-tau antibody RD4. We thank Dr. Marco Losa for the generous provision of secondary antibodies for light chain typing and Magdalena Bialkowska, Lisa Caflisch, Berre Doğançay, Julie Domange, Marigona Imeri, Lorène Mottier, Rea Müller, Antonella Rosati, Dezirae Schneider, and Anne Wiedmer for help with the high-throughput assays. Insightful advice about programming in R software was provided by Reto Guadagnini.

Clark, Frances K.;Weissbart, Gauthier;Wang, Xihang;Harline, Kate;Li, Chun-Biu;Formosa-Jordan, Pau;Roeder, Adrienne H. K.

doi: 10.1371/journal.pbio.3003469pmid: 41183025

Introduction The Arabidopsis thaliana (hereafter Arabidopsis) mature leaf blade epidermis contains three main cell types: stomatal guard cells, trichomes, and pavement cells [1]. Stomatal guard cells surround stomatal pores through which gas exchange occurs, and trichomes are large branched hair cells that serve to discourage herbivory, among other functions [2]. All other epidermal cells in the mature leaf blade epidermis (the expanded part of the leaf between the midrib and the margin) are classified as pavement cells. However, pavement cells are not a homogeneous group of cells, but rather exhibit a variety of sizes, ploidies, and shapes [3,4]. Much research has focused on the patterning of stomata [5–7] and trichomes [8,9], leading to important insights into how the regulation of intercellular signaling, cell fate specification, the cell cycle, and polarized cell division orientation give rise to their spatial arrangement. However, the patterning of pavement cells is understudied. In particular, little is known about how some pavement cells are specified to become larger and more highly polyploid than others. Pavement cell-size patterning has been studied in the Arabidopsis sepal. Pavement cells in the sepal vary in size and ploidy, with some cells reaching up to 800 μm in length (Fig 1A) and having ploidies up to 32C [10]. These very large pavement cells that have a characteristic highly anisotropic shape and bulge out of the epidermis have been named “giant cells” [10], and these form when a cell endoreduplicates early during growth [10]. Endoreduplicating cells replicate their DNA but do not enter mitosis or divide and instead continue to grow and increase their ploidy. Once a cell enters endoreduplication, it is thought to terminally differentiate and almost never re-enters the mitotic cycle [10]. Similar numbers of giant cells form on sepals within an Arabidopsis plant and among plants, but the precise spatial arrangement of giant cells differs from sepal to sepal. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. The genetic pathway that regulates giant cell development in sepals. (A–G) Cell area heat maps (in μm2) of the abaxial (outer) surface of a stage 14 adult sepal of (A) wild type, (B) acr4-2, (C) atml1-3, (D) dek1-4, (E) lgo-2, (F) LGO-OX (pATML1::LGO), and (G) ATML1-OX (pPDF1::FLAG-ATML1). Scale bar represents 100 µm. (H) The ordering of genes into a genetic pathway according to double-mutant phenotypic analysis. The underlying data for this figure can be found at Open Science Framework (osf.io), https://doi.org/10.17605/OSF.IO/RFCWS. https://doi.org/10.1371/journal.pbio.3003469.g001 Forward-genetic screens have identified the genes involved in sepal giant cell patterning, and double mutant analysis has allowed these genes to be ordered within a genetic pathway [10–13] (Fig 1). The homeodomain leucine zipper Class IV transcription factor Arabidopsis thaliana MERISTEM LAYER1 (ATML1) promotes giant cell specification in a dose-dependent manner [11,12]. Loss of ATML1 function in sepals greatly reduces giant cell number, and overexpression of ATML1 leads to ectopic giant cell formation (Fig 1A, 1C, and 1G) [11,12]. ATML1 protein concentration fluctuates in the protodermal nuclei of developing sepals [12]. High concentrations of ATML1 reached during the G2 phase of the cell cycle are strongly correlated with giant cell differentiation, consistent with a model in which an ATML1 concentration that surpasses a threshold in G2 results in giant cell specification, early endoreduplication, and giant cell differentiation [12]. The receptor-like kinase ARABIDOPSIS CRINKLY 4 (ACR4) functions upstream of ATML1 to promote giant cell formation [11,12,14–16] (Fig 1B and 1H). Loss of function of ACR4 leads to a modest reduction in the number of giant cells [11] (Fig 1A and 1B). The calpain protease DEFECTIVE KERNEL (DEK1) and the CDK inhibitor LOSS OF GIANT CELLS FROM ORGANS (LGO; also known as SIAMESE-RELATED 1, SMR1) function genetically downstream of ATML1 to promote giant cell formation [12] (Fig 1H). A hypomorphic mutant dek1 allele (dek1-4) results in the complete loss of giant cells from sepals [11] (Fig 1D). Similarly, sepals from plants homozygous for a loss-of-function mutation in LGO have no giant cells [10,11] (Fig 1E), and overexpression of LGO increases giant cell number [11] (Fig 1F). It is unknown whether this genetic pathway affects cell size only in the sepal or whether it is also a more general mechanism of epidermal cell-size patterning in other organs. Leaf pavement cell size is affected by the family of CDK inhibitors that includes LGO, known as the SIAMESE/SIAMESE-RELATED (SIM/SMR) family [17,18]. SMR proteins bind to cyclin CDK complexes and inhibit their phosphorylation of downstream targets [18]. lgo mutants lack large pavement cells and have a reduction in leaf cell endoreduplication compared with wild type [10,17,18]. In lgo mutants, pavement cells that should be mature continue to divide [19]. Furthermore, overexpression of the closely related paralog of LGO, SIM, results in larger and more highly endoreduplicated leaf epidermal pavement cells [17]. In sepals, LGO upregulates defense response genes, including glucosinolate biosynthesis genes [20], whereas in leaves, ATML1 promotes the formation of ER bodies, which contain components of the glucosinolate system, in large pavement cells [21], suggesting a common role for large cells in defense response. Whether the same upstream components of the sepal giant cell pathway also function in leaf cell-size patterning has not been thoroughly investigated. One study compared pavement cell size in dek1-4 and wild-type cotyledons and found no evidence that the cells differed in ploidy [22]. However, true leaves were not examined. In leaves and sepals, it is unknown whether giant cells exhibit a spatially ordered pattern across the organ, or if instead their spatial arrangement is random. Other epidermal cell types are nonrandomly distributed across the leaf tissue. For instance, trichomes do not form in adjacent cells due to lateral inhibition mediated by diffusible signals [9], and stomata rarely differentiate in adjacent cells due to both lineage-specific division orientation and intercellular signaling [23]. In contrast to stomata and trichomes, sepal giant cells can be in contact with one another. However, it is unknown whether such giant cell contacts are likely to be formed by chance. Due to their large shapes, quantifying the spatial arrangement of giant cells has remained challenging, and standard methods for assessing point pattern randomness are not applicable [24–27]. Here, we imaged and analyzed large areas of leaves to obtain a holistic understanding of both the cell-size distributions and the spatial arrangements of epidermal pavement cells in the leaf blade. We discovered that the genetic pathway that controls sepal giant cell formation also has a broader role in patterning epidermal pavement cell size in leaves. We quantified the spatial organization of giant cells using simulated randomized tissues and found that giant cells tend to cluster together in both mature leaves and sepals more than expected by chance, reflecting the tissue’s developmental history. Using modeling and data analysis, we found that giant cells emerge randomly in space at early stages of development, but the division of surrounding cells causes the spatial pattern to become nonrandom and clustered within the context of the whole tissue over time. Our computational modeling supports the notion that a nonrandom clustered pattern can emerge in a proliferating tissue over developmental time in a cell-autonomous and stochastic manner. Results Arabidopsis leaves exhibit a large range of pavement cell sizes, similar to sepals In sepals, giant cells are easily visible because they are highly elongated (S1A Fig). Similarly, we observe large and highly anisotropic cells in cauline leaves (S1B Fig). In rosette leaves, pavement cells of the epidermis are jigsaw puzzle-piece shaped with lobes and necks, such that cell size is not readily apparent by eye (S1C Fig); however, heterogeneity in pavement cell sizes has been previously observed [4,28]. Therefore, we wondered to what extent the distribution of cell sizes observed in sepals, ranging from giant cells to small cells, also occurs in rosette leaves. We imaged large sections of the blade (excluding midrib and margin cells) of leaf 1 or 2 from wild-type plants at 25 days postgermination (dpg) when the leaves are fully expanded and mature. Leaves 1 and 2 initiate simultaneously and are indistinguishable; therefore, we refer to them interchangeably as leaf 1 or 2. We measured the area of the epidermal cells of leaves 1 or 2 and sepals on the abaxial (bottom) side (Fig 2A and 2C). Throughout our analysis, we use the term size to mean cell area because area has been shown to be a more relevant measure of cell size than volume in highly vacuolated plant epidermal cells [4]. We observed that the abaxial cell-size distributions for both sepals and leaves are asymmetric, with long tails representing large cells (Fig 2E). Sepal giant cells are larger outliers in size and, consequently, the cell-size distribution in the sepal has a more extended tail than in the leaf (Fig 2E). Still, we observed that the cell-size range in the leaf and sepal are similar and the largest cells of the sepal are about the same size as the largest cells of the leaf (Fig 2A–2E). We conclude that Arabidopsis leaves have a diverse range of cell sizes characterized by a long-tailed distribution, similar to the abaxial side of sepals. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Abaxial and adaxial cell-size distribution in the wild-type leaf and sepal epidermis; size correlates with DNA content. (A–D) Cell area heat maps (in µm2) of (A) abaxial surface of wild-type sepal, (B) adaxial surface of wild-type sepal, (C) abaxial surface of 25-dpg wild-type leaf 1 or 2 (cell density: 234 cells mm−2), and (D) adaxial surface of 25-dpg wild-type leaf 1 or 2 (cell density: 156 cells mm−2). Scale bars represent 100 µm. (E) Violin and strip plots of cell areas of abaxial and adaxial sides of 25-dpg wild-type leaves (two pooled replicates) and adult wild-type sepals (three pooled replicates). (F) Adaxial side (green) and abaxial side (purple) of 25-dpg leaf cell area versus DNA content as measured by H2B-TFP total nuclear fluorescence, with R2 = 0.85 for the abaxial side and R2 = 0.82 for the adaxial side (one of two replicates). See S2 Fig for replicates. The underlying data for this figure can be found at Open Science Framework (osf.io), https://doi.org/10.17605/OSF.IO/RFCWS. https://doi.org/10.1371/journal.pbio.3003469.g002 Large cells are formed on the adaxial side as well as the abaxial side of the leaf In sepals, giant cells are restricted to the abaxial (outer) surface (Figs 2A, 2B, and S2A–S2D). We asked whether there was a difference in cell size between adaxial (top) and abaxial (bottom) surfaces of the leaf. Large cells of similar size are formed on both the adaxial and abaxial surfaces of the leaf, in contrast to the sepal (Figs 2A–2E and S2A–S2F). In addition, across the whole cell-size distribution, the cell density is lower and many cells are slightly more expanded on the adaxial side than the abaxial side (Figs 2C, 2D, S2E, and S2F). There are a greater number of stomata and stomatal lineage cells (very small cells) on the abaxial side compared with the adaxial side (Figs 2C–2E, S2E, and S2F). We also observed that the abaxial cells are more lobed than the adaxial cells (Figs 2C, 2D, S2E, and S2F). Despite slight differences, the cell-size distributions of the abaxial and adaxial sides of the leaf are quite similar, particularly in the tails, where both sides exhibit a similar range of larger cells, in contrast to the sepal, where only the abaxial side has very large cells. Cell area correlates with DNA content Cell area and ploidy are positively correlated in leaf epidermal cells [4]. A recent study showed that fluorescence levels of histone fluorescence reporters measured from microscopy images are a good proxy to infer nuclear ploidy in Arabidopsis cotyledons [29]. Hence, to validate whether cell area and ploidy were correlated in our leaves, we measured DNA content by quantifying total fluorescence of Histone 2B-TFP (pUBQ10::H2B-TFP) within each cell nucleus of the 25-dpg leaf images. For this reporter, the H2B-TFP signal distribution was noisy and continuous, not divided into four discrete ploidy peaks, providing an approximation of ploidy, not exact DNA content. As expected, a strong linear correlation between H2B-TFP fluorescence and cell area was observed for both the abaxial pavement cells (R2 = 0.85 and 0.91; n = 2) and the adaxial pavement cells (R2 = 0.79 and 0.82; n = 2) (Figs 2F and S2G). Therefore, we focus on analyzing cell size, and infer that large cell size indicates high ploidy. We wondered whether cells of similar size on the abaxial and adaxial side of the same leaf also have a similar DNA content. We found that cells of similar DNA content are larger on the adaxial side than on the abaxial side (Figs 2F and S2G), suggesting that adaxial cells have expanded more than abaxial cells, as noted above. Because the largest sepal cells and the largest leaf cells had approximately the same areas, we asked whether the DNA content of these cells was also similar. We found that the total H2B-TFP fluorescence values of the largest cells were very similar between sepal and leaf, suggesting that these largest cells are similar in ploidy (S2H Fig). Cell-size patterning emerges at the tip and progresses basipetally as the leaf differentiates To determine how the cell-size pattern emerges in the leaf during development, we imaged both the adaxial and abaxial surfaces of each leaf at different stages of development. From 5 to 9 dpg, cell size increases greatly (S3 Fig), as expected. At day 5, cells throughout the blade are fairly homogeneous in size, with a few cells starting to expand near the distal tip, and the large cells of the margin and overlying midrib already apparent (Fig 3A). We focus on pavement cells in the blade and exclude margin and midrib cells from further analysis. The cell-size pattern consisting of large cells interspersed between small cells progressively develops basipetally from the tip (Fig 3A–3C), whereas at the base the cells remain uniformly small. The progression of cell-size patterning down the leaf is consistent with the well-established basipetal wavefront of differentiation and cessation of cell division [30]. The cell area distributions showed that more large cells appear throughout development and the maximal cell size increases (Fig 3A, 3B, and 3D). By 9 dpg, cell size has been patterned almost to the base of the leaf (Fig 3A and 3B). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Cell-size patterning occurs as a basipetal wave simultaneously on the adaxial and abaxial sides of the leaf. (A–B) Cell area heat maps (in µm2) of leaf 1 or 2 at different stages of development on (A) the abaxial side and on (B) the adaxial side of the same leaf at 5, 6, 7, 8, and 9 dpg (half leaf). Unsegmented regions on adaxial leaves correspond to trichomes, which were not considered in this analysis. Each stage is associated with a distinct heat map color range. Scale bars represent 50 µm at 5 and 6 dpg, and 100 µm at 7, 8, and 9 dpg. (C) Spatial positions of the largest cells (those above an area threshold, see below) on the abaxial (purple points) and adaxial (green points) sides of the same leaf at 6, 7, 8, and 9 dpg. Area thresholds for each leaf were determined from the 98th percentile cell area of the abaxial side. (Note that the midrib and margin cells are included in these overlays of large cell positions.) (D) Violin plots of cell areas (in µm2) of the largest cells (as defined in (C) and excluding margin and midrib cells) on abaxial and adaxial sides of leaves at different developmental stages. Abaxial and adaxial sides are from the same leaf. See also S3 Fig for the leaves shown to scale. The underlying data for this figure can be found at Open Science Framework (osf.io), https://doi.org/10.17605/OSF.IO/RFCWS. https://doi.org/10.1371/journal.pbio.3003469.g003 We next asked whether the wavefront of cell-size patterning progresses basipetally at the same rate on the abaxial and adaxial sides of the leaf. Using images of both the abaxial and adaxial sides of the same leaf, we plotted the positions of the centers of the largest cells on both sides (including margin and midrib cells for landmarks) (Fig 3C). Large pavement cells are observed in the same proximal–distal region on abaxial and adaxial sides during development (Fig 3C). The region expands in the proximal direction as development progresses. This suggests that the wavefront of patterning and differentiation is coordinated across the abaxial/adaxial axis of the leaf. The sepal giant cell specification pathway also patterns giant cells in leaves Because the cell-size distributions have similarities in leaves and sepals, we tested whether the giant cell specification pathway in sepals (Fig 1H) also functions in the leaf to pattern cell size. We imaged leaf 1 or 2 at both 9 and 25 dpg from wild-type and giant cell pathway mutants. At 9 dpg, patterning has just extended to the base of the leaf, and the leaf is still small enough that we could image the whole upper abaxial quadrant to determine the pattern over a large fraction of the leaf blade (Figs 4A, S4, and S5). At 25 dpg, the leaf is fully differentiated, fully expanded, and the pattern is established (Figs 4B, S6, and S7). We found that cell-size patterning in the leaf is similarly affected in the mutants at both 9 and 25 dpg as in the mature sepal (Figs 4, S8A, and S8B). Notably, the largest cells show similar variations in their numbers across genotypes. Similar to the sepal, the size of the largest cells is moderately reduced in acr4-2 mutants (Figs 1B and 4A–4D), and more greatly reduced in atml1-3 mutants (Figs 1C and 4A–4D). The reduction in large cells is drastic in dek1-4 and lgo-2 mutant sepals and leaves, resulting in the absence of a long tail in the cell-size distribution (Figs 1D–1E and 4A–4D). For these genotypes, the number of medium-sized cells is also substantially decreased (Figs 1D, 1E, 4C, and 4D). Conversely, the overexpression of ATML1 (ATML1-OX) or LGO (LGO-OX) leads to an increase in the size of large cells and in fewer small cells compared with wild type, as in the sepal (Figs 1F, 1G, and 4A–4D). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. The sepal giant cell specification pathway also patterns cell size in leaves. (A, B) Cell area heat maps (in µm2) of the upper abaxial quadrant of leaf 1 or 2 at 9 dpg (A) and of a section on the abaxial side of leaf 1 or 2 at 25 dpg (B) for the following genotypes: wild type, acr4-2, atml1-3, dek1-4, lgo-2, LGO-OX (pATML1::LGO), and ATML1-OX (pPDF1::FLAG-ATML1). Scale bars represent 100 µm. Cell area heat maps of other replicates are shown in S4–S7 Figs. (C, D) Violin plots of cell area distributions on a log10 scale for 9-dpg replicates (C) and for 25-dpg replicates (D). Stomata were classified and removed in (C, D). Violin plots of individual replicates are shown in S8A and S8B Fig. (E) Wasserstein distance plot of normalized cell-size distributions (see Materials and methods) for all 9- and 25-dpg replicates displayed as Euclidean distances embedded in 2D. The 25-dpg replicates are indicated by circular dots and 9-dpg replicates by triangular dots. The Wasserstein distance plot for 9- and 25-dpg and Wasserstein statistical tests among replicates are shown in S4, S6 and S8 Figs. Associated with S4–S8 Figs. The underlying data for this figure can be found at Open Science Framework (osf.io), https://doi.org/10.17605/OSF.IO/RFCWS. https://doi.org/10.1371/journal.pbio.3003469.g004 To quantify the variations in the number of large cells precisely, we quantitatively defined leaf giant cells on the basis of a cell area threshold. Specifically, we first classified pavement cells and stomata using a Support Vector Machine classifier based on features of cell shape (see Materials and methods, S9 Fig). Next, a cell-size threshold was established in the mature sepal and in the leaf, at both 9 and 25 dpg, using the atml1-3 mutants, which are known to have very few giant cells in sepals (see Materials and methods, Figs 1C and S9). Those cells in the 9- and 25-dpg leaves as well as in the sepal that exceeded their associated threshold were categorized as giant cells (see cell-type classification outcomes in S10 and S11 Figs). On the basis of this definition, we performed a quantitative comparison and statistically compared the number of giant cells per unit area among genotypes in leaves. Two-sample, two-tailed t tests showed that in the 9-dpg leaf and the mature leaf, wild type had significantly more giant cells than lgo-2 (9 dpg: p = 0.002, 25 dpg: p = 0.003), dek1-4 (9 dpg: p = 0.002, 25 dpg: p = 0.002), atml1-3 (9 dpg: p = 0.002, 25 dpg: p = 0.005), and acr4-2 (9 dpg: p = 0.010, 25 dpg: p = 0.044). Conversely, LGO-OX had significantly more giant cells than wild type (9 dpg: p = 0.001, 25 dpg: p = 0.003). However, no statistically significant difference in the number of giant cells per unit area was observed between wild type and ATML1-OX (9 dpg: p = 0.213, 25 dpg: p = 0.75). Because the giant cells in ATML1-OX are so much bigger than wild-type giant cells, each of the giant cells in a given area of ATML1-OX leaf takes up a large amount of space, resulting in few giant cells per unit area despite the fact that most of the unit area is occupied by giant cells. We sought to quantify what was apparent visually by comparing the fractional area occupied by giant cells between ATML1-OX and wild type and found that the fractional area occupied by giant cells was significantly higher in ATML1-OX (9 dpg: p < 0.005, 25 dpg: p < 0.005). Thus, in ATML1-OX the number of giant cells is not changed, but the fractional area covered by giant cells is increased. Collectively, the similarities in the variation between the number of giant cells in the leaf and the sepal indicates that the sepal giant cell specification pathway also regulates the formation of giant cells in leaves. Giant cell mutants affect the entire cell-size distribution We observed that not only are giant cells affected in these mutants, but other aspects of the cell-size distribution are also affected. For example, the number of medium-sized cells in lgo-2 and dek1-4 is reduced in addition to the number of giant cells (Fig 4A–4D) and, correspondingly, the number of small cells is increased in these mutants. To statistically analyze the difference in cell-size distributions, we conducted a principal coordinate analysis based on the Wasserstein distances between cell-size distributions (termed Wasserstein distance plot), which showed the difference between leaf samples according to their cell-size distributions on a 2-dimensional plane (S4H, S6H, and S8C–S8F Figs, see Materials and methods). Samples clustered according to genotype, indicating that genotype controls the cell-size distribution. We observed a progressive increase in the number of giant cells along the first principal coordinate V1 from lgo-2 mutants to ATML1-OX and LGO-OX (S4H and S6H Figs). ATML1-OX and LGO-OX were distant from each other in this plot, which might partly reflect the fact that LGO-OX has more giant cells, whereas ATML1-OX has fewer but larger giant cells. When we created the combined Wasserstein distance plot using normalized cell-size distributions from both 9- and 25-dpg leaves (see Materials and methods), the samples continued to group according to genotype rather than developmental stage, further supporting the idea that these genes have affected the cell-size distribution by 9 dpg (Fig 4E). Thus, we conclude that these genes affect the entire cell-size distribution. However, some differences in the cell-size distribution are apparent between 9-dpg and mature 25-dpg leaves. Firstly, at 9 dpg, dek1-4, and lgo-2 mutants are very similar; however, in the fully mature 25-dpg leaves, the lgo-2 cell-size range is notably smaller than that in the dek1-4 mutant (Fig 4A–4D), suggesting that lgo-2 cells continue to divide after 9 dpg. In addition, the small cells in lgo-2 mutants were more uniform in size than all of the other genotypes because the typical small stomatal lineage cells that encircle the stomata in mature leaves were fewer in lgo-2 (Fig 4A–4D). This altered cell-size distribution relates to the previous finding that LGO affects pavement cell differentiation in these stomatal lineage ground cells and that cells undergo division for a longer time in the absence of LGO [19]. Secondly, although at 9 dpg the LGO-OX giant cells were slightly smaller than the ATML1-OX giant cells, at 25 dpg, the LGO-OX giant cells were nearly equivalent in size to ATML1-OX giant cells (Fig 4A–4D). In addition, we observed that more pavement cells were larger in LGO-OX, whereas only a few cells became giant in ATML1-OX (Fig 4A–4D). ATML1-OX leaves had a few connected giant cells separating large islands of small cells, whereas LGO-OX leaves showed more giant cells interspersed among smaller clusters of small cells (Fig 4A and 4B and S10 and S11) ). These phenotypic differences might reflect either inherent differences in ATML1 and LGO activities or the fact that ATML1 and LGO overexpression transgenes are under the control of different promoters that might have differences in activity at different developmental stages. Relationship between the size and shape of cells and organs In plants, compensation is the process by which organ size is maintained when cell number is altered by an accompanying change in cell size [31]. We observed compensation in the leaf giant cell mutants (S12 Fig). Mature leaves of the mutants acr4-2, atml1-3, dek1-4, and lgo-2 are similar in size to wild-type leaves. Having fewer giant cells is compensated by having more small pavement cells (Fig 4). However, ATML1-OX and LGO-OX mature leaves, which have much larger cells (see, e.g., Fig 4B and 4D), are smaller than wild type (S12N–S12P Fig). Therefore, only partial compensation for having fewer cells by having larger cells is observed in ATML1-OX and LGO-OX plants. Additionally, ATML1-OX leaves are narrower than those of wild type and LGO-OX (S12A, S12F, S12G, S12I, S12N, and S12O Fig). We also observed that giant cells are more directionally elongated in ATML1-OX than in other genotypes (Figs 4A, 4B, S4F, S4G, S5F, S5G, S6F, S6G, S7G and S7H), reflecting the elongated shape of the leaf. This suggests the existence of a relationship between giant cell shape and leaf morphology. Likewise, wild-type cauline leaves are both narrower and more elongated than wild-type rosette leaves, and also have more anisotropic elongated giant cells than in rosette leaves (S1 Fig). This observation supports the idea that cell shape reflects the anisotropy of the growing tissue [32]. