PLoS Biology

doi: 10.1371/journal.pbio.1001930pmid: 25116939

Introduction Meiosis is a modified cell cycle where two rounds of chromosome segregation follow a single S phase, resulting in the production of haploid gametes. Recombination is a key step in meiosis I, as it results in genetic crossover (CO) formation, which establishes physical links between the homologues cytologically visible as chiasmata [1],[2]. In most species, each chromosome pair has at least one CO (referred to as the obligatory CO), which is required to hold the homologues together during the first meiotic division, ensuring their correct segregation. In most organisms, homologues that lack a CO often segregate improperly, leading to the formation of aneuploid gametes [3]. Meiotic recombination can also lead to gene conversion not associated with COs (NCOs) [4]. Meiotic recombination is initiated by the induction of DNA double-strand breaks (DSBs) catalysed by SPO11 [5]. DSBs are then resected by exonucleases to generate 3′ single-stranded DNA molecules (ssDNA). In the subsequent step, RecA homologues RAD51 and DMC1 assemble on the ssDNA to form nucleoprotein filaments. These filaments search for homologous sequences and trigger single-strand invasions [6] to generate displacement loop (D-loop) recombination intermediates [7]. Depending on the way these D-loop intermediates are processed, different recombination products can be formed. For example, capture of the second DSB end leads to the formation of a double Holliday junction that can be resolved to generate either a non-CO (NCO) or a CO [8]–[10]. Alternatively, NCOs can also be formed when a single strand end is displaced after priming a limited amount of DNA synthesis, annealing with the other DSB end in a process called synthesis-dependent strand annealing (SDSA) [11]. In most organisms, when multiple COs occur on the same chromosome, they are distributed nonrandomly: One CO prevents other COs from occurring close by, in a distance-dependent manner. This phenomenon results in COs being more evenly spaced along chromosomes than would be expected if they occurred randomly. The term used to describe this phenomenon is CO interference [12],[13]. In budding yeast, two kinds of COs are known to coexist: class I COs, which are interference-sensitive COs and whose formation depends on the ZMM proteins (Zip1, Zip2, Zip3, Zip4, Msh4, Msh5 and Mer3) in addition to Mlh1 and Mlh3, and class II COs, which are not subject to interference and depend on Mus81 and Eme1/Mms4 [10]. Arabidopsis thaliana, like yeast and mammals, has two recombination pathways: one that exhibits CO interference and another one that does not [14]–[20]. In A. thaliana, disruption of genes acting in the interference-sensitive pathway causes a loss of approximately 85% of COs [21]. In addition, there is evidence that the MUS81 gene accounts for some, but not all, of the 15% MSH4-independent COs, suggesting that MUS81 is involved in a secondary subset of meiotic COs that are interference insensitive [14],[22]. Very little information is available on the mechanisms controlling interference and the number and distribution of COs during meiosis in general [23],[24]. Eukaryotes possess a highly conserved mechanism to control protein degradation mediated by the action of the ubiquitin (Ub) proteasome system (UPS) [25]. In this system, E3 Ub ligases are required to ubiquitylate specific protein targets. Cullin RING ligases (CRLs) are the largest class of E3 ligases. Several mechanisms control CRL activity: It can be activated by covalent attachment of the Ub-like protein NEDD8/RUB (a process called neddylation or rubylation) [26],[27] or inhibited by the COP9 signalosome-directed deneddylation [28]. Neddylation/rubylation has been shown to play a crucial role in processes such as morphogenesis in mice [29], cell division in budding yeast [30], embryogenesis in C. elegans [31], meiosis to mitosis transition in C. elegans [32], and response to various plant hormones [33],[34] including auxin [35]–[38]. However, neddylation/rubylation had not been connected to homologous recombination (HR). Cullin RING Ligase 4 (CRL4) is associated with DNA repair in plants and humans; the DDB1-CUL4ADDB2 E3 ligase initiates nucleotide excision repair (NER) by recognizing damaged chromatin with concomitant ubiquitylation of core histones at the lesion site [39]–[41]. Additionally, CUL4A plays a role in meiotic recombination and spermatogenesis in mice [42],[43]. Inactivation of cul4a affected male fertility, with increased death of pachytene/diplotene cells and defects in MLH1 dissociation from the SCs. Here we show that the E1 enzyme of the neddylation complex, AXR1, is a major regulator of meiotic recombination in Arabidopsis. In axr1 mutants, the average number of meiotic COs is unchanged; they are still under the control of the ZMM proteins, but they tend to cluster together and no longer follow the obligatory CO rule. We were able to show that this recombination defect is correlated with strong synapsis defects. In addition, we found that this deregulation of CO localisation is likely mediated by a CRL4. Results The axr1 Mutants Are Meiosis-Defective In the process of screening A. thaliana T-DNA (Agrobacterium tumefaciens transferred DNA) insertional lines for meiotic defects, we isolated three mutants [EGS344, EIC174, and EVM8 (Ws-4 strain); Figure 1 and Figure S1] allelic for disruption in At1g05180, the AXR1 gene, previously shown to encode the E1 enzyme of the Arabidopsis neddylation complex [44]. Another insertion line in At1g05180 available in the public collection (http://signal.salk.edu/) Sail_904E06 (N877898, Col-0 strain) and the historical axr1 allele (axr1-12/N3076, Col-0 ecotype [44], with a single nucleotide substitution in exon 11 of At1g05180) were also included in this study (Figure 1). Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 1. The AXR1 gene and axr1 mutations. The arrow indicates the orientation of the open reading frame. Exons are shown as boxes (pink, UTR; black, CDS). In the EGS344 mutant, a large deletion associated with an insertion of the exogenous Agrobaterium Ti plasmid disrupts the AXR1 gene from nucleotide 91 (40 bp 5′ to the ATG). In the EVM8 mutant, an in-frame deletion of 312 bp between exons 3 and 4 generates a 20 aa truncated protein. In EIC174, a single nucleotide insertion (A) in exon 6 (position 1364 of the genomic sequence, corresponding to nt 688 in the cDNA) leads to a premature stop codon (a 222 aa protein is produced instead of the 540 aa protein in wild type). In axr1-12, corresponding to the N3076 line, a single C-T nucleotide substitution at position 1295 of the cDNA leads to a premature stop codon (415 aa instead of 540), as described by Leyser et al. [44]. In N877898, corresponding to the Sail_904E06 line, a T-DNA insertion occurred in intron 11. References used for this figure are Tair accession 4010763662 for the genomic sequence and Tair accession 4010730885 for the cDNA sequence. https://doi.org/10.1371/journal.pbio.1001930.g001 The mutant plants all show the same vegetative phenotypes as previously described for axr1 mutants: They are dwarfed, excessively branched, with small rosettes and crinkled leaves (shown for N877898 in Figure 2A–B and in Figure S2 for the other alleles) [45],[46]. They also have small flowers and short fruits, indicating fertility defects (Figure S2). Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 2. axr1 developmental defects. Five- (A) or nine- (B) week-old wild-type (wt) or axr1 (N877898) mutant plants. axr1 mutants are dwarfed, strongly branched, and have short siliques. Alexander staining (C–D) reveals round pollen grains, with a red cytoplasm reflecting viable male gametophytes in wild type (C), whereas axr1 (N877898) anthers (D) contain a mixture of viable and dead (uncoloured, arrows) pollen grains. DIC microscopy of male meiosis products (E and F) reveals tetrads of microspores in wild-type (E) and unbalanced tetrads or polyads in axr1 (F, N877898). https://doi.org/10.1371/journal.pbio.1001930.g002 We examined the reproductive development of these mutants and found that all alleles showed a high level of male and female gametophyte abortion [shown for N8779898 male gametophytes (pollen grains) in Figure 2D]. In plants, male gametogenesis occurs in the anthers where groups of meiocytes undergo meiosis synchronously, each producing four haploid cells (called microspores). The four products of each meiosis remain temporally encased in a common callose wall, forming tetrads of microspores that can be visualised after tissue clearing (Figure 2E). Each microspore is then released from its tetrad and continues to develop into a mature pollen grain (the male gametophyte) containing the male gametes. Study of the early stages of pollen development in axr1 revealed the presence of abnormal meiotic products. Instead of the regular tetrahedral structure observed in the wild-type, asymmetric tetrads (containing four daughter cells of unequal size) or “polyads” (containing more than four products) were observed (Figure 2F), suggesting that the meiotic program is disrupted in these mutants. To confirm that the reduced fertility was caused by a defect in meiosis, we investigated male meiosis via chromosome spreading and DAPI (4′,6-diamidino-2-phenylindole) staining (Figure 3). During wild-type meiotic prophase I (Figure 3A–D), DNA fibres of each sister chromatid are organised as chromatin loops connected to a common protein axis (the axial element [AE]) [47]. When chromosomes start to condense at leptotene, they become visible as threads (Figure 3A). At this stage, meiotic recombination is initiated by the formation of a large number of DNA DSBs (not shown). HR repairs these breaks concomitantly with the progression of synapsis, the close association of the homologous chromosome axes through the polymerisation of the central element (CE) of the synaptonemal complex (SC). Synapsis begins at zygotene (not shown) and is complete by pachytene, when complete alignment of homologous pairs can be detected in DAPI-stained chromosomes (Figure 3B). DNA repair and recombination are thought to be achieved during pachytene, yielding at least one CO per homologous chromosome pair. At diplotene (Figure 3C), when the CE of the SC is depolymerised, the homologous chromosomes are therefore connected to each other by COs in which chromatids from homologous chromosomes have been exchanged. These connections between homologous chromosomes become apparent only at diakinesis (Figure 3D, arrows), when chromosomes are sufficiently condensed. At this stage in Arabidopsis, chiasmata (the cytological manifestations of COs) cannot be scored precisely, but chiasma-carrying chromosome arms can sometimes be identified based on bivalent appearance (see Figure 3D, arrows). Next, condensation proceeds and, at metaphase I, the five Arabidopsis bivalents are easily distinguishable, aligned on the metaphase plate (Figure 3E). During anaphase I, sister chromatid cohesion is released from chromosome arms, allowing homologous chromosomes to segregate to the two opposite cellular poles (Figure 3F). The second meiotic division then separates the sister chromatids, generating four pools of five chromosomes (Figure 3G and 3H), which gives rise to the tetrads of four spores (Figure 2E). Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 3. axr1 mutants show normal meiotic progression but reduced bivalent formation at metaphase I. DAPI staining of meiotic chromosomes in wild type (A–H) and axr1 (N877898, I–P). At the onset of meiotic prophase I (A and I), chromosomes can be identified. Chromosome alignment and synapsis then proceeds, leading eventually to the pachytene stage in wild type (B), where homologous chromosomes are synapsed along their entire length. This association can be observed in axr1 (J, enlarged regions) but remains partial. Then, the SC disappears at diplotene (C and K), condensation proceeds, and bivalents can be identified in wild type at diakinesis (D), but this stage is rarely observed in axr1 (L). At metaphase I, the five Arabidopsis bivalents can be identified in wild type (E), segregating at anaphase I (F). In axr1, a mixture of bivalents and univalents are observed (M), leading to subsequent improper segregation at anaphase I (N). Sister chromatids segregate at meiosis II (G and O), leading to balanced tetrads in wild type (H), unbalanced tetrads (not shown) or polyads in axr1 (P). At metaphase I, univalents (u) can be distinguished from ring bivalents (where a chiasma occurred in each of the two chromosome arms, *) and from rod bivalents (where only one chromosome arm shows a chiasma, #). Arrows in (D) indicate some of the chiasma-containing arms. Bar, 10 µm. https://doi.org/10.1371/journal.pbio.1001930.g003 In A. thaliana axr1 mutants, the leptotene and zygotene stages appeared similar to those in the wild type. However, no pachytene cells were identified in the 457 meiocytes analysed, in contrast to wild type, where this stage is present in approximately 35% of the cells (n = 334). Instead, we observed pachytene-like stages, with only partial chromosome alignment (Figure 3J). This suggests that axr1 is defective in synapsis. Diplotene cells were indistinguishable from those in the wild type (Figure 3K). Then, chromosome condensation could be followed until metaphase I, although diakinesis stages were rarely observed (1% of all stage cells, n = 457 for N877898, 12% in wt, n = 334) (Figure 3L). At wild-type metaphase I, the five typical Arabidopsis bivalents could be observed aligned on the metaphase plate (Figure 3E). Each bivalent was composed of two homologous chromosomes connected by chiasmata either on one chromosome arm (rod bivalent, Figure 3E#) or on both pairs of chromosome arms (ring bivalent, Figure 3E*). Chiasma numbers could therefore be estimated based on the bivalent structure. However, because multiple COs on a single arm cannot be cytologically differentiated from single COs, these estimates only correspond to a minimum chiasma number (MCN; Figure 4, Table S1). Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 4. COs in axr1 are largely ZMM-dependent. For each axr1 allele and for their respective wild-type strains (Ws-4 for EGS344, EIC174, EVM8, and Col-0 for N8777898 and N3076), and for a combination of multiple mutants, the level of bivalent formation as well as the MCN per cell were measured. For multiple mutant analyses, the N877898 allele was used in a Col-0 background and EGS344 in a Ws-4 background. The complete dataset can be found in Table S1. https://doi.org/10.1371/journal.pbio.1001930.g004 In axr1 mutants, we observed reduced bivalent formation, and instead of five bivalents, a mixture of bivalents and univalents could be identified (Figure 3M). The reduction in bivalent formation resulted in chromosome mis-segregation during subsequent anaphase I (Figure 3N), whereas the second meiotic division separated sister chromatids (Figure 3O), giving rise to a variable number of daughter cells containing aberrant numbers of chromosomes (Figure 3P). We quantified the decrease in bivalent formation as well as the MCN at metaphase I from all axr1 mutants and their respective wild-type accessions (Figure 4, Table S1). On average, axr1 mutants had 78% of the wild-type number of bivalents for the Col-0 background and 52% for the Ws background. In terms of the chiasma number, axr1 mutants displayed a residual level of 56% and 41% of the wild-type levels for Col-0 and Ws strains, respectively (Figure 4). Within a single ecotype (Col-0 or Ws), all alleles were statistically different from the wild type but not different from each other. Finally, when the partitioning of the residual chiasmata in axr1 was analysed, we observed that a large proportion of metaphase I cells showed both ring bivalents (at least two chiasmata) together with univalents (no chiasma) (42% of the N877898 cells, n = 47), showing that in axr1, the obligatory CO is lost. To further analyse the bivalent shortage observed in axr1, we used fluorescence in situ hybridization (FISH) analyses on PMCs. Metaphase I chromosomes were labelled with probes for the 45S and 5S rDNA repeats, allowing specific identification of chromosomes 1, 2, and 4 (Figure 5). Chromosomes 3 and 5 could not be discriminated from each other with these probes and were pooled. First, we observed that in axr1 as in wild type, bivalents were always formed between homologous chromosomes (n = 147 bivalents for axr1, n = 165 for wt). Then, we considered each bivalent individually and determined which pair of chromosomes was involved in its formation. As shown in Figure 5D, in axr1, as in the wild type, each pair of chromosomes was equally involved in bivalent formation, showing that the decrease in bivalent formation observed in axr1 affected all chromosomes in the same way. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 5. Bivalent shortage has a similar effect on each pair of chromosomes. Fluorescent in situ hybridisation (FISH) on metaphase I cells was performed with probes directed against the 45S (green) and the 5S (red) rDNA, which allow the identification of chromosomes 1 (unlabelled), 2 (green labelled), and 4 (green and red labelled), whereas chromosomes 3 and 5 cannot be distinguished (red labelled). In wild type, each chromosome pair represents 20% of the total number of bivalents (A and D, centre circle, in light, n = 21 cells). In axr1 (B and D, N877898 allele, external circle, n = 28), the proportion of each bivalent pair is the same as in wild type. Bar = 5 µm. https://doi.org/10.1371/journal.pbio.1001930.g005 axr1 COs Are ZMM-Dependent In wild-type Arabidopsis, the majority of COs (85%–90%, depending on the genetic background Col-0 versus Ws-4) depend on the ZMM proteins (MSH4, MSH5, MER3, ZIP4, SHOC1/ZIP2, HEI10, and PTD) as well as on MLH1 and MLH3 [21],[48], whereas MUS81 is responsible for 10%–15% of the remaining COs [14],[22]. We measured bivalent formation frequencies and the chiasma frequencies in various genetic combinations compared to the single axr1 mutant (Figure 4, Table S1). For all the zmmaxr1 double mutants (except mer3axr1) and regardless of strain (Col-0 versus Ws-4), the level of bivalent formation was reduced by more than 95% with hardly any bivalents observed (from 0.13 to 0.18 bivalent per cell; Table S1), showing that almost all the COs in axr1 are ZMM-dependent. We also analysed the bivalent frequency in the axr1mus81 double mutant, which was the same as for the axr1 single mutant (3.77±1.03 against 3.75±1.12; p = 0.9) (Figure 4). We then quantified bivalent frequency in the axr1msh5mus81 triple mutant and observed, as expected, a dramatic decrease in bivalent formation compared to axr1mus81 (Figure 4). No difference could be detected between the axr1msh5mus81 triple mutant and the axr1msh5 double mutant (p = 0.2). These results show that CO formation in axr1 mutants is almost exclusively dependent on ZMM proteins, whereas the MUS81 pathway plays only a limited role, if any. Class I COs Are Mislocalised in the axr1 Mutant To further analyse recombination events in axr1, we immunolabelled chromosomes with antibodies directed against HEI10 and MLH1, two markers of class I COs in Arabidopsis [48],[49]. MLH1 foci can be seen from late pachytene to diakinesis [49], whereas HEI10 is first loaded early during prophase on a large number of sites forming foci of different sizes on chromosomes. A limited number of these foci then remain (Figure 6A and B) at sites that correspond to class I COs where they co-localise with MLH1 until the end of prophase [48]. We therefore counted HEI10 and MLH1 foci in late pachytene and diplotene cells in wild type and axr1. Surprisingly, the average foci number per cell was not different between wild type and axr1, for either HEI10 (8.30±0.29, n = 54 and 7.49±0.40, n = 84, p = 0.15) or MLH1 (8.61±0.29, n = 33 and 7.58±0.54, n = 91, respectively, p = 0.263). In addition, we confirmed that these foci localise to chiasma-containing arms at diakinesis (Figure 6E and F and Figure S3), showing that they are likely to mark CO sites in axr1 as in wild type [49]. We also observed that there was higher variability in the numbers of HEI10 and MLH1 foci in axr1 than in wild type (Figure 6G), with the coefficient of variation (standard deviation divided by the mean) varying from 26% (HEI10, wt) to 50% (HEI10, axr1) or from 19% (MLH1, wt) to 68% (MLH1, axr1). Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 6. The average number of class I COs is similar in wild type and axr1. HEI10 or MLH1 was immunolocalised on acetic acid spread chromosomes from wild-type (A, B, and E, Col-0) or axr1 (C, D, and F) meiocytes from late pachytene to diakinesis. In axr1 (N877898 allele), the average number of HEI10 or MLH1 foci per cell is similar to that in wild type (G). Bar  =  5 µM. https://doi.org/10.1371/journal.pbio.1001930.g006 Another striking feature of axr1 was the frequent occurrence at the pachytene-like and diplotene stages of portions of paired chromosome axes where adjacent HEI10 and MLH1 foci could be seen (Figure 6C, D, arrows and Figure 7A, arrows). Forty-seven percent (HEI10, n = 60) or 53% (MLH1, n = 66) of the cells had at least two foci localised on the same portion of a chromosome axis, whereas in wild type, this scenario occurred only in 7% (HEI10, n = 57) or 3% of the cells (MLH1, n = 39) (Figure 7B). In addition, although we never observed more than two adjacent foci in wild type, we observed 22% (HEI10) and 13% (MLH1) of the cells with more than two adjacent foci, with a maximum of five adjacent HEI10 foci observed in axr1 (Figure 7B). Therefore, although the average level of class I COs is the same in axr1 and in wild type (Figure 6G), these class I COs tend to cluster together in at least 50% of the axr1 cells. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 7. Class I COs tend to cluster in axr1. (A) Examples of adjacent HEI10 or MLH1 foci in wild-type (Wt, Col-0) and in axr1 (N877898 allele) acetic acid spread meiotic chromosomes. (B) Proportion of pachytene and diplotene cells where adjacent foci were observed on the same chromosome axis pair (Wt, wild type; axr1, N877898 allele) (0, no evidence of adjacent foci; 2, two adjacent foci, etc.). Some of these situations are indicated by arrows in panel A. Bar  =  5 µM. https://doi.org/10.1371/journal.pbio.1001930.g007 We then estimated the scale at which this clustering arises. The distance between clustered foci was measured and compared to the total length of chromosome axis. The distance between two adjacent foci was on average 1/400 of the total axis length of a cell, ranging from 1/1600 of the genome to a maximum of 1/90 of the genome (Figure S4A). Extrapolated in DNA distance, with the additional assumption that genome condensation is homogeneous, the distance between two adjacent foci in a cluster is therefore expected to vary from 150 kb to 3,000 kb, with an average of 625 kb. We also observed that the distance between two adjacent foci does not vary significantly in clusters with exactly two foci compared with clusters with more than two foci. As a consequence, cluster size increases proportionally with the number of foci present in the cluster (Figure S4B). The size of the clusters was on average 1/200 of the genome for HEI10 foci (n = 14, 1,200 kb) and 1/300 for MLH1 (n = 21, 800 kb). Finally, we examined whether the clustered foci displayed interference, as might be expected for class I COs. We thus considered the hypothesis H0 that the foci in clusters are not subject to interference. The test was based on the distribution of distances between adjacent foci, specifically using the coefficient of variation for the statistical test and comparing to 105 simulations under H0 (see Materials and Methods). For the clusters of three or more foci (Table S2), we rejected the H0 hypothesis of no interference for MLH1 foci (p = 0.0024 based on seven clusters), for HEI10 foci (p = 0.0028 based on six clusters), and when pooling the MLH1 and HEI10 data (p = 2.4×10−5 based on 13 clusters). Specifically, inside clusters, MLH1 and HEI10 foci are more evenly distributed than at random, showing that COs within clusters still interfere. Taken together, these results show that the shortage in bivalent formation observed in axr1 mutants is not due to a general decrease in CO formation but rather to a mislocalisation of class I COs that tend to cluster together. Measurement of Recombination Rates in axr1 Mutants The level of genetic recombination on several chromosomal intervals was measured using the Fluorescent-Tagged Lines (FTL) tool developed by Copenhaver et al. [50]. The FTL system is a visual assay based on segregation of genetically linked fluorescent proteins expressed in the pollen grains of the quartet mutant (qrt1), in which the pollen grains remain attached as tetrads. With these lines, a large number of meiotic products can be visually scored and then a subset of multiple CO events can be identified (two-, three-, and four-strand double COs in adjacent intervals and four-strand double COs within a single interval) ([50] and Table S3B). Six different intervals were used, either on chromosome 3 (I3b and I3c) or 5 (I5a, I5b, I5c, and I5d), with sizes ranging from 1,200 to 4,900 kb (Table S3A). We first measured recombination rates for each interval using the standard Perkins genetic mapping equation [51]. As shown in Table 1, recombination rates in axr1 vary differently depending on the interval tested, from 70% to 180% of the wild-type level. On average, axr1 shows an increase in recombination, but these data should be taken with caution, as recombination measurements rely only on a subset of tetrads (the viable tetrads). Out of the six intervals considered, intervals located close to the telomeres (I3b and I5b) showed the most significant increase in recombination, whereas proximal intervals appeared less affected. This could indicate that the level of recombination is affected differently according to the location on the chromosomes, although additional data will be required to determine if telomere proximity increases CO frequency in the mutant. Download: PPT PowerPoint slide PNG larger image TIFF original image Table 1. Recombination rates and interinterval interference. https://doi.org/10.1371/journal.pbio.1001930.t001 We then used the FTL data to estimate interference between COs occurring in adjacent intervals (Table 2 and Table S4). We calculated the Interference Ratio (IR) as defined by Malkova et al. [18], which compares the genetic length of one interval with and without the presence of a simultaneous event in the neighbouring interval. When the occurrence of a CO in one interval reduces the probability of a CO occurring in the adjacent interval, the IR is less than 1, indicating (positive) CO interference. When COs in the two adjacent intervals are independent of each other, the IR is 1, and if the presence of one CO in an interval increases the probability of an additional CO in the adjacent interval, the IR is greater than 1, indicating negative interference. As shown in Table 2, all wild-type IRs were less than 1, in agreement with the presence of CO interference. For axr1, however, all IRs increased dramatically and were statistically significantly different to wild type (p<0.0001, Table 2 and Table S4). In addition, all axr1 IR values were greater than 1, although only one pair of intervals tested was significantly different from 1 (I5a I5b, first data set, IR = 1.63, p = 4×10−3). Therefore, in axr1, adjacent COs appear to occur more frequently than in wild type, which is in agreement with the previously observed clustering of class I COs scored cytologically (Figure 7). The cytologically observed clustering is occurring at a very small scale, namely a few hundred kb (on average 1,200 kb for HEI10 foci and 800 kb for MLH1 foci, see above), whereas in FTLs pairs of intervals correspond to more than 3,000 kb (I5cd, I3bc) and up to 7,500 kb (I5ab). Consequently, most of the clusters are expected to be present within a single interval and to only occasionally affect two adjacent intervals, which could explain why only one pair of intervals showed significant negative interference. Download: PPT PowerPoint slide PNG larger image TIFF original image Table 2. Intra-interval interference. https://doi.org/10.1371/journal.pbio.1001930.t002 Double CO events within a single interval can be detected using the FTLs if the two COs involve four different chromatids (Table S3B) because they will generate nonparental ditype (NPD) tetrads [50]. Interference within single intervals can be estimated by comparing the observed number of double COs (NPD frequency) to the expected number of double COs under the hypothesis of no interference [52]. The ratio between these two numbers (NPDr) gives the strength of interference within the considered interval, even if an important proportion of multiple COs will be silent. We calculated NPDr for all intervals considered for wild type and axr1 (Table 1 and Table S5). In wild type, the NPDr indicated strong interference (NPDr close to 0.3) within all the intervals (except for I3c, which is too small for statistically meaningful data, Tables S3A and S5). In axr1, however, the NPDr increased systematically (between 0.7 and 1.47) and was mostly greater than 1. For two intervals (I5a and I5b), the NPDr values of 2.69 and 1.63 were statistically significant (p<0.01), showing negative interference (Table 2). Thus, genetic analyses allowed us to measure negative interference in several of the intervals tested, confirming the CO clustering observed in cytology. Recombination Initiation Is Not Modified in axr1 Mutants To verify whether the recombination defect in axr1 could be linked to a defect in recombination initiation, we used two methods to investigate DSB formation. We first introgressed the axr1 mutation into a rad51 mutant, defective for meiotic DSB repair. In this mutant, DSBs are formed but are then repaired abnormally, leading to significant chromosomal defects (such as chromosome bridges and chromosome fragmentation) during anaphase I (Figure S5A). These chromosomal defects persisted in axr1rad51, showing that DSBs are present in the axr1 mutant (Figure S5B). Second, we analysed the nuclear distribution of the DMC1 protein, a meiosis-specific recombinase that forms foci at recombination sites. The dynamics and number of AtDMC1 foci in axr1 (237±40, n = 7) were indistinguishable from wild type (234±89, n = 28) (t, p = 0.9) (Figure S5). Thus, the meiotic defects observed in axr1 are not correlated with a decrease in the amount of recombination initiation events. Synapsis Is Strongly Defective in axr1 But Chromosome Axes Are Normal During meiotic prophase, chromosomes are structured in the context of a protein axis (the AE), which is crucial for most meiotic events, including meiotic recombination and synapsis [53],[54]. The meiotic chromosome axis is composed of specific AE proteins, such as ASY1 and cohesion proteins (REC8 and SCC3, [55],[56]). In wild-type meiotic cells, cohesins are loaded as early as premeiotic G1, whereas ASY1 appears at leptotene first as foci, then as a linear signal throughout the entire chromosome length (Figure S6A), in a pattern similar to that of cohesins (Figure S6C, [56]). As shown in Figure S6, the signal observed in axr1 mutants cannot be differentiated from wild type, showing that no major alteration of the axis can be detected in axr1 mutants. We then analysed the progression of synapsis by immunolocalisation of ZYP1, the A. thaliana CE component [57]. In wild type, ZYP1 appeared on chromosomes as foci that quickly elongated to yield a mixture of foci and short stretches of ZYP1 (Figure 8A,B, red signal and Figure S7). Synapsis then progressed until complete synapsis was reached, defining the pachytene stage (Figure 8C,D and Figure S7). In axr1, the early stages of synapsis could not be distinguished from wild type, showing a mix of foci and short ZYP1 stretches (Figure 8E,I and Figure S7). As meiosis progressed, ZYP1 elongation could be detected (Figure 8F–L and Figure S7), but full synapsis was never achieved (n = 66), confirming the synapsis defect detected after DAPI staining of meiocyte spreads (Figure 3). In addition, in approximately half of the cells, ZYP1 signals appeared strongly perturbed, uneven in thickness and forming dotted lines rather than a homogeneous continuous signal (Figure 8J or G and Figure S7). In some cases, only short and thick ZYP1 stretches were detected. These could correspond to ZYP1 poly-complexes rather than to CE polymerisation (Figure 8L and Figure S7). Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 8. Synapsis is strongly perturbed in axr1, but HEI10 dynamics during early prophase are unchanged. ZYP1 and HEI10 proteins were co-immunolocalised on lipsol-spread chromosomes from wild-type (A–D) and axr1 (N877989 allele, E–L) meiotic cells. The overlay of both signals is shown here (ZYP1 in red, HEI10 in green), but single channels can be found in Figures S7 and S8. In wild type as in axr1, ZYP1 appears on chromosomes as foci that quickly elongate, yielding a mixture of foci and short stretches (A, B, E, F, and I). Synapsis then progresses until complete synapsis is reached in wild type, defining the pachytene stage (C–D). In axr1, ZYP1 elongation can be detected, but full synapsis was never achieved (G, H, and K). In axr1, the ZYP1 signal is often uneven in thickness or forms dotted lines rather than a homogeneous and continuous signal (J and G). In addition, in some cases, only short and thick ZYP1 stretches were detected which could correspond to ZYP1 polycomplexes (L). During early zygotene, in wild type as in axr1, HEI10 forms numerous foci of variable sizes on chromatin (A, E, and I). Then, although synapsis progresses, combinations of large and small foci are observed, forming “strings of pearls” on ZYP1 stretches (B, F, and J, arrows). As meiosis progresses, a few bigger and brighter HEI10 foci can be observed in wild type (D) and in axr1 (G, H, K, and L), which generally co-exist with smaller and fainter HEI1O foci (C, D, G, H, and K). Whereas this latter HEI10 pattern is associated with complete synapsis in wild type (C–D), synapsis is only partial in axr1 (G, H, K, and L). Bar = 2 µm. https://doi.org/10.1371/journal.pbio.1001930.g008 CO maturation and Synapsis Can Be Uncoupled in axr1 To follow the progression of meiotic recombination events, we co-immunolocalised ZYP1 and HEI10, using a lipsol spreading protocol that has the advantage of allowing the simultaneous detection of these two proteins [58] but also the disadvantage of preventing examination of prophase after pachytene [59]. As mentioned above, HEI10 is detected as foci on meiotic chromosomes from leptotene to diakinesis, and its dynamics reflect the progression from early recombination intermediates to mature class I COs [48]. During leptotene and early zygotene, HEI10 forms numerous foci of variable size on chromatin (Figure 8A and Figure S8). Then, during synapsis initiation, bigger and brighter HEI10 foci appear, often co-localising with synapsed regions (Figure 8B and Figure S8). At this stage and later on, a combination of large and small foci are observed, forming “strings of HEI10 pearls” on ZYP1 stretches (Figure 8B,C and Figure S8B,C, arrows). At late pachytene, only a few bright HEI10 foci, corresponding to mature class I COs, are retained (Figure 8D and Figure S8). Nevertheless, during most of the pachytene stage, bright HEI10 foci are present, together with faint HEI10 signal marking the CE (Figure S8C,D). In axr1, the dynamics of HEI10 progression were the same as in wild type with HEI10 detected as multiple foci during early prophase stages (Figure 8E,I and Figure S8). Brighter foci then appeared as synapsis progressed, also forming a string of pearls on ZYP1 stretches (Figure 8J and Figure S8, arrows). A subset of very bright foci was retained at the later stages (Figure 8G,H,K,L and Figure S8). We noticed that at these late stages (based on the HEI10 pattern), the level of synapsis varied considerably from one cell to another. In addition, although these late HEI10 foci were always observed on ZYP1 stretches, the reverse was not true and ZYP1 stretches without late HEI10 signals were observed (see, for example, Figure 8H, where four late HEI10 foci are clustered on a single ZYP1 stretch, whereas many ZYP1 stretches are deprived of HEI10 foci). Therefore, it appears that class I CO clustering in axr1 is correlated with strong synapsis defects, but cannot be explained by the limited extension of the SC. Meiotic Defects in the axr1 Mutant Are Epistatic to Those of a Cullin 4 Mutant Because neddylation is known to regulate the activity of CRLs, we investigated whether axr1 meiotic defects are dependent on a specific CRL. In A. thaliana only four cullins are neddylated: cullin 1, cullin 3A, cullin 3B, and cullin 4 [33]. To identify possible AXR1 downstream players, we scored cullin-deficient lines for meiotic defects. Complete suppression of any of cullin functions (null cul1 or cul4 or the double cul3a cul3b mutants) is lethal, but various genetic backgrounds deficient in cullin activities are available We first investigated meiosis of the auxin response defective cul1 mutant alleles—cul1–6 [60], axr6-2/N3818 [61], and axr6-3/eta1 [62]—and observed perfectly normal meiosis (not shown). Next, considering cullin 3 activity, we analysed the CUL3a/3b hypomorphic mutant [cul3w (cul3a3cul3b1)] described for its defects in various aspects of the ethylene biosynthesis pathway and root development [63]. cul3w plants also showed normal meiotic development of male meiocytes (not shown). Finally, we analysed the cul4-1 mutant in which a T-DNA is inserted occurred in the 12th exon of the gene, leading to aberrant CUL4 mRNA expression, which varies depending on the developmental stage [64]. We observed significant male and female gametophyte abortion in cul4-1 (shown for the male, compare Figure 9A to Figure 2C). Although in wild type only balanced tetrads of microspores were observed, asymmetric tetrads and polyads were seen in cul4-1 mutants (compare Figure 9B to Figure 2E). Male meiosis was then investigated. The first stages of meiosis proceeded normally in cul4-1 mutants, however we observed metaphase I phenotypes reminiscent of the axr1 defects, with a large proportion of cells showing a clear reduction in bivalent formation (Figure 9C). The MCN per meiotic cell in cul4-1 (6±3.2, n = 71) was significantly different from wild type (8.9±0.9, n = 51, p<0.0001), and slightly different from axr1 (5.1±1.5, n = 74, p = 0.02). Nevertheless, the number of MCN per cell in cul4-1 was far more variable than in axr1 (Figure 9E), due to an overrepresentation of cells with wild-type levels of chiasmata (Figure 9D,E). We then introgressed the axr1 mutation (N877898) into cul4-1 and found that the double mutant cannot be distinguished from the single axr1 in terms of meiotic phenotype (not shown), the average level of MCN per cell (4.9±1.8, n = 98, p = 0.412), and in terms of variability of the values (Figure 9E), showing that axr1 is epistatic to cul4-1. Overall, our results suggest that AXR1 acts during meiotic recombination through the activation of a CRL4 complex. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 9. CULLIN4 is involved in meiotic recombination in the same pathway as axr1. In the cul4-1 mutant, a mixture of viable (purple) and dead (arrow) pollen grains can be seen in the anthers after Alexander staining (A). This is correlated with the production of aberrant tetrads and polyads of microspores (B). DAPI staining of the meiotic chromosomes revealed a defect at metaphase I (C) in bivalent formation, which is quantified in (E). Bar  =  10 µM. https://doi.org/10.1371/journal.pbio.1001930.g009 The axr1 Mutants Are Meiosis-Defective In the process of screening A. thaliana T-DNA (Agrobacterium tumefaciens transferred DNA) insertional lines for meiotic defects, we isolated three mutants [EGS344, EIC174, and EVM8 (Ws-4 strain); Figure 1 and Figure S1] allelic for disruption in At1g05180, the AXR1 gene, previously shown to encode the E1 enzyme of the Arabidopsis neddylation complex [44]. Another insertion line in At1g05180 available in the public collection (http://signal.salk.edu/) Sail_904E06 (N877898, Col-0 strain) and the historical axr1 allele (axr1-12/N3076, Col-0 ecotype [44], with a single nucleotide substitution in exon 11 of At1g05180) were also included in this study (Figure 1). Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 1. The AXR1 gene and axr1 mutations. The arrow indicates the orientation of the open reading frame. Exons are shown as boxes (pink, UTR; black, CDS). In the EGS344 mutant, a large deletion associated with an insertion of the exogenous Agrobaterium Ti plasmid disrupts the AXR1 gene from nucleotide 91 (40 bp 5′ to the ATG). In the EVM8 mutant, an in-frame deletion of 312 bp between exons 3 and 4 generates a 20 aa truncated protein. In EIC174, a single nucleotide insertion (A) in exon 6 (position 1364 of the genomic sequence, corresponding to nt 688 in the cDNA) leads to a premature stop codon (a 222 aa protein is produced instead of the 540 aa protein in wild type). In axr1-12, corresponding to the N3076 line, a single C-T nucleotide substitution at position 1295 of the cDNA leads to a premature stop codon (415 aa instead of 540), as described by Leyser et al. [44]. In N877898, corresponding to the Sail_904E06 line, a T-DNA insertion occurred in intron 11. References used for this figure are Tair accession 4010763662 for the genomic sequence and Tair accession 4010730885 for the cDNA sequence. https://doi.org/10.1371/journal.pbio.1001930.g001 The mutant plants all show the same vegetative phenotypes as previously described for axr1 mutants: They are dwarfed, excessively branched, with small rosettes and crinkled leaves (shown for N877898 in Figure 2A–B and in Figure S2 for the other alleles) [45],[46]. They also have small flowers and short fruits, indicating fertility defects (Figure S2). Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 2. axr1 developmental defects. Five- (A) or nine- (B) week-old wild-type (wt) or axr1 (N877898) mutant plants. axr1 mutants are dwarfed, strongly branched, and have short siliques. Alexander staining (C–D) reveals round pollen grains, with a red cytoplasm reflecting viable male gametophytes in wild type (C), whereas axr1 (N877898) anthers (D) contain a mixture of viable and dead (uncoloured, arrows) pollen grains. DIC microscopy of male meiosis products (E and F) reveals tetrads of microspores in wild-type (E) and unbalanced tetrads or polyads in axr1 (F, N877898). https://doi.org/10.1371/journal.pbio.1001930.g002 We examined the reproductive development of these mutants and found that all alleles showed a high level of male and female gametophyte abortion [shown for N8779898 male gametophytes (pollen grains) in Figure 2D]. In plants, male gametogenesis occurs in the anthers where groups of meiocytes undergo meiosis synchronously, each producing four haploid cells (called microspores). The four products of each meiosis remain temporally encased in a common callose wall, forming tetrads of microspores that can be visualised after tissue clearing (Figure 2E). Each microspore is then released from its tetrad and continues to develop into a mature pollen grain (the male gametophyte) containing the male gametes. Study of the early stages of pollen development in axr1 revealed the presence of abnormal meiotic products. Instead of the regular tetrahedral structure observed in the wild-type, asymmetric tetrads (containing four daughter cells of unequal size) or “polyads” (containing more than four products) were observed (Figure 2F), suggesting that the meiotic program is disrupted in these mutants. To confirm that the reduced fertility was caused by a defect in meiosis, we investigated male meiosis via chromosome spreading and DAPI (4′,6-diamidino-2-phenylindole) staining (Figure 3). During wild-type meiotic prophase I (Figure 3A–D), DNA fibres of each sister chromatid are organised as chromatin loops connected to a common protein axis (the axial element [AE]) [47]. When chromosomes start to condense at leptotene, they become visible as threads (Figure 3A). At this stage, meiotic recombination is initiated by the formation of a large number of DNA DSBs (not shown). HR repairs these breaks concomitantly with the progression of synapsis, the close association of the homologous chromosome axes through the polymerisation of the central element (CE) of the synaptonemal complex (SC). Synapsis begins at zygotene (not shown) and is complete by pachytene, when complete alignment of homologous pairs can be detected in DAPI-stained chromosomes (Figure 3B). DNA repair and recombination are thought to be achieved during pachytene, yielding at least one CO per homologous chromosome pair. At diplotene (Figure 3C), when the CE of the SC is depolymerised, the homologous chromosomes are therefore connected to each other by COs in which chromatids from homologous chromosomes have been exchanged. These connections between homologous chromosomes become apparent only at diakinesis (Figure 3D, arrows), when chromosomes are sufficiently condensed. At this stage in Arabidopsis, chiasmata (the cytological manifestations of COs) cannot be scored precisely, but chiasma-carrying chromosome arms can sometimes be identified based on bivalent appearance (see Figure 3D, arrows). Next, condensation proceeds and, at metaphase I, the five Arabidopsis bivalents are easily distinguishable, aligned on the metaphase plate (Figure 3E). During anaphase I, sister chromatid cohesion is released from chromosome arms, allowing homologous chromosomes to segregate to the two opposite cellular poles (Figure 3F). The second meiotic division then separates the sister chromatids, generating four pools of five chromosomes (Figure 3G and 3H), which gives rise to the tetrads of four spores (Figure 2E). Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 3. axr1 mutants show normal meiotic progression but reduced bivalent formation at metaphase I. DAPI staining of meiotic chromosomes in wild type (A–H) and axr1 (N877898, I–P). At the onset of meiotic prophase I (A and I), chromosomes can be identified. Chromosome alignment and synapsis then proceeds, leading eventually to the pachytene stage in wild type (B), where homologous chromosomes are synapsed along their entire length. This association can be observed in axr1 (J, enlarged regions) but remains partial. Then, the SC disappears at diplotene (C and K), condensation proceeds, and bivalents can be identified in wild type at diakinesis (D), but this stage is rarely observed in axr1 (L). At metaphase I, the five Arabidopsis bivalents can be identified in wild type (E), segregating at anaphase I (F). In axr1, a mixture of bivalents and univalents are observed (M), leading to subsequent improper segregation at anaphase I (N). Sister chromatids segregate at meiosis II (G and O), leading to balanced tetrads in wild type (H), unbalanced tetrads (not shown) or polyads in axr1 (P). At metaphase I, univalents (u) can be distinguished from ring bivalents (where a chiasma occurred in each of the two chromosome arms, *) and from rod bivalents (where only one chromosome arm shows a chiasma, #). Arrows in (D) indicate some of the chiasma-containing arms. Bar, 10 µm. https://doi.org/10.1371/journal.pbio.1001930.g003 In A. thaliana axr1 mutants, the leptotene and zygotene stages appeared similar to those in the wild type. However, no pachytene cells were identified in the 457 meiocytes analysed, in contrast to wild type, where this stage is present in approximately 35% of the cells (n = 334). Instead, we observed pachytene-like stages, with only partial chromosome alignment (Figure 3J). This suggests that axr1 is defective in synapsis. Diplotene cells were indistinguishable from those in the wild type (Figure 3K). Then, chromosome condensation could be followed until metaphase I, although diakinesis stages were rarely observed (1% of all stage cells, n = 457 for N877898, 12% in wt, n = 334) (Figure 3L). At wild-type metaphase I, the five typical Arabidopsis bivalents could be observed aligned on the metaphase plate (Figure 3E). Each bivalent was composed of two homologous chromosomes connected by chiasmata either on one chromosome arm (rod bivalent, Figure 3E#) or on both pairs of chromosome arms (ring bivalent, Figure 3E*). Chiasma numbers could therefore be estimated based on the bivalent structure. However, because multiple COs on a single arm cannot be cytologically differentiated from single COs, these estimates only correspond to a minimum chiasma number (MCN; Figure 4, Table S1). Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 4. COs in axr1 are largely ZMM-dependent. For each axr1 allele and for their respective wild-type strains (Ws-4 for EGS344, EIC174, EVM8, and Col-0 for N8777898 and N3076), and for a combination of multiple mutants, the level of bivalent formation as well as the MCN per cell were measured. For multiple mutant analyses, the N877898 allele was used in a Col-0 background and EGS344 in a Ws-4 background. The complete dataset can be found in Table S1. https://doi.org/10.1371/journal.pbio.1001930.g004 In axr1 mutants, we observed reduced bivalent formation, and instead of five bivalents, a mixture of bivalents and univalents could be identified (Figure 3M). The reduction in bivalent formation resulted in chromosome mis-segregation during subsequent anaphase I (Figure 3N), whereas the second meiotic division separated sister chromatids (Figure 3O), giving rise to a variable number of daughter cells containing aberrant numbers of chromosomes (Figure 3P). We quantified the decrease in bivalent formation as well as the MCN at metaphase I from all axr1 mutants and their respective wild-type accessions (Figure 4, Table S1). On average, axr1 mutants had 78% of the wild-type number of bivalents for the Col-0 background and 52% for the Ws background. In terms of the chiasma number, axr1 mutants displayed a residual level of 56% and 41% of the wild-type levels for Col-0 and Ws strains, respectively (Figure 4). Within a single ecotype (Col-0 or Ws), all alleles were statistically different from the wild type but not different from each other. Finally, when the partitioning of the residual chiasmata in axr1 was analysed, we observed that a large proportion of metaphase I cells showed both ring bivalents (at least two chiasmata) together with univalents (no chiasma) (42% of the N877898 cells, n = 47), showing that in axr1, the obligatory CO is lost. To further analyse the bivalent shortage observed in axr1, we used fluorescence in situ hybridization (FISH) analyses on PMCs. Metaphase I chromosomes were labelled with probes for the 45S and 5S rDNA repeats, allowing specific identification of chromosomes 1, 2, and 4 (Figure 5). Chromosomes 3 and 5 could not be discriminated from each other with these probes and were pooled. First, we observed that in axr1 as in wild type, bivalents were always formed between homologous chromosomes (n = 147 bivalents for axr1, n = 165 for wt). Then, we considered each bivalent individually and determined which pair of chromosomes was involved in its formation. As shown in Figure 5D, in axr1, as in the wild type, each pair of chromosomes was equally involved in bivalent formation, showing that the decrease in bivalent formation observed in axr1 affected all chromosomes in the same way. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 5. Bivalent shortage has a similar effect on each pair of chromosomes. Fluorescent in situ hybridisation (FISH) on metaphase I cells was performed with probes directed against the 45S (green) and the 5S (red) rDNA, which allow the identification of chromosomes 1 (unlabelled), 2 (green labelled), and 4 (green and red labelled), whereas chromosomes 3 and 5 cannot be distinguished (red labelled). In wild type, each chromosome pair represents 20% of the total number of bivalents (A and D, centre circle, in light, n = 21 cells). In axr1 (B and D, N877898 allele, external circle, n = 28), the proportion of each bivalent pair is the same as in wild type. Bar = 5 µm. https://doi.org/10.1371/journal.pbio.1001930.g005 axr1 COs Are ZMM-Dependent In wild-type Arabidopsis, the majority of COs (85%–90%, depending on the genetic background Col-0 versus Ws-4) depend on the ZMM proteins (MSH4, MSH5, MER3, ZIP4, SHOC1/ZIP2, HEI10, and PTD) as well as on MLH1 and MLH3 [21],[48], whereas MUS81 is responsible for 10%–15% of the remaining COs [14],[22]. We measured bivalent formation frequencies and the chiasma frequencies in various genetic combinations compared to the single axr1 mutant (Figure 4, Table S1). For all the zmmaxr1 double mutants (except mer3axr1) and regardless of strain (Col-0 versus Ws-4), the level of bivalent formation was reduced by more than 95% with hardly any bivalents observed (from 0.13 to 0.18 bivalent per cell; Table S1), showing that almost all the COs in axr1 are ZMM-dependent. We also analysed the bivalent frequency in the axr1mus81 double mutant, which was the same as for the axr1 single mutant (3.77±1.03 against 3.75±1.12; p = 0.9) (Figure 4). We then quantified bivalent frequency in the axr1msh5mus81 triple mutant and observed, as expected, a dramatic decrease in bivalent formation compared to axr1mus81 (Figure 4). No difference could be detected between the axr1msh5mus81 triple mutant and the axr1msh5 double mutant (p = 0.2). These results show that CO formation in axr1 mutants is almost exclusively dependent on ZMM proteins, whereas the MUS81 pathway plays only a limited role, if any. Class I COs Are Mislocalised in the axr1 Mutant To further analyse recombination events in axr1, we immunolabelled chromosomes with antibodies directed against HEI10 and MLH1, two markers of class I COs in Arabidopsis [48],[49]. MLH1 foci can be seen from late pachytene to diakinesis [49], whereas HEI10 is first loaded early during prophase on a large number of sites forming foci of different sizes on chromosomes. A limited number of these foci then remain (Figure 6A and B) at sites that correspond to class I COs where they co-localise with MLH1 until the end of prophase [48]. We therefore counted HEI10 and MLH1 foci in late pachytene and diplotene cells in wild type and axr1. Surprisingly, the average foci number per cell was not different between wild type and axr1, for either HEI10 (8.30±0.29, n = 54 and 7.49±0.40, n = 84, p = 0.15) or MLH1 (8.61±0.29, n = 33 and 7.58±0.54, n = 91, respectively, p = 0.263). In addition, we confirmed that these foci localise to chiasma-containing arms at diakinesis (Figure 6E and F and Figure S3), showing that they are likely to mark CO sites in axr1 as in wild type [49]. We also observed that there was higher variability in the numbers of HEI10 and MLH1 foci in axr1 than in wild type (Figure 6G), with the coefficient of variation (standard deviation divided by the mean) varying from 26% (HEI10, wt) to 50% (HEI10, axr1) or from 19% (MLH1, wt) to 68% (MLH1, axr1). Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 6. The average number of class I COs is similar in wild type and axr1. HEI10 or MLH1 was immunolocalised on acetic acid spread chromosomes from wild-type (A, B, and E, Col-0) or axr1 (C, D, and F) meiocytes from late pachytene to diakinesis. In axr1 (N877898 allele), the average number of HEI10 or MLH1 foci per cell is similar to that in wild type (G). Bar  =  5 µM. https://doi.org/10.1371/journal.pbio.1001930.g006 Another striking feature of axr1 was the frequent occurrence at the pachytene-like and diplotene stages of portions of paired chromosome axes where adjacent HEI10 and MLH1 foci could be seen (Figure 6C, D, arrows and Figure 7A, arrows). Forty-seven percent (HEI10, n = 60) or 53% (MLH1, n = 66) of the cells had at least two foci localised on the same portion of a chromosome axis, whereas in wild type, this scenario occurred only in 7% (HEI10, n = 57) or 3% of the cells (MLH1, n = 39) (Figure 7B). In addition, although we never observed more than two adjacent foci in wild type, we observed 22% (HEI10) and 13% (MLH1) of the cells with more than two adjacent foci, with a maximum of five adjacent HEI10 foci observed in axr1 (Figure 7B). Therefore, although the average level of class I COs is the same in axr1 and in wild type (Figure 6G), these class I COs tend to cluster together in at least 50% of the axr1 cells. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 7. Class I COs tend to cluster in axr1. (A) Examples of adjacent HEI10 or MLH1 foci in wild-type (Wt, Col-0) and in axr1 (N877898 allele) acetic acid spread meiotic chromosomes. (B) Proportion of pachytene and diplotene cells where adjacent foci were observed on the same chromosome axis pair (Wt, wild type; axr1, N877898 allele) (0, no evidence of adjacent foci; 2, two adjacent foci, etc.). Some of these situations are indicated by arrows in panel A. Bar  =  5 µM. https://doi.org/10.1371/journal.pbio.1001930.g007 We then estimated the scale at which this clustering arises. The distance between clustered foci was measured and compared to the total length of chromosome axis. The distance between two adjacent foci was on average 1/400 of the total axis length of a cell, ranging from 1/1600 of the genome to a maximum of 1/90 of the genome (Figure S4A). Extrapolated in DNA distance, with the additional assumption that genome condensation is homogeneous, the distance between two adjacent foci in a cluster is therefore expected to vary from 150 kb to 3,000 kb, with an average of 625 kb. We also observed that the distance between two adjacent foci does not vary significantly in clusters with exactly two foci compared with clusters with more than two foci. As a consequence, cluster size increases proportionally with the number of foci present in the cluster (Figure S4B). The size of the clusters was on average 1/200 of the genome for HEI10 foci (n = 14, 1,200 kb) and 1/300 for MLH1 (n = 21, 800 kb). Finally, we examined whether the clustered foci displayed interference, as might be expected for class I COs. We thus considered the hypothesis H0 that the foci in clusters are not subject to interference. The test was based on the distribution of distances between adjacent foci, specifically using the coefficient of variation for the statistical test and comparing to 105 simulations under H0 (see Materials and Methods). For the clusters of three or more foci (Table S2), we rejected the H0 hypothesis of no interference for MLH1 foci (p = 0.0024 based on seven clusters), for HEI10 foci (p = 0.0028 based on six clusters), and when pooling the MLH1 and HEI10 data (p = 2.4×10−5 based on 13 clusters). Specifically, inside clusters, MLH1 and HEI10 foci are more evenly distributed than at random, showing that COs within clusters still interfere. Taken together, these results show that the shortage in bivalent formation observed in axr1 mutants is not due to a general decrease in CO formation but rather to a mislocalisation of class I COs that tend to cluster together. Measurement of Recombination Rates in axr1 Mutants The level of genetic recombination on several chromosomal intervals was measured using the Fluorescent-Tagged Lines (FTL) tool developed by Copenhaver et al. [50]. The FTL system is a visual assay based on segregation of genetically linked fluorescent proteins expressed in the pollen grains of the quartet mutant (qrt1), in which the pollen grains remain attached as tetrads. With these lines, a large number of meiotic products can be visually scored and then a subset of multiple CO events can be identified (two-, three-, and four-strand double COs in adjacent intervals and four-strand double COs within a single interval) ([50] and Table S3B). Six different intervals were used, either on chromosome 3 (I3b and I3c) or 5 (I5a, I5b, I5c, and I5d), with sizes ranging from 1,200 to 4,900 kb (Table S3A). We first measured recombination rates for each interval using the standard Perkins genetic mapping equation [51]. As shown in Table 1, recombination rates in axr1 vary differently depending on the interval tested, from 70% to 180% of the wild-type level. On average, axr1 shows an increase in recombination, but these data should be taken with caution, as recombination measurements rely only on a subset of tetrads (the viable tetrads). Out of the six intervals considered, intervals located close to the telomeres (I3b and I5b) showed the most significant increase in recombination, whereas proximal intervals appeared less affected. This could indicate that the level of recombination is affected differently according to the location on the chromosomes, although additional data will be required to determine if telomere proximity increases CO frequency in the mutant. Download: PPT PowerPoint slide PNG larger image TIFF original image Table 1. Recombination rates and interinterval interference. https://doi.org/10.1371/journal.pbio.1001930.t001 We then used the FTL data to estimate interference between COs occurring in adjacent intervals (Table 2 and Table S4). We calculated the Interference Ratio (IR) as defined by Malkova et al. [18], which compares the genetic length of one interval with and without the presence of a simultaneous event in the neighbouring interval. When the occurrence of a CO in one interval reduces the probability of a CO occurring in the adjacent interval, the IR is less than 1, indicating (positive) CO interference. When COs in the two adjacent intervals are independent of each other, the IR is 1, and if the presence of one CO in an interval increases the probability of an additional CO in the adjacent interval, the IR is greater than 1, indicating negative interference. As shown in Table 2, all wild-type IRs were less than 1, in agreement with the presence of CO interference. For axr1, however, all IRs increased dramatically and were statistically significantly different to wild type (p<0.0001, Table 2 and Table S4). In addition, all axr1 IR values were greater than 1, although only one pair of intervals tested was significantly different from 1 (I5a I5b, first data set, IR = 1.63, p = 4×10−3). Therefore, in axr1, adjacent COs appear to occur more frequently than in wild type, which is in agreement with the previously observed clustering of class I COs scored cytologically (Figure 7). The cytologically observed clustering is occurring at a very small scale, namely a few hundred kb (on average 1,200 kb for HEI10 foci and 800 kb for MLH1 foci, see above), whereas in FTLs pairs of intervals correspond to more than 3,000 kb (I5cd, I3bc) and up to 7,500 kb (I5ab). Consequently, most of the clusters are expected to be present within a single interval and to only occasionally affect two adjacent intervals, which could explain why only one pair of intervals showed significant negative interference. Download: PPT PowerPoint slide PNG larger image TIFF original image Table 2. Intra-interval interference. https://doi.org/10.1371/journal.pbio.1001930.t002 Double CO events within a single interval can be detected using the FTLs if the two COs involve four different chromatids (Table S3B) because they will generate nonparental ditype (NPD) tetrads [50]. Interference within single intervals can be estimated by comparing the observed number of double COs (NPD frequency) to the expected number of double COs under the hypothesis of no interference [52]. The ratio between these two numbers (NPDr) gives the strength of interference within the considered interval, even if an important proportion of multiple COs will be silent. We calculated NPDr for all intervals considered for wild type and axr1 (Table 1 and Table S5). In wild type, the NPDr indicated strong interference (NPDr close to 0.3) within all the intervals (except for I3c, which is too small for statistically meaningful data, Tables S3A and S5). In axr1, however, the NPDr increased systematically (between 0.7 and 1.47) and was mostly greater than 1. For two intervals (I5a and I5b), the NPDr values of 2.69 and 1.63 were statistically significant (p<0.01), showing negative interference (Table 2). Thus, genetic analyses allowed us to measure negative interference in several of the intervals tested, confirming the CO clustering observed in cytology. Recombination Initiation Is Not Modified in axr1 Mutants To verify whether the recombination defect in axr1 could be linked to a defect in recombination initiation, we used two methods to investigate DSB formation. We first introgressed the axr1 mutation into a rad51 mutant, defective for meiotic DSB repair. In this mutant, DSBs are formed but are then repaired abnormally, leading to significant chromosomal defects (such as chromosome bridges and chromosome fragmentation) during anaphase I (Figure S5A). These chromosomal defects persisted in axr1rad51, showing that DSBs are present in the axr1 mutant (Figure S5B). Second, we analysed the nuclear distribution of the DMC1 protein, a meiosis-specific recombinase that forms foci at recombination sites. The dynamics and number of AtDMC1 foci in axr1 (237±40, n = 7) were indistinguishable from wild type (234±89, n = 28) (t, p = 0.9) (Figure S5). Thus, the meiotic defects observed in axr1 are not correlated with a decrease in the amount of recombination initiation events. Synapsis Is Strongly Defective in axr1 But Chromosome Axes Are Normal During meiotic prophase, chromosomes are structured in the context of a protein axis (the AE), which is crucial for most meiotic events, including meiotic recombination and synapsis [53],[54]. The meiotic chromosome axis is composed of specific AE proteins, such as ASY1 and cohesion proteins (REC8 and SCC3, [55],[56]). In wild-type meiotic cells, cohesins are loaded as early as premeiotic G1, whereas ASY1 appears at leptotene first as foci, then as a linear signal throughout the entire chromosome length (Figure S6A), in a pattern similar to that of cohesins (Figure S6C, [56]). As shown in Figure S6, the signal observed in axr1 mutants cannot be differentiated from wild type, showing that no major alteration of the axis can be detected in axr1 mutants. We then analysed the progression of synapsis by immunolocalisation of ZYP1, the A. thaliana CE component [57]. In wild type, ZYP1 appeared on chromosomes as foci that quickly elongated to yield a mixture of foci and short stretches of ZYP1 (Figure 8A,B, red signal and Figure S7). Synapsis then progressed until complete synapsis was reached, defining the pachytene stage (Figure 8C,D and Figure S7). In axr1, the early stages of synapsis could not be distinguished from wild type, showing a mix of foci and short ZYP1 stretches (Figure 8E,I and Figure S7). As meiosis progressed, ZYP1 elongation could be detected (Figure 8F–L and Figure S7), but full synapsis was never achieved (n = 66), confirming the synapsis defect detected after DAPI staining of meiocyte spreads (Figure 3). In addition, in approximately half of the cells, ZYP1 signals appeared strongly perturbed, uneven in thickness and forming dotted lines rather than a homogeneous continuous signal (Figure 8J or G and Figure S7). In some cases, only short and thick ZYP1 stretches were detected. These could correspond to ZYP1 poly-complexes rather than to CE polymerisation (Figure 8L and Figure S7). Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 8. Synapsis is strongly perturbed in axr1, but HEI10 dynamics during early prophase are unchanged. ZYP1 and HEI10 proteins were co-immunolocalised on lipsol-spread chromosomes from wild-type (A–D) and axr1 (N877989 allele, E–L) meiotic cells. The overlay of both signals is shown here (ZYP1 in red, HEI10 in green), but single channels can be found in Figures S7 and S8. In wild type as in axr1, ZYP1 appears on chromosomes as foci that quickly elongate, yielding a mixture of foci and short stretches (A, B, E, F, and I). Synapsis then progresses until complete synapsis is reached in wild type, defining the pachytene stage (C–D). In axr1, ZYP1 elongation can be detected, but full synapsis was never achieved (G, H, and K). In axr1, the ZYP1 signal is often uneven in thickness or forms dotted lines rather than a homogeneous and continuous signal (J and G). In addition, in some cases, only short and thick ZYP1 stretches were detected which could correspond to ZYP1 polycomplexes (L). During early zygotene, in wild type as in axr1, HEI10 forms numerous foci of variable sizes on chromatin (A, E, and I). Then, although synapsis progresses, combinations of large and small foci are observed, forming “strings of pearls” on ZYP1 stretches (B, F, and J, arrows). As meiosis progresses, a few bigger and brighter HEI10 foci can be observed in wild type (D) and in axr1 (G, H, K, and L), which generally co-exist with smaller and fainter HEI1O foci (C, D, G, H, and K). Whereas this latter HEI10 pattern is associated with complete synapsis in wild type (C–D), synapsis is only partial in axr1 (G, H, K, and L). Bar = 2 µm. https://doi.org/10.1371/journal.pbio.1001930.g008 CO maturation and Synapsis Can Be Uncoupled in axr1 To follow the progression of meiotic recombination events, we co-immunolocalised ZYP1 and HEI10, using a lipsol spreading protocol that has the advantage of allowing the simultaneous detection of these two proteins [58] but also the disadvantage of preventing examination of prophase after pachytene [59]. As mentioned above, HEI10 is detected as foci on meiotic chromosomes from leptotene to diakinesis, and its dynamics reflect the progression from early recombination intermediates to mature class I COs [48]. During leptotene and early zygotene, HEI10 forms numerous foci of variable size on chromatin (Figure 8A and Figure S8). Then, during synapsis initiation, bigger and brighter HEI10 foci appear, often co-localising with synapsed regions (Figure 8B and Figure S8). At this stage and later on, a combination of large and small foci are observed, forming “strings of HEI10 pearls” on ZYP1 stretches (Figure 8B,C and Figure S8B,C, arrows). At late pachytene, only a few bright HEI10 foci, corresponding to mature class I COs, are retained (Figure 8D and Figure S8). Nevertheless, during most of the pachytene stage, bright HEI10 foci are present, together with faint HEI10 signal marking the CE (Figure S8C,D). In axr1, the dynamics of HEI10 progression were the same as in wild type with HEI10 detected as multiple foci during early prophase stages (Figure 8E,I and Figure S8). Brighter foci then appeared as synapsis progressed, also forming a string of pearls on ZYP1 stretches (Figure 8J and Figure S8, arrows). A subset of very bright foci was retained at the later stages (Figure 8G,H,K,L and Figure S8). We noticed that at these late stages (based on the HEI10 pattern), the level of synapsis varied considerably from one cell to another. In addition, although these late HEI10 foci were always observed on ZYP1 stretches, the reverse was not true and ZYP1 stretches without late HEI10 signals were observed (see, for example, Figure 8H, where four late HEI10 foci are clustered on a single ZYP1 stretch, whereas many ZYP1 stretches are deprived of HEI10 foci). Therefore, it appears that class I CO clustering in axr1 is correlated with strong synapsis defects, but cannot be explained by the limited extension of the SC. Meiotic Defects in the axr1 Mutant Are Epistatic to Those of a Cullin 4 Mutant Because neddylation is known to regulate the activity of CRLs, we investigated whether axr1 meiotic defects are dependent on a specific CRL. In A. thaliana only four cullins are neddylated: cullin 1, cullin 3A, cullin 3B, and cullin 4 [33]. To identify possible AXR1 downstream players, we scored cullin-deficient lines for meiotic defects. Complete suppression of any of cullin functions (null cul1 or cul4 or the double cul3a cul3b mutants) is lethal, but various genetic backgrounds deficient in cullin activities are available We first investigated meiosis of the auxin response defective cul1 mutant alleles—cul1–6 [60], axr6-2/N3818 [61], and axr6-3/eta1 [62]—and observed perfectly normal meiosis (not shown). Next, considering cullin 3 activity, we analysed the CUL3a/3b hypomorphic mutant [cul3w (cul3a3cul3b1)] described for its defects in various aspects of the ethylene biosynthesis pathway and root development [63]. cul3w plants also showed normal meiotic development of male meiocytes (not shown). Finally, we analysed the cul4-1 mutant in which a T-DNA is inserted occurred in the 12th exon of the gene, leading to aberrant CUL4 mRNA expression, which varies depending on the developmental stage [64]. We observed significant male and female gametophyte abortion in cul4-1 (shown for the male, compare Figure 9A to Figure 2C). Although in wild type only balanced tetrads of microspores were observed, asymmetric tetrads and polyads were seen in cul4-1 mutants (compare Figure 9B to Figure 2E). Male meiosis was then investigated. The first stages of meiosis proceeded normally in cul4-1 mutants, however we observed metaphase I phenotypes reminiscent of the axr1 defects, with a large proportion of cells showing a clear reduction in bivalent formation (Figure 9C). The MCN per meiotic cell in cul4-1 (6±3.2, n = 71) was significantly different from wild type (8.9±0.9, n = 51, p<0.0001), and slightly different from axr1 (5.1±1.5, n = 74, p = 0.02). Nevertheless, the number of MCN per cell in cul4-1 was far more variable than in axr1 (Figure 9E), due to an overrepresentation of cells with wild-type levels of chiasmata (Figure 9D,E). We then introgressed the axr1 mutation (N877898) into cul4-1 and found that the double mutant cannot be distinguished from the single axr1 in terms of meiotic phenotype (not shown), the average level of MCN per cell (4.9±1.8, n = 98, p = 0.412), and in terms of variability of the values (Figure 9E), showing that axr1 is epistatic to cul4-1. Overall, our results suggest that AXR1 acts during meiotic recombination through the activation of a CRL4 complex. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 9. CULLIN4 is involved in meiotic recombination in the same pathway as axr1. In the cul4-1 mutant, a mixture of viable (purple) and dead (arrow) pollen grains can be seen in the anthers after Alexander staining (A). This is correlated with the production of aberrant tetrads and polyads of microspores (B). DAPI staining of the meiotic chromosomes revealed a defect at metaphase I (C) in bivalent formation, which is quantified in (E). Bar  =  10 µM. https://doi.org/10.1371/journal.pbio.1001930.g009 Discussion AXR1 Controls the Localisation of Class I COs During Meiosis We observed that in axr1 mutants, meiotic nondisjunction is correlated with defects in bivalent formation. However, our results indicate that the general level of meiotic recombination in axr1 is close to that of wild type, as we showed that COs are mostly under the control of the ZMM pathway and that their average number, revealed by MLH1 and HEI10 foci, is unchanged. Furthermore, these CO events show a completely aberrant distribution in axr1. First, cytogenetic data showed that clustered MLH1 or HEI10 foci are observed in approximately 50% of the meiocytes (Figure 7B). Second, genetic data showed that adjacent COs in most tested intervals no longer display genetic interference, showing that CO distribution is abnormal in axr1. More strikingly, in several intervals, strong significant negative interference was detected, a genetic demonstration of CO clustering. We can therefore conclude that COs in axr1 tend to cluster together and that the observed shortage in bivalent formation is not due to a global decrease in meiotic recombination but rather mislocalisation of these events, resulting in a loss of the obligatory CO. Very little information is available on the mechanisms that control CO distribution during meiosis. Nevertheless, it has been known for a long time that COs are not randomly distributed among chromosomes, as in most organisms, adjacent COs display interference and therefore tend to be evenly spaced within chromosomes [13]. In addition, the phenomenon of the “obligatory CO” (or CO assurance) ensures the formation of at least one CO per bivalent, whatever the total number of CO precursors per cell. The relationship between these two phenomena is still under debate [23], but recent modelling analyses suggested that the obligatory CO is a direct consequence of interference [65]. Numerous mutants with altered interference were described, but they nearly always also change CO rates, either because of increased MUS81-dependant COs [66]–[68] or because they are defective in the ZMM CO pathway (see, for example, [48],[69]). One possible exception is the Saccharomyces cerevisiae pch2 mutant, for which two independent studies showed that CO interference is alleviated without changes to meiotic recombination rates, at least on the smallest yeast chromosome (III) [70],[71]. Nevertheless, the generalisation of this observation to the whole genome seems unlikely [71]. To our knowledge, axr1 is therefore the first mutant that specifically modifies the localisation of class I COs, changing interference among them and resulting in the loss of the obligatory CO, but without changing the global average number of CO events. Thus, AXR1 is a key regulator of meiotic recombination outcomes. From our data, we can exclude that the meiotic defects observed in axr1 are due to a major decrease in DSB formation or to a drastic mislocalisation of these events (Figure S5). We have also shown that axr1 meiotic defects are not associated with major chromosome axis defects (Figure S6), but instead with major perturbations in the polymerisation of the SC CE (Figure 8). The relationship between CO control and SC polymerisation is a long-standing question in the field of meiosis [72]. In yeast, SC polymerisation is not necessary for CO interference, as it occurs after CO patterns have been imposed [73],[74]. In Caenorhabditis elegans, however, it was recently shown that the SC central region limits the formation of COs and imposes total interference [75]. This could also be the case in rice, where zep1 mutants (ZEP1 being the rice CE ZIP1 homologue) show an increase in chiasma formation at diakinesis [76]. This suggests that, in plants as in C. elegans, SC polymerisation could be necessary to limit CO formation. However, in Arabidopsis, ZYP1 appears to be required to prevent non-HR rather than acting on homologous CO formation [57]. To further complicate our understanding of the relationship between polymerisation of the SC CE and CO controls, in yeast, SC polymerisation requires the stabilisation of recombination intermediates by the ZMM proteins [77],[78]. In Arabidopsis, SC polymerisation is also dependent on the formation of HR intermediates, as no synapsis is observed either in spo11, dmc1, or rad51 mutants where recombination is either not initiated or is blocked at the invasion step. However, Arabidopsis zmm mutants all display normal synapsis [20],[48],[69], showing that SC polymerisation in these species depends on the formation of recombination intermediates, but not on their stabilisation by the ZMMs. The limited synapsis progression observed in axr1 mutants therefore suggests that recombination only proceeds far enough in a limited fraction of the genome where SC can polymerise and COs are formed. This appears to explain what is seen in Figure 8K, where mature HEI10 foci are concentrated on a few ZYP1 stretches. In that sense, aberrant SC formation illustrates that either recombination is blocked in a portion of the genome or that only a limited portion of the genome is competent to support recombination maturation, resulting in the loss of the obligatory CO. Nevertheless, the observation of nuclei where mature HEI10 foci are clustered on a single ZYP1 stretch, whereas the level of synapsis is high (as illustrated on Figure 8H) shows that the amount of ZYP1 polymerisation can be uncoupled from clustering of mature HEI10 foci. This suggests that CO clustering in axr1 is not only a consequence of limited synapsis progression. It is interesting to note that, within clusters of more than two class I CO foci, the distance among foci is not random, showing that they still display interference. During wild-type meiosis, no more than two adjacent foci could be scored, showing that these events are either less frequent or much more distant than in axr1. In this latter case, our results suggest that interference strength is considerably modified in axr1, resulting in CO clustering, at least in some areas of the genome. In Arabidopsis, interference strength is not uniform within chromosomes and increases toward the chromosome extremities [79]. This suggests that regional modification of interference parameters could be affected in axr1. In the future, it will therefore be crucial to determine whether CO clustering in axr1 is region-specific or not. Regardless, the average number of final CO events is unchanged in axr1, suggesting that (i) the total number of class I COs is precisely controlled, (ii) this control is still active in axr1, and (iii) the mechanism underlying this CO homeostasis is independent of the obligatory CO mechanism. AXR1 Acts on Meiotic Recombination Through Activation of a CRL4 Complex Neddylation stimulates several subclasses of cullin RING Ub ligases. We provide evidence that during meiotic recombination neddylation acts through cullin 4 activation to regulate the localisation of class I COs. Cullin 4 is a widely conserved cullin, involved in a large range of cellular and developmental controls, many of which are associated with genome integrity maintenance [80]. CRL4 complexes are composed of a CUL4 scaffold, a small RING domain containing RBX1 protein, a WD40-like repeat-containing adaptor DDB1 (DNA-damage binding 1), and a substrate receptor subunit called DWD (DDB1-binding WD40 protein) or DCAF (DDB1- and CUL4-associated factor) [80]. Evidence for CRL4 functions in genome integrity control come from multiple sources and concern mostly cell responses to UV damage and replication controls by regulating the accumulation of the replication licensing factor CDT1. For example, DDB1- and/or CUL4A-depleted human cells accumulate DSBs and have an activated ATM-ATR cell cycle checkpoint [81]. The budding yeast cul8 mutants (cullin 8 is thought to be the functional homologue of cullin 4 in S. cerevisae) also accumulate DNA damage [82]. In fission yeast, mutation in Ddb1 increases the spontaneous mutation rate by more than 20-fold and prevents premeiotic S phase entry [83]. CRL4 activity is also required for the NER pathway by controlling the detection and processing of DNA lesions induced by UV in plants [41],[84]–[86], but also in mammals, as loss of CUL4A in mice leads to an increase in susceptibility to UV skin cancer [87]. Evidence for the role of CRL4 complexes in DSB repair was also provided in Drosophila, where DDB1 depletion promotes loss of heterozygosity in somatic cells [88]. In addition, CRL4s complexes may also be involved in HR regulation, as in fission yeast, ddb1 mutants are defective in HR probably by regulating the pool of available dNTPs [89]. Interestingly, we observed that hardly any COs are retained in double axr1zmm mutants, suggesting that the MUS81 recombination pathway may be shut down in axr1. Because this pathway accounts for only a small proportion of all COs in Arabidopsis [14],[22], this disruption would not have a strong impact on meiosis. However, it could have a dramatic effect on somatic DNA repair, as the MUS81 pathway is one of the major pathways of somatic HR in eukaryotes [90]. It would therefore be interesting to study the involvement of AXR1 in somatic DNA recombination, above all considering that Dohmann and collaborators observed DSB accumulation in axr1 (axr1-3 and axr1-12) somatic cells [91]. Considering the crucial role of CRL4s in genome maintenance and the activation of DNA repair pathways including HR, it is hardly surprising to find that it is involved in the regulation of meiotic HR in Arabidopsis. Considering the conservation of CRL4 functions across kingdoms, it is likely that the regulation of meiotic recombination by one (or several) CRL4 complex(es) will be also observed in other eukaryotes. Indeed, two converging studies in mice recently showed that cullin 4 is also required for meiosis also in mammals, as depletion of Cul4a (one of the two mammalian Cul4 genes) led to male infertility [42],[43]. Whether this infertility is associated with early recombination [42] or later CO resolution defects [43] is still under debate. Nevertheless, the observation that MLH1 foci number is unchanged in cul4a but that a fraction of meiotic cells show pachytene bivalents without any MLH1 foci [43] is reminiscent of our data on axr1. Therefore, we propose that neddylation is acting on one or several CRL4 complex(es) to regulate the localisation of class I COs not only in Arabidopsis but also in mammals. In A. thaliana, there are more than 85 substrate receptor DWD domain proteins that can assemble with DDB1A or DDB1B or directly with CUL4-RBX1 to form CRL4 complexes [92],[93]. Further studies will be necessary to identify which of these is acting during meiosis. AXR1 Controls the Localisation of Class I COs During Meiosis We observed that in axr1 mutants, meiotic nondisjunction is correlated with defects in bivalent formation. However, our results indicate that the general level of meiotic recombination in axr1 is close to that of wild type, as we showed that COs are mostly under the control of the ZMM pathway and that their average number, revealed by MLH1 and HEI10 foci, is unchanged. Furthermore, these CO events show a completely aberrant distribution in axr1. First, cytogenetic data showed that clustered MLH1 or HEI10 foci are observed in approximately 50% of the meiocytes (Figure 7B). Second, genetic data showed that adjacent COs in most tested intervals no longer display genetic interference, showing that CO distribution is abnormal in axr1. More strikingly, in several intervals, strong significant negative interference was detected, a genetic demonstration of CO clustering. We can therefore conclude that COs in axr1 tend to cluster together and that the observed shortage in bivalent formation is not due to a global decrease in meiotic recombination but rather mislocalisation of these events, resulting in a loss of the obligatory CO. Very little information is available on the mechanisms that control CO distribution during meiosis. Nevertheless, it has been known for a long time that COs are not randomly distributed among chromosomes, as in most organisms, adjacent COs display interference and therefore tend to be evenly spaced within chromosomes [13]. In addition, the phenomenon of the “obligatory CO” (or CO assurance) ensures the formation of at least one CO per bivalent, whatever the total number of CO precursors per cell. The relationship between these two phenomena is still under debate [23], but recent modelling analyses suggested that the obligatory CO is a direct consequence of interference [65]. Numerous mutants with altered interference were described, but they nearly always also change CO rates, either because of increased MUS81-dependant COs [66]–[68] or because they are defective in the ZMM CO pathway (see, for example, [48],[69]). One possible exception is the Saccharomyces cerevisiae pch2 mutant, for which two independent studies showed that CO interference is alleviated without changes to meiotic recombination rates, at least on the smallest yeast chromosome (III) [70],[71]. Nevertheless, the generalisation of this observation to the whole genome seems unlikely [71]. To our knowledge, axr1 is therefore the first mutant that specifically modifies the localisation of class I COs, changing interference among them and resulting in the loss of the obligatory CO, but without changing the global average number of CO events. Thus, AXR1 is a key regulator of meiotic recombination outcomes. From our data, we can exclude that the meiotic defects observed in axr1 are due to a major decrease in DSB formation or to a drastic mislocalisation of these events (Figure S5). We have also shown that axr1 meiotic defects are not associated with major chromosome axis defects (Figure S6), but instead with major perturbations in the polymerisation of the SC CE (Figure 8). The relationship between CO control and SC polymerisation is a long-standing question in the field of meiosis [72]. In yeast, SC polymerisation is not necessary for CO interference, as it occurs after CO patterns have been imposed [73],[74]. In Caenorhabditis elegans, however, it was recently shown that the SC central region limits the formation of COs and imposes total interference [75]. This could also be the case in rice, where zep1 mutants (ZEP1 being the rice CE ZIP1 homologue) show an increase in chiasma formation at diakinesis [76]. This suggests that, in plants as in C. elegans, SC polymerisation could be necessary to limit CO formation. However, in Arabidopsis, ZYP1 appears to be required to prevent non-HR rather than acting on homologous CO formation [57]. To further complicate our understanding of the relationship between polymerisation of the SC CE and CO controls, in yeast, SC polymerisation requires the stabilisation of recombination intermediates by the ZMM proteins [77],[78]. In Arabidopsis, SC polymerisation is also dependent on the formation of HR intermediates, as no synapsis is observed either in spo11, dmc1, or rad51 mutants where recombination is either not initiated or is blocked at the invasion step. However, Arabidopsis zmm mutants all display normal synapsis [20],[48],[69], showing that SC polymerisation in these species depends on the formation of recombination intermediates, but not on their stabilisation by the ZMMs. The limited synapsis progression observed in axr1 mutants therefore suggests that recombination only proceeds far enough in a limited fraction of the genome where SC can polymerise and COs are formed. This appears to explain what is seen in Figure 8K, where mature HEI10 foci are concentrated on a few ZYP1 stretches. In that sense, aberrant SC formation illustrates that either recombination is blocked in a portion of the genome or that only a limited portion of the genome is competent to support recombination maturation, resulting in the loss of the obligatory CO. Nevertheless, the observation of nuclei where mature HEI10 foci are clustered on a single ZYP1 stretch, whereas the level of synapsis is high (as illustrated on Figure 8H) shows that the amount of ZYP1 polymerisation can be uncoupled from clustering of mature HEI10 foci. This suggests that CO clustering in axr1 is not only a consequence of limited synapsis progression. It is interesting to note that, within clusters of more than two class I CO foci, the distance among foci is not random, showing that they still display interference. During wild-type meiosis, no more than two adjacent foci could be scored, showing that these events are either less frequent or much more distant than in axr1. In this latter case, our results suggest that interference strength is considerably modified in axr1, resulting in CO clustering, at least in some areas of the genome. In Arabidopsis, interference strength is not uniform within chromosomes and increases toward the chromosome extremities [79]. This suggests that regional modification of interference parameters could be affected in axr1. In the future, it will therefore be crucial to determine whether CO clustering in axr1 is region-specific or not. Regardless, the average number of final CO events is unchanged in axr1, suggesting that (i) the total number of class I COs is precisely controlled, (ii) this control is still active in axr1, and (iii) the mechanism underlying this CO homeostasis is independent of the obligatory CO mechanism. AXR1 Acts on Meiotic Recombination Through Activation of a CRL4 Complex Neddylation stimulates several subclasses of cullin RING Ub ligases. We provide evidence that during meiotic recombination neddylation acts through cullin 4 activation to regulate the localisation of class I COs. Cullin 4 is a widely conserved cullin, involved in a large range of cellular and developmental controls, many of which are associated with genome integrity maintenance [80]. CRL4 complexes are composed of a CUL4 scaffold, a small RING domain containing RBX1 protein, a WD40-like repeat-containing adaptor DDB1 (DNA-damage binding 1), and a substrate receptor subunit called DWD (DDB1-binding WD40 protein) or DCAF (DDB1- and CUL4-associated factor) [80]. Evidence for CRL4 functions in genome integrity control come from multiple sources and concern mostly cell responses to UV damage and replication controls by regulating the accumulation of the replication licensing factor CDT1. For example, DDB1- and/or CUL4A-depleted human cells accumulate DSBs and have an activated ATM-ATR cell cycle checkpoint [81]. The budding yeast cul8 mutants (cullin 8 is thought to be the functional homologue of cullin 4 in S. cerevisae) also accumulate DNA damage [82]. In fission yeast, mutation in Ddb1 increases the spontaneous mutation rate by more than 20-fold and prevents premeiotic S phase entry [83]. CRL4 activity is also required for the NER pathway by controlling the detection and processing of DNA lesions induced by UV in plants [41],[84]–[86], but also in mammals, as loss of CUL4A in mice leads to an increase in susceptibility to UV skin cancer [87]. Evidence for the role of CRL4 complexes in DSB repair was also provided in Drosophila, where DDB1 depletion promotes loss of heterozygosity in somatic cells [88]. In addition, CRL4s complexes may also be involved in HR regulation, as in fission yeast, ddb1 mutants are defective in HR probably by regulating the pool of available dNTPs [89]. Interestingly, we observed that hardly any COs are retained in double axr1zmm mutants, suggesting that the MUS81 recombination pathway may be shut down in axr1. Because this pathway accounts for only a small proportion of all COs in Arabidopsis [14],[22], this disruption would not have a strong impact on meiosis. However, it could have a dramatic effect on somatic DNA repair, as the MUS81 pathway is one of the major pathways of somatic HR in eukaryotes [90]. It would therefore be interesting to study the involvement of AXR1 in somatic DNA recombination, above all considering that Dohmann and collaborators observed DSB accumulation in axr1 (axr1-3 and axr1-12) somatic cells [91]. Considering the crucial role of CRL4s in genome maintenance and the activation of DNA repair pathways including HR, it is hardly surprising to find that it is involved in the regulation of meiotic HR in Arabidopsis. Considering the conservation of CRL4 functions across kingdoms, it is likely that the regulation of meiotic recombination by one (or several) CRL4 complex(es) will be also observed in other eukaryotes. Indeed, two converging studies in mice recently showed that cullin 4 is also required for meiosis also in mammals, as depletion of Cul4a (one of the two mammalian Cul4 genes) led to male infertility [42],[43]. Whether this infertility is associated with early recombination [42] or later CO resolution defects [43] is still under debate. Nevertheless, the observation that MLH1 foci number is unchanged in cul4a but that a fraction of meiotic cells show pachytene bivalents without any MLH1 foci [43] is reminiscent of our data on axr1. Therefore, we propose that neddylation is acting on one or several CRL4 complex(es) to regulate the localisation of class I COs not only in Arabidopsis but also in mammals. In A. thaliana, there are more than 85 substrate receptor DWD domain proteins that can assemble with DDB1A or DDB1B or directly with CUL4-RBX1 to form CRL4 complexes [92],[93]. Further studies will be necessary to identify which of these is acting during meiosis. Materials and Methods Plant Material Ws-4 lines (including EGS344, EIC174, and EVM8) were obtained from the Versailles collection of Arabidopsis T-DNA transformants available at http://www-ijpb.versailles.inra.fr/en/sgap/equipes/variabilite/crg/[94]. Col-0 lines [including N877898 (Sail_904E06) and N3076 = axr1-12] were obtained from the collection of T-DNA mutants from the Salk Institute Genomic Analysis Laboratory (Columbia accession) (SIGnAL, http://signal.salk.edu/cgi-bin/tdnaexpress) [95] and provided by NASC (http://nasc.nott.ac.uk/). Other mutant alleles used in this study are as follows: msh4Ws (EXY25) [48], msh5Col (SALK_026553) [96]; hei10Ws (EQO124) [48], zip4Col (SALK_068052) [69]; mer3Col (mer3-2, SALK_091560) [20], mlh1Col (SK_25975) [48], mus81Col (SALK_107515) [14], rad51Col (Gabi_134A01) [97], mre11Col (mre11-4, Salk_067823), cul1-6 Col [60], axr6-2 Col (N3818) [61], axr6-3 Col (eta1) [62], cul3w Col [63], and cul4-1 Col [64]. Growth Conditions Plants were grown in a greenhouse (photoperiod 16 h/d and 8 h/night; temperature 20°C day and night; humidity 70%). AXR1 Cloning Screening for A. thaliana T-DNA (A. tumefaciens transferred DNA) insertions that provoke meiotic defects, we isolated three mutant lines: EGS344, EIC174, and EVM8. They all segregated 3∶1 for reduced fertility, meiotic defects, and a bushy vegetative phenotype. Linkage analysis (as described by Grelon et al. [98]) showed that none of the mutations were linked with a T-DNA insertion. We therefore undertook a rough positional cloning of the three mutations as described by De Muyt et al. [99]. The most closely linked marker was chr1_02991901 for all three mutants (based on 31 F2 mutant plants for EVM8, 31 for EGS344, and 31 for EIC174). Fine gene mapping was then carried out as described by De Muyt et al. [99] using chromosome 1 microsatellite markers located between 1,243,352 and 1,573,000 bp. Among the predicted genes by TAIR10 SeqViewer server (http://www.arabidopsis.org/), we retained AXR1 (At1G05180) as the best candidate, as axr1 mutants were previously shown to display the same vegetative developmental defects as EGS344, EIC174, and EVM8 [44],[46]. Sequencing of At1g05180 in the three mutant lines showed that all three are disrupted in this open reading frame (see below). We further analysed the axr1 reference allele (axr1-12) and another insertion line (Sail_904E06) available in the public databases (http://signal.salk.edu/). They all displayed the same meiotic phenotype as the previously isolated lines. Molecular Characterisation of axr1 Alleles Sequencing of At1g05180 in the EIC174 mutant line revealed a single nucleotide insertion in exon 6 (position 1364 of the genomic sequence, corresponding to nt 688 in the cDNA), leading to a premature stop codon (a 222 aa protein is produced instead of 540 aa in wild type). In the EGS344 mutant, a deletion of 898 bp (from nucleotide 91 of the genomic sequence) together with an insertion of Agrobacterium plasmid Ti DNA disrupts At1g05180 (Figure S1). In the EVM8 line, an in-frame deletion of 312 bp occurred between exons 3 and 4, generating a 20 aa deleted protein. Details are shown in Figure S1. In axr1-12, corresponding to the N3076 line, a single C-T nucleotide substitution in position 1295 of the cDNA occurred, leading to a premature stop codon (415 aa instead of 540), as described by Leyser et al. [44]. In N877898, corresponding to the Sail_904E06 line, a T-DNA insertion occurred in intron 11. Sequence references are as follows: Tair Accession 4010763662 for the genomic sequence, and Tair Accession 4010730885 for the cDNA sequence. PCR Genotyping of Mutant Lines For EGS344 and EVM8, wild-type alleles were amplified with primers 05180-P1 (ACCCTGATTGAAGAAAAGTCT) and 05180-P2 (CGGAGGTCGTCAAGAAAA) (60°C, 30 PCR cycles, 1,200 bp). The EGS344 mutant allele was amplified with primers 05180-P1 and 05180-AgroP1 (ACATCACAGCACCTCGATCCTGG) (60°C, 30 PCR cycles, 300 bp). The EVM8 mutant allele was amplified with 05180-P1 and 05180-P2 (60°C, 30 PCR cycles, 980 bp) For N877898, the wild-type allele was amplified with primers N877898U and N877898L (60°C, 30 PCR cycles, 957 bp). The mutant allele was amplified with primers N877898L and Lb3SAIL (TAGCATCTGAATTTCATAACCAATCTCGATACAC) (60°C, 30 PCR cycles, 500 bp). For all other genotypes, the primer list and PCR amplification conditions are shown in Table S6. Genetic Analyses Recombination and interference measurements. The six intervals used in this study correspond to intervals I5a, I5b, I5c, I5d I3b, and I3c, described by Berchowitz et al. [50] and in Table S3A. We produced plants qrt−/− N877898+/−, and qrt−/− N877898+/− RYC/RYC. We crossed these two plants and in the progeny analysed tetrad fluorescence of semi-sterile plants qrt−/− N877898−/− RYC/+++ or fertile plants either qrt−/− N877898+/− RYC/+++ or qrt−/− N877898+/+ RYC/+++. Plants were grown in a greenhouse. Tetrad analyses were carried out as described in [50]. The resulting tetrad data (Table S3B) were analysed as described by Berchowitz et al. [50]. In brief, map distances were calculated using the Perkins mapping equation based on the measurement of the frequency of tetratype (T), parental (P), and NPD combinations of markers [d(cM) = (100 [6NPD+T])/(2[P+NPD+T])] [51]. Interference was then measured by comparing CO frequency in an interval when the adjacent interval had no CO to the CO frequency when the adjacent interval does have a CO, as described by Malkova et al. [18]. We calculated the ratio of these genetic distances and statistically compared these ratios as described by Berchowitz et al. [50] and using Stahl Lab Online tools (http://www.molbio.uoregon.edu/~fstahl/). Another estimate of interinterval interference via the coefficient of coincidence is shown in Table S4B. In addition, we estimated the level of intra-interval interference by calculating the NPD ratios (Table S5), which compares the number of observed double COs within a single interval (NPD) to the expected number of double COs under the hypothesis of no interference [52],[100]. Interference analysis of axr1 using the HEI10 and MLH1 foci patterns. Given a cluster of at least three MLH1 or HEI10 foci, we asked whether the internal foci were distributed as happens in the absence of interference. Because these clusters were small, under the hypothesis H0 of no interference, the foci should be distributed uniformly. Considering first the clusters formed with three foci, let d1 and d2 be the two distances between adjacent foci. After normalization, we have d1+d2 = 1. Then, we introduced the statistic S that corresponds to the sum of the squared centred deviations, normalized by the variance under H0. Because the mean (respectively, the variance) of a uniformly distributed random variable in [0;1] is 0.5 (respectively, 1/12), S = [(d1–0.5)2+(d2–0.5)2] * 12. This statistic can be generalized to clusters with k+2 foci (k+1 distances, d1+d2+…+dk+1 = 1): S = [(d1–1/(k+1))2+(d2–1/(k+1))2+… +(dk+1–1/(k+1))2] * (k+1)2 * (k+2)/k. The normalization is chosen so that under H0 each distance contributes to the same average to the statistic, regardless of the value of k. The total statistic S for the test is simply obtained by summing over all clusters. It is a random variable whose distribution we obtained by direct simulation under H0 of 105 datasets having the same values of k as in the experimental measurements. Small experimental values of S correspond to foci more regularly distributed than expected. The p value for our test is given by the proportion of simulated statistics smaller than that of the experimental data. Antibodies The anti-ASY1 polyclonal antibody was described by Armstrong et al. [101]. It was used at a dilution of 1∶500. The anti-ZYP1 polyclonal antibody was described by Higgins et al. [57]. It was used at a dilution of 1∶500. The anti-DMC1, anti-MLH1, and anti-HEI10 antibodies were described by Chelysheva et al. in [69], [49], and [48], respectively. These were used at a dilution of 1∶20, 1∶200, and 1∶200, respectively. The anti-REC8 polyclonal antibody was described by Cromer et al. [102] and the anti-SCC3 by Chelysheva et al. [56]. These were used at a dilution of 1∶250 and 1∶500, respectively. Microscopy Comparison of the early stages of microsporogenesis and the development of PMCs was carried out as described in Grelon et al. [98]. Preparation of prophase stage spreads for immunocytology was performed using Carnoy's fixative and acetic acid chromosome spreads [59], except for DMC1 detection and double HEI10/ZYP1 immunolabelling where lipsol spreading and paraformaldehyde fixation were used [58]. Chiasma numbers were assessed by analysing metaphase I spread PMC chromosomes stained with DAPI, as described by Sanchez-Moran et al. [103]. In brief, a rod bivalent stands for a single chiasma, whereas a ring bivalent as two (one on each arm). Observations were made as described by Chelysheva et al. [48]. Plant Material Ws-4 lines (including EGS344, EIC174, and EVM8) were obtained from the Versailles collection of Arabidopsis T-DNA transformants available at http://www-ijpb.versailles.inra.fr/en/sgap/equipes/variabilite/crg/[94]. Col-0 lines [including N877898 (Sail_904E06) and N3076 = axr1-12] were obtained from the collection of T-DNA mutants from the Salk Institute Genomic Analysis Laboratory (Columbia accession) (SIGnAL, http://signal.salk.edu/cgi-bin/tdnaexpress) [95] and provided by NASC (http://nasc.nott.ac.uk/). Other mutant alleles used in this study are as follows: msh4Ws (EXY25) [48], msh5Col (SALK_026553) [96]; hei10Ws (EQO124) [48], zip4Col (SALK_068052) [69]; mer3Col (mer3-2, SALK_091560) [20], mlh1Col (SK_25975) [48], mus81Col (SALK_107515) [14], rad51Col (Gabi_134A01) [97], mre11Col (mre11-4, Salk_067823), cul1-6 Col [60], axr6-2 Col (N3818) [61], axr6-3 Col (eta1) [62], cul3w Col [63], and cul4-1 Col [64]. Growth Conditions Plants were grown in a greenhouse (photoperiod 16 h/d and 8 h/night; temperature 20°C day and night; humidity 70%). AXR1 Cloning Screening for A. thaliana T-DNA (A. tumefaciens transferred DNA) insertions that provoke meiotic defects, we isolated three mutant lines: EGS344, EIC174, and EVM8. They all segregated 3∶1 for reduced fertility, meiotic defects, and a bushy vegetative phenotype. Linkage analysis (as described by Grelon et al. [98]) showed that none of the mutations were linked with a T-DNA insertion. We therefore undertook a rough positional cloning of the three mutations as described by De Muyt et al. [99]. The most closely linked marker was chr1_02991901 for all three mutants (based on 31 F2 mutant plants for EVM8, 31 for EGS344, and 31 for EIC174). Fine gene mapping was then carried out as described by De Muyt et al. [99] using chromosome 1 microsatellite markers located between 1,243,352 and 1,573,000 bp. Among the predicted genes by TAIR10 SeqViewer server (http://www.arabidopsis.org/), we retained AXR1 (At1G05180) as the best candidate, as axr1 mutants were previously shown to display the same vegetative developmental defects as EGS344, EIC174, and EVM8 [44],[46]. Sequencing of At1g05180 in the three mutant lines showed that all three are disrupted in this open reading frame (see below). We further analysed the axr1 reference allele (axr1-12) and another insertion line (Sail_904E06) available in the public databases (http://signal.salk.edu/). They all displayed the same meiotic phenotype as the previously isolated lines. Molecular Characterisation of axr1 Alleles Sequencing of At1g05180 in the EIC174 mutant line revealed a single nucleotide insertion in exon 6 (position 1364 of the genomic sequence, corresponding to nt 688 in the cDNA), leading to a premature stop codon (a 222 aa protein is produced instead of 540 aa in wild type). In the EGS344 mutant, a deletion of 898 bp (from nucleotide 91 of the genomic sequence) together with an insertion of Agrobacterium plasmid Ti DNA disrupts At1g05180 (Figure S1). In the EVM8 line, an in-frame deletion of 312 bp occurred between exons 3 and 4, generating a 20 aa deleted protein. Details are shown in Figure S1. In axr1-12, corresponding to the N3076 line, a single C-T nucleotide substitution in position 1295 of the cDNA occurred, leading to a premature stop codon (415 aa instead of 540), as described by Leyser et al. [44]. In N877898, corresponding to the Sail_904E06 line, a T-DNA insertion occurred in intron 11. Sequence references are as follows: Tair Accession 4010763662 for the genomic sequence, and Tair Accession 4010730885 for the cDNA sequence. PCR Genotyping of Mutant Lines For EGS344 and EVM8, wild-type alleles were amplified with primers 05180-P1 (ACCCTGATTGAAGAAAAGTCT) and 05180-P2 (CGGAGGTCGTCAAGAAAA) (60°C, 30 PCR cycles, 1,200 bp). The EGS344 mutant allele was amplified with primers 05180-P1 and 05180-AgroP1 (ACATCACAGCACCTCGATCCTGG) (60°C, 30 PCR cycles, 300 bp). The EVM8 mutant allele was amplified with 05180-P1 and 05180-P2 (60°C, 30 PCR cycles, 980 bp) For N877898, the wild-type allele was amplified with primers N877898U and N877898L (60°C, 30 PCR cycles, 957 bp). The mutant allele was amplified with primers N877898L and Lb3SAIL (TAGCATCTGAATTTCATAACCAATCTCGATACAC) (60°C, 30 PCR cycles, 500 bp). For all other genotypes, the primer list and PCR amplification conditions are shown in Table S6. Genetic Analyses Recombination and interference measurements. The six intervals used in this study correspond to intervals I5a, I5b, I5c, I5d I3b, and I3c, described by Berchowitz et al. [50] and in Table S3A. We produced plants qrt−/− N877898+/−, and qrt−/− N877898+/− RYC/RYC. We crossed these two plants and in the progeny analysed tetrad fluorescence of semi-sterile plants qrt−/− N877898−/− RYC/+++ or fertile plants either qrt−/− N877898+/− RYC/+++ or qrt−/− N877898+/+ RYC/+++. Plants were grown in a greenhouse. Tetrad analyses were carried out as described in [50]. The resulting tetrad data (Table S3B) were analysed as described by Berchowitz et al. [50]. In brief, map distances were calculated using the Perkins mapping equation based on the measurement of the frequency of tetratype (T), parental (P), and NPD combinations of markers [d(cM) = (100 [6NPD+T])/(2[P+NPD+T])] [51]. Interference was then measured by comparing CO frequency in an interval when the adjacent interval had no CO to the CO frequency when the adjacent interval does have a CO, as described by Malkova et al. [18]. We calculated the ratio of these genetic distances and statistically compared these ratios as described by Berchowitz et al. [50] and using Stahl Lab Online tools (http://www.molbio.uoregon.edu/~fstahl/). Another estimate of interinterval interference via the coefficient of coincidence is shown in Table S4B. In addition, we estimated the level of intra-interval interference by calculating the NPD ratios (Table S5), which compares the number of observed double COs within a single interval (NPD) to the expected number of double COs under the hypothesis of no interference [52],[100]. Interference analysis of axr1 using the HEI10 and MLH1 foci patterns. Given a cluster of at least three MLH1 or HEI10 foci, we asked whether the internal foci were distributed as happens in the absence of interference. Because these clusters were small, under the hypothesis H0 of no interference, the foci should be distributed uniformly. Considering first the clusters formed with three foci, let d1 and d2 be the two distances between adjacent foci. After normalization, we have d1+d2 = 1. Then, we introduced the statistic S that corresponds to the sum of the squared centred deviations, normalized by the variance under H0. Because the mean (respectively, the variance) of a uniformly distributed random variable in [0;1] is 0.5 (respectively, 1/12), S = [(d1–0.5)2+(d2–0.5)2] * 12. This statistic can be generalized to clusters with k+2 foci (k+1 distances, d1+d2+…+dk+1 = 1): S = [(d1–1/(k+1))2+(d2–1/(k+1))2+… +(dk+1–1/(k+1))2] * (k+1)2 * (k+2)/k. The normalization is chosen so that under H0 each distance contributes to the same average to the statistic, regardless of the value of k. The total statistic S for the test is simply obtained by summing over all clusters. It is a random variable whose distribution we obtained by direct simulation under H0 of 105 datasets having the same values of k as in the experimental measurements. Small experimental values of S correspond to foci more regularly distributed than expected. The p value for our test is given by the proportion of simulated statistics smaller than that of the experimental data. Recombination and interference measurements. The six intervals used in this study correspond to intervals I5a, I5b, I5c, I5d I3b, and I3c, described by Berchowitz et al. [50] and in Table S3A. We produced plants qrt−/− N877898+/−, and qrt−/− N877898+/− RYC/RYC. We crossed these two plants and in the progeny analysed tetrad fluorescence of semi-sterile plants qrt−/− N877898−/− RYC/+++ or fertile plants either qrt−/− N877898+/− RYC/+++ or qrt−/− N877898+/+ RYC/+++. Plants were grown in a greenhouse. Tetrad analyses were carried out as described in [50]. The resulting tetrad data (Table S3B) were analysed as described by Berchowitz et al. [50]. In brief, map distances were calculated using the Perkins mapping equation based on the measurement of the frequency of tetratype (T), parental (P), and NPD combinations of markers [d(cM) = (100 [6NPD+T])/(2[P+NPD+T])] [51]. Interference was then measured by comparing CO frequency in an interval when the adjacent interval had no CO to the CO frequency when the adjacent interval does have a CO, as described by Malkova et al. [18]. We calculated the ratio of these genetic distances and statistically compared these ratios as described by Berchowitz et al. [50] and using Stahl Lab Online tools (http://www.molbio.uoregon.edu/~fstahl/). Another estimate of interinterval interference via the coefficient of coincidence is shown in Table S4B. In addition, we estimated the level of intra-interval interference by calculating the NPD ratios (Table S5), which compares the number of observed double COs within a single interval (NPD) to the expected number of double COs under the hypothesis of no interference [52],[100]. Interference analysis of axr1 using the HEI10 and MLH1 foci patterns. Given a cluster of at least three MLH1 or HEI10 foci, we asked whether the internal foci were distributed as happens in the absence of interference. Because these clusters were small, under the hypothesis H0 of no interference, the foci should be distributed uniformly. Considering first the clusters formed with three foci, let d1 and d2 be the two distances between adjacent foci. After normalization, we have d1+d2 = 1. Then, we introduced the statistic S that corresponds to the sum of the squared centred deviations, normalized by the variance under H0. Because the mean (respectively, the variance) of a uniformly distributed random variable in [0;1] is 0.5 (respectively, 1/12), S = [(d1–0.5)2+(d2–0.5)2] * 12. This statistic can be generalized to clusters with k+2 foci (k+1 distances, d1+d2+…+dk+1 = 1): S = [(d1–1/(k+1))2+(d2–1/(k+1))2+… +(dk+1–1/(k+1))2] * (k+1)2 * (k+2)/k. The normalization is chosen so that under H0 each distance contributes to the same average to the statistic, regardless of the value of k. The total statistic S for the test is simply obtained by summing over all clusters. It is a random variable whose distribution we obtained by direct simulation under H0 of 105 datasets having the same values of k as in the experimental measurements. Small experimental values of S correspond to foci more regularly distributed than expected. The p value for our test is given by the proportion of simulated statistics smaller than that of the experimental data. Antibodies The anti-ASY1 polyclonal antibody was described by Armstrong et al. [101]. It was used at a dilution of 1∶500. The anti-ZYP1 polyclonal antibody was described by Higgins et al. [57]. It was used at a dilution of 1∶500. The anti-DMC1, anti-MLH1, and anti-HEI10 antibodies were described by Chelysheva et al. in [69], [49], and [48], respectively. These were used at a dilution of 1∶20, 1∶200, and 1∶200, respectively. The anti-REC8 polyclonal antibody was described by Cromer et al. [102] and the anti-SCC3 by Chelysheva et al. [56]. These were used at a dilution of 1∶250 and 1∶500, respectively. Microscopy Comparison of the early stages of microsporogenesis and the development of PMCs was carried out as described in Grelon et al. [98]. Preparation of prophase stage spreads for immunocytology was performed using Carnoy's fixative and acetic acid chromosome spreads [59], except for DMC1 detection and double HEI10/ZYP1 immunolabelling where lipsol spreading and paraformaldehyde fixation were used [58]. Chiasma numbers were assessed by analysing metaphase I spread PMC chromosomes stained with DAPI, as described by Sanchez-Moran et al. [103]. In brief, a rod bivalent stands for a single chiasma, whereas a ring bivalent as two (one on each arm). Observations were made as described by Chelysheva et al. [48]. Supporting Information Figure S1. Molecular characterisation of the axr1 alleles. In EIC174, a single nucleotide (A) insertion occurred in exon 6 (position 1364 of the genomic sequence, in red), leading to a premature stop codon (a 222 aa protein is produced instead of 540 aa in wild type). In EVM8, a large in-frame deletion of 312 bp (in blue) generates a 20 aa truncated protein. In EGS344, a 907 bp deletion (in green) associated with an Agrobacterium Ti plasmid DNA insertion (*) occurred in the 5′ end of the gene. https://doi.org/10.1371/journal.pbio.1001930.s001 (DOCX) Figure S2. Phenotype of axr1 mature plants. (A) axr1 mutants display strong vegetative defects. All plants shown are 6 wk old. Upper panel, axr1 allelic series. Lower panel, axr1 mutants compared to their respective wild-type strain (Col-0 or Ws-4). (B) axr1 mutants produce less seeds than wild type. https://doi.org/10.1371/journal.pbio.1001930.s002 (TIF) Figure S3. MLH1 foci localise on chiasma-containing arms in wild type and axr1. MLH1 was immunolocalised on acetic acid spread chromosomes from wild type (wt, Col-0) and axr1 (N877898) at diakinesis. Because adjacent univalents cannot be distinguished from bivalents, we selected only the MLH1-labelled bivalent-like structures (arrows) and scored where MLH1 foci occurred. In both genotypes, 100% of the MLH1 foci (n = 246 for wild type, n = 44 for axr1) were detected on connected chromosome arms and never on a free chromosome. https://doi.org/10.1371/journal.pbio.1001930.s003 (TIF) Figure S4. Class I CO cluster characterisation. (A) The distance between two adjacent foci does not vary significantly with the cluster type. The distance between two adjacent foci (measured in µm and divided by the total size of the chromosome axis) was measured in clusters containing two, three, or four clustered foci. The x axis is the total number of foci in the considered clusters. All measures were undertaken in the N877898 allele. (B) Class I CO cluster lengths increase proportionally with the foci number. In the graph below, the mean size of the HEI10 or MLH1 clusters (calculated in µm and divided by the total size of the chromosome axis) is given according to the number of foci present in the cluster. All measures were undertaken in the N877898 allele. https://doi.org/10.1371/journal.pbio.1001930.s004 (TIF) Figure S5. Early recombination events are not altered in axr1. (A and B) DAPI staining of anaphase I meiocytes from rad51 (A) and axr1rad51 (B) mutants. rad51 is defective in meiotic DSB repair as shown by the major chromosomal defects observed at meiosis (fragmentation, A). Introgression of the axr1 mutation (N877898 allele) does not rescue these defects (B), showing that meiotic DSB are present in axr1. Bars = 10 µm. (C–F) Lipsol chromosome spreads of Arabidopsis meiocytes stained with DAPI (C and E) and immunolabelled with the anti-DMC1 antibody (D and F). axr1 meiocytes (N877898 allele, E and F) show wild-type–like DMC1 staining. Bar = 5 µm. https://doi.org/10.1371/journal.pbio.1001930.s005 (TIF) Figure S6. No major axis defect can be detected in axr1. Immunolocalisation of ASY1 (A and B), REC8 (C and D), and SCC3 (E and F) in wild type (A, C, and E) and axr1 (B, D, and F) prophase meiotic cells (N877898 allele). Bar = 5 µm. https://doi.org/10.1371/journal.pbio.1001930.s006 (TIF) Figure S7. Synapsis is strongly perturbed in axr1. ZYP1 was immunolocalised on lipsol spread chromosomes from wild-type (A–D) and axr1 (N877989 allele, E–L) meiotic cells. This figure corresponds to the red channel from Figure 8. https://doi.org/10.1371/journal.pbio.1001930.s007 (TIF) Figure S8. HEI10 dynamics during early prophase is unchanged in axr1. HEI10 was immunolocalised on lipsol spread chromosomes from wild-type (A–D) and axr1 (N877989 allele, E–L) meiotic cells. This figure corresponds to the green channel from Figure 8. https://doi.org/10.1371/journal.pbio.1001930.s008 (TIF) Table S1. Average MCN and average bivalent number per meiocyte. https://doi.org/10.1371/journal.pbio.1001930.s009 (DOCX) Table S2. Interfoci distance within class I CO clusters. In axr1 (N877898 allele), HEI10 and MLH1 foci form clusters in approximately half of the pachytene/diplotene meiocytes. To estimate interference between adjacent foci within the clusters, we measured the distance between two adjacent foci in clusters containing more than two foci. https://doi.org/10.1371/journal.pbio.1001930.s010 (DOCX) Table S3. Recombination dataset. (A) The localisation of the six FTL intervals used in this study are shown below on the five Arabidopsis chromosomes. They correspond to three pairs of linked intervals—I5a I5b, I5c I5d, and I3b I3c—described by Berchowitz and Copenhaver [50]. The position (in bp) of each transgene encoding red, yellow, or cyan fluorescent proteins (filled circles) is shown. The size of each interval is given in the accompanying table. The map was obtained using the Chromosome Map Tool from http://www.arabidopsis.org/jsp/ChromosomeMap/tool.jsp. (B) To measure recombination rates, we produced plants homozygous for the quartet (qrt) mutation, heterozygous for axr1 (N877898 allele), and carrying or not three linked fluorescent markers (R, red; Y, yellow; C, cyan): qrt−/− axr1+/− and qrt−/− axr1+/− RYC/RYC. The qrt mutation allows the four pollen grains from a single meiosis to be maintained together. We crossed these two plants and in the progeny analysed tetrad fluorescence of mutant plants (qrt−/− axr1−/− RYC/+++) or fertile plants (wt) (either qrt−/− axr1+/− RYC/+++ or qrt−/− axr1+/+ RYC/+++). Plants were grown in a greenhouse, and tetrad analyses were carried out as described by Berchowitz and colleagues [50]. The distribution of the markers within the tetrads and the resulting distribution of colours vary depending on the number, localisation, and chromatids involved in recombination. The different possibilities are indicated by drawings and the number of tetrads in each phenotypic class is shown in the tables below. Two replicates were made for the I5a I5b intervals. https://doi.org/10.1371/journal.pbio.1001930.s011 (DOCX) Tables S4. Interinterval interference analyses. Genetic interference among COs occurring in two linked intervals was calculated using two methods: (A) The IR. The IR [18] compares the genetic size of an interval (d(Ia), in cM calculated using the Perkins equation [51]) when a CO occurs in an adjacent interval (d(Ia) with CO in Ib) to the genetic size of the same interval when no CO occurs in the adjacent interval (d(Ia) without CO in Ib). The ratio of these two distances, called the IR, gives a measurement of the strength of interference between the two intervals. When there is no interference, the ratio is equal to 1, whereas the ratio is below 1 when there is interference. Ratios above 1 indicate negative interference, indicative of more adjacent C0s than expected. We compared each IR to 1 [P(IR = 1)] and compared mutant IR to wild-type IR [P(IR = IR wt)]. Calculations and statistical analyses were performed according to Berchowitz and Copenhaver [50] using Stahl Lab Online tools (http://www.molbio.uoregon.edu/~fstahl/). In wild type, IRs were statistically below 1 for all pairs of intervals considered, indicative of interference among COs. In axr1, IRs increased systematically and were always greater than 1, but the difference with 1 was significant (p<0.01) only for intervals I5a I5b and only for one replicate of the experiment. (B) The coefficient of coincidence. The coefficient of coincidence (c.o.c.) compares the observed frequency of double COs in the two adjacent intervals [f(Ia and Ib) observed] to the expected frequency of double COs if there were no interference [f(Ia and Ib) expected]. This last frequency is the product of the frequency of COs in each single interval [f(Ia) and f(Ib)]. The c.o.c. corresponds to (f observed/f expected). When interference is absent, the c.o.c. is equal to 1. When interference is total, the c.o.c. is equal to 0. For wild type, this index varied from 0.37 to 0.63, indicative of interference. In axr1 the index was always above 1, which shows that double COs in adjacent intervals are more frequent than expected. https://doi.org/10.1371/journal.pbio.1001930.s012 (DOCX) Table S5. Intra-interval interference analyses. Interference was measured within a single interval, by comparing the observed number of double COs [based on NPD frequency (NPD observed) to the expected number of double COs under the hypothesis of no interference (NPD expected)] [52]. The ratio between these two figures (NPDr) gives the strength of interference within the considered interval. The NPD tetrads correspond to h, i, j, k, and l classes from Table S2. In wild type, NPDr indicate strong interference (NPDr close to 0.3) within all the intervals (except for I3c, which is too small to give statistically meaningful figures). In axr1, NPDrs increased systematically (between 0.7 and 2.69) and were generally greater than 1, indicative of a trend toward negative interference (more double COs in a single interval than expected). However, statistical analyses on NPDr (Stahl Lab Online tools, http://www.molbio.uoregon.edu/~fstahl/) showed that these values are statistically different from 1 (p<0.01) only on I5a and I5b (one of the two replicates). https://doi.org/10.1371/journal.pbio.1001930.s013 (DOCX) Table S6. Primer sequences and PCR conditions for mutant genotyping. https://doi.org/10.1371/journal.pbio.1001930.s014 (DOCX) Acknowledgments We are grateful to Wayne Crismani, Eric Jenczewski, Christine Mézard, Raphaël Mercier, and Wojtek Pawlowski for helpful discussions and constructive reading of the manuscript. We wish to thank Gregory Copenhaver for providing FTL lines and Pascal Genschik for providing cul3w and cul4-1 lines. We also wish to thank LKG Scientific Editing & Translation for correcting the manuscript.

