PLoS Biology

doi: 10.1371/journal.pbio.2001259pmid: 27932848

Introduction Translational research is often afflicted by a fundamental problem: a limited understanding of disease mechanisms prevents effective targeting of new treatments [1]. The linear commercialization model, led by large pharmaceutical companies and sponsored research projects [2], is an increasingly outdated and ineffective approach to remedying the problem [3–5]. Instead, biomedical stakeholders are implementing dispersed, network-based approaches to innovation [6,7] and emulating effective models from other high-tech industries [8,9]. Nowhere is this of more importance than in the field of neurodegenerative and neuropsychiatric disease [1]. Various “open” research models (re-)emerged over the past several decades to help accelerate research progress, with varying degrees of success [7,10]. Open Science at an institutional level is one such model. It represents both a “nuanced approach to dissemination of university knowledge” [4] sought by Nicol et al. and an evolution of the role of universities in the innovation system [11]. Seeking to accelerate research advances and reimagine its role in the community, the Montreal Neurological Institute (Neuro) announced in the spring of 2016 that it is launching a five-year experiment during which it will adopt Open Science across the institution, including all of its labs. This article explores the potential of Open Science in general, and at the Neuro and in Montreal in particular. Given the experimental nature of the initiative, the Neuro is funding projects to independently measure, compare, and assess its own performance and that of its Open Science initiative. While the Neuro’s story is unique in that it is the first institution to adopt an open science model across the entire spectrum of its research, which includes clinical work, it will also provide a window into the future applications of Open Science more generally. In adopting Open Science at the institutional level, the Neuro hopes to achieve benefits beyond that for research; the initiative provides a foundation for multiple parties—researchers in and outside of McGill, patient organizations, regulators, and industry—to engage in neurobiological research and to engage in local “knowledge-based economic development” [11]. These benefits are expected to extend beyond the institution itself into the wider community. Open Science provides advantages over proprietary models of innovation in this respect by enhancing partnerships, lowering transaction costs, and encouraging local innovation where the subject matter of the innovation contains significant degrees of “sticky” (knowledge that is less expensive to use in place than to move elsewhere) [12], including tacit, knowledge. Definition The term Open Science is the culmination of the last decade’s embrace of “open innovation,” “open access,” and “open data” [7]. By adding the eschewal of patent protection in addition to the features of these models, it better accords with the general public’s conception of the word “open.” The Open Science model aims at accelerating discovery, innovation, and research by encouraging rapid multilateral sharing of data, ideas, and materials without the limitations imposed by patent protection. Institutions adopting an Open Science platform seek to break down barriers between researchers, datasets, and partners to create dynamic knowledge hubs and eliminate artificial bottlenecks imposed on upstream research. The Neuro expects that partners will include researchers around the globe, other research institutions, commercial entities, and patients themselves. The Open Science model allows the “ingredients for successful innovation—skilled individuals, resources, and financing—to come together.” [5] The Neuro’s Open Science model rests on principles that encompass a pledge to not seek patent rights over any of its research outputs at the institutional level—it currently applies for approximately five patents per year—while respecting each individual researcher’s independence to do so at her own expense. The model also promises open sharing of results, with the exception of clinical data supporting a regulatory application in line with the US Institute of Medicine’s recommendation [13]. Researchers will also be able to access associated metadata, physical biosamples, including through a soon-to-be established Neuro biobank, and other scientific materials. Drawing on tools already used by the McConnell Brain Imaging Centre, which currently has over 30,000 registered users worldwide and that is involved in international collaborations such as BigBrain (https://www.mcgill.ca/bic/home), the Neuro will develop a sharing infrastructure. The principles underlying the Open Science initiative recognize, however, that patient privacy and consent may, in some circumstances, limit what the Neuro shares. Finally, the Neuro expects its partners to uphold the same open principles in relation to the work they do directly with the Neuro (as opposed to their own follow-on or concurrent research). The Neuro possesses unique characteristics that provide a particularly rich environment in which to test the benefits and weigh the costs of Open Science at the institutional level. These include housing clinicians, basic research, and sophisticated brain imaging facilities in a single institution. This improves the Neuro’s ability to link individual patient data, samples, and cells with clinical research and high-tech tools such as brain imaging databases without compromising researchers’ independence in setting research programs and pursuing discoveries. As the Open Science initiative takes form, the research group will explicitly and transparently collect data about its concrete effects on research, collaboration, model uptake by other institutions, and the local economy [8,11]. Enhancing Partnerships The Neuro’s initiative provides universities, policy-makers, and