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. Giant cells are more clustered than expected in a randomized null model both in the wild-type leaf and sepal. (A) Scheme summarizing the method used to assess the randomness of the cellular patterns. Each segmentation is computationally randomized using the dmSET method into 400 randomized tissues where cell positions (and orientation in the case of the leaf) have been randomly shuffled (left; see Materials and methods). To statistically assess the extent to which the segmented image shows a random giant cell pattern, a quantitative observable (middle) extracted from the segmentation is compared with the same observable computed in all randomized tissues, forming the estimated ‘null distribution’ (right). (B) Example of a representative segmentation of a wild-type leaf 25 dpg (top left) and a wild-type sepal (bottom left) and one of their randomized tissue (randomization) images (right). (C) Mean number of giant cell neighbors (also referred to as giant neighbors) per giant cell (also referred to as giant) in leaves (top) and sepals (bottom). The value extracted from the segmentations (in red) was statistically tested against all the values extracted from the 400 pooled randomizations (in gray). The mean number of giant cell neighbors per giant cell is higher than expected in a randomized null model, and the null hypothesis can be rejected (p-value <0.05), indicating that giant cells are clustered. (D) Distributions of the number of giant cell neighbors for all giant cells found in all replicates of segmentations (in red) and randomizations (in gray) in leaves (top) and sepals (bottom). Total number of giant cells counted (excluding giant cells at the image border) in the analysis: n = 68 (leaf, segmentations), n = 68 × 400 (leaf, randomizations), n = 74 (sepal, segmentations), n = 74 × 400 (sepal, randomizations). See also S14, S15, S16, S21, and S22 Figs. The code and data associated with this figure can be found at Open Science Framework (osf.io), https://doi.org/10.17605/OSF.IO/RFCWS. https://doi.org/10.1371/journal.pbio.3003469.g005 Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 6. Different cell sizes display different spatial patterns in the wild-type leaf. The method used to assess the randomness of the giant cell patterns (Fig 5) was applied here on different pavement cell-size populations within the mature 25-dpg leaf: (A–C) giant, (D–F) mid-size (around 5,000 µm2), (G–I) small (smallest pavement cells), and (J–L) random (randomly selected pavement cells). The number of cells in each category was determined such that the total cell area of the cell population was approximately equal to the area occupied by the giant cells. (A, D, G, J) Example of representative segmentation of a 25-dpg wild-type leaf (left) and one of its corresponding randomized tissues (right), where cell locations have been computationally shuffled. Cells colored in magenta represent the cells within the studied pavement cell-size population. (B, E, H, K) Mean number of cell neighbors per cell within the same-size population. (B) The mean number of giant cell neighbors per giant cell is higher than expected by chance (p < 0.05), indicating that giant cells are clustered. Same data as in Fig 5C, top. (E) Middle-size cells are less clustered than giant cells and more randomly organized (the null hypothesis cannot be rejected, p = 0.195). (H) The mean number of small cell neighbors per small cell is significantly higher than in the randomized tissues (p < 0.05), highlighting that small cells form clusters. (K) As expected, the randomly selected pavement cells (with area > 2,000 µm2) show a value that falls right in the center of the null distribution (p = 0.445). (C, F, I, L) Distributions of the number of cell neighbors belonging to the studied cell population per cell of that population in the segmentations (in red) and the randomizations (in gray). All six replicates were pooled together. Total number of cells in cell populations counted in the analysis: n = 68 (giant cells, segmentations), n = 68 × 400 (giant cells, randomizations), n = 199 (middle-size cells, segmentations), n = 199 × 400 (middle-size cells, randomizations), n = 639 (small cells, segmentations), n = 639 × 400 (small cells, randomizations), n = 162 (random cells, segmentations), n = 162 × 400 (random cells, randomizations). The code and data associated with this figure can be found at Open Science Framework (osf.io), https://doi.org/10.17605/OSF.IO/RFCWS. https://doi.org/10.1371/journal.pbio.3003469.g006 Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 7. A cell-autonomous stochastic model can recapitulate giant cell clustering because of cell divisions of surrounding cells. (A) Cartoon of the computational model for giant cell patterning. ATML1 activates a target (LGO). If the target is above a certain threshold during the G2 cell-cycle phase (line changing from white to magenta in the cartoon of the time course), it prevents cell division and instead drives entry of the cell into endoreduplication and giant cell formation. (B) Snapshots of the simulated growing sepal, at three different time points. Color codes indicate the cell ploidy levels. Scale bars represent the same size in arbitrary units. (C) A rectangular section of the simulation output (e.g., see rectangle shown in (B)) is used to quantify the giant cell pattern. “Segmentation” refers to one simulation output (left) and “Randomization” to one randomization of the simulated output (right). Giant cells, labeled in magenta, were defined by a size threshold (see Materials and methods). (D) Mean number of giant cell neighbors (also referred to as giant neighbors) per giant cell (also referred to as giant) in the simulations (called segmentation in red) and in their randomizations (in gray). The mean number of giant cell neighbors per giant cell is higher than expected in a randomized null model (p < 0.05), indicating a clustered pattern of giant cells. (E) Distribution of the number of giant cell neighbors per giant cell. The results obtained in (D, E) are comparable with the results in experimental sepal replicates (Fig 5C and 5D). Five simulation outputs with five different initial conditions were performed and combined for the analysis. Total number of giant cells (excluding giant cells at the image border) counted in the analysis: n = 42 (segmentations), n = 42 × 400 (randomizations). (F, G) Statistical assessment of the randomness of the giant cell pattern (comparing the “segmentations” in red with the randomized tissues in gray) at initial time (top) and final time point (bottom) in (F) the simulations (at t = 55 and t = 135) and (G) the real tissues (at stage 4 and stage 9). At the initial time point, the null hypothesis could not be rejected but the mean giant cell neighbors per giant cell became significantly greater than in a randomized null model (p < 0.05) at the final time point. To the right of panel (F), a cartoon represents two neighboring giant cells surrounded by an increasing number of cells as the tissue develops. All five replicates (in simulations) and three replicates (in experimental data) were pooled. Total number of giant cells counted in the analysis: n = 83 (simulations, segmentations), n = 83 × 400 (simulations, randomizations), n = 49 (experimental data, segmentations), n = 49 × 400 (experimental data, randomizations). The dataset in (G) was also used for an independent analysis in Hervieux and colleagues 2016. See randomization snapshots related to this figure in S20 Fig. The code and data associated with this figure can be found at Open Science Framework (osf.io), https://doi.org/10.17605/OSF.IO/RFCWS. https://doi.org/10.1371/journal.pbio.3003469.g007 Spatial patterning of giant cells within the leaf blade In wild-type plants, giant cells vary in position from sepal to sepal and from leaf to leaf [10–12]. An open question has been whether the spatial organization of giant cells is random, or whether there is an underlying order. Classically, many specialized cell types such as stomata and trichomes are spaced such that they are not in direct contact to one another [23,33]. Giant cells are frequently adjacent to each other and, therefore, it is clear that there is not a strong lateral inhibition between them. We set out to determine firstly whether giant cell position is correlated with underlying vasculature and secondly, how giant cells are spatially positioned relative to one another. Giant cells are not preferentially positioned overlying the vasculature We wondered whether giant cell positioning was correlated with the position of leaf vasculature for two reasons. Firstly, we observed that large, highly endoreduplicated cells overlie the midrib of the leaf, extending all the way to the leaf tip (S13A Fig). We wondered whether giant cells might be similarly preferentially located over the other veins. Secondly, we observed that large, highly endoreduplicated cells often appear to “peel” away from the midrib, as if following vascular branches (S13A Fig). This phenomenon is most common in ATML1-OX leaves (S13C–S13F Fig). To investigate whether giant cells overlie veins, we traced the veins from the original confocal image onto the heat map of cell area for a 9-dpg wild-type half leaf and four ATML1-OX half leaves. We found that many giant cells do not overlie the vasculature (S13B–S13F Fig). Specifically, we noted that the points where giant cells peel off the midrib often do not align with where veins extend from the midvein. Furthermore, the orientation of giant cells do not follow the direction of the veins (S13B–S13F Fig). Instead, veins in ATML1-OX plants frequently pass through patches of small cells (S13C–S13F Fig). We conclude that vascular and giant cell patterns are not obviously correlated. Giant cells are clustered more often than expected by chance A cell-autonomous and stochastic mechanism has been proposed to explain giant cell formation in the sepal [12]. However, it remains unknown whether giant cells are randomly arranged within the tissue. To statistically assess the randomness of the pattern, we needed a random reference (or null model) to compare with our experimental replicates. Previous studies addressing this problem considered cells as points [26,34], or used a regular hexagonal grid to build a null model [35]. In our case, these assumptions are not applicable due to the complexity of giant cell shapes and the heterogeneity of cell shapes and sizes that affect cellular arrangements [36]. Therefore, we used the dmSET image-based method [36,37] to generate randomized tissues from the real segmented images (Figs 5A and S14), allowing to randomly shuffle cell positions by preserving cell sizes and shapes of the original tissues (S15 Fig, see Materials and methods). We generated 400 randomized tissues for each biological replicate for both the wild-type sepal and 25-dpg leaf. Several measures (S16 Fig), such as the mean number of giant cell neighbors per giant cell, which captures the amount of contacts between giant cells, were computed in the experimental data and the corresponding randomized tissues. To statistically assess the randomness of the giant cell pattern, these measures in the real biological tissues (segmentation) were compared with the same measures in all the randomized tissues (randomizations), which formed a null distribution (Figs 5A and S16). In the randomized tissues, cell sizes were well preserved, but cell shapes were affected in the leaf (Figs 5A, 5B, S15D, and S15E). To ensure that these shape artifacts did not introduce bias in our analyses, we tested our method on a randomly selected population of cells in both leaves and sepals (see Materials and methods) and confirmed the absence of significant bias in the null models (S17 Fig). Additionally, we reconstructed the original leaf tissues with shape artifacts similar to those in the randomized tissues (S18 Fig), and found that giant cell connectivity was largely preserved and that the results remained consistent (S18 Fig) (see Materials and methods). For both wild-type 25-dpg leaves and mature sepals, when considering the six pooled replicates, the mean number of giant cell neighbors per giant cell was greater than in a randomized null model, and the null hypothesis could be rejected (p < 0.05) (Fig 5C). This result shows the presence of clustering among giant cells both in the leaf and the sepal. It was less probable to find isolated giant cells, and more probable to find giant cells in contact with two or more other giant cells compared with what was expected by chance (Fig 5D). Similar results were found in the leaf using an alternative randomization method we developed based on cutting and merging cells (S19 Fig). The nonrandom pattern of giant cells was also supported by the analysis of other spatial measures (S16 Fig). The similar distribution of the number of giant cell neighbors in leaves and sepals (Fig 5C and 5D) reflects a similar spatial organization, supporting the idea of common patterning mechanisms. Different cell sizes are organized into different spatial patterns To investigate whether the clustered pattern is exclusive to giant cells, we applied the same analysis to distinct sub-populations of pavement cells in the leaf tissues. Four populations of pavement cells were defined: giant cells (Fig 6A–6C), middle-sized cells (Fig 6D–6F), small cells (Fig 6G–6I), and a control population of randomly selected pavement cells of any size (Fig 6J–6L). In contrast to the clustered pattern of giant cells (Fig 6A–6C), middle-sized pavement cells exhibited a more random organization (Fig 6D–6F; the null hypothesis could not be rejected, with p = 0.195). Conversely, small pavement cells showed a clustered organization (Fig 6G–6I), because the mean number of neighbors between small pavement cells significantly exceeded the value observed in the randomized tissues. Notably, these small cells were clustered around the stomata, and their spatial arrangement is probably a consequence of the stomatal patterning process. A random cellular pattern was found in randomly selected pavement cells, as expected (Fig 6J–6L). Overall, these analyses highlight a relationship between pavement cell size and cell spatial organization within the tissue. Furthermore, these findings underscore the distinctive clustered arrangement of giant cells in comparison to middle-sized and randomly selected pavement cells. It was previously shown that, except for cells with low neighbour numbers, pavement cells mostly follow the theoretical topological laws expected from space-filling (i.e. entropic) considerations, with larger cells being on average surrounded by smaller ones in young spch leaf tissues [38]. Our analyses on more mature wild-type leaves reveal that larger cells are surrounded by larger cells (and have fewer neighbors) than what is expected by purely random space‐filling given by our null model (Figs 6 and S18D). A cell-autonomous stochastic model can recapitulate giant cell clustering To investigate how giant cell clustering emerges during leaf and sepal epidermal development, we wondered whether the existing cell-autonomous and stochastic model for giant cell specification in sepals [12] could also recapitulate the clustered feature of the giant cell pattern. In this multicellular computational model, the concentration of ATML1 stochastically fluctuates and is regulated by a self-catalytic feedback loop. ATML1 regulates the expression of a downstream cell-cycle regulator target (Fig 7A). At the end of a cell cycle, a cell either divides or endoreduplicates if the ATML1 target exceeds a specific threshold during the G2 phase. We used this model [12] to investigate the resulting spatial organization of giant cells in simulated tissues (Fig 7A and 7B; see Materials and methods). To assess the randomness of the simulated giant cell pattern, we applied the same method as in the experimental images (Fig 5A) to images of the final simulation time point (Fig 7B and 7C). Giant cells were also defined by a size threshold, which was established such that all cells of ploidy 16C or above were considered to be giant (see Materials and methods). We observed that the mean number of giant cell neighbors per giant cell was greater than expected if giant cells were randomly distributed (p < 0.05, Fig 7D), showing that the current cell-autonomous model can also produce a clustered giant cell pattern. Furthermore, the distribution of the number of giant cell neighbors per giant cell (Fig 7E) was similar to the distribution observed in the experimental sepals (Fig 5D, bottom). This raises the question of what mechanisms are responsible for cell clustering in a cell-autonomous, multicellular model of dividing cells. Cell division contributes to the clustering of giant cells To understand how the giant cell clustering behavior emerges in our computational model, we analyzed how the cellular spatial pattern changes over time within the tissue. We hypothesized that the initial giant cell pattern arises randomly throughout the epidermis, due to the stochastic nature of ATML1 concentration fluctuations that trigger endoreduplication, and occasionally lead to giant cell contacts. As non-giant cells continue to divide, giant cells would appear more clustered in the fully grown tissue. To test this hypothesis in our simulations, we selected the first-arising giant cells (see Materials and methods) and quantified their spatial organization both at an early time point and at the end of the simulation (Figs 7F,S20A, and S20B). We found that these giant cells were more randomly distributed at the initial time point, as the null hypothesis could not be rejected (p = 0.185, Fig 7F), whereas they were more clustered compared with the randomized tissues at the final time point (p < 0.05, Fig 7F). Indeed, although the giant cell contacts were preserved over time (red bar in Fig 7F), we observed a shift in the null distribution of the mean number of giant neighbors per giant cell between the initial and the final time point (Fig 7F). As new cells arise from cell division, the number of potential cellular configurations (i.e., the number of possible spatial cellular arrangements) increases, which decreases the probability of observing giant cell clusters under a random model where all cells have random positions. Therefore, even if giant cell contacts are preserved, their arrangement in the context of the entire tissue becomes more clustered over time. To investigate the emergence of the giant cell spatial pattern over time in real tissues, we used time-lapse data of developing sepals [39], where cells were tracked over time, and we similarly quantified the patterns of the first-arising giant cells at the first available time point (sepal at stage 4, 24-h time point) and a later one (sepal at stage 9, 120-h time point; see Materials and methods) (Figs 7G, S20C, and S20D). Similar to the simulations, we observed that giant cells were more randomly distributed in younger sepals and were more clustered in the more developed sepals when compared with the randomized tissues (Fig 7G). This analysis indicates that the stochastic and cell-autonomous model is a plausible model to explain the spatial organization of giant cells. Moreover, it shows that cell clustering can emerge in a growing tissue without the need for cell–cell communication but instead as a result of cell divisions. Arabidopsis leaves exhibit a large range of pavement cell sizes, similar to sepals In sepals, giant cells are easily visible because they are highly elongated (S1A Fig). Similarly, we observe large and highly anisotropic cells in cauline leaves (S1B Fig). In rosette leaves, pavement cells of the epidermis are jigsaw puzzle-piece shaped with lobes and necks, such that cell size is not readily apparent by eye (S1C Fig); however, heterogeneity in pavement cell sizes has been previously observed [4,28]. Therefore, we wondered to what extent the distribution of cell sizes observed in sepals, ranging from giant cells to small cells, also occurs in rosette leaves. We imaged large sections of the blade (excluding midrib and margin cells) of leaf 1 or 2 from wild-type plants at 25 days postgermination (dpg) when the leaves are fully expanded and mature. Leaves 1 and 2 initiate simultaneously and are indistinguishable; therefore, we refer to them interchangeably as leaf 1 or 2. We measured the area of the epidermal cells of leaves 1 or 2 and sepals on the abaxial (bottom) side (Fig 2A and 2C). Throughout our analysis, we use the term size to mean cell area because area has been shown to be a more relevant measure of cell size than volume in highly vacuolated plant epidermal cells [4]. We observed that the abaxial cell-size distributions for both sepals and leaves are asymmetric, with long tails representing large cells (Fig 2E). Sepal giant cells are larger outliers in size and, consequently, the cell-size distribution in the sepal has a more extended tail than in the leaf (Fig 2E). Still, we observed that the cell-size range in the leaf and sepal are similar and the largest cells of the sepal are about the same size as the largest cells of the leaf (Fig 2A–2E). We conclude that Arabidopsis leaves have a diverse range of cell sizes characterized by a long-tailed distribution, similar to the abaxial side of sepals. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Abaxial and adaxial cell-size distribution in the wild-type leaf and sepal epidermis; size correlates with DNA content. (A–D) Cell area heat maps (in µm2) of (A) abaxial surface of wild-type sepal, (B) adaxial surface of wild-type sepal, (C) abaxial surface of 25-dpg wild-type leaf 1 or 2 (cell density: 234 cells mm−2), and (D) adaxial surface of 25-dpg wild-type leaf 1 or 2 (cell density: 156 cells mm−2). Scale bars represent 100 µm. (E) Violin and strip plots of cell areas of abaxial and adaxial sides of 25-dpg wild-type leaves (two pooled replicates) and adult wild-type sepals (three pooled replicates). (F) Adaxial side (green) and abaxial side (purple) of 25-dpg leaf cell area versus DNA content as measured by H2B-TFP total nuclear fluorescence, with R2 = 0.85 for the abaxial side and R2 = 0.82 for the adaxial side (one of two replicates). See S2 Fig for replicates. The underlying data for this figure can be found at Open Science Framework (osf.io), https://doi.org/10.17605/OSF.IO/RFCWS. https://doi.org/10.1371/journal.pbio.3003469.g002 Large cells are formed on the adaxial side as well as the abaxial side of the leaf In sepals, giant cells are restricted to the abaxial (outer) surface (Figs 2A, 2B, and S2A–S2D). We asked whether there was a difference in cell size between adaxial (top) and abaxial (bottom) surfaces of the leaf. Large cells of similar size are formed on both the adaxial and abaxial surfaces of the leaf, in contrast to the sepal (Figs 2A–2E and S2A–S2F). In addition, across the whole cell-size distribution, the cell density is lower and many cells are slightly more expanded on the adaxial side than the abaxial side (Figs 2C, 2D, S2E, and S2F). There are a greater number of stomata and stomatal lineage cells (very small cells) on the abaxial side compared with the adaxial side (Figs 2C–2E, S2E, and S2F). We also observed that the abaxial cells are more lobed than the adaxial cells (Figs 2C, 2D, S2E, and S2F). Despite slight differences, the cell-size distributions of the abaxial and adaxial sides of the leaf are quite similar, particularly in the tails, where both sides exhibit a similar range of larger cells, in contrast to the sepal, where only the abaxial side has very large cells. Cell area correlates with DNA content Cell area and ploidy are positively correlated in leaf epidermal cells [4]. A recent study showed that fluorescence levels of histone fluorescence reporters measured from microscopy images are a good proxy to infer nuclear ploidy in Arabidopsis cotyledons [29]. Hence, to validate whether cell area and ploidy were correlated in our leaves, we measured DNA content by quantifying total fluorescence of Histone 2B-TFP (pUBQ10::H2B-TFP) within each cell nucleus of the 25-dpg leaf images. For this reporter, the H2B-TFP signal distribution was noisy and continuous, not divided into four discrete ploidy peaks, providing an approximation of ploidy, not exact DNA content. As expected, a strong linear correlation between H2B-TFP fluorescence and cell area was observed for both the abaxial pavement cells (R2 = 0.85 and 0.91; n = 2) and the adaxial pavement cells (R2 = 0.79 and 0.82; n = 2) (Figs 2F and S2G). Therefore, we focus on analyzing cell size, and infer that large cell size indicates high ploidy. We wondered whether cells of similar size on the abaxial and adaxial side of the same leaf also have a similar DNA content. We found that cells of similar DNA content are larger on the adaxial side than on the abaxial side (Figs 2F and S2G), suggesting that adaxial cells have expanded more than abaxial cells, as noted above. Because the largest sepal cells and the largest leaf cells had approximately the same areas, we asked whether the DNA content of these cells was also similar. We found that the total H2B-TFP fluorescence values of the largest cells were very similar between sepal and leaf, suggesting that these largest cells are similar in ploidy (S2H Fig). Cell-size patterning emerges at the tip and progresses basipetally as the leaf differentiates To determine how the cell-size pattern emerges in the leaf during development, we imaged both the adaxial and abaxial surfaces of each leaf at different stages of development. From 5 to 9 dpg, cell size increases greatly (S3 Fig), as expected. At day 5, cells throughout the blade are fairly homogeneous in size, with a few cells starting to expand near the distal tip, and the large cells of the margin and overlying midrib already apparent (Fig 3A). We focus on pavement cells in the blade and exclude margin and midrib cells from further analysis. The cell-size pattern consisting of large cells interspersed between small cells progressively develops basipetally from the tip (Fig 3A–3C), whereas at the base the cells remain uniformly small. The progression of cell-size patterning down the leaf is consistent with the well-established basipetal wavefront of differentiation and cessation of cell division [30]. The cell area distributions showed that more large cells appear throughout development and the maximal cell size increases (Fig 3A, 3B, and 3D). By 9 dpg, cell size has been patterned almost to the base of the leaf (Fig 3A and 3B). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Cell-size patterning occurs as a basipetal wave simultaneously on the adaxial and abaxial sides of the leaf. (A–B) Cell area heat maps (in µm2) of leaf 1 or 2 at different stages of development on (A) the abaxial side and on (B) the adaxial side of the same leaf at 5, 6, 7, 8, and 9 dpg (half leaf). Unsegmented regions on adaxial leaves correspond to trichomes, which were not considered in this analysis. Each stage is associated with a distinct heat map color range. Scale bars represent 50 µm at 5 and 6 dpg, and 100 µm at 7, 8, and 9 dpg. (C) Spatial positions of the largest cells (those above an area threshold, see below) on the abaxial (purple points) and adaxial (green points) sides of the same leaf at 6, 7, 8, and 9 dpg. Area thresholds for each leaf were determined from the 98th percentile cell area of the abaxial side. (Note that the midrib and margin cells are included in these overlays of large cell positions.) (D) Violin plots of cell areas (in µm2) of the largest cells (as defined in (C) and excluding margin and midrib cells) on abaxial and adaxial sides of leaves at different developmental stages. Abaxial and adaxial sides are from the same leaf. See also S3 Fig for the leaves shown to scale. The underlying data for this figure can be found at Open Science Framework (osf.io), https://doi.org/10.17605/OSF.IO/RFCWS. https://doi.org/10.1371/journal.pbio.3003469.g003 We next asked whether the wavefront of cell-size patterning progresses basipetally at the same rate on the abaxial and adaxial sides of the leaf. Using images of both the abaxial and adaxial sides of the same leaf, we plotted the positions of the centers of the largest cells on both sides (including margin and midrib cells for landmarks) (Fig 3C). Large pavement cells are observed in the same proximal–distal region on abaxial and adaxial sides during development (Fig 3C). The region expands in the proximal direction as development progresses. This suggests that the wavefront of patterning and differentiation is coordinated across the abaxial/adaxial axis of the leaf. The sepal giant cell specification pathway also patterns giant cells in leaves Because the cell-size distributions have similarities in leaves and sepals, we tested whether the giant cell specification pathway in sepals (Fig 1H) also functions in the leaf to pattern cell size. We imaged leaf 1 or 2 at both 9 and 25 dpg from wild-type and giant cell pathway mutants. At 9 dpg, patterning has just extended to the base of the leaf, and the leaf is still small enough that we could image the whole upper abaxial quadrant to determine the pattern over a large fraction of the leaf blade (Figs 4A, S4, and S5). At 25 dpg, the leaf is fully differentiated, fully expanded, and the pattern is established (Figs 4B, S6, and S7). We found that cell-size patterning in the leaf is similarly affected in the mutants at both 9 and 25 dpg as in the mature sepal (Figs 4, S8A, and S8B). Notably, the largest cells show similar variations in their numbers across genotypes. Similar to the sepal, the size of the largest cells is moderately reduced in acr4-2 mutants (Figs 1B and 4A–4D), and more greatly reduced in atml1-3 mutants (Figs 1C and 4A–4D). The reduction in large cells is drastic in dek1-4 and lgo-2 mutant sepals and leaves, resulting in the absence of a long tail in the cell-size distribution (Figs 1D–1E and 4A–4D). For these genotypes, the number of medium-sized cells is also substantially decreased (Figs 1D, 1E, 4C, and 4D). Conversely, the overexpression of ATML1 (ATML1-OX) or LGO (LGO-OX) leads to an increase in the size of large cells and in fewer small cells compared with wild type, as in the sepal (Figs 1F, 1G, and 4A–4D). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. The sepal giant cell specification pathway also patterns cell size in leaves. (A, B) Cell area heat maps (in µm2) of the upper abaxial quadrant of leaf 1 or 2 at 9 dpg (A) and of a section on the abaxial side of leaf 1 or 2 at 25 dpg (B) for the following genotypes: wild type, acr4-2, atml1-3, dek1-4, lgo-2, LGO-OX (pATML1::LGO), and ATML1-OX (pPDF1::FLAG-ATML1). Scale bars represent 100 µm. Cell area heat maps of other replicates are shown in S4–S7 Figs. (C, D) Violin plots of cell area distributions on a log10 scale for 9-dpg replicates (C) and for 25-dpg replicates (D). Stomata were classified and removed in (C, D). Violin plots of individual replicates are shown in S8A and S8B Fig. (E) Wasserstein distance plot of normalized cell-size distributions (see Materials and methods) for all 9- and 25-dpg replicates displayed as Euclidean distances embedded in 2D. The 25-dpg replicates are indicated by circular dots and 9-dpg replicates by triangular dots. The Wasserstein distance plot for 9- and 25-dpg and Wasserstein statistical tests among replicates are shown in S4, S6 and S8 Figs. Associated with S4–S8 Figs. The underlying data for this figure can be found at Open Science Framework (osf.io), https://doi.org/10.17605/OSF.IO/RFCWS. https://doi.org/10.1371/journal.pbio.3003469.g004 To quantify the variations in the number of large cells precisely, we quantitatively defined leaf giant cells on the basis of a cell area threshold. Specifically, we first classified pavement cells and stomata using a Support Vector Machine classifier based on features of cell shape (see Materials and methods, S9 Fig). Next, a cell-size threshold was established in the mature sepal and in the leaf, at both 9 and 25 dpg, using the atml1-3 mutants, which are known to have very few giant cells in sepals (see Materials and methods, Figs 1C and S9). Those cells in the 9- and 25-dpg leaves as well as in the sepal that exceeded their associated threshold were categorized as giant cells (see cell-type classification outcomes in S10 and S11 Figs). On the basis of this definition, we performed a quantitative comparison and statistically compared the number of giant cells per unit area among genotypes in leaves. Two-sample, two-tailed t tests showed that in the 9-dpg leaf and the mature leaf, wild type had significantly more giant cells than lgo-2 (9 dpg: p = 0.002, 25 dpg: p = 0.003), dek1-4 (9 dpg: p = 0.002, 25 dpg: p = 0.002), atml1-3 (9 dpg: p = 0.002, 25 dpg: p = 0.005), and acr4-2 (9 dpg: p = 0.010, 25 dpg: p = 0.044). Conversely, LGO-OX had significantly more giant cells than wild type (9 dpg: p = 0.001, 25 dpg: p = 0.003). However, no statistically significant difference in the number of giant cells per unit area was observed between wild type and ATML1-OX (9 dpg: p = 0.213, 25 dpg: p = 0.75). Because the giant cells in ATML1-OX are so much bigger than wild-type giant cells, each of the giant cells in a given area of ATML1-OX leaf takes up a large amount of space, resulting in few giant cells per unit area despite the fact that most of the unit area is occupied by giant cells. We sought to quantify what was apparent visually by comparing the fractional area occupied by giant cells between ATML1-OX and wild type and found that the fractional area occupied by giant cells was significantly higher in ATML1-OX (9 dpg: p < 0.005, 25 dpg: p < 0.005). Thus, in ATML1-OX the number of giant cells is not changed, but the fractional area covered by giant cells is increased. Collectively, the similarities in the variation between the number of giant cells in the leaf and the sepal indicates that the sepal giant cell specification pathway also regulates the formation of giant cells in leaves. Giant cell mutants affect the entire cell-size distribution We observed that not only are giant cells affected in these mutants, but other aspects of the cell-size distribution are also affected. For example, the number of medium-sized cells in lgo-2 and dek1-4 is reduced in addition to the number of giant cells (Fig 4A–4D) and, correspondingly, the number of small cells is increased in these mutants. To statistically analyze the difference in cell-size distributions, we conducted a principal coordinate analysis based on the Wasserstein distances between cell-size distributions (termed Wasserstein distance plot), which showed the difference between leaf samples according to their cell-size distributions on a 2-dimensional plane (S4H, S6H, and S8C–S8F Figs, see Materials and methods). Samples clustered according to genotype, indicating that genotype controls the cell-size distribution. We observed a progressive increase in the number of giant cells along the first principal coordinate V1 from lgo-2 mutants to ATML1-OX and LGO-OX (S4H and S6H Figs). ATML1-OX and LGO-OX were distant from each other in this plot, which might partly reflect the fact that LGO-OX has more giant cells, whereas ATML1-OX has fewer but larger giant cells. When we created the combined Wasserstein distance plot using normalized cell-size distributions from both 9- and 25-dpg leaves (see Materials and methods), the samples continued to group according to genotype rather than developmental stage, further supporting the idea that these genes have affected the cell-size distribution by 9 dpg (Fig 4E). Thus, we conclude that these genes affect the entire cell-size distribution. However, some differences in the cell-size distribution are apparent between 9-dpg and mature 25-dpg leaves. Firstly, at 9 dpg, dek1-4, and lgo-2 mutants are very similar; however, in the fully mature 25-dpg leaves, the lgo-2 cell-size range is notably smaller than that in the dek1-4 mutant (Fig 4A–4D), suggesting that lgo-2 cells continue to divide after 9 dpg. In addition, the small cells in lgo-2 mutants were more uniform in size than all of the other genotypes because the typical small stomatal lineage cells that encircle the stomata in mature leaves were fewer in lgo-2 (Fig 4A–4D). This altered cell-size distribution relates to the previous finding that LGO affects pavement cell differentiation in these stomatal lineage ground cells and that cells undergo division for a longer time in the absence of LGO [19]. Secondly, although at 9 dpg the LGO-OX giant cells were slightly smaller than the ATML1-OX giant cells, at 25 dpg, the LGO-OX giant cells were nearly equivalent in size to ATML1-OX giant cells (Fig 4A–4D). In addition, we observed that more pavement cells were larger in LGO-OX, whereas only a few cells became giant in ATML1-OX (Fig 4A–4D). ATML1-OX leaves had a few connected giant cells separating large islands of small cells, whereas LGO-OX leaves showed more giant cells interspersed among smaller clusters of small cells (Fig 4A and 4B and S10 and S11) ). These phenotypic differences might reflect either inherent differences in ATML1 and LGO activities or the fact that ATML1 and LGO overexpression transgenes are under the control of different promoters that might have differences in activity at different developmental stages. Relationship between the size and shape of cells and organs In plants, compensation is the process by which organ size is maintained when cell number is altered by an accompanying change in cell size [31]. We observed compensation in the leaf giant cell mutants (S12 Fig). Mature leaves of the mutants acr4-2, atml1-3, dek1-4, and lgo-2 are similar in size to wild-type leaves. Having fewer giant cells is compensated by having more small pavement cells (Fig 4). However, ATML1-OX and LGO-OX mature leaves, which have much larger cells (see, e.g., Fig 4B and 4D), are smaller than wild type (S12N–S12P Fig). Therefore, only partial compensation for having fewer cells by having larger cells is observed in ATML1-OX and LGO-OX plants. Additionally, ATML1-OX leaves are narrower than those of wild type and LGO-OX (S12A, S12F, S12G, S12I, S12N, and S12O Fig). We also observed that giant cells are more directionally elongated in ATML1-OX than in other genotypes (Figs 4A, 4B, S4F, S4G, S5F, S5G, S6F, S6G, S7G and S7H), reflecting the elongated shape of the leaf. This suggests the existence of a relationship between giant cell shape and leaf morphology. Likewise, wild-type cauline leaves are both narrower and more elongated than wild-type rosette leaves, and also have more anisotropic elongated giant cells than in rosette leaves (S1 Fig). This observation supports the idea that cell shape reflects the anisotropy of the growing tissue [32]. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. Giant cells are more clustered than expected in a randomized null model both in the wild-type leaf and sepal. (A) Scheme summarizing the method used to assess the randomness of the cellular patterns. Each segmentation is computationally randomized using the dmSET method into 400 randomized tissues where cell positions (and orientation in the case of the leaf) have been randomly shuffled (left; see Materials and methods). To statistically assess the extent to which the segmented image shows a random giant cell pattern, a quantitative observable (middle) extracted from the segmentation is compared with the same observable computed in all randomized tissues, forming the estimated ‘null distribution’ (right). (B) Example of a representative segmentation of a wild-type leaf 25 dpg (top left) and a wild-type sepal (bottom left) and one of their randomized tissue (randomization) images (right). (C) Mean number of giant cell neighbors (also referred to as giant neighbors) per giant cell (also referred to as giant) in leaves (top) and sepals (bottom). The value extracted from the segmentations (in red) was statistically tested against all the values extracted from the 400 pooled randomizations (in gray). The mean number of giant cell neighbors per giant cell is higher than expected in a randomized null model, and the null hypothesis can be rejected (p-value <0.05), indicating that giant cells are clustered. (D) Distributions of the number of giant cell neighbors for all giant cells found in all replicates of segmentations (in red) and randomizations (in gray) in leaves (top) and sepals (bottom). Total number of giant cells counted (excluding giant cells at the image border) in the analysis: n = 68 (leaf, segmentations), n = 68 × 400 (leaf, randomizations), n = 74 (sepal, segmentations), n = 74 × 400 (sepal, randomizations). See also S14, S15, S16, S21, and S22 Figs. The code and data associated with this figure can be found at Open Science Framework (osf.io), https://doi.org/10.17605/OSF.IO/RFCWS. https://doi.org/10.1371/journal.pbio.3003469.g005 Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 6. Different cell sizes display different spatial patterns in the wild-type leaf. The method used to assess the randomness of the giant cell patterns (Fig 5) was applied here on different pavement cell-size populations within the mature 25-dpg leaf: (A–C) giant, (D–F) mid-size (around 5,000 µm2), (G–I) small (smallest pavement cells), and (J–L) random (randomly selected pavement cells). The number of cells in each category was determined such that the total cell area of the cell population was approximately equal to the area occupied by the giant cells. (A, D, G, J) Example of representative segmentation of a 25-dpg wild-type leaf (left) and one of its corresponding randomized tissues (right), where cell locations have been computationally shuffled. Cells colored in magenta represent the cells within the studied pavement cell-size population. (B, E, H, K) Mean number of cell neighbors per cell within the same-size population. (B) The mean number of giant cell neighbors per giant cell is higher than expected by chance (p < 0.05), indicating that giant cells are clustered. Same data as in Fig 5C, top. (E) Middle-size cells are less clustered than giant cells and more randomly organized (the null hypothesis cannot be rejected, p = 0.195). (H) The mean number of small cell neighbors per small cell is significantly higher than in the randomized tissues (p < 0.05), highlighting that small cells form clusters. (K) As expected, the randomly selected pavement cells (with area > 2,000 µm2) show a value that falls right in the center of the null distribution (p = 0.445). (C, F, I, L) Distributions of the number of cell neighbors belonging to the studied cell population per cell of that population in the segmentations (in red) and the randomizations (in gray). All six replicates were pooled together. Total number of cells in cell populations counted in the analysis: n = 68 (giant cells, segmentations), n = 68 × 400 (giant cells, randomizations), n = 199 (middle-size cells, segmentations), n = 199 × 400 (middle-size cells, randomizations), n = 639 (small cells, segmentations), n = 639 × 400 (small cells, randomizations), n = 162 (random cells, segmentations), n = 162 × 400 (random cells, randomizations). The code and data associated with this figure can be found at Open Science Framework (osf.io), https://doi.org/10.17605/OSF.IO/RFCWS. https://doi.org/10.1371/journal.pbio.3003469.g006 Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 7. A cell-autonomous stochastic model can recapitulate giant cell clustering because of cell divisions of surrounding cells. (A) Cartoon of the computational model for giant cell patterning. ATML1 activates a target (LGO). If the target is above a certain threshold during the G2 cell-cycle phase (line changing from white to magenta in the cartoon of the time course), it prevents cell division and instead drives entry of the cell into endoreduplication and giant cell formation. (B) Snapshots of the simulated growing sepal, at three different time points. Color codes indicate the cell ploidy levels. Scale bars represent the same size in arbitrary units. (C) A rectangular section of the simulation output (e.g., see rectangle shown in (B)) is used to quantify the giant cell pattern. “Segmentation” refers to one simulation output (left) and “Randomization” to one randomization of the simulated output (right). Giant cells, labeled in magenta, were defined by a size threshold (see Materials and methods). (D) Mean number of giant cell neighbors (also referred to as giant neighbors) per giant cell (also referred to as giant) in the simulations (called segmentation in red) and in their randomizations (in gray). The mean number of giant cell neighbors per giant cell is higher than expected in a randomized null model (p < 0.05), indicating a clustered pattern of giant cells. (E) Distribution of the number of giant cell neighbors per giant cell. The results obtained in (D, E) are comparable with the results in experimental sepal replicates (Fig 5C and 5D). Five simulation outputs with five different initial conditions were performed and combined for the analysis. Total number of giant cells (excluding giant cells at the image border) counted in the analysis: n = 42 (segmentations), n = 42 × 400 (randomizations). (F, G) Statistical assessment of the randomness of the giant cell pattern (comparing the “segmentations” in red with the randomized tissues in gray) at initial time (top) and final time point (bottom) in (F) the simulations (at t = 55 and t = 135) and (G) the real tissues (at stage 4 and stage 9). At the initial