doi: 10.1371/journal.pbio.1001919pmid: 25093743

Axons rarely regrow after a severe spinal cord injury, in part because of inhibitory signals associated with myelin, which surrounds and insulates the axon. These signals bind to the Nogo receptor, which can then bind to a variety of co-receptors, including a protein called p75. Together, this complex triggers a cascade of intracellular signals that ultimately inhibits axonal sprouting and prevents regeneration. Understanding how p75 is regulated, therefore, may shed light on new strategies for promoting recovery from nerve damage. In this issue of PLOS Biology, Marçal Vilar, Tsung-Chang Sung, and Kuo-Fen Lee show that p75 may lose its affinity for the Nogo receptor through interaction with a second protein, whose structure is closely related to p75 itself. p75 bears an intracellular region (the chillingly named “death domain”) which, when it occurs in many related proteins, allows the protein bearing it to associate with others like it and to trigger signaling. One such protein is p45, and so the authors asked whether p45 and p75 might interact. In extracts from mouse cerebellum, they found that, indeed, the two proteins formed a complex and that p45's expression pattern in response to injury matched that of p75. Increasing the expression of p45, they noticed, reduced the association of p75 and the Nogo receptor, an effect dependent on both intracellular and transmembrane domains of p45, indicating the likely site of physical association between the two proteins (Figure 1). Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 1. Following nerve injury, increasing p45 levels in neurons promotes nerve growth by blocking the formation of p75 homodimers that are critical for transmitting inhibitory signals from myelin-associated inhibitors such as Nogo. https://doi.org/10.1371/journal.pbio.1001919.g001 To confirm that hypothesis, the authors next characterized in detail the intracellular domain of p75. They found that in a gel, purified p75 displayed two different molecular weights, one twice that of the other, suggesting it forms a homodimer. Next, they systematically mutated specific amino acids throughout the protein and used nuclear magnetic resonance spectroscopy to map the effect on binding. As they suspected, the death domain played a key role in binding the two monomers together. A covalent cysteine linkage between the two further stabilized the homodimer. That stability contributed to the dimer's ability to bind to the Nogo receptor. Next, they used similar techniques to characterize p45's intracellular regions and to show that p45 linked to p75 through the death domains on each. By changing the concentrations of the two in solution, they found that p45 could disrupt the homodimer, leading to formation of a p45–p75 heterodimer, an effect that depended in part on cysteine–cysteine interactions within the transmembrane domains of each. Finally, they showed that the disruption of the p75 homodimer, along with the formation of the p45–p75 heterodimer, reduced the association of p75 with the Nogo receptor and, therefore, reduced the complex's ability to trigger the signaling that inhibits axon growth, thereby promoting axonal regrowth. While mice express p45, humans do not, because of a stop codon mutation, raising the question of whether this regulatory system is absent in humans, or whether other, as-yet-unidentified proteins play a similar role in regulating the p75–Nogo receptor interaction. In either case, the ability to disrupt that interaction, either directly or by targeting those putative proteins, may offer a new avenue for spinal cord repair. Vilar M, Sung T-C, Chen Z, García-Carpio I, Fernandez EM, et al. (2014) Heterodimerization of p45–p75 Modulates p75 Signaling: Structural Basis and Mechanism of Action. doi:10.1371/journal.pbio.1001918

doi: 10.1371/journal.pbio.1001938pmid: 25157871

The inner cell mass of the early embryo, or blastocyst, is composed of a population of undifferentiated cells that will eventually give rise to all the tissues of the adult animal. By studying the processes that drive the differentiation of these cells, scientists hope to gain a better understanding of how these processes can go awry and glean insights into how we might counter the deficiencies and diseases that result. For instance, studying the development of spinal cord neurons could someday bring about new therapies for the treatment of spinal cord injuries or neurodegenerative diseases. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 1. Stem cell–derived neuromesodermal precursors (green) transplanted into mouse embryos (shown in cross-section here) contribute to mesoderm, including Pax3 expressing dermomyotome (red). Tissue revealed with DAPI (blue). Image Credit: Filip J. Wymeersch. https://doi.org/10.1371/journal.pbio.1001938.g001 Observations made in developing mouse and chicken embryos have shown that spinal cord neurons differentiate from a population of progenitor cells, neuromesodermal precursors (NMPs), located in the posterior of the developing embryo. NMPs can also differentiate into mesoderm (an early embryonic tissue that gives rise to muscle and connective tissues, along with other cell types), and accordingly, NMPs express both neural and mesodermal markers. But little else is known about NMPs because it is difficult to reach and study the region of the embryo in which these cells appear. In a collaborative study between the Briscoe and Wilson labs, published this month in PLOS Biology, Mina Gouti, Anestis Tsakiridis, and colleagues explore how these cells develop and demonstrate a culture system to produce NMPs in vitro. To study NMP development, Gouti et al. employed mouse embryonic stem cells (mESCs). It is known that in vitro monolayer cultures of mESCs secrete a protein called fibroblast growth factor (Fgf) that prompts their differentiation into neural progenitor cells, a type of stem cell that is able to produce cells of the neuronal lineage. By default, these neural progenitor cells express markers typical of the forebrain, but by adding other chemicals to the cultures, researchers can prod the cells to instead display markers from different brain regions, including the hindbrain and brainstem. This culture system has therefore allowed scientists to ask questions about the development of these different brain regions that would be difficult or impossible to address in intact embryos. Until now, the chemical signals that drive differentiation of NMPs were unknown. However, previous studies have shown that, during development, cells in the posterior of the embryo are exposed to Fgf and another secreted signaling protein called wingless (Wnt). Therefore, Gouti et al. reasoned that Fgf and Wnt may cooperate to guide development of NMPs, and could also be able to induce mESCs to differentiate into NMPs in vitro. Indeed, the authors found that if they followed the in vitro protocol to produce cells with hindbrain identity, but added in a brief, early pulse of Wnt signaling before the cells assumed neural identity, they could obtain cells expressing spinal cord–related genes. Examination of gene expression patterns in these cells showed that they expressed both the neural marker Sox2 and the mesoderm marker Brachyury—two proteins characteristic of NMPs. Also like NMPs, the in vitro–cultured cells had a dual developmental potential. They could be prompted to differentiate into mesoderm if the Fgf and Wnt pulse was followed by exposure to Wnt in the absence of Fgf. Furthermore, when the in vitro–derived NMPs were engrafted into developing chick embryos, the descendants of these cells were found in both spinal cord– and mesoderm-derived tissues. These data showed that the authors had successfully reconstituted in culture the conditions needed to produce NMPs. With slight adjustments, Gouti and colleagues were also able to adapt and refine their in vitro NMP culture protocol for use with other types of stem cells: mouse epiblast stem cells and human embryonic stem cells. Like their mESC counterparts, these other stem cells could be induced to take on NMP identity after exposure to Fgf and a brief pulse of Wnt, demonstrating that the developmental signals for the generation of NMPs are shared across multiple species. Having discovered the conditions necessary to generate bona fide NMPs in vitro, Gouti et al. next took advantage of this system to investigate how Wnt and Fgf contribute to the developmental fate of embryonic spinal cord and trunk mesoderm. Ultimately, they showed that the mesoderm marker Brachyury is necessary to maintain the dual developmental potential of NMPs. In its absence, Wnt signaling was no longer able to promote mesoderm fate, and instead, cells with spinal cord identity were always formed. Together, these experiments advance our understanding of the cells that give rise to embryonic spinal cord and trunk mesoderm development. Gouti M, Tsakiridis A, Wymeersch FJ, Huang Y, Kleinjung J, et al. (2014) In Vitro Generation of Neuromesodermal Progenitors Reveals Distinct Roles for Wnt Signalling in the Specification of Spinal Cord and Paraxial Mesoderm Identity. doi:10.1371/journal.pbio.1001937

doi: 10.1371/journal.pbio.1001922pmid: 25093404

Spider silk is wonderful stuff—light as the breeze and stretchy yet stronger than steel. People can manufacture synthetic fibers, such as Kevlar, that come close but can't begin to match the process spiders use. Their silk proteins, called spidroins, rapidly convert from the soluble form to solid fibers at ambient temperatures and with water as the solvent. Not only is this beyond us, we don't even know how spiders do it. Now, in this issue of PLOS Biology, new research by Anna Rising, Jan Johansson, and colleagues shows that silk formation involves structural shifts at either end of the spidroin and that these shifts are completely different, overturning the hypothesis that these protein terminals play similar roles. Spidroins are big proteins of up to 3,500 amino acids that contain mostly repetitive sequences, and the nonrepetitive N- and C-terminal domains at opposite ends are thought to regulate conversion to silk. These terminal domains are unique to spider silk and are highly conserved among spiders. Spidroins have a helical and unordered structure when stored as soluble proteins in silk glands, but, when converted to silk, they contain β-sheets that lend mechanical stability. We know that there is a pH gradient across the spider silk gland, which narrows from a tail to a sac to a slender duct, and that silk forms at a precise site in the duct. However, further details of spider silk production have been elusive. To discover the basis for the pH gradient in the silk gland, the researchers studied the orb weaver spider Nephila clavipes, which reaches 7.6 centimeters wide and spins webs up to 3 meters across. Poking ion-selective microelectrodes into the silk gland revealed that the pH fell from 7.6 to 5.7 between the base of the tail and halfway down the duct, which was as far as the electrodes fit. This pH gradient is much steeper than previously thought. The microelectrodes also showed that bicarbonate ions rise between the base of the tail and the far end of the sac. Taken together, these ion patterns suggest that the pH gradient is due to carbonic anhydrase, an enzyme that converts carbon dioxide and water to bicarbonate and hydrogen ions. In support of this idea, a carbonic anhydrase inhibitor called methazolamide collapsed the pH gradient. Next, the researchers investigated how the pH range across the silk gland affects silk proteins, using N- and C-terminal domains isolated from the spidroins of another orb weaver, Araneus ventricosus (Figure 1). They found that both domains undergo structural changes at the pH found in the duct. Importantly, this is also where carbonic anhydrase activity is concentrated. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 1. An Araneus spider in its orb web. Image credit: Anna Rising. https://doi.org/10.1371/journal.pbio.1001922.g001 The researchers also found pH had opposite effects on the two domains' stability to temperature and urea, which was a surprise given that the domains had been suggested to have a similar impact on silk formation. The N-terminal dimerized at pH 6—which is found in the beginning of the duct—and became increasingly stable as the pH dropped along the duct. In contrast, the C-terminal domain destabilized as the pH dropped, gradually unfolding until it formed the β-sheets characteristic of silk at pH 5.5. What makes the C-terminal domain unfold? This end of the spidroin has a salt bridge joining two helices, and mutations that interfere with this bridge also destabilize the C-terminal domain. Underscoring its importance, the salt bridge is conserved in spidroins from several spider species. The researchers believe that protonating amino acids in the salt bridge would also destabilize the C-terminal domain, facilitating conversion of spidroins from helices to β-sheet fibrils. These findings led the researchers to propose a new “lock and trigger” model for spider silk formation. Gradual dimerization of the N-terminal domains could lock spidroins into multimers, while the β-sheet fibrils at the C-terminals could serve as nuclei that trigger rapid polymerization of spidroins into fibers. Interestingly, the C-terminal β-sheets are similar to those in the amyloid fibrils characteristic of diseases such as Alzheimer disease. This work brings us closer to unraveling the mystery of spider silk, explaining how it can form so quickly—faster than a meter per second—as well as how its formation can be confined to the spinning duct. Moreover, because the N- and C-terminal domains of spidroins are found nowhere else, this lock and trigger formation is likely unique to spider silk. Besides being essential to producing biomimetic spidroin fibers, knowing how spiders spin silk could give insights into natural ways of hindering the amyloid fibrils associated with disease. Andersson M, Chen G, Otikovs M, Landreh M, Nordling K, et al. (2014) Carbonic Anhydrase Generates CO2 and H+ That Drive Spider Silk Formation Via Opposite Effects on the Terminal Domains. doi:10.1371/journal.pbio.1001921

doi: 10.1371/journal.pbio.1001927pmid: 25116919

The deepest branch on the evolutionary tree is between bacteria and archaea (eukaryotes arrived late, probably as a fusion of the two). These two domains of life share fundamental aspects of their biochemistries, including a genetic code and their reliance on cross membrane proton gradients to drive adenosine triphosphate (ATP) synthesis via the ATP synthase enzyme. Yet, the membranes themselves are profoundly different. In bacteria, glycerol phosphate links to an unbranched fatty acid via an ester bond (forming the same phospholipid structure found in all eukaryotes), while in archaea it links to a (typically) branched isoprenoid via an ether bond. The stereochemistry of the resulting molecules also differs, all of which strongly suggests that these membrane components evolved after the two diverged from the last universal common ancestor (LUCA). As logical as it is, this conclusion has met with much head-scratching among evolutionary biologists, since it suggests that LUCA itself did not possess a modern membrane. However, if that is the case, how did it harvest energy, and what could drive its archaeal and bacterial progeny to diverge in the structure of their membrane lipids? In this issue of PLOS Biology, Victor Sojo, Andrew Pomiankowski, and Nick Lane develop a mathematical model of bioenergetics in a LUCA-like cell, with a membrane lipid that could be a precursor of both types of modern membranes. Leakiness is a critical feature of this membrane, and the authors offer a plausible scheme for its evolution into the two distinct membranes found in bacteria and archaea today. In modern cells, the proton gradients that drive ATP synthesis are generated by proton pumps in the membrane. However, like the membranes themselves, these pumps differ in archaea and bacteria. One possibility is that LUCA could have used proton gradients but not generated them itself and therefore might have relied on natural proton gradients. However, that leads straight to another problem. The influx of protons down a natural gradient can’t go on forever, or even very long, since the buildup of positive charge halts the electrostatic drive behind the process—hence, the authors reasoned, the value of a leaky membrane. Entering protons could either leak out or be neutralized by hydroxide ions leaking in, thereby maintaining the gradient and the ability to make ATP. Fatty acids without their glycerol phosphate head groups form just such a leaky membrane, since protons attaching to the weak negative charge on the fatty acid head neutralize it, allowing it to randomly flip back and forth within the membrane, carrying the proton out with it. As a source of proton gradients, LUCA most likely relied on naturally occurring pH differences like those found in the oceans, where alkaline fluids seep from deep sea vents into relatively acidic seawater. The model they built then posited a cell in contact with a constant flow of protons on one side (from seawater), a constant flow of alkaline fluid on the other (potentially from a vent), and a leaky membrane containing an ATP synthase. They found that with a 3-unit pH gradient (i.e., a 1000-fold concentration gradient of protons) and the ATP synthase comprising 1% of the membrane, the cell could drive synthesis of ATP. Movement away from deep sea vents into environments without natural proton gradients would have been impossible for this type of leaky cell, but the authors suggest that a second membrane protein—a sodium-proton antiporter (SPAP)—could have laid the first steps. SPAP exchanges a positively charged proton on one side of the membrane for a positively charged sodium on the other. Sodium doesn’t neutralize the fatty acid in the same manner and consequently leaks through much less. The continuous flux of protons through SPAP could therefore add a sodium gradient to the natural proton gradient, giving cells more power and therefore allowing them to survive on smaller proton gradients, facilitating early spread and divergence. SPAP is found in both archaea and bacteria, and the authors suggest it arose in LUCA. Modern cells don’t just exploit proton gradients; they actively create them as well, and here, the existence of SPAP emerges as a crucial preadaptation for the evolution of the modern cell. In the absence of SPAP, pumping protons across a leaky membrane gives no benefit, as most of the protons immediately leak back through the membrane. Making the membrane less leaky doesn’t help, as that cuts off cells from the natural proton gradient, undermining their power. However, pumping in the presence of SPAP does pay. For every proton pumped out, a little extra energy is retained in a sodium gradient. That makes pumping with SPAP beneficial even with a leaky membrane. The authors showed that it reaped even bigger rewards as membranes tightened, giving for the first time an advantage to modern phospholipid membranes. LUCA’s membrane may have contained both fatty acids and isoprenoids, and tightening is the natural consequence of adding a glycerol phosphate head to either membrane lipid. Based on their model, the authors suggest that the stereochemical differences between archaea and bacteria arose randomly in separate populations (which diverged with the benefit of SPAP). The authors point out that both the archaeal version and the bacterial version of glycerol phosphate arise from the same precursor (dihydroxyacetone phosphate). The two opposing stereochemistries arise from attacking the central atom from two opposing sides, likely a random event initially but one that eventually became fixed in separate populations. Whether or not this model is correct in detail, it offers a robust and potentially testable hypothesis that goes a long way in explaining why archaea and bacteria are so alike and yet so different. Sojo V, Pomiankowski A, Lane N (2014) A Bioenergetic Basis for Membrane Divergence in Archaea and Bacteria. doi:10.1371/journal.pbio.1001926 Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 1. Why do bacteria and archaea differ so fundamentally in their membranes and other traits? It makes more sense if they started out living on natural proton gradients. https://doi.org/10.1371/journal.pbio.1001927.g001

doi: 10.1371/journal.pbio.1001932pmid: 25137065

Introduction The concept of the cortical “column” is one of the few organising principles for cortical circuits that we have, yet the characteristic orientation columns in the primary visual cortex (V1) of the cat and monkey appear to be completely absent in rodent V1. In place of the ordered maps of orientation seen in cat and monkey, the distribution of orientation preferences in rodent V1 appears to be essentially random [1]–[3]. This “salt-and-pepper” arrangement in the rodent must reflect differences in the wiring of superficial layer neurons in rodents compared to cat and monkey. Another striking difference between V1 of mouse and those of cat and monkey is the tuning properties of inhibitory neurons. While in cat and monkey the receptive fields of smooth (putative GABAergic inhibitory) neurons are typically orientation selective [4]–[10], with only occasional exceptions [11], in the mouse they are essentially weakly tuned [12],[13] (but see Runyan and colleagues [14]). A third striking difference is that neurons in mouse V1 receive many more synapses on average [15],[16] than a neuron in primary visual cortex of cat [17],[18] or monkey [15],[19]–[21]. In rodent barrel cortex a significant proportion of these synapses are probably contributed by neighbouring pyramidal cells, which form their synapses on the basal dendrites [22],[23]. In cat, superficial layer pyramidal neurons are estimated to receive more than 60% of the excitatory synapses from their neighbouring pyramidal neurons [24]. This suggests that positive feedback loops are more likely between superficial layer pyramidal cells than between pyramidal cells in other layers, whose principal projections tend to project out of their home layers (see Douglas and Martin [25]). By implication, the large number of excitatory synapses per neuron in the mouse may require a stronger component of recurrent inhibition. Clear evidence for an enhanced inhibitory component in the recurrent circuit came from a recent ultrastructural study by Bock et al. [26] designed to investigate whether the broadly tuned receptive fields of GABAergic inhibitory neurons could be explained by the convergence of input from excitatory neurons with different orientation preferences. This work involved partial reconstruction of 13 pyramidal cells and one smooth (putative GABAergic inhibitory) neuron in a single 50 µm thick section of V1 from a mouse that had undergone calcium imaging in vivo [26]. Their main conclusion was that pyramidal cells of different orientation preferences converged on individual smooth neurons. No synapses were formed between any of the 13 pyramidal cells. A remarkable statistic from Bock et al. was that 51% of the synapses formed by the pyramidal axons were targeting smooth neurons. This is a staggeringly high proportion, and it implies a very different wiring strategy from the cat or monkey V1, where the proportion of excitatory synapses formed by layer 2/3 pyramidal neurons with smooth neurons is 5% [27] and 19% [28], respectively. The results of Bock et al. thus raise the question of whether this arises because there are proportionately more smooth neuron targets in the mouse, or whether pyramidal cells select smooth neurons as their targets in a way they do not in the cat or monkey. To answer these questions we made detailed analyses not just of the synaptic targets of superficial layer pyramidal cells, but also of the content of the neuropil in mouse V1 in our material, and made the same analyses of the neuropil in the material of Bock et al. [26]. Our analyses indicate that the superficial pyramidal cells do not connect randomly to dendrites in the neuropil, as Braitenberg and Schüz [29] have proposed (and named “Peters' rule”), but instead form a far higher proportion of their synapses with neighbouring smooth neurons than would be expected by chance. Results Our goal was to determine the proportion of pyramidal cell synapses with smooth and spiny neurons and to determine whether the composition of the neuropil reflected the proportion of pyramidal axon targets found experimentally. To replicate the results of Bock et al. [26], we used the same mouse strain and, as they did, used in vivo calcium imaging with two-photon microscopy (2PM) to record the responses of V1 neurons to visual stimulation. In contrast to the Bock study, however, we did not reconstruct the unlabelled axons by serial section electron microscopy (EM). Instead, we reduced the load of EM reconstruction considerably by filling individual imaged neurons with biocytin by electroporation after their functional characterization using 2PM, and making a correlated light/electron microscopic examination of their axons (Figure 1). Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 1. Targeted electroporation of functionally identified neurons in mouse V1. (A) Upper panel: Example two-photon image of V1 (286×286 µm field of view); neurons were labelled with the calcium indicator OGB-1/AM (green) and astrocytes with sulforhodamine (SR101, red). The dotted squares show the limits of the areas seen in (B) and (C). Lower panel: Averaged calcium signals of the selected neuron in response to drifting gratings. Scale bars: horizontal: 10 s, vertical: 20% ΔF/F. (B) Targeted electroporation of the neuron shown in (A). Upper panel: Two-photon image of the red channel before electroporation; the targeted neuron appears as a black hole. The glass pipette used for electroporation, filled with biocytin and Alexa Fluor 594 (see Materials and Methods) appears in white. Lower panel: Same picture after electroporation. (C) Calcium imaging after electroporation. The electroporated neuron contains OGB-1/AM and Alexa Fluor 594 and consequently appears in orange. Lower panel: Averaged calcium signals of the electroporated neuron. Scale bars: horizontal: 10 s, vertical: 5% ΔF/F. Notice that the neuron has kept its orientation tuning. The dotted square shows the limits of the area seen in (B). (D) Side view of blood vessels and the electroporated cell in a two-photon image stack in vivo (left) and in a post-hoc light microscopy stack (right). The arrows indicate recognizable common features between the two stacks. (E) Blood vessels reconstruction from the two-photon stack (upper panels) and the light microscopy stack (lower panels). Left pictures represent a top view of the reconstruction, right pictures are side views. Notice the similarity between the two-photon and the light microscopy reconstructions. All scale bars are 20 µm. https://doi.org/10.1371/journal.pbio.1001932.g001 We successfully characterized the morphology and synaptic ultrastructure of six neurons in layer 2/3 of five mice. The visual tuning properties were obtained for five neurons in five mice. In one mouse, a second serendipitously filled neuron was also reconstructed. Figure 1 shows the steps from imaging to electroporation to recovering the functionally characterised neuron. After 2PM calcium imaging of the neuronal population and functional characterization using drifting gratings, a reliably responsive and selective neuron was selected for electroporation. The white arrow in Figure 1A shows the neuron selected, which was tuned for vertically oriented drifting gratings (Circular Variance Index [CVI] = 0.62; Direction Selectivity Index [DSI] = 0.17). We then used targeted electroporation to label this neuron (Figure 1B). We observed that the electroporated neuron maintained its selectivity and responsiveness (Figure 1C, Figure S1). By aligning a stack of 2PM images with the serial 80 µm thick sections imaged with bright field light microscopy (LM), the recorded neuron was identified and recovered for morphological examination (Figure 1D, E). Figure 2 shows the full extent of the axon contained within the single 80 µm thick section containing the cell body. The position of the soma and the distribution of the axons were very similar to those reconstructed by Bock et al. [26]. The position of the cell body relative to the laminae is indicated by a triangle and the laminar borders are indicated with dashed and dotted lines. The circles on the black axons show the positions of the boutons that formed synapses. Filled circles indicate synapses formed with spines and open circles indicate synapses formed with dendritic shafts. Arrowheads indicate boutons where the composition of the surrounding neuropil was analysed. The traces below each reconstruction are averages from calcium imaging and show that all five neurons were orientation tuned and/or directionally biased. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 2. Structural and functional mapping of layer 2/3 pyramidal neurons. Each panel (A–F) shows the morphology and physiology of each of the single neurons analysed in this study. It contains a camera lucida drawing of the axonal arbour contained within a single 80 µm section (black lines). The triangles mark the location of the soma and the circles mark the location of the axonal varicosities investigated with light-electron correlated microscopy. The layer 1–2 border is displayed as a dashed line and the layer 3–4 border as a dotted line. The number of dendritic spine and shaft targets was counted in the neuropil surrounding the varicosities indicated with arrowheads. The response of the neurons to oriented gratings is displayed under the reconstruction together with the Circular Variance Index (CVI) and the Direction Selectivity Index (DSI). Black lines denote mean responses and gray lines individual trials. https://doi.org/10.1371/journal.pbio.1001932.g002 Serial ultrathin sections were taken through the axon to examine 21–31 boutons per neuron. These segments of the axon were correlated with the light microscopy (LM) reconstructions to define the precise position of the synapses and their targets along the axon. A total of 163 boutons were investigated in the EM. They formed a total of 170 synapses (148 boutons formed one synapse, 11 boutons formed two synapses, and four boutons formed no synapses). The different targets of the pyramidal axons were classified by standard criteria [27]. Figure 3 shows two examples of spines (Figure 3A, B) forming synapses with the biocytin-labelled boutons, which are electron-dense and filled with vesicles. The large postsynaptic density (arrow head) indicates a typical asymmetric synapse formed by pyramidal neurons. Two unlabelled vesicle-filled boutons forming asymmetric synapses are also indicated in Figure 3A (arrowheads) for comparison. In Figure 3C the bouton formed an asymmetric synapse with the dendritic shaft of a smooth neuron. Unlike the dendrites of spiny neurons, where most asymmetric synapses are formed with spines, the asymmetric synapses formed with dendrites of smooth (i.e. spine-free) neurons are naturally found on the shafts (arrow heads in Figure 3C). In all cases the classification of the targets was made on the basis of serial section analyses of the postsynaptic dendrite. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 3. Electron micrographs of labelled boutons forming synapses with dendritic spines (A, B) and a dendritic shaft (C). Black arrowheads indicate synapses. https://doi.org/10.1371/journal.pbio.1001932.g003 By reconstructing the axons at the LM level, we were able to identify the particular branch segments that contained the synapses examined with subsequent EM. Figure 4 shows a summary dendrogram that reveals the branch ordering of the axons and the relative location of the 170 synapses examined on the axons. As in Figure 2, the target type is indicated by closed circles for spines and open circles for dendritic shafts. The axon leaving the soma descends vertically before branching and forming collaterals with boutons in layers 2 and 3. The dendrogram shows that the synapses we sampled were found on all orders of the branches, even on the main descending axon. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 4. Summary dendrogram of layer 2/3 axon with the location of investigated synapses per branch order. Circles indicate the location of the investigated synaptic boutons. Colours represent different neurons; targeting of dendritic spines is indicated by filled circles and dendritic shafts by empty circles. https://doi.org/10.1371/journal.pbio.1001932.g004 A total of 126 synapses were formed with spiny neurons (120 formed with dendritic spines and 6 with dendritic shafts) and 44 with dendritic shafts of smooth neurons. The data for each neuron in terms of target type are plotted in the histograms of Figure 5A. These histograms show that although spines formed the majority of targets, the variance between individual neurons was surprisingly high. If Peters' rule [29] applied, we would expect the proportion of different targets to reflect the local average proportions of smooth and spiny neurons in layer 2/3. The question was whether this high variance reflected some local heterogeneities in distribution of targets in the neuropil, or whether it was due to specific targeting of smooth neurons by some pyramidal cells. We tested this using the unbiased disector counting technique (see Materials and Methods [30],[31]) to determine the distribution of asymmetric synapses formed with spiny or smooth neurons in the neuropil at the vicinity of each of the reconstructed neurons. These results show that in the neuropil many more synapses were formed with spines than were found for the labelled axons (Figure 5C). Next, we explored the possibility that the observed specificity was due to the fact that labelled boutons formed synapses in regions of the neuropil where there were more dendritic shaft targets. Again using the physical disector method we placed a 5 µm×5 µm sampling square centred on labelled boutons (randomly selected boutons indicated by arrowheads in Figure 2). The results plotted in Figure 5B show that in the region around any labelled bouton, virtually all the synapses were formed with spines. This was again different from the distribution of targets of the labelled boutons, which formed significantly more synapses with smooth neurons than Peters' rule would predict. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 5. Histograms showing the percentage of the different types of post-synaptic targets of (A) layer 2/3 labelled neurons, (B) unlabelled boutons in the neuropil surrounding labelled boutons, and (C) unlabelled boutons from random locations in the neuropil. Targets were considered as smooth dendritic shaft, spiny dendritic shaft, and dendritic spine. https://doi.org/10.1371/journal.pbio.1001932.g005 Finally, to test whether the difference in targeting between labelled axons and the unlabelled neuropil could be due to a random process, we ran a simulation of an axon growing through a virtual neuropil and connecting to its targets by chance. The location of targets in the neuropil was uniformly distributed, as found in the large disectors (30 µm×30 µm), and each simulation was performed 10,000 times. When the simulations used the percentage of smooth dendritic targets collected from locations surrounding labelled boutons (Figure 5B), the Monte Carlo analysis (Figure S2) revealed that, with the exception of neuron M20 (p = 0.077) the other neurons showed a strong statistical difference (p = 0) between the number of inhibitory targets observed experimentally and that predicted from a random process. When the simulations used the percentage of smooth dendritic targets collected from random locations in the neuropil (Figure 5C), the Monte Carlo analysis (Figure S3) revealed that, with the exception of neuron M31 (p = 0.48), the other neurons showed strong a statistical difference (p = 0) between the number of inhibitory targets observed experimentally and that predicted from a random process. We also tested whether the biases observed by Bock et al. [26] followed the same trend as our data. We applied the unbiased disector method on their web-based data to estimate the proportion of synapses formed with spiny and smooth neurons in the neuropil of the superficial cortical layers of their mouse (pie charts in Figure 6, right column). We found that 80% of the targets were on spiny dendrites (35 synapses on spines and one on a spiny shaft) and 20% on smooth dendrites (nine synapses). Our analyses of their data indicate that the axons contained in their reconstructed volume targeted far more smooth neurons than would be expected from the composition of the neuropil through which they passed (compare Figure 6, data from imaged neurons in the lower pie chart with disector counts in the upper pie chart). Thus, although on average our labelled neurons formed proportionately fewer synapses with smooth neurons than did those of Bock et al. [26], in both studies the proportion of targeted smooth neurons was far higher than would be expected on the basis of random connectivity. Thus the data from both studies indicate that these superficial layer pyramidal cells in mouse V1 appear to select smooth neurons as their targets. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 6. (A) Summary diagram of the distribution of post-synaptic targets in layer 2/3 of mouse visual cortex. (B) Comparison with the study of Bock et al. (2011) [26]. https://doi.org/10.1371/journal.pbio.1001932.g006 Discussion Our goal was to establish whether the salt-and-pepper representation of orientation in rodent V1 is reflected in the synaptic connections formed by the superficial layer pyramidal cells. After 2PM calcium imaging, individual pyramidal neurons were labelled with biocytin, sectioned, and reconstructed with LM. The synaptic targets of their axons as well the synaptic complement of the surrounding neuropil were quantified using EM. Previous physiological studies suggested that pyramidal cells connect specifically to one another [32]–[36] and to GABAergic neurons [37]. Moreover, in mouse V1, the probability of pyramidal neurons connecting to neighbouring fast-spiking interneurons is much higher than the probability of pyramidal neurons connecting to each other [38]. Our data further indicate that pyramidal cells make specific connections with smooth, putative GABAergic neurons. A previous combined 2PM calcium imaging and electron microscopy study by Bock et al. [26] of 13 pyramidal cells in one mouse indicated that the pyramidal cells formed a consistently high proportion (50%) of their synapses with smooth, putative GABAergic neurons. This is an astonishingly high fraction, since more extensive analyses of superficial layer pyramidal cells in V1 of other species indicate that typically 20% or fewer of the synapses are formed on smooth neurons. One explanation for the data from Bock et al. might be that the neuropil of mouse V1 contains a higher proportion of smooth neurons than other species. This seems not to be the explanation since no major differences have been noted in the proportion of pyramidal cells and smooth neurons in the superficial layers of rodent, cat, or monkey V1 [20],[39],[40]. The critical question is then whether the result obtained from the single section in one mouse by Bock et al [26] is an outlier, or whether it really reflects a wiring strategy to increase the local component of recurrent inhibition in V1. The composition of the neuropil based on our own samples and those of Bock et al. [26] indicates that the superficial layer pyramidal cells in mouse V1 form a significantly higher proportion of their synapses with smooth, putative inhibitory neurons than would be predicted by Peters' rule [29], which assumes that axons and dendrites connect in the proportions in which they are found in the neuropil. Our disector counts indicated that virtually all the unlabelled synapses in the neuropil within a 5 µm radius of any of the labelled pyramidal cell boutons were formed with spines, not smooth dendrites. Bock et al. [26] concluded that geometry dominates over function, since the proximity of two pyramidal cells, not their receptive field similarity, was the strongest indicator that their axons would converge onto a smooth neuron. However, our own results, and our new analyses of the cortical tissue from Bock et al., indicates that the pyramidal cell connections to smooth neurons are far from being determined purely by geometry, for if geometry were the sole determinant, the pyramidal neurons should connect to smooth neurons in proportion to their occurrence in the neuropil. Instead, some pyramidal cells preferentially formed a subset of their synapses with smooth neurons. What is unexplained, however, is why the variance across the pyramidal cells is so high. In this context it is noteworthy that all the pyramidal cells in the mouse of Bock et al. had consistently high proportions of smooth targets, as did the mouse in which we examined two pyramidal cells. This suggests that the source of variance might not be within the individual, but between individuals of the same strain. These interesting observations across the two studies raise both a warning and an interesting challenge as to how we might discover the principles by which mouse brain wires itself if such high variance does exist between individuals. Our results and those of Bock et al. have implications for the functional architecture of mouse visual cortex (Figure 7) and its operation. If layer 2/3 smooth neurons receive more than their fair share of synapses from local pyramidal cells than would be expected from Peters' rule, this implies that they receive proportionally fewer synapses from other excitatory projections into layers 2 and 3. These other excitatory inputs arise from spiny neurons in layer 4 and 5 of V1 as well as other cortical areas and subcortical nuclei, like the thalamus. By this argument, pyramidal cells then have proportionately fewer synapses to devote to connections to other pyramidal neurons in the same layer. If this is the case, then in layer 2/3 of mouse visual cortex one might expect proportionately less recurrent excitation from within these layers than is present in the cat. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 7. Comparison between the circuit of layer 2/3 of visual cortex of cat and monkey with the circuit of the mouse. Inhibitory neurons are represented as blue disks and excitatory neurons as orange triangles. Arrows indicate synaptic connections, and the thickness of the arrow represents its weight in terms of number of connections. Black bars represent the orientation preference of each of the neurons, and a black disk indicates an untuned or poorly tuned neuron. https://doi.org/10.1371/journal.pbio.1001932.g007 Smooth neurons, like basket cells and chandelier cells, form their axonal arbours largely within the same layer as the cell body. Therefore, the smooth neurons targeted by our labelled pyramidal cell axons most likely are recurrently inhibiting the pyramidal cells that excite them. The fact that in mouse the smooth neurons have more convergent input from neurons with a variety of orientation tuning produces a circuit configuration that is very reminiscent of a winner-take-all (WTA) circuit. In this circuit, excitatory neurons have a map of some parameter (e.g. orientation [35],[36]), and the inhibitory neurons receive input from all excitatory neurons in the map and provide inhibition proportional to the overall excitation in the circuit [41]. While this study focuses on mouse V1, previous work on superficial layer pyramidal cells in V1 of cat and monkey gave dramatically different results to those presented here. In monkey V1, McGuire et al. [28] have shown that the axons of intracellularly filled layer 2/3 pyramidal neurons formed 19% of their targets with smooth dendritic shafts (28% if one considers spiny and dendritic shafts). As we did, McGuire et al. [28] analysed the neuropil surrounding one of their neurons, but unlike us found no evidence for preferential targeting of smooth GABAergic neurons by the superficial layer pyramidal neurons (see also Beaulieu and colleagues for other counts of targets in monkey V1 neuropil [20]). In cat V1, Kisvarday et al. [27] found that the axons of intracellularly filled layer 3 pyramidal neurons formed only 5% of their synapses with GABAergic neurons. They did not analyse the neuropil surrounding the labelled neurons, but in a different study Beaulieu and Colonnier analysed the neuropil of cat layer 2/3 of and found that 18% (mean of layers 2, 3A, and 3B) of the asymmetric synapses are formed with dendritic shafts, some of which may be of spiny neurons [18]. These data strongly suggested that in cat there is no preferential targeting of inhibition by layer 2/3 pyramidal neurons, unlike what we, and Bock et al. [26], now find for mouse V1. This was also the conclusion of a theoretical study by Stepanyants and colleagues [42], who found that the results of Kisvarday et al. [27] were consistent with Peters' rule. The conclusion of Stepanyants et al. makes it very clear that in the cat the proportion of GABAergic smooth neurons that are targets of superficial layer pyramidal axons is well below that of the mouse V1. One idea for the generation of orientation “columns” in cat is that the orientation selectivity of neurons is created in layer 4 and then simply fed-forward to neurons in the superficial and deep layers [43]. In the macaque monkey, the situation is somewhat different, because most layer 4C neurons have non-oriented receptive fields, whereas neurons in the superficial and deep layers are orientation selective and form an orderly map of orientation, as in the cat. Development of the acolumnar salt-and-pepper arrangement of rodent V1 demands a high degree of specificity if it were to be achieved by feed-forward connections alone. Here, the stronger bias in the connections to smooth cells in the mouse may reflect increased demands on the inhibitory circuitry to shape the receptive field mediated by the superficial layer pyramidal cells. Materials and Methods Animal Preparation All animal procedures were carried out according to the guidelines of the University of Zurich, and were approved by the Cantonal Veterinary Office. C57BL/6 mice (2–4 months old, of either sex) were either first sedated with chlorprothixene (Sigma; 0.2 mg/mouse) and anaesthetized with urethane (0.5–1.0 g/kg) or anaesthetized by 2.7 ml/kg of a solution containing one part fentanyl citrate and fluanisone (Hypnorm; Janssen-Cilag, UK) and one part midazolam (Hypnovel; Roche, Switzerland) in two parts of water, both delivered by intraperitoneal injections. Atropine (0.3 mg/kg) and dexamethasone (2 mg/kg) were administered subcutaneously to reduce secretions and oedema. Lactate-Ringer solution was regularly injected subcutaneously to prevent dehydration. Pinch reflexes were used to assess the depth of anaesthesia. Two-Photon Guided Staining The location of the primary visual area, V1, was determined by stereotaxic coordinates (V1 monocular segment – 1.0 mm anterior to lambda and 2.5 mm lateral from the midline [44]) and confirmed by subsequent intrinsic imaging. Briefly, the skull above the estimated visual cortex was carefully thinned until a noticeable transparency of the bone was achieved. We then illuminated the cortical surface with 630-nm LED light, presented drifting gratings for 5 s, and collected reflectance images through a 4× objective with a CCD camera (Toshiba TELI CS3960DCL). Intrinsic signal changes were analysed as fractional reflectance changes relative to the pre-stimulus average. V1 was the largest area active during visual stimulation at a location in accordance with stereotaxic coordinates. After identification with intrinsic imaging, a small craniotomy (from 500 µm×500 µm to 1 mm×1 mm) was opened above V1, the dura removed and the exposed cortex superfused with artificial cerebrospinal fluid (ACSF) (135 mM NaCl, 5.4 mM KCl, 5 mM Hepes, 1.8 mM CaCl2, 1 mM MgCl2, pH 7.2, with NaOH). Calcium indicator loading was performed using the “multi cell bolus loading” technique [45]. Briefly, 50 µg of the acetoxymethyl (AM) ester form of the calcium-sensitive fluorescent dye Oregon Green BAPTA-1 (OGB-1; Invitrogen, Basel, Switzerland) were dissolved in 2 µl of DMSO plus 20% Pluronic F-127 (BASF, Germany) and diluted with 37 µl standard pipette solution (150 mM NaCl, 2.5 mM KCl, 10 mM Hepes, pH 7.2) yielding a final OGB-1 concentration of about 1 mM. 1 µl of Alexa Fluor 594 (Invitrogen; 2 mM stock solution in distilled water) was added for visualization of the pipette during 2PM guided staining. The dye was pressure ejected under visual control through a glass pipette (4–5 MΩ) at a depth between 150–300 µm to stain layer 2/3 neurons. Brief application of sulforhodamine 101 (SR101; Invitrogen) to the exposed neocortical surface resulted in co-labelling of the astrocytic network [46]. Following dye injection the craniotomy was filled with agarose (type III-A, Sigma; 1% in ACSF) and covered with an immobilized glass cover slip. Visual Stimulation Visual stimuli were presented on a 7-inch TFT monitor (75 Hz refresh rate) 7 cm in front of the right eye roughly at 60° along the body axis of the anesthetized mouse. For the majority of the study, the visual stimuli were full contrast square wave gratings generated by the VisionEgg software [47] moving for 3 s in eight different directions spaced by 5 s blank (grey screen presentation). The temporal frequency (TF) was 0.5 to 1 Hertz (Hz) and spatial frequency (SF) was 0.02 to 0.05 cycles per degree (cyc/°), which have been shown to activate most neurons. For one animal, the stimulation used was full contrast square wave gratings moving back and forth during 4 s for each of four orientations. Two-Photon Calcium Imaging Calcium transients were acquired using a custom-built two-photon microscope equipped with a 40× water immersion objective (LUMPlanFl/IR; 0.8 NA; Olympus 2). 128×128 pixel frames or 256×256 pixel frames were acquired at rates from 2 to 4 Hz using custom written software (LabView; National Instruments, USA). Calcium Signal Analysis Data were analysed with ImageJ (National Institute of Mental Health, NIH) and MATLAB (Mathworks). Cells were detected manually by drawing a region of interest around cell bodies. Relative percentage changes in fluorescence (ΔF/F) were calculated using as baseline the blank just before each stimulation. Traces were filtered using a Savitzky-Golay filtering approach. Responses were calculated by averaging 3–6 points around the peak fluorescence change (time window of 1.5 s around the peak) for each stimulation epoch. Circular Variance We defined a selectivity criterion using circular variance over gratings responses. Circular variance is defined as , where θ is the average drift direction of the grating: This measure of circular variance combines aspects of amplitude modulation and tuning width and takes into account all the responses to each direction of drift [48]. To use it as an index comparable to orientation selectivity indexes, the values given in this manuscript are 1 – circular variance (see Niell and Stryker [49]) referred to in the text as CVI, for Circular Variance Index). Consequently, a perfectly tuned neuron would have a CVI value close to 1, and a perfectly untuned neuron close to 0. For further analysis of selectivity we used the Direction Selectivity Index (DSI; see below). Direction Selectivity We determined the direction selectivity as previously described [48],[49]. It is defined as:Where Rpref is the response at the preferred angle θpref and Ropposite is the responses at the opposite direction θpref+π. If DSI >0.5, the neuron is considered direction selective. Targeted Electroporation Glass pipettes of resistance from 4 to 6 MΩ were filled with a standard pipette solution containing 2%–5% biocytin. These concentrations of biocytin were reached by mixing 4% biocytin (ε-Biotinoyl-L-Lysine; Invitrogen) diluted in some cases with the red dye Alexa 594 (20 µM; Invitrogen) with a solution of 0.8 to 1.5% 5-(and-6)-Tetramethylrhodamine biocytin (Biocytin TMR; Invitrogen). The tip of the pipette was placed near the selected neuron for electroporation and a loose seal was formed to record extracellular spikes. Spikes were recorded at 5 kHz using a patch-clamp amplifier (npi, Reutlingen, Germany) and Spike2 software (CED, Cambridge, UK). Once a stable configuration was reached, pulses from 300 to 400 mV of 10 ms duration were applied until successful electroporation was verified visually by uptake of the red indicator dye. In addition, we verified in some experiments the viability of the neuron by retesting the responses to visual stimulation after a recovery period of 10–20 min. This recovery period allows sufficient time for the pores formed during the electroporation to reseal, which usually occurs within 1 min [50]. Perfusion and Histology At the end of the experiment the mouse was given an extra dose of anaesthesia and perfused transcardially with normal 0.9% NaCl solution, followed by a warm solution of 4% paraformaldehyde (w/v), 0.5% glutaraldehyde (v/v) and 15% saturated solution of picric acid (v/v) in 0.1 M PB pH 7.4. After fixation the mouse was perfused with solutions of 10%, 20%, and 30% sucrose in 0.1 M phosphate buffer (PB). Once the brain was removed it was allowed to sink in a 30% sucrose solution in 0.1 M PB to provide cryoprotection and then freeze-thawed in liquid nitrogen. The brains were then washed in 0.1 M PB for at least 2 h to allow them to recover from the shrinkage provoked by the incubation in sucrose solution. Sections containing V1 were cut at 80 µm in the coronal plane and collected in 0.1 M PB. After cutting, the sections were washed several times in buffer in order to remove any remaining fixative. To reveal biocytin the sections were washed in TBS and then incubated overnight (5°C) with an avidin-biotin complex (Vector ABC kit – Elite). The peroxidase activity was identified using 3-diaminobenzidine tetrahydrochloride (DAB) with nickel intensification. After assessment by LM, regions of tissue containing the imaged area were treated with 1% osmium tetroxide in 0.1 M PB, dehydrated through alcohols (1% uranyl acetate in the 70% alcohol) and propylene oxide, and flat mounted in Durcupan (Fluka) on glass slides. Postmortem Light and Electron Correlated Microscopy Serial light micrographs were taken from the osmicated sections at different magnifications, and the blood vessel pattern surrounding labelled neurons was reconstructed using TrakEM2 [51]. A similar blood vessel reconstruction was done on the 2PM stacks acquired in vivo. These reconstructions were used to find the recorded neurons in the osmicated histological sections. Finally the micrographs taken from histological sections were superimposed on the 2PM images to confirm the correspondence of the recorded neurons. The dendritic arbour and the proximal axon the neurons of interest were then reconstructed first in 2D using a drawing tube attached to a light microscope, and then in 3D from serial light micrographs using TrakEM2 [51]. Afterward, the tissue was serially resectioned at 50 nm thickness and collected on Pioloform-coated single slot copper grids. The axons of labelled neurons were then found in the ultrathin sections, and synapse connectivity between labelled axons and neuropil targets investigated with transmission electron microscopy (TEM). Synapses and associated structures were classified using conventional criteria [52],[53]. Counts of Dendritic Targets Estimations of the percentage of dendritic targets (spines or shafts) were performed at the EM level using the physical disector method [30]. The disector was composed of two serial sections of known thickness (50 nm) separated by one intervening section. Synapses that disappeared from reference to lookup section were counted and the target was classified as dendritic spine or shaft as in [54]. Both sections were used as reference and lookup doubling the number of disectors per site. Electron micrographs were collected at a magnification of (13,500×, pixel size 2.5 nm) with a digital camera (11 megapixels, Morada, Soft Imaging Systems). Four sets of counts were performed. The first set was done on randomly selected location in the neuropil surroundings the labelled neurons. The disectors had a size of 5 µm×5 µm and were sampled from the first intact section of every fourth grid (each grid contained eight sections on average). The sampling sites (five sites per animal) and grids were selected according to a systematic random sampling scheme [31],[55]. The second set was done on the neuropil surrounding the labelled boutons of recorded neurons. The counts were done in six randomly selected boutons per neuron and the disectors had a size of 5 µm×5 µm. The third set was from a single animal and the exact 2D location of the synapses was also collected for use in the Monte Carlo simulations described below. Three randomly located large disectors (size 30 µm×30 µm) were collected. The fourth set was collected from the dataset of Bock et al. [26] which was made available through CATMAID [56]. The disectors had a size of 12.7 µm×6.8 µm and the sampling sites (eight sampling sites) and sections were chosen according to a systematic random sampling scheme. Monte Carlo Simulations A Monte Carlo analysis was performed to test whether the observed statistics of synaptic targets by labelled axons could be due to a random process. We ran a simulation in MATLAB (Mathworks) of an axon growing through a virtual neuropil of size 200 µm×200 µm×200 µm. Each simulation was run 10,000 times with the parameters from each labelled neuron/neuropil and was terminated when the virtual axon reached the number of synapses reconstructed for each labelled neuron. The result of each simulation was the proportion of smooth dendritic targets on the virtual axon. The location of targets in the virtual neuropil followed a uniform distribution as found in the biological data obtained from three large disectors (30 µm×30 µm). The proportion of spiny and smooth dendritic targets in the virtual neuropil were taken from the counts shown in Figure 5 and the density of synapses used was 109 synapses/mm3 following the work of Schüz and Palm [16]. The p-value estimate was given by the proportion of simulations that showed results larger than or equal to the measurements made in the real neurons. Animal Preparation All animal procedures were carried out according to the guidelines of the University of Zurich, and were approved by the Cantonal Veterinary Office. C57BL/6 mice (2–4 months old, of either sex) were either first sedated with chlorprothixene (Sigma; 0.2 mg/mouse) and anaesthetized with urethane (0.5–1.0 g/kg) or anaesthetized by 2.7 ml/kg of a solution containing one part fentanyl citrate and fluanisone (Hypnorm; Janssen-Cilag, UK) and one part midazolam (Hypnovel; Roche, Switzerland) in two parts of water, both delivered by intraperitoneal injections. Atropine (0.3 mg/kg) and dexamethasone (2 mg/kg) were administered subcutaneously to reduce secretions and oedema. Lactate-Ringer solution was regularly injected subcutaneously to prevent dehydration. Pinch reflexes were used to assess the depth of anaesthesia. Two-Photon Guided Staining The location of the primary visual area, V1, was determined by stereotaxic coordinates (V1 monocular segment – 1.0 mm anterior to lambda and 2.5 mm lateral from the midline [44]) and confirmed by subsequent intrinsic imaging. Briefly, the skull above the estimated visual cortex was carefully thinned until a noticeable transparency of the bone was achieved. We then illuminated the cortical surface with 630-nm LED light, presented drifting gratings for 5 s, and collected reflectance images through a 4× objective with a CCD camera (Toshiba TELI CS3960DCL). Intrinsic signal changes were analysed as fractional reflectance changes relative to the pre-stimulus average. V1 was the largest area active during visual stimulation at a location in accordance with stereotaxic coordinates. After identification with intrinsic imaging, a small craniotomy (from 500 µm×500 µm to 1 mm×1 mm) was opened above V1, the dura removed and the exposed cortex superfused with artificial cerebrospinal fluid (ACSF) (135 mM NaCl, 5.4 mM KCl, 5 mM Hepes, 1.8 mM CaCl2, 1 mM MgCl2, pH 7.2, with NaOH). Calcium indicator loading was performed using the “multi cell bolus loading” technique [45]. Briefly, 50 µg of the acetoxymethyl (AM) ester form of the calcium-sensitive fluorescent dye Oregon Green BAPTA-1 (OGB-1; Invitrogen, Basel, Switzerland) were dissolved in 2 µl of DMSO plus 20% Pluronic F-127 (BASF, Germany) and diluted with 37 µl standard pipette solution (150 mM NaCl, 2.5 mM KCl, 10 mM Hepes, pH 7.2) yielding a final OGB-1 concentration of about 1 mM. 1 µl of Alexa Fluor 594 (Invitrogen; 2 mM stock solution in distilled water) was added for visualization of the pipette during 2PM guided staining. The dye was pressure ejected under visual control through a glass pipette (4–5 MΩ) at a depth between 150–300 µm to stain layer 2/3 neurons. Brief application of sulforhodamine 101 (SR101; Invitrogen) to the exposed neocortical surface resulted in co-labelling of the astrocytic network [46]. Following dye injection the craniotomy was filled with agarose (type III-A, Sigma; 1% in ACSF) and covered with an immobilized glass cover slip. Visual Stimulation Visual stimuli were presented on a 7-inch TFT monitor (75 Hz refresh rate) 7 cm in front of the right eye roughly at 60° along the body axis of the anesthetized mouse. For the majority of the study, the visual stimuli were full contrast square wave gratings generated by the VisionEgg software [47] moving for 3 s in eight different directions spaced by 5 s blank (grey screen presentation). The temporal frequency (TF) was 0.5 to 1 Hertz (Hz) and spatial frequency (SF) was 0.02 to 0.05 cycles per degree (cyc/°), which have been shown to activate most neurons. For one animal, the stimulation used was full contrast square wave gratings moving back and forth during 4 s for each of four orientations. Two-Photon Calcium Imaging Calcium transients were acquired using a custom-built two-photon microscope equipped with a 40× water immersion objective (LUMPlanFl/IR; 0.8 NA; Olympus 2). 128×128 pixel frames or 256×256 pixel frames were acquired at rates from 2 to 4 Hz using custom written software (LabView; National Instruments, USA). Calcium Signal Analysis Data were analysed with ImageJ (National Institute of Mental Health, NIH) and MATLAB (Mathworks). Cells were detected manually by drawing a region of interest around cell bodies. Relative percentage changes in fluorescence (ΔF/F) were calculated using as baseline the blank just before each stimulation. Traces were filtered using a Savitzky-Golay filtering approach. Responses were calculated by averaging 3–6 points around the peak fluorescence change (time window of 1.5 s around the peak) for each stimulation epoch. Circular Variance We defined a selectivity criterion using circular variance over gratings responses. Circular variance is defined as , where θ is the average drift direction of the grating: This measure of circular variance combines aspects of amplitude modulation and tuning width and takes into account all the responses to each direction of drift [48]. To use it as an index comparable to orientation selectivity indexes, the values given in this manuscript are 1 – circular variance (see Niell and Stryker [49]) referred to in the text as CVI, for Circular Variance Index). Consequently, a perfectly tuned neuron would have a CVI value close to 1, and a perfectly untuned neuron close to 0. For further analysis of selectivity we used the Direction Selectivity Index (DSI; see below). Direction Selectivity We determined the direction selectivity as previously described [48],[49]. It is defined as:Where Rpref is the response at the preferred angle θpref and Ropposite is the responses at the opposite direction θpref+π. If DSI >0.5, the neuron is considered direction selective. Targeted Electroporation Glass pipettes of resistance from 4 to 6 MΩ were filled with a standard pipette solution containing 2%–5% biocytin. These concentrations of biocytin were reached by mixing 4% biocytin (ε-Biotinoyl-L-Lysine; Invitrogen) diluted in some cases with the red dye Alexa 594 (20 µM; Invitrogen) with a solution of 0.8 to 1.5% 5-(and-6)-Tetramethylrhodamine biocytin (Biocytin TMR; Invitrogen). The tip of the pipette was placed near the selected neuron for electroporation and a loose seal was formed to record extracellular spikes. Spikes were recorded at 5 kHz using a patch-clamp amplifier (npi, Reutlingen, Germany) and Spike2 software (CED, Cambridge, UK). Once a stable configuration was reached, pulses from 300 to 400 mV of 10 ms duration were applied until successful electroporation was verified visually by uptake of the red indicator dye. In addition, we verified in some experiments the viability of the neuron by retesting the responses to visual stimulation after a recovery period of 10–20 min. This recovery period allows sufficient time for the pores formed during the electroporation to reseal, which usually occurs within 1 min [50]. Perfusion and Histology At the end of the experiment the mouse was given an extra dose of anaesthesia and perfused transcardially with normal 0.9% NaCl solution, followed by a warm solution of 4% paraformaldehyde (w/v), 0.5% glutaraldehyde (v/v) and 15% saturated solution of picric acid (v/v) in 0.1 M PB pH 7.4. After fixation the mouse was perfused with solutions of 10%, 20%, and 30% sucrose in 0.1 M phosphate buffer (PB). Once the brain was removed it was allowed to sink in a 30% sucrose solution in 0.1 M PB to provide cryoprotection and then freeze-thawed in liquid nitrogen. The brains were then washed in 0.1 M PB for at least 2 h to allow them to recover from the shrinkage provoked by the incubation in sucrose solution. Sections containing V1 were cut at 80 µm in the coronal plane and collected in 0.1 M PB. After cutting, the sections were washed several times in buffer in order to remove any remaining fixative. To reveal biocytin the sections were washed in TBS and then incubated overnight (5°C) with an avidin-biotin complex (Vector ABC kit – Elite). The peroxidase activity was identified using 3-diaminobenzidine tetrahydrochloride (DAB) with nickel intensification. After assessment by LM, regions of tissue containing the imaged area were treated with 1% osmium tetroxide in 0.1 M PB, dehydrated through alcohols (1% uranyl acetate in the 70% alcohol) and propylene oxide, and flat mounted in Durcupan (Fluka) on glass slides. Postmortem Light and Electron Correlated Microscopy Serial light micrographs were taken from the osmicated sections at different magnifications, and the blood vessel pattern surrounding labelled neurons was reconstructed using TrakEM2 [51]. A similar blood vessel reconstruction was done on the 2PM stacks acquired in vivo. These reconstructions were used to find the recorded neurons in the osmicated histological sections. Finally the micrographs taken from histological sections were superimposed on the 2PM images to confirm the correspondence of the recorded neurons. The dendritic arbour and the proximal axon the neurons of interest were then reconstructed first in 2D using a drawing tube attached to a light microscope, and then in 3D from serial light micrographs using TrakEM2 [51]. Afterward, the tissue was serially resectioned at 50 nm thickness and collected on Pioloform-coated single slot copper grids. The axons of labelled neurons were then found in the ultrathin sections, and synapse connectivity between labelled axons and neuropil targets investigated with transmission electron microscopy (TEM). Synapses and associated structures were classified using conventional criteria [52],[53]. Counts of Dendritic Targets Estimations of the percentage of dendritic targets (spines or shafts) were performed at the EM level using the physical disector method [30]. The disector was composed of two serial sections of known thickness (50 nm) separated by one intervening section. Synapses that disappeared from reference to lookup section were counted and the target was classified as dendritic spine or shaft as in [54]. Both sections were used as reference and lookup doubling the number of disectors per site. Electron micrographs were collected at a magnification of (13,500×, pixel size 2.5 nm) with a digital camera (11 megapixels, Morada, Soft Imaging Systems). Four sets of counts were performed. The first set was done on randomly selected location in the neuropil surroundings the labelled neurons. The disectors had a size of 5 µm×5 µm and were sampled from the first intact section of every fourth grid (each grid contained eight sections on average). The sampling sites (five sites per animal) and grids were selected according to a systematic random sampling scheme [31],[55]. The second set was done on the neuropil surrounding the labelled boutons of recorded neurons. The counts were done in six randomly selected boutons per neuron and the disectors had a size of 5 µm×5 µm. The third set was from a single animal and the exact 2D location of the synapses was also collected for use in the Monte Carlo simulations described below. Three randomly located large disectors (size 30 µm×30 µm) were collected. The fourth set was collected from the dataset of Bock et al. [26] which was made available through CATMAID [56]. The disectors had a size of 12.7 µm×6.8 µm and the sampling sites (eight sampling sites) and sections were chosen according to a systematic random sampling scheme. Monte Carlo Simulations A Monte Carlo analysis was performed to test whether the observed statistics of synaptic targets by labelled axons could be due to a random process. We ran a simulation in MATLAB (Mathworks) of an axon growing through a virtual neuropil of size 200 µm×200 µm×200 µm. Each simulation was run 10,000 times with the parameters from each labelled neuron/neuropil and was terminated when the virtual axon reached the number of synapses reconstructed for each labelled neuron. The result of each simulation was the proportion of smooth dendritic targets on the virtual axon. The location of targets in the virtual neuropil followed a uniform distribution as found in the biological data obtained from three large disectors (30 µm×30 µm). The proportion of spiny and smooth dendritic targets in the virtual neuropil were taken from the counts shown in Figure 5 and the density of synapses used was 109 synapses/mm3 following the work of Schüz and Palm [16]. The p-value estimate was given by the proportion of simulations that showed results larger than or equal to the measurements made in the real neurons. Supporting Information Figure S1. Examples of responses before and after electroporation for the cells M20 and M21. Black traces are the averaged responses to drifting gratings. Stimulation onsets are indicated by orange dotted lines. https://doi.org/10.1371/journal.pbio.1001932.s001 (TIF) Figure S2. Histograms showing the distribution of the percentage of targets formed by virtual layer 2/3 axons with smooth dendrites obtained with Monte Carlo simulations. The percentage of the available smooth dendritic targets in the neuropil was taken from the disectors collected from locations surrounding labelled boutons (Figure 5B). https://doi.org/10.1371/journal.pbio.1001932.s002 (EPS) Figure S3. Histograms showing the distribution of the percentage of targets formed by virtual layer 2/3 axons with smooth dendrites obtained with Monte Carlo simulations. The percentage of the available smooth dendritic targets in the neuropil was taken from the disectors collected from random location in the neuropil (Figure 5C). https://doi.org/10.1371/journal.pbio.1001932.s003 (EPS) Acknowledgments We would like to thank Stefan Roth for providing the original code to run the Monte Carlo simulation and to German Koestinger for help implementing it. We would also like to thank the Clay Reid lab in Harvard, specially Davi D. Bock and Wei-Chung Lee, the Pittsburgh Supercomputing Center, and the Open Connectome Project for making the dataset from Bock et al. (2011) [26] available for our estimations of the percentage of dendritic targets.