firms with the opportunity to evaluate whether Open Science enhances research and local economic growth. Specifically, it will allow these communities to examine two hypotheses. The first of these hypotheses is that the Open Science initiative will attract new private partners. In fact, it has already done so. A number of these collaborated on a recent CDN$84 million grant from the Canadian federal government, and the Neuro is engaged in negotiations over significant partnerships (that will be announced when complete). In line with the Neuro itself, these partners are seeking solutions to struggling drug development and social responses to mental health issues. For example, because of the Open Science initiative, Thermo Fisher Scientific agreed to partner with the Neuro to develop reagents, including antibodies and knock-out cell lines, to accelerate research into a number of neurodegenerative diseases. Potential partners recognize the promise of the Open Science model, in particular the commercial benefits of sharing the risk of early-stage research, and accelerating the speed of research progress [6,8,14,15]. Similar forces launched the Structural Genomics Consortium (SGC) and the Allen Institute for Brain Science [9]. The Allen Institute is particularly pertinent, as it also focuses on neuroscience research and successfully implemented an open science policy almost 15 years ago. Its work has led to the creation of brain atlases, among other research tools, all of which are openly available, and some of which serve as standards in the field. The Neuro aims to develop this approach further by eliminating patents, moving into clinical work and associated data and by avoiding organizational challenges by preserving research independence. As the experience with the SGC illustrates, pharmaceutical companies have, in particular, expressed interest in participating in open science projects [5]. Allowing for Knowledge Spillover A second hypothesis is that the Neuro’s institution-based approach will draw companies to the Montreal region, where the Neuro is based, leading to the creation of a local knowledge hub with attendant jobs and attracting other firms with complementary specialties [16]. The Ubisoft-anchored video game cluster in Montreal illustrates this effect [16]. One of the expectations of the Neuro’s Open Science model is that it will lead firms to develop complementary downstream applications, creating advantages for translational research and developing a local knowledge hub. Because these applications will be complementary to the core research program, permitting partner firms to seek intellectual property protection does not sacrifice, or impede, scientific norms of openness. Open Science at the Neuro will allow these partners to pursue the legal avenues they choose on these complementary or downstream innovations. Further, Open Science also stimulates new research avenues: Murray et al. found that interaction with new partners uninhibited by restrictive intellectual property protection “encourag[es] the establishment of entirely new research directions,” and “reduce[s] the fixed cost of ‘entering’ a particular research area to conduct these investigations,” [17] while avoiding stigmatization of interesting avenues for follow-on research. “Faculty consulting” with private companies outside of the university’s purview causes a similar effect [11]. More generally, scholars have found that encouraging crossdisciplinary integration of expertise is a crucial component of overcoming roadblocks in research [11]. By maximizing ease of access and attracting new collaborators, many of whom are not specialized in the neurobiology domain, the Neuro will create an “interdisciplinary communit[y] made up of a heterogeneous range of members,” expected to accelerate the progression of neuroscience research and the growth of a “knowledge society” [14]. The Neuro’s aim is to build new algorithms, apps, and innovative software that will stimulate more activity, attract more firms and partners, and create a snowball effect. Future collaborations may form across a range of fields from finance to physics, to visualization software. Furthermore, the exchange of knowledge between specialists in different fields should help to avoid replication issues down the road and software bugs by diversifying the number of people with different backgrounds who look at given results. Lowering Transaction Costs A further expectation worth testing is that Open Science lowers transaction costs, such as contract negotiations, court challenges, and intellectual property management that collectively impose a serious burden on universities [4]. Not only should decreasing these transaction costs attract new (particularly smaller) [9] partners by virtue of the increased simplicity of forming partnerships, but doing so can also be expected to accelerate research progress. More efficient resource allocation and decreasing management costs of institutions that specialize in basic research—not legal and business strategies—ought to further decrease costs [11]. Taking advantage of this cost reduction, the Neuro will be in a position to engage in open collaborations with firms, research consortia, and institutions around the world. Not being limited or constrained by the Neuro’s patents, new partner firms can develop business opportunities faster and at a lower cost than if the Neuro maintained a proprietary approach to its research. Nevertheless, to fully access the benefits of Open Science, institutions need to invest in technicians, interoperability standards, and new infrastructure [8]. One would expect the cost savings from eliminating both patent protection and extended negotiations with partners to more than offset these costs [12]. Eliminating patent protections over life sciences research tools in particular will avoid raising the cost of “exploratory research that may enable the future creation of many applications, including