time point, the null hypothesis could not be rejected but the mean giant cell neighbors per giant cell became significantly greater than in a randomized null model (p < 0.05) at the final time point. To the right of panel (F), a cartoon represents two neighboring giant cells surrounded by an increasing number of cells as the tissue develops. All five replicates (in simulations) and three replicates (in experimental data) were pooled. Total number of giant cells counted in the analysis: n = 83 (simulations, segmentations), n = 83 × 400 (simulations, randomizations), n = 49 (experimental data, segmentations), n = 49 × 400 (experimental data, randomizations). The dataset in (G) was also used for an independent analysis in Hervieux and colleagues 2016. See randomization snapshots related to this figure in S20 Fig. The code and data associated with this figure can be found at Open Science Framework (osf.io), https://doi.org/10.17605/OSF.IO/RFCWS. https://doi.org/10.1371/journal.pbio.3003469.g007 Spatial patterning of giant cells within the leaf blade In wild-type plants, giant cells vary in position from sepal to sepal and from leaf to leaf [10–12]. An open question has been whether the spatial organization of giant cells is random, or whether there is an underlying order. Classically, many specialized cell types such as stomata and trichomes are spaced such that they are not in direct contact to one another [23,33]. Giant cells are frequently adjacent to each other and, therefore, it is clear that there is not a strong lateral inhibition between them. We set out to determine firstly whether giant cell position is correlated with underlying vasculature and secondly, how giant cells are spatially positioned relative to one another. Giant cells are not preferentially positioned overlying the vasculature We wondered whether giant cell positioning was correlated with the position of leaf vasculature for two reasons. Firstly, we observed that large, highly endoreduplicated cells overlie the midrib of the leaf, extending all the way to the leaf tip (S13A Fig). We wondered whether giant cells might be similarly preferentially located over the other veins. Secondly, we observed that large, highly endoreduplicated cells often appear to “peel” away from the midrib, as if following vascular branches (S13A Fig). This phenomenon is most common in ATML1-OX leaves (S13C–S13F Fig). To investigate whether giant cells overlie veins, we traced the veins from the original confocal image onto the heat map of cell area for a 9-dpg wild-type half leaf and four ATML1-OX half leaves. We found that many giant cells do not overlie the vasculature (S13B–S13F Fig). Specifically, we noted that the points where giant cells peel off the midrib often do not align with where veins extend from the midvein. Furthermore, the orientation of giant cells do not follow the direction of the veins (S13B–S13F Fig). Instead, veins in ATML1-OX plants frequently pass through patches of small cells (S13C–S13F Fig). We conclude that vascular and giant cell patterns are not obviously correlated. Giant cells are clustered more often than expected by chance A cell-autonomous and stochastic mechanism has been proposed to explain giant cell formation in the sepal [12]. However, it remains unknown whether giant cells are randomly arranged within the tissue. To statistically assess the randomness of the pattern, we needed a random reference (or null model) to compare with our experimental replicates. Previous studies addressing this problem considered cells as points [26,34], or used a regular hexagonal grid to build a null model [35]. In our case, these assumptions are not applicable due to the complexity of giant cell shapes and the heterogeneity of cell shapes and sizes that affect cellular arrangements [36]. Therefore, we used the dmSET image-based method [36,37] to generate randomized tissues from the real segmented images (Figs 5A and S14), allowing to randomly shuffle cell positions by preserving cell sizes and shapes of the original tissues (S15 Fig, see Materials and methods). We generated 400 randomized tissues for each biological replicate for both the wild-type sepal and 25-dpg leaf. Several measures (S16 Fig), such as the mean number of giant cell neighbors per giant cell, which captures the amount of contacts between giant cells, were computed in the experimental data and the corresponding randomized tissues. To statistically assess the randomness of the giant cell pattern, these measures in the real biological tissues (segmentation) were compared with the same measures in all the randomized tissues (randomizations), which formed a null distribution (Figs 5A and S16). In the randomized tissues, cell sizes were well preserved, but cell shapes were affected in the leaf (Figs 5A, 5B, S15D, and S15E). To ensure that these shape artifacts did not introduce bias in our analyses, we tested our method on a randomly selected population of cells in both leaves and sepals (see Materials and methods) and confirmed the absence of significant bias in the null models (S17 Fig). Additionally, we reconstructed the original leaf tissues with shape artifacts similar to those in the randomized tissues (S18 Fig), and found that giant cell connectivity was largely preserved and that the results remained consistent (S18 Fig) (see Materials and methods). For both wild-type 25-dpg leaves and mature sepals, when considering the six pooled replicates, the mean number of giant cell neighbors per giant cell was greater than in a randomized null model, and the null hypothesis could be rejected (p < 0.05) (Fig 5C). This result shows the presence of clustering among giant cells both in the leaf and the sepal. It was less probable to find isolated giant cells, and more probable to find giant cells in contact with two or more other giant cells compared with what was expected by chance (Fig 5D). Similar results were found in the leaf using an alternative randomization method we developed based on cutting and merging cells (S19 Fig). The nonrandom pattern of giant cells was also supported by the analysis of other spatial measures (S16 Fig). The similar distribution of the number of giant cell neighbors in leaves and sepals (Fig 5C and 5D) reflects a similar spatial organization, supporting the idea of common patterning mechanisms. Different cell sizes are organized into different spatial patterns To investigate whether the clustered pattern is exclusive to giant cells, we applied the same analysis to distinct sub-populations of pavement cells in the leaf tissues. Four populations of pavement cells were defined: giant cells (Fig 6A–6C), middle-sized cells (Fig 6D–6F), small cells (Fig 6G–6I), and a control population of randomly selected pavement cells of any size (Fig 6J–6L). In contrast to the clustered pattern of giant cells (Fig 6A–6C), middle-sized pavement cells exhibited a more random organization (Fig 6D–6F; the null hypothesis could not be rejected, with p = 0.195). Conversely, small pavement cells showed a clustered organization (Fig 6G–6I), because the mean number of neighbors between small pavement cells significantly exceeded the value observed in the randomized tissues. Notably, these small cells were clustered around the stomata, and their spatial arrangement is probably a consequence of the stomatal patterning process. A random cellular pattern was found in randomly selected pavement cells, as expected (Fig 6J–6L). Overall, these analyses highlight a relationship between pavement cell size and cell spatial organization within the tissue. Furthermore, these findings underscore the distinctive clustered arrangement of giant cells in comparison to middle-sized and randomly selected pavement cells. It was previously shown that, except for cells with low neighbour numbers, pavement cells mostly follow the theoretical topological laws expected from space-filling (i.e. entropic) considerations, with larger cells being on average surrounded by smaller ones in young spch leaf tissues [38]. Our analyses on more mature wild-type leaves reveal that larger cells are surrounded by larger cells (and have fewer neighbors) than what is expected by purely random space‐filling given by our null model (Figs 6 and S18D). A cell-autonomous stochastic model can recapitulate giant cell clustering To investigate how giant cell clustering emerges during leaf and sepal epidermal development, we wondered whether the existing cell-autonomous and stochastic model for giant cell specification in sepals [12] could also recapitulate the clustered feature of the giant cell pattern. In this multicellular computational model, the concentration of ATML1 stochastically fluctuates and is regulated by a self-catalytic feedback loop. ATML1 regulates the expression of a downstream cell-cycle regulator target (Fig 7A). At the end of a cell cycle, a cell either divides or endoreduplicates if the ATML1 target exceeds a specific threshold during the G2 phase. We used this model [12] to investigate the resulting spatial organization of giant cells in simulated tissues (Fig 7A and 7B; see Materials and methods). To assess the randomness of the simulated giant cell pattern, we applied the same method as in the experimental images (Fig 5A) to images of the final simulation time point (Fig 7B and 7C). Giant cells were also defined by a size threshold, which was established such that all cells of ploidy 16C or above were considered to be giant (see Materials and methods). We observed that the mean number of giant cell neighbors per giant cell was greater than expected if giant cells were randomly distributed (p < 0.05, Fig 7D), showing that the current cell-autonomous model can also produce a clustered giant cell pattern. Furthermore, the distribution of the number of giant cell neighbors per giant cell (Fig 7E) was similar to the distribution observed in the experimental sepals (Fig 5D, bottom). This raises the question of what mechanisms are responsible for cell clustering in a cell-autonomous, multicellular model of dividing cells. Cell division contributes to the clustering of giant cells To understand how the giant cell clustering behavior emerges in our computational model, we analyzed how the cellular spatial pattern changes over time within the tissue. We hypothesized that the initial giant cell pattern arises randomly throughout the epidermis, due to the stochastic nature of ATML1 concentration fluctuations that trigger endoreduplication, and occasionally lead to giant cell contacts. As non-giant cells continue to divide, giant cells would appear more clustered in the fully grown tissue. To test this hypothesis in our simulations, we selected the first-arising giant cells (see Materials and methods) and quantified their spatial organization both at an early time point and at the end of the simulation (Figs 7F,S20A, and S20B). We found that these giant cells were more randomly distributed at the initial time point, as the null hypothesis could not be rejected (p = 0.185, Fig 7F), whereas they were more clustered compared with the randomized tissues at the final time point (p < 0.05, Fig 7F). Indeed, although the giant cell contacts were preserved over time (red bar in Fig 7F), we observed a shift in the null distribution of the mean number of giant neighbors per giant cell between the initial and the final time point (Fig 7F). As new cells arise from cell division, the number of potential cellular configurations (i.e., the number of possible spatial cellular arrangements) increases, which decreases the probability of observing giant cell clusters under a random model where all cells have random positions. Therefore, even if giant cell contacts are preserved, their arrangement in the context of the entire tissue becomes more clustered over time. To investigate the emergence of the giant cell spatial pattern over time in real tissues, we used time-lapse data of developing sepals [39], where cells were tracked over time, and we similarly quantified the patterns of the first-arising giant cells at the first available time point (sepal at stage 4, 24-h time point) and a later one (sepal at stage 9, 120-h time point; see Materials and methods) (Figs 7G, S20C, and S20D). Similar to the simulations, we observed that giant cells were more randomly distributed in younger sepals and were more clustered in the more developed sepals when compared with the randomized tissues (Fig 7G). This analysis indicates that the stochastic and cell-autonomous model is a plausible model to explain the spatial organization of giant cells. Moreover, it shows that cell clustering can emerge in a growing tissue without the need for cell–cell communication but instead as a result of cell divisions. Discussion We investigated pavement cell-size patterning in the Arabidopsis leaf epidermis. We found that the same genetic pathway that controls giant cell formation in sepals also controls cell size and giant cell formation in the leaf. Specifically, the receptor-like kinase ACR4, the transcription factor ATML1, the calpain protease DEK1, and the CDK inhibitor LGO are important for the formation of leaf giant cells. Just as in the sepal, overexpression of LGO results in an increased number of giant cells and overexpression of ATML1 leads to a larger area occupied by giant cells. Although giant cells are only present on the abaxial epidermis of sepals, they are present on both the abaxial and adaxial surfaces of leaves. We observed that giant cells are scattered across the surface, sometimes in contact with one another, in both leaves and sepals. Our analysis demonstrated that giant cells are more likely to be in contact than in a randomized tissue null model in both organs. Many patterning systems rely on cell–cell communication to generate proper spacing [7,40], and the emergence of clustered patterns in certain cell types is often attributed to cell–cell communication mechanisms in static tissues [36,41,42]. However, giant cell specification occurs within the context of tissue growth and cell division. Therefore, it is important to consider the influence of these dynamic factors as well. We revisited our previous cell-autonomous model for giant cell specification in which ATML1 stochastically fluctuates, and confirmed that giant cell clustering could arise in that model as a result of cell division, without the need for cell–cell communication. To understand how clustering emerges, we tracked giant cells from their initial emergence both in our modeled tissue and in published experimental time-lapse data of growing sepals [39], and analyzed the evolution of their spatial pattern over time. This analysis suggested that the giant cell pattern initially arises randomly in space in the primordium and becomes more clustered in the context of the fully grown tissue. Therefore, the decrease in the randomness of the giant cell pattern over time appears to be caused by the division of cells surrounding the giant cells, including dividing stomata lineage cells. Rather than resulting from active cell–cell communication, giant cell clustering reflects the history of cell division and tissue growth. Nevertheless, in a proliferating tissue, other mechanisms might operate at the same time that result in giant cell clustering. For instance, correlative effects on cells belonging to the same lineage might influence cell fate decisions, e.g., due to the inheritance of molecular factors from mother to daughter cells [43]. Although we show that giant cell clustering can occur as a result of surrounding cell divisions, whether or not cell–cell communication or other patterning mechanisms may also exert some effect on giant cell clustering is yet to be determined and warrants further study. In the past, researchers have attempted to increase organ size by increasing cell size by promoting endoreduplication, but these efforts have not been successful [44]. This is because compensation occurs, in which smaller cell size is accompanied by an increase in cell number, so that organ size is relatively conserved [45,46]. Consistent with this, we observed that leaf 1 or 2 of wild type, atml1-3, and lgo-2 plants are approximately the same size at maturity. Furthermore, instead of having larger leaves, the ATML1-OX and LGO-OX genotypes that have larger cells actually have slightly smaller leaves than the wild type at maturity. These observations are consistent with what is observed for sepals where ATML1-OX and LGO-OX sepals are slightly smaller than wild-type sepals [10,45]. We have previously shown that mitotic division substitutes for endoreduplication to compensate and maintain organ size in mutants lacking giant cells [10]. Our images suggest that this mechanism also operates in leaves. Although giant cell number does not greatly influence organ size, organ shape is altered in sepals and leaves. ATML1-OX and LGO-OX sepals are narrower than those of wild type and curve outward, so that the bud opens prematurely [11]. We speculate that the anisotropy of sepal giant cells drives the change in sepal shape. In ATML1-OX leaves, where giant cells are highly anisotropic, we observe a similar change in leaf shape, in which ATML1-OX leaf 1 and 2 are more pointed and oblong compared with the rounded wild-type leaf 1 and 2. By contrast, giant cells in LGO-OX leaves are more isotropic and similar to wild-type giant cells; correspondingly LGO-OX leaves are more rounded. Our results suggest that ATML1 is sufficient to induce anisotropic cell growth, whereas LGO is not. Tang and colleagues (2023) have shown that the change in shape between rounded juvenile rosette leaves and more elongated adult rosette leaves is accompanied by the appearance of highly anisotropic giant cells at the leaf base [32]. However, they showed that loss of these directional, elongated giant cells does not change adult rosette leaf shape; the adult rosette leaf 7 remains elongated in lgo-2 leaves where giant cells are not present [32]. Thus, the relationship between giant cell shape, anisotropic growth, and organ shape is complex. Further work at the single-cell level will be needed to elucidate the influence of giant cells on the shapes of different tissues. Despite the similarities between cell-size patterning in leaves and sepals, subtle differences also exist. Firstly, the distribution of epidermal cell sizes in the leaf is broader than in the sepal, where cells are fairly uniformly small except for a scattering of giant cells (Fig 2E). Secondly, leaves have giant cells on both abaxial and adaxial blades, whereas sepals have giant cells only on the abaxial side and not on the adaxial side that faces the petals. The petal blade does not have giant cells on either abaxial or adaxial sides [47]; thus, sepals might be an organ whose identity is transitional between vegetative and floral organs [48]. We observe a similar phase change in the anisotropy of giant cells. Rosette leaf giant cells are jigsaw puzzle-piece shaped and relatively isotropic. Later in the plant life cycle, giant cells in cauline leaves begin to be more anisotropic along the proximal–distal axis and start to resemble sepal giant cells. This supports the hypothesis that cauline leaves represent an intermediate state between rosette leaves and sepals [49]. Finally, sepal giant cells are highly anisotropic along the proximal–distal axis. Although sepals and leaves have notable yet subtle differences in cell size, cell-size patterning is regulated by the same developmental pathway in both organs. The genetic pathway that regulates giant cell specification has been co-opted from the epidermal specification pathway, which is a developmental pathway necessary for epidermal and thus plant development [14,50,51]. Without proper epidermal specification, the plant embryo will not progress past the globular stage of development [50–52]. The fact that this fundamental epidermal developmental pathway also patterns giant cells illustrates a common theme in development, namely, that regulatory proteins are commonly reused for more than one developmental process [53]. Taken together, our analysis and theoretical work on patterning during tissue growth highlights that unexpected effects can occur, and that these are difficult to infer from the canonical view of pattern formation arising in a static tissue. In this instance, an initially random pattern of giant cells becomes nonrandom as the surrounding cells divide. Thus, the effects of cell proliferation might also be important to determine the spatial distribution of specialized cell types in other tissues. Materials and methods Plant growth conditions All seeds were sown on LM111 soil in pots and were stratified in the dark for 3 days at 4 °C. The pots were then transferred to Percival plant growth chambers set to 60% humidity, 22 °C temperature, and 24-h light provided by Philips 800 Series 32-Watt fluorescent bulbs (f32t8/tl841) (~100 μmol m−2 s−1). dpg were counted from the time the pots were transferred to plant growth chambers. Cloning fluorescent nuclear markers To create a teal fluorescent (TFP) nuclear marker ubiquitously expressed under the UBIQUITIN 10 promoter (pUBQ10::H2B-TFP; pAR393), an H2B-TFP fusion with an AAAPAAAAA linker was generated by PCR. TFP was amplified by PCR with primers oAR440 (5′-gct gcc gct cca gct gca gct gcc gct ATG GTT TCT AAG GGA GAA GAA ACT ACT ATG-3′) and oAR438 (5′-cct cga gtc aCT TAT AAA GTT CAT CCA TAC CAT CAG TAG-3′). The lower-case letters in the primer sequences represent linkers, restriction sites, and cloning sequences that were added to the gene sequences. H2B was PCR amplified with oAR369 (5′-CAC CGG ATC CAC AAT GGC GAA GGC AGA TAA G-3′) and oAR439 (5′-agc ggc agc tgc agc tgg agc ggc agc AGA ACT CGT AAA CTT CGT AAC CGC CTT AG-3′). Sequences encoding H2B-TFP were fused via overlapping PCR with oAR369 and oAR438 primers. The H2B-TFP PCR product was cloned into pENTR D TOPO to create the pAR198 entry clone. H2B-TFP was recombined into pUB-Dest with LR Clonase II according to the manufacturer’s instructions to generate pUBQ10::H2B-TFP (pAR393). pAR393 was transformed into Arabidopsis Col-0 plants expressing the pLH13 p35S::mCitrine-RCI2A yellow fluorescent plasma membrane marker [45] via Agrobacterium tumefaciens (strain GV3101)-mediated floral dipping [54] and selection with glufosinate-ammonium (“Basta” Neta Scientific OAK-044851-25g). Generation of mutant plant lines containing fluorescent plasma membrane and nuclear markers for imaging Mutant alleles and overexpression transgenes were crossed to plants expressing fluorescent cell-membrane and nuclear markers to obtain plants for imaging. The following mutant alleles were used: acr4-2, atml1-3, dek1-4, and lgo-2. In addition, two lines overexpressing either ATML1 (pPDF1::FLAG-ATML1) or LGO (pATML1::LGO) in a Col-0 background were used. All of these alleles/transgenes are in the Columbia-0 (Col-0) accession. acr4-2 (SAIL_240_B04) contains a T-DNA insertion in the codon of the second of seven 39-amino-acid repeats of the beta propeller extracellular domain, which is upstream of the transmembrane domain and the kinase domain and is therefore presumed to be loss-of-function allele. The acr4-2 mutant was obtained from Gwyneth Ingram [14], who obtained it from Syngenta [55]. The atml1-3 allele is a T-DNA insertion in the homeodomain and is a loss-of-function mutant [11]. The atml1-3 was obtained from the Arabidopsis Biological Resource Center (ABRC; accession number SALK_033408) [11]. The dek1-4 allele contains a point mutation that changes a conserved arginine to a cysteine within calpain domain III (ABRC accession CS68904) [11]. Complete loss of function of DEK1 is lethal [50]; therefore, dek1-4 must retain some function. The dek1-4 phenotype is recessive and therefore, dek1-4 is likely hypomorphic [11]. The dek1-4 mutant was originally isolated in the Landsberg erecta accession and was subsequently back-crossed twice into Col-0 [12]. lgo-2 contains a T-DNA insertion within the coding sequence of the gene and is a loss-of-function allele [10]. lgo-2 was obtained from the ABRC (accession number SALK_033905 and is available as a homozygous mutant as accession CS69160). pPDF1::FLAG-ATML1 (ATML1-OX) was obtained from Gwyneth Ingram [56]. pATML1::LGO (LGO-OX) has been deposited for distribution at ABRC under accession CS69162 [11,45]. acr4-2, atml1-3, dek1-4, lgo-2, pPDF1::FLAG-ATML1, and pATML1::LGO were each crossed to plants