doi: 10.1371/journal.pbio.1001935pmid: 25157590

Introduction Deleterious, but non-lethal mutations are constantly generated and can hitchhike with adaptive mutations [1]. Consequently, such deleterious alleles are widespread in eukaryotic populations [2],[3]. For example, as high as 12% of the coding SNPs in yeast populations are deleterious [2]. Many of the observed functional variation in this species yield proteins with compromised or no activities [2], or lead to complete loss of genes with significant contribution to fitness (Text S1). Deleterious loss-of-function variants may occasionally revert to wild type, eventually perish from the population, or become compensated by mutations elsewhere in the genome. The third possibility, termed compensatory evolution, is the focus of our study. Theoretical works suggest that mutant subpopulations can cross fitness valleys by the simultaneous fixation of a compensatory mutation in the population [4],[5]. This process can also work in large populations and is facilitated by linkage of the two alleles [5]. Compensatory evolution appears to be common at many levels of molecular interactions. It is involved in the maintenance of RNA and protein secondary structures, it mitigates the costs of antibiotic resistance [6],[7], and allows rapid fitness recovery in populations with accumulated deleterious mutation loads [7]–[9]. Compensatory regulatory mutations also act to stabilize gene expression levels across species [10],[11], and conserve DNA-encoded nucleosome organization [12]. The most detailed experimental analyses on compensatory mutations for fixed deleterious mutations were performed in DNA bacteriophages [8],[13]–[15], bacteria [16],[17], and Caenorhabditis elegans [7],[9]. Three major patterns emerged from these studies. As the target size for compensatory mutations is typically much larger than that for reversion, compensation is more likely than reversion of deleterious mutations [13]. The rate of compensatory evolution increased with the severity of the deleterious fitness effects, and was not limited to functionally interacting partners of the mutated gene [15]. As regards the potential pleiotropic effects of compensatory evolution, our knowledge is rather limited, not least because it demands detailed exploration of the underlying molecular mechanisms of compensation. Compensatory mutations may enhance fitness either by reducing the need for the gene with the compromised function, or by restoring the efficiency of the affected molecular function [18]. For compensation of fitness costs of antibiotic resistance conferring mutations, restoration of function was the most common mechanism [18], but in other systems the relative importance of functional substitution and restoration is unknown. In the case of functional restoration (e.g., by enhanced dosage of a redundant duplicate of the disrupted gene), one might expect limited pleiotropic fitness effects of compensatory mutations across environmental conditions. Compensatory evolution following gene loss is of special interest [17]. Gene loss may be initiated by genetic drift and/or selection through antagonistic pleiotropy [17],[19]. As reversion to the wild-type state is less likely, gene loss may promote genetic changes that drive the populations to new adaptive peaks (Figure 1). It's reasonable to assume that compensatory mutations are generally specific to the gene defect, and multiple molecular mechanisms can restore fitness. Therefore, independently evolving populations carrying an inactivated gene are expected to diverge from each other. Moreover, if compensation mainly proceeds by reducing the need for the disrupted molecular function then compensatory evolution could have a large impact on cellular physiology and survival upon environmental change. Accordingly, the beneficial effects of compensatory mutations may frequently be conditional, and subsequent changes to the environment can reveal the hidden genetic variation across populations (Figure 1). The goal of the current study was to test this hypothesis by an integrated systems biology approach. Specifically, we aimed to determine the potential of the Saccharomyces cerevisiae genome to compensate for gene loss through compensatory evolution and to explore the long-term consequences of this process. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 1. Compensatory evolution on the adaptive landscape. Schematic representation of the impact of compensatory evolution on the fitness landscape. The x and y axes on the landscape locate the network of neighboring genotypic states, while the z axis defines fitness in a single environment. Gene loss leads to a fitness valley (from WT to KO), while compensatory evolution can drive the population to different adaptive peaks (Ev1 versus Ev2). The upper fitness landscape shows the environment where compensatory evolution took place. The dashed arrow represents the original gene deletion event. Yellow lines represent different evolutionary routes. WT, wild type; KO, ancestor strain with a gene deletion. https://doi.org/10.1371/journal.pbio.1001935.g001 Results Rapid Compensatory Evolution Following Gene Loss Is Common We initiated laboratory evolutionary experiments with 187 haploid single gene knock-out mutant strains, all of which initially showed slow (but non-zero) growth compared to the wild-type control in a standard laboratory medium (Figure 2A, for selection criteria, see Materials and Methods). These genes cover a wide range of molecular processes and functions (Table S1). Populations were cultivated in parallel (four replicate populations for each null mutation), resulting in 748 independently evolving lines. 0.5% of each culture was diluted into fresh medium every 48 hours, and populations were propagated for approximately 400 generations. To control for potential adaptation unrelated to compensatory evolution, we also established 22 populations starting from the isogenic wild-type genotype, referred to as evolving wild types. Next, all starting and evolved populations were subjected to high-throughput fitness measurements by monitoring growth rates in liquid cultures. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 2. Compensation of fitness loss during laboratory evolution. (A) Experimental scheme to estimate evolutionary compensation of gene defects. See text for details. (B) Distribution of relative fitness improvement (RFI) of the knock-out mutant strains and the evolving control lineages (Table S1), where RFI = (evolved fitness/initial fitness)−1. (C) Relative compensation (RC) of the compensated knock-out mutant strains (Table S1), where RC is the fraction of the initial fitness defect that was compensated for during laboratory evolution (see Materials and Methods). (D) Compensation does not depend on pleiotropy (Table S1). The bars indicate mean ± standard error, Wilcoxon rank sum test p-values for the three comparisons are: 0.71, 0.44, and 0.36, respectively. (E) Genotypes with lower initial fitness were more likely to be compensated for during laboratory evolution (Table S1). Lines were divided into groups by initial fitness, the fraction of compensated lines among all the lines in the group is shown as bars (chi-squared test for trend in proportions, p<10−13, number of lines in the groups from left to right: 38, 56, 201, 337). https://doi.org/10.1371/journal.pbio.1001935.g002 Fitness may increase during the course of laboratory evolution as a result of general adaptation to the environment and/or accumulation of compensatory mutations that suppress the deleterious effects of gene inactivation. Under the assumption that compensatory evolution was the dominant force in our experiments, fitness should not increase by the same extent in all lineages: genotypes that carry deleterious null mutations are further away from the optimal state and are hence expected to show large fitness gains (Figure 2A); this was indeed so. On average, the evolving wild-type control populations showed a small, but significant 5% fitness improvement. By contrast, the fitness of populations carrying a deleterious null mutation improved by 23% on average (Figure 2B), and many of them approximated wild-type fitness (Figure 2C; Table S1). On the basis of fitness measurements at multiple time points during laboratory evolution (see Methods), we also report that individual fitness trajectories often showed a saturating trend during the course of laboratory evolution (Figure S1). The difference in fitness improvement is not due to the elevated mutation rate of mutant genotypes for two reasons. First, a previous study conducted a genome-wide screen with the aim to identify genes in S. cerevisiae that influence the rate of mutations [20]. While a large number of such genes have been found, only four of them were present in our gene set (Δrad54, Δrad52, Δmre11, and Δrad50). Second, fitness improvements of the corresponding single gene knock-out strains did not differ from the rest of the dataset (one-tailed Wilcoxon rank sum test, p = 0.89). As previously [16], we defined compensatory evolution as a fitness increase that is disproportionally large relative to that in the evolving wild-type lines. Using this definition, 68% of the genotypes showed evidence of compensatory evolution (i.e., at least one of the four independently evolving populations fulfilled the above criteria). The corresponding genes cover a wide range of molecular and cellular processes (Table S1). Impact of Gene Pleiotropy and Dispensability on the Propensity for Compensation Next, we compared the fitness improvements between evolved lines founded from the same gene deletion genotype versus those founded from different genotypes. This analysis revealed that not all genes were equally likely to be compensated as fitness gain differed significantly across genotypes (ANOVA, F(186) = 3.9, p<10−14) (see also Figure S2). It has been previously suggested that as mutations with especially large fitness effects tend to disrupt a broader range of molecular processes [21], such mutations may influence the number of mutational targets where compensatory evolution can occur [13]. We compiled three datasets that estimate different aspects of gene pleiotropy [22], including fitness under diverse environmental conditions (environmental pleiotropy), the number of protein-protein interactions (network pleiotropy), and the number of biological processes associated with a gene (multifunctionality). The extent of evolutionary compensation did not depend on any of the above mentioned features (Figure 2D). However, consistent with results of prior small-scale bacterial and viral evolutionary studies [13],[16], null mutations with more severe defects were more likely to be compensated (Figure 2E). This pattern probably reflects that the availability of compensatory mutations across the genome strongly depends on the fitness effect of the deleted gene. We provide a simple explanation of this phenomenon in the Discussion. Compensatory Evolution Promotes Genomic Diversification To investigate the genomic changes underlying compensatory evolution, we re-sequenced the complete genomes of 41 independently evolved lines and the 14 corresponding ancestors, all of which showed large fitness improvements (Table S1). We focused on de novo mutations that accumulated during the course of laboratory evolution. Large-scale duplications (including segmental or whole chromosome duplication) were observed in 22% of the laboratory evolved lines. On average, six point mutations and 0.5 small insertions or deletions per clone were detected (Figure 3A; Table S2). The ratio of non-synonymous to synonymous mutations was significantly higher than expected by chance (p = 0.003, see Materials and Methods), indicating that the accumulation of these mutations was driven by adaptive evolution. On average, pairs of evolutionary lines founded from the same genotype shared 5.3% of their mutated genes, while the same figure was 0.1% for lines founded from different genotypes (Table S2). This result is in contrast to results of a prior bacterial study [23], where a strong signature of parallel evolution emerged at the gene level across parallel evolving laboratory populations. Despite the rarity of parallel evolution at the molecular level, a major unifying trend emerged: evolution preferentially affected genes that are functionally related to that of the disrupted gene (Figure 3B). Moreover, when the null mutation affected a protein complex subunit, another subunit of the same complex was mutated 10 times more often than expected by chance (Figure 3B). Taken together, these results indicate that deletion of any single gene drives adaptive genetic changes specific to the functional defect incurred. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 3. Genomic analyses of evolutionary compensation. (A) Distribution of different mutational events (Table S2). The inlet shows the color coding and the average value of total mutational events per genotype. (B) The originally deleted gene and the gene with identified de novo mutation participated more often in the same protein complex, were more often assigned to the same functional category and showed significantly more similar genetic interaction and expression profile than expected by random shuffling of the knock-out gene–mutated gene network. Dashed line represents no enrichment; */**/*** indicates p-value<0.05/0.01/0.001, respectively. The x axis is logarithmically scaled. (C) Δrpl6b evolved lines showed duplication of the chromosomal region (or the complete chromosome) carrying a duplicate with redundant function (RPL6A). The gene positions are marked by arrows below the corresponding chromosome, copy numbers are shown by color codes. (D) Dosage compensation of Δrpl6b by increased copy number of RPL6A (Table S5). Copy number of RPL6A was increased by transforming the RPL6A bearing plasmid of the MoBY ORF Library. As the vector carries a selectable marker and a yeast centromere, the plasmid is present in one to three copies per cell. As a control, strains were transformed with the empty centromeric plasmid. Relative fitness was measured as colony sizes on agar plates, values were normalized to the wild-type control with a single genomic copy of RPL6A. All strains were grown on synthetic complete medium without uracil to select for the plasmids. Error bars show standard error. https://doi.org/10.1371/journal.pbio.1001935.g003 Pre-Existing Genetic Redundancy Has No Major Impact on Compensatory Evolution Although duplicated genes with partially overlapping function are frequent in the yeast genome, we found no evidence that genetic changes affecting a duplicate of the disrupted gene provide a general mechanism of compensation in our evolved lines. First, our dataset contains 128 genes showing evidence for compensation, and only 25% of these genes have a duplicate in the yeast genome (i.e., at least 30% amino acid similarity between the two copies). This figure is a gross overestimate, as it includes very distant duplicates that most likely diverged functionally (Materials and Methods). Second, the subset of genes with a gene duplicate were not more likely to be compensated during laboratory evolution than the rest of the dataset (Chi-squared test, p = 0.54). Third, genome sequence analysis of the evolved lines revealed only one clear example where evolution proceeded through increasing the dosage of a gene duplicate with redundant function of the deleted gene (Figure 3C). All three studied evolved lines of Δrpl6b showed an increased copy number of the left arm of Chromosome XIII (Figure 3C). RPL6B is a non-essential gene and encodes a ribosomal 60S subunit protein L6B. The duplicated genomic regions of Δrpl6b evolved lines carry RPL6A, a duplicate copy of RPL6B. The two genes share 94% amino acid identity, have highly overlapping functions, and deletion of both genes confer a synthetic lethal phenotype [24]. On the basis of these observations, we propose that doubling the copy number of RPL6A through segmental duplication could be partly responsible for the improved fitness in the evolved lines carrying the RPL6B deletion. The hypothesis was tested by increasing the copy number of RPL6A in wild-type and Δrpl6b genetic backgrounds, respectively. As expected, an enhanced copy number of RPL6A substantially improved the fitness of Δrpl6b, but not that of the wild type (Figure 3D). Compensatory Evolution Does Not Restore Wild-Type Genomic Expression State Compensatory evolution may restore wild-type physiology or generate novel alterations with respect to prior physiological states [25]. To investigate the relative contribution of these processes, eight genotypes carrying a deleterious gene deletion and one corresponding evolved line were selected for transcriptome analysis (see Materials and Methods for selection criteria). Using DNA microarrays, the global gene expression states were compared between the wild-type, the ancestral line, and the evolved lines carrying the same gene deletion (Figure 4A and 4B). As expected from prior studies [26], inactivation of genes with high fitness contribution altered the expression of a large number of genes across the genome (ranging between 81 to 588) (see Table S3). Next, the transcriptomic profiles were compared by calculating all pairwise combinations of Euclidean distances. The wild-type, the ancestral line, and the corresponding evolved lines generally showed substantial differences in their transcriptome profiles (Figure 4B), indicating that compensatory evolution drives the cell towards novel genomic expression states. Importantly, transcriptome profile distances between different genotypes was always higher than distances between replicate measurements of the same genotype (Figure 4B), implying that the substantial differences observed between evolved lines and wild-type cannot be attributed to measurement noise. As a further support, typically only 10%–30% of the genes with altered expression in the ancestral lines showed significant shift towards the wild-type expression level in the corresponding evolved lines (Figure 4C). Hence, despite substantial fitness improvements (>75% for all cases investigated), the majority of the gene expression changes due to gene deletion remained unrestored during evolution. These patterns were not attributable to growth rate regulated gene expression or copy number variation in the evolved lines (Figure S3). Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 4. Comparisons of the transcriptome profiles of wild-type, ancestor, and evolved lines. (A) Heatmaps of transcriptome profiles of deletion mutants Δrpl43a, Δpop2, Δmdm34, Δrsc2, Δifm1, Δrpb9, and Δbud20 and their corresponding evolved lines. For each deletion mutant, the fold-changes (FC) are shown for the ancestor strain versus the wild type, the evolved strain versus the wild type and the evolved strain versus the ancestor strain (Table S3). Color scales as indicated. Individual transcripts are depicted if they change significantly (FC>1.7, p<0.05) at least once in one of these comparisons. (B) The Euclidean distances of microarray profiles of the evolved evolutionary line from its ancestor and from wild type (WT) were calculated and normalized to the ancestor–wild type distance for each genotype. The distances of the points in the figure are proportional to the calculated profile distances. For each genotype triplet, distances were calculated on the basis of those genes that are differentially expressed in at least one of the pairwise comparisons. For each deletion strain, the edges of the triangle represent Euclidean distances of log2 mRNA expression fold-changes between the wild-type (WT), ancestor (anc), and evolved (evo) lines. To calculate these distances we used the average of four replicate expression measurements (two biological and two technical replicates). Circles around average values represent the Euclidean distance between the two biological replicates (calculated as the average based on the two technical replicates). For each genotype triplet, distances were calculated on the basis of those genes that are differentially expressed (FC>1.7, p<0.05) in at least one of the pairwise comparisons (Table S6). (C) Within the subset of genes that showed expression change upon gene deletion, the barplot shows the fraction of these genes that changed expression during evolution in the opposite direction (i.e., evolution towards restoration of wild-type expression level; see inset). With one major exception (lines disrupted in mdm34), only a small fraction of the expression changes were restored in the evolved lines (Table S6). The threshold for expression change was 1.7-fold-change and p<0.05, as in [62]. https://doi.org/10.1371/journal.pbio.1001935.g004 Compensatory Evolution Generates Diverse Growth Phenotypes across Environments Taken together, compensatory evolution following gene loss did not restore wild-type genomic expression and promoted genomic divergence across populations. Are these evolutionary outcomes phenotypically completely equivalent? This problem was first addressed by monitoring the fitness of 237 evolved populations in 14 environmental settings, including previously tested nutrients and stress factors [27]. Prior to evolution, genotypes carrying a gene deletion generally displayed slow growth in most environments (Table S1). The situation was far more complex following laboratory evolution. Considering all possible pairs of population-environment combinations, fitness improved in 52%, and declined in 8% of the cases (Figure 5A). Moreover, independently evolved populations carrying the same disrupted gene showed more fitness variation across the 14 tested conditions than in the environment they had been exposed to during laboratory evolution (Figure 5B, p<10−7), while evolved wild-type populations did not show such a difference (p = 0.93, coefficient of variations compared by Z-test). Furthermore, the degree of fitness variation across conditions was especially high for gene deletions that showed large fitness gains during compensatory evolution (Spearman rho = 0.36, p = 10−4) (Figure 5C). These results indicate that the level of discernible heterogeneity in fitness was relatively low in the evolved populations founded from the same genotype, but the variation can be uncovered upon environmental change. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 5. Large-scale phenotypic screen of evolved lines. (A) Fitness trade-offs in evolved lines carrying a deletion across 14 environments (Table S1). Lines are ranked according to the number of environments in which they display improved fitness (brown). Grey and black dots indicate conditions where the fitness of the line is statistically equal or lower, respectively, than that of the corresponding ancestor. (B) Fitness variation in independently evolving lines carrying the same gene deletion. The figure shows the coefficient of variation in the in the medium of selection (YPD) versus all other media (Table S1). The difference is highly significant (Wilcoxon rank sum test p-value<10−7). The bars indicate mean of the coefficients of variations ± standard error. (C) Gene deletions showing larger fitness gains have higher variance of fitness between replicate lines across other environments (Spearman rank correlation, rho = 0.36, p = 0.0001). Each point represents a gene deletion genotype. The x-axis shows the mean of the fitness gains of the parallel evolving replicates of a given gene deletion, while the y-axis shows the mean of the coefficient of variations measured in each alternative media between the parallel evolving replicates after 104 days of lab evolution (Table S1). The gray line indicates fit by linear regression. https://doi.org/10.1371/journal.pbio.1001935.g005 Finally, our analysis revealed a few instances where the laboratory evolved lines displayed significantly higher than wild-type fitness in specific environments (Table S1). Most notably, the evolved Δrpl6b and Δatp11 lines displayed 24%–26% fitness increase compared to that of the wild type in a medium containing sodium chloride (Table S1), a result that was confirmed by additional independent colony size assays with high replicate number (n = 20, Wilcoxon rank-sum test p<10−4 in all cases). Moreover, the fitnesses of these lines in this medium surpassed all that of the 22 evolved wild-type controls. These results are all the more remarkable, as the corresponding ancestral Δrpl6b and Δatp11 strains showed fitness values significantly lower than wild type under all environmental conditions considered. These preliminary results indicate that gene loss can promote adaptive evolution towards novel environments, a possibility that will be explored further in a future work. A Case Study Reveals the Fitness Cost and Condition Dependence of Compensatory Evolution Next, we conducted an in-depth genetic analysis with the MDM34 deletion with the aim of deciphering the molecular mechanisms and/or potential fitness costs of compensatory mutations (Text S1). This gene codes for a component of the ERMES protein complex, and is involved in the exchange of phospholipids between mitochondria and the endoplasmatic reticulum (Figure 6A). Disruption of this gene yields impaired cardiolipin synthesis [28], as an insufficient amount of unsaturated fatty acids reaches the mitochondria (Figure 6A). Laboratory-evolved lines carrying deletion in this gene substantially improved fitness in the medium of selection (Table S1), but the putative cellular mechanisms of compensation were remarkably different across populations (Figures 6A and S4). The native copy of MDM34 was reinserted into the ancestral line and four evolved lines carrying the same deletion (Δmdm34). The analysis revealed that the net effect of mutations in three evolved lines were deleterious in the presence of MDM34 (Figure 6B). Next, we concentrated on a specific mutation observed in MGA2, a gene involved in the regulation of unsaturated fatty acid biosynthesis (Figure 6A; Text S1). Inserting the observed mutations (mga2-1) into wild type and Δmdm34 resulted in very similar conclusions. mga2-1 and Δmdm34 showed strong sign-epistasis [29]: they were independently deleterious but significantly less so when they occurred together (Figure 6C). Moreover, the capacity of mga2-1 to compensate the loss of MDM34 was restricted to non-acidic conditions (Figure 6C), probably because of the misregulation of the corresponding stress-induced pathway under low pH (Text S1). Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 6. Compensation of the MDM34 gene deletion. (A) The cardiolipin synthesis pathway with an emphasis on the ERMES complex. The complex tethers the endoplasmatic reticulum to the mitochondria, and is central for the transfer of phospholipids between the two compartments. De novo mutations in the independent evolutionary lines affected different, but related cellular subsystems, including upregulation of the unsaturated fatty acid synthesis (MGA2), another step of the cardiolipin synthesis pathway downstream of the ERMES complex (MDM35), and another mitochondrial transport process (CRC1), which most likely affects respiration by modulating the interaction between carnitine and cardiolipin. For further details on the underlying mechanisms see Text S1. The green arrow represents transcriptional upregulation; the dashed arrow indicates indirect positive effect. The mutations in MGA2, MDM35, and CRC1 genes were found in Δmdm34 evolved lines 1, 3, and 4, respectively. (B) The cumulative fitness effects of the compensatory mutations in Δmdm34 and “wild type” (Δmdm34+MDM34 reintroduced) backgrounds (Table S7). (C) Epistatic interactions between mutations in two environments (Table S7). The bars in (B) and (C) indicate means ± standard error. Arrows indicate fitness costs and the extent of compensation. https://doi.org/10.1371/journal.pbio.1001935.g006 Evolutionary Compensation by Loss-of-Function Mutation Our dataset contains 21 independent point mutations that occurred during laboratory evolution and generated in-frame stop codons. Most notably, a mutation in WHI2 emerged in an evolving Δrpb9 line, which shortened the coding region from 480 to 133 codons, and hence most likely resulted in a non-functional protein. To test the impact of loss of WHI2 function on fitness and compensation, Δwhi2 was introduced into Δrpb9 cells using synthetic genetic array methodology (Figure 7A and 7B) [30]. In agreement with expectation, deletion of WHI2 partly suppressed the harmful effect of the RPB9 deletion (Figure 7B). RPB9 is an RNA polymerase II subunit, and its deletion leads to elevated transcriptional error rate [31] and in turn, to proteotoxic stress [32], which can result in cell cycle arrest [33]. WHI2 is known to be required for general stress response [34] and cell cycle arrest [35]. We speculate that less stringent cell cycle control due to WHI2 deletion is favorable in Δrpb9 (see also [36]). Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 7. Environment-dependent compensation by a loss-of-function mutation. (A) Δrpb9 and Δwhi2 mutations were crossed by SGA using haploid parental strains as shown. To compare the double mutant Δrpb9 Δwhi2 with the wild-type control and corresponding single mutants, the resistance cassettes required by the SGA method were introduced into wild-type and single mutants by crossing them with parental strains where the corresponding resistance cassettes reside at a non-functional locus (Δhis3::KanMX4 and Δho::NatMX4). (B) Relative fitness was measured as colony sizes on YPD and YPD supplemented with cycloheximide (CYC), values were normalized to WT. The arrow shows the extent of compensation of Δrpb9 by Δwhi2 on glucose medium (Wilcoxon rank sum test p = 0.005, error bars show standard error) (Table S8). (C) Relative fitness of Δrpb9 replicate evolving line 2 and Δrpb9 Δwhi2 double mutant were measured as colony sizes grown on different media. Genotypes are indicated on the left, the growth media are indicated above the heat map. For media composition and abbreviations, see Table S4. Values are normalized to Δrpb9 ancestor. Log2 values are shown according to the color coding (Table S8). https://doi.org/10.1371/journal.pbio.1001935.g007 Next, the fitness impact of WHI2 deletion was evaluated across 14 environments. The fitnesses of the Δrpb9 Δwhi2 strain varied strongly across conditions, and showed correlation with that of the evolved Δrpb9 line, which carried the WHI2 non-sense mutations (Spearman rho = 0.77, p<0.005) (see Figure 7C). Most notably, the compensation of Δrpb9 by Δwhi2 was completely abolished in the presence of cycloheximide (Figure 7B). We conclude that the compensatory effect of WHI2 deletion is plastic across environments. Rapid Compensatory Evolution Following Gene Loss Is Common We initiated laboratory evolutionary experiments with 187 haploid single gene knock-out mutant strains, all of which initially showed slow (but non-zero) growth compared to the wild-type control in a standard laboratory medium (Figure 2A, for selection criteria, see Materials and Methods). These genes cover a wide range of molecular processes and functions (Table S1). Populations were cultivated in parallel (four replicate populations for each null mutation), resulting in 748 independently evolving lines. 0.5% of each culture was diluted into fresh medium every 48 hours, and populations were propagated for approximately 400 generations. To control for potential adaptation unrelated to compensatory evolution, we also established 22 populations starting from the isogenic wild-type genotype, referred to as evolving wild types. Next, all starting and evolved populations were subjected to high-throughput fitness measurements by monitoring growth rates in liquid cultures. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 2. Compensation of fitness loss during laboratory evolution. (A) Experimental scheme to estimate evolutionary compensation of gene defects. See text for details. (B) Distribution of relative fitness improvement (RFI) of the knock-out mutant strains and the evolving control lineages (Table S1), where RFI = (evolved fitness/initial fitness)−1. (C) Relative compensation (RC) of the compensated knock-out mutant strains (Table S1), where RC is the fraction of the initial fitness defect that was compensated for during laboratory evolution (see Materials and Methods). (D) Compensation does not depend on pleiotropy (Table S1). The bars indicate mean ± standard error, Wilcoxon rank sum test p-values for the three comparisons are: 0.71, 0.44, and 0.36, respectively. (E) Genotypes with lower initial fitness were more likely to be compensated for during laboratory evolution (Table S1). Lines were divided into groups by initial fitness, the fraction of compensated lines among all the lines in the group is shown as bars (chi-squared test for trend in proportions, p<10−13, number of lines in the groups from left to right: 38, 56, 201, 337). https://doi.org/10.1371/journal.pbio.1001935.g002 Fitness may increase during the course of laboratory evolution as a result of general adaptation to the environment and/or accumulation of compensatory mutations that suppress the deleterious effects of gene inactivation. Under the assumption that compensatory evolution was the dominant force in our experiments, fitness should not increase by the same extent in all lineages: genotypes that carry deleterious null mutations are further away from the optimal state and are hence expected to show large fitness gains (Figure 2A); this was indeed so. On average, the evolving wild-type control populations showed a small, but significant 5% fitness improvement. By contrast, the fitness of populations carrying a deleterious null mutation improved by 23% on average (Figure 2B), and many of them approximated wild-type fitness (Figure 2C; Table S1). On the basis of fitness measurements at multiple time points during laboratory evolution (see Methods), we also report that individual fitness trajectories often showed a saturating trend during the course of laboratory evolution (Figure S1). The difference in fitness improvement is not due to the elevated mutation rate of mutant genotypes for two reasons. First, a previous study conducted a genome-wide screen with the aim to identify genes in S. cerevisiae that influence the rate of mutations [20]. While a large number of such genes have been found, only four of them were present in our gene set (Δrad54, Δrad52, Δmre11, and Δrad50). Second, fitness improvements of the corresponding single gene knock-out strains did not differ from the rest of the dataset (one-tailed Wilcoxon rank sum test, p = 0.89). As previously [16], we defined compensatory evolution as a fitness increase that is disproportionally large relative to that in the evolving wild-type lines. Using this definition, 68% of the genotypes showed evidence of compensatory evolution (i.e., at least one of the four independently evolving populations fulfilled the above criteria). The corresponding genes cover a wide range of molecular and cellular processes (Table S1). Impact of Gene Pleiotropy and Dispensability on the Propensity for Compensation Next, we compared the fitness improvements between evolved lines founded from the same gene deletion genotype versus those founded from different genotypes. This analysis revealed that not all genes were equally likely to be compensated as fitness gain differed significantly across genotypes (ANOVA, F(186) = 3.9, p<10−14) (see also Figure S2). It has been previously suggested that as mutations with especially large fitness effects tend to disrupt a broader range of molecular processes [21], such mutations may influence the number of mutational targets where compensatory evolution can occur [13]. We compiled three datasets that estimate different aspects of gene pleiotropy [22], including fitness under diverse environmental conditions (environmental pleiotropy), the number of protein-protein interactions (network pleiotropy), and the number of biological processes associated with a gene (multifunctionality). The extent of evolutionary compensation did not depend on any of the above mentioned features (Figure 2D). However, consistent with results of prior small-scale bacterial and viral evolutionary studies [13],[16], null mutations with more severe defects were more likely to be compensated (Figure 2E). This pattern probably reflects that the availability of compensatory mutations across the genome strongly depends on the fitness effect of the deleted gene. We provide a simple explanation of this phenomenon in the Discussion. Compensatory Evolution Promotes Genomic Diversification To investigate the genomic changes underlying compensatory evolution, we re-sequenced the complete genomes of 41 independently evolved lines and the 14 corresponding ancestors, all of which showed large fitness improvements (Table S1). We focused on de novo mutations that accumulated during the course of laboratory evolution. Large-scale duplications (including segmental or whole chromosome duplication) were observed in 22% of the laboratory evolved lines. On average, six point mutations and 0.5 small insertions or deletions per clone were detected (Figure 3A; Table S2). The ratio of non-synonymous to synonymous mutations was significantly higher than expected by chance (p = 0.003, see Materials and Methods), indicating that the accumulation of these mutations was driven by adaptive evolution. On average, pairs of evolutionary lines founded from the same genotype shared 5.3% of their mutated genes, while the same figure was 0.1% for lines founded from different genotypes (Table S2). This result is in contrast to results of a prior bacterial study [23], where a strong signature of parallel evolution emerged at the gene level across parallel evolving laboratory populations. Despite the rarity of parallel evolution at the molecular level, a major unifying trend emerged: evolution preferentially affected genes that are functionally related to that of the disrupted gene (Figure 3B). Moreover, when the null mutation affected a protein complex subunit, another subunit of the same complex was mutated 10 times more often than expected by chance (Figure 3B). Taken together, these results indicate that deletion of any single gene drives adaptive genetic changes specific to the functional defect incurred. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 3. Genomic analyses of evolutionary compensation. (A) Distribution of different mutational events (Table S2). The inlet shows the color coding and the average value of total mutational events per genotype. (B) The originally deleted gene and the gene with identified de novo mutation participated more often in the same protein complex, were more often assigned to the same functional category and showed significantly more similar genetic interaction and expression profile than expected by random shuffling of the knock-out gene–mutated gene network. Dashed line represents no enrichment; */**/*** indicates p-value<0.05/0.01/0.001, respectively. The x axis is logarithmically scaled. (C) Δrpl6b evolved lines showed duplication of the chromosomal region (or the complete chromosome) carrying a duplicate with redundant function (RPL6A). The gene positions are marked by arrows below the corresponding chromosome, copy numbers are shown by color codes. (D) Dosage compensation of Δrpl6b by increased copy number of RPL6A (Table S5). Copy number of RPL6A was increased by transforming the RPL6A bearing plasmid of the MoBY ORF Library. As the vector carries a selectable marker and a yeast centromere, the plasmid is present in one to three copies per cell. As a control, strains were transformed with the empty centromeric plasmid. Relative fitness was measured as colony sizes on agar plates, values were normalized to the wild-type control with a single genomic copy of RPL6A. All strains were grown on synthetic complete medium without uracil to select for the plasmids. Error bars show standard error. https://doi.org/10.1371/journal.pbio.1001935.g003 Pre-Existing Genetic Redundancy Has No Major Impact on Compensatory Evolution Although duplicated genes with partially overlapping function are frequent in the yeast genome, we found no evidence that genetic changes affecting a duplicate of the disrupted gene provide a general mechanism of compensation in our evolved lines. First, our dataset contains 128 genes showing evidence for compensation, and only 25% of these genes have a duplicate in the yeast genome (i.e., at least 30% amino acid similarity between the two copies). This figure is a gross overestimate, as it includes very distant duplicates that most likely diverged functionally (Materials and Methods). Second, the subset of genes with a gene duplicate were not more likely to be compensated during laboratory evolution than the rest of the dataset (Chi-squared test, p = 0.54). Third, genome sequence analysis of the evolved lines revealed only one clear example where evolution proceeded through increasing the dosage of a gene duplicate with redundant function of the deleted gene (Figure 3C). All three studied evolved lines of Δrpl6b showed an increased copy number of the left arm of Chromosome XIII (Figure 3C). RPL6B is a non-essential gene and encodes a ribosomal 60S subunit protein L6B. The duplicated genomic regions of Δrpl6b evolved lines carry RPL6A, a duplicate copy of RPL6B. The two genes share 94% amino acid identity, have highly overlapping functions, and deletion of both genes confer a synthetic lethal phenotype [24]. On the basis of these observations, we propose that doubling the copy number of RPL6A through segmental duplication could be partly responsible for the improved fitness in the evolved lines carrying the RPL6B deletion. The hypothesis was tested by increasing the copy number of RPL6A in wild-type and Δrpl6b genetic backgrounds, respectively. As expected, an enhanced copy number of RPL6A substantially improved the fitness of Δrpl6b, but not that of the wild type (Figure 3D). Compensatory Evolution Does Not Restore Wild-Type Genomic Expression State Compensatory evolution may restore wild-type physiology or generate novel alterations with respect to prior physiological states [25]. To investigate the relative contribution of these processes, eight genotypes carrying a deleterious gene deletion and one corresponding evolved line were selected for transcriptome analysis (see Materials and Methods for selection criteria). Using DNA microarrays, the global gene expression states were compared between the wild-type, the ancestral line, and the evolved lines carrying the same gene deletion (Figure 4A and 4B). As expected from prior studies [26], inactivation of genes with high fitness contribution altered the expression of a large number of genes across the genome (ranging between 81 to 588) (see Table S3). Next, the transcriptomic profiles were compared by calculating all pairwise combinations of Euclidean distances. The wild-type, the ancestral line, and the corresponding evolved lines generally showed substantial differences in their transcriptome profiles (Figure 4B), indicating that compensatory evolution drives the cell towards novel genomic expression states. Importantly, transcriptome profile distances between different genotypes was always higher than distances between replicate measurements of the same genotype (Figure 4B), implying that the substantial differences observed between evolved lines and wild-type cannot be attributed to measurement noise. As a further support, typically only 10%–30% of the genes with altered expression in the ancestral lines showed significant shift towards the wild-type expression level in the corresponding evolved lines (Figure 4C). Hence, despite substantial fitness improvements (>75% for all cases investigated), the majority of the gene expression changes due to gene deletion remained unrestored during evolution. These patterns were not attributable to growth rate regulated gene expression or copy number variation in the evolved lines (Figure S3). Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 4. Comparisons of the transcriptome profiles of wild-type, ancestor, and evolved lines. (A) Heatmaps of transcriptome profiles of deletion mutants Δrpl43a, Δpop2, Δmdm34, Δrsc2, Δifm1, Δrpb9, and Δbud20 and their corresponding evolved lines. For each deletion mutant, the fold-changes (FC) are shown for the ancestor strain versus the wild type, the evolved strain versus the wild type and the evolved strain versus the ancestor strain (Table S3). Color scales as indicated. Individual transcripts are depicted if they change significantly (FC>1.7, p<0.05) at least once in one of these comparisons. (B) The Euclidean distances of microarray profiles of the evolved evolutionary line from its ancestor and from wild type (WT) were calculated and normalized to the ancestor–wild type distance for each genotype. The distances of the points in the figure are proportional to the calculated profile distances. For each genotype triplet, distances were calculated on the basis of those genes that are differentially expressed in at least one of the pairwise comparisons. For each deletion strain, the edges of the triangle represent Euclidean distances of log2 mRNA expression fold-changes between the wild-type (WT), ancestor (anc), and evolved (evo) lines. To calculate these distances we used the average of four replicate expression measurements (two biological and two technical replicates). Circles around average values represent the Euclidean distance between the two biological replicates (calculated as the average based on the two technical replicates). For each genotype triplet, distances were calculated on the basis of those genes that are differentially expressed (FC>1.7, p<0.05) in at least one of the pairwise comparisons (Table S6). (C) Within the subset of genes that showed expression change upon gene deletion, the barplot shows the fraction of these genes that changed expression during evolution in the opposite direction (i.e., evolution towards restoration of wild-type expression level; see inset). With one major exception (lines disrupted in mdm34), only a small fraction of the expression changes were restored in the evolved lines (Table S6). The threshold for expression change was 1.7-fold-change and p<0.05, as in [62]. https://doi.org/10.1371/journal.pbio.1001935.g004 Compensatory Evolution Generates Diverse Growth Phenotypes across Environments Taken together, compensatory evolution following gene loss did not restore wild-type genomic expression and promoted genomic divergence across populations. Are these evolutionary outcomes phenotypically completely equivalent? This problem was first addressed by monitoring the fitness of 237 evolved populations in 14 environmental settings, including previously tested nutrients and stress factors [27]. Prior to evolution, genotypes carrying a gene deletion generally displayed slow growth in most environments (Table S1). The situation was far more complex following laboratory evolution. Considering all possible pairs of population-environment combinations, fitness improved in 52%, and declined in 8% of the cases (Figure 5A). Moreover, independently evolved populations carrying the same disrupted gene showed more fitness variation across the 14 tested conditions than in the environment they had been exposed to during laboratory evolution (Figure 5B, p<10−7), while evolved wild-type populations did not show such a difference (p = 0.93, coefficient of variations compared by Z-test). Furthermore, the degree of fitness variation across conditions was especially high for gene deletions that showed large fitness gains during compensatory evolution (Spearman rho = 0.36, p = 10−4) (Figure 5C). These results indicate that the level of discernible heterogeneity in fitness was relatively low in the evolved populations founded from the same genotype, but the variation can be uncovered upon environmental change. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 5. Large-scale phenotypic screen of evolved lines. (A) Fitness trade-offs in evolved lines carrying a deletion across 14 environments (Table S1). Lines are ranked according to the number of environments in which they display improved fitness (brown). Grey and black dots indicate conditions where the fitness of the line is statistically equal or lower, respectively, than that of the corresponding ancestor. (B) Fitness variation in independently evolving lines carrying the same gene deletion. The figure shows the coefficient of variation in the in the medium of selection (YPD) versus all other media (Table S1). The difference is highly significant (Wilcoxon rank sum test p-value<10−7). The bars indicate mean of the coefficients of variations ± standard error. (C) Gene deletions showing larger fitness gains have higher variance of fitness between replicate lines across other environments (Spearman rank correlation, rho = 0.36, p = 0.0001). Each point represents a gene deletion genotype. The x-axis shows the mean of the fitness gains of the parallel evolving replicates of a given gene deletion, while the y-axis shows the mean of the coefficient of variations measured in each alternative media between the parallel evolving replicates after 104 days of lab evolution (Table S1). The gray line indicates fit by linear regression. https://doi.org/10.1371/journal.pbio.1001935.g005 Finally, our analysis revealed a few instances where the laboratory evolved lines displayed significantly higher than wild-type fitness in specific environments (Table S1). Most notably, the evolved Δrpl6b and Δatp11 lines displayed 24%–26% fitness increase compared to that of the wild type in a medium containing sodium chloride (Table S1), a result that was confirmed by additional independent colony size assays with high replicate number (n = 20, Wilcoxon rank-sum test p<10−4 in all cases). Moreover, the fitnesses of these lines in this medium surpassed all that of the 22 evolved wild-type controls. These results are all the more remarkable, as the corresponding ancestral Δrpl6b and Δatp11 strains showed fitness values significantly lower than wild type under all environmental conditions considered. These preliminary results indicate that gene loss can promote adaptive evolution towards novel environments, a possibility that will be explored further in a future work. A Case Study Reveals the Fitness Cost and Condition Dependence of Compensatory Evolution Next, we conducted an in-depth genetic analysis with the MDM34 deletion with the aim of deciphering the molecular mechanisms and/or potential fitness costs of compensatory mutations (Text S1). This gene codes for a component of the ERMES protein complex, and is involved in the exchange of phospholipids between mitochondria and the endoplasmatic reticulum (Figure 6A). Disruption of this gene yields impaired cardiolipin synthesis [28], as an insufficient amount of unsaturated fatty acids reaches the mitochondria (Figure 6A). Laboratory-evolved lines carrying deletion in this gene substantially improved fitness in the medium of selection (Table S1), but the putative cellular mechanisms of compensation were remarkably different across populations (Figures 6A and S4). The native copy of MDM34 was reinserted into the ancestral line and four evolved lines carrying the same deletion (Δmdm34). The analysis revealed that the net effect of mutations in three evolved lines were deleterious in the presence of MDM34 (Figure 6B). Next, we concentrated on a specific mutation observed in MGA2, a gene involved in the regulation of unsaturated fatty acid biosynthesis (Figure 6A; Text S1). Inserting the observed mutations (mga2-1) into wild type and Δmdm34 resulted in very similar conclusions. mga2-1 and Δmdm34 showed strong sign-epistasis [29]: they were independently deleterious but significantly less so when they occurred together (Figure 6C). Moreover, the capacity of mga2-1 to compensate the loss of MDM34 was restricted to non-acidic conditions (Figure 6C), probably because of the misregulation of the corresponding stress-induced pathway under low pH (Text S1). Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 6. Compensation of the MDM34 gene deletion. (A) The cardiolipin synthesis pathway with an emphasis on the ERMES complex. The complex tethers the endoplasmatic reticulum to the mitochondria, and is central for the transfer of phospholipids between the two compartments. De novo mutations in the independent evolutionary lines affected different, but related cellular subsystems, including upregulation of the unsaturated fatty acid synthesis (MGA2), another step of the cardiolipin synthesis pathway downstream of the ERMES complex (MDM35), and another mitochondrial transport process (CRC1), which most likely affects respiration by modulating the interaction between carnitine and cardiolipin. For further details on the underlying mechanisms see Text S1. The green arrow represents transcriptional upregulation; the dashed arrow indicates indirect positive effect. The mutations in MGA2, MDM35, and CRC1 genes were found in Δmdm34 evolved lines 1, 3, and 4, respectively. (B) The cumulative fitness effects of the compensatory mutations in Δmdm34 and “wild type” (Δmdm34+MDM34 reintroduced) backgrounds (Table S7). (C) Epistatic interactions between mutations in two environments (Table S7). The bars in (B) and (C) indicate means ± standard error. Arrows indicate fitness costs and the extent of compensation. https://doi.org/10.1371/journal.pbio.1001935.g006 Evolutionary Compensation by Loss-of-Function Mutation Our dataset contains 21 independent point mutations that occurred during laboratory evolution and generated in-frame stop codons. Most notably, a mutation in WHI2 emerged in an evolving Δrpb9 line, which shortened the coding region from 480 to 133 codons, and hence most likely resulted in a non-functional protein. To test the impact of loss of WHI2 function on fitness and compensation, Δwhi2 was introduced into Δrpb9 cells using synthetic genetic array methodology (Figure 7A and 7B) [30]. In agreement with expectation, deletion of WHI2 partly suppressed the harmful effect of the RPB9 deletion (Figure 7B). RPB9 is an RNA polymerase II subunit, and its deletion leads to elevated transcriptional error rate [31] and in turn, to proteotoxic stress [32], which can result in cell cycle arrest [33]. WHI2 is known to be required for general stress response [34] and cell cycle arrest [35]. We speculate that less stringent cell cycle control due to WHI2 deletion is favorable in Δrpb9 (see also [36]). Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 7. Environment-dependent compensation by a loss-of-function mutation. (A) Δrpb9 and Δwhi2 mutations were crossed by SGA using haploid parental strains as shown. To compare the double mutant Δrpb9 Δwhi2 with the wild-type control and corresponding single mutants, the resistance cassettes required by the SGA method were introduced into wild-type and single mutants by crossing them with parental strains where the corresponding resistance cassettes reside at a non-functional locus (Δhis3::KanMX4 and Δho::NatMX4). (B) Relative fitness was measured as colony sizes on YPD and YPD supplemented with cycloheximide (CYC), values were normalized to WT. The arrow shows the extent of compensation of Δrpb9 by Δwhi2 on glucose medium (Wilcoxon rank sum test p = 0.005, error bars show standard error) (Table S8). (C) Relative fitness of Δrpb9 replicate evolving line 2 and Δrpb9 Δwhi2 double mutant were measured as colony sizes grown on different media. Genotypes are indicated on the left, the growth media are indicated above the heat map. For media composition and abbreviations, see Table S4. Values are normalized to Δrpb9 ancestor. Log2 values are shown according to the color coding (Table S8). https://doi.org/10.1371/journal.pbio.1001935.g007 Next, the fitness impact of WHI2 deletion was evaluated across 14 environments. The fitnesses of the Δrpb9 Δwhi2 strain varied strongly across conditions, and showed correlation with that of the evolved Δrpb9 line, which carried the WHI2 non-sense mutations (Spearman rho = 0.77, p<0.005) (see Figure 7C). Most notably, the compensation of Δrpb9 by Δwhi2 was completely abolished in the presence of cycloheximide (Figure 7B). We conclude that the compensatory effect of WHI2 deletion is plastic across environments. Discussion Our work addresses one of the most long-standing debates in evolution. Since the early 1920s, Ronald Fisher pioneered the view that adaptation is by and large a hill climbing process: it proceeds through progressive accumulation of beneficial mutations [37],[38]. However, as slightly deleterious mutations are far more abundant, they have a significant contribution to genetic variation in natural populations [2]. In the long run, the wealth of such detrimental mutations is expected to promote fixation of compensatory mutations elsewhere in the genome. This work focused on a specific aspect of this problem, and asked whether deleterious gene loss events promote adaptive genetic changes and what the side consequences of such a process might be. To systematically study compensatory evolution following gene loss, we initiated laboratory evolutionary experiments with over 180 haploid yeast genotypes, all of which initially displayed slow growth owing to the deletion of a single gene, and investigated the genomic and phenotypic capacities of the evolved lines in detail. Thanks to the exceptionally large-scale analysis of our study, the following major conclusions can be drawn. First, compensatory evolution following gene loss was pervasive: 68% of the deleterious, but non-lethal gene disruptions were compensated through the accumulation of adaptive mutations elsewhere in the genome (Figure 2B). Furthermore, in agreement with prior bacterial studies [16],[17], the process was strikingly rapid. As the set of disrupted genes are functionally very diverse (Table S1), it appears that defects in a broad range of molecular processes can readily be compensated during evolution.However, we and others [17] also found that not all genotypes are equally likely to be recovered during laboratory evolution. Therefore, future works should clarify the exact molecular, functional, and systems level gene properties that influence compensability. Second, our large-scale study indicates that the extent of fitness loss due to gene disruption is one if not the strongest predictor of compensatory evolution (Figure 2E). Although this relationship has been observed previously in small-scale studies [16], the reasons remained largely unknown. One may argue that the spread of compensatory mutations with mild beneficial effects would have taken many more than 400 generations to reach fixation [16]. Although this explanation cannot be excluded, there is another intriguing possibility [13]. Consistent with Fisher's geometric model [37],[38], fitness improvement in populations close to an optimal state can only be achieved by relatively rare mutations with small effects. However, when a population with a gene defect is further away from a fitness peak, compensatory evolution may proceed through a wider range of mutations, including ones that have deleterious side effects. Two lines of evidence are consistent with this scenario. Compensatory evolution has associated pleiotropic effects (Figures 5 and 6C). Moreover, the theory predicts that compensatory mutations should be especially frequent in the case of strongly deleterious null mutations. An analysis based on data of a prior genome-wide genetic interaction study [21] suggests that it may indeed be so (Figure 8). Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 8. Strongly deleterious gene deletions can be suppressed by a large number of other null mutations according to a genome-wide genetic interaction study. The plot shows the relationship between the fitness of a given single-gene deletion strain and the fraction of other genes across the genome whose deletion suppresses the fitness effect of this mutation (Table S9). Boxplots present the median and first and third quartiles, with whiskers showing either the maximum (minimum) value or 1.5 times the interquartile range of the data. Spearman correlation on raw data: rho = −0.69, p<10−16, n = 3880. We note that using the fraction of suppressive interactions among all genetic interactions displayed by a given gene yields a very similar result (rho = −0.69, p<10−16), indicating that the relationship is not simply due to the fact that slow-growing strains generally display especially large numbers of both positive and negative interactions [21]. Information on suppression genetic interactions and single-deletion fitness comes from a global genetic interaction map of yeast [21]. Suppression interactions were defined as in previously [70]. In brief, deletion of gene B suppresses deletion of gene A if their fitness values obey the following rules: FA<FB and FAB>FA+σA (where FA, FB, and FAB are the fitness measures of single deletants A, B, and the double deletant AB, respectively, and σA is the standard deviation of FA). One important caveat is that as this simple analysis considers null mutations only, the results should be considered preliminary. https://doi.org/10.1371/journal.pbio.1001935.g008 Third, genomic analysis of the evolved lines revealed that deletion of any single gene drives adaptive genetic changes specific to the functional defect incurred (Figure 3B), and consequently convergent evolution at the molecular level was extremely rare. In agreement with a prior bacterial evolutionary study [17], we found that gene duplication has only a minor role during compensatory evolution following gene loss. A more general issue is the extent to which mutations that affect gene expression could alone recover fitness [17],[39]. Although genetic changes in putative promoter regions were not overrepresented in our dataset (Binomial test, p = 0.87), 21 observed point mutations generated in-frame stop codons, most likely yielding proteins with compromised or no activities (see also Figure 7). These results indicate that fitness recovery following gene loss can partly be achieved purely through inactivation of other genes. Fourth, compensatory evolution promoted divergence of genomic diversification, and shifted the evolved population towards novel genomic expression states (Figure 4B). Despite substantial fitness improvements, the majority of the gene expression changes due to gene deletion remained unrestored during evolution. This finding is consistent with prior works arguing that no clear relationship exists between the change in mRNA expression of a gene and its requirement for growth in the same condition [40]. Fifth, independently evolved populations showed substantial fitness variation across environments that they had not been exposed to during laboratory evolution (Figure 5). These results suggest that accumulation of adaptive mutations during compensatory evolution generated substantial genetic differences between populations, and this variation can be uncovered upon environmental change. Taken together, several lines of evidence indicate that fitness gains in the evolved lines reflect accumulation of gene specific compensatory mutations rather than a global adaptation: (i) evolving wild-type control populations showed only minor changes in fitness, (ii) the rate of adaptation was genotype specific, (ii) convergence at the molecular across genotypes was extremely rare, (iv) evolution preferentially affected genes that are functionally related to that of the disrupted gene, and (v) compensatory mutations had no beneficial impact in a wild-type genetic background. The above results encouraged us to distinguish between two evolutionary scenarios. Organisms may attempt to restore the disrupted molecular function through mutations in genes with redundant functions (functional restoration). Alternatively, they may aim to minimize the cellular damage incurred by gene disruption (functional replacement). While the possibility of full functional restoration cannot be excluded, the rarity of compensation through mutations in gene duplicates and the plasticity of compensatory mutational effects across environments are consistent with the second scenario. Indeed, our work demonstrates that gene loss promotes genetic changes that have a large impact on evolutionary diversification, genomic expression, and viability upon environmental change. An important implication of our study is that the beneficial effects of compensatory mutations should frequently be conditional, and subsequent changes to the environment can reveal the hidden fitness effects (beneficial and detrimental alike). Lack of restoration of fitness across environments is broadly consistent with the emerging view that epistatic interactions are plastic across conditions [41],[42]. The perspective offered in this work leads to the re-formulation of several fundamental questions. First, it sheds light on an evolutionary paradox: while core cellular processes are generally conserved during evolution [43], the constituent genes are partly different across related species with similar lifestyles. We propose that gene content variation across species is partly due to the action of compensatory evolution and may not need to reflect changes in environmental conditions and the consequent passive loss of genes. Although the exact population genetic conditions facilitating this process remain to be elucidated, several observations are consistent with this view. Most notably, the phylogenetic conservation of indispensable genes depends on how easily the gene can be functionally replaced through enhanced expression of other genes [44]. Second, it has been suggested that deleterious mutations may act as stepping stones in adaptive evolution by providing access to fitness peaks that are not otherwise accessible [45],[46]. Indeed, our analysis revealed a few instances where the laboratory evolved lines displayed significantly higher than wild-type fitness in specific environments. Finally, given the prevalence of gene loss events during tumorigenesis, future work should elucidate whether similar processes drive the somatic evolution of cancer [47]. Materials and Methods Yeast Strains and Media All strains used in this study were derived from the BY4741 S. cerevisiae parental strain. Non-essential single-gene deletion strains from the haploid yeast deletion collection [40] (MATa; his3Δ 1; leu2Δ 0; met15Δ 0; ura3Δ 0; xxx::KanMX4) were used to systematically identify all gene disruptions with a significant growth defect. Slow-growing mutants were identified in two steps. An earlier study identified 671 gene deletants in diploid background, which showed a significant fitness defect on both rich and synthetic media [48]. We thus measured fitness of the corresponding MATa haploid strains by recording their growth curves in liquid media. We identified 187 deletants showing at least 10% growth rate defect, which constituted the set of ancestral strains subjected to laboratory evolution (for details of growth measurements see below). The slow-growing yeast deletants used in this study are listed in Table S1. The evolutionary experiment was conducted using rich liquid medium (YPD, 1% yeast extract, 2% peptone, 2% glucose). Solid media were prepared using 2% agar, which were found to be optimal for reproducible colony size measurement. Details on the media used in the phenotypic profiling experiment can be found in Table S4. Oleic acid and stearic acid was dissolved in DMSO as a 100 mM stock and added to the medium after autoclaving to a final concentration of 0.1 mM. Laboratory Evolution Compensatory adaptation refers to fitness gains in a gene deletion strain that are greater than fitness gains occurring in an isogenic wild-type strain. We conducted a series of laboratory evolutionary experiments using four independent populations of each of the 187 slow-growing deletants along with 22 independent lineages of an isogenic wild-type strain (referred to as evolving wild types). The YOR202W deletion strain was used as evolving wild-type control because the fitness of this strain is indistinguishable from the BY4741 parental wild-type strain [19]. Moreover, this strain carries the KanMX4 cassette in the nonfunctional his3Δ1 allele, thus it was possible to control for the reported mutation-generating effect of the KanMX4 cassette [36]. All strains were inoculated into randomly selected positions of 96-well plates. Four wells in different positions were not inoculated by cells to help plate identification and orientation. Cells were grown in standard laboratory rich media to minimize selection pressure originating from nutrient limitation. The presence of the KanXM4 cassette was not selected for during the evolutionary experiment, since G418 was omitted from the medium for two reasons. First, using G418 at 200 mg/l concentration decreases the growth rate of the unevolved wild-type control strain (unpublished data) and might lead to selection for increased resistance. Second, the usage of the drug at a growth-limiting concentration may induce mutagenesis through environmental stress response. To provide optimal growth conditions, plates were covered with sandwich cover (Enzyscreeen BV), shaken at 350 rpm, and incubated at 30°C. Using a handheld replicator, ∼105 cells (∼0.5 µl sample volume) were transferred every second day to 100 µl of fresh medium in 96-well plates resulting in ∼7.6 generations between transfers. The experiment was run for 104 days (∼400 generations total) and samples from days 0, 26, 52, 78, and 104 were frozen in 15% glycerol and kept at −80°C until fitness measurement. Cross-contamination events were regularly checked by PCR and visual inspection of empty wells (unpublished data). High-Throughput Fitness Measurements We used established protocols specifically designed to measure fitness in yeast populations [49]. Growth was assayed by monitoring the optical density (OD600) of liquid cultures of each strain using 384-well microtiter plates containing YPD medium (as during the evolutionary experiments). We used relative growth rate as a proxy for relative fitness (see below). Compared to laborious competition based fitness assays, this protocol allows estimating growth rate on a relatively large scale in an environment that is nearly identical to the one used in the evolutionary experiments. Growth Curve Recording Starter cultures were inoculated from frozen samples using 96-well plates. The starter plates were grown for 48 hours under identical conditions to the evolutionary experiment. 384-well plates filled with 60 µl rich medium per well were inoculated for growth curve recording from the starter plates using pintool with 1.58 mm floating pins. The pintool was moved by a Microlab Starlet liquid handling workstation (Hamilton Bonaduz AG) to provide uniform inoculum across all samples. The median blank corrected initial OD600 of the wells was 0.027. Each 384-well plate were inoculated with four different starter plates: one plate having the unevolved wild-type control as a reference strain in all wells in order to estimate various within-plate measurement biases, and three plates containing the same set of mutants from three of the five time points of the evolutionary experiment. The 384-well plates were incubated at 30°C in an STX44 (LiCONiC AG) automated incubator with alternating shaking speed every minute between 1,000 rpm and 1,200 rpm. Plates were transferred by a Microlab Swap 420 robotic arm (Hamilton Bonaduz AG) to Powerwave XS2 plate readers (BioTek Instruments Inc) every 20 minutes and cell growth was followed by recording the optical density at 600 nm. Six technical replicate measurements were executed on all strains sampled from each time-point of the evolutionary experiment. Measurements with growth curve irregularities were automatically removed. Only those strains were further analyzed where at least four technical replicate measurements remained after this quality control step. Growth Curve Analysis Growth rate was calculated from the obtained growth curves following an established procedure [49],[50]. To eliminate potential within-plate effects that might cause measurement bias, growth rates were normalized by the growth rate of neighboring reference wells that contained the wild-type controls. For each strain and each evolutionary time point, relative fitness was calculated as the median of the normalized growth rates of the technical replicates divided by the median growth rate of the wild-type controls. At day 0, the technical replicate measurements of the isogenic independently evolving lines were combined to calculate median ancestral fitness since by that time these populations had no independent evolutionary history. Stringent criteria were used to define the set of ancestor strains with substantial growth rate defect: a minimum of 10% fitness drop was required compared to the wild-type controls; significance was determined by one-tailed Wilcoxon rank sum test, p-value was corrected with a false discovery rate of 0.05. Identifying Lines Showing a Significant Compensatory Adaptation To determine whether the fitness defect of a given knock-out strain became compensated during the evolutionary experiment two criteria must have been met: First, the growth rate improvement had to be significant (one-tailed Wilcoxon rank sum test, p-value corrected with a false discovery rate of 0.05). Second, the growth rate increment of the knock-out strain had to be disproportionally larger than that of the evolving wild-type control strains. To test whether fitness gain in a knockout is higher than those occurring in the evolving control lines, we first fitted a normal distribution to the fitness improvement values of the evolving control lines. Next, we defined a fitness improvement cutoff, so that the probability that an evolving control line would show an improvement at least that high is less than 0.05. To evaluate the extent of evolutionary compensation, a relative compensation index was calculated according to the following formula:where WT and Δ means median normalized growth rate of the evolving wild-type control and the knock-out strain, respectively, measured before (start) and after (end) the evolutionary experiment. Thus, a relative compensation of 1 indicates that the knock-out strain reached the same fitness after evolution as the evolving wild-type control strains. See Table S1 for the whole dataset. Phenotypic Profiling across Environmental Conditions To study the pleiotropic effects of compensatory adaptation, we measured the fitnesses of 237 evolved lines carrying a single gene deletion, all evolved wild-type control lines along with the corresponding ancestors across various environmental conditions. As this experiment demands high-throughput analyses (over 14,000 data points), fitness was estimated by colony size on solid agar media. Moreover, it allowed direct comparison of the reliability of our measurements to results of a previous study (Figure S5). We prepared solid agar media of 14 different compositions to expose the strains to fundamentally diverse environments and to obtain sufficient throughput. Our list of 14 growth media was primarily based on a previous study [27] and included various carbon sources and stress conditions (Table S4). A robotized replicating system was set up for colony size based fitness measurement. The system consists of a Microlab Starlet liquid handling workstation (Hamilton Bonaduz AG) equipped with a pintool with 768 pins (S&P Robotics Inc) and a custom-made pintool sterilization station. Several aspects of the replication procedure had been experimentally customized to achieve uniform, reproducible inoculation of yeast cells. Fitness of the ancestor (day 0) and evolved strains (day 104) was approximated by measuring colony sizes of ordered arrays of strains at 768 density. First, four different 96-well plates of the evolutionary experiment were scaled up to arrays of 384 colonies: one having the unevolved wild-type control in all positions, and three different plates of the mutant set from the same time point. Then pairs of 384 arrays with corresponding strains from day 0 and 104 were combined to reach 768 density. With this set up, all evolving replicate lines derived from the same ancestral genotype from both day 0 and day 104 were grown on the same 768 plate to exclude potential plate-to-plate variations when comparing colony growth of ancestor and evolved lines. Four technical replicates of these 768 arrays were transferred into each of the 14 different media. After acclimatization to the media at 30°C for 48 hours the plates were replicated again onto the same type of media and photographed after 48 hours of incubation at 30°C. Digital images were processed to calculate colony sizes, and potential systematic biases in colony growth were eliminated (Text S1). For each growth environment, fitness of each original knock-out genotype at day zero and each independently evolving line at day 104 was determined as the median of the size of replicate colonies. The reliability of our experimental setup and data processing was confirmed by comparing the fitness measurements of ancestral knock-out strains with the published data of Dudley and colleagues (Figure S5) [27]. To determine whether an ancestor genotype shows a significantly altered fitness compared to the wild-type control in a given environment, we used a Wilcoxon rank sum test (with p-value corrected for each condition with a false discovery rate of 0.05). The same statistical test was used to determine whether the fitness of an evolved line is different from that of its ancestor in a given environment. See result in Table S1. Genome Sequencing To reveal the underlying molecular mechanisms of compensation, we subjected 41 strains to whole-genome re-sequencing. Our list of sequenced strains primarily included genotypes with large initial fitness defect, substantial fitness improvement and gradual fitness increase over the course of evolution. To be able to detect parallel evolution at the molecular level, we selected two to four independently evolving lines of each ancestor genotype for sequencing. Overall, 41 evolved lines from 14 deletion strains were chosen along with their corresponding ancestor strains. Candidates were re-streaked and single clones were isolated and their fitness increase was confirmed by growth curve recording. Genomic DNA was prepared using a glass bead lysis protocol: clones were inoculated into 5 ml YPD+G418 (200 mg/l) and grown to saturation at 30°C. Cells were pelleted and resuspended in 500 µl of lyis buffer (1% SDS, 50 mM EDTA, 100 mM Tris [pH 8]). Cells were mechanically disrupted by vortexing for 3 minutes at high speed with 500 µl glass bead (500 µm, acid washed). After adding 275 µl 7 M ammonium acetate, samples were incubated at 65°C for 5 minutes, followed by a second incubation on ice for 5 minutes. The samples were extracted with chloroform∶isoamyl alcohol (24∶1) and centrifuged for 10 minutes. The aqueous layer was transferred into a new tube and precipitated with 1 ml isopropanol, pelleted and washed with 70% ethanol, and resuspended in 500 µl RNaseA solution (50 ng/ml). After 30 minutes RNaseA treatment at room temperature, samples were chloroform∶isoamyl alcohol (24∶1) extracted, precipitated with 50 µl sodium acetate (3 M [pH 5.2]) and 1,250 µl ethanol, pelleted and washed with 70% ethanol. Finally, the genomic DNA was dissolved in water. The steps of re-sequencing was done by the UD-GenoMed Medical Genomic Technologies Ltd: amplified genomic shotgun libraries were run on the Illumina HighScan SC with 1×100 bp single read module resulting in an average coverage of about 80×. Reads were aligned to the S. cerevisiae EF4 genome assembly using the BWA software package [51] having the genomic repeats masked using RepeatMasking [52]. Variant calling was performed using the GATK software package [53]. Genomic single-nucleotide polymorphisms with less than 200 phred-scaled quality score or lower than 0.3 mutant/reference ratio were ignored. Duplications of large chromosomal segments or whole chromosomes were identified as increased read coverage of certain regions. Elevated read coverage of regions with a minimum of 25 kb length were accepted as duplications if both the Control-FREEC [54] (Wilcoxon rank-sum test, p<0.01) and the CNV-seq [55] (p<0.0001) software predicted significant alteration from the read coverage of the reference genome. Our primary aim was to analyze de novo mutational events. De novo mutations were identified as alterations from the reference genome specifically found in the evolved lines but not present in the ancestral strains. Mutations, which occurred before our evolutionary experiment but after the gene knock-out, are referred to as secondary ancestor mutations. These mutations were identified in the ancestral strains as SNPs and indels present only in the corresponding ancestor strain, not in any other ancestral strains. The rationale behind this consideration is not to classify mutations accumulated in the parental strain of the mutant library prior to the generation of the knock-out strain as a secondary ancestor mutation. The list of identified mutations can be found in Table S2. Ratio of Non-Synonymous to Synonymous SNPs Whole-genome re-sequencing revealed that 86% of SNPs in the coding regions were non-synonymous. To statistically test whether the ratio of non-synonymous to synonymous SNPs was higher than expected based on a neutral model of evolution, we employed the method of Barrick and colleagues [56]. Briefly, we took all different point mutations observed in protein coding regions and calculated the probability that 86% or more substitutions would result in a non-synonymous substitution if it occurred in a random coding position. The excess of non-synonymous substitution observed in the evolved genomes was significant (p = 0.003). Datasets Used for Bioinformatic Analysis To test whether the extent of evolutionary compensation is influenced by the disrupted gene's pleiotropy, we used three complementary measures of gene pleiotropy. Environmental pleiotropy of a non-essential gene was defined as the number of unique conditions in which the removal of the gene resulted in a fitness defect according to Dudley and colleagues [27]. Network pleiotropy was measured as the total number of protein-protein interactions reported in the BioGRID database [57]. Finally, multifunctionality of a gene was calculated on the basis of a set of GO terms considered to be specific by yeast geneticists, as previously described [58]. To investigate whether mutations accumulated during compensatory evolution preferentially affected genes that are functionally related to the disrupted gene, we used different measures of functional relatedness: co-membership within stable protein complexes, shared functional category, genetic interaction profile similarity, co-expression, and paralogy. For protein complexes we used the manually curated dataset based on tandem affinity purification/mass spectrometry studies (YHTP2008) from the Wodak lab [59]. For functional categories, the MIPS Functional Catalogue Database was downloaded [60]. Genetic interaction profile similarities were obtained from a large-scale genetic interaction screen study [21]. The authors calculated the genetic interaction profile for a given gene deletion genotype as the list of genetic interaction scores detected across all other genes in their dataset. The genetic interaction profile similarity between two genes was defined as the Pearson correlation value of the two genetic interaction profiles [21]. For calculating co-expression data, 247 normalized microarray datasets from the M3D database [61] were used to create an expression profile for each gene. In case of multiple replicates per experiment, the average normalized values were calculated, and employed further. For each gene pair, co-expression value was calculated as the Pearson correlation coefficient between the two expression profiles. Paralog gene pairs were identified by performing all-against-all BLASTP similarity searches of yeast open reading frames. We defined two genes as paralogs if (i) the BLAST score had an expected value E<10−8, (ii) alignment length exceeded 100 residues, (iii) sequence similarity was >30%, and (iv) they were not parts of transposons. Gene Expression Analysis Eight evolved lines were selected for microarray analysis, all of them showing high fitness following evolution (at least 20% initial fitness defect compared to the wild-type control and at least 20% fitness improvement as a result of the evolutionary process). The corresponding ancestral strains and the wild-type control were also subjected to gene expression profiling. Table S3 contains the list of strains. Candidates were re-streaked and single clones were isolated and their fitness increase was confirmed by growth curve recording. Two independent colonies of the wild-type control, evolved, and corresponding ancestor knock-out strains were inoculated into 15 ml YPD and grown overnight at 30°C. The saturated populations were diluted to an OD600 of 0.15 in 60 ml YPD and grown to early mid-log phase (OD600 0.6±0.05) in 250 ml Erlenmeyer flasks with 220 rpm shaking at 30°C. Cells were harvested by centrifugation (4,000 rpm, 3 min, 30°C) and immediately frozen in liquid nitrogen after removal of supernatant. Total RNA was prepared by hot acidic phenol extraction and cleaned up using the QIAGEN's RNAeasy kit. All steps after RNA isolation were automated using robotic liquid handlers as described previously [62]. Dual-channel 70-mer oligonucleotide arrays were used with a common reference pool of wild-type RNA. Quality control, normalization, and dye-bias correction was performed as described earlier [62]. The reported fold change is the average of the four replicate mutant profiles versus the average of all wild-type controls. A total of 58 transcripts showed stochastic changes in wild-type profiles and were excluded from the analyses. Differentially expressed genes were defined as those showing a 1.7-fold abundance change and a p-value<0.05 when comparing two strains. The raw dataset is available online at ArrayExpress (http://www.ebi.ac.uk/arrayexpress/, accession number E-MTAB-2352). Robustness of Results of the Transcriptome Analysis to Growth Rate Related Genes and Copy Number Variations All transcriptome comparisons of the wild-type, knockout, and evolved strains were repeated on a dataset where CNVs, genes showing expression response to aneuploidy, and growth rate related genes were excluded. CNVs were identified on the basis of the read coverage of the genome sequence data (Table S2) with the exception of one strain (Δrpl43a), which was not sequenced. In the case of Δrpl43a, whole chromosome duplication was predicted on the basis of visual inspection of expression profiles. The position of partial chromosome duplication was predicted by the Charm algorithm [63]. In evolved strains carrying aneuploid chromosomes, genes showing expression response to that particular aneuploidy were excluded from the transcriptome comparisons (data on the transcriptome effects of aneuploidy were obtained from [64]). Genes showing significant expression response to changes in growth rate were also excluded, as defined previously [65] on the basis of the growth rate measurements of Brauer and colleagues [66]. Strain Modifications to Investigate the Fitness Costs and Epistatic Effects of Compensatory Mutations The evolved lines of Δmdm34 were chosen for in-depth genetic analysis. The fitness cost of the set of compensatory mutations accumulated in the evolved Δmdm34 lineages was measured in wild-type genetic background. To this end, the MDM34 gene was re-introduced into the ancestor and evolved Δmdm34 lineages according to the delitto perfetto method [67]. First, the KanMX4 cassette in the ancestor and evolved Δmdm34 lineages was swapped with the CORE-UH cassette, containing the KlURA3 and hyg markers. Then the MDM34 open reading frame with longer than 0.3 kb flanking regions on both sides was amplified from the unevolved wild-type control strain and transformed into the cells to replace the CORE-UH cassette. The replacement of the KlURA3 marker was counter-selected using 5-FOA containing medium. The loss of hygr was confirmed, the site and orientation of gene replacement was verified by PCR and the sequence of the MDM34 gene was determined by capillary sequencing. In a second analysis, a point mutation identified in the MGA2 gene in one of the evolved Δmdm34 lineages was reinserted into both the wild-type and ancestor Δmdm34 background. This specific point mutation changes the 750th codon of MGA2 from GAT to TAT resulting in the incorporation of tyrosine instead of aspartic acid. We refer to the mutant allele as mga2-1. Using the delitto perfetto method [67], we introduced this point mutation into the unevolved wild-type control strain. First, the CORE-UH cassette was inserted into the genome at the desired position of the SNP. Then, two complementary oligonucleotides of 81 bp length with the sequence of the region of interest and the SNP in the 41st position were transformed. The replacement of the KlURA3 marker with the missense SNP was counter-selected using 5-FOA containing medium, loss of hygr was confirmed, and the result of the site-directed mutagenesis was verified by capillary sequencing. Attempts to introduce the mga2-1 mutation into the ancestor Δmdm34 strain in this way were not successful, presumably due to the severe slow growth of the intermediate strain that lacks both MDM34 and MGA2 gene in a functional form. To complement this, a helper plasmid with MDM34 gene (MoBY ORF Library [68]) was transformed into the cells prior to the site directed mutagenesis [69]. Because of the presence of the URA3 marker on the helper plasmid, the CORE-Hp53 cassette was used in this experiment. The steps of mutagenesis were similar as without the helper plasmid, which was removed by passaging cells through 5-FOA afterwards. qPCR Method Yeast samples were grown in 20 ml YPD medium to mid-log phase (0.8 OD600 value). RNA was extracted from 107 yeast cells by acidic phenol method using TRI Reagent Protocol (Sigma-Aldrich Co). The RNA samples were concentrated by the NucleoSpin RNA Plant Kit (Macherey-Nagel), according to the manufacturer's instructions. A total of 500 ng RNA was used as a template to prepare cDNA using the Maxima First Strand cDNA Synthesis kit (Thermo Scientific). Reactions without template were set up to detect contaminations of the reagents used in the cDNA synthesis. qPCR reactions were set up in 20 µl volume, using the following templates: no template control, 10 ng non-transcribed RNA and cDNA transcribed from 10 ng RNA. The qPCR reactions were run in a Bioer LineK Gene device, using 2× Maxima SYBR Green qPCR Master Mix (Thermo Scientific). All samples had three technical replicates. Gene expression was determined in arbitrary units using a standard curve fitted on triplicates of a four-step 10-fold dilution series. OLE1 expression level was determined relative to TUB1 expression level as an internal control. All control reactions, not treated with reverse transcriptase or not having template, gave Ct values at least 10 cycles higher than the corresponding samples. Yeast Strains and Media All strains used in this study were derived from the BY4741 S. cerevisiae parental strain. Non-essential single-gene deletion strains from the haploid yeast deletion collection [40] (MATa; his3Δ 1; leu2Δ 0; met15Δ 0; ura3Δ 0; xxx::KanMX4) were used to systematically identify all gene disruptions with a significant growth defect. Slow-growing mutants were identified in two steps. An earlier study identified 671 gene deletants in diploid background, which showed a significant fitness defect on both rich and synthetic media [48]. We thus measured fitness of the corresponding MATa haploid strains by recording their growth curves in liquid media. We identified 187 deletants showing at least 10% growth rate defect, which constituted the set of ancestral strains subjected to laboratory evolution (for details of growth measurements see below). The slow-growing yeast deletants used in this study are listed in Table S1. The evolutionary experiment was conducted using rich liquid medium (YPD, 1% yeast extract, 2% peptone, 2% glucose). Solid media were prepared using 2% agar, which were found to be optimal for reproducible colony size measurement. Details on the media used in the phenotypic profiling experiment can be found in Table S4. Oleic acid and stearic acid was dissolved in DMSO as a 100 mM stock and added to the medium after autoclaving to a final concentration of 0.1 mM. Laboratory Evolution Compensatory adaptation refers to fitness gains in a gene deletion strain that are greater than fitness gains occurring in an isogenic wild-type strain. We conducted a series of laboratory evolutionary experiments using four independent populations of each of the 187 slow-growing deletants along with 22 independent lineages of an isogenic wild-type strain (referred to as evolving wild types). The YOR202W deletion strain was used as evolving wild-type control because the fitness of this strain is indistinguishable from the BY4741 parental wild-type strain [19]. Moreover, this strain carries the KanMX4 cassette in the nonfunctional his3Δ1 allele, thus it was possible to control for the reported mutation-generating effect of the KanMX4 cassette [36]. All strains were inoculated into randomly selected positions of 96-well plates. Four wells in different positions were not inoculated by cells to help plate identification and orientation. Cells were grown in standard laboratory rich media to minimize selection pressure originating from nutrient limitation. The presence of the KanXM4 cassette was not selected for during the evolutionary experiment, since G418 was omitted from the medium for two reasons. First, using G418 at 200 mg/l concentration decreases the growth rate of the unevolved wild-type control strain (unpublished data) and might lead to selection for increased resistance. Second, the usage of the drug at a growth-limiting concentration may induce mutagenesis through environmental stress response. To provide optimal growth conditions, plates were covered with sandwich cover (Enzyscreeen BV), shaken at 350 rpm, and incubated at 30°C. Using a handheld replicator, ∼105 cells (∼0.5 µl sample volume) were transferred every second day to 100 µl of fresh medium in 96-well plates resulting in ∼7.6 generations between transfers. The experiment was run for 104 days (∼400 generations total) and samples from days 0, 26, 52, 78, and 104 were frozen in 15% glycerol and kept at −80°C until fitness measurement. Cross-contamination events were regularly checked by PCR and visual inspection of empty wells (unpublished data). High-Throughput Fitness Measurements We used established protocols specifically designed to measure fitness in yeast populations [49]. Growth was assayed by monitoring the optical density (OD600) of liquid cultures of each strain using 384-well microtiter plates containing YPD medium (as during the evolutionary experiments). We used relative growth rate as a proxy for relative fitness (see below). Compared to laborious competition based fitness assays, this protocol allows estimating growth rate on a relatively large scale in an environment that is nearly identical to the one used in the evolutionary experiments. Growth Curve Recording Starter cultures were inoculated from frozen samples using 96-well plates. The starter plates were grown for 48 hours under identical conditions to the evolutionary experiment. 384-well plates filled with 60 µl rich medium per well were inoculated for growth curve recording from the starter plates using pintool with 1.58 mm floating pins. The pintool was moved by a Microlab Starlet liquid handling workstation (Hamilton Bonaduz AG) to provide uniform inoculum across all samples. The median blank corrected initial OD600 of the wells was 0.027. Each 384-well plate were inoculated with four different starter plates: one plate having the unevolved wild-type control as a reference strain in all wells in order to estimate various within-plate measurement biases, and three plates containing the same set of mutants from three of the five time points of the evolutionary experiment. The 384-well plates were incubated at 30°C in an STX44 (LiCONiC AG) automated incubator with alternating shaking speed every minute between 1,000 rpm and 1,200 rpm. Plates were transferred by a Microlab Swap 420 robotic arm (Hamilton Bonaduz AG) to Powerwave XS2 plate readers (BioTek Instruments Inc) every 20 minutes and cell growth was followed by recording the optical density at 600 nm. Six technical replicate measurements were executed on all strains sampled from each time-point of the evolutionary experiment. Measurements with growth curve irregularities were automatically removed. Only those strains were further analyzed where at least four technical replicate measurements remained after this quality control step. Growth Curve Analysis Growth rate was calculated from the obtained growth curves following an established procedure [49],[50]. To eliminate potential within-plate effects that might cause measurement bias, growth rates were normalized by the growth rate of neighboring reference wells that contained the wild-type controls. For each strain and each evolutionary time point, relative fitness was calculated as the median of the normalized growth rates of the technical replicates divided by the median growth rate of the wild-type controls. At day 0, the technical replicate measurements of the isogenic independently evolving lines were combined to calculate median ancestral fitness since by that time these populations had no independent evolutionary history. Stringent criteria were used to define the set of ancestor strains with substantial growth rate defect: a minimum of 10% fitness drop was required compared to the wild-type controls; significance was determined by one-tailed Wilcoxon rank sum test, p-value was corrected with a false discovery rate of 0.05. Identifying Lines Showing a Significant Compensatory Adaptation To determine whether the fitness defect of a given knock-out strain became compensated during the evolutionary experiment two criteria must have been met: First, the growth rate improvement had to be significant (one-tailed Wilcoxon rank sum test, p-value corrected with a false discovery rate of 0.05). Second, the growth rate increment of the knock-out strain had to be disproportionally larger than that of the evolving wild-type control strains. To test whether fitness gain in a knockout is higher than those occurring in the evolving control lines, we first fitted a normal distribution to the fitness improvement values of the evolving control lines. Next, we defined a fitness improvement cutoff, so that the probability that an evolving control line would show an improvement at least that high is less than 0.05. To evaluate the extent of evolutionary compensation, a relative compensation index was calculated according to the following formula:where WT and Δ means median normalized growth rate of the evolving wild-type control and the knock-out strain, respectively, measured before (start) and after (end) the evolutionary experiment. Thus, a relative compensation of 1 indicates that the knock-out strain reached the same fitness after evolution as the evolving wild-type control strains. See Table S1 for the whole dataset. Phenotypic Profiling across Environmental Conditions To study the pleiotropic effects of compensatory adaptation, we measured the fitnesses of 237 evolved lines carrying a single gene deletion, all evolved wild-type control lines along with the corresponding ancestors across various environmental conditions. As this experiment demands high-throughput analyses (over 14,000 data points), fitness was estimated by colony size on solid agar media. Moreover, it allowed direct comparison of the reliability of our measurements to results of a previous study (Figure S5). We prepared solid agar media of 14 different compositions to expose the strains to fundamentally diverse environments and to obtain sufficient throughput. Our list of 14 growth media was primarily based on a previous study [27] and included various carbon sources and stress conditions (Table S4). A robotized replicating system was set up for colony size based fitness measurement. The system consists of a Microlab Starlet liquid handling workstation (Hamilton Bonaduz AG) equipped with a pintool with 768 pins (S&P Robotics Inc) and a custom-made pintool sterilization station. Several aspects of the replication procedure had been experimentally customized to achieve uniform, reproducible inoculation of yeast cells. Fitness of the ancestor (day 0) and evolved strains (day 104) was approximated by measuring colony sizes of ordered arrays of strains at 768 density. First, four different 96-well plates of the evolutionary experiment were scaled up to arrays of 384 colonies: one having the unevolved wild-type control in all positions, and three different plates of the mutant set from the same time point. Then pairs of 384 arrays with corresponding strains from day 0 and 104 were combined to reach 768 density. With this set up, all evolving replicate lines derived from the same ancestral genotype from both day 0 and day 104 were grown on the same 768 plate to exclude potential plate-to-plate variations when comparing colony growth of ancestor and evolved lines. Four technical replicates of these 768 arrays were transferred into each of the 14 different media. After acclimatization to the media at 30°C for 48 hours the plates were replicated again onto the same type of media and photographed after 48 hours of incubation at 30°C. Digital images were processed to calculate colony sizes, and potential systematic biases in colony growth were eliminated (Text S1). For each growth environment, fitness of each original knock-out genotype at day zero and each independently evolving line at day 104 was determined as the median of the size of replicate colonies. The reliability of our experimental setup and data processing was confirmed by comparing the fitness measurements of ancestral knock-out strains with the published data of Dudley and colleagues (Figure S5) [27]. To determine whether an ancestor genotype shows a significantly altered fitness compared to the wild-type control in a given environment, we used a Wilcoxon rank sum test (with p-value corrected for each condition with a false discovery rate of 0.05). The same statistical test was used to determine whether the fitness of an evolved line is different from that of its ancestor in a given environment. See result in Table S1. Genome Sequencing To reveal the underlying molecular mechanisms of compensation, we subjected 41 strains to whole-genome re-sequencing. Our list of sequenced strains primarily included genotypes with large initial fitness defect, substantial fitness improvement and gradual fitness increase over the course of evolution. To be able to detect parallel evolution at the molecular level, we selected two to four independently evolving lines of each ancestor genotype for sequencing. Overall, 41 evolved lines from 14 deletion strains were chosen along with their corresponding ancestor strains. Candidates were re-streaked and single clones were isolated and their fitness increase was confirmed by growth curve recording. Genomic DNA was prepared using a glass bead lysis protocol: clones were inoculated into 5 ml YPD+G418 (200 mg/l) and grown to saturation at 30°C. Cells were pelleted and resuspended in 500 µl of lyis buffer (1% SDS, 50 mM EDTA, 100 mM Tris [pH 8]). Cells were mechanically disrupted by vortexing for 3 minutes at high speed with 500 µl glass bead (500 µm, acid washed). After adding 275 µl 7 M ammonium acetate, samples were incubated at 65°C for 5 minutes, followed by a second incubation on ice for 5 minutes. The samples were extracted with chloroform∶isoamyl alcohol (24∶1) and centrifuged for 10 minutes. The aqueous layer was transferred into a new tube and precipitated with 1 ml isopropanol, pelleted and washed with 70% ethanol, and resuspended in 500 µl RNaseA solution (50 ng/ml). After 30 minutes RNaseA treatment at room temperature, samples were chloroform∶isoamyl alcohol (24∶1) extracted, precipitated with 50 µl sodium acetate (3 M [pH 5.2]) and 1,250 µl ethanol, pelleted and washed with 70% ethanol. Finally, the genomic DNA was dissolved in water. The steps of re-sequencing was done by the UD-GenoMed Medical Genomic Technologies Ltd: amplified genomic shotgun libraries were run on the Illumina HighScan SC with 1×100 bp single read module resulting in an average coverage of about 80×. Reads were aligned to the S. cerevisiae EF4 genome assembly using the BWA software package [51] having the genomic repeats masked using RepeatMasking [52]. Variant calling was performed using the GATK software package [53]. Genomic single-nucleotide polymorphisms with less than 200 phred-scaled quality score or lower than 0.3 mutant/reference ratio were ignored. Duplications of large chromosomal segments or whole chromosomes were identified as increased read coverage of certain regions. Elevated read coverage of regions with a minimum of 25 kb length were accepted as duplications if both the Control-FREEC [54] (Wilcoxon rank-sum test, p<0.01) and the CNV-seq [55] (p<0.0001) software predicted significant alteration from the read coverage of the reference genome. Our primary aim was to analyze de novo mutational events. De novo mutations were identified as alterations from the reference genome specifically found in the evolved lines but not present in the ancestral strains. Mutations, which occurred before our evolutionary experiment but after the gene knock-out, are referred to as secondary ancestor mutations. These mutations were identified in the ancestral strains as SNPs and indels present only in the corresponding ancestor strain, not in any other ancestral strains. The rationale behind this consideration is not to classify mutations accumulated in the parental strain of the mutant library prior to the generation of the knock-out strain as a secondary ancestor mutation. The list of identified mutations can be found in Table S2. Ratio of Non-Synonymous to Synonymous SNPs Whole-genome re-sequencing revealed that 86% of SNPs in the coding regions were non-synonymous. To statistically test whether the ratio of non-synonymous to synonymous SNPs was higher than expected based on a neutral model of evolution, we employed the method of Barrick and colleagues [56]. Briefly, we took all different point mutations observed in protein coding regions and calculated the probability that 86% or more substitutions would result in a non-synonymous substitution if it occurred in a random coding position. The excess of non-synonymous substitution observed in the evolved genomes was significant (p = 0.003). Datasets Used for Bioinformatic Analysis To test whether the extent of evolutionary compensation is influenced by the disrupted gene's pleiotropy, we used three complementary measures of gene pleiotropy. Environmental pleiotropy of a non-essential gene was defined as the number of unique conditions in which the removal of the gene resulted in a fitness defect according to Dudley and colleagues [27]. Network pleiotropy was measured as the total number of protein-protein interactions reported in the BioGRID database [57]. Finally, multifunctionality of a gene was calculated on the basis of a set of GO terms considered to be specific by yeast geneticists, as previously described [58]. To investigate whether mutations accumulated during compensatory evolution preferentially affected genes that are functionally related to the disrupted gene, we used different measures of functional relatedness: co-membership within stable protein complexes, shared functional category, genetic interaction profile similarity, co-expression, and paralogy. For protein complexes we used the manually curated dataset based on tandem affinity purification/mass spectrometry studies (YHTP2008) from the Wodak lab [59]. For functional categories, the MIPS Functional Catalogue Database was downloaded [60]. Genetic interaction profile similarities were obtained from a large-scale genetic interaction screen study [21]. The authors calculated the genetic interaction profile for a given gene deletion genotype as the list of genetic interaction scores detected across all other genes in their dataset. The genetic interaction profile similarity between two genes was defined as the Pearson correlation value of the two genetic interaction profiles [21]. For calculating co-expression data, 247 normalized microarray datasets from the M3D database [61] were used to create an expression profile for each gene. In case of multiple replicates per experiment, the average normalized values were calculated, and employed further. For each gene pair, co-expression value was calculated as the Pearson correlation coefficient between the two expression profiles. Paralog gene pairs were identified by performing all-against-all BLASTP similarity searches of yeast open reading frames. We defined two genes as paralogs if (i) the BLAST score had an expected value E<10−8, (ii) alignment length exceeded 100 residues, (iii) sequence similarity was >30%, and (iv) they were not parts of transposons. Gene Expression Analysis Eight evolved lines were selected for microarray analysis, all of them showing high fitness following evolution (at least 20% initial fitness defect compared to the wild-type control and at least 20% fitness improvement as a result of the evolutionary process). The corresponding ancestral strains and the wild-type control were also subjected to gene expression profiling. Table S3 contains the list of strains. Candidates were re-streaked and single clones were isolated and their fitness increase was confirmed by growth curve recording. Two independent colonies of the wild-type control, evolved, and corresponding ancestor knock-out strains were inoculated into 15 ml YPD and grown overnight at 30°C. The saturated populations were diluted to an OD600 of 0.15 in 60 ml YPD and grown to early mid-log phase (OD600 0.6±0.05) in 250 ml Erlenmeyer flasks with 220 rpm shaking at 30°C. Cells were harvested by centrifugation (4,000 rpm, 3 min, 30°C) and immediately frozen in liquid nitrogen after removal of supernatant. Total RNA was prepared by hot acidic phenol extraction and cleaned up using the QIAGEN's RNAeasy kit. All steps after RNA isolation were automated using robotic liquid handlers as described previously [62]. Dual-channel 70-mer oligonucleotide arrays were used with a common reference pool of wild-type RNA. Quality control, normalization, and dye-bias correction was performed as described earlier [62]. The reported fold change is the average of the four replicate mutant profiles versus the average of all wild-type controls. A total of 58 transcripts showed stochastic changes in wild-type profiles and were excluded from the analyses. Differentially expressed genes were defined as those showing a 1.7-fold abundance change and a p-value<0.05 when comparing two strains. The raw dataset is available online at ArrayExpress (http://www.ebi.ac.uk/arrayexpress/, accession number E-MTAB-2352). Robustness of Results of the Transcriptome Analysis to Growth Rate Related Genes and Copy Number Variations All transcriptome comparisons of the wild-type, knockout, and evolved strains were repeated on a dataset where CNVs, genes showing expression response to aneuploidy, and growth rate related genes were excluded. CNVs were identified on the basis of the read coverage of the genome sequence data (Table S2) with the exception of one strain (Δrpl43a), which was not sequenced. In the case of Δrpl43a, whole chromosome duplication was predicted on the basis of visual inspection of expression profiles. The position of partial chromosome duplication was predicted by the Charm algorithm [63]. In evolved strains carrying aneuploid chromosomes, genes showing expression response to that particular aneuploidy were excluded from the transcriptome comparisons (data on the transcriptome effects of aneuploidy were obtained from [64]). Genes showing significant expression response to changes in growth rate were also excluded, as defined previously [65] on the basis of the growth rate measurements of Brauer and colleagues [66]. Strain Modifications to Investigate the Fitness Costs and Epistatic Effects of Compensatory Mutations The evolved lines of Δmdm34 were chosen for in-depth genetic analysis. The fitness cost of the set of compensatory mutations accumulated in the evolved Δmdm34 lineages was measured in wild-type genetic background. To this end, the MDM34 gene was re-introduced into the ancestor and evolved Δmdm34 lineages according to the delitto perfetto method [67]. First, the KanMX4 cassette in the ancestor and evolved Δmdm34 lineages was swapped with the CORE-UH cassette, containing the KlURA3 and hyg markers. Then the MDM34 open reading frame with longer than 0.3 kb flanking regions on both sides was amplified from the unevolved wild-type control strain and transformed into the cells to replace the CORE-UH cassette. The replacement of the KlURA3 marker was counter-selected using 5-FOA containing medium. The loss of hygr was confirmed, the site and orientation of gene replacement was verified by PCR and the sequence of the MDM34 gene was determined by capillary sequencing. In a second analysis, a point mutation identified in the MGA2 gene in one of the evolved Δmdm34 lineages was reinserted into both the wild-type and ancestor Δmdm34 background. This specific point mutation changes the 750th codon of MGA2 from GAT to TAT resulting in the incorporation of tyrosine instead of aspartic acid. We refer to the mutant allele as mga2-1. Using the delitto perfetto method [67], we introduced this point mutation into the unevolved wild-type control strain. First, the CORE-UH cassette was inserted into the genome at the desired position of the SNP. Then, two complementary oligonucleotides of 81 bp length with the sequence of the region of interest and the SNP in the 41st position were transformed. The replacement of the KlURA3 marker with the missense SNP was counter-selected using 5-FOA containing medium, loss of hygr was confirmed, and the result of the site-directed mutagenesis was verified by capillary sequencing. Attempts to introduce the mga2-1 mutation into the ancestor Δmdm34 strain in this way were not successful, presumably due to the severe slow growth of the intermediate strain that lacks both MDM34 and MGA2 gene in a functional form. To complement this, a helper plasmid with MDM34 gene (MoBY ORF Library [68]) was transformed into the cells prior to the site directed mutagenesis [69]. Because of the presence of the URA3 marker on the helper plasmid, the CORE-Hp53 cassette was used in this experiment. The steps of mutagenesis were similar as without the helper plasmid, which was removed by passaging cells through 5-FOA afterwards. qPCR Method Yeast samples were grown in 20 ml YPD medium to mid-log phase (0.8 OD600 value). RNA was extracted from 107 yeast cells by acidic phenol method using TRI Reagent Protocol (Sigma-Aldrich Co). The RNA samples were concentrated by the NucleoSpin RNA Plant Kit (Macherey-Nagel), according to the manufacturer's instructions. A total of 500 ng RNA was used as a template to prepare cDNA using the Maxima First Strand cDNA Synthesis kit (Thermo Scientific). Reactions without template were set up to detect contaminations of the reagents used in the cDNA synthesis. qPCR reactions were set up in 20 µl volume, using the following templates: no template control, 10 ng non-transcribed RNA and cDNA transcribed from 10 ng RNA. The qPCR reactions were run in a Bioer LineK Gene device, using 2× Maxima SYBR Green qPCR Master Mix (Thermo Scientific). All samples had three technical replicates. Gene expression was determined in arbitrary units using a standard curve fitted on triplicates of a four-step 10-fold dilution series. OLE1 expression level was determined relative to TUB1 expression level as an internal control. All control reactions, not treated with reverse transcriptase or not having template, gave Ct values at least 10 cycles higher than the corresponding samples. Supporting Information Figure S1. Fitness trajectories often show a saturating trend by day 104 of the evolution experiment. Fitness was measured at five time points during laboratory evolution (at day 0, 26, 52, 78, and 104), and fitness improvements were tested for each line and time interval (Wilcoxon rank-sum test, with a p-value cut-off of 0.05, see Methods and Table S10). (A) focuses on lines that showed one significant fitness improvement during the four 26-day time intervals. There is a strong (5-fold) depletion of lines that showed a fitness improvement in the last time step of the evolutionary experiment (eight out 159 cases, 40 expected, Chi-square test, p<10−8), indicating saturating compensatory evolution. (B) Representative examples of fitness trajectories showing a saturating trend (replicate lines of six genotypes are depicted). https://doi.org/10.1371/journal.pbio.1001935.s001 (TIF) Figure S2. The extent of compensatory evolution in knock-outs is genotype-specific. Here, we tested whether there are inherent differences in the propensity for compensation across genotypes (i.e., lines carrying different gene deletions). We defined compensatory evolution as a fitness increase that is disproportionally large relative to that in the evolving wild-type lines (Table S1). Accordingly, genotypes can be classified into three major categories on the basis of the fraction of corresponding lines fulfilling the above criteria (none, mixed, all). To assess the degree of departure from random expectation a randomization protocol was used. It generated a distribution of the above three categories under the assumption that all genotypes are equally likely to gain high fitness during the course of laboratory evolution. Specifically, the matrix of lines was shuffled one thousand times (gray bars) and the above categories were recalculated. The analysis revealed a strong enrichment of genotypes where all lines were compensated (“all”) and genotypes where none of the lines were compensated (“none”), while the “mixed” category was relatively rare (a). This result is not simply due to the fact that null mutations with more severe defects are especially likely to be compensated for. When only genotypes with similar initial fitness defects were considered, the trend remained (b,c,d). The four plots show the observed and randomly expected distributions a, for the whole dataset; b, c, d, for initial fitness ranges <0.7, 0.7–0.8, >0.8, respectively. Genotypes where either all or none of the evolutionary lines showed compensation are significantly enriched in all four cases, the corresponding Chi-square test p-values for a, b, c, and d are <10−20, 0.013, 7×10−6 and 10−8, respectively. https://doi.org/10.1371/journal.pbio.1001935.s002 (TIF) Figure S3. Global transcriptome changes following compensatory evolution. (A and B) were prepared by reproducing the main results of Figure 4, after excluding genes from the transcriptome profiles that (i) show copy number changes in the evolved lines, (ii) change expression level in aneuploid lines [13], or (iii) whose expression level depends on cellular growth rate (for details see Materials and Method). (A) The Euclidean distances of microarray profiles of the evolved evolutionary line from its ancestor and from wild type (WT) were calculated and normalized to the ancestor–wild type distance for each genotype (Table S11). The distances of the points on the figure are proportional to the calculated profile distances. For each genotype triplet, distances were calculated on the basis of those genes that are differentially expressed in at least one of the pairwise comparisons. (B) The figure focuses on the subset of genes that showed expression change upon gene deletion, and shows the fraction of these genes that changed expression during evolution in the opposite direction (i.e., evolution towards restoration of wild-type expression level; see inset). With one major exception (Δmdm34), only a small fraction of the expression changes were restored in the evolved lines (Table S11). The threshold for expression change was 1.7-fold-change and p<0.05, as previously described [14]. https://doi.org/10.1371/journal.pbio.1001935.s003 (TIF) Figure S4. Pleiotropic effects and mechanism of compensation of Δmdm34. (A) Diversity of pleiotropic effects in independently evolved lines. Relative fitness across environments of isolated clones of independently evolving lines founded from the same Δmdm34 genotype were measured as colony sizes grown on different media (Table S12). Genotypes are indicated on the left, the growth media are indicated above the heat map. For media composition and abbreviations, see Table S4. Values were normalized to that of the ancestral Δmdm34 strain in the corresponding environment. In (A) and (D) log2 values are shown according to the color coding. (B) Quantitative PCR confirmation of upregulation of OLE1 in both the evolved line carrying the mga2-1 mutation and in the Δmdm34 mga2-1 double mutant strain (Table S12). OLE1 expression was measured relative to TUB1 as an internal control and expression values were normalized to Δmdm34 ancestor. Error bars show standard error. (C) Addition of oleic acid to the medium suppresses the fitness defect of Δmdm34, but does not affect the fitness of the evolved line carrying the mga2-1 mutation or the strain carrying both Δmdm34 and mga2-1 mutations. Fitness was measured as colony sizes relative to unevolved wild-type control on rich media supplemented with DMSO as solvent control (non-treated), 0.1 mM oleic acid and 0.1 mM stearic acid (Table S12). For each genotype relative fitness change compared to the corresponding non-treated strain is shown. Error bars show standard error. (D) A specific point mutation in MGA2 recapitulates the pleiotropic effects of compensatory evolution observed in evolved line 1. Relative fitnesses of Δmdm34 evolving line 1, and Δmdm34 mga2-1 double mutant were measured as colony sizes grown on different media (Table S12). Genotypes are indicated on the left, the growth media are indicated above the heat map. For media composition and abbreviations, see Table S4. Values were normalized to that of the ancestral Δmdm34 strain in the corresponding environment. https://doi.org/10.1371/journal.pbio.1001935.s004 (TIF) Figure S5. Validation of the phenotypic profiling experiment. We compared our colony size measurements (Table S1) of the ancestral knockout strains to a published fitness profiling of the yeast deletion collection [4]. In the environments that match the published study, we find a good agreement between our data and the classification of Dudley and colleagues [4]. In each environment, knockouts present in our dataset were labeled as “no defect” versus “no/slow growth” based on Dudley and colleague's data. A significant difference was found between the two groups in our continuous fitness measurement (y-axis) in each of the environments (one-tailed Wilcoxon rank-sum test; */**/*** indicates p-value<0.05/0.01/0.001, respectively). https://doi.org/10.1371/journal.pbio.1001935.s005 (TIF) Table S1. Fitness of strains in various environments. The table includes fitness values of ancestor and evolved strains as measured in liquid YPD and in different agar media. Pleiotropy measures and GO process terms of the deleted genes are also presented. https://doi.org/10.1371/journal.pbio.1001935.s006 (XLSX) Table S2. Mutations identified by Illumina next generation sequencing. The table contains all the identified de novo and ancestral mutations in the sequenced genomes. https://doi.org/10.1371/journal.pbio.1001935.s007 (XLS) Table S3. Microarray analysis results. The table contains microarray data on all ancestral and evolved lines subjected to microarray analysis. https://doi.org/10.1371/journal.pbio.1001935.s008 (XLS) Table S4. Composition of media used for phenotypic profiling. Each media contained 1% yeast extract, 2% pepton, 2% agar, and different carbon sources. Some media also contained growth inhibitors as indicated. Concentration of drug inhibitors were set to have a minor but detectable growth inhibitory effect on the evolving wild-type control (unpublished data). The list of 14 growth media was primarily based on a previous study [4]. https://doi.org/10.1371/journal.pbio.1001935.s009 (XLSX) Table S5. Data supporting Figure 3D. The table contains fitness measurements supporting dosage compensation of Δrpl6b by increased copy number of RPL6A. https://doi.org/10.1371/journal.pbio.1001935.s010 (XLSX) Table S6. Data supporting Figure 4B and 4C. (4B) Euclidean distances between pairs of wild-type evolved and ancestor knock-out strains, and also between the corresponding biological replicates. (4C) Categories of expression changes for each gene in the eight evolved knockout strains. Genes, which show initial expression change in the knockout can be categorized as restored, if expression during evolution goes in the opposite direction or unrestored if not. The category “other” includes genes not showing initial expression change. https://doi.org/10.1371/journal.pbio.1001935.s011 (XLSX) Table S7. Data supporting Figure 6. The table contains fitness (colony size) data employed for epistasis analysis between the mdm34 gene deletion and the mutations accumulated in the evolving strains (Figure 6B) and between the mdm34 gene deletion and one particular compensatory mutation (“mga2-1”) (Figure 6C) in both YPD and acetic acid, respectively. https://doi.org/10.1371/journal.pbio.1001935.s012 (XLSX) Table S8. Data supporting Figure 7. The table contains colony size measurement data on the environment-dependent compensation of the deletion of rpb9 by a loss-of-function mutation of whi2. https://doi.org/10.1371/journal.pbio.1001935.s013 (XLSX) Table S9. Data supporting Figure 8. Table includes single-gene knockout fitness and relative frequency of suppressing mutations for 3880 non-essential yeast genes. https://doi.org/10.1371/journal.pbio.1001935.s014 (XLSX) Table S10. Data supporting Figure S1. The table contains data on the fitness trajectories of the evolving strains. Fitness was measured at day 0, 26, 52, 78, and 104. The columns “improved day x-y” show whether there is a statistically significant fitness improvement between day x and y, as assessed by one-sided Wilcoxon tests (with false discovery rate correction, p<0.05 cutoff). https://doi.org/10.1371/journal.pbio.1001935.s015 (XLSX) Table S11. Data supporting Figure S3. (S3A) Euclidean distances between pairs of wild-type, evolved, and ancestor knock-outs, after excluding genes from the transcriptome profiles that (i) show copy number changes in the evolved lines, (ii) change expression level in aneuploid lines, or (iii) whose expression level depends on cellular growth rate. (S3B) Table includes categories of expression changes for each gene in the eight evolved knockout strains, excluding genes from the transcriptome profiles that (i) show copy number changes in the evolved lines, (ii) change expression level in aneuploid lines, or (iii) whose expression level depends on cellular growth rate. Genes displaying an initial expression change in the knockout can be categorized as restored, if its expression level changes in the opposite direction during evolution, or unrestored. The category “other” includes genes that did not display an initial expression change. https://doi.org/10.1371/journal.pbio.1001935.s016 (XLSX) Table S12. Data supporting Figure S4. The table contains data on the pleiotropic effects and mechanism of compensation of the deletion strain Δmdm34. https://doi.org/10.1371/journal.pbio.1001935.s017 (XLSX) Text S1. Additional analyses supporting the prevalence of and mechanisms underlying compensatory evolution following gene loss. The text includes a bioinformatic analyses of deleterious loss-of-function variants in natural yeast populations, a case study on compensatory mutations, and a brief description of image analysis used for measuring the extent of compensatory evolution. https://doi.org/10.1371/journal.pbio.1001935.s018 (DOC) Acknowledgments We thank Charles Boone and Francesca Storici for sharing plasmids and yeast strains and Andrásné Borka for technical assistance. We are also grateful to Laurence D. Hurst for his insightful comments.