expressing both a p35S::mCitrine-RCI2A fluorescent plasma membrane marker (pLH13) and a pUBQ::H2B-TFP fluorescent nuclear marker (pAR393). The F2 progeny were genotyped for acr4-2, atml1-3, dek1-4, and lgo-2 (primer sequences in S1 Table) and lines were isolated that were homozygous for these alleles and that also expressed both pUBQ::H2B-TFP and p35S::mCitrine-RCI2A. We could not obtain atml1-3 homozygous plants that also contained the p35S::mCitrine-RCI2A transgene after crossing, which was probably because the ATML1 gene was linked to the insertion site of the p35S::mCitrine-RCI2A transgene. To obtain plants expressing p35S::mCitrine-RCI2A in a homozygous atml1-3 background, the plasmid containing p35S::mCitrine-RCI2A was transformed into atml1-3 homozygous plants with pUBQ::H2B-TFP through Agrobacterium tumefaciens (strain GV3101)-mediated floral dipping [53]. A T1 line was chosen that strongly expressed the mCitrine membrane signal and this line was used for future experiments. For the overexpression transgenes pPDF1::FLAG-ATML1 and pATML1::LGO, seeds were collected from F2 plants and F3 plants were genotyped for pPDF1::FLAG-ATML1 or pATML1::LGO (S1 Table). Those F2 plants that produced only F3 plants having pPDF1::FLAG-ATML1 or pATML1::LGO were isolated as homozygous for pPDF1::FLAG-ATML1 or pATML1::LGO, respectively. Those lines with pPDF1::FLAG-ATML1 or pATML1::LGO homozygous and that expressed both pUBQ::H2B-TFP and p35S::mCitrine-RCI2A were used for imaging. Sample preparation for imaging Leaves and sepals were mounted in 0.01% (v/v) Triton X-100 for imaging. Leaves were imaged between two coverslips, and sepals were imaged on a slide with a coverslip. Curvy leaves were cut with a razor blade to ensure they could be placed flat under the coverslip. Samples were imaged immediately after preparation. Sepals were imaged at stage 14 [57]. Imaging with confocal microscopy A ZEISS LSM 710 Axio Examiner confocal microscope with a W Plan-Apochromat 20×/1.0 DIC D-0.17 M27 75 mm water-immersion objective lens was used to image leaf 1 or 2 of the Arabidopsis rosette and mature (stage 14) sepals. A 458 nm laser was used to excite TFP in pUBQ::H2B-TFP (collection range 463–500 nm) and a 514 nm laser was used to excite mCitrine in p35S::mCitrine-RCI2A (collection range 525–645 nm). Images were captured with a 1× zoom. The gain and laser power varied slightly between images to accommodate slight differences in signal intensity between samples. Each image was composed of several tiles. The dimensions of each voxel were 0.415 µm (x) by 0.415 µm (y) by 1 µm (z). For the images to compare cell size with nuclear fluorescence on the abaxial and adaxial faces of the same organ (two leaf replicates and three sepal replicates; Figs 2 and S2), the 458-nm laser power and gain used for imaging TFP in pUBQ::H2B-TFP were adjusted so that the TFP signal was below saturation and was then held constant for all images. Leaf areas were calculated from confocal images of entire leaves taken using a 2.5× objective for each 9-dpg and 25-dpg leaf replicate. Image processing Tiles were stitched in the horizontal direction by ZEISS stitching software (overlap of 5% and threshold of 0.7) and in the vertical direction with MorphoGraphX [58,59] using the process “Stacks/Multistack/Merge Stacks” (parameters: method = max; interpolation = linear). Assembled images were saved as MorphoGraphX stack files. A surface mesh was created in MorphoGraphX from each image to perform segmentation and analysis on the epidermis. First, extraneous parts of the image were removed with the Voxel Edit tool. (Such extraneous parts of the image include trichomes on the adaxial images and pollen grains/nematode eggs on some leaf images.) Then, an image was subjected to Gaussian Blur using the process “Stack/Filter/Gaussian Blur Stack” (parameters: x = 2; y = 2; z = 2). Next, the tissue surface was identified with the process “Stack/Morphology/Edge Detect” (parameters: threshold varied between 2,300 and 7,000 according to individual image brightness; multiplier = 2.0; adapt factor = 0.3; fill value = 30,000). These steps extracted a surface of the leaf. The process “Stack/Morphology/Fill Holes” was applied to some images when holes were apparent in the surface (parameters: x-radius = 20; y-radius = 20; threshold = 10,000; depth = 0; fill value = 30,000). This surface was then used to generate a mesh with the process “Mesh/Creation/Marching Cube Surface” (parameters: cube size = 5 μm; threshold = 20,000). The mesh was smoothed with “Mesh/Structure/Smooth mesh” (parameters: number of passes varied between 20–45; Walls Only = no). The mesh obtained was then subdivided with the process “Mesh/Structure/Subdivide” either once or twice depending on its size. In order to obtain the cell membrane signal on the surface, the process “Meshes/Signal/Project Signal” (parameters: min/max distances ranged between 5 and 15 μm; MinSig = 0.0; MaxSig = 60,000) was used to project the mCitrine signal from the p35S::mCitrine-RCI2A plasma membrane marker onto the mesh at an optimal depth. The depth range yielding the clearest cell membrane signals with minimal distortion was selected. To perform cell segmentation, each individual cell in the leaf was first manually identified with a cell label marking (seed). Using these seeds, watershed segmentation was performed using the process “Meshes/Segmentation/Watershed Segmentation.” Adjacent pairs of stomatal guard cells were segmented together to form a single cell; these were classified as stomata. Errors in segmentation were identified and corrected by removing the label for those cells and reseeding. The Heat Map processes computed the cell area and other morphological cell features as well as the position of every cell. The cell area data were exported into a data table file for each image. In addition, other cellular shape features were computed and exported into a data table for the cell type classification (Materials and methods: cell type classification). To analyze the nuclear signal from images of leaves expressing pUBQ::H2B-TFP, vertical stitching of tiles (already horizontally stitched with Zeiss stitching software) was performed in MorphoGraphX (max method and linear interpolation). Because the TFP reporter was expressed under the UBIQUITIN 10 promoter, TFP was localized in nuclei of the mesophyll cells in addition to cells of the epidermis. Mesophyll nuclei were removed with the Voxel Edit tool. Nuclei were identified as being from the mesophyll by lining up the nuclear signal images with their corresponding membrane signal images and comparing the nuclei within the bounds of each epidermal cell membrane. When compared with an epidermal cell nucleus, mesophyll cell nuclei were often dimmer and lower down and therefore, excess nuclei were removed according to these criteria so that each epidermal cell had one nucleus. When it was ambiguous which of two nuclei in a single cell was from the mesophyll or epidermis, both nuclei were removed and excluded from the analysis. Segmentation of the nuclei was performed in MorphoGraphX so that the total signal could be calculated for each nucleus. To do so, the confocal image was first subjected to “Stack/Filters/Brighten Darken” (parameter: 1). Next, a gaussian blur was performed using “Stack/Filters/Gaussian Blur” (parameters: x = 1, y = 1, z = 1), followed by a binarization with the process “Stack/Filters/Binarize” (parameters: threshold = 2,000), which functioned to select pixels above a threshold value to identify the edges of each nucleus. A lower threshold value was chosen so that we could identify the entire nuclei even for dim nuclei. The Voxel edit tool was used to separate nuclei that inflated into one another. We then created a mesh from the binarized image using the process “Mesh/Creation/Marching Cubes 3D” (parameters: cube size = 1, min voxel = 0, smooth passes = 3, label = 0). To ensure that the mesh covered all fluorescence of each nucleus, we expanded the mesh using “Mesh/Structure/Shrink Mesh” with a negative value (parameter: distance = −1). Individual nuclei were manually seeded and then the watershed segmentation was performed with the process “Mesh/Segmentation/Watershed Segmentation” to identify each nucleus. The Heat Map function calculated the total H2B-TFP fluorescence within each nucleus, as a representation of DNA content. To study correlations between total nuclear H2B-TFP signal and cell size, individual cells from cell area meshes were matched with their constituent nuclei from nuclear signal meshes using MorphoGraphX parent tracking. For the leaf replicates, total nuclear H2B-TFP signal was calculated for as many cells as possible from both the abaxial and adaxial sides. For the sepal replicates, total nuclear H2B-TFP signal was calculated only on the abaxial side and only for the largest cells. To create the heat maps overlaid with vasculature in S13 Fig, confocal images of leaves expressing p35S::mCitrine-RCI2A were used to create surfaces and were segmented as described above to create cell area heat maps. The mCitrine-RCI2A confocal images were found to have signal in the vasculature, so that the trajectories of veins could be traced in images from the abaxial surface of the image. The mCitrine-RCI2A confocal images were transformed around the z-axis in MorphoGraphX. For each leaf, the cell area heat map and the mCitrine-RCI2A confocal image transformed around the z-axis were aligned in MorphoGraphX and PNG screen captures were taken of each. These PNGs were then loaded into Adobe Illustrator and the veins were traced in white onto the heat maps. Please note that wild-type 25 dpg leaf replicates 1, 3, and 4, lgo-2 25 dpg leaf replicates 1 and 2, and LGO-OX replicates 1, 2, and 3 were used for an independent analysis of cell shape (specifically lobeyness) [60]. Statistical analysis To analyze the relationship between total nuclear H2B-TFP signal (DNA content) and cell area for the leaves in Figs 2 and S2, linear regressions were performed on R statistical software (https://www.r-project.org/). To compare the total nuclear H2B-TFP signal (DNA content) of the cells of largest area between sepals and leaves, the cell area at the 98th percentile was calculated for each of the three abaxial sepal replicates and these three cell areas were averaged for an area threshold of 4,308 µm2. Cell area versus total nuclear H2B-TFP signal (DNA content) was plotted for cells above this 4,308 µm2 area threshold for the abaxial sepals and the abaxial and adaxial leaves. To compare positions of the largest cells on the abaxial and adaxial sides of each leaf at different stages of development (Fig 3), the abaxial and adaxial images were aligned in MorphoGraphX. Then, cell area heat maps were created and the x and y coordinates of the center of each cell were calculated (Fig 3C). Cell area thresholds for each leaf were determined from the 98th percentile cell area of the abaxial side, and the positions of cells above these area thresholds were plotted for the abaxial and adaxial images of each leaf. R statistical software was used to analyze the cell-size distributions and create the violin plots and Wasserstein plots. The threshold for significance was set to alpha = 0.05. To create the Wasserstein plots, a Wasserstein test was performed between each pair of replicate distributions. A test statistic (also known as Wasserstein distance) and p-values were returned for every test. The p-values are listed in S8C and S8D Fig. Classical multidimensional scaling was performed to create a 2D coordinate for each replicate distribution based on the Wasserstein distances, and points from these coordinates were plotted. To ensure that the distances between 2D points adequately reflected the Wasserstein distances among replicate distributions, we plotted the Wasserstein distances against the Euclidean distances between points (S8E and S8F Fig). The linear relationships between Wasserstein distances and Euclidean distances showed that the 2D graph accurately represents the differences between distributions. To create the Wasserstein plot of the combined 9- and 25-dpg cell area data, cell areas of each replicate in each genotype were normalized by the mean cell area for that replicate. In this way, each replicate has a mean of 1. This eliminated the difference in the mean values of the cell size between the 9- and 25-dpg leaves, such that all higher moments of the area distributions could be compared rather than the absolute sizes. To statistically compare the differences in the number of giant cells across genotypes in the leaf at 9 dpg and at 25 dpg, two-sample, two-tailed t-tests that assumed equal variance were performed on the number of giant cells per segmented area between wild type and the different genotypes. To statistically assess the randomness of the cellular patterns, see section “Statistical analysis of the cellular patterns” below. Cell type classification To automatically distinguish stomata from pavement cells, a supervised classification algorithm was used based on cell shape features (S9 Fig). Cell shape features were computed from each 2.5D mesh using the MorphoGraphX [58,59] process “Mesh/Heat Map/Analysis/Cell Analysis 2D” and were extracted with “Mesh/Attributes/Save as CSV” into a data table. Three distinct training datasets were created using a single wild-type replicate—one for the sepal, one for the leaf at 25 dpg and one for the leaf at 9 dpg. To get the different training datasets, we manually selected some pavement cells and stomata and labeled them as different cell types, ran the classification processes available within MorphoGraphX, and manually corrected the cells that were wrongly identified. These training datasets were then used to train a supervised learning algorithm (Support Vector Machine quadratic) using the Classification Learner App in MATLAB [61,62]. The following cellular shape features were selected to train the classifier in the 25-dpg leaf: area, average radius, length of the major axis, maximum radius, perimeter, circularity, lobeyness (ratio of the cell perimeter over that of its convex hull; convex shapes have lobeyness 1) [59], and rectangularity (ratio of the cell area over the area of the minimum bounding rectangle in the cell) [59]. For the sepal, the aspect ratio and the length of the minor axis were also taken into account. For the 9-dpg leaf, where the variety of cell types was more complex, three cell types were defined (pavement cell, meristemoid, and stomata) and meristemoid and stomata were combined in a postprocessing script. The shape features used to train the classifier were area, average radius, minimum radius, perimeter, circularity, lobeyness, and visibility stomata (it counts the proportion of straight lines that connects the cell outline without passing through a cell boundary) [59,63]. To automatically predict cell types in all replicates, a developed MATLAB script containing the trained classifier and a postclassifier filter, which corrects for potentially wrong predictions on the basis of known shape criteria, was applied. Manual corrections were finally performed, in which misclassified cells were re-labeled with their correct cell type. Giant cells were defined by a cell-size threshold (S9 Fig). Because a few giant cells were expected in atml1-3 mutants, atml1-3 mutants were used as a reference to build this threshold. Fewer than 0.7% of the pavement cells were considered to be giant cells in atml1-3 tissues, which was supported through visual observation in the sepal. Consequently, the giant cell-size threshold was set as that corresponding to the average between the 99.3rd percentile cell-size value with the cell-size value immediately above it in the distribution, taking into account the data of three atml1-3 pooled replicates. For consistency, the same method was applied to the sepal, and to 9- and 25-dpg leaves, which gave three different threshold values (sepal: 5,290 µm2, leaf 9-dpg: 2,570 µm2, leaf 25-dpg: 14,160 µm2). The percentiles were only calculated on rectangular sections (omitting cells at the outline of the organs) of the sepal to maintain consistency across different organs. Classification output examples in different genotypes are shown in S10 and S11 Figs. Randomization of the experimental images To assess the randomness of the cellular patterns, it was essential to establish a random reference, or null model, against which the observed pattern could be compared. To produce the required random reference, the image-based method dmSET [36,37] was applied to generate 400 synthetic random equivalent tissues from each segmented image. Cell positions and orientations were randomly shuffled into new images (named randomizations) [37], while preserving individual approximate cell shapes and sizes (S15 Fig). Only the incomplete cells at the border of the images were fixed. This randomization method avoids potential biases arising from the heterogeneity of cell sizes and shapes in the tissues, which affects the number of neighboring cells. We ensured that cellular properties, and more specifically cell area and cell lobeyness (defined as the perimeter of the cell divided by the perimeter of its convex hull), were approximately conserved in the randomized leaf tissues (S15D and S15E Fig; the Pearson coefficient was 0.98 for cell area correlation and was 0.94 for cell lobeyness correlation). In the sepal randomized tissues, cell orientations were constrained between −π/6 and +π/6 compared with their initial orientation, to maintain the anisotropy of the tissue. Cell shape properties were also approximately conserved (S15I and S15J Fig). A custom-made MATLAB script was subsequently applied to both original and randomized images to correct errors introduced by the dmSET method and to compute cell shape properties and cellular network information that was used to quantify the cellular pattern. Before randomizing the different sepal and leaf replicate images (Figs 5 and S14), each 2.5D mesh was first converted into a 2D pixel image using the process “Stack/Mesh Interaction/Mesh To Image” (with a pixel size of 1μm) in MorphoGraphX. Subsequently, a square crop (in the leaf) or rectangular crop (in the sepal) that maximized the tissue section was performed in the segmented images. These 2D segmented images were then randomized using the dmSET method. To study the change in the giant cell spatial pattern over time, published time-lapse sepal data were used [39] that were randomized at two different time points (sepal at stage 4: 24 h, and at stage 8: 96 h). Cell segmentation and cell lineage tracking were already performed in [39]. Using MorphoGraphX, sepal cells were manually selected at the later time point, and the exact corresponding mother cells at the first time point were established using the lineage tracking analysis from [39]. In order to quantify the spatial pattern of the same giant cells at two different time points, giant cells at both stages were defined as the pavement cells that did not divide during this period of time. This approach allowed the comparison of the change in tissue organization consistently at two different time points. Then, the 2.5D meshes were projected into 2D images. These images were subsequently randomized using the dmSET method. Here, to facilitate the study of the same giant cells over time, the images were not cropped and the entire studied tissue was randomized, including the cells at the edges. To achieve this, the background region, located outside the tissue of interest, was considered as a single cell that remained fixed in the randomized tissues. Examples of randomizations are shown in S20C and S20D Fig. Three different sepal replicates were used for these analyses of the time-lapse data. “Segmentation” and “Randomization” images appearing in figures such as Figs 5, 6, S14, S17, S18, and S20 were produced with a Python script using the multi-labeled images generated by dmSET. Several approaches were used to test the robustness of our method and ensure that there was no bias in our null model. In the randomized tissues, cell sizes were well preserved (S15D Fig), but cell shapes were affected in the leaf (S15E Fig). Because the leaf pavement cells in the randomizations exhibit different shapes (more convex shapes and noisier edges) compared with the original cells, we developed an additional alternative method for generating randomized tissues (S19 Fig), called Cut and Merge Cells. In this approach, each leaf replicate was first over-segmented using MorphoGraphX (autoseeding with r = 4 μm combined with manual seeding) to create templates composed of small fragments of pavement cells (S19A Fig). These templates were then used to automatically generate a random pattern of giant cells, preserving their original sizes and numbers (S19A Fig). Giant cells were created recursively: their positions were randomly allocated to a first pavement cell fragment, which was then merged with neighboring cell fragments until the original giant cell size was reached (S19A Fig). Similar results were obtained by using this randomization method (S19B and S19C Fig), confirming the tendency of giant cells to cluster more than would be expected by chance. However, while cell shapes produced by this method exhibited less noisy edges, they remained less lobed and more convex compared with those in the original segmentation (S19F Fig). Moreover, this method does not randomize all cells in the tissue, such as stomata. We wondered whether the shape artifacts introduced by the dmSET method could introduce bias into our null model by affecting the contacts between cells in the randomized tissues differently than in the original segmentations. To explore this possibility, we generated “reconstructions” of the original images using the same dmSET method as for the randomizations, except that each cell’s position from the initial segmentation was preserved (with some added uniform noise having a maximum of ±5 μm in the y and x directions) (S18A–S18C, S18E, and S18F Fig). This resulted in cell shape artifacts that were nearly identical between the reconstructions and the randomizations (S18J–S18L Fig), making them more comparable. Indeed, the differences in cell lobeyness (defined as the perimeter of the cell divided by the perimeter of its convex hull) from the original tissues were similar between the reconstructions and randomizations for large cells (S18J Fig). The number of cell neighbors differed slightly from the original segmentations due to the introduced shape artifacts (S18M and S18N Fig), but we found that giant cell contacts remained mostly the same in the reconstructions. Specifically, we observed approximately 6% fewer contacts between giant cells in the reconstructions compared with the segmentations. When comparing reconstructions with randomizations (which are comparable because of their similar cell shapes), the results remained consistent (S18O Fig), with giant cells being more clustered than expected by chance (p = 0.005). In the sepal, cell shape was quite preserved (S15I and S15J Fig; the Pearson coefficient was >0.99 for cell area and perimeter). Sepal cell edges were noisier in the randomized tissues, with a higher cell perimeter (S15J Fig); this could potentially lead to more contacts in the randomized tissues, which would not affect our conclusion. Furthermore, we evaluated our method by using a randomly selected population of pavement cells in both leaves and sepals (S17 Fig). The null hypothesis could not be rejected and this result was robust across five different random patterns in all replicates (one is shown in S17 Fig). Specifically, we obtained p-values ranging from 0.395 to 0.488 in leaves and from 0.178 to 0.335 in sepals. The number of giant cell contacts was either slightly more or less than the expected mean random value, attributable to variability in the random patterns. This indicates that the artificial random pattern did not deviate from true randomness, suggesting that our null models do not present a significant bias. Statistical analysis of the cellular patterns By comparing a spatial observable in the cellular network of the actual segmentation with the corresponding observable in the cellular networks of the 400 generated randomized tissues, whether the considered observable is likely to be observed by chance can be statistically tested [36]. Hence, the