doi: 10.1371/journal.pbio.1001918pmid: 25093680

Introduction The neurotrophin receptor p75 is a member of the tumor necrosis factor receptor (TNFR) superfamily and has four extracellular cysteine rich domains, a single transmembrane (TM) domain, and an intracellular domain (ICD) comprising a juxtamembrane and a death domain (DD) [1]–[5]. Depending on co-receptor partners and cellular contexts, p75 may play seemingly opposing effects in multiple systems. For example, p75 interacts with Trk receptors to promote neurotrophin-dependent nerve growth. In contrast, p75 has been shown to play a role in apoptosis when binding to pro-neurotrophins and with the co-receptor sortilin [4]. In addition, p75 inhibits nerve growth mediated by myelin-associated inhibitors via functioning in part as a co-receptor for the GPI-linked neuronal Nogo-66 receptor (NgR) [6] or another non-NgR molecule that is yet to be identified [7],[8]. Elucidation of the mechanisms that modulate p75-mediated signaling may increase our understanding of neural development and nerve injury. Upon nerve injury in adult mammals, factors at the injury site such as myelin-associated inhibitors inhibit regeneration of injured axons, resulting in permanent disability. Axon regeneration is blocked by the presence of multiple types of nerve growth inhibitors, such as myelin-associated inhibitors from damaged myelins, chondroitin sulphate proteoglycans, and repulsive axon-guidance molecules expressed by reactive glial cells [9]–[12]. The structurally dissimilar myelin-associated inhibitors Nogo66, MAG, and OMgp inhibit axon growth by binding to the NgR, a GPI-linked protein, which then transduces the inhibitory signal into the cell by binding to co-receptors with intracellular signaling domains, such as p75 [13],[14] or TROY [15],[16]. LINGO-1 also plays a role in NgR signaling [17]. Downstream from their receptor binding, these myelin inhibitors trigger inhibition of axonal growth through the activation of the small GTPase Rho [18]–[21] in a protein kinase C (PKC)-dependent manner [22]. Targeting this complex has been described to lead to the promotion of neurite outgrowth, oligodendrocyte proliferation and differentiation, and inhibition of cell death. p45 is highly homologous in sequence to p75. It is also called neurotrophin receptor homologue 2 (NRH2) [25], neurotrophin receptor alike DD protein (NRADD) [24], or p75-like apoptosis inducing DD protein (PLAIDD) [23]. P45 displays strong sequence similarity to p75 in the TM, juxtamembrane, and DD regions [26]. P45 contains a truncated and short extracellular domain (ECD) with no neurotrophin binding domain. It has been shown previously that p45 associates with p75 and with TrkA receptors [23],[25],[27]. In addition, p45 participates in the trafficking of sortilin to the plasma membrane [27]. However, its role in other p75-regulated signaling pathways has not been studied. In this study, we have explored the modulation of p75/NgR signaling. The results indicate that p45 heterodimerizes with p75 and, thereby, impedes the formation of p75 homodimer that is required for the p75/NgR complex formation and its downstream activation of RhoA GTPase. In addition, we found that p45 binds p75 through both the TM and the ICDs. Furthermore, we showed that a cysteine–cysteine interaction within the TM domain of p45 and p75 is required for stabilization of their heterodimer formation. The results reveal a new mechanism of modulating p75-mediated inhibitory signaling via heterodimer formation with a member of the TNFR superfamily such as p45. Results p45 Forms a Stable Complex with p75 and Is Up-Regulated upon Nerve Injury p45 contains a DD (Figure 1A). Because the function of DDs is to bind the DD of other members of the DD superfamily in order to transduce signals [28], we investigated whether p45 can interact with other members of the DD superfamily, such as p75, FADD, TNFR1, and Fas. As shown in Figure 1B, p45 can be co-immunoprecipitated with p75 or FADD when co-expressed in 293 cells, but not with caspase-8, Fas, or TNFR1. Recently we characterized the interaction of p45 with FADD and its role upon spinal cord injury [29]. In the present study, we focus on the interaction of p45 with p75. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 1. Interactions between p75 and p45. (A) A schematic diagram showing domains of p45 in comparison with p75. TM, transmembrane domain; DD, putative death domain; PDZ, putative PDZ binding domain. The degrees of identity and homology in amino acid residues between mouse p75 and p45 are shown in percentages. (B) P45 forms a complex with FADD and p75. V5-tagged TNFR1, human Fas, mouse Fas, p75, FADD, or Caspase-8 was transfected into CrmA/Flag-p45/293 stable cells. The lysates were immunoprecipitated with anti-Flag antibodies and immunoblotted with anti-V5 antibodies. (C) P7 cerebellum extracts were immunprecipitated with anti-p75 ECD antibody (9651) or anti-p45 ECD antibody (6750) followed by immnuoblotting with anti-p75 antibody (Buster). T, total lysate. (D) Western blotting analysis of p45 and p75 expression in the brain, spinal cord, and DRG of P0 and adult mice. (E) Increased expression of p45 and p75 in the spinal cord and sciatic nerve following sciatic nerve injury. (Top) Spinal cord sections were immunostained with anti-p45 or -p75 antibodies. p45 and p75 immunoreactivities were markedly increased in the ipsilateral side as compared to the contralateral side. Higher magnification indicated that expression of p45 and p75 is increased in motor neurons. (F) Longitudinal sections of crushed and uncrushed sciatic nerves were immunostained with antibodies against p45, p75, or neurofilament. The levels of p45 and p75 were markedly increased in the distal (D) portion of sciatic nerves as compared to the proximal (P) end and the uncrushed (UC) nerve. https://doi.org/10.1371/journal.pbio.1001918.g001 p45 and p75 share a high degree of amino acid similarity in their TM domain (94%), including conserved cysteine residues [30], and the ICD (50%) (Figure 1A). However, the ECD of p45 is short and has no binding sites for neurotrophins (Figure 1A). p45 and p75 are expressed in both the peripheral nervous system (PNS) and central nervous system (CNS) during development [26]. We found that p75 and p45 form an immunocomplex in cerebellum extracts (Figure 1C). The expression level of p45 is high in embryonic but is significantly reduced in adult tissues (Figure 1D). However, p45 is up-regulated after sciatic nerve injury in the spinal cord and sciatic nerve in a similar fashion as p75 (Figure 1E,F). Because p45 shares a similar protein sequence with p75 and has similar up-regulated expression patterns with p75 after injury, we decided to investigate whether p45 regulates signaling mechanisms involving p75. p45 Interferes with p75/NgR Signaling p75 mediates nerve growth inhibition by myelin-associated inhibitors via functioning in part as a co-receptor for GPI-linked neuronal NgR [6]. Because p45 and p75 form a stable complex (Figure 1B), we investigated whether p45 interferes with or enhances the formation of the complex between p75 and NgR and subsequent signaling through the p75/NgR complex. When HEK293 cells were co-transfected with p75 and Flag-tagged human NgR (Flag-hNgR) expression constructs followed by immunoprecipitation with anti-p75ICD antibodies, we found that p45 markedly reduced the levels of complex formation between p75 and NgR (Figure 2A) and it does in a concentration-dependent manner (Figure 2B). As a control, p45 is not able to bind to NgR directly (Figure S1). To further characterize the domains of p45 involved in such inhibition, we made several deletion constructs of p45 (Figure S2 and Table S1). As shown in Figure S2, both p45 intracellular and TM domains are necessary for p45 inhibition of p75/NgR complex formation. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 2. p45 interferes with the p75-NgR interaction and signaling. (A) Lysates from HEK293 cells co-transfected with vectors expressing p45, p75, and Flag-tagged human NogoR (Flag-hNgR) were immunoprecipitated with anti-p75 antibodies and probed with anti-Flag antibodies. The results showed that p75 and hNgR form a complex and presence of p45 reduces the formation of p75/hNgR complex. (B) Co-transfection with varying amount of p45-expressing vectors markedly reduced p75/hNgR complex formation in a concentration-dependent fashion. (C) Quantification of the increase of RhoA activity after MAG-Fc treatment. In wild-type (WT) CGN cultures, when treated with MAG-Fc, the RhoA activity increased by 50%. However, in CGN cultures derived from Thy1-p45 transgenic mice, the increase in RhoA activity induced by MAG-Fc treatment is completely abolished. The value is expressed as percent over control. The value is derived from three independent experiments. * p<0.05, Student's t test. The data can be found in Table S1. (D) CGNs were seeded on glass coverslips coated with inhibitory substrates, grown for 14–18 h, and immunofluorescently stained with Tuj1 (Red) and anti-p45 (Green) antibodies. Scale bar, 50 µm. (E) Quantitative analysis of neurite length from the outgrowth assay, using Nogo66-GST, myelin, or HEK293 cells expressing MAG as inhibitory substrates. The data are represented as mean ± SEM (* p<0.001) and can be found in Table S1. Overexpression of p45 promoted neurite outgrowth of CGNs on inhibitory substrates. https://doi.org/10.1371/journal.pbio.1001918.g002 We examined the possibility that p45 would antagonize signaling through the p75/NgR complex. Previous results demonstrated that p75 is required for MAG-induced RhoA activation through NgR [20],[31]. RhoA activation is necessary for neurite outgrowth inhibition mediated by myelin-associated inhibitors. Thus, we examined whether overexpression of p45 blunts signaling through the p75/NgR complex. Postnatal day 7 (P7) cerebellar granule neurons (CGNs) that express low levels of endogenous p45 were transfected with full-length, capped p45 RNAs containing poly(A) tails generated by an in vitro transcription system. As shown in Figure S3A, the level of p45 protein was markedly elevated 24 h following RNA transfection. The cultures were then serum-starved and treated with Fc or MAG-Fc proteins. The level of activated RhoA was measured. As illustrated in Figure S3B, overexpression of p45 blocks MAG-Fc–induced RhoA activation. We also quantitatively measured RhoA activation using the G-LISA kit (Cytoskeleton Inc.) on CGN cultures from wild-type mice and Thy1-p45 transgenic mice that consistently overexpress p45 under a Thy1 promoter [29]. As shown in Figure 2C, MAG-Fc treatment of the WT CGNs induced 50% increase in the RhoA activity, whereas the MAG-Fc–induced RhoA activation is completely abolished in Thy1-p45 CGNs (Table S1). These results suggest that p45 is capable of effectively blocking RhoA activation through the p75/NgR complex. We then examined whether overexpression of p45 prevents neurite outgrowth inhibition induced by Nogo66, MAG, or CNS myelin. P7 CGNs were transfected with p45 RNAs, plated onto dishes previously coated with different substrates, and allowed to grow overnight. The cultures were double immunostained with antibodies against p45 (green) and neurotubulin (TuJ1, red) (Figure 2D). The neurite length of control CGNs and transfected CGNs that display increased p45-immunoreactivity over control CGNs was measured. As shown in Figure 2E, neurite outgrowth inhibition elicited by Nogo66 is alleviated by p45 overexpression (Table S1). Similarly, p45 overexpression significantly promotes neurite outgrowth that was otherwise inhibited when cultured on dishes coated with CNS myelin or MAG-expressing cells (Figure 2E). These results support the idea that p45 promotes neurite outgrowth. It is worth noting that despite NgR has been implicated in mediating nerve growth inhibition induced by myelin inhibitors in culture, neurite outgrowth of CGNs from NgR null mutants is still inhibited by myelin inhibitors [7],[8]. Recent results suggest that NgR is required only for the acute growth cone-collapsing but not chronic growth-inhibitory actions of myelin inhibitors [32]. Furthermore, no measurable corticospinal tract regeneration was observed in mice lacking all Nogo isoforms [33]–[35] (but see Cafferty et al. [36]). In contrast, inhibition of nerve growth by myelin inhibitors is significantly reduced in p75-deficient CGNs [7],[8]. These results raise the possibility that a yet to be identified receptor mediates myelin inhibitor activity through p75. Dimerization of p75DD in Solution To understand the mechanism by which the p75–p45 interaction regulates p75-dependent signaling, we first characterized biophysically the ICD domain of p75. Because the DD of p75 is a protein–protein interaction motif and often is involved in functionally essential homo- and hetero-associations [37]–[40], we studied the oligomerization behavior of p75ICD comprising residues 290–418. We used gel filtration chromatography of purified p75ICD to analyze the oligomerization state of p75ICD in phosphate buffer at pH 8.0. Purification of p75ICD from bacteria yielded two peaks in the elution profile of gel filtration (Figure 3A, black lines) with the presence of some high molecular weight aggregates (Figure 3A, asterisk). After running SDS-PAGE of the peak fractions in reducing and nonreducing conditions, we found that the elution fraction I corresponds to a covalently disulfide bond dimer and fraction II to a monomer molecular weight (Figure 3B). When we purified p75ICD with the presence of DTT, a single peak was observed that eluted between the monomer and dimer peaks (Figure 3A). The same behavior is observed in the presence of iodoacetamide, which blocks free cysteines, in the lysis buffer (Figure S4). The estimated molecular weight of this fraction from the gel filtration yields a mass of 35 kDa (p75ICD MW is 16 kDa), suggesting the presence of a dimer. We carried out an analytical ultracentrifugation analysis of purified p75ICD in the presence of DTT, using equilibrium sedimentation and velocity experiments of recombinant p75ICD in the same buffer (Figure 3C). Our ultracentrifugation confirms that p75ICD behaves in solution as a single species with a molecular weight of 30.7 kDa, close to the theoretical dimer of p75ICD (see Figure 3C legend). Altogether, we conclude that p75ICD is a noncovalent dimer that during purification or in oxidative conditions dimerizes through a disulfide bond (Figure S5). Recently the crystal structure of a covalent disulfide p75ICD dimer, purified from bacteria in oxidative conditions (in the presence of DTNB), has been described [41]. The dimer is mediated by a covalent disulfide bond through Cys379. We think that the dimer found in this work is the same dimer, because when we made p75-C379A mutant, gel filtration gives a monomer peak (Figure S6). Whether this covalent disulfide dimer is formed in the reducing conditions encountered inside cells and what its biological function is need to be further investigated. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 3. p75ICD homodimerization. (A) Size exclusion gel-filtration chromatography of p75ICD. The elution profile reveals the presence of a mixture of monomers and dimers in the case of p75ICD (black lines). In the presence of DTT, only one elution peak in the gel filtration chromatogram is seen (gray line). Molecular weight standards are shown above the chromatogram. (B) Coomassie blue staining of reducing and nonreducing SDS-PAGE of the fractions collected in gel filtration as shown in (A). The presence of a protein band corresponding to a p75ICD dimer in the nonreducing SDS-PAGE is shown with an arrowhead. The migration of p45ICD is shown as a reference. (C) Analytical ultracentrifugation data on p75ICD in PBS (pH 8.0). (Top) Overlay of successive sedimentation velocity profiles recorded at ∼10 min intervals, represented by different colors. The solid lines represent the direct fitting of the data to a two-species model by the Svedberg program. (Bottom) Sedimentation velocity AUC profiles and the c(S) distributions for p75ICD (at 0.1, 0.3, and 1.0 mg ml−1). The residual differences between the experimental data and the fit for each point are shown above. Theoretical p75ICD MW  = 16.5 kDa. Fitting data MW  = 30.7±1.2 kDa. https://doi.org/10.1371/journal.pbio.1001918.g003 p75ICD Dimer Interface Mapping by NMR NMR titration experiments were performed at different p75ICD concentrations in the presence of DTT to avoid formation of the disulfide dimer. Concentration-dependent chemical shift changes in [42] TROSY spectra [43] of 15N-labeled p75ICD were then used to map the homodimer interface (Figure 4A). Figure 4B shows several examples of concentration-induced chemical shift changes. The appearance of only one peak suggests a slow monomer–dimer equilibrium for NMR time scale measurements. This pattern of change is indicative of a weak binding that we can estimate from plotting changes in the chemical shift of interface residues. We obtained a Kd of ∼100 µM (Figure 4C). Such low binding affinities are typically observed for DD-type interactions [44]. Residues with the most pronounced chemical shift changes—that is, L360, E363, Q367, H370, D372, F374, T375, C379, H376, E377, A383, L384, L385, and W388, (Figure 4A)—were then mapped onto the reported NMR structure of p75ICD (Figure 4D) [45]. These residues are all located on one side of the DD, in particular within helices α3 and α4. Together with the presence of only one TROSY cross-peak per 15N-1H moiety, these results suggest the formation of a symmetrical p75-DD homodimer. Because helices α3 and α4 contain charged amino acid residues, the homodimer formation may be in part due to electrostatic interactions. Recently the crystal structure of an asymmetrical dimer of p75-DD has been described [41]. In that structure, R404 from monomer A interacts with S373, H376, and E377 of monomer B. In our NMR titration study, chemical shift changes of R404 are not observed (Figure 4A), although changes in the chemical shift of S373 (small) and E377 and H376 (big) are clearly visible (Figure 4A). These results suggest that in solution the symmetrical dimer is favored, although we cannot exclude the possibility that p75ICD with different conformations is present in solution. Sometimes crystallization favors conformations that are better packed but are not necessarily the prevalent conformations in solution. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 4. Determination of p75ICD dimer interface by NMR. (A) Concentration-dependent chemical shift changes of 15N-labeled p75ICD observed in [42]-TROSY spectra at p75ICD concentrations of 10 µM, 100 µM, and 500 µM. (B) Some examples of residue peaks of p75ICD at the different concentrations of 10 µM (purple), 100 µM (green), and 500 µM (red). The cross-peaks are labeled to the corresponding residues. (C) Determination of monomer-dimer kd using the changes in the chemical shift of different residues from the dimer interface plotted versus p75ICD concentration. (D) Concentration-dependent chemical shift changes from data in (B) are mapped onto the 3D structure of p75DD (PDB code 1NGR; [45]). The 3D structure is represented by a ribbon diagram and by a surface representation. Residues with chemical shift perturbations ΔCS, larger than 0.2 ppm, are displayed and colored in green. ΔCS  = 25[Δ(δ(1H))2 + Δ(δ(15N))2]0.5, where δ(1H) and δ(15N) are the chemical shifts in part per million (ppm) along the ω2(1H) and ω1(15N) dimensions, respectively. (E) Immunoprecipitation experiments of wild-type or some p75 mutants in the dimer interface demonstrating the role of those residues in p75 homodimerization. https://doi.org/10.1371/journal.pbio.1001918.g004 To further characterize and confirm the homodimer interface in the full-length p75, we made p75 mutant constructs containing amino acid replacements at different residues, which showed high concentration-dependent chemical shift changes in the NMR studies of p75ICD (Figure 4A). The emphasis of the amino acid replacements was charge changes due to the potential electrostatic nature of the interaction (D372R, H376E, and E377R; Figure 4E). Wild-type or mutant p75 constructs were co-transfected with a Flag-tagged construct that contained only the TM domain and the ICD of p75 (Flag-ΔECD-p75) in HEK293 cells. The presence of the p75 dimers was measured by co-immunoprecipitation with an anti-Flag antibody and detection of full-length p75 with an anti-p75 antibody. As shown in Figure 3G, wild-type p75 forms a dimer with Flag-ΔECD-p75. In comparison to wild-type p75, the p75 mutant E377R shows a significant decrease in dimer formation, suggesting the importance of this residue and the negative charge in the homodimer interface. In contrast, the mutant H376E has a stronger binding than wild-type p75 (Figure 4E). The role of H376E mutation in dimer formation suggests that the dimer formation could be dependent on the ionization of H376 and then on the pH of the solution. The fact the mutation H376E favors the dimerization suggests that an electrostatic interaction plays a role in homodimerization. NgR Interaction Prefers the Presence of p75 Dimers Stabilized by DD and Cys257 Very little is known about how p75 and NgR interact from a structural point of view. To shed light on this and to understand how p45 modulates p75/NgR signaling, we first investigated how p75 and NgR interact with each other. We used the following p75 constructs: (1) p75 dimerization mutants E377R, D372R, and H376E (Figure 4), which exhibit a significant reduction or increase of p75 homodimer formation, and (2) p75-C257A, a mutant in the TM domain of p75, which although able to form dimers, is not functional upon NGF binding [30]. When co-transfected, mutants E377R and D372R showed less interaction with NgR (Figure 5A). However, mutant H376E, which promotes p75 homodimer formation, displayed a significant increase in its capability to bind NgR (Figure 5A and Table S1). We then determined whether C257 plays a role in the interaction with NgR. When co-transfected with NgR, the p75-C257A/NgR interaction was markedly impaired, although some interaction was observed (Figure 5B and Table S1). When we performed the co-immunoprecipitation of NgR with p75-wt and ran a nonreducing SDS-PAGE, we found the majority of p75-wts that were co-immunoprecipitated with NgR were in the form of dimers, whereas NgR co-immunoprecipitated a small and similar amount of p75-wt and p75-C257A in the form of monomers (Figure 5C). This suggests that NgR and p75 form a complex that is better stabilized with p75-wt than with p75-C257A. In addition, because p75-C257A is still able to form dimers as inferred by crosslinking [30],[46], these results suggest that p75-wt dimers mediated by C257 have a preferred conformation for binding to NgR. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 5. p75 stabilized dimers through both DD and TM domain are required for NgR interaction. (A) Immunoprecipitation experiments of wild-type or mutant p75 with NgR that is Flag-tagged in HEK293 cells. The same mutants described in Figure 4 that are not co-immunoprecipitated with Flag-ΔECD-p75 are not able to be co-immunoprecipitated with Flag-tagged NgR either, suggesting a role of DD dimer stabilization in NgR binding. WT, 100%; D372R, 61.8%±3.64%, N = 5; *** p<0.0001; H376E, 125.4%±9.13%, N = 5; * p<0.1; E377R, 27.4%±4.79%, N = 5; *** p<0.0001. The data can be found in Table S1. (B) Immunoprecipitation experiments of wild-type or p75-C257A with NgR-Flag in HEK293 cells. The p75-C257A interaction with NgR is impaired in comparison to p75 wild type. WT,100%, C257A, 17.67%±4.67%, N = 3; *** p<0.0001. The data can be found in Table S1. (C) Immunoprecipitation experiments of wild-type or p75-C257A with NgR-Flag in HEK293 cells and nonreducing SDS-PAGE followed by Western blot. The NgR interaction to p75 dimers is preferred to p75 monomers. The presence of dimers (d) and monomers (m) is labeled in the blot. https://doi.org/10.1371/journal.pbio.1001918.g005 From these data we conclude that it is not the mere dimerization of p75 but the conformation stabilized by both disulfide bond and DD electrostatic interactions that is preferred for NgR interaction. NMR Solution Structure of p45ICD Our data (Figure S2) and previous published data from other authors have mapped the interaction between p45 and p75 to the TM and the ICD [27]. To map the binding interface of p45 and p75 ICDs, first we solved the three-dimensional solution structure of mouse p45ICD by NMR spectroscopy (Figure 6). The NMR studies showed that p45ICD contains a flexible domain at the N terminus (residues 75–140) that could not be assigned because they display limited chemical shift dispersions, and a folded domain at the C terminus (141–218) (Figure S7). Figure 6A presents the three-dimensional NMR structure of the folded DD domain obtained from the experimental restraints (Table S2). The regions with a secondary structure are the best defined and typical for a DD, and p45DD is composed of six α-helix disposed in a specific orientation (Figure 6A) comprising residues 141–147 (α1), 154–167 (α2), 169–180 (α3), 182–190 (α4), 200–207 (α5), and 212–218 (α6). A DALI [47] search revealed p75DD as the closest structural relative with an rmsd of 2.7 Å, followed by other members of the DD family (Table S3). The DD of p75 and p45 share many structural features and the same arrangement of all the six α-helices, which is not surprising as p75DD and p45DD are homologues. Only the length of the loop between α1 and α2 is longer in p45DD because of an insertion of four amino acid residues in this segment. When compared with p75DD, the longer loop reorients α1 in respect to α2 and α3 and brings residue E153 of the loop in close neighborhood to other negative charged residues (Figure 6B). Together with some additional amino acid residue differences, this small structural reorientation changes significantly the charge distribution around helix α3 of p45DD (Figure 5B). The negative charged region of p45DD is formed by E153, E160, E170, E173, and D178. In the equivalent region of p75DD (Figure 6B), the negative charged E363, E369, D372, and E377 are located in a more balanced environment surrounded by positive charged residues R358, H370, and H376, which are positively charged depending of their specific pka. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 6. Three-dimensional NMR structure of p45DD. (A) Superposition of 20 conformers representing the 3D NMR structure (left) and ribbon diagram of the lowest energy conformer highlighting the α-helices in red and yellow (right). (Bottom) 13Cα chemical shift deviation from their corresponding "random coil" values Δδ(13Cα) of p45ICD (residues 130–228). Segments of positive deviations are indicative of helical secondary structure. The location of the six α-helices of p45 are represented by cylinders and labeled accordingly. The asterisk indicates the unusual chemical shift of R211 attributed to the salt bridge between R211 and D213 as well as E160. (B) Electrostatic potential of p45DD and p75DD in a surface representation indicates a highly negative patch around helix α3 of p45DD. The same orientation as in (A) is used. (C) Sequence alignment of DEDs of mouse PEA-15 (Q62048), human FADD-DED (Q13158), human Caspase-8 (Q14790), molluscum contagiosum virus MCV-159 (Q98325), and death domain from rat p75DD (NP_036742) and mouse p45DD (NP_080288). The positions of helices are indicated by the diagram below the p45DD sequence. The conserved motif RxDΦ at the beginning of helix 6 and the conserved residues (EL) in helix 2 are indicated in bold. https://doi.org/10.1371/journal.pbio.1001918.g006 Another interesting feature of the sequence of p45DD is the presence of the RxDΦ motif (x, any residues; Φ, a hydrophobic residue) at the beginning of helix α6 (Figure 5C), which is typically observed in death effector domains (DEDs), such as the DED from PEA-15, FADD, Caspase-8, and others [48],[49]. In DEDs, this motif has been suggested to stabilize the DED fold, because it participates in a salt-bridged network between the arginine side chain and the aspartic acid side chain of the RxDΦ motif, and a glutamic acid side chain located in the helix α2 (for instance, R72, D74, and E19 in FADD-DED) (Figure 6C) [48]. Such a charged network is also present in the three-dimensional structure of p45DD between residues R211, D213, and E160, as indicated by the large downfield shift of the Hε for R211 of p45, indicative of a charged interaction (Figure 6A, the asterisk indicates the downshift of R211). A similar shift has been observed for R72 of FADD-DED [48]. The presence of this salt bridge is an unexpected feature of p45DD, because this motif is not found in any other DD. p75DD has a similar RxDΦ sequence, RADI (highlighted in black in Figure 6C), and from this argument, Park et al. [50] has suggested that p75DD is a DED, not a DD. However, in our hands, we did not see a large shift of the arginine of p75 by NMR, like in p45 and in PEA, suggesting that maybe this arginine is not forming a salt bridge. In fact, in the amino acid sequence of p75, the analogue residue for E160 involved in the salt bridge in p45 is a His residue (H217) (Figure 6C). Nevertheless, the possibility that p45 and p75 DDs are actually DEDs or a chameleon between DD and DED should be considered in future research. Mapping the p75DD/p45DD Interface Interaction by NMR We then started the characterization of the p75ICD–p45ICD interaction. Using NMR chemical shift perturbation experiments with p45ICD, the binding site of p75ICD on p45ICD was mapped to α2/α3 in the DD of p45 (Figure S8). To further characterize the interaction between p45DD and p75DD in the corresponding full-length proteins, p45 point mutations at some residues with significant chemical shift changes were constructed (i.e., H164E, E173R, C177H, D178A, and D178R). The presence of a full-length p45–p75 complex in correspondingly transfected HEK293 cells was measured by co-immunoprecipitation of p75 with wild-type or mutant p45 (Figure 7A). Wild-type p45 formed a complex with p75. Of the entire series of p45 mutant constructs studied, E173R and D178R showed a significant decrease in complex formation, whereas C177H and D178A showed a significant increase in complex formation (Figure 7A). NMR experiments with purified p45 mutants showed that they were well folded (Figure S9). These data are consistent with the structural studies and suggest that p45 forms a complex with p75 through its helix α3 of the DD. Next, the amino acid residues in p75ICD that interact with p45ICD (Figure S4) were characterized. In the chemical shift perturbation experiment, p45ICD is titrated against 10 µM of 15N-labeled p75DD. Chemical shift changes on p75DD are observed in amino acid residues located close to and on helix α3 of p75DD (i.e., L360, E363, Q367, H370, D372, F374, T375, H376, E377, A378, A383, L384, L385, W388), indicating helix α3 is the binding site for p45DD on p75DD (Figure 7B). The effect on the interaction between p45 and p75 by some of the amino acid replacements located in this side of p75DD (D372R, H376E, and E377R) supports the NMR-derived interface (Figure 7B). A comparison of the p45DD binding site on p75DD (Figure 7A) with the p75DD homodimer binding site (Figure 4D) shows that the two sites overlap. The presence of an overlapping binding site on p75 suggests that p45 is binding to a monomeric p75 by forming a heterodimer. We tried to purify a stable complex between p75ICD and p45ICD by gel filtration, but we were unsuccessful. It could be possible that p45ICD interaction inhibits the formation of p75 dimers or multimers, but the interaction is too weak to yield sufficient p75/p45 heterodimers. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 7. p75/p45 interaction is promoted by DD and TM domain. (A) Co-immunoprecipitation experiments of wild-type or mutant p45 with Myc-p75 in HEK293 cells. p45 mutants show differential binding to p75. The p75DD-dependent chemical shift changes of p45DD are mapped onto the 3D structure of p45DD. (B) Co-immunoprecipitation experiments of wild-type or p75 mutants with Flag-p45 in HEK293 cells showed differential binding to p45. The p45DD-dependent chemical shift changes of p75DD are mapped onto the 3D structure of p45DD. (C) Protein sequences of p75 and p45 TM domains are highly conserved. The conserved cysteine residue is highlighted in a red box. (D) The cysteine residues in the TM domain of both p75 and p45 form a covalent disulfide dimer between p75 and p45. Co-immunoprecipitations of either p75 wild-type or the p75 TM domain mutant (p75-C257A) with p45 wild type were analyzed in HEK293 cells and in reducing and nonreducing SDS-PAGE followed by Western blot. p75 and p45 form a heterodimer sensitive to DTT (arrow). (E) Co-immunoprecipitations of either p75 wild type or the p75 TM domain mutant (p75-C257A) with p45 wild type or p45 C58A mutant were analyzed in HEK293 cells, indicating that both p75-C257A and p45-C58A TM domain mutants diminish the interaction between p75 and p45. https://doi.org/10.1371/journal.pbio.1001918.g007 p45DD Forms a Heterodimer with p75DD by Breaking the p75DD Dimer The presence of a partially overlapping binding site on p75DD for homodimerization with p75DD and heterodimerization with p45DD suggests that p45DD may break a p75DD dimer by competition. To get insights into this potential mechanism, the p45ICD-induced chemical shift perturbations of p75ICD were measured at high (dimeric) and low (monomeric) concentrations of p75ICD. p45ICD binds p75ICD even at high p75ICD concentrations, where p75ICD is mainly homodimeric (Figure S6). As demonstrated above, A378 is participating in the homodimer interface of p75DD because chemical shift differences between low and high concentrated p75ICD were observed (Figure S10). However, the addition of p45ICD at low concentrations of p75ICD did not result in any chemical shift changes of A378, indicating that A378 is not part of the p45–p75 binding site. In contrast, the addition of p45ICD at a high (dimeric) concentration of p75ICD generated chemical shift perturbations of the 15N-1H moiety of A378 to values identical to p75ICD at a low (monomeric) concentration in absence of p45ICD (Figure S10). Our explanation of these findings is as follows. Although A378 is participating in p75DD homodimer formation, it is not involved in p45DD binding. However, the presence of p45 breaks the p75 dimer and forms a p45–p75 heterodimer and thus shifts A378 from a dimeric environment to a monomeric environment (Figure S10). Complementary information about the breaking of a p75 homodimer into a p45–p75 heterodimer can be extracted by the p75ICD- and p45ICD-dependent chemical shift perturbations of T375. T375 has different chemical shifts at high and low concentrations of p75ICD, indicative of its participation in the p75DD homodimer formation. However, upon the addition of p45DD, the chemical shifts of T375 move to a new position that is independent of the p75ICD concentration (at least at the concentration window studied here; Figure S10). This result suggests that T375 is involved in both p45DD and p75DD binding. Furthermore, the data indicate that p45 is able to compete with the p75DD homodimer by the formation of a p45DD–p75DD heterodimer. Other residues of p75 showed similar behavior, suggesting that p45DD is able to break the relatively weak p75DD homodimer by forming a heterodimer p45DD–p75DD. p75 and p45 Heterodimerize Through Their TM Domain by p75–C257/p45–C58 Interaction The results from above suggest that the p45 modulation of p75 signaling is increased with the presence of the TM domain of p45. The TM domain of p75 self-associates [30]. Because the TM domain of p45 is highly homologous to p75-TM (Figure 7C), we asked if p45-TM was able to bind p75 through its TM. Co-immunoprecipitation experiments were conducted in HEK293 cells transfected with p75 wt and p45, and analysis in nonreducing SDS-PAGE showed a band recognized by both p75 and p45 antibodies and with a molecular weight corresponding to a heterodimer p75–p45 (Figure 7D). This band was lost when p75-C257A was co-transfected with p45. These results suggest that p75 and p45 can form a heterodimer through C257 from p75 and another cysteine residue from p45. Due to the homology in the TM domain, we mutated the cysteine equivalent in p45, Cys58 (Figure 7C). As shown in Figure 7E, p45-C58A did not form a complex with wt p75. Only when the blot was exposed longer did we see an interaction between p45 and p75, presumably by interaction through their intracellular DDs. p45 Forms a Stable Complex with p75 and Is Up-Regulated upon Nerve Injury p45 contains a DD (Figure 1A). Because the function of DDs is to bind the DD of other members of the DD superfamily in order to transduce signals [28], we investigated whether p45 can interact with other members of the DD superfamily, such as p75, FADD, TNFR1, and Fas. As shown in Figure 1B, p45 can be co-immunoprecipitated with p75 or FADD when co-expressed in 293 cells, but not with caspase-8, Fas, or TNFR1. Recently we characterized the interaction of p45 with FADD and its role upon spinal cord injury [29]. In the present study, we focus on the interaction of p45 with p75. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 1. Interactions between p75 and p45. (A) A schematic diagram showing domains of p45 in comparison with p75. TM, transmembrane domain; DD, putative death domain; PDZ, putative PDZ binding domain. The degrees of identity and homology in amino acid residues between mouse p75 and p45 are shown in percentages. (B) P45 forms a complex with FADD and p75. V5-tagged TNFR1, human Fas, mouse Fas, p75, FADD, or Caspase-8 was transfected into CrmA/Flag-p45/293 stable cells. The lysates were immunoprecipitated with anti-Flag antibodies and immunoblotted with anti-V5 antibodies. (C) P7 cerebellum extracts were immunprecipitated with anti-p75 ECD antibody (9651) or anti-p45 ECD antibody (6750) followed by immnuoblotting with anti-p75 antibody (Buster). T, total lysate. (D) Western blotting analysis of p45 and p75 expression in the brain, spinal cord, and DRG of P0 and adult mice. (E) Increased expression of p45 and p75 in the spinal cord and sciatic nerve following sciatic nerve injury. (Top) Spinal cord sections were immunostained with anti-p45 or -p75 antibodies. p45 and p75 immunoreactivities were markedly increased in the ipsilateral side as compared to the contralateral side. Higher magnification indicated that expression of p45 and p75 is increased in motor neurons. (F) Longitudinal sections of crushed and uncrushed sciatic nerves were immunostained with antibodies against p45, p75, or neurofilament. The levels of p45 and p75 were markedly increased in the distal (D) portion of sciatic nerves as compared to the proximal (P) end and the uncrushed (UC) nerve. https://doi.org/10.1371/journal.pbio.1001918.g001 p45 and p75 share a high degree of amino acid similarity in their TM domain (94%), including conserved cysteine residues [30], and the ICD (50%) (Figure 1A). However, the ECD of p45 is short and has no binding sites for neurotrophins (Figure 1A). p45 and p75 are expressed in both the peripheral nervous system (PNS) and central nervous system (CNS) during development [26]. We found that p75 and p45 form an immunocomplex in cerebellum extracts (Figure 1C). The expression level of p45 is high in embryonic but is significantly reduced in adult tissues (Figure 1D). However, p45 is up-regulated after sciatic nerve injury in the spinal cord and sciatic nerve in a similar fashion as p75 (Figure 1E,F). Because p45 shares a similar protein sequence with p75 and has similar up-regulated expression patterns with p75 after injury, we decided to investigate whether p45 regulates signaling mechanisms involving p75. p45 Interferes with p75/NgR Signaling p75 mediates nerve growth inhibition by myelin-associated inhibitors via functioning in part as a co-receptor for GPI-linked neuronal NgR [6]. Because p45 and p75 form a stable complex (Figure 1B), we investigated whether p45 interferes with or enhances the formation of the complex between p75 and NgR and subsequent signaling through the p75/NgR complex. When HEK293 cells were co-transfected with p75 and Flag-tagged human NgR (Flag-hNgR) expression constructs followed by immunoprecipitation with anti-p75ICD antibodies, we found that p45 markedly reduced the levels of complex formation between p75 and NgR (Figure 2A) and it does in a concentration-dependent manner (Figure 2B). As a control, p45 is not able to bind to NgR directly (Figure S1). To further characterize the domains of p45 involved in such inhibition, we made several deletion constructs of p45 (Figure S2 and Table S1). As shown in Figure S2, both p45 intracellular and TM domains are necessary for p45 inhibition of p75/NgR complex formation. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 2. p45 interferes with the p75-NgR interaction and signaling. (A) Lysates from HEK293 cells co-transfected with vectors expressing p45, p75, and Flag-tagged human NogoR (Flag-hNgR) were immunoprecipitated with anti-p75 antibodies and probed with anti-Flag antibodies. The results showed that p75 and hNgR form a complex and presence of p45 reduces the formation of p75/hNgR complex. (B) Co-transfection with varying amount of p45-expressing vectors markedly reduced p75/hNgR complex formation in a concentration-dependent fashion. (C) Quantification of the increase of RhoA activity after MAG-Fc treatment. In wild-type (WT) CGN cultures, when treated with MAG-Fc, the RhoA activity increased by 50%. However, in CGN cultures derived from Thy1-p45 transgenic mice, the increase in RhoA activity induced by MAG-Fc treatment is completely abolished. The value is expressed as percent over control. The value is derived from three independent experiments. * p<0.05, Student's t test. The data can be found in Table S1. (D) CGNs were seeded on glass coverslips coated with inhibitory substrates, grown for 14–18 h, and immunofluorescently stained with Tuj1 (Red) and anti-p45 (Green) antibodies. Scale bar, 50 µm. (E) Quantitative analysis of neurite length from the outgrowth assay, using Nogo66-GST, myelin, or HEK293 cells expressing MAG as inhibitory substrates. The data are represented as mean ± SEM (* p<0.001) and can be found in Table S1. Overexpression of p45 promoted neurite outgrowth of CGNs on inhibitory substrates. https://doi.org/10.1371/journal.pbio.1001918.g002 We examined the possibility that p45 would antagonize signaling through the p75/NgR complex. Previous results demonstrated that p75 is required for MAG-induced RhoA activation through NgR [20],[31]. RhoA activation is necessary for neurite outgrowth inhibition mediated by myelin-associated inhibitors. Thus, we examined whether overexpression of p45 blunts signaling through the p75/NgR complex. Postnatal day 7 (P7) cerebellar granule neurons (CGNs) that express low levels of endogenous p45 were transfected with full-length, capped p45 RNAs containing poly(A) tails generated by an in vitro transcription system. As shown in Figure S3A, the level of p45 protein was markedly elevated 24 h following RNA transfection. The cultures were then serum-starved and treated with Fc or MAG-Fc proteins. The level of activated RhoA was measured. As illustrated in Figure S3B, overexpression of p45 blocks MAG-Fc–induced RhoA activation. We also quantitatively measured RhoA activation using the G-LISA kit (Cytoskeleton Inc.) on CGN cultures from wild-type mice and Thy1-p45 transgenic mice that consistently overexpress p45 under a Thy1 promoter [29]. As shown in Figure 2C, MAG-Fc treatment of the WT CGNs induced 50% increase in the RhoA activity, whereas the MAG-Fc–induced RhoA activation is completely abolished in Thy1-p45 CGNs (Table S1). These results suggest that p45 is capable of effectively blocking RhoA activation through the p75/NgR complex. We then examined whether overexpression of p45 prevents neurite outgrowth inhibition induced by Nogo66, MAG, or CNS myelin. P7 CGNs were transfected with p45 RNAs, plated onto dishes previously coated with different substrates, and allowed to grow overnight. The cultures were double immunostained with antibodies against p45 (green) and neurotubulin (TuJ1, red) (Figure 2D). The neurite length of control CGNs and transfected CGNs that display increased p45-immunoreactivity over control CGNs was measured. As shown in Figure 2E, neurite outgrowth inhibition elicited by Nogo66 is alleviated by p45 overexpression (Table S1). Similarly, p45 overexpression significantly promotes neurite outgrowth that was otherwise inhibited when cultured on dishes coated with CNS myelin or MAG-expressing cells (Figure 2E). These results support the idea that p45 promotes neurite outgrowth. It is worth noting that despite NgR has been implicated in mediating nerve growth inhibition induced by myelin inhibitors in culture, neurite outgrowth of CGNs from NgR null mutants is still inhibited by myelin inhibitors [7],[8]. Recent results suggest that NgR is required only for the acute growth cone-collapsing but not chronic growth-inhibitory actions of myelin inhibitors [32]. Furthermore, no measurable corticospinal tract regeneration was observed in mice lacking all Nogo isoforms [33]–[35] (but see Cafferty et al. [36]). In contrast, inhibition of nerve growth by myelin inhibitors is significantly reduced in p75-deficient CGNs [7],[8]. These results raise the possibility that a yet to be identified receptor mediates myelin inhibitor activity through p75. Dimerization of p75DD in Solution To understand the mechanism by which the p75–p45 interaction regulates p75-dependent signaling, we first characterized biophysically the ICD domain of p75. Because the DD of p75 is a protein–protein interaction motif and often is involved in functionally essential homo- and hetero-associations [37]–[40], we studied the oligomerization behavior of p75ICD comprising residues 290–418. We used gel filtration chromatography of purified p75ICD to analyze the oligomerization state of p75ICD in phosphate buffer at pH 8.0. Purification of p75ICD from bacteria yielded two peaks in the elution profile of gel filtration (Figure 3A, black lines) with the presence of some high molecular weight aggregates (Figure 3A, asterisk). After running SDS-PAGE of the peak fractions in reducing and nonreducing conditions, we found that the elution fraction I corresponds to a covalently disulfide bond dimer and fraction II to a monomer molecular weight (Figure 3B). When we purified p75ICD with the presence of DTT, a single peak was observed that eluted between the monomer and dimer peaks (Figure 3A). The same behavior is observed in the presence of iodoacetamide, which blocks free cysteines, in the lysis buffer (Figure S4). The estimated molecular weight of this fraction from the gel filtration yields a mass of 35 kDa (p75ICD MW is 16 kDa), suggesting the presence of a dimer. We carried out an analytical ultracentrifugation analysis of purified p75ICD in the presence of DTT, using equilibrium sedimentation and velocity experiments of recombinant p75ICD in the same buffer (Figure 3C). Our ultracentrifugation confirms that p75ICD behaves in solution as a single species with a molecular weight of 30.7 kDa, close to the theoretical dimer of p75ICD (see Figure 3C legend). Altogether, we conclude that p75ICD is a noncovalent dimer that during purification or in oxidative conditions dimerizes through a disulfide bond (Figure S5). Recently the crystal structure of a covalent disulfide p75ICD dimer, purified from bacteria in oxidative conditions (in the presence of DTNB), has been described [41]. The dimer is mediated by a covalent disulfide bond through Cys379. We think that the dimer found in this work is the same dimer, because when we made p75-C379A mutant, gel filtration gives a monomer peak (Figure S6). Whether this covalent disulfide dimer is formed in the reducing conditions encountered inside cells and what its biological function is need to be further investigated. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 3. p75ICD homodimerization. (A) Size exclusion gel-filtration chromatography of p75ICD. The elution profile reveals the presence of a mixture of monomers and dimers in the case of p75ICD (black lines). In the presence of DTT, only one elution peak in the gel filtration chromatogram is seen (gray line). Molecular weight standards are shown above the chromatogram. (B) Coomassie blue staining of reducing and nonreducing SDS-PAGE of the fractions collected in gel filtration as shown in (A). The presence of a protein band corresponding to a p75ICD dimer in the nonreducing SDS-PAGE is shown with an arrowhead. The migration of p45ICD is shown as a reference. (C) Analytical ultracentrifugation data on p75ICD in PBS (pH 8.0). (Top) Overlay of successive sedimentation velocity profiles recorded at ∼10 min intervals, represented by different colors. The solid lines represent the direct fitting of the data to a two-species model by the Svedberg program. (Bottom) Sedimentation velocity AUC profiles and the c(S) distributions for p75ICD (at 0.1, 0.3, and 1.0 mg ml−1). The residual differences between the experimental data and the fit for each point are shown above. Theoretical p75ICD MW  = 16.5 kDa. Fitting data MW  = 30.7±1.2 kDa. https://doi.org/10.1371/journal.pbio.1001918.g003 p75ICD Dimer Interface Mapping by NMR NMR titration experiments were performed at different p75ICD concentrations in the presence of DTT to avoid formation of the disulfide dimer. Concentration-dependent chemical shift changes in [42] TROSY spectra [43] of 15N-labeled p75ICD were then used to map the homodimer interface (Figure 4A). Figure 4B shows several examples of concentration-induced chemical shift changes. The appearance of only one peak suggests a slow monomer–dimer equilibrium for NMR time scale measurements. This pattern of change is indicative of a weak binding that we can estimate from plotting changes in the chemical shift of interface residues. We obtained a Kd of ∼100 µM (Figure 4C). Such low binding affinities are typically observed for DD-type interactions [44]. Residues with the most pronounced chemical shift changes—that is, L360, E363, Q367, H370, D372, F374, T375, C379, H376, E377, A383, L384, L385, and W388, (Figure 4A)—were then mapped onto the reported NMR structure of p75ICD (Figure 4D) [45]. These residues are all located on one side of the DD, in particular within helices α3 and α4. Together with the presence of only one TROSY cross-peak per 15N-1H moiety, these results suggest the formation of a symmetrical p75-DD homodimer. Because helices α3 and α4 contain charged amino acid residues, the homodimer formation may be in part due to electrostatic interactions. Recently the crystal structure of an asymmetrical dimer of p75-DD has been described [41]. In that structure, R404 from monomer A interacts with S373, H376, and E377 of monomer B. In our NMR titration study, chemical shift changes of R404 are not observed (Figure 4A), although changes in the chemical shift of S373 (small) and E377 and H376 (big) are clearly visible (Figure 4A). These results suggest that in solution the symmetrical dimer is favored, although we cannot exclude the possibility that p75ICD with different conformations is present in solution. Sometimes crystallization favors conformations that are better packed but are not necessarily the prevalent conformations in solution. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 4. Determination of p75ICD dimer interface by NMR. (A) Concentration-dependent chemical shift changes of 15N-labeled p75ICD observed in [42]-TROSY spectra at p75ICD concentrations of 10 µM, 100 µM, and 500 µM. (B) Some examples of residue peaks of p75ICD at the different concentrations of 10 µM (purple), 100 µM (green), and 500 µM (red). The cross-peaks are labeled to the corresponding residues. (C) Determination of monomer-dimer kd using the changes in the chemical shift of different residues from the dimer interface plotted versus p75ICD concentration. (D) Concentration-dependent chemical shift changes from data in (B) are mapped onto the 3D structure of p75DD (PDB code 1NGR; [45]). The 3D structure is represented by a ribbon diagram and by a surface representation. Residues with chemical shift perturbations ΔCS, larger than 0.2 ppm, are displayed and colored in green. ΔCS  = 25[Δ(δ(1H))2 + Δ(δ(15N))2]0.5, where δ(1H) and δ(15N) are the chemical shifts in part per million (ppm) along the ω2(1H) and ω1(15N) dimensions, respectively. (E) Immunoprecipitation experiments of wild-type or some p75 mutants in the dimer interface demonstrating the role of those residues in p75 homodimerization. https://doi.org/10.1371/journal.pbio.1001918.g004 To further characterize and confirm the homodimer interface in the full-length p75, we made p75 mutant constructs containing amino acid replacements at different residues, which showed high concentration-dependent chemical shift changes in the NMR studies of p75ICD (Figure 4A). The emphasis of the amino acid replacements was charge changes due to the potential electrostatic nature of the interaction (D372R, H376E, and E377R; Figure 4E). Wild-type or mutant p75 constructs were co-transfected with a Flag-tagged construct that contained only the TM domain and the ICD of p75 (Flag-ΔECD-p75) in HEK293 cells. The presence of the p75 dimers was measured by co-immunoprecipitation with an anti-Flag antibody and detection of full-length p75 with an anti-p75 antibody. As shown in Figure 3G, wild-type p75 forms a dimer with Flag-ΔECD-p75. In comparison to wild-type p75, the p75 mutant E377R shows a significant decrease in dimer formation, suggesting the importance of this residue and the negative charge in the homodimer interface. In contrast, the mutant H376E has a stronger binding than wild-type p75 (Figure 4E). The role of H376E mutation in dimer formation suggests that the dimer formation could be dependent on the ionization of H376 and then on the pH of the solution. The fact the mutation H376E favors the dimerization suggests that an electrostatic interaction plays a role in homodimerization. NgR Interaction Prefers the Presence of p75 Dimers Stabilized by DD and Cys257 Very little is known about how p75 and NgR interact from a structural point of view. To shed light on this and to understand how p45 modulates p75/NgR signaling, we first investigated how p75 and NgR interact with each other. We used the following p75 constructs: (1) p75 dimerization mutants E377R, D372R, and H376E (Figure 4), which exhibit a significant reduction or increase of p75 homodimer formation, and (2) p75-C257A, a mutant in the TM domain of p75, which although able to form dimers, is not functional upon NGF binding [30]. When co-transfected, mutants E377R and D372R showed less interaction with NgR (Figure 5A). However, mutant H376E, which promotes p75 homodimer formation, displayed a significant increase in its capability to bind NgR (Figure 5A and Table S1). We then determined whether C257 plays a role in the interaction with NgR. When co-transfected with NgR, the p75-C257A/NgR interaction was markedly impaired, although some interaction was observed (Figure 5B and Table S1). When we performed the co-immunoprecipitation of NgR with p75-wt and ran a nonreducing SDS-PAGE, we found the majority of p75-wts that were co-immunoprecipitated with NgR were in the form of dimers, whereas NgR co-immunoprecipitated a small and similar amount of p75-wt and p75-C257A in the form of monomers (Figure 5C). This suggests that NgR and p75 form a complex that is better stabilized with p75-wt than with p75-C257A. In addition, because p75-C257A is still able to form dimers as inferred by crosslinking [30],[46], these results suggest that p75-wt dimers mediated by C257 have a preferred conformation for binding to NgR. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 5. p75 stabilized dimers through both DD and TM domain are required for NgR interaction. (A) Immunoprecipitation experiments of wild-type or mutant p75 with NgR that is Flag-tagged in HEK293 cells. The same mutants described in Figure 4 that are not co-immunoprecipitated with Flag-ΔECD-p75 are not able to be co-immunoprecipitated with Flag-tagged NgR either, suggesting a role of DD dimer stabilization in NgR binding. WT, 100%; D372R, 61.8%±3.64%, N = 5; *** p<0.0001; H376E, 125.4%±9.13%, N = 5; * p<0.1; E377R, 27.4%±4.79%, N = 5; *** p<0.0001. The data can be found in Table S1. (B) Immunoprecipitation experiments of wild-type or p75-C257A with NgR-Flag in HEK293 cells. The p75-C257A interaction with NgR is impaired in comparison to p75 wild type. WT,100%, C257A, 17.67%±4.67%, N = 3; *** p<0.0001. The data can be found in Table S1. (C) Immunoprecipitation experiments of wild-type or p75-C257A with NgR-Flag in HEK293 cells and nonreducing SDS-PAGE followed by Western blot. The NgR interaction to p75 dimers is preferred to p75 monomers. The presence of dimers (d) and monomers (m) is labeled in the blot. https://doi.org/10.1371/journal.pbio.1001918.g005 From these data we conclude that it is not the mere dimerization of p75 but the conformation stabilized by both disulfide bond and DD electrostatic interactions that is preferred for NgR interaction. NMR Solution Structure of p45ICD Our data (Figure S2) and previous published data from other authors have mapped the interaction between p45 and p75 to the TM and the ICD [27]. To map the binding interface of p45 and p75 ICDs, first we solved the three-dimensional solution structure of mouse p45ICD by NMR spectroscopy (Figure 6). The NMR studies showed that p45ICD contains a flexible domain at the N terminus (residues 75–140) that could not be assigned because they display limited chemical shift dispersions, and a folded domain at the C terminus (141–218) (Figure S7). Figure 6A presents the three-dimensional NMR structure of the folded DD domain obtained from the experimental restraints (Table S2). The regions with a secondary structure are the best defined and typical for a DD, and p45DD is composed of six α-helix disposed in a specific orientation (Figure 6A) comprising residues 141–147 (α1), 154–167 (α2), 169–180 (α3), 182–190 (α4), 200–207 (α5), and 212–218 (α6). A DALI [47] search revealed p75DD as the closest structural relative with an rmsd of 2.7 Å, followed by other members of the DD family (Table S3). The DD of p75 and p45 share many structural features and the same arrangement of all the six α-helices, which is not surprising as p75DD and p45DD are homologues. Only the length of the loop between α1 and α2 is longer in p45DD because of an insertion of four amino acid residues in this segment. When compared with p75DD, the longer loop reorients α1 in respect to α2 and α3 and brings residue E153 of the loop in close neighborhood to other negative charged residues (Figure 6B). Together with some additional amino acid residue differences, this small structural reorientation changes significantly the charge distribution around helix α3 of p45DD (Figure 5B). The negative charged region of p45DD is formed by E153, E160, E170, E173, and D178. In the equivalent region of p75DD (Figure 6B), the negative charged E363, E369, D372, and E377 are located in a more balanced environment surrounded by positive charged residues R358, H370, and H376, which are positively charged depending of their specific pka. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 6. Three-dimensional NMR structure of p45DD. (A) Superposition of 20 conformers representing the 3D NMR structure (left) and ribbon diagram of the lowest energy conformer highlighting the α-helices in red and yellow (right). (Bottom) 13Cα chemical shift deviation from their corresponding "random coil" values Δδ(13Cα) of p45ICD (residues 130–228). Segments of positive deviations are indicative of helical secondary structure. The location of the six α-helices of p45 are represented by cylinders and labeled accordingly. The asterisk indicates the unusual chemical shift of R211 attributed to the salt bridge between R211 and D213 as well as E160. (B) Electrostatic potential of p45DD and p75DD in a surface representation indicates a highly negative patch around helix α3 of p45DD. The same orientation as in (A) is used. (C) Sequence alignment of DEDs of mouse PEA-15 (Q62048), human FADD-DED (Q13158), human Caspase-8 (Q14790), molluscum contagiosum virus MCV-159 (Q98325), and death domain from rat p75DD (NP_036742) and mouse p45DD (NP_080288). The positions of helices are indicated by the diagram below the p45DD sequence. The conserved motif RxDΦ at the beginning of helix 6 and the conserved residues (EL) in helix 2 are indicated in bold. https://doi.org/10.1371/journal.pbio.1001918.g006 Another interesting feature of the sequence of p45DD is the presence of the RxDΦ motif (x, any residues; Φ, a hydrophobic residue) at the beginning of helix α6 (Figure 5C), which is typically observed in death effector domains (DEDs), such as the DED from PEA-15, FADD, Caspase-8, and others [48],[49]. In DEDs, this motif has been suggested to stabilize the DED fold, because it participates in a salt-bridged network between the arginine side chain and the aspartic acid side chain of the RxDΦ motif, and a glutamic acid side chain located in the helix α2 (for instance, R72, D74, and E19 in FADD-DED) (Figure 6C) [48]. Such a charged network is also present in the three-dimensional structure of p45DD between residues R211, D213, and E160, as indicated by the large downfield shift of the Hε for R211 of p45, indicative of a charged interaction (Figure 6A, the asterisk indicates the downshift of R211). A similar shift has been observed for R72 of FADD-DED [48]. The presence of this salt bridge is an unexpected feature of p45DD, because this motif is not found in any other DD. p75DD has a similar RxDΦ sequence, RADI (highlighted in black in Figure 6C), and from this argument, Park et al. [50] has suggested that p75DD is a DED, not a DD. However, in our hands, we did not see a large shift of the arginine of p75 by NMR, like in p45 and in PEA, suggesting that maybe this arginine is not forming a salt bridge. In fact, in the amino acid sequence of p75, the analogue residue for E160 involved in the salt bridge in p45 is a His residue (H217) (Figure 6C). Nevertheless, the possibility that p45 and p75 DDs are actually DEDs or a chameleon between DD and DED should be considered in future research. Mapping the p75DD/p45DD Interface Interaction by NMR We then started the characterization of the p75ICD–p45ICD interaction. Using NMR chemical shift perturbation experiments with p45ICD, the binding site of p75ICD on p45ICD was mapped to α2/α3 in the DD of p45 (Figure S8). To further characterize the interaction between p45DD and p75DD in the corresponding full-length proteins, p45 point mutations at some residues with significant chemical shift changes were constructed (i.e., H164E, E173R, C177H, D178A, and D178R). The presence of a full-length p45–p75 complex in correspondingly transfected HEK293 cells was measured by co-immunoprecipitation of p75 with wild-type or mutant p45 (Figure 7A). Wild-type p45 formed a complex with p75. Of the entire series of p45 mutant constructs studied, E173R and D178R showed a significant decrease in complex formation, whereas C177H and D178A showed a significant increase in complex formation (Figure 7A). NMR experiments with purified p45 mutants showed that they were well folded (Figure S9). These data are consistent with the structural studies and suggest that p45 forms a complex with p75 through its helix α3 of the DD. Next, the amino acid residues in p75ICD that interact with p45ICD (Figure S4) were characterized. In the chemical shift perturbation experiment, p45ICD is titrated against 10 µM of 15N-labeled p75DD. Chemical shift changes on p75DD are observed in amino acid residues located close to and on helix α3 of p75DD (i.e., L360, E363, Q367, H370, D372, F374, T375, H376, E377, A378, A383, L384, L385, W388), indicating helix α3 is the binding site for p45DD on p75DD (Figure 7B). The effect on the interaction between p45 and p75 by some of the amino acid replacements located in this side of p75DD (D372R, H376E, and E377R) supports the NMR-derived interface (Figure 7B). A comparison of the p45DD binding site on p75DD (Figure 7A) with the p75DD homodimer binding site (Figure 4D) shows that the two sites overlap. The presence of an overlapping binding site on p75 suggests that p45 is binding to a monomeric p75 by forming a heterodimer. We tried to purify a stable complex between p75ICD and p45ICD by gel filtration, but we were unsuccessful. It could be possible that p45ICD interaction inhibits the formation of p75 dimers or multimers, but the interaction is too weak to yield sufficient p75/p45 heterodimers. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 7. p75/p45 interaction is promoted by DD and TM domain. (A) Co-immunoprecipitation experiments of wild-type or mutant p45 with Myc-p75 in HEK293 cells. p45 mutants show differential binding to p75. The p75DD-dependent chemical shift changes of p45DD are mapped onto the 3D structure of p45DD. (B) Co-immunoprecipitation experiments of wild-type or p75 mutants with Flag-p45 in HEK293 cells showed differential binding to p45. The p45DD-dependent chemical shift changes of p75DD are mapped onto the 3D structure of p45DD. (C) Protein sequences of p75 and p45 TM domains are highly conserved. The conserved cysteine residue is highlighted in a red box. (D) The cysteine residues in the TM domain of both p75 and p45 form a covalent disulfide dimer between p75 and p45. Co-immunoprecipitations of either p75 wild-type or the p75 TM domain mutant (p75-C257A) with p45 wild type were analyzed in HEK293 cells and in reducing and nonreducing SDS-PAGE followed by Western blot. p75 and p45 form a heterodimer sensitive to DTT (arrow). (E) Co-immunoprecipitations of either p75 wild type or the p75 TM domain mutant (p75-C257A) with p45 wild type or p45 C58A mutant were analyzed in HEK293 cells, indicating that both p75-C257A and p45-C58A TM domain mutants diminish the interaction between p75 and p45. https://doi.org/10.1371/journal.pbio.1001918.g007 p45DD Forms a Heterodimer with p75DD by Breaking the p75DD Dimer The presence of a partially overlapping binding site on p75DD for homodimerization with p75DD and heterodimerization with p45DD suggests that p45DD may break a p75DD dimer by competition. To get insights into this potential mechanism, the p45ICD-induced chemical shift perturbations of p75ICD were measured at high (dimeric) and low (monomeric) concentrations of p75ICD. p45ICD binds p75ICD even at high p75ICD concentrations, where p75ICD is mainly homodimeric (Figure S6). As demonstrated above, A378 is participating in the homodimer interface of p75DD because chemical shift differences between low and high concentrated p75ICD were observed (Figure S10). However, the addition of p45ICD at low concentrations of p75ICD did not result in any chemical shift changes of A378, indicating that A378 is not part of the p45–p75 binding site. In contrast, the addition of p45ICD at a high (dimeric) concentration of p75ICD generated chemical shift perturbations of the 15N-1H moiety of A378 to values identical to p75ICD at a low (monomeric) concentration in absence of p45ICD (Figure S10). Our explanation of these findings is as follows. Although A378 is participating in p75DD homodimer formation, it is not involved in p45DD binding. However, the presence of p45 breaks the p75 dimer and forms a p45–p75 heterodimer and thus shifts A378 from a dimeric environment to a monomeric environment (Figure S10). Complementary information about the breaking of a p75 homodimer into a p45–p75 heterodimer can be extracted by the p75ICD- and p45ICD-dependent chemical shift perturbations of T375. T375 has different chemical shifts at high and low concentrations of p75ICD, indicative of its participation in the p75DD homodimer formation. However, upon the addition of p45DD, the chemical shifts of T375 move to a new position that is independent of the p75ICD concentration (at least at the concentration window studied here; Figure S10). This result suggests that T375 is involved in both p45DD and p75DD binding. Furthermore, the data indicate that p45 is able to compete with the p75DD homodimer by the formation of a p45DD–p75DD heterodimer. Other residues of p75 showed similar behavior, suggesting that p45DD is able to break the relatively weak p75DD homodimer by forming a heterodimer p45DD–p75DD. p75 and p45 Heterodimerize Through Their TM Domain by p75–C257/p45–C58 Interaction The results from above suggest that the p45 modulation of p75 signaling is increased with the presence of the TM domain of p45. The TM domain of p75 self-associates [30]. Because the TM domain of p45 is highly homologous to p75-TM (Figure 7C), we asked if p45-TM was able to bind p75 through its TM. Co-immunoprecipitation experiments were conducted in HEK293 cells transfected with p75 wt and p45, and analysis in nonreducing SDS-PAGE showed a band recognized by both p75 and p45 antibodies and with a molecular weight corresponding to a heterodimer p75–p45 (Figure 7D). This band was lost when p75-C257A was co-transfected with p45. These results suggest that p75 and p45 can form a heterodimer through C257 from p75 and another cysteine residue from p45. Due to the homology in the TM domain, we mutated the cysteine equivalent in p45, Cys58 (Figure 7C). As shown in Figure 7E, p45-C58A did not form a complex with wt p75. Only when the blot was exposed longer did we see an interaction between p45 and p75, presumably by interaction through their intracellular DDs. Discussion The presented biophysical and biological characterization of p45 and p75 interactions shows that a p45–p75 heterodimer is formed, using both the TM and the ICDs. p45 binding to p75 inhibits RhoA activation and increases neurite outgrowth. In light of these data, the following mechanistic model of p45 action is proposed (Figure 8). NgR binding to p75 recruits intracellular proteins that activate RhoA signaling. In the presence of p45, however, p45–p75 heterodimers are formed, stabilized by the Cys257–Cys58 interaction within the TM domain and enhanced by the interaction of cytoplasmic domains. In this heterodimer conformation, the interaction of NgR with p75 is impaired and this translates into diminished RhoA signaling. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 8. Model illustrating p45 inhibition of p75/NgR signaling. p75 is a constitutive dimer in the membrane, stabilized by the Cys257 at the TM domain, where it can bind to NgR complex and activate RhoA signaling and axonal growth collapse. When p45 binds to p75 through both the TM domain (Cys257–Cys58) and DD interactions, p75 downstream signaling is inhibited. https://doi.org/10.1371/journal.pbio.1001918.g008 p45 interaction with p75 has been described previously [23],[27]. However, a prominent role of the TM domain in that interaction was not suggested. The interaction between p45 and p75 is decreased by a deletion of the p75ICD, but not entirely blocked, suggesting that the TM domain also has a prominent role in the p75/p45 interaction (see figure 3C in [27]). It is also interesting to note that p45 has been suggested previously to interact with TrkA and modulate its activity [25]. In that publication, the authors made several deletion constructs of p45 and suggested that the TM domain of p45 was needed for the p45–TrkA interaction [25],[51]. That observation, together with our data, suggests that the TM domains of neurotrophin receptors are starting to be recognized as very important for their function. One important question that remains to be answered is where and how in the cell the p45/p75 heterodimer could be formed. It is difficult to imagine that p45 could break a covalently formed p75 homodimer, however free monomeric p75 is present in the cell membrane [30]. One possibility is that the heterodimer is formed in the ER during the translation and maturation of both proteins. Kim et al. have shown that p45 is involved in the trafficking of sortilin to the plasma membrane, indicating that the p45/sortilin interaction takes place in the ER membrane [27]. The p75/p45 interaction could also occur in the ER membrane during receptor maturation. This implies that it requires the synthesis of new proteins for p45 activity, and only when new p75 and p45 molecules are synthesized will they have the opportunity to form the heterodimers. Such a situation is plausible because both p75 and p45 are up-regulated upon nerve injury and they are expressed in the same cells (Figure S11). In that situation, at the plasma membrane, it could be possible to have different oligomers of p75—namely, p75 monomers, p75 homodimers, and p75/p45 heterodimers. The function of those species may be different, contributing to the complexity of p75 signaling. We showed that p75ICD exists in a monomer–dimer equilibrium mediated by electrostatic interactions, and we postulate that p75 dimerization is pH-dependent or promoted by the presence of counterions, like phosphate in our buffer. This could reconcile the contradictory results reported by previous studies of p75ICD structure. NMR structural studies detected only the monomeric p75ICD species at a pH in the range of 6–7 and in plain water with no counterions [45], solution conditions in which p75ICD will not be favored to self-associate. In contrast, X-ray structures of p75ICD revealed a dimer [41], but the buffers used for crystallization were at or above pH 7.0 and contained 1.1–1.4 M sodium malonate or sodium citrate, as stabilizing counterions. Thus, we conclude that p75 dimerization is dependent on the pH and the presence of counterions. It is interesting to compare the homo- and hetero-dimeric interactions found here with other protein oligomerizations that involve DDs and DEDs. Although DDs and DEDs were first identified in proteins that mediate programmed cell death, they are now recognized to act as protein interaction domains in a variety of cellular signaling pathways [28]. In the DD subfamily, low sequence homology produces diverse interaction surfaces, enabling binding specificity within a subfamily [52]. Protein–protein interactions of DDs have been thereby thought to be predominantly homotypic among different adaptors, as shown here for p75DD homodimer formation, although some examples of heterotypic interactions have been demonstrated (reviewed in [28]), including the p45–p75 presented here. These results suggest that DDs employ diverse mechanisms for interactions [53]. They can be classified into three types of interactions (reviewed in [52]). Type I interaction is exemplified by the procaspase-9 CARD:Apaf-1 CARD complex [54], whereas the type II interaction is represented by the Pelle DD:Tube DD complex [53], and the type III interaction is proposed to exist in the Fas DD:FADD DD complex [55]. p75 and p45 DD interaction appears symmetrical based on chemical shift data, but one cannot exclude an asymmetrical interaction that uses the same regions of the interface; as p75 is a homodimer and p45 and p75 binding sites overlap only in parts, the p45–p75 interaction could not be totally symmetric. Because p75DD interacts with itself through residues located between helices α-3 and the loop connecting α-3 and α-4, they belong to a type III interaction [48]. The asymmetric type I, II, and III interactions between DDs are conserved in all current structures of oligomeric DD signaling complexes [55]–[57]. These interactions likely represent the predominant mechanism of DD polymerization. Here, p45DD acts as an inhibitor of those interactions, shutting off or modulating the p75 signal strength. Interestingly, both p45DD and p75DD are promiscuous and can interact also with other proteins through their DD. Although p75DD is able to interact with downstream targets, such as the CARD of RIP-2 [58], p45DD appears to interact with FADD, thereby reducing FADD-mediated cell death [29]. Recently the regions of p75DD involved in the three different p75 signaling paradigms has been mapped by mutagenesis—namely, apoptosis, NF-κB activation, and Rho signaling [59]. For p75/p45 heterodimers, p45DD will occupy the region very close to where the CARD domain of RIP-2 is binding to p75DD, according to previous data [59]. Strikingly, RIP-2 binding to p75 is necessary for Rho-GDI release and RhoA inhibition. The data suggest that p45DD binding to p75 will release RhoGDI and inhibit RhoA activation. This is in agreement with our data that p45 binding to p75 inhibits RhoA activity. Recently it has been described that p75 could adopt two different conformations, a symmetrical dimer, stabilized by a cysteine disulfide bond, and an asymmetrical dimer [41]. The authors proposed that p75 could be in equilibrium between both conformations unless oxidant conditions inside the cell promote the formation of the disulfide bond [41]. Our NMR data suggest that the symmetrical conformation is the predominant form at least in solution, because interaction between residues from helix 3 and helixes 5–6 are not seen in our conditions. Further investigation will be needed to understand which conformation belongs to the active receptor. The fact that p45 is able to bind and to block the symmetrical interface suggest a well-designed and potent p75 inhibitor. Apart from the p75 signaling, p45 might play additional roles of an inhibitory nature. In particular, because the DD of p45 appears to be important for p45/p75 interaction and several members of the TNFR family contain DDs such as TNF-R1 or CD95, which have been shown to play important roles in SCI [60],[61], it is intriguing to speculate whether p45 may antagonize the activity of some of these receptors upon SCI by binding to their DD domains as well. Thus, the promiscuous structural nature of p45 may facilitate functional recovery after SCI by inhibiting multiple signaling pathways that are detrimental for neuronal survival and nerve regeneration. In summary, p45 presents an example of a new antagonizing mechanism by which an interaction mediated by the TM and cytoplasmic domains is able to inhibit p75 disulfide dimer formation and function. Such knowledge provides a new glimpse into our understanding of the multiple distinct activities and signaling capabilities of p75. Materials and Methods Ethics Statement In this study, animal protocols were approved by the Institutional Animal Care and Use Committee (IACUC) of the Salk Institute and the Council on Accreditation of the Association for Assessment and Accreditation of Laboratory Animal Care (AAALAC International). Euthanasia will be performed by methods specified and approved by the IACUC Panel on Euthanasia. Antibodies The following antibodies were used in the present studies. Anti-p75 antibodies (anti-p75ECD epitope: 9651, a gift from Dr. Moses Chao, New York University; anti-p75ICD epitope: Buster, a gift from Dr. Phil Barker, Montreal Neurological Institute), anti-p45 antibodies (anti-p45ECD epitope, 6750 and antip45ICD epitope, 6655 were generated in house), and anti-V5 antibody were purchased from Life Technologies (R960-25); anti-Falg M2 antibody was purchased from Sigma (F1804); and anti-HA antibodies were purchased from Roche (clone 3F10, 11867423001) and Santa Cruz Biotechnology (sc-805). Protein Construct Expression and Purification For the recombinant expression of p45ICD, Escherichia coli BL21(DE3) freshly transformed with the pGST-p45 expression vector, which encodes a N-terminal GST tag and a thrombin cleavage site, was used. Two liters of M9 minimal medium containing either (NH4)2SO4 or (15NH4)2SO4 and either glucose or 13C-glucose as the sole nitrogen and carbon sources were inoculated in the presence of ampicillin (100 µg/ml) with 30 ml of preculture of pGST-p45–containing E. coli BL21(DE3) cells, which had been grown at 37 °C overnight. At OD600 = 0.6, expression was induced with 1 mM isopropyl L-D-galactopyranoside. The cells were harvested after 4 h, and the pellet was resuspended in 30 ml of PBS. After sonication, the cell lysate was centrifuged at 20,000 g for 30 min, and the supernatant was incubated with Glutathione Sepharose (Amersham) in PBS with DTT 1 mM at 4 °C for 1 h. After several washes with PBS, the GST fusion protein was removed by thrombin digestion overnight at room temperature while still bound to the sepharose resin. The digest was incubated with benzamidine resin to remove the thrombin (Sigma). The supernatant contained free p45ICD with 95% purity according to SDS-PAGE. For the recombinant expression of human p75ICD, E. coli BL21 (DE3) freshly transformed with the pHisp75 expression vector, which encodes a N-terminal 22-aa affinity tag containing a six-histidine sequence and a thrombin cleavage site, was used. Two liters of unlabeled or stable-isotope–labeled minimal medium (see above) containing kanamycin (50 µg/ml) were inoculated with 30 ml of preculture of pHisp75-containing E. coli BL21(DE3) cells that had been grown at 37 °C overnight. At OD600 = 0.6, expression was induced with 1 mM isopropyl L-D-galactopyranoside. The cells were harvested after 4 h, and the pellet was resuspended in 30 ml buffer A (25 mM Tris-HCl, pH 8.0/500 mM NaCl/10 mM β-mercaptoethanol). After sonication, the cell lysate was centrifuged at 20,000 g for 30 min, and the supernatant was applied to a Ni2+-charged NTA column (Qiagen, Chatsworth, CA). The fusion protein was eluted with a stepwise gradient of 0–500 mM imidazole in buffer A. After dialysis against PBS (pH 8.0), the N-terminal fusion tail was removed by thrombin cleavage performed as described above. The supernatant contained free p75ICD with 95% purity according to SDS-PAGE. The mutant p75 constructs were expressed and purified accordingly. The protein constructs were concentrated using a 10 kDa centriprep amicon concentrator. The concentrations of all proteins used in this study were determined from their absorbance at 280 nm by using molar extinction coefficients calculated from the Expasy protein server software (http://www.expasy.ch). If not stated otherwise, all biophysical experiments were measured in PBS pH 8, 100 mM NaCl. Size Exclusion Chromatography Ni-NTA purified p75ICD and GST-sepharose purified p45ICD were loaded on an S200-Superdex gel filtration column at 4 °C and eluted isocratically in PBS pH 8.0 buffer at 0.7 ml/min. Analytical Ultracentrifugation Sedimentation equilibrium measurements of samples of p75ICD and p45ICD were conducted at concentrations of 10 µM, 300 µM, and 700 µM. Data were collected at four different speeds (10,000, 14,000, 20,000, and 28,000 rpm). NMR Spectroscopy The NMR experiments were carried out on a Bruker DRX700 spectrometer at 25 °C by using protein solutions that contained PBS pH 8.0, 100 mM NaCl, 1 mM sodium azide, and 95%/5% H2O/D2O. Sequential assignment and structure determination was performed with the standard protocol for 13C, 15N-labeled proteins [62]. Hence, sequential assignments of backbone resonances of 15N,13C-labeled p45ICD and 15N,13C-labeled p75ICD were obtained from HNCAcodedCB, HNCAcodedCO [63]. HNCA [42] and 15N-resolved [64] NOESY [65] spectra. The side chain signals of p45ICD were assigned from HCCH-COSY [66] and 13C-resolved [64] NOESY experiments. Aromatic side chain assignments were obtained with 2D [64] NOESY in D2O [64]. Distance constraints for the calculation of the 3D structure were derived from 3D 13C-,15N-resolved [64] NOESY and 2D [64] NOESY spectra recorded with a mixing time of 80 ms. [42]-TROSY [43] spectra with parameters as described below were measured for p75ICD mutants. The data were analyzed using the CARA software program (www.nmr.ch). Chemical Shift Perturbation Experiments For the chemical shift perturbation experiments, [42]-TROSY spectra [43] of stable isotope-labeled p45ICD or p75ICD were measured with t1,max = 88 ms, t2,max = 98 ms, and a data size of 200×1,024 complex points. [42]-TROSY experiments of 13C,15N-labeled p45ICD were performed at protein concentrations of 0.1 mM and 0.5 mM in order to study the oligomerization state of p45ICD. [42]-TROSY experiments of 15N-labeled p75ICD were performed at protein concentrations of 10 mM, 0.1 mM, 0.2 mM, 0.5 mM, and 2 mM to study the oligomerization state of p75ICD and to elucidate the homodimer interface. [42]-TROSY experiments of 15N-labeled p75ICD were performed at a protein concentration of 10 mM free and in the presence of 0.1 mM unlabeled p45ICD to study the p45ICD–p75ICD heterodimer interface. [42]-TROSY experiments of 15N-labeled p75ICD were performed at a protein concentration of 0.5 mM at a 1:0 mixture of 15N-labeled p75ICD and unlabeled p45ICD followed by stepwise addition of unlabeled p45ICD up to a p45ICD concentration of 2 mM (protein ratios were 1:0, 1:0.5, 1:1, 1:2, and 1:4). The same NMR setups were used in titration experiments performed to investigate the binding site of p75ICD on p45ICD. The titration experiments were started with a 1:0 mixture of 13C,15N-labeled p45ICD, and unlabeled p75ICD was added stepwise from 0 mM to 2 mM (protein ratios were 1:0.5, 1:1, 1:2, and 1:4). Structure Calculation We observed 2,130 NOEs in the NOESY spectra, leading to 1,065 meaningful distance restraints and 372 angle restraints (Table S2). For the structure calculation, the program CYANA was used [67],[68], followed by restrained energy minimization using the program INSIGHT. CYANA initially generated 100 conformers, and the 20 conformers with the lowest energy were used to represent the three-dimensional NMR structure. The 20 refined conformers showed small residual constraint violations that are compatible with the observed NOEs and the short interatomic distances (Table S2). Similar energy values were obtained for all 20 conformers. The quality of the structures is reflected by the RMSD values of 0.65 Å relative to the mean coordinates of p45 residues 141–218 (see Table S2 and Figure 5A). The bundle of 20 conformers representing the NMR structure is deposited in the PDB database under accession no. 2IB1. Transfection and Immunoprecipitation Constructs containing NgR, full-length p45 or p75, as well as deletion and amino-acid point mutants were transfected into HEK293 cells by TransFectin transfection reagents (BioRad). Transfected cells were collected and lysed in RIPA buffer (150 mM NaCl, 1% NP-40, 0.5% DOC, 0.1% SDS, 50 mM Tris pH 8.0). Lysates were immunoprecipitated with antibodies described in the text. Samples were analyzed using SDS-PAGE and Western blots. For quantitative analysis, gel images were analyzed by Image J program (NIH). Statistical analyses and data graph were done using Prism software. RNA Transfection and Neurite Outgrowth Assay Full-length p45 RNA with capping at the 5′ end and poly(A) sequences was transcribed in vitro using the mMessage-mMACHINE kit (Ambion). RNA was transfected into CGNs with the TransMessenger transfection reagent (Qiagen). We found that this protocol achieves 50%–70% transfection efficiency. Neurite outgrowth assay using CGNs was carried out as previously described [14]. RhoA Assay CGN culture was performed as previously described [13],[14]. Briefly, p5–p7 cerebella were isolated from WT and Thy1-p45 mice. Neurons were plated on Poly-D-Lysine–coated six-well tissue culture dishes at the density of 5 million cells per well. Cells are allowed to grow for 48 h and then starved in basal medium eagle (BME, Life Technologies) for 7 h before being treated with preclustered MAG-Fc at the concentration of 2 µg/ml for 15 min. Preclustered human IgG was used as the control. The preclustering is achieved by incubating the MAG-Fc with an anti-human-Fc antibody at a 2:1 molar ratio in BME for 30 min in 37 °C. Cells are then lysed on ice according to the manufacturer's suggestion, and the RhoA assay was performed following the instructions of the G-LISA (absorbance based) kit. The RhoA activity is measured by reading the 490 nm absorbance using a 96-well plate reader. We have also used the Millipore (Upstate) RhoA assay kit for a pull-down of activated RhoA by Western blotting analysis. Ethics Statement In this study, animal protocols were approved by the Institutional Animal Care and Use Committee (IACUC) of the Salk Institute and the Council on Accreditation of the Association for Assessment and Accreditation of Laboratory Animal Care (AAALAC International). Euthanasia will be performed by methods specified and approved by the IACUC Panel on Euthanasia. Antibodies The following antibodies were used in the present studies. Anti-p75 antibodies (anti-p75ECD epitope: 9651, a gift from Dr. Moses Chao, New York University; anti-p75ICD epitope: Buster, a gift from Dr. Phil Barker, Montreal Neurological Institute), anti-p45 antibodies (anti-p45ECD epitope, 6750 and antip45ICD epitope, 6655 were generated in house), and anti-V5 antibody were purchased from Life Technologies (R960-25); anti-Falg M2 antibody was purchased from Sigma (F1804); and anti-HA antibodies were purchased from Roche (clone 3F10, 11867423001) and Santa Cruz Biotechnology (sc-805). Protein Construct Expression and Purification For the recombinant expression of p45ICD, Escherichia coli BL21(DE3) freshly transformed with the pGST-p45 expression vector, which encodes a N-terminal GST tag and a thrombin cleavage site, was used. Two liters of M9 minimal medium containing either (NH4)2SO4 or (15NH4)2SO4 and either glucose or 13C-glucose as the sole nitrogen and carbon sources were inoculated in the presence of ampicillin (100 µg/ml) with 30 ml of preculture of pGST-p45–containing E. coli BL21(DE3) cells, which had been grown at 37 °C overnight. At OD600 = 0.6, expression was induced with 1 mM isopropyl L-D-galactopyranoside. The cells were harvested after 4 h, and the pellet was resuspended in 30 ml of PBS. After sonication, the cell lysate was centrifuged at 20,000 g for 30 min, and the supernatant was incubated with Glutathione Sepharose (Amersham) in PBS with DTT 1 mM at 4 °C for 1 h. After several washes with PBS, the GST fusion protein was removed by thrombin digestion overnight at room temperature while still bound to the sepharose resin. The digest was incubated with benzamidine resin to remove the thrombin (Sigma). The supernatant contained free p45ICD with 95% purity according to SDS-PAGE. For the recombinant expression of human p75ICD, E. coli BL21 (DE3) freshly transformed with the pHisp75 expression vector, which encodes a N-terminal 22-aa affinity tag containing a six-histidine sequence and a thrombin cleavage site, was used. Two liters of unlabeled or stable-isotope–labeled minimal medium (see above) containing kanamycin (50 µg/ml) were inoculated with 30 ml of preculture of pHisp75-containing E. coli BL21(DE3) cells that had been grown at 37 °C overnight. At OD600 = 0.6, expression was induced with 1 mM isopropyl L-D-galactopyranoside. The cells were harvested after 4 h, and the pellet was resuspended in 30 ml buffer A (25 mM Tris-HCl, pH 8.0/500 mM NaCl/10 mM β-mercaptoethanol). After sonication, the cell lysate was centrifuged at 20,000 g for 30 min, and the supernatant was applied to a Ni2+-charged NTA column (Qiagen, Chatsworth, CA). The fusion protein was eluted with a stepwise gradient of 0–500 mM imidazole in buffer A. After dialysis against PBS (pH 8.0), the N-terminal fusion tail was removed by thrombin cleavage performed as described above. The supernatant contained free p75ICD with 95% purity according to SDS-PAGE. The mutant p75 constructs were expressed and purified accordingly. The protein constructs were concentrated using a 10 kDa centriprep amicon concentrator. The concentrations of all proteins used in this study were determined from their absorbance at 280 nm by using molar extinction coefficients calculated from the Expasy protein server software (http://www.expasy.ch). If not stated otherwise, all biophysical experiments were measured in PBS pH 8, 100 mM NaCl. Size Exclusion Chromatography Ni-NTA purified p75ICD and GST-sepharose purified p45ICD were loaded on an S200-Superdex gel filtration column at 4 °C and eluted isocratically in PBS pH 8.0 buffer at 0.7 ml/min. Analytical Ultracentrifugation Sedimentation equilibrium measurements of samples of p75ICD and p45ICD were conducted at concentrations of 10 µM, 300 µM, and 700 µM. Data were collected at four different speeds (10,000, 14,000, 20,000, and 28,000 rpm). NMR Spectroscopy The NMR experiments were carried out on a Bruker DRX700 spectrometer at 25 °C by using protein solutions that contained PBS pH 8.0, 100 mM NaCl, 1 mM sodium azide, and 95%/5% H2O/D2O. Sequential assignment and structure determination was performed with the standard protocol for 13C, 15N-labeled proteins [62]. Hence, sequential assignments of backbone resonances of 15N,13C-labeled p45ICD and 15N,13C-labeled p75ICD were obtained from HNCAcodedCB, HNCAcodedCO [63]. HNCA [42] and 15N-resolved [64] NOESY [65] spectra. The side chain signals of p45ICD were assigned from HCCH-COSY [66] and 13C-resolved [64] NOESY experiments. Aromatic side chain assignments were obtained with 2D [64] NOESY in D2O [64]. Distance constraints for the calculation of the 3D structure were derived from 3D 13C-,15N-resolved [64] NOESY and 2D [64] NOESY spectra recorded with a mixing time of 80 ms. [42]-TROSY [43] spectra with parameters as described below were measured for p75ICD mutants. The data were analyzed using the CARA software program (www.nmr.ch). Chemical Shift Perturbation Experiments For the chemical shift perturbation experiments, [42]-TROSY spectra [43] of stable isotope-labeled p45ICD or p75ICD were measured with t1,max = 88 ms, t2,max = 98 ms, and a data size of 200×1,024 complex points. [42]-TROSY experiments of 13C,15N-labeled p45ICD were performed at protein concentrations of 0.1 mM and 0.5 mM in order to study the oligomerization state of p45ICD. [42]-TROSY experiments of 15N-labeled p75ICD were performed at protein concentrations of 10 mM, 0.1 mM, 0.2 mM, 0.5 mM, and 2 mM to study the oligomerization state of p75ICD and to elucidate the homodimer interface. [42]-TROSY experiments of 15N-labeled p75ICD were performed at a protein concentration of 10 mM free and in the presence of 0.1 mM unlabeled p45ICD to study the p45ICD–p75ICD heterodimer interface. [42]-TROSY experiments of 15N-labeled p75ICD were performed at a protein concentration of 0.5 mM at a 1:0 mixture of 15N-labeled p75ICD and unlabeled p45ICD followed by stepwise addition of unlabeled p45ICD up to a p45ICD concentration of 2 mM (protein ratios were 1:0, 1:0.5, 1:1, 1:2, and 1:4). The same NMR setups were used in titration experiments performed to investigate the binding site of p75ICD on p45ICD. The titration experiments were started with a 1:0 mixture of 13C,15N-labeled p45ICD, and unlabeled p75ICD was added stepwise from 0 mM to 2 mM (protein ratios were 1:0.5, 1:1, 1:2, and 1:4). Structure Calculation We observed 2,130 NOEs in the NOESY spectra, leading to 1,065 meaningful distance restraints and 372 angle restraints (Table S2). For the structure calculation, the program CYANA was used [67],[68], followed by restrained energy minimization using the program INSIGHT. CYANA initially generated 100 conformers, and the 20 conformers with the lowest energy were used to represent the three-dimensional NMR structure. The 20 refined conformers showed small residual constraint violations that are compatible with the observed NOEs and the short interatomic distances (Table S2). Similar energy values were obtained for all 20 conformers. The quality of the structures is reflected by the RMSD values of 0.65 Å relative to the mean coordinates of p45 residues 141–218 (see Table S2 and Figure 5A). The bundle of 20 conformers representing the NMR structure is deposited in the PDB database under accession no. 2IB1. Transfection and Immunoprecipitation Constructs containing NgR, full-length p45 or p75, as well as deletion and amino-acid point mutants were transfected into HEK293 cells by TransFectin transfection reagents (BioRad). Transfected cells were collected and lysed in RIPA buffer (150 mM NaCl, 1% NP-40, 0.5% DOC, 0.1% SDS, 50 mM Tris pH 8.0). Lysates were immunoprecipitated with antibodies described in the text. Samples were analyzed using SDS-PAGE and Western blots. For quantitative analysis, gel images were analyzed by Image J program (NIH). Statistical analyses and data graph were done using Prism software. RNA Transfection and Neurite Outgrowth Assay Full-length p45 RNA with capping at the 5′ end and poly(A) sequences was transcribed in vitro using the mMessage-mMACHINE kit (Ambion). RNA was transfected into CGNs with the TransMessenger transfection reagent (Qiagen). We found that this protocol achieves 50%–70% transfection efficiency. Neurite outgrowth assay using CGNs was carried out as previously described [14]. RhoA Assay CGN culture was performed as previously described [13],[14]. Briefly, p5–p7 cerebella were isolated from WT and Thy1-p45 mice. Neurons were plated on Poly-D-Lysine–coated six-well tissue culture dishes at the density of 5 million cells per well. Cells are allowed to grow for 48 h and then starved in basal medium eagle (BME, Life Technologies) for 7 h before being treated with preclustered MAG-Fc at the concentration of 2 µg/ml for 15 min. Preclustered human IgG was used as the control. The preclustering is achieved by incubating the MAG-Fc with an anti-human-Fc antibody at a 2:1 molar ratio in BME for 30 min in 37 °C. Cells are then lysed on ice according to the manufacturer's suggestion, and the RhoA assay was performed following the instructions of the G-LISA (absorbance based) kit. The RhoA activity is measured by reading the 490 nm absorbance using a 96-well plate reader. We have also used the Millipore (Upstate) RhoA assay kit for a pull-down of activated RhoA by Western blotting analysis. Supporting Information Figure S1. p45 does not bind to NgR. A plasmid expressing V5-tagged p45 or V5-tagged p75 were co-transfected in 293T cells with a plasmid encoding for Flag-NgR. Western blots show that p75 and NgR interact upon co-transfection and co-immunoprecipitation with Flag antibody (M2). However, p45 does not co-immunoprecipitate with Flag-NgR, suggesting p45 modulates p75/NgR signaling through p75, not directly interacting with NgR. https://doi.org/10.1371/journal.pbio.1001918.s001 (TIF) Figure S2. The inhibition of p75/NgR interaction by p45 requires the TM and ICD domains of p45. Different p45 deletion mutants were co-transfected with p75- and hNgR-expressing vectors. p45 devoid of the ECD has similar blocking activity as the full-length p45. In contrast, constructs without the ICD or TM display a much lower blocking activity. CTL, 100%; *** p<0.0001; p45-FL, 54.25±7.69, N = 4, compared to p45-ECD-TM, 65.25±4.69, N = 4; unpaired t test, ns; F test, ns. p45-ICD, 67±4.55, N = 4, compared to p45-TM-ICD, 45.5±5.78, N = 4, unpaired t test: * p<0.1; F test, ns. p45-ECD-TM, 65.25±4.69, N = 4, compared to p45-TM-ICD, 45.5±5.78, N = 4, * p<0.1. The data can be found in Table S1. https://doi.org/10.1371/journal.pbio.1001918.s002 (TIF) Figure S3. Overexpression of p45 inhibits MAG-Fc–induced RhoA activation. (A) Increased p45 protein levels in P5–P7 CGNs following transfection of p45 RNA. (B) Following transfection with the p45 RNA, GCNs were treated with MAG-Fc and subjected to a RhoA activity assay. Overexpression of p45 inhibited MAG-Fc–induced RhoA activation. https://doi.org/10.1371/journal.pbio.1001918.s003 (TIF) Figure S4. Iodoacetamide purification of p75-ICD. SDS-PAGE in reducing and nonreducing conditions of p75-ICD purified from E. coli using iodoacetamide, a blocking agent of free cysteines, in the lysis buffer. The absence of dimerized p75-ICD in these conditions suggests that p75-ICD dimerization is produced during the purification as a result of oxidation of free cysteines. https://doi.org/10.1371/journal.pbio.1001918.s004 (TIF) Figure S5. p75-ICD covalent disulfide dimmer formation in the presence of hydrogen perxiode. Nonreducing and reducing SDS-PAGE of purified p75-ICD from E. coli was incubated with hydrogen peroxide (10 mM) during the indicated time points. Note that p75-ICD purified from E. coli (without DTT, t = 0 min) has already some amount of disulfide dimer. https://doi.org/10.1371/journal.pbio.1001918.s005 (TIF) Figure S6. p75-Cys379 is responsible for disulfide dimerization of p75-ICD. Gel filtration profile of purified p75-ICD WT in reducing (blue) and nonreducing conditions (dark blue) and of purified p75-C379S (green). The elution of p75-C379S is indicative of a monomeric p75-ICD. https://doi.org/10.1371/journal.pbio.1001918.s006 (TIF) Figure S7. Summary of NOEs. Observed NOEs are summarized for p45ICD. Sequential NOEs are indicated by thick horizontal bars. The thickness of the bar is proportional to the magnitude of the NOE intensity. Thin horizontal bars indicate long-distance NOEs. https://doi.org/10.1371/journal.pbio.1001918.s007 (TIF) Figure S8. Insights into p45ICD-p75ICD heterodimer formation. p45ICD-dependent chemical shift changes versus the amino acid sequence observed in stable isotope-labeled p75ICD at (B) 10 µM and (C) 2 mM p75ICD concentration. The bar plot represents the normalized change of the chemical shifts of p75ICD in the absence and presence of p45ICD in the [15N,1H]-TROSY spectrum using the equation N = 25[Δ(δ(1H))2 + Δ(δ(15N))2]0.5, where δ(1H) and δ(15N) are the chemical shifts in part per million (ppm) along the ω2(1H) and ω2(15N) dimensions, respectively. Perturbations larger than 0.2 ppm are labeled. https://doi.org/10.1371/journal.pbio.1001918.s008 (TIF) Figure S9. 2D-NMR of selected 15N-labelled p45-ICD mutants expressed in E. coli. p45 DD mutants were expressed and purified as 15N-labelled proteins and analyzed by NMR spectra, indicating that all p45 mutants are correctly folded like p45-WT. https://doi.org/10.1371/journal.pbio.1001918.s009 (TIF) Figure S10. Insights into p45–p75 heterodimer formation from NMR. (A) NMR analysis of p45-DD and p75-DD interaction suggests a heterodimer formation. A378, red colored lines represent the cross-peak of A378 in the [15N,1H] TROSY spectrum at a high concentration of p75ICD (homodimer), and the corresponding cross-peak at a low concentration (monomer) of p75ICD is represented by green dashed lines, respectively. Upon p45ICD addition to a highly concentrated p75ICD sample, the cross-peak of A378 colored as blue lines shifted to the position of the monomer. In contrast, upon p45ICD addition to the sample with low p75ICD concentration, the cross-peak of A378—represented as blue dashed lines—did not shift. These findings indicate that A378 is not part of the p45–p75 interface and that p45 breaks the p75 homodimer. T375, the cross-peaks of T375 in the [15N,1H]-TROSY spectra at high and low p75ICD concentrations and in the presence and absence of p45ICD are displayed with the same color code as for A378. The addition of p45ICD at high p75ICD concentrations results in a shift of the cross-peak of T375 (red cross-peak to blue cross-peak). The cross-peak of T375 at low p75ICD concentrations is also shifted upon addition of p45ICD (cross-peak represented by green dashed lines to cross-peak represented with blue dashed lines). Because the position of the cross-peak of T375 in the presence of p45 is independent of the p75ICD concentration, T375 appears to be part of the p45–p75 interface as well as part of the p75–p75 interface. https://doi.org/10.1371/journal.pbio.1001918.s010 (TIF) Figure S11. Co-localization of p75 and p45 expression in lumbar spinal motorneurons after sciatic nerve crush. Confocal images of immunofluorescence staining of spinal cord sections from mice with the sciatic nerve crush. (A) p75 staining, (B) p45 staining, and (C) merge of (A) and (B). Some p75-expressing motor neurons also express p45. https://doi.org/10.1371/journal.pbio.1001918.s011 (TIF) Table S1. Data for Figures 2, 5, and S2. https://doi.org/10.1371/journal.pbio.1001918.s012 (XLSX) Table S2. NMR and structural statistics of the 3D structure of p45-DD. https://doi.org/10.1371/journal.pbio.1001918.s013 (DOCX) Table S3. Structural homology search with DALI Server showing the top 10 matches. https://doi.org/10.1371/journal.pbio.1001918.s014 (DOCX) Acknowledgments We thank B. Zheng for critical reading of the manuscript, J. Rivier for p45 peptides, J. Vaugh for generating anti-p45 antibodies, and L. Kevin Wang and Zhigang He for myelin-associated inhibitors and for the protocol for granule neuron cultures and nerve growth inhibition assay.

doi: 10.1371/journal.pbio.1001924pmid: 25093574

The ultimate cause of Alzheimer's disease (AD) may be accumulation of the excess proteins found in plaques and tangles in the brain, but the immediate causes of cognitive dysfunction in the disease likely lie several steps downstream from the disruption of protein homeostasis. One implicated pathway involves striatal-enriched protein tyrosine phosphatase, or STEP, a neuron-specific enzyme that, among other jobs, regulates the trafficking of synaptic glutamate receptors and the activity of a group of widely active kinases. STEP is overactive in AD, in part because it isn't degraded fast enough, and its overactivity disrupts the post-synaptic events that underlie learning and memory. In animal models of AD, knocking out STEP improves cognition. Thus, STEP inhibition is a potential target for treatment of AD. In this issue of PLOS Biology, Jian Xu, Paul Lombroso, and colleagues report their discovery of a new class of STEP inhibitor—a discovery that involved a small but significant bit of serendipity—and demonstrate its potential in an AD animal model. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 1. From impurity to drug. On the left is the normal cyclic eight-atom form of elemental sulfur, present as an impurity in some of the drug samples. In the center is the related compound benzopentathiepin, which contains five sulfur atoms attached in a loop to a benzene ring. On the right is a derivative of benzopentathein, TC-2153, which has more desirable pharmacological properties. https://doi.org/10.1371/journal.pbio.1001924.g001 The authors began by conducting a high-throughput screen of 150,000 compounds, testing the ability of each to inhibit STEP's phosphatase activity. As is usual in such screens, a number of good candidates emerged. These were winnowed down to eight, chosen for their high activity at low concentration and favorable properties, such as likely ability to cross the blood-brain barrier and absence of known toxic moieties, all important for developing a centrally active drug. Following standard practice, next, they synthesized the molecules from scratch, and here got a surprise—the compounds displayed little STEP inhibitory activity. Some chemical detective work revealed the true inhibitor was elemental sulfur, S8, present as an impurity in the commercially obtained samples used in the screening. This ring compound doesn't make a good drug, so the authors investigated a structurally related compound, benzopentathiepin, containing a ring of six carbons fused to a ring of five sulfurs. A derivative, TC-2153, was known to have low toxicity and was likely to cross the blood-brain barrier, and, they found, was a potent inhibitor of STEP. They showed that TC-2153 increased the phosphorylation state of multiple STEP substrates, both in cell culture and in mice. It was also relatively specific for STEP, with little inhibition of related phosphatases in noncortical neurons (STEP is restricted to the central nervous system), or in the periphery (where STEP is not expressed). Exactly how TC-2153 inhibits STEP is still under investigation, but it appears to interact with sulfurs in the enzyme's catalytic site, forming an irreversible covalent linkage with them. To test whether TC-2153 could reverse some of the cognitive effects of STEP overactivity, the authors turned to the “triple transgenic” mouse model, with mutations in three genes known to cause AD: presenilin 1, amyloid precursor protein, and tau. Compared to vehicle, intraperitoneal injection of TC-2153 improved spatial working memory, novel object recognition, and reference memory, all standard tests of cognitive function in AD models. The treatment had no effect on either a-beta, found in amyloid plaques outside of cortical neurons, or phospho-tau, found in neurofibrillary tangles inside them, indicating that the beneficial effect of TC-2153 was not due to alteration of events upstream of STEP overactivity. The fortuitous discovery of this novel class of protein tyrosine phosphatase inhibitors is likely to lead to further development of potential drugs for AD, as well as several other neuropsychiatric and neurodegenerative diseases in which STEP is overactive. Whether any of these can become effective treatments for these diseases is, of course, a far more challenging question, given the disappointing track record of drug discovery in this field to date. But the demonstration that cognitive dysfunction can be ameliorated through inhibition of this enzyme may lead to better understanding of how that dysfunction arises, which by itself is no small…step. Xu J, Chatterjee M, Baguley TD, Brouillette J, Kurup P, et al. (2014) Inhibitor of the Tyrosine Phosphatase STEP Reverses Cognitive Deficits in a Mouse Model of Alzheimer's Disease. doi:10.1371/journal.pbio.1001923