use of this method on observables measuring distances or contacts between the studied cells allows the assessment of whether the arrangement of the cells within the tissue is random, clustered or dispersed (Fig 5A). To quantify the patterns, a custom-made Python script was used to extract pertinent observables (i.e., spatial quantities) from the cellular network, which used the NetworkX Python library [64]. In this manuscript, we mainly focused on the number of giant cell neighbors per giant cell to quantify the number of local contacts between giant cells. Other observables have been quantified, such as the minimum shortest path between giant cells, and the number of giant cells in a cluster (S16 Fig). When dealing with cropped images, giant cells (or any cell population studied, see Fig 6) at the border of the image were not considered in the analysis. The number of giant cell neighbors was extracted for every giant cell, and the mean number of giant cell neighbors per giant cell across all giant cells was computed within each experimental replicate. Similarly to the methodology described by the authors of the dmSET method [36], the mean value extracted from the segmentation image was compared with the approximated null distribution formed by the 400 mean values extracted from the randomized images. We first performed the analysis on each replicate independently (S21 and S22 Figs). As the cell-size distributions in the different replicates showed similarities across replicates (Figs 5, S21, and S22), replicates were pooled to increase the sample size and statistical power. Six image replicates were used for both the leaf and sepal wild-type (Fig 5), and three replicates were used for the wild-type sepal time-lapsed images (Fig 7). To test the null hypothesis assessing the randomness of the observed metric, a p-value p was obtained as the ratio of the number of random images (defined here as one random image resulting from pooling one random image per replicate) displaying the same or a more extreme value than the one obtained in the segmentation replicates (one-sided test). If the value fell within the null distribution with an associated high p-value (p > 0.05), the null hypothesis could not be rejected, indicating that the observed quantity could likely be expected by chance. If the value fell outside the null distribution, we assigned p < 0.0025, with 0.0025 corresponding to the inverse of the number of random images (400) used to create the null distribution. In addition, the distribution of the number of giant cell neighbors for all giant cells from the pooled experimental replicates was studied, which provided more insights into their spatial organization. This was compared qualitatively with the distribution expected in a random tissue, extracted from the 400 randomized tissues of all replicates. All plots derived from these analyses were performed with Python, with the use of the matplotlib [65] and seaborn packages [66]. Mathematical model for giant cell fate commitment and numerical simulations To simulate the giant cell fate decisions, our published stochastic and cell-autonomous multicellular model in a growing tissue was used [12]. In that model, the transcription factor ATML1 stochastically fluctuates and drives the expression of its target LGO. In the simulated growing tissue, cells divide using a timer with some stochasticity. When the timer of a cell reaches a threshold ΘC,S, cells undergo the S-phase, and therefore cells transition from being diploid (2C) to tetraploid (4C). By default, cells that reach a second and higher timer threshold ΘC,D will undergo division. However, those cells that have reached a certain LGO concentration threshold ΘT after undergoing the S-phase, considered to be in the G2 phase, will not divide and are maintained in an endoreduplication cycle, which increases their ploidy, becoming giant cells. To model the dynamics of the concentrations for ATML1 and LGO, and the Timer variable, we use chemical Langevin equations [67], which are differential equations with a corresponding deterministic part, consisting of production, degradation and regulatory terms, followed by a stochastic part modeling thermodynamic fluctuations that contains a square root, whose radicand has the sum of the absolute values of the production, degradation, and regulatory terms. The dynamics of ATML1 concentration, LGO concentration and Timer variable in cell i, denoted by [ATML1]i, [Target]i and Timeri, respectively, follow the Langevin equations given by (1)(2)(3) Eqn. (1) stands for the rate of change of ATML1 in cell i, and its terms on the right-hand side describe constitutive expression, self-activation (implemented via a Hill function) [68], linear degradation, and the corresponding stochastic term in ATML1; Eqn. (2), stands for the rate of change of the Target in cell i, and its terms on the right-hand side describe ATML1-induced expression of the Target (using also a Hill function), linear degradation, and the corresponding stochastic term in the Target; Eqn. (3) stands for the rate of change of the Timer in cell i, and its terms on the right-hand side are a constitutive production and the corresponding stochastic term. The parameters of the equations are as follows: PX is the basal production rate for the X variable (where X is either A for ATML1, T for Target concentration or C for the Timer variable), VX is the prefactor of the ATML1-dependent production rate for the X variable, KX is the ATML1 concentration at which the ATML1-dependent production rate for the variable X has its half-maximal value, nX is the Hill coefficient, and GX is the linear degradation rate for the X variable. εi(t) is a normalized cell area, εi(t) = E0Ei(t), where E0 is an effective cell area, and Ei(t) is the area of cell i in arbitrary units. ηXi is a random Gaussian variable with zero mean that fulfills〈ηXi(t)ηX′j(t′)〉 = δ(t − t′) δXX′δij, where i and j are cell indices, X and X′ the modeled variables, δXX′ and δij are Kronecker deltas and δ(t − t′) is the Dirac delta function. Upon cell division, the Timer was reset. To implement the resetting, the following rule was applied at each time step: (4) where Ui is a uniform randomly distributed number in the interval [0, 0.5) and ΘC,D is the cell division threshold for the Timer. The multicellular template on which the simulations were run and initial conditions were the same as in Meyer and colleagues (2017). Initial conditions for ATML1 and Target were randomly uniformly distributed in the interval of [0,1) and [0,0.1), respectively. The Timer initial conditions were set in correlation with the cell areas in the initial template with some stochasticity, as performed in Meyer and colleagues (2017). The differences between the used initial conditions were just in the ATML1, Target, and Timer initial cellular values, determined by different random numbers. Tissue growth and division were also implemented as in Meyer and colleagues (2017). The multicellular tissue grows anisotropically, to emulate the patterning process in the sepal. This anisotropic growth was implemented by imposing a displacement of the vertices with respect to the center of mass of the tissue, with a given radial and vertical exponential rate. After each simulation step, dilution effects due to growth in the modeled variables were taken into account. Cells divided using the shortest path rule together with the constraint of having the division plane through the center of mass of the cell. We assumed that molecules are homogeneously distributed within cells, and therefore, upon cell division, sister cells have the same concentration of ATML1 and Target variables at birth, but can have different cell sizes. Numerical simulations were performed with Tissue software [12,13,69], and the integration was performed using an Îto interpretation of the Langevin equations with a Heun algorithm [70]. Integration was performed with a time step dt = 0.1, and simulations were stopped at time 135. Parameter values for the simulations are given in S2 Table. Parameters were chosen such that the wild-type behavior reported in Meyer and colleagues (2017) could be recapitulated. The outcome of the simulation in Fig 7B was displayed using Paraview software [71]. We recently proposed a more detailed model of the ATML1 regulatory network to study how giant cell specification and cell fate maintenance depend on VLCFA [13], which is still a stochastic and cell-autonomous model. Here, however, for the sake of simplicity, and the intention of using a minimal, stochastic, and cell-autonomous phenomenological model, the former ATML1 model was used [12]. Randomizations of the outcomes from the numerical simulations To assess the randomness of the giant cell pattern in the numerical simulations (Fig 7), the same method was employed as that used for the experimental images. Although randomizations of the tissues were performed similarly (see the “Randomization of the experimental images” section above), a Python script was developed to display the output of the simulation as a multi-labeled image, where each cell was colored with a different label. These images could therefore be randomized using the dmSET method [36,37]. To compare the simulated giant cell pattern (Fig 7B) with the giant cell pattern found in the experimental mature sepals (Fig 5B), the output image was cropped using the maximal rectangle in the tissue, and giant cells were also defined by a size threshold, ensuring that all cells with a ploidy of 16C or higher were categorized as giant cells (Fig 7C). The few 8C cells that exceeded this threshold were also considered as giant cells. To study the change in the giant cell spatial pattern over time (Fig 7F), the same simulations were used, but only the first-arising giant cells (i.e., cells that stopped dividing after time t = 55 of the simulations for being committed to endoreduplicate) were studied. The same method was used to assess the randomness of the cellular patterns on these giant cells both at time t = 55 and time t = 135. Here, instead of cropping the image, the whole tissue was randomized (using the dmSET method), including the cells at the edges, such that exactly the same giant cells were considered at both time points. Examples of randomizations are shown in S20A and S20B Fig. The analysis was performed over five simulation replicates, with different cellular random initial conditions. Related “Segmentation” and “Randomization” images appearing in Figs 7 and S20 were produced with a Python script using the multi-labeled images generated by dmSET. Plant growth conditions All seeds were sown on LM111 soil in pots and were stratified in the dark for 3 days at 4 °C. The pots were then transferred to Percival plant growth chambers set to 60% humidity, 22 °C temperature, and 24-h light provided by Philips 800 Series 32-Watt fluorescent bulbs (f32t8/tl841) (~100 μmol m−2 s−1). dpg were counted from the time the pots were transferred to plant growth chambers. Cloning fluorescent nuclear markers To create a teal fluorescent (TFP) nuclear marker ubiquitously expressed under the UBIQUITIN 10 promoter (pUBQ10::H2B-TFP; pAR393), an H2B-TFP fusion with an AAAPAAAAA linker was generated by PCR. TFP was amplified by PCR with primers oAR440 (5′-gct gcc gct cca gct gca gct gcc gct ATG GTT TCT AAG GGA GAA GAA ACT ACT ATG-3′) and oAR438 (5′-cct cga gtc aCT TAT AAA GTT CAT CCA TAC CAT CAG TAG-3′). The lower-case letters in the primer sequences represent linkers, restriction sites, and cloning sequences that were added to the gene sequences. H2B was PCR amplified with oAR369 (5′-CAC CGG ATC CAC AAT GGC GAA GGC AGA TAA G-3′) and oAR439 (5′-agc ggc agc tgc agc tgg agc ggc agc AGA ACT CGT AAA CTT CGT AAC CGC CTT AG-3′). Sequences encoding H2B-TFP were fused via overlapping PCR with oAR369 and oAR438 primers. The H2B-TFP PCR product was cloned into pENTR D TOPO to create the pAR198 entry clone. H2B-TFP was recombined into pUB-Dest with LR Clonase II according to the manufacturer’s instructions to generate pUBQ10::H2B-TFP (pAR393). pAR393 was transformed into Arabidopsis Col-0 plants expressing the pLH13 p35S::mCitrine-RCI2A yellow fluorescent plasma membrane marker [45] via Agrobacterium tumefaciens (strain GV3101)-mediated floral dipping [54] and selection with glufosinate-ammonium (“Basta” Neta Scientific OAK-044851-25g). Generation of mutant plant lines containing fluorescent plasma membrane and nuclear markers for imaging Mutant alleles and overexpression transgenes were crossed to plants expressing fluorescent cell-membrane and nuclear markers to obtain plants for imaging. The following mutant alleles were used: acr4-2, atml1-3, dek1-4, and lgo-2. In addition, two lines overexpressing either ATML1 (pPDF1::FLAG-ATML1) or LGO (pATML1::LGO) in a Col-0 background were used. All of these alleles/transgenes are in the Columbia-0 (Col-0) accession. acr4-2 (SAIL_240_B04) contains a T-DNA insertion in the codon of the second of seven 39-amino-acid repeats of the beta propeller extracellular domain, which is upstream of the transmembrane domain and the kinase domain and is therefore presumed to be loss-of-function allele. The acr4-2 mutant was obtained from Gwyneth Ingram [14], who obtained it from Syngenta [55]. The atml1-3 allele is a T-DNA insertion in the homeodomain and is a loss-of-function mutant [11]. The atml1-3 was obtained from the Arabidopsis Biological Resource Center (ABRC; accession number SALK_033408) [11]. The dek1-4 allele contains a point mutation that changes a conserved arginine to a cysteine within calpain domain III (ABRC accession CS68904) [11]. Complete loss of function of DEK1 is lethal [50]; therefore, dek1-4 must retain some function. The dek1-4 phenotype is recessive and therefore, dek1-4 is likely hypomorphic [11]. The dek1-4 mutant was originally isolated in the Landsberg erecta accession and was subsequently back-crossed twice into Col-0 [12]. lgo-2 contains a T-DNA insertion within the coding sequence of the gene and is a loss-of-function allele [10]. lgo-2 was obtained from the ABRC (accession number SALK_033905 and is available as a homozygous mutant as accession CS69160). pPDF1::FLAG-ATML1 (ATML1-OX) was obtained from Gwyneth Ingram [56]. pATML1::LGO (LGO-OX) has been deposited for distribution at ABRC under accession CS69162 [11,45]. acr4-2, atml1-3, dek1-4, lgo-2, pPDF1::FLAG-ATML1, and pATML1::LGO were each crossed to plants expressing both a p35S::mCitrine-RCI2A fluorescent plasma membrane marker (pLH13) and a pUBQ::H2B-TFP fluorescent nuclear marker (pAR393). The F2 progeny were genotyped for acr4-2, atml1-3, dek1-4, and lgo-2 (primer sequences in S1 Table) and lines were isolated that were homozygous for these alleles and that also expressed both pUBQ::H2B-TFP and p35S::mCitrine-RCI2A. We could not obtain atml1-3 homozygous plants that also contained the p35S::mCitrine-RCI2A transgene after crossing, which was probably because the ATML1 gene was linked to the insertion site of the p35S::mCitrine-RCI2A transgene. To obtain plants expressing p35S::mCitrine-RCI2A in a homozygous atml1-3 background, the plasmid containing p35S::mCitrine-RCI2A was transformed into atml1-3 homozygous plants with pUBQ::H2B-TFP through Agrobacterium tumefaciens (strain GV3101)-mediated floral dipping [53]. A T1 line was chosen that strongly expressed the mCitrine membrane signal and this line was used for future experiments. For the overexpression transgenes pPDF1::FLAG-ATML1 and pATML1::LGO, seeds were collected from F2 plants and F3 plants were genotyped for pPDF1::FLAG-ATML1 or pATML1::LGO (S1 Table). Those F2 plants that produced only F3 plants having pPDF1::FLAG-ATML1 or pATML1::LGO were isolated as homozygous for pPDF1::FLAG-ATML1 or pATML1::LGO, respectively. Those lines with pPDF1::FLAG-ATML1 or pATML1::LGO homozygous and that expressed both pUBQ::H2B-TFP and p35S::mCitrine-RCI2A were used for imaging. Sample preparation for imaging Leaves and sepals were mounted in 0.01% (v/v) Triton X-100 for imaging. Leaves were imaged between two coverslips, and sepals were imaged on a slide with a coverslip. Curvy leaves were cut with a razor blade to ensure they could be placed flat under the coverslip. Samples were imaged immediately after preparation. Sepals were imaged at stage 14 [57]. Imaging with confocal microscopy A ZEISS LSM 710 Axio Examiner confocal microscope with a W Plan-Apochromat 20×/1.0 DIC D-0.17 M27 75 mm water-immersion objective lens was used to image leaf 1 or 2 of the Arabidopsis rosette and mature (stage 14) sepals. A 458 nm laser was used to excite TFP in pUBQ::H2B-TFP (collection range 463–500 nm) and a 514 nm laser was used to excite mCitrine in p35S::mCitrine-RCI2A (collection range 525–645 nm). Images were captured with a 1× zoom. The gain and laser power varied slightly between images to accommodate slight differences in signal intensity between samples. Each image was composed of several tiles. The dimensions of each voxel were 0.415 µm (x) by 0.415 µm (y) by 1 µm (z). For the images to compare cell size with nuclear fluorescence on the abaxial and adaxial faces of the same organ (two leaf replicates and three sepal replicates; Figs 2 and S2), the 458-nm laser power and gain used for imaging TFP in pUBQ::H2B-TFP were adjusted so that the TFP signal was below saturation and was then held constant for all images. Leaf areas were calculated from confocal images of entire leaves taken using a 2.5× objective for each 9-dpg and 25-dpg leaf replicate. Image processing Tiles were stitched in the horizontal direction by ZEISS stitching software (overlap of 5% and threshold of 0.7) and in the vertical direction with MorphoGraphX [58,59] using the process “Stacks/Multistack/Merge Stacks” (parameters: method = max; interpolation = linear). Assembled images were saved as MorphoGraphX stack files. A surface mesh was created in MorphoGraphX from each image to perform segmentation and analysis on the epidermis. First, extraneous parts of the image were removed with the Voxel Edit tool. (Such extraneous parts of the image include trichomes on the adaxial images and pollen grains/nematode eggs on some leaf images.) Then, an image was subjected to Gaussian Blur using the process “Stack/Filter/Gaussian Blur Stack” (parameters: x = 2; y = 2; z = 2). Next, the tissue surface was identified with the process “Stack/Morphology/Edge Detect” (parameters: threshold varied between 2,300 and 7,000 according to individual image brightness; multiplier = 2.0; adapt factor = 0.3; fill value = 30,000). These steps extracted a surface of the leaf. The process “Stack/Morphology/Fill Holes” was applied to some images when holes were apparent in the surface (parameters: x-radius = 20; y-radius = 20; threshold = 10,000; depth = 0; fill value = 30,000). This surface was then used to generate a mesh with the process “Mesh/Creation/Marching Cube Surface” (parameters: cube size = 5 μm; threshold = 20,000). The mesh was smoothed with “Mesh/Structure/Smooth mesh” (parameters: number of passes varied between 20–45; Walls Only = no). The mesh obtained was then subdivided with the process “Mesh/Structure/Subdivide” either once or twice depending on its size. In order to obtain the cell membrane signal on the surface, the process “Meshes/Signal/Project Signal” (parameters: min/max distances ranged between 5 and 15 μm; MinSig = 0.0; MaxSig = 60,000) was used to project the mCitrine signal from the p35S::mCitrine-RCI2A plasma membrane marker onto the mesh at an optimal depth. The depth range yielding the clearest cell membrane signals with minimal distortion was selected. To perform cell segmentation, each individual cell in the leaf was first manually identified with a cell label marking (seed). Using these seeds, watershed segmentation was performed using the process “Meshes/Segmentation/Watershed Segmentation.” Adjacent pairs of stomatal guard cells were segmented together to form a single cell; these were classified as stomata. Errors in segmentation were identified and corrected by removing the label for those cells and reseeding. The Heat Map processes computed the cell area and other morphological cell features as well as the position of every cell. The cell area data were exported into a data table file for each image. In addition, other cellular shape features were computed and exported into a data table for the cell type classification (Materials and methods: cell type classification). To analyze the nuclear signal from images of leaves expressing pUBQ::H2B-TFP, vertical stitching of tiles (already horizontally stitched with Zeiss stitching software) was performed in MorphoGraphX (max method and linear interpolation). Because the TFP reporter was expressed under the UBIQUITIN 10 promoter, TFP was localized in nuclei of the mesophyll cells in addition to cells of the epidermis. Mesophyll nuclei were removed with the Voxel Edit tool. Nuclei were identified as being from the mesophyll by lining up the nuclear signal images with their corresponding membrane signal images and comparing the nuclei within the bounds of each epidermal cell membrane. When compared with an epidermal cell nucleus, mesophyll cell nuclei were often dimmer and lower down and therefore, excess nuclei were removed according to these criteria so that each epidermal cell had one nucleus. When it was ambiguous which of two nuclei in a single cell was from the mesophyll or epidermis, both nuclei were removed and excluded from the analysis. Segmentation of the nuclei was performed in MorphoGraphX so that the total signal could be calculated for each nucleus. To do so, the confocal image was first subjected to “Stack/Filters/Brighten Darken” (parameter: 1). Next, a gaussian blur was performed using “Stack/Filters/Gaussian Blur” (parameters: x = 1, y = 1, z = 1), followed by a binarization with the process “Stack/Filters/Binarize” (parameters: threshold = 2,000), which functioned to select pixels above a threshold value to identify the edges of each nucleus. A lower threshold value was chosen so that we could identify the entire nuclei even for dim nuclei. The Voxel edit tool was used to separate nuclei that inflated into one another. We then created a mesh from the binarized image using the process “Mesh/Creation/Marching Cubes 3D” (parameters: cube size = 1, min voxel = 0, smooth passes = 3, label = 0). To ensure that the mesh covered all fluorescence of each nucleus, we expanded the mesh using “Mesh/Structure/Shrink Mesh” with a negative value (parameter: distance = −1). Individual nuclei were manually seeded and then the watershed segmentation was performed with the process “Mesh/Segmentation/Watershed Segmentation” to identify each nucleus. The Heat Map function calculated the total H2B-TFP fluorescence within each nucleus, as a representation of DNA content. To study correlations between total nuclear H2B-TFP signal and cell size, individual cells from cell area meshes were matched with their constituent nuclei from nuclear signal meshes using MorphoGraphX parent tracking. For the leaf replicates, total nuclear H2B-TFP signal was calculated for as many cells as possible from both the abaxial and adaxial sides. For the sepal replicates, total nuclear H2B-TFP signal was calculated only on the abaxial side and only for the largest cells. To create the heat maps overlaid with vasculature in S13 Fig, confocal images of leaves expressing p35S::mCitrine-RCI2A were used to create surfaces and were segmented as described above to create cell area heat maps. The mCitrine-RCI2A confocal images were found to have signal in the vasculature, so that the trajectories of veins could be traced in images from the abaxial surface of the image. The mCitrine-RCI2A confocal images were transformed around the z-axis in MorphoGraphX. For each leaf, the cell area heat map and the mCitrine-RCI2A confocal image transformed around the z-axis were aligned in MorphoGraphX and PNG screen captures were taken of each. These PNGs were then loaded into Adobe Illustrator and the veins were traced in white onto the heat maps. Please note that wild-type 25 dpg leaf replicates 1, 3, and 4, lgo-2 25 dpg leaf replicates 1 and 2, and LGO-OX replicates 1, 2, and 3 were used for an independent analysis of cell shape (specifically lobeyness) [60]. Statistical analysis To analyze the relationship between total nuclear H2B-TFP signal (DNA content) and cell area for the leaves in Figs 2 and