doi: 10.1371/journal.pbio.1001926pmid: 25116890

Introduction Reconstructing the traits of the last universal common ancestor (LUCA) requires constraining the relationships between the three domains of life, the archaea, bacteria, and eukaryotes. Recent phylogenetic studies show that eukaryotes are secondarily derived: they are genomic chimeras, arising from an endosymbiosis between a bacterium and an archaeal host cell [1]–[5]. The divergence between the two primary domains, the archaea and the bacteria, is now seen as the deepest branch in the tree of life [1],[6]–[8]. The properties of LUCA are most parsimoniously those shared by bacteria and archaea. This leads straight to a serious paradox. Archaea and bacteria share core biochemistry, including the genetic code, transcription machinery, and ribosomal translation [9], but differ for unknown reasons in fundamental traits including cell membrane [10] and cell wall [11], glycolysis [12], ion pumping [13], and even DNA replication [14]. The differences in membrane lipids may be the key to this major unsolved problem in biology. Phospholipid side chains are typically isoprenoids in archaea and fatty acids in bacteria [15]. While this could reflect adaptive evolution [16], archaea and bacteria also differ in the stereochemistry of the glycerol-phosphate headgroup [10]. Archaeal lipids have an sn-glycerol-1-phosphate (G1P) headgroup, while bacteria use the mirror structure sn-glycerol-3-phosphate (G3P) (Figure 1). There is no persuasive selective explanation for these opposite stereochemistries [10],[13],[17]. The enzymes involved, glycerol-1-phosphate-dehydrogenase (G1PDH) in archaea and glycerol-3-phosphate-dehydrogenase (G3PDH) in bacteria, bear no phylogenetic resemblance, suggesting they arose independently [10]. If so, then LUCA did not possess a modern membrane—a seemingly improbable conclusion, given the central importance of membranes to cells [10],[17],[18]. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 1. Membrane lipids of archaea and bacteria. Archaeal lipids (left) are typically composed of isoprenoid chains linked by ether bonds to an sn-glycerol-1-phosphate (G1P) backbone. The chirality of the two glycerol backbones is fully conserved within each clade not only in structure but in their unrelated synthetic enzymes. Although ether linkages have been observed in bacterial membranes [15] and isoprenoids are common to all three domains, bacterial lipids (right) are typically composed of fatty acids in ester linkage to an sn-glycerol-3-phosphate (G3P) skeleton. Despite widespread horizontal gene transfer, no bacterium has been observed with the archaeal enantiomer, or vice versa [10]. https://doi.org/10.1371/journal.pbio.1001926.g001 Set against this paradoxical difference in membrane composition is the universality of membrane bioenergetics [19]. Essentially all cells power ATP synthesis through chemiosmotic coupling, in which the ATP synthase (ATPase) is powered by electrochemical differences in H+ or Na+ concentration across membranes [20]. The ATPase is universally conserved [21] and shares the same deep phylogenetic split as the ribosome, implying that both were present in LUCA [22]–[24]. The deepest branches in the tree of life are entirely populated by autotrophs [1],[6],[7],[12],[25], which also depend on chemiosmotic coupling to drive carbon metabolism via proteins such as the energy-converting hydrogenase (Ech) and ferredoxin [26]. But there are serious objections to the idea that LUCA was chemiosmotic. Pumping protons across membranes requires sophisticated proteins, which are only useful in membranes impermeable to protons [27]. Unlike the ATPase, no ion pumps are universally conserved [13]. The pathways for heme and quinone synthesis (the major cofactors of respiratory proteins) also differ in archaea and bacteria, although their distribution is complicated by lateral gene transfer, as is reconstruction of the phylogenetic origins of respiratory ion pumps [13]. But it seems likely that both lipid membranes and active pumping are evolutionarily distinct in archaea and bacteria [9],[11]. It is hard to reconcile these fundamental differences with the universality of the ATPase. On the face of it, LUCA was chemiosmotic, yet did not have a modern phospholipid membrane or active ion pumps. A possible resolution is that LUCA exploited natural (geochemically sustained) proton gradients [18],[28],[29]. However, the hypothesis that natural proton gradients could drive carbon and energy metabolism in LUCA, in the absence of active ion pumps, faces a serious drawback. Because fluids are electrically balanced, the transfer of H+ ions down a concentration gradient, from an acid solution into a cell, transfers positive charge into the cell, generating a membrane potential that opposes further influx. The system swiftly reaches electrochemical (Donnan) equilibrium, in which electrical charges and concentration differences are offset [30]. Equilibrium is death: natural proton gradients could only drive carbon and energy metabolism in LUCA if such equilibrium is avoided—in effect, if protons accumulating inside a cell can leave again. Membrane permeability could be critical to maintaining disequilibrium in any system with continuous flow, as leaky membranes impose less of a barrier to the continued flux of H+, OH−, and other ions [19]. The feasibility of this hypothesis depends on the dynamics of ion fluxes that are unknown. We have therefore built a model to estimate quantitative differences in free energy (−ΔG) across lipid membranes exposed to natural proton gradients. We consider a cell exposed simultaneously to alkaline fluids and relatively acidic water (Figure 2). Our model is independent of any particular setting, but requires continuous laminar flow with limited mixing (as found in microporous alkaline hydrothermal vents [18],[19],[24],[31]–[33] and potentially other environments), allowing sharp gradients of several pH units to be maintained across short distances of 1–2 µm. In general, we assume that the external pH does not change on either side of the cell, as external fluids are replenished by continuous flow from large reservoirs (e.g., hydrothermal fluids or the ocean), but we do also consider mixing. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 2. The model. A cell with a semi-permeable membrane sits at the interface between an alkaline and an acidic fluid. The fluids are continuously replenished and otherwise separated by an inorganic barrier. Hydroxide ions (OH−) can flow into the cell from the alkaline side by simple diffusion across the membrane, with protons (H+) entering in a similar manner from the acidic side. Other ions (Na+, K+, Cl−, not shown) diffuse similarly, as a function of their permeability, charge, and respective internal and external concentrations on each side. Inside the protocell, H+ and OH− can neutralize into water, or leave towards either side. Internal pH thus depends on the water equilibrium and relative influxes of each ion. A protein capable of exploiting the natural proton gradient sits on the acidic side, allowing energy assimilation via ATP production, or carbon assimilation via CO2 fixation. https://doi.org/10.1371/journal.pbio.1001926.g002 Protons enter the cell through membrane proteins, and directly through the lipid phase of the membrane. The overall rate of proton influx depends on the difference in proton concentration and electrical charge (upon proton entry) between the outside and inside of the cell, the kinetics of the membrane protein (e.g., ATPase), the number of membrane proteins (given as a proportion of the surface area), the proton permeability of the lipid phase of the membrane, and the rate of loss of protons from inside the cell (see Materials and Methods). For simplicity, we assume that gradient-exploiting membrane proteins are only present on the acid face of the cell. Proton loss from inside the cell therefore depends on the rate of influx of OH− from alkaline fluids, which neutralize protons within the cell, and the rate of loss of protons across the lipid phase to the alkaline exterior (Figure 2). We also consider membrane permeability to Na+, K+, and Cl− ions, which move charge, and hence influence the electrochemical potential difference and the rate of proton flux. By calculating the overall proton flux on the basis of these parameters, we estimate changes in the steady-state proton concentration inside the cell relative to the outside, giving the free energy (−ΔG) available to drive carbon and energy metabolism. Our findings allow us to propose a new and tightly constrained bioenergetic route map leading from a leaky LUCA dependent on natural proton gradients, to the first archaea and bacteria with highly distinct ion-tight phospholipid membranes. These bioenergetic considerations give striking insights into the nature of LUCA, and the deep divergence between archaea and bacteria. Results Free-Energy Availability Depends on Membrane Permeability The model shows that cells with 1% ATPase in a proton-tight membrane with glycerol-phosphate headgroups (giving an H+ permeability <10−5 cm/s, like extant archaea and bacteria [34]), collapse natural proton gradients within seconds (Figure 3A and 3B). The magnitude of the pH gradient depends on the environmental setting. To constrain possibilities we considered pH values commensurate with alkaline hydrothermal vents, but the same principles apply to any other setting with dynamic pH gradients across short distances. The early oceans may have been mildly acidic, as low as pH 5, and alkaline fluids as high as pH 11 [35] but we conservatively set a 3 pH-unit gradient, with the “acid” at pH 7 and alkaline fluids at pH 10. Nonetheless, collapse of the gradient was evident in proton-tight membranes across a range of gradients (Figure 3B). Protons enter through the ATPase faster than they can exit or be neutralized by OH−, so H+ influx rapidly reaches electrochemical equilibrium. In contrast, leaky protocells (equivalent to fatty-acid vesicles without glycerol phosphate headgroups) in a 7∶10 pH gradient with 1% ATPase in the membrane retain nearly all the free energy available, having a −ΔG only ∼17% lower than an open system (i.e., a single membrane containing the same number of membrane proteins, separating a continuous flux of acid and alkaline fluids; Figure 3A). This is because proton flux through the ATPase is ∼4 orders of magnitude faster than through the lipid phase, even with a high proton permeability of 10−2 cm/s (based on the kinetics of proton-flux through the ATPase, see Materials and Methods and Table S1). Leaky cells in natural proton gradients of 3 pH units therefore have sufficient free energy to drive ATP synthesis. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 3. Dynamics of free-energy change (−ΔG) in cells powered by natural proton gradients. (A) Proton-permeable vesicles (≥10−4 cm/s) have only a small loss of free-energy compared with an open system (pH gradient 7∶10, 1% ATPase). Reduced membrane permeability (≤10−4 cm/s), including permeabilities equivalent to modern membranes (<10−5 cm/s), collapse the gradient within seconds. (B) At low permeability (10−6 cm/s), −ΔG collapses regardless of gradient size. Within seconds, H+ flux through ATPase equilibrates with the acidic fluids. (C) The collapse of −ΔG is more extensive the greater the amount of membrane-bound ATPase, even with a leaky membrane (10−3 cm/s). (D) With Ech, the collapse of the natural gradient is similar to that of the ATPase, showing that natural proton gradients can power energy (ATPase) and carbon (Ech) metabolism, given 1%–5% enzyme in membrane. Na+ permeability was kept 6 orders of magnitude higher than that of H+ throughout all simulations in this and all figures of the article. Except in (B), all results were calculated in a pH gradient 7∶10. https://doi.org/10.1371/journal.pbio.1001926.g003 Even leaky cells are sensitive to the amount of membrane protein, with higher proportions of ATPase collapsing the gradient (Figure 3C). In this case, the rate of H+ entry through ATPase covering 10%–50% of the membrane surface area is substantially faster than the rate of clearance of H+ from inside the cell (and reaction with OH−), collapsing −ΔG. However, 1%–5% ATPase in a leaky membrane (10−3 cm/s) retains a −ΔG of close to 20 kJ/mol (Figure 3A and 3C). With 3–4 protons translocated per ATP synthesized (Table S1), this gives a −ΔG for ATP hydrolysis of 60 to 80 kJ/mol, similar to modern cells and sufficient to drive intermediary biochemistry, including aminoacyl adenylation in protein synthesis [36]. This assumes the same stoichiometry as the modern ATPase (3–4 protons per ATP). Because the kinetics of early enzymes would arguably not have been as honed by evolution as their modern equivalents, we used 10% of modern proton flux rates. However, this difference in efficiency actually has limited impact on the model compared with modern flux rates (Figure S1); increasing the stoichiometry of the ATPase has a similarly small effect (Figure S2). We did not estimate rates of ATP synthesis, as that would require additional assumptions about concentrations of ATP, ADP, and phosphate, as well as the rates of ATP consumption and growth; these are almost impossible to constrain at present. The same principles apply to carbon metabolism. We consider whether the membrane protein Ech could drive carbon reduction by H2 in natural proton gradients. Ech uses the proton-motive force to drive carbon metabolism in some archaea and bacteria via the reduction of ferredoxin [26]. As with the ATPase, cells with 1%–5% Ech in the membrane retain most of the free energy available from a 7∶10 pH gradient (Figure 3D). Higher concentrations of Ech (10%–50%) collapse −ΔG even more than the ATPase, as the rate of proton flux through Ech is double that of the ATPase, and its surface area is slightly smaller, so there are more proton pores per unit surface area (Table S1). Such high concentrations of Ech or ATPase are in any case improbable, and not relevant to modern cells, but demonstrate the range of conditions in which natural gradients can in principle drive carbon and energy metabolism. Given a 7∶10 pH gradient, it is therefore feasible to have 1%–5% Ech and 1%–5% ATPase in the membrane, driving both carbon and energy metabolism in cells with leaky membranes. But incorporation of either G1P or G3P glycerol-phosphate headgroups (found in archaea and bacteria respectively), or racemic mixtures of archaeal and bacterial lipids (which, surprisingly, are as impermeable to protons as standard membranes [37]), are not favored because they reduce the proton permeability of the membrane and so collapse the energetic driving force. Glycerol-phosphate headgroups in particular decrease proton permeability, as they prevent fatty acid flip-flop across the membrane (see Discussion). Pumping Ions across Leaky Membranes Does Not Give a Sustained Increase in Free Energy If leaky cells with low amounts of ATPase and Ech (1%–5%) are viable in natural proton gradients, but cells with phospholipid membranes are not, then the evolution of active pumping becomes a paradox: pumping protons across a proton-permeable membrane does not increase free energy (−ΔG), because the protons immediately return through the lipid phase of the membrane. We demonstrate this using a model of a simple H2-dependent proton pump (equivalent to Ech operating in reverse, as found in some simple bacteria and archaea [26]). We find that in a 7∶10 pH gradient −ΔG falls as membrane permeability decreases from 10−2 to 10−6 cm/s (Figure 4A). −ΔG here depends on two factors: active pumping and the natural pH gradient. As membrane permeability falls, the contribution of the natural pH gradient also falls, undermining −ΔG. In contrast, the benefit of pumping increases, as fewer protons return through the lipid phase. The balance between these two factors depends on the strength of pumping (which equates to the number of pumps, i.e., % surface area). However, even when the pump occupies 5% of the membrane surface area, pumping H+ gives no advantage until a modern permeability of 10−5 cm/s, i.e., there is no benefit to improving permeability across 1,000-fold (Figure 4A). Thus, there is no selective pressure to drive either the origin of pumping or the evolution of modern proton-tight membrane lipids in natural proton gradients. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 4. Pumping H+ or Na+ does not offer a sustained selective advantage. (A) Pumping H+ in a membrane with 1% ATPase causes a sustained loss in −ΔG as membrane permeability decreases with 1% pump. Even with 5% pump, −ΔG does not change over 3 orders of magnitude, and pumping only improves −ΔG near modern membrane permeability (≤10−5 cm/s). (B) Pumping less-permeable Na+ is initially better, adding to the natural gradient, but the early benefit is lost as membranes become tighter, due to the collapse of the natural H+ gradient. In the absence of a gradient, pumping both H+ (C) and Na+ (D) offers a sustained advantage to tightening up membranes, but given a minimal requirement of around 15–20 kJ/mol to power aminoacyl adenylation, the energy attained is not sufficient to power intermediary biochemistry. https://doi.org/10.1371/journal.pbio.1001926.g004 Pumping Na+ works better across leaky membranes (Figure 4B), as lipid membranes are ∼6 orders of magnitude less permeable to Na+ than to H+ (due to fatty acid flip-flop; see Discussion) [34]. However, as with pumping H+, −ΔG falls as the membrane becomes less permeable, because the contribution of the natural gradient also declines, giving no continuous selective advantage to pumping Na+. With a proton permeability <10−5 cm/s, there is no advantage to pumping Na+ at a pump density of 1%–5% surface area compared with leaky protocells lacking a pump. Pumping Na+ therefore offers an initial advantage, but there is no sustained selection pressure for tightening membrane permeability to modern values. Neither is there any advantage in the absence of a natural pH gradient. This would apply to the evolution of chemiosmotic coupling in any setting that lacks natural gradients. Under this condition, pumping either H+ (Figure 4C) or Na+ (Figure 4D) offers a steadily amplifying advantage as membrane permeability falls. However, without an external pH gradient, −ΔG is low, the rise with reduced permeability is meager, and remains well below the 15–20 kJ/mol required by modern cells to drive processes like aminoacyl adenylation for protein synthesis [36]. Cells with permeable membranes (10−2–10−4 cm/s) are therefore unlikely to be viable unless powered by some other means [23],[27]. Hence in either the presence or absence of pH gradients, there is no sustained selection pressure to drive the evolution of either active pumping or modern membranes. Promiscuous H+/Na+ Bioenergetics Facilitates Spread and Is Prerequisite for Active Pumping Our model shows that leaky membranes were necessary to survive in natural proton gradients but that pumping protons across such leaky membranes is fruitless. Yet free-living cells require ion-tight membranes and active pumping for bioenergetics. What drove this evolutionary change? We hypothesize that a necessary first step was adding Na+ as an additional “promiscuous” coupling ion. A non-electrogenic sodium-proton (1Na+/1H+) antiporter (SPAP), found widely in cells, could in principle use a natural H+ gradient to generate a biochemical Na+ gradient. Exchanging Na+ for H+ does not alter membrane potential directly, but the difference in lipid permeability of the two ions alters ion flux, with significant effects on −ΔG. Because lipid membranes are ∼6 orders of magnitude less permeable to Na+ than to H+ [34], fewer Na+ ions can pass through the lipid phase of the membrane, so the Na+ gradient does not dissipate as quickly. As a result, Na+ flux becomes more tightly funneled through membrane proteins, improving the coupling of the membrane without changing its chemistry [19]. Because the H+ gradient is sustained geochemically, SPAP simply adds a Na+ gradient to the natural H+ gradient. Taking advantage of mixed Na+/H+ gradients requires promiscuity of membrane proteins for both ions, which is indeed the case for several contemporary bioenergetic proteins, including the ATPase [38] and Ech [26] (see Discussion). SPAP increases proton influx, initially lowering −ΔG (Figure 5A). However, the coupled extrusion of relatively impermeable Na+ ions increases −ΔG by ∼60% within minutes in a 7∶10 gradient, saturating when SPAP covers ∼5% of the membrane surface area (Figure 5A). Importantly, the free energy available from pH gradients declines in more acidic conditions. −ΔG is greatest with a 7∶10 gradient, lower at 6∶9, and nearly zero with a 5∶8 gradient, despite the three-order-of-magnitude correspondence (Figure 5B). This asymmetry arises because H+ and OH− flux through the membrane depends on concentrations as well as gradient size [39]. Comparatively high acidity and low alkalinity increases H+ influx but hinders OH− neutralization, collapsing the H+ gradient. Because Na+ extrusion through SPAP depends on the natural H+ gradient, SPAP increases −ΔG in relatively alkaline regions (pH 7–10 and 6–9) but has little effect on −ΔG in more acidic regions (pH 5–8), making acidic regions less favorable for colonization, even with SPAP. When the rate of H+ influx does not collapse the proton gradient, SPAP significantly increases −ΔG, allowing survival in shallower pH gradients (Figure 5C). If a −ΔG>15 kJ/mol is needed for growth, 5%–10% SPAP allows cells to grow in 50-fold weaker gradients (e.g., 8.5∶10; Figure 5C), a significant ecological advantage, facilitating spread. This general principle holds whatever the actual value of −ΔG needed for growth in early cells. The advantage offered by SPAP also applies to fluctuations in gradient size (e.g., due to mixing of fluids). −ΔG plainly fluctuates with the pH front even in the presence of SPAP; but SPAP still increases −ΔG even with considerable fluctuations in pH (Figures S3 and S4). Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 5. SPAP significantly increases free energy. (A) Because external Na+ concentration (0.4 M) is higher than H+ concentration (10−7 M), SPAP initially collapses −ΔG, and it takes minutes for the 1∶1 H+∶Na+ exchange to increase −ΔG; eventually it renders an increase of ∼60%. (B) The greatest increases are attained in relatively alkaline pH 7∶10 environments, saturating as % surface area rises. Despite equivalent gradient sizes, the absolute difference in H+ and OH− concentrations means a 6∶9 gradient gives a lower −ΔG, as the rate of H+ influx is greater while neutralizing OH− influx is lower. A 5∶8 gradient undermines −ΔG further, with or without SPAP. (C) SPAP facilitates colonization of environments with weaker proton gradients. 1% SPAP pushes −ΔG above 20 kJ/mol in a 7.5∶10 gradient, whereas 10% SPAP salvages an otherwise unviable 8∶10 gradient. All simulations with 1% promiscuous ATPase, no pump, no Ech, and H+ permeability 10−3 cm/s. https://doi.org/10.1371/journal.pbio.1001926.g005 Crucially, SPAP is also a necessary preadaptation for the active pumping of protons, and for decreasing membrane permeability towards modern values. Whereas pumping H+ in the absence of SPAP gives no sustained benefit in terms of −ΔG, the presence of SPAP in a leaky membrane allows pumping of H+ to pay dividends. −ΔG now markedly increases with decreasing permeability (Figure 6A), for the first time giving a sustained selective advantage to higher levels of pumping and tighter membranes. As in the absence of SPAP, −ΔG depends on two factors: the power of the pump (which varies with the proportion of surface area covered) and the natural pH gradient. As membrane permeability falls, the contribution of the natural pH gradient also falls. While 1% pump cannot sustain −ΔG when the contribution of the gradient is lost, 5% H+ pump gives a steadily amplifying advantage to lowering membrane permeability (Figure 6A). Much the same applies to pumping Na+ (Figure 6B). The lower permeability of Na+ gives an initial benefit to pumping this ion, but this is lost as the membrane becomes tighter, even with 5% pump (Figure 6B). This lower efficacy is due to the much higher external concentration of Na+. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 6. SPAP gives a sustained benefit to pumping favoring tighter membranes and allowing free living. (A) The combination of SPAP with 5% H+ pump gives a sustained increase in −ΔG as membrane permeability decreases, for the first time favoring the evolution of modern proton-tight phospholipid membranes. In contrast, 1% H+ pump gives an initial benefit, but provides insufficient power to sustain −ΔG as the gradient is lost with decreasing permeability. (B) The combination of SPAP with both 1% and 5% Na+ pump provides an initial benefit, but neither provides enough power to sustain −ΔG with decreasing permeability. (C) SPAP facilitates colonization of smaller gradients, ultimately making it possible to survive, after the evolution of tight membranes, in the total absence of a gradient (D); cells could not survive without a gradient unless relatively proton-tight membranes were already in place, as −ΔG falls well below the 15–20 kJ/mol threshold upon losing the gradient with a leaky membrane. All simulations assume 1% SPAP. Legend in (B) is common to all panels. https://doi.org/10.1371/journal.pbio.1001926.g006 With active pumping, tighter membranes, and SPAP, cells could colonize more acidic regions (Figure S5), regions with weaker gradients (Figure 6C), and ultimately survive in the absence of a gradient altogether (Figure 6D). With no external pH gradient, SPAP interconverts efficiently between H+ and Na+, making it feasible to pump either ion (Figure 6D). These cells are now modern in that they have a fully functional chemiosmotic circuit and proton-tight membranes, and hence could evolve the traits required to leave the natural gradients for the external world. We propose that this process occurred independently in divergent populations that had spread widely using SPAP to colonize regions with weak gradients (see Discussion). These independent populations subsequently evolved into the two main branches of early life, the archaea and bacteria [1]. Free-Energy Availability Depends on Membrane Permeability The model shows that cells with 1% ATPase in a proton-tight membrane with glycerol-phosphate headgroups (giving an H+ permeability <10−5 cm/s, like extant archaea and bacteria [34]), collapse natural proton gradients within seconds (Figure 3A and 3B). The magnitude of the pH gradient depends on the environmental setting. To constrain possibilities we considered pH values commensurate with alkaline hydrothermal vents, but the same principles apply to any other setting with dynamic pH gradients across short distances. The early oceans may have been mildly acidic, as low as pH 5, and alkaline fluids as high as pH 11 [35] but we conservatively set a 3 pH-unit gradient, with the “acid” at pH 7 and alkaline fluids at pH 10. Nonetheless, collapse of the gradient was evident in proton-tight membranes across a range of gradients (Figure 3B). Protons enter through the ATPase faster than they can exit or be neutralized by OH−, so H+ influx rapidly reaches electrochemical equilibrium. In contrast, leaky protocells (equivalent to fatty-acid vesicles without glycerol phosphate headgroups) in a 7∶10 pH gradient with 1% ATPase in the membrane retain nearly all the free energy available, having a −ΔG only ∼17% lower than an open system (i.e., a single membrane containing the same number of membrane proteins, separating a continuous flux of acid and alkaline fluids; Figure 3A). This is because proton flux through the ATPase is ∼4 orders of magnitude faster than through the lipid phase, even with a high proton permeability of 10−2 cm/s (based on the kinetics of proton-flux through the ATPase, see Materials and Methods and Table S1). Leaky cells in natural proton gradients of 3 pH units therefore have sufficient free energy to drive ATP synthesis. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 3. Dynamics of free-energy change (−ΔG) in cells powered by natural proton gradients. (A) Proton-permeable vesicles (≥10−4 cm/s) have only a small loss of free-energy compared with an open system (pH gradient 7∶10, 1% ATPase). Reduced membrane permeability (≤10−4 cm/s), including permeabilities equivalent to modern membranes (<10−5 cm/s), collapse the gradient within seconds. (B) At low permeability (10−6 cm/s), −ΔG collapses regardless of gradient size. Within seconds, H+ flux through ATPase equilibrates with the acidic fluids. (C) The collapse of −ΔG is more extensive the greater the amount of membrane-bound ATPase, even with a leaky membrane (10−3 cm/s). (D) With Ech, the collapse of the natural gradient is similar to that of the ATPase, showing that natural proton gradients can power energy (ATPase) and carbon (Ech) metabolism, given 1%–5% enzyme in membrane. Na+ permeability was kept 6 orders of magnitude higher than that of H+ throughout all simulations in this and all figures of the article. Except in (B), all results were calculated in a pH gradient 7∶10. https://doi.org/10.1371/journal.pbio.1001926.g003 Even leaky cells are sensitive to the amount of membrane protein, with higher proportions of ATPase collapsing the gradient (Figure 3C). In this case, the rate of H+ entry through ATPase covering 10%–50% of the membrane surface area is substantially faster than the rate of clearance of H+ from inside the cell (and reaction with OH−), collapsing −ΔG. However, 1%–5% ATPase in a leaky membrane (10−3 cm/s) retains a −ΔG of close to 20 kJ/mol (Figure 3A and 3C). With 3–4 protons translocated per ATP synthesized (Table S1), this gives a −ΔG for ATP hydrolysis of 60 to 80 kJ/mol, similar to modern cells and sufficient to drive intermediary biochemistry, including aminoacyl adenylation in protein synthesis [36]. This assumes the same stoichiometry as the modern ATPase (3–4 protons per ATP). Because the kinetics of early enzymes would arguably not have been as honed by evolution as their modern equivalents, we used 10% of modern proton flux rates. However, this difference in efficiency actually has limited impact on the model compared with modern flux rates (Figure S1); increasing the stoichiometry of the ATPase has a similarly small effect (Figure S2). We did not estimate rates of ATP synthesis, as that would require additional assumptions about concentrations of ATP, ADP, and phosphate, as well as the rates of ATP consumption and growth; these are almost impossible to constrain at present. The same principles apply to carbon metabolism. We consider whether the membrane protein Ech could drive carbon reduction by H2 in natural proton gradients. Ech uses the proton-motive force to drive carbon metabolism in some archaea and bacteria via the reduction of ferredoxin [26]. As with the ATPase, cells with 1%–5% Ech in the membrane retain most of the free energy available from a 7∶10 pH gradient (Figure 3D). Higher concentrations of Ech (10%–50%) collapse −ΔG even more than the ATPase, as the rate of proton flux through Ech is double that of the ATPase, and its surface area is slightly smaller, so there are more proton pores per unit surface area (Table S1). Such high concentrations of Ech or ATPase are in any case improbable, and not relevant to modern cells, but demonstrate the range of conditions in which natural gradients can in principle drive carbon and energy metabolism. Given a 7∶10 pH gradient, it is therefore feasible to have 1%–5% Ech and 1%–5% ATPase in the membrane, driving both carbon and energy metabolism in cells with leaky membranes. But incorporation of either G1P or G3P glycerol-phosphate headgroups (found in archaea and bacteria respectively), or racemic mixtures of archaeal and bacterial lipids (which, surprisingly, are as impermeable to protons as standard membranes [37]), are not favored because they reduce the proton permeability of the membrane and so collapse the energetic driving force. Glycerol-phosphate headgroups in particular decrease proton permeability, as they prevent fatty acid flip-flop across the membrane (see Discussion). Pumping Ions across Leaky Membranes Does Not Give a Sustained Increase in Free Energy If leaky cells with low amounts of ATPase and Ech (1%–5%) are viable in natural proton gradients, but cells with phospholipid membranes are not, then the evolution of active pumping becomes a paradox: pumping protons across a proton-permeable membrane does not increase free energy (−ΔG), because the protons immediately return through the lipid phase of the membrane. We demonstrate this using a model of a simple H2-dependent proton pump (equivalent to Ech operating in reverse, as found in some simple bacteria and archaea [26]). We find that in a 7∶10 pH gradient −ΔG falls as membrane permeability decreases from 10−2 to 10−6 cm/s (Figure 4A). −ΔG here depends on two factors: active pumping and the natural pH gradient. As membrane permeability falls, the contribution of the natural pH gradient also falls, undermining −ΔG. In contrast, the benefit of pumping increases, as fewer protons return through the lipid phase. The balance between these two factors depends on the strength of pumping (which equates to the number of pumps, i.e., % surface area). However, even when the pump occupies 5% of the membrane surface area, pumping H+ gives no advantage until a modern permeability of 10−5 cm/s, i.e., there is no benefit to improving permeability across 1,000-fold (Figure 4A). Thus, there is no selective pressure to drive either the origin of pumping or the evolution of modern proton-tight membrane lipids in natural proton gradients. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 4. Pumping H+ or Na+ does not offer a sustained selective advantage. (A) Pumping H+ in a membrane with 1% ATPase causes a sustained loss in −ΔG as membrane permeability decreases with 1% pump. Even with 5% pump, −ΔG does not change over 3 orders of magnitude, and pumping only improves −ΔG near modern membrane permeability (≤10−5 cm/s). (B) Pumping less-permeable Na+ is initially better, adding to the natural gradient, but the early benefit is lost as membranes become tighter, due to the collapse of the natural H+ gradient. In the absence of a gradient, pumping both H+ (C) and Na+ (D) offers a sustained advantage to tightening up membranes, but given a minimal requirement of around 15–20 kJ/mol to power aminoacyl adenylation, the energy attained is not sufficient to power intermediary biochemistry. https://doi.org/10.1371/journal.pbio.1001926.g004 Pumping Na+ works better across leaky membranes (Figure 4B), as lipid membranes are ∼6 orders of magnitude less permeable to Na+ than to H+ (due to fatty acid flip-flop; see Discussion) [34]. However, as with pumping H+, −ΔG falls as the membrane becomes less permeable, because the contribution of the natural gradient also declines, giving no continuous selective advantage to pumping Na+. With a proton permeability <10−5 cm/s, there is no advantage to pumping Na+ at a pump density of 1%–5% surface area compared with leaky protocells lacking a pump. Pumping Na+ therefore offers an initial advantage, but there is no sustained selection pressure for tightening membrane permeability to modern values. Neither is there any advantage in the absence of a natural pH gradient. This would apply to the evolution of chemiosmotic coupling in any setting that lacks natural gradients. Under this condition, pumping either H+ (Figure 4C) or Na+ (Figure 4D) offers a steadily amplifying advantage as membrane permeability falls. However, without an external pH gradient, −ΔG is low, the rise with reduced permeability is meager, and remains well below the 15–20 kJ/mol required by modern cells to drive processes like aminoacyl adenylation for protein synthesis [36]. Cells with permeable membranes (10−2–10−4 cm/s) are therefore unlikely to be viable unless powered by some other means [23],[27]. Hence in either the presence or absence of pH gradients, there is no sustained selection pressure to drive the evolution of either active pumping or modern membranes. Promiscuous H+/Na+ Bioenergetics Facilitates Spread and Is Prerequisite for Active Pumping Our model shows that leaky membranes were necessary to survive in natural proton gradients but that pumping protons across such leaky membranes is fruitless. Yet free-living cells require ion-tight membranes and active pumping for bioenergetics. What drove this evolutionary change? We hypothesize that a necessary first step was adding Na+ as an additional “promiscuous” coupling ion. A non-electrogenic sodium-proton (1Na+/1H+) antiporter (SPAP), found widely in cells, could in principle use a natural H+ gradient to generate a biochemical Na+ gradient. Exchanging Na+ for H+ does not alter membrane potential directly, but the difference in lipid permeability of the two ions alters ion flux, with significant effects on −ΔG. Because lipid membranes are ∼6 orders of magnitude less permeable to Na+ than to H+ [34], fewer Na+ ions can pass through the lipid phase of the membrane, so the Na+ gradient does not dissipate as quickly. As a result, Na+ flux becomes more tightly funneled through membrane proteins, improving the coupling of the membrane without changing its chemistry [19]. Because the H+ gradient is sustained geochemically, SPAP simply adds a Na+ gradient to the natural H+ gradient. Taking advantage of mixed Na+/H+ gradients requires promiscuity of membrane proteins for both ions, which is indeed the case for several contemporary bioenergetic proteins, including the ATPase [38] and Ech [26] (see Discussion). SPAP increases proton influx, initially lowering −ΔG (Figure 5A). However, the coupled extrusion of relatively impermeable Na+ ions increases −ΔG by ∼60% within minutes in a 7∶10 gradient, saturating when SPAP covers ∼5% of the membrane surface area (Figure 5A). Importantly, the free energy available from pH gradients declines in more acidic conditions. −ΔG is greatest with a 7∶10 gradient, lower at 6∶9, and nearly zero with a 5∶8 gradient, despite the three-order-of-magnitude correspondence (Figure 5B). This asymmetry arises because H+ and OH− flux through the membrane depends on concentrations as well as gradient size [39]. Comparatively high acidity and low alkalinity increases H+ influx but hinders OH− neutralization, collapsing the H+ gradient. Because Na+ extrusion through SPAP depends on the natural H+ gradient, SPAP increases −ΔG in relatively alkaline regions (pH 7–10 and 6–9) but has little effect on −ΔG in more acidic regions (pH 5–8), making acidic regions less favorable for colonization, even with SPAP. When the rate of H+ influx does not collapse the proton gradient, SPAP significantly increases −ΔG, allowing survival in shallower pH gradients (Figure 5C). If a −ΔG>15 kJ/mol is needed for growth, 5%–10% SPAP allows cells to grow in 50-fold weaker gradients (e.g., 8.5∶10; Figure 5C), a significant ecological advantage, facilitating spread. This general principle holds whatever the actual value of −ΔG needed for growth in early cells. The advantage offered by SPAP also applies to fluctuations in gradient size (e.g., due to mixing of fluids). −ΔG plainly fluctuates with the pH front even in the presence of SPAP; but SPAP still increases −ΔG even with considerable fluctuations in pH (Figures S3 and S4). Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 5. SPAP significantly increases free energy. (A) Because external Na+ concentration (0.4 M) is higher than H+ concentration (10−7 M), SPAP initially collapses −ΔG, and it takes minutes for the 1∶1 H+∶Na+ exchange to increase −ΔG; eventually it renders an increase of ∼60%. (B) The greatest increases are attained in relatively alkaline pH 7∶10 environments, saturating as % surface area rises. Despite equivalent gradient sizes, the absolute difference in H+ and OH− concentrations means a 6∶9 gradient gives a lower −ΔG, as the rate of H+ influx is greater while neutralizing OH− influx is lower. A 5∶8 gradient undermines −ΔG further, with or without SPAP. (C) SPAP facilitates colonization of environments with weaker proton gradients. 1% SPAP pushes −ΔG above 20 kJ/mol in a 7.5∶10 gradient, whereas 10% SPAP salvages an otherwise unviable 8∶10 gradient. All simulations with 1% promiscuous ATPase, no pump, no Ech, and H+ permeability 10−3 cm/s. https://doi.org/10.1371/journal.pbio.1001926.g005 Crucially, SPAP is also a necessary preadaptation for the active pumping of protons, and for decreasing membrane permeability towards modern values. Whereas pumping H+ in the absence of SPAP gives no sustained benefit in terms of −ΔG, the presence of SPAP in a leaky membrane allows pumping of H+ to pay dividends. −ΔG now markedly increases with decreasing permeability (Figure 6A), for the first time giving a sustained selective advantage to higher levels of pumping and tighter membranes. As in the absence of SPAP, −ΔG depends on two factors: the power of the pump (which varies with the proportion of surface area covered) and the natural pH gradient. As membrane permeability falls, the contribution of the natural pH gradient also falls. While 1% pump cannot sustain −ΔG when the contribution of the gradient is lost, 5% H+ pump gives a steadily amplifying advantage to lowering membrane permeability (Figure 6A). Much the same applies to pumping Na+ (Figure 6B). The lower permeability of Na+ gives an initial benefit to pumping this ion, but this is lost as the membrane becomes tighter, even with 5% pump (Figure 6B). This lower efficacy is due to the much higher external concentration of Na+. Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 6. SPAP gives a sustained benefit to pumping favoring tighter membranes and allowing free living. (A) The combination of SPAP with 5% H+ pump gives a sustained increase in −ΔG as membrane permeability decreases, for the first time favoring the evolution of modern proton-tight phospholipid membranes. In contrast, 1% H+ pump gives an initial benefit, but provides insufficient power to sustain −ΔG as the gradient is lost with decreasing permeability. (B) The combination of SPAP with both 1% and 5% Na+ pump provides an initial benefit, but neither provides enough power to sustain −ΔG with decreasing permeability. (C) SPAP facilitates colonization of smaller gradients, ultimately making it possible to survive, after the evolution of tight membranes, in the total absence of a gradient (D); cells could not survive without a gradient unless relatively proton-tight membranes were already in place, as −ΔG falls well below the 15–20 kJ/mol threshold upon losing the gradient with a leaky membrane. All simulations assume 1% SPAP. Legend in (B) is common to all panels. https://doi.org/10.1371/journal.pbio.1001926.g006 With active pumping, tighter membranes, and SPAP, cells could colonize more acidic regions (Figure S5), regions with weaker gradients (Figure 6C), and ultimately survive in the absence of a gradient altogether (Figure 6D). With no external pH gradient, SPAP interconverts efficiently between H+ and Na+, making it feasible to pump either ion (Figure 6D). These cells are now modern in that they have a fully functional chemiosmotic circuit and proton-tight membranes, and hence could evolve the traits required to leave the natural gradients for the external world. We propose that this process occurred independently in divergent populations that had spread widely using SPAP to colonize regions with weak gradients (see Discussion). These independent populations subsequently evolved into the two main branches of early life, the archaea and bacteria [1]. Discussion Our model suggests a resolution to the long-standing paradox that membrane bioenergetics are universal, but membranes are fundamentally different [19]. In so doing, the model gives a striking insight into the deep evolutionary split between archaea and bacteria. It reveals that the late and divergent evolution of impermeable membranes could have arisen as a simple outcome of LUCA's exploitation of natural proton gradients. Our model applies in principle to any environment in which sharp differences in proton concentration are sustained over short distances, one concrete example being alkaline hydrothermal vents [18],[24],[31]–[33]. Given the membrane proteins Ech and ATPase, we show that natural proton gradients could have sustained both carbon and energy metabolism in LUCA (Figure 3C and 3D). However, to do so, LUCA had to have very leaky membranes, the only way to avoid deadly electrochemical equilibrium (Figure 3A). Our results indicate that LUCA did not have modern phospholipids. The addition of glycerol-phosphate headgroups is specifically precluded by the requirement for high proton-permeability in natural gradients (Figure 3A). Addition of a glycerol-phosphate headgroup reduces proton permeability substantially, as the polar headgroup cannot cross the hydrophobic interior of the membrane [40]. In contrast, lipid membranes composed of mixed amphiphiles, including fatty acids, have much greater proton permeability, through “flip-flop.” In flip-flop, protonation of a negatively charged fatty acid eliminates its charge, allowing the neutral residue to migrate across the hydrophobic membrane to the inside [41]. Deprotonation on the relatively alkaline interior rapidly dissipates proton gradients, explaining the high proton permeability of fatty acid vesicles [41]. Flip-flop is not possible with Na+, which remains ionic in the presence of a negatively charged amphiphile, hence its lower permeability [34]. Our results indicate that LUCA was sophisticated in terms of genes and proteins, but did not have a modern phospholipid membrane. However, LUCA must have had a stable lipid bilayer membrane composed of mixed amphiphiles, probably including fatty acids and isoprenes (some of which are found in both archaea and bacteria [15]). A lipid bilayer membrane is undoubtedly necessary for the function of membrane proteins such as the ATPase and Ech [42]. The actual permeability of membranes is difficult to determine experimentally, as H+ permeability depends in part on the permeability of counter-ions, and therefore varies with the composition of solutions used in measurements. Values of phospholipid membrane H+ permeability range from 10−4 cm/s [43] to 10−10 cm/s [44],[45], with a consensus favoring a value of between 10−4 to 10−6 cm/s [34]. The H+ permeability of fatty acid vesicles is higher, in the range of 10−2 to 10−3 cm/s or even greater [41]. These values are for standard temperature, 25°C (298 K). Both H+ and Na+ permeability rise substantially with temperature, by approximately 1 order of magnitude for every 20°C increase between 20°C and 100°C, although the actual values depend on membrane composition [45]. The membrane permeability also depends on the kinetics of membrane proteins, which likewise vary with temperature. We have used standard temperature for enzyme kinetics. How these values would vary with temperature is difficult to estimate, as the kinetics of enzymes adapted to low temperatures would differ from those in thermophiles if placed in the same membrane at the same temperature. However, our simulations of efficiency and stoichiometry (Figures S1 and S2) suggest that the effect should be substantially less than that of lipid permeability. We are therefore confident that our results apply generally, despite these uncertainties. We stress that our argument relates to the principle of energy transduction in natural proton gradients, not to the specific values used for membrane permeability. The key point is that leaky membranes were essential to transduce natural proton gradients, and there was no advantage to be gained by the evolution of proton-tight phospholipid membranes, whether at low or high temperatures. This leads to a paradox. Pumping either H+ or Na+ over leaky membranes gives no sustained advantage when membrane permeability is lowered over 1,000-fold (Figure 4A and 4B). That precludes the evolution of either active ion pumps or modern proton-tight membranes in a LUCA dependent on natural proton gradients. We hypothesize that the evolution of a SPAP was the key innovation that favored the independent evolution of active ion pumps and phospholipid membranes in bacteria and archaea. SPAP has two major effects that made this possible. First, SPAP favors divergence, through adding a Na+ gradient to the geochemically sustained H+ gradient. Because lipid membranes are much less permeable to Na+ ions, these preferentially flow back through membrane proteins, thereby increasing free-energy availability by up to 60% (Figure 5A). For this additional Na+ gradient to be useful, membrane proteins must be promiscuous for Na+ and H+, which is the case for some primitive ATPase enzymes [38] and for Ech [26]. While the ATPase generally specializes either for H+ or Na+ today, only a few amino acid changes are required to switch from one form to the other [46]. Phylogenetic trees of the ATPase suggest that the H+-dependent and Na+-dependent forms are interleaved, implying greater promiscuity in early evolution [24]. The reason probably relates to the close similarity in ionic radius and charge of Na+ without its hydration shell (the form in which it usually passes through membrane proteins) and the hydronium ion, H3O+ (the form in which H+ is most commonly found in solution). Thus it is likely that addition of a Na+ gradient to a natural H+ gradient by SPAP would indeed increase the free energy available to the cell as a usable electrochemical difference. This enabled cells to survive in 50-fold lower gradients (Figure 5C), or with intermittent gradients and mixing (Figures S3 and S4), facilitating spread and divergence. Second, SPAP gives a continuous selective advantage to actively pumping protons even across a leaky membrane (Figure 6A). This advantage amplifies steadily as membrane permeability decreases, all the way towards values for largely impermeable modern membranes (Figure 6A). Our results lead us to suggest that the SPAP is ancestral and must have been present in LUCA. Phylogenetic analysis is consistent with this prediction. BLAST [47] results show a match for archaeon Methanocaldococcus jannaschii's Mj1275 SPAP to an equivalent or very closely related protein in at least one member of 35 out of all 37 prokaryotic phyla reported to date (Table S2). The two bacterial clades with a missing match are to date single-member phyla whose only known species may have either lost the gene over time, had it diverge beyond observable similarity to the M. jannaschii ortholog, or simply have not been fully annotated in the databases yet. This confirms our prediction of the universality of SPAP in spite of the stark dissimilarity in membranes, and paves the way for closer phylogenetic analysis of these antiporters and related proteins. We note that the early operation of SPAP would have the effect of lowering the intracellular Na+ concentration substantially below ambient seawater concentration, explaining how cells that evolved in the ocean could nonetheless be optimized to low intracellular Na+ and high K+ concentration. The operation of antiporters (and possibly symporters), driven by natural proton gradients, could in principle have modulated intracellular ionic composition to the low-Na+–high-K+ characteristic of most modern cells, leading to selective optimization of protein function without the need for a specific terrestrial environment with a particular ionic balance [27]. These considerations are also consistent with the universality of SPAP across prokaryotic phyla. Our analysis demonstrates that active ion pumps almost certainly arose after SPAP, and only then did selection favor the evolution of ion-tight membranes with glycerol phosphate headgroups. Given that SPAP in itself facilitated the spread and colonization of regions with shallower (Figure 5C) or more intermittent gradients (Figures S3 and S4), pumping is expected to arise independently in more than one population, as observed [13],[19]. Only when active ion pumping had evolved was there any benefit to incorporating glycerol-phosphate headgroups, thereby reducing membrane permeability (Figure 6A). Phospholipid biosynthesis involves nucleophilic attack on the prochiral carbonyl center of dihydroxyacetone phosphate [10]. This can be achieved from either side of the molecule, giving rise to opposite stereochemistries of the central carbon in glycerol phosphate (Figure 1). The enzymes involved, G1PDH in archaea and G3PDH in bacteria appear to have taken these alternatives by chance in independent populations that had already evolved distinct ion pumps. Thus we posit that the ancestors of archaea and bacteria evolved both ion pumps and phospholipid membranes independently, the latter on the basis of a simple binary choice in the orientation of nucleophilic attack on dihydroxyacetone phosphate. We conclude that the membranes of LUCA were necessarily leaky, composed of mixed amphiphiles (including fatty acids) but lacking glycerol-phosphate headgroups. Fatty-acid vesicles have long been considered plausible protocells because of their simplicity, stability, and dynamic ability to grow [48]–[50], but are generally thought unsuitable for chemiosmotic coupling due to their high proton permeability [27],[51]. Leaky membranes have therefore generally been interpreted in terms of heterotrophic origins of life [52]. In contrast, we find that high proton permeability was in fact indispensable to drive both carbon and energy metabolism in natural proton gradients, consistent with autotrophic origins; and this requirement for leaky membranes in turn precluded the early evolution of phospholipid membranes (Figure 7). Our model offers a selective basis for the universality of membrane bioenergetics and the ATPase, while elucidating the paradoxical differences in membranes and active ion pumps. The deep disparity between archaea and bacteria in carbon and energy metabolism [19],[53], and in membrane lipid stereochemistry [10], reflects two independent origins of active pumping in divergent populations (Figure 7). Download: PPT PowerPoint slide PNG larger image TIFF original image Figure 7. Divergence of archaea and bacteria. (A) Ions cross the membrane in response to concentration gradients and electrical potential. OH− neutralizes incoming protons. The H+ gradient can drive energy metabolism via ATPase, and carbon metabolism via Ech (not shown). (B) SPAP generates a Na+ gradient from the H+ gradient. As Na+ is less permeable than H+, SPAP improves coupling, given promiscuity of membrane proteins for H+ and Na+. (C) Membrane pumps generate gradients by extruding H+ or Na+ ions. (D) Exploiting natural gradients demands high membrane permeability, but pumping with SPAP drives the evolution of tighter membranes, facilitating colonization of less alkaline environments. (E) Impermeable membranes funnel ion flow through bioenergetic proteins, independent of natural gradients. (F) From bottom up, SPAP favors divergence, selection for active pumping and tighter membranes. Pumping and phospholipid membranes arose independently in archaea and bacteria. https://doi.org/10.1371/journal.pbio.1001926.g007 The conclusion that LUCA had leaky membranes, and that modern phospholipid membranes evolved later and independently in archaea and bacteria, provides a framework for interpreting other dichotomies between archaea and bacteria. The late and independent evolution of glycolysis but not gluconeogenesis [12] is entirely consistent with LUCA being powered by natural proton gradients across leaky membranes. Several discordant traits are likely to be linked to the late evolution of cell membranes, notably the cell wall, whose synthesis depends on the membrane [11] and DNA replication [14]. In the latter case, the fingers-thumb-palm motif at the active site of DNA polymerase enzymes [54] and the structure of the replication fork [55] are superficially similar in archaea and bacteria, yet most proteins involved in DNA replication, including the principal replicative polymerases, bear no phylogenetic resemblance [14],[56],[57]. That implies either independent origins [14] or inscrutably deep divergence compared with the plainly homologous transcription and translation machinery [56],[57]. Because the bacterial replicon is attached to the plasma membrane during cell division [58]–[60], this complex presumably arose after (or coevolved with) the bacterial membrane, which must have driven a deep phylogenetic disparity, even if DNA replication had arisen in LUCA. Thus key facets of the fundamental split between archaea and bacteria could be linked to the late origin of phospholipid membranes, for these bioenergetic reasons. While it is difficult to prove that these bioenergetic factors really did account for the deepest branch in the tree of life, they do offer a robust and testable framework that can explain the paradoxical character of LUCA and the stark differences between archaea and bacteria. Materials and Methods General Description of the Model Cells were modeled half embedded in the alkaline fluid, with the other half exposed to the comparatively acidic fluid. This produced an inward proton gradient from the acidic side, sustained by the constant neutralization with OH− from the alkaline side (Figure 2). Only the two external pH values are fixed; the internal pH is then arrived at in response to the fluxes of H+ and OH− across the membrane, which in turn depends on permeability, the respective concentrations of each ion, and flow through the membrane proteins. Equation 1 describes the various ways in which protons could enter or leave the cell at every time step: by simple diffusion across the membrane on either side, and through any of the membrane proteins, namely the ATPase, SPAP, pump, or Ech.(1)Total concentrations of H+ and OH− were calculated at every time step by neutralization and equilibration to the dissociation constant of water. External fluids were assumed to be part of comparatively large bodies of water, with their acidity and alkalinity sustained by large-scale geological or meteorological processes; thus their concentrations of H+, OH−, and other ions were assumed constant. Analogous equations were used for other ions. Table S1 describes the parameters chosen for the results presented in the text, unless otherwise stated. We anticipate that enzymes could not have reached their current reaction rate values at the early stages of evolution that we are considering, so for the results presented in the main text we have consistently used 10% of the current turnover rates referenced in Table S1. A series of results using modern (100%) turnover rates are presented in Figure S1 for comparison. Flux through the Membrane Membrane flux JS of a neutral substance S was modeled using a traditional passive diffusion equation [61](2)where PS is the permeability of the substance, A is the area of the membrane, and [S]ext and [S]int are the external and internal concentrations respectively. To account for the effect of membrane potential Δψ on the behavior of charged particles, ion diffusion was modeled using the Goldman-Hodgkin-Katz flux equation [39],[62](3)where zs is the charge of the substance, F and R are the Faraday and gas constants, respectively, and T is the temperature. Electrical membrane potential Δψ was in turn modeled using the Goldman-Hodgkin-Katz voltage equation [39],[62](4)for the concentration of each cation and anion present. Internal protons and hydroxide were equilibrated using the dissociation constant of water. Free Energy (ΔG) Calculations The available free energy ΔG from the H+ gradient was modeled with the traditional equation used by Mitchell [20](5)An analogous equation was used for the Na+ gradient. The power of ATP to catalyze biochemical reactions in the cell comes not specifically from hydrolysis of the molecule itself but from the degree to which the ATP/ADP ratio is shifted from thermodynamic equilibrium; that is, the energy available from ATP hydrolysis varies with the ATP/ADP ratio [30]. The equilibrium constant and thus the energy required for ATP synthesis depends on the concentrations of ADP, phosphate, and magnesium ion, as well as pH [20],[30], but with the exception of pH these values are unknown for the systems modeled, as are rates of ATP hydrolysis. We have therefore used Equation 5 to calculate the size of the electrochemical gradient (ΔG) as a function of the H+ and Na+ gradients and the electrical membrane potential (Δψ). The steady-state ΔG in turn gives an indication of how far from equilibrium the ATP/ADP ratio could be pushed. With 3–4 protons translocated per ATP, a steady-state ΔG of −20 kJ/mol is large enough to drive the ATP/ADP ratio to a disequilibrium of 10 orders of magnitude, equivalent to that found in modern cells [30]. We calculated steady-state ΔG as a function of the size of the H+ and Na+ gradients and the electrical membrane potential (Δψ) between the acid fluid and the inside of the cell. These factors in turn depend on steady-state rates of proton flux into and out of the cell via the lipid phase of the membrane (specified by its H+ and Na+ permeability and surface area) and through the ATPase. We calculated the maximum flux of H+ or Na+ flux through the ATPase on the basis of the maximum possible number of ions translocated per second. Maximum ion flux is based on the reported maximum turnover rate of ATPase (Table S1), i.e., the maximum number of ATP molecules that each ATPase unit can synthesize in one second when operating at top speed, multiplied by 3.3, the number of H+ or Na+ required to synthesize 1 ATP (Table S1). This number was then multiplied by the number of ATPase units in the system, estimated from the membrane surface area assigned to this protein in each simulation (e.g., 1%, 5%, etc.) and the reported surface area of the membrane-integral FO subunit (Table S1). We further assumed that the actual flux rate of H+ and Na+ through the ATPase would also depend on the driving force itself, ΔG, i.e., the size of the H+/Na+ gradient and the electrical membrane potential (Δψ). We assumed that the ATPase obeys hyperbolic Michaelis-Menten dynamics, commonly the case in enzyme kinetics [63] and reported for the ATPase [64], such that H+/Na+ flux asymptotically approaches the maximum turnover rate when the driving force is large, again assuming that flux rate is unconstrained by ADP availability. Thus, increasing ΔG beyond a threshold cannot increase H+/Na+ flux beyond the maximum turnover rate, so flux rate must saturate. The hyperbolic curve was modeled to reach saturation slightly beyond −20 kJ/mol, a gradient large enough to drive the ATP/ADP ratio to 10 orders of magnitude disequilibrium in modern cells [30] and equivalent to a membrane potential of around 200 mV, close to a maximum for modern lipid membranes, given the low capacitance of thin lipid membranes. This number, between zero and one, was finally multiplied by the maximum flux of H+ or Na+, described above, to determine the influx of each of the two ions through the ATPase. When added to H+/Na+ flux rates across the lipid phase, the steady-state H+/Na+ flux through the ATPase gave a steady-state ΔG available to drive ATP synthesis. Full promiscuity of the ATPase to Na+ and H+ was assumed, with preference of one ion over the other depending solely on their respective gradient sizes. The Ech was modeled analogously. Modeling the Sodium-Proton Antiporter and Pump SPAP was modeled to respond to the H+ and Na+ gradients, exchanging ions in the direction determined by the larger of the two gradients. Δψ was assumed to affect SPAP speed but not direction [65]. Since the H+ gradient is reversed on the alkaline side, we assumed the SPAP, ATPase, and Ech operated only on the acidic side. The pump was modeled as a generic system able to extrude either H+ or Na+, dependent on the concentration of hydrogen gas (H2), and responding to the opposing gradient, thus making it easier to pump protons against an alkaline fluid, and more difficult against an acidic fluid. Source Code A running example of the code can be found at http://www.ucl.ac.uk/~rmhknjl/research/membranedivergence This code can be run directly from any typical computer with an Internet connection. Additionally, it can be downloaded and run locally (at no significant increase in speed) from http://github.com/UCL/membranedivergence BLAST Searches The primary amino acid sequence of the M. jannaschii Mj1275 Na+/H+ antiporter (SPAP) was obtained from the NCBI protein sequence database. Mj1275 is one of three known SPAP genes in archaeon M. jannaschii, the other two being Mj0057 and Mj1521 [66]. The first belongs to the NapA family, while the latter two are in the NhaP family. Phylogenetic analysis was performed on these three genes as well as the two common Escherichia coli SPAP genes, NhaA and NhaB [67],[68], using the NCBI-BLASTp server [47] with standard parameters, filtering for each prokaryotic phylum (considering each of the proteobacteria as a separate clade). Results for Mj1275 showed the highest hit rate (Table S2), possibly hinting that it is closest to the ancestral form of the SPAP. Results for the other genes are not shown. General Description of the Model Cells were modeled half embedded in the alkaline fluid, with the other half exposed to the comparatively acidic fluid. This produced an inward proton gradient from the acidic side, sustained by the constant neutralization with OH− from the alkaline side (Figure 2). Only the two external pH values are fixed; the internal pH is then arrived at in response to the fluxes of H+ and OH− across the membrane, which in turn depends on permeability, the respective concentrations of each ion, and flow through the membrane proteins. Equation 1 describes the various ways in which protons could enter or leave the cell at every time step: by simple diffusion across the membrane on either side, and through any of the membrane proteins, namely the ATPase, SPAP, pump, or Ech.(1)Total concentrations of H+ and OH− were calculated at every time step by neutralization and equilibration to the dissociation constant of water. External fluids were assumed to be part of comparatively large bodies of water, with their acidity and alkalinity sustained by large-scale geological or meteorological processes; thus their concentrations of H+, OH−, and other ions were assumed constant. Analogous equations were used for other ions. Table S1 describes the parameters chosen for the results presented in the text, unless otherwise stated. We anticipate that enzymes could not have reached their current reaction rate values at the early stages of evolution that we are considering, so for the results presented in the main text we have consistently used 10% of the current turnover rates referenced in Table S1. A series of results using modern (100%) turnover rates are presented in Figure S1 for comparison. Flux through the Membrane Membrane flux JS of a neutral substance S was modeled using a traditional passive diffusion equation [61](2)where PS is the permeability of the substance, A is the area of the membrane, and [S]ext and [S]int are the external and internal concentrations respectively. To account for the effect of membrane potential Δψ on the behavior of charged particles, ion diffusion was modeled using the Goldman-Hodgkin-Katz flux equation [39],[62](3)where zs is the charge of the substance, F and R are the Faraday and gas constants, respectively, and T is the temperature. Electrical membrane potential Δψ was in turn modeled using the Goldman-Hodgkin-Katz voltage equation [39],[62](4)for the concentration of each cation and anion present. Internal protons and hydroxide were equilibrated using the dissociation constant of water. Free Energy (ΔG) Calculations The available free energy ΔG from the H+ gradient was modeled with the traditional equation used by Mitchell [20](5)An analogous equation was used for the Na+ gradient. The power of ATP to catalyze biochemical reactions in the cell comes not specifically from hydrolysis of the molecule itself but from the degree to which the ATP/ADP ratio is shifted from thermodynamic equilibrium; that is, the energy available from ATP hydrolysis varies with the ATP/ADP ratio [30]. The equilibrium constant and thus the energy required for ATP synthesis depends on the concentrations of ADP, phosphate, and magnesium ion, as well as pH [20],[30], but with the exception of pH these values are unknown for the systems modeled, as are rates of ATP hydrolysis. We have therefore used Equation 5 to calculate the size of the electrochemical gradient (ΔG) as a function of the H+ and Na+ gradients and the electrical membrane potential (Δψ). The steady-state ΔG in turn gives an indication of how far from equilibrium the ATP/ADP ratio could be pushed. With 3–4 protons translocated per ATP, a steady-state ΔG of −20 kJ/mol is large enough to drive the ATP/ADP ratio to a disequilibrium of 10 orders of magnitude, equivalent to that found in modern cells [30]. We calculated steady-state ΔG as a function of the size of the H+ and Na+ gradients and the electrical membrane potential (Δψ) between the acid fluid and the inside of the cell. These factors in turn depend on steady-state rates of proton flux into and out of the cell via the lipid phase of the membrane (specified by its H+ and Na+ permeability and surface area) and through the ATPase. We calculated the maximum flux of H+ or Na+ flux through the ATPase on the basis of the maximum possible number of ions translocated per second. Maximum ion flux is based on the reported maximum turnover rate of ATPase (Table S1), i.e., the maximum number of ATP molecules that each ATPase unit can synthesize in one second when operating at top speed, multiplied by 3.3, the number of H+ or Na+ required to synthesize 1 ATP (Table S1). This number was then multiplied by the number of ATPase units in the system, estimated from the membrane surface area assigned to this protein in each simulation (e.g., 1%, 5%, etc.) and the reported surface area of the membrane-integral FO subunit (Table S1). We further assumed that the actual flux rate of H+ and Na+ through the ATPase would also depend on the driving force itself, ΔG, i.e., the size of the H+/Na+ gradient and the electrical membrane potential (Δψ). We assumed that the ATPase obeys hyperbolic Michaelis-Menten dynamics, commonly the case in enzyme kinetics [63] and reported for the ATPase [64], such that H+/Na+ flux asymptotically approaches the maximum turnover rate when the driving force is large, again assuming that flux rate is unconstrained by ADP availability. Thus, increasing ΔG beyond a threshold cannot increase H+/Na+ flux beyond the maximum turnover rate, so flux rate must saturate. The hyperbolic curve was modeled to reach saturation slightly beyond −20 kJ/mol, a gradient large enough to drive the ATP/ADP ratio to 10 orders of magnitude disequilibrium in modern cells [30] and equivalent to a membrane potential of around 200 mV, close to a maximum for modern lipid membranes, given the low capacitance of thin lipid membranes. This number, between zero and one, was finally multiplied by the maximum flux of H+ or Na+, described above, to determine the influx of each of the two ions through the ATPase. When added to H+/Na+ flux rates across the lipid phase, the steady-state H+/Na+ flux through the ATPase gave a steady-state ΔG available to drive ATP synthesis. Full promiscuity of the ATPase to Na+ and H+ was assumed, with preference of one ion over the other depending solely on their respective gradient sizes. The Ech was modeled analogously. Modeling the Sodium-Proton Antiporter and Pump SPAP was modeled to respond to the H+ and Na+ gradients, exchanging ions in the direction determined by the larger of the two gradients. Δψ was assumed to affect SPAP speed but not direction [65]. Since the H+ gradient is reversed on the alkaline side, we assumed the SPAP, ATPase, and Ech operated only on the acidic side. The pump was modeled as a generic system able to extrude either H+ or Na+, dependent on the concentration of hydrogen gas (H2), and responding to the opposing gradient, thus making it easier to pump protons against an alkaline fluid, and more difficult against an acidic fluid. Source Code A running example of the code can be found at http://www.ucl.ac.uk/~rmhknjl/research/membranedivergence This code can be run directly from any typical computer with an Internet connection. Additionally, it can be downloaded and run locally (at no significant increase in speed) from http://github.com/UCL/membranedivergence BLAST Searches The primary amino acid sequence of the M. jannaschii Mj1275 Na+/H+ antiporter (SPAP) was obtained from the NCBI protein sequence database. Mj1275 is one of three known SPAP genes in archaeon M. jannaschii, the other two being Mj0057 and Mj1521 [66]. The first belongs to the NapA family, while the latter two are in the NhaP family. Phylogenetic analysis was performed on these three genes as well as the two common Escherichia coli SPAP genes, NhaA and NhaB [67],[68], using the NCBI-BLASTp server [47] with standard parameters, filtering for each prokaryotic phylum (considering each of the proteobacteria as a separate clade). Results for Mj1275 showed the highest hit rate (Table S2), possibly hinting that it is closest to the ancestral form of the SPAP. Results for the other genes are not shown. Supporting Information Figure S1. Comparison of different enzyme turnover rates. We assume that membrane proteins in LUCA had lower turnover rates than those in modern archaea and bacteria. For all the results in the main text, turnover rates were modeled at 10% of modern values (see Table S1 for these values). The figure shows that with ATPase, SPAP, and pump, the behavior is similar when turnover is set at 10%, 50%, and 100% for each protein. Parameters: 5% pump, 1% ATPase, 1% SPAP, pH gradient 7∶10. https://doi.org/10.1371/journal.pbio.1001926.s001 (TIF) Figure S2. Effect of higher H+-to-ATP stoichiometry in the ATPase. Lowering the efficiency of the ATPase by increasing the number of H+ necessary to synthesize one ATP molecule has a minor effect on the simulation results. Almost halving efficiency to 6 H+ per ATP lowers −ΔG by less than 1%. https://doi.org/10.1371/journal.pbio.1001926.s002 (TIF) Figure S3. Effect of fluctuations in external acidic pH, while holding external alkaline pH constant at pH 10. We considered the effect of mixing, with alkaline fluids causing local fluctuations in the pH of the acidic side. These were taken to occur on a scale of seconds, causing meaningful perturbations to the pH gradient and −ΔG. (A) Increases in the pH of the acidic side shrink the exploitable gradient. (B) With 1% ATPase and no SPAP or pump in the membrane, pH fluctuations are followed swiftly by corresponding changes in −ΔG. Circles on the y axis show the −ΔG values at stasis at pHacidic 7. Histograms in (C) show the frequency distributions for the corresponding curves in (B), with the vertical lines denoting the values for stasis at pH 7 (solid black) and mean of the corresponding curve (dashed grey). (D) Although responses are somewhat slower, addition of 5% SPAP makes fluctuations more survivable by increasing power overall. (E) is analogous to (C). See Figure S4 for similar fluctuations in the alkaline side. https://doi.org/10.1371/journal.pbio.1001926.s003 (TIF) Figure S4. Effect of fluctuations in external alkaline pH, while holding external acidic pH constant at pH 7. Qualitatively similar behavior to that of Figure S3 was observed when fluctuations occur on the alkaline side. https://doi.org/10.1371/journal.pbio.1001926.s004 (TIF) Figure S5. Pumping in the presence of SPAP facilitates adaptation to more acidic regions. All three curves show a steady increase in −ΔG with 5% pump in equivalent pH gradients (each of 3 pH units) with decreasing membrane permeability. In relatively alkaline conditions (pH 7∶10 and 6∶9) the benefit of pumping increases with decreasing permeability, but is relatively modest. In more acidic environments (pH 5∶8) there is initially a relatively greater payback to pumping as membrane permeability decreases. The reason is that at high membrane permeability (10−2 cm/s) and relatively acidic pH (5∶8), there is a fast influx of H+ (from the acidic side) and a slow influx of OH− (from the alkaline side), leading to the collapse of −ΔG. Pumping across a very leaky membrane gives little benefit even with SPAP (−ΔG is very low). Lowering membrane permeability limits H+ influx and enhances the benefits of pumping, giving a greater relative benefit in acidic conditions (pH 5∶8). In contrast, with tight membranes (10−6 cm/s), cells are powered almost exclusively by their own pumps, with little contribution from the external gradient (−ΔG collapses in the absence of a pump; see Figure 3A and 3B). Cells in relatively alkaline (6∶9 and 7∶10) environments now gain slightly more from pumping. The reason is that the opposing external H+ concentration is greater at pH 5∶8, so pumping H+ out is harder than at pH 6∶9 or 7∶10. The figure thus shows a transition from a highly permeable gradient-powered system on the left to a low permeability pump-powered system on the right. https://doi.org/10.1371/journal.pbio.1001926.s005 (TIF) Table S1. Parameters in the model and references. https://doi.org/10.1371/journal.pbio.1001926.s006 (DOC) Table S2. BLAST-search results for matches of the archaeal M. jannaschii Mj1275 SPAP to at least one member of each of the 37 known prokaryotic phyla. https://doi.org/10.1371/journal.pbio.1001926.s007 (DOC) Acknowledgments We are grateful to Bill Martin, Mike Russell, Peter Rich, Frank Harold, Ian Booth, and Don Braben for their stimulating discussions and comments on the manuscript.