S2, linear regressions were performed on R statistical software (https://www.r-project.org/). To compare the total nuclear H2B-TFP signal (DNA content) of the cells of largest area between sepals and leaves, the cell area at the 98th percentile was calculated for each of the three abaxial sepal replicates and these three cell areas were averaged for an area threshold of 4,308 µm2. Cell area versus total nuclear H2B-TFP signal (DNA content) was plotted for cells above this 4,308 µm2 area threshold for the abaxial sepals and the abaxial and adaxial leaves. To compare positions of the largest cells on the abaxial and adaxial sides of each leaf at different stages of development (Fig 3), the abaxial and adaxial images were aligned in MorphoGraphX. Then, cell area heat maps were created and the x and y coordinates of the center of each cell were calculated (Fig 3C). Cell area thresholds for each leaf were determined from the 98th percentile cell area of the abaxial side, and the positions of cells above these area thresholds were plotted for the abaxial and adaxial images of each leaf. R statistical software was used to analyze the cell-size distributions and create the violin plots and Wasserstein plots. The threshold for significance was set to alpha = 0.05. To create the Wasserstein plots, a Wasserstein test was performed between each pair of replicate distributions. A test statistic (also known as Wasserstein distance) and p-values were returned for every test. The p-values are listed in S8C and S8D Fig. Classical multidimensional scaling was performed to create a 2D coordinate for each replicate distribution based on the Wasserstein distances, and points from these coordinates were plotted. To ensure that the distances between 2D points adequately reflected the Wasserstein distances among replicate distributions, we plotted the Wasserstein distances against the Euclidean distances between points (S8E and S8F Fig). The linear relationships between Wasserstein distances and Euclidean distances showed that the 2D graph accurately represents the differences between distributions. To create the Wasserstein plot of the combined 9- and 25-dpg cell area data, cell areas of each replicate in each genotype were normalized by the mean cell area for that replicate. In this way, each replicate has a mean of 1. This eliminated the difference in the mean values of the cell size between the 9- and 25-dpg leaves, such that all higher moments of the area distributions could be compared rather than the absolute sizes. To statistically compare the differences in the number of giant cells across genotypes in the leaf at 9 dpg and at 25 dpg, two-sample, two-tailed t-tests that assumed equal variance were performed on the number of giant cells per segmented area between wild type and the different genotypes. To statistically assess the randomness of the cellular patterns, see section “Statistical analysis of the cellular patterns” below. Cell type classification To automatically distinguish stomata from pavement cells, a supervised classification algorithm was used based on cell shape features (S9 Fig). Cell shape features were computed from each 2.5D mesh using the MorphoGraphX [58,59] process “Mesh/Heat Map/Analysis/Cell Analysis 2D” and were extracted with “Mesh/Attributes/Save as CSV” into a data table. Three distinct training datasets were created using a single wild-type replicate—one for the sepal, one for the leaf at 25 dpg and one for the leaf at 9 dpg. To get the different training datasets, we manually selected some pavement cells and stomata and labeled them as different cell types, ran the classification processes available within MorphoGraphX, and manually corrected the cells that were wrongly identified. These training datasets were then used to train a supervised learning algorithm (Support Vector Machine quadratic) using the Classification Learner App in MATLAB [61,62]. The following cellular shape features were selected to train the classifier in the 25-dpg leaf: area, average radius, length of the major axis, maximum radius, perimeter, circularity, lobeyness (ratio of the cell perimeter over that of its convex hull; convex shapes have lobeyness 1) [59], and rectangularity (ratio of the cell area over the area of the minimum bounding rectangle in the cell) [59]. For the sepal, the aspect ratio and the length of the minor axis were also taken into account. For the 9-dpg leaf, where the variety of cell types was more complex, three cell types were defined (pavement cell, meristemoid, and stomata) and meristemoid and stomata were combined in a postprocessing script. The shape features used to train the classifier were area, average radius, minimum radius, perimeter, circularity, lobeyness, and visibility stomata (it counts the proportion of straight lines that connects the cell outline without passing through a cell boundary) [59,63]. To automatically predict cell types in all replicates, a developed MATLAB script containing the trained classifier and a postclassifier filter, which corrects for potentially wrong predictions on the basis of known shape criteria, was applied. Manual corrections were finally performed, in which misclassified cells were re-labeled with their correct cell type. Giant cells were defined by a cell-size threshold (S9 Fig). Because a few giant cells were expected in atml1-3 mutants, atml1-3 mutants were used as a reference to build this threshold. Fewer than 0.7% of the pavement cells were considered to be giant cells in atml1-3 tissues, which was supported through visual observation in the sepal. Consequently, the giant cell-size threshold was set as that corresponding to the average between the 99.3rd percentile cell-size value with the cell-size value immediately above it in the distribution, taking into account the data of three atml1-3 pooled replicates. For consistency, the same method was applied to the sepal, and to 9- and 25-dpg leaves, which gave three different threshold values (sepal: 5,290 µm2, leaf 9-dpg: 2,570 µm2, leaf 25-dpg: 14,160 µm2). The percentiles were only calculated on rectangular sections (omitting cells at the outline of the organs) of the sepal to maintain consistency across different organs. Classification output examples in different genotypes are shown in S10 and S11 Figs. Randomization of the experimental images To assess the randomness of the cellular patterns, it was essential to establish a random reference, or null model, against which the observed pattern could be compared. To produce the required random reference, the image-based method dmSET [36,37] was applied to generate 400 synthetic random equivalent tissues from each segmented image. Cell positions and orientations were randomly shuffled into new images (named randomizations) [37], while preserving individual approximate cell shapes and sizes (S15 Fig). Only the incomplete cells at the border of the images were fixed. This randomization method avoids potential biases arising from the heterogeneity of cell sizes and shapes in the tissues, which affects the number of neighboring cells. We ensured that cellular properties, and more specifically cell area and cell lobeyness (defined as the perimeter of the cell divided by the perimeter of its convex hull), were approximately conserved in the randomized leaf tissues (S15D and S15E Fig; the Pearson coefficient was 0.98 for cell area correlation and was 0.94 for cell lobeyness correlation). In the sepal randomized tissues, cell orientations were constrained between −π/6 and +π/6 compared with their initial orientation, to maintain the anisotropy of the tissue. Cell shape properties were also approximately conserved (S15I and S15J Fig). A custom-made MATLAB script was subsequently applied to both original and randomized images to correct errors introduced by the dmSET method and to compute cell shape properties and cellular network information that was used to quantify the cellular pattern. Before randomizing the different sepal and leaf replicate images (Figs 5 and S14), each 2.5D mesh was first converted into a 2D pixel image using the process “Stack/Mesh Interaction/Mesh To Image” (with a pixel size of 1μm) in MorphoGraphX. Subsequently, a square crop (in the leaf) or rectangular crop (in the sepal) that maximized the tissue section was performed in the segmented images. These 2D segmented images were then randomized using the dmSET method. To study the change in the giant cell spatial pattern over time, published time-lapse sepal data were used [39] that were randomized at two different time points (sepal at stage 4: 24 h, and at stage 8: 96 h). Cell segmentation and cell lineage tracking were already performed in [39]. Using MorphoGraphX, sepal cells were manually selected at the later time point, and the exact corresponding mother cells at the first time point were established using the lineage tracking analysis from [39]. In order to quantify the spatial pattern of the same giant cells at two different time points, giant cells at both stages were defined as the pavement cells that did not divide during this period of time. This approach allowed the comparison of the change in tissue organization consistently at two different time points. Then, the 2.5D meshes were projected into 2D images. These images were subsequently randomized using the dmSET method. Here, to facilitate the study of the same giant cells over time, the images were not cropped and the entire studied tissue was randomized, including the cells at the edges. To achieve this, the background region, located outside the tissue of interest, was considered as a single cell that remained fixed in the randomized tissues. Examples of randomizations are shown in S20C and S20D Fig. Three different sepal replicates were used for these analyses of the time-lapse data. “Segmentation” and “Randomization” images appearing in figures such as Figs 5, 6, S14, S17, S18, and S20 were produced with a Python script using the multi-labeled images generated by dmSET. Several approaches were used to test the robustness of our method and ensure that there was no bias in our null model. In the randomized tissues, cell sizes were well preserved (S15D Fig), but cell shapes were affected in the leaf (S15E Fig). Because the leaf pavement cells in the randomizations exhibit different shapes (more convex shapes and noisier edges) compared with the original cells, we developed an additional alternative method for generating randomized tissues (S19 Fig), called Cut and Merge Cells. In this approach, each leaf replicate was first over-segmented using MorphoGraphX (autoseeding with r = 4 μm combined with manual seeding) to create templates composed of small fragments of pavement cells (S19A Fig). These templates were then used to automatically generate a random pattern of giant cells, preserving their original sizes and numbers (S19A Fig). Giant cells were created recursively: their positions were randomly allocated to a first pavement cell fragment, which was then merged with neighboring cell fragments until the original giant cell size was reached (S19A Fig). Similar results were obtained by using this randomization method (S19B and S19C Fig), confirming the tendency of giant cells to cluster more than would be expected by chance. However, while cell shapes produced by this method exhibited less noisy edges, they remained less lobed and more convex compared with those in the original segmentation (S19F Fig). Moreover, this method does not randomize all cells in the tissue, such as stomata. We wondered whether the shape artifacts introduced by the dmSET method could introduce bias into our null model by affecting the contacts between cells in the randomized tissues differently than in the original segmentations. To explore this possibility, we generated “reconstructions” of the original images using the same dmSET method as for the randomizations, except that each cell’s position from the initial segmentation was preserved (with some added uniform noise having a maximum of ±5 μm in the y and x directions) (S18A–S18C, S18E, and S18F Fig). This resulted in cell shape artifacts that were nearly identical between the reconstructions and the randomizations (S18J–S18L Fig), making them more comparable. Indeed, the differences in cell lobeyness (defined as the perimeter of the cell divided by the perimeter of its convex hull) from the original tissues were similar between the reconstructions and randomizations for large cells (S18J Fig). The number of cell neighbors differed slightly from the original segmentations due to the introduced shape artifacts (S18M and S18N Fig), but we found that giant cell contacts remained mostly the same in the reconstructions. Specifically, we observed approximately 6% fewer contacts between giant cells in the reconstructions compared with the segmentations. When comparing reconstructions with randomizations (which are comparable because of their similar cell shapes), the results remained consistent (S18O Fig), with giant cells being more clustered than expected by chance (p = 0.005). In the sepal, cell shape was quite preserved (S15I and S15J Fig; the Pearson coefficient was >0.99 for cell area and perimeter). Sepal cell edges were noisier in the randomized tissues, with a higher cell perimeter (S15J Fig); this could potentially lead to more contacts in the randomized tissues, which would not affect our conclusion. Furthermore, we evaluated our method by using a randomly selected population of pavement cells in both leaves and sepals (S17 Fig). The null hypothesis could not be rejected and this result was robust across five different random patterns in all replicates (one is shown in S17 Fig). Specifically, we obtained p-values ranging from 0.395 to 0.488 in leaves and from 0.178 to 0.335 in sepals. The number of giant cell contacts was either slightly more or less than the expected mean random value, attributable to variability in the random patterns. This indicates that the artificial random pattern did not deviate from true randomness, suggesting that our null models do not present a significant bias. Statistical analysis of the cellular patterns By comparing a spatial observable in the cellular network of the actual segmentation with the corresponding observable in the cellular networks of the 400 generated randomized tissues, whether the considered observable is likely to be observed by chance can be statistically tested [36]. Hence, the use of this method on observables measuring distances or contacts between the studied cells allows the assessment of whether the arrangement of the cells within the tissue is random, clustered or dispersed (Fig 5A). To quantify the patterns, a custom-made Python script was used to extract pertinent observables (i.e., spatial quantities) from the cellular network, which used the NetworkX Python library [64]. In this manuscript, we mainly focused on the number of giant cell neighbors per giant cell to quantify the number of local contacts between giant cells. Other observables have been quantified, such as the minimum shortest path between giant cells, and the number of giant cells in a cluster (S16 Fig). When dealing with cropped images, giant cells (or any cell population studied, see Fig 6) at the border of the image were not considered in the analysis. The number of giant cell neighbors was extracted for every giant cell, and the mean number of giant cell neighbors per giant cell across all giant cells was computed within each experimental replicate. Similarly to the methodology described by the authors of the dmSET method [36], the mean value extracted from the segmentation image was compared with the approximated null distribution formed by the 400 mean values extracted from the randomized images. We first performed the analysis on each replicate independently (S21 and S22 Figs). As the cell-size distributions in the different replicates showed similarities across replicates (Figs 5, S21, and S22), replicates were pooled to increase the sample size and statistical power. Six image replicates were used for both the leaf and sepal wild-type (Fig 5), and three replicates were used for the wild-type sepal time-lapsed images (Fig 7). To test the null hypothesis assessing the randomness of the observed metric, a p-value p was obtained as the ratio of the number of random images (defined here as one random image resulting from pooling one random image per replicate) displaying the same or a more extreme value than the one obtained in the segmentation replicates (one-sided test). If the value fell within the null distribution with an associated high p-value (p > 0.05), the null hypothesis could not be rejected, indicating that the observed quantity could likely be expected by chance. If the value fell outside the null distribution, we assigned p < 0.0025, with 0.0025 corresponding to the inverse of the number of random images (400) used to create the null distribution. In addition, the distribution of the number of giant cell neighbors for all giant cells from the pooled experimental replicates was studied, which provided more insights into their spatial organization. This was compared qualitatively with the distribution expected in a random tissue, extracted from the 400 randomized tissues of all replicates. All plots derived from these analyses were performed with Python, with the use of the matplotlib [65] and seaborn packages [66]. Mathematical model for giant cell fate commitment and numerical simulations To simulate the giant cell fate decisions, our published stochastic and cell-autonomous multicellular model in a growing tissue was used [12]. In that model, the transcription factor ATML1 stochastically fluctuates and drives the expression of its target LGO. In the simulated growing tissue, cells divide using a timer with some stochasticity. When the timer of a cell reaches a threshold ΘC,S, cells undergo the S-phase, and therefore cells transition from being diploid (2C) to tetraploid (4C). By default, cells that reach a second and higher timer threshold ΘC,D will undergo division. However, those cells that have reached a certain LGO concentration threshold ΘT after undergoing the S-phase, considered to be in the G2 phase, will not divide and are maintained in an endoreduplication cycle, which increases their ploidy, becoming giant cells. To model the dynamics of the concentrations for ATML1 and LGO, and the Timer variable, we use chemical Langevin equations [67], which are differential equations with a corresponding deterministic part, consisting of production, degradation and regulatory terms, followed by a stochastic part modeling thermodynamic fluctuations that contains a square root, whose radicand has the sum of the absolute values of the production, degradation, and regulatory terms. The dynamics of ATML1 concentration, LGO concentration and Timer variable in cell i, denoted by [ATML1]i, [Target]i and Timeri, respectively, follow the Langevin equations given by (1)(2)(3) Eqn. (1) stands for the rate of change of ATML1 in cell i, and its terms on the right-hand side describe constitutive expression, self-activation (implemented via a Hill function) [68], linear degradation, and the corresponding stochastic term in ATML1; Eqn. (2), stands for the rate of change of the Target in cell i, and its terms on the right-hand side describe ATML1-induced expression of the Target (using also a Hill function), linear degradation, and the corresponding stochastic term in the Target; Eqn. (3) stands for the rate of change of the Timer in cell i, and its terms on the right-hand side are a constitutive production and the corresponding stochastic term. The parameters of the equations are as follows: PX is the basal production rate for the X variable (where X is either A for ATML1, T for Target concentration or C for the Timer variable), VX is the prefactor of the ATML1-dependent production rate for the X variable, KX is the ATML1 concentration at which the ATML1-dependent production rate for the variable X has its half-maximal value, nX is the Hill coefficient, and GX is the linear degradation rate for the X variable. εi(t) is a normalized cell area, εi(t) = E0Ei(t), where E0 is an effective cell area, and Ei(t) is the area of cell i in arbitrary units. ηXi is a random Gaussian variable with zero mean that fulfills〈ηXi(t)ηX′j(t′)〉 = δ(t − t′) δXX′δij, where i and j are cell indices, X and X′ the modeled variables, δXX′ and δij are Kronecker deltas and δ(t − t′) is the Dirac delta function. Upon cell division, the Timer was reset. To implement the resetting, the following rule was applied at each time step: (4) where Ui is a uniform randomly distributed number in the interval [0, 0.5) and ΘC,D is the cell division threshold for the Timer. The multicellular template on which the simulations were run and initial conditions were the same as in Meyer and colleagues (2017). Initial conditions for ATML1 and Target were randomly uniformly distributed in the interval of [0,1) and [0,0.1), respectively. The Timer initial conditions were set in correlation with the cell areas in the initial template with some stochasticity, as performed in Meyer and colleagues (2017). The differences between the used initial conditions were just in the ATML1, Target, and Timer initial cellular values, determined by different random numbers. Tissue growth and division were also implemented as in Meyer and colleagues (2017). The multicellular tissue grows anisotropically, to emulate the patterning process in the sepal. This anisotropic growth was implemented by imposing a displacement of the vertices with respect to the center of mass of the tissue, with a given radial and vertical exponential rate. After each simulation step, dilution effects due to growth in the modeled variables were taken into account. Cells divided using the shortest path rule together with the constraint of having the division plane through the center of mass of the cell. We assumed that molecules are homogeneously distributed within cells, and therefore, upon cell division, sister cells have the same concentration of ATML1 and Target variables at birth, but can have different cell sizes. Numerical simulations were performed with Tissue software [12,13,69], and the integration was performed using an Îto interpretation of the Langevin equations with a Heun algorithm [70]. Integration was performed with a time step dt = 0.1, and simulations were stopped at time 135. Parameter values for the simulations are given in S2 Table. Parameters were chosen such that the wild-type behavior reported in Meyer and colleagues (2017) could be recapitulated. The outcome of the simulation in Fig 7B was displayed using Paraview software [71]. We recently proposed a more detailed model of the ATML1 regulatory network to study how giant cell specification and cell fate maintenance depend on VLCFA [13], which is still a stochastic and cell-autonomous model. Here, however, for the sake of simplicity, and the intention of using a minimal, stochastic, and cell-autonomous phenomenological model, the former ATML1 model was used [12]. Randomizations of the outcomes from the numerical simulations To assess the randomness of the giant cell pattern in the numerical simulations (Fig 7), the same method was employed as that used for the experimental images. Although randomizations of the tissues were performed similarly (see the “Randomization of the experimental images” section above), a Python script was developed to display the output of the simulation as a multi-labeled image, where each cell was colored with a different label. These images could therefore be randomized using the dmSET method [36,37]. To compare the simulated giant cell pattern (Fig 7B) with the giant cell pattern found in the experimental mature sepals (Fig 5B), the output image was cropped using the maximal rectangle in the tissue, and giant cells were also defined by a size threshold, ensuring that all cells with a ploidy of 16C or higher were categorized as giant cells (Fig 7C). The few 8C cells that exceeded this threshold were also considered as giant cells. To study the change in the giant cell spatial pattern over time (Fig 7F), the same simulations were used, but only the first-arising giant cells (i.e., cells that stopped dividing after time t = 55 of the simulations for being committed to endoreduplicate) were studied. The same method was used to assess the randomness of the cellular patterns on these giant cells both at time t = 55 and time t = 135. Here, instead of cropping the image, the whole tissue was randomized (using the dmSET method), including the cells at the edges, such that exactly the same giant cells were considered at both time points. Examples of randomizations are shown in S20A and S20B Fig. The analysis was performed over five simulation replicates, with different cellular random initial conditions. Related “Segmentation” and “Randomization” images appearing in Figs 7 and S20 were produced with a Python script using the multi-labeled images generated by dmSET. Supporting information S1 Fig. Cauline leaves have elongated giant cells similar to sepals. (A) Abaxial side of a wild-type mature sepal expressing a cell membrane marker (p35S::mCitrine-RCI2A). (B) Tip section of the abaxial side of a wild-type cauline leaf expressing a cell membrane marker (p35S::mCitrine-RCI2A). (C) A developing abaxial side of a wild-type rosette leaf 1 or 2 at 8 dpg expressing a cell membrane marker (p35S::mCitrine-RCI2A). Scale bars associated with the overview images (bottom) represent 200 µm, and scale bars associated with the magnified images represent 100 µm. The underlying data for this figure can be found at Open Science Framework (osf.io), https://doi.org/10.17605/OSF.IO/RFCWS. https://doi.org/10.1371/journal.pbio.3003469.s001 (TIF) S2 Fig. Replicates of abaxial and adaxial cell-size distribution in the wild-type leaf and sepal epidermis; cell size correlates with DNA content. Cell area heat maps (in µm2) of (A, C) abaxial surfaces of wild-type sepals, (B, D) adaxial surfaces of wild-type sepals, (E) abaxial surface of 25-dpg wild-type leaf 1 or 2 (cell density: 284 cells mm−2) and (F) adaxial surface of 25-dpg wild-type leaf 1 or 2 (cell density: 177 cells mm−2). Scale bars represent 100 µm. (G) Abaxial and adaxial side of 25-dpg leaf cell area versus DNA content (one of two replicates) as measured by H2B-TFP total nuclear fluorescence, with R2 = 0.91 for the abaxial side and R2 = 0.79 for the adaxial side. Associated with Fig 2. (H) Cell area of the largest cells (area >4,308 µm2) versus DNA content as measured by H2B-TFP total nuclear fluorescence in both the abaxial and adaxial side of the 25-dpg leaf (red) and in the abaxial side of the adult sepal (blue). The underlying data for this figure can be found at Open Science Framework (osf.io), https://doi.org/10.17605/OSF.IO/RFCWS. https://doi.org/10.1371/journal.pbio.3003469.s002 (TIF) S3 Fig. Cell-size patterning occurs as a basipetal wave simultaneously in the adaxial and abaxial sides of the leaf. (A–B) Cell area heat maps (in µm2) for wild-type leaf 1 or 2 leaves at 5–9 dpg on (A) the abaxial side of the leaf and on (B) the adaxial side of the same leaf. Scale bars represent 100 µm. Leaves are to scale and have the same heat map color range. Associated with Fig 3. The underlying data for this figure can be found at Open Science Framework (osf.io), https://doi.org/10.17605/OSF.IO/RFCWS. https://doi.org/10.1371/journal.pbio.3003469.s003 (TIF) S4 Fig. The sepal giant cell specification pathway also patterns cell size in 9-dpg leaves, replicate 2. Cell area heat maps (in µm2) of the upper abaxial quadrant of leaf 1 or 2 at 9 dpg for the genotypes (A) wild type, (B) acr4-2, (C) atml1-3, (D) dek1-4, (E) lgo-2, (F) LGO-OX (pATML1::LGO), and (G) ATML1-OX (pPDF1::ATML1). Scale bar represents 100 µm. Second replicate associated with Fig 4. (H) 2D Wasserstein distance plot for 9-dpg replicates. Cell area heat maps of other replicates are shown in Figs 4 and S5. The Wasserstein statistical tests among replicates are shown in S8 Fig. The underlying data for this figure can be found at Open Science Framework (osf.io), https://doi.org/10.17605/OSF.IO/RFCWS. https://doi.org/10.1371/journal.pbio.3003469.s004 (TIF) S5 Fig. The sepal giant cell specification pathway also patterns cell size in 9-dpg leaves, replicate 3. Cell area heat maps (in µm2) of the upper abaxial quadrant of leaf 1 or 2 at 9 dpg for the genotypes (A) wild type, (B) acr4-2, (C) atml1-3, (D) dek1-4, (E) lgo-2, (F) LGO-OX (pATML1::LGO), and (G) ATML1-OX (pPDF1::ATML1). Scale bar represents 100 µm. Third replicate associated with Fig 4. The underlying data for this figure can be found at Open Science Framework (osf.io), https://doi.org/10.17605/OSF.IO/RFCWS. https://doi.org/10.1371/journal.pbio.3003469.s005 (TIF) S6 Fig. The sepal giant cell specification pathway also patterns cell size in 25-dpg mature leaves, replicate 2. Cell area heat maps (in µm2) of a region approximately midway between midrib and margin and between tip and base on the abaxial side of leaf 1 or 2 at 25 dpg for the genotypes (A) wild type, (B) acr4-2, (C) atml1-3, (D) dek1-4, (E) lgo-2, (F) LGO-OX (pATML1::LGO), and (G) ATML1-OX (pPDF1::ATML1). Scale bar represents 100 µm. Second replicate associated with Fig 4. (H) Wasserstein distances for 25-dpg replicates displayed as Euclidean distances embedded in 2D. Cell area heat maps of other replicates are shown in Figs 4 and S7. Datasets from (E) and (F) are also used for an independent analysis in Trozzi and colleagues (2023). The underlying data for this figure can be found at Open Science Framework (osf.io), https://doi.org/10.17605/OSF.IO/RFCWS. https://doi.org/10.1371/journal.pbio.3003469.s006 (TIF) S7 Fig. The sepal giant cell specification pathway also patterns cell size in 25-dpg mature leaves, replicate 3. Cell area heat maps (in µm2) of an area approximately midway between midrib and margin and between tip and base on the abaxial side of leaf 1 or 2 at 25 dpg for the genotypes (A, B) wild type (two replicates), (C) acr4-2, (D) atml1-3, (E) dek1-4, (F) lgo-2, (G) LGO-OX (pATML1::LGO), and (H) ATML1-OX (pPDF1::ATML1). Scale bar is 100 µm. Third replicate associated with Fig 4. Datasets from (A), (B), (F), and (G) are also used for an independent analysis in Trozzi and colleagues (2023). The underlying data for this figure can be found at Open Science Framework (osf.io), https://doi.org/10.17605/OSF.IO/RFCWS. https://doi.org/10.1371/journal.pbio.3003469.s007 (TIF) S8 Fig. Statistical tests on the cell-size distributions and statistical tests in young and mature leaves. (A, B) Violin plots of cell area densities on a log10 scale for individual replicates of (A) 9-dpg and (B) 25-dpg leaves. Stomata were removed in both (A) and (B). Associated with Fig 4C and 4D. (C, D) p-values of the Wasserstein tests for all the replicate pair comparisons for (C) 9-dpg and (D) 25-dpg leaves. (E, F) Wasserstein test statistics plotted against their corresponding Euclidean distances for all (E) 9-dpg and (F) 25-dpg test pairs. Associated with Fig 4. The underlying data for this figure can be found at Open Science Framework (osf.io), https://doi.org/10.17605/OSF.IO/RFCWS. https://doi.org/10.1371/journal.pbio.3003469.s008 (TIF) S9 Fig. Classification of cell types in the leaf and the sepal. (A–F) Steps for cell-type classification, with a wild-type sepal (A–C) and a wild-type leaf 25 dpg (D–F) as examples. First, cellular shape features were computed from the segmented meshes in MorphoGraphX (A, D) to train a classification algorithm (Support Vector Machine [SVM] quadratic) and predict the stomata (blue) and the pavement (green) cell types (B, E). Heat map colors represent the cell area in (A, D). Second, to classify giant cells (C, F), a size threshold was established based on observations in atml1-3 sepals, such as a small fraction of pavement cells (0.7% on average) are giant cells in atml1-3 mutants (see G, H). Heat map colors represent cell type in (B, C, E, F). (G, H) Cell size in three wild-type replicates and the three atml1-3 replicates that were used to construct the threshold in leaf 25 dpg (G) and in the sepal (H). Points are colored according to the cell-type classification. Cells exceeding the defined size threshold (see Materials and methods) are defined as giant cells (magenta). Associated with S10 Fig. The code and data associated with this figure can be found at Open Science Framework (osf.io), https://doi.org/10.17605/OSF.IO/RFCWS. https://doi.org/10.1371/journal.pbio.3003469.s009 (TIF) S10 Fig. Cell type classification in the leaf and sepal. (A, B) Segmented meshes of one replicate for each genotype after cell type classification in the mature sepal (A) and the mature leaf (B). Cells are colored with their corresponding cell type: pavement cells (in green), stomata (in blue) and giant cells (in magenta). Stomata and pavement cells were first classified using a trained classification algorithm based on cell shape features. Giant cells were defined as the largest cells, using a size threshold based on atml1-3 mutants (see Materials and methods). Scale bars represent 200 µm. See also S9 and S11 Figs. The code and data associated with this figure can be found at Open Science Framework (osf.io), https://doi.org/10.17605/OSF.IO/RFCWS. https://doi.org/10.1371/journal.pbio.3003469.s010 (TIF) S11 Fig. Cell type classification in the 9-dpg leaves. Meshes of one replicate for each genotype in the 9-dpg leaf sections. Cells are colored with their corresponding cell type: pavement cells (in green), stomata (in blue) and giant cells (in magenta). Stomata and pavement cells were first classified using a trained SVM classification algorithm based on cell shape features. Because meristemoids and stomata are difficult to distinguish at this stage, both were classified as stomata. Giant cells were defined by a cell-size threshold based on atml1-3 mutants (see Materials and methods and S9 Fig). Scale bar represents 200 µm. Associated with S10 Fig. The code and data associated with this figure can be found at Open Science Framework (osf.io), https://doi.org/10.17605/OSF.IO/RFCWS. https://doi.org/10.1371/journal.pbio.3003469.s011 (TIF) S12 Fig. Cell-size patterning has little effect on leaf size, except in ATML1-OX and LGO-OX. (A–G) Images of 9 dpg full leaves with p35S::mCitrine-RCI2A used for segmenting the upper quadrant (Fig 4). One replicate of each genotype is shown. Leaves of similar sizes were chosen to developmentally stage match as closely as possible. (H) Leaf areas of three replicates of 9-dpg leaves for each genotype. (I–O) Images of 25 dpg full leaves with p35S::mCitrine-RCI2A used for imaging and segmenting the mature cells (Fig 4). One replicate of each genotype is shown. (P) Leaf areas of three replicates of 25-dpg leaves for each genotype. Scale bar is 0.5 mm for (A–G) and 1 mm for (I–O). The underlying data for this figure can be found at Open Science Framework (osf.io), https://doi.org/10.17605/OSF.IO/RFCWS. https://doi.org/10.1371/journal.pbio.3003469.s012 (TIF) S13 Fig. Giant cells are not preferentially positioned overlying the vasculature. (A) Top: Cell area heat map of the abaxial upper quadrant of a 9-dpg wild-type leaf. Large cells overlying the midrib that extend up to the leaf tip are boxed in white. Large cells that align as if along an underlying vein are circled in white. Bottom: Cell area heat map of the abaxial midrib region of a 9-dpg wild-type leaf. A large pavement cell extending out from a large midrib cell as if following an underlying vein peeling off the midrib. (B) Cell area heat map of the abaxial side of half of a 9-dpg wild-type leaf with the underlying vasculature in white. The colored heat map is associated with the color bar in (A). (C–F) Cell area heat map of the abaxial sides of halves of ATML1-OX leaves with the underlying vasculature in white (four replicates; see Materials and methods). Colored heat maps are associated with the color bar in (F). All scale bars represent 100 µm. The underlying data for this figure can be found at Open Science Framework (osf.io), https://doi.org/10.17605/OSF.IO/RFCWS. https://doi.org/10.1371/journal.pbio.3003469.s013 (TIF) S14 Fig. Examples of randomizations in the leaf and in the sepal. (A) One replicate of the leaf after cell type classification and three corresponding randomized tissues as an example. (B) One replicate of the sepal after cell type classification and three corresponding randomized tissues as an example. Associated with Fig 5. The code and data associated with this figure can be found at Open Science Framework (osf.io), https://doi.org/10.17605/OSF.IO/RFCWS. https://doi.org/10.1371/journal.pbio.3003469.s014 (TIF) S15 Fig. Conservation of cellular features between the segmented tissue and the corresponding randomized tissues. (A) Example of a segmented tissue and a corresponding randomized tissue, generated using the dmSET method, in the leaf 25 dpg (the replicate used is the same as in Fig 5B). (B, C) Comparison of the cell center coordinates (in µm) between the real tissue (segmentation) and a randomized tissue (dmSET randomization) shown in (A). (D, E) Comparison of cell shape features between all real tissues (segmentation) and one of their randomized tissues (dmSET randomization): (D) cell area (in µm2) and (E) cell lobeyness (defined as the perimeter of the cell divided by the perimeter of its convex hull). (F) Example of a segmented tissue and a corresponding randomized tissue, generated using the dmSET method, in the sepal (the replicate used is the same as in Fig 5B). (G, H) Comparison of the cell center coordinates (in µm) between the real tissue (segmentation) and a randomized tissue (dmSET randomization) shown in (F). (I, J) cell shape features between all real tissues (segmentation) and one of their randomized tissues (dmSET randomization): cell area (I) and cell perimeter (J). The color bars associated with (B–C), (D–E), (G–H), and (I–J) represent the cell areas (in µm2). Each dot represents one cell. Associated with Fig 5. The code and data associated with this figure can be found at Open Science Framework (osf.io), https://doi.org/10.17605/OSF.IO/RFCWS. https://doi.org/10.1371/journal.pbio.3003469.s015 (TIF) S16 Fig. Different spatial observables were studied to statistically assess the spatial organization of giant cells. Three measures were extracted from the cellular network of the segmentations (red) and statistically compared against the randomizations (gray). (A–D) The number of giant cell neighbors per giant, as a mean over giant cells (A, C) and as a distribution (B, D), as presented in Fig 5C and 5D, respectively. (E–H) The minimum shortest path (min. SP) between two giant cells (in path length, 1 meaning that the pair of giant cells are in contact) as a mean over giant cells (E, G) and as a distribution (F, H). (I, K) Fraction of giant cells in contact and (J, L) number of giant cells per giant cell cluster as a distribution. The top panels (A, B, E, F, I, J) show the result of the pattern quantification in the wild-type leaf 25 dpg and the bottom panels (C, D, G, H, K, L) show the results in the wild-type sepal. Analyses were performed over six pooled replicates. Total number of giant cells counted (excluding giant cells at the image border) in the analysis: n = 68 (leaf, segmentations), n = 68 × 400 (leaf, randomizations), n = 74 (sepal, segmentations), n = 74 × 400 (sepal, randomizations). The code and data associated with this figure can be found at Open Science Framework (osf.io), https://doi.org/10.17605/OSF.IO/RFCWS. https://doi.org/10.1371/journal.pbio.3003469.s016 (TIF) S17 Fig. Assessment of the dmSET method on a random cell pattern. (A, B) To identify potential biases in our null model due to artifacts in randomized tissues, an artificial random cellular pattern was created in all leaf replicates by randomly selecting a population of pavement cells with an area larger than 2,000 µm2 in leaves. (A) Example of representative segmentation of a 25-dpg wild-type leaf (left) and one of its corresponding randomized tissue (right). Randomly selected cells are labeled in magenta. (B) Mean number of cell neighbors per cell within the randomly selected cells in the real tissues (segmentations) and in the randomized tissues (randomizations). The null hypothesis could not be rejected, showing that the artificial random pattern does not deviate significantly from randomness (see Materials and methods). (C, D) Similar to (A–B) in the sepals by randomly selecting a population of pavement cells with an area larger than 100 µm2. Total number of cells considered in the analysis: n = 10 × 6 (leaf, segmentations), n = 10 × 6 × 400 (leaf, randomizations), n = 30 × 6 (sepal, segmentations), n = 30 × 6 × 400 (sepal, randomizations). The code and data associated with this figure can be found at Open Science Framework (osf.io), https://doi.org/10.17605/OSF.IO/RFCWS. https://doi.org/10.1371/journal.pbio.3003469.s017 (TIF) S18 Fig. Reconstruction of the tissues using dmSET to investigate the effects of shape artifacts in randomized tissues. (A–C) Example of (A) an original segmentation of a 25-dpg wild-type leaf, (B) a reconstruction, and (C) one randomized tissue. All leaf replicates were reconstructed as shown in (C) by using the same dmSET used for the generation of randomizations (see Materials and methods), but constraining each cell’s position to its original location plus a small amount of noise. (D) Average cell neighboring areas versus cell areas in real tissues and in randomizations. (E) Cell positions (in µm) in the reconstruction of all replicates are nearly equal to segmentations. (F) Cell positions (in µm) in one randomization of all replicates are randomly shuffled. (G) Relative error on cell areas between the segmentation (or reconstruction) and the randomizations. (H) Cell areas (in µm2) in reconstructions versus segmentation. (I) Cell areas (in µm2) in randomizations versus segmentation. (G–I) show that cell areas are similar in segmentations, reconstructions and in the randomized tissues. (J) Relative error on cell lobeyness between the segmentation (or reconstruction) and the randomizations. (K) Cell lobeyness in reconstructions versus segmentation. (L) Cell lobeyness in randomizations versus segmentation. The color bar in (E) denoting cell areas is also associated with panels (F, H, I, K, L, and N). (J–L) show that cell shapes are affected in a similar manner in reconstructions and original segmentations for cells larger than 10,000 µm2. (M) Number of neighbors as a function of cell areas in reconstructions and in segmentations are similar. (N) Number of neighbors in reconstructions versus segmentation. (O) Mean number of giant cell neighbors per giant cell (as shown in Fig 5C) with the comparison with the tissues “reconstructions”. (M–O) show that the cell connectivity is affected in the reconstruction due to the shape artifacts, but that giant cell contacts are not significantly different and the results remain the same when comparing the reconstructions with the randomizations (see Materials and methods). The six replicates were compared together with one corresponding randomized tissue in (D–N). ρ indicates Pearson correlation coefficient in (E, F, H, I, K, L, and N). Two-sample t tests were performed in (G, J, M) at each interval: p < 0.05 (*), p < 0.01 (**), p < 0.001 (***). The code and data associated with this figure can be found at Open Science Framework (osf.io), https://doi.org/10.17605/OSF.IO/RFCWS. https://doi.org/10.1371/journal.pbio.3003469.s018 (TIF) S19 Fig. Giant cells are more clustered than in a randomized null model using the Cut and Merge Cells method. (A) Overview of the alternative method referred to as Cut and Merge Cells (CMC) to generate random giant cell patterns (see Materials and methods). Each leaf replicate was over-segmented to create templates made of small pieces of pavement cells. These templates were then used to automatically generate a random pattern of giant cells by preserving their sizes and numbers (see Materials and methods). (B) Example of one initial segmentation of a 25-dpg leaf (left) and one corresponding CMC randomization (right). (C, D) Quantification of the giant cell patterns to compare with Fig 5C and 5D. (C) Mean number of giant cell neighbors per giant cell in the real tissues (segmentation) and in their randomizations (in gray). (D) Distributions of the number of giant cell neighbors for all giant cells found in all replicates of segmentations (in red) and randomizations (in gray). (E) Left: Comparison of cell areas of giant cells in the dmSET randomizations and in the segmentations. Right: Comparison of cell areas of giant cells in the CMC randomizations and in the segmentations. (F) Left: Comparison of cell lobeyness (see Materials and methods) of giant cells in the dmSET randomizations and in the segmentations. Right: Comparison of cell lobeyness of giant cells in the CMC randomizations and in the segmentations. Six replicates and one randomization per replicate were considered in (E, F). The color bars in (E, F) denote cell areas in µm2. The code and data associated with this figure can be found at Open Science Framework (osf.io), https://doi.org/10.17605/OSF.IO/RFCWS. https://doi.org/10.1371/journal.pbio.3003469.s019 (TIF) S20 Fig. Examples of randomizations at initial and final time points in simulations and in time-lapse imaging sepal data. One replicate is shown on the left and three corresponding randomized tissues on the right as an example. (A, B) Numerical simulation at time t = 55 (initial time) (A) and at time t = 135 (final time) (B). (C, D) Sepal from time-lapse imaging data at two different stages called here initial time (C) and final time (D). Associated with Fig 7. The real sepals in (C) and (D) were also used for an independent analysis in Hervieux and colleagues (2016). The code and data associated with this figure can be found at Open Science Framework (osf.io), https://doi.org/10.17605/OSF.IO/RFCWS. https://doi.org/10.1371/journal.pbio.3003469.s020 (TIF) S21 Fig. Quantification of the giant cell pattern in the individual replicates of the leaf. On the left, each replicate is displayed after the cell type classification. In the middle: the corresponding results of the cellular pattern quantification showing the mean giant number of neighbors per giant, where the tissue segmentations (in red) were statistically compared with the tissue randomizations, containing 400 randomized tissues per replicate (in gray). On the right: the corresponding distributions of the number of giant cell neighbors per giant cell, both for the actual tissue (red) and the randomizations (gray). Associated with Fig 5. The code and data associated with this figure can be found at Open Science Framework (osf.io), https://doi.org/10.17605/OSF.IO/RFCWS. https://doi.org/10.1371/journal.pbio.3003469.s021 (TIF) S22 Fig. Quantification of the giant cell pattern in the individual replicates of the sepal. On the left, each replicate is displayed (rotated 90°) after the cell type classification. In the middle: the corresponding results of the cellular pattern quantification showing the mean giant number of neighbors per giant, where the tissue segmentations (in red) are statistically compared with the randomizations, containing 400 randomized tissues per replicate (in gray). On the right: the corresponding distributions of the number of giant cell neighbors per giant cell, both for the actual tissue (red) and the randomizations (gray). Associated with Fig 5. The code and data associated with this figure can be found at Open Science Framework (osf.io), https://doi.org/10.17605/OSF.IO/RFCWS. https://doi.org/10.1371/journal.pbio.3003469.s022 (TIF) S1 Table. Primers for genotyping mutants and overexpression lines. https://doi.org/10.1371/journal.pbio.3003469.s023 (PDF) S2 Table. Parameter values used for the simulations. Parameters used for the simulations shown in Figs 7 and S20. All units are arbitrary. https://doi.org/10.1371/journal.pbio.3003469.s024 (PDF) Acknowledgments We thank Nicholas J. Russell, John Chandler, Josep Mercadal, Elise Laruelle, and Philippe Andrey for critical comments on the manuscript. We thank Stephen Starkman, Emily Phung, and Isabel Delo for assistance with segmentation, and Violeta Gibelli for assistance with cell classification corrections. We thank Richard Smith and Soeren Strauss for their help in using MorphoGraphX; we are grateful that Richard Smith created a process to generate 2D images from meshes. We also thank Philippe Andrey for his insights on image analysis and statistics and Elise Laruelle for her help in using the dmSET randomization code, as well as the Cornell Statistics Consulting Unit, specifically May Boggess, for her help with statistics and coding.