those that still are undreamt of.” [14] Resources that are thus freed up can be reallocated to support research and innovation instead. Encourage and Intensify the Accumulation of “Sticky,” Including Tacit, Knowledge A final advantage of the Neuro Open Science initiative is the expectation that it will accelerate the generation of sticky knowledge in the Montreal area. Knowledge is produced by universities but this is not often efficiently translated into local economic benefit. While spin-off companies provide one means through which to translate university-generated knowledge to the local economy [11], they are not the only means of doing so. The Neuro’s knowledge hub—tools and approaches to analyzing the combination of genetic, brain imaging, and behavioral data, links between big data and individual patients, cell lines, and clinical expertise—creates knowledge that will be deeply embedded in Montreal and its surrounding region. This knowledge is not only key to identifying more promising targets for drug development and community mental health care [8] but is also sticky to Montreal as it is the result of positive externalities of innovation. It resides as tacit knowledge in the minds [14] and interaction of individuals living and working in Montreal [11]. Only by locating some operations in Montreal can firms fully take advantage of the ideas founded on the Neuro’s Open Science initiative [16]. Conclusion By encouraging new partnerships, reducing transaction costs burdening upstream research, and creating sticky knowledge specific to Montreal, the Neuro Open Science initiative is designed to promote local innovation development and dissemination of university knowledge [11]. The unique features of this initiative, namely the elimination of patent protection and its clinical institution-based nature, make it particularly well-suited to achieving these goals. Thanks to the Neuro’s commitment to independent and transparent monitoring, we will learn whether these hypothesized benefits turn into reality. The movement away from traditional research models has begun, using other project-based “open” initiatives, and the Neuro Open Science initiative provides an important new piece in the puzzle of improving the efficacy of translational research. By itself, it may be able to accelerate some research. But to fully achieve the benefits of Open Science, other institutions will need to not only follow, but expand on, the Neuro’s lead. Acknowledgments ERG would like to acknowledge the research and writing contributions of Kendra Levasseur. He was funded by PACEOMICS. The PACEOMICS project received funding from Genome Canada, Genome Alberta, Genome Quebec, the Canadian Institutes for Health Research, and Alberta Innovates—Health Solutions and the Social Sciences and Humanities Research Council.

doi: 10.1371/journal.pbio.2000234pmid: 28027292

Introduction An important issue in evolutionary biology is understanding how the continuous-time process of speciation can lead to discrete entities—species. There is usually no ambiguity about species delineation when distant lineages are compared. The continuous nature of the divergence process, however, causes endless debates about the species status of closely related lineages [1]. A number of definitions of species have thus been introduced over the 20th century, each of them using its own criteria—morphological, ecological, phylogenetic, biological, evolutionary, or genotypic. A major problem is that distinct markers do not diverge in time at the same rate [2]. For instance, in some taxa, morphological differences evolve faster than the expression of hybrid fitness depression, which in turn typically establishes long before genome-wide reciprocal monophyly [3]. In other groups, morphology is almost unchanged between lineages that show high levels of molecular divergence [4]. The erratic behavior and evolution of the various criteria is such that in a wide range of between-lineage divergence—named the grey zone of the speciation continuum—distinct species concepts do not converge to the same conclusions regarding species delineation [2]. Besides taxonomic aspects, the grey zone has raised an intense controversy regarding the genetic mechanisms involved in the formation of species [5–7]. Of particular importance is the question of gene flow between diverging lineages. How isolated must two gene pools be for speciation to begin? How long does gene flow persist as lineages diverge? Is speciation a gradual process of gene flow interruption or a succession of periods of isolation and periods of contact? These questions are not only central in the speciation literature but also relevant to the debate about species delineation, with the ability of individuals to exchange genes being at the heart of the biological concept of species. As genomic data have become easier and less expensive to obtain, sophisticated computational approaches have been developed to perform historical inferences in speciation genomics (i.e., estimate the time of ancestral separation in two gene pools, changes in effective population size over evolutionary time, and the history of gene flow between the considered lineages [8–10]). Simulation-based approximate Bayesian computation (ABC) methods are particularly flexible and have recently attracted an increased attention in speciation genomics. One strength of ABC approaches is their ability to deal with complex, hopefully realistic models of speciation and test for the presence or absence of ongoing introgression between sister lineages. This is achieved by simulating molecular data under alternative models of speciation with or without current introgression and choosing among models based on their relative posterior probabilities [11]. Migration tends to homogenize allele content and frequency between diverging populations. This homogenizing effect, however, is often expected to only affect a fraction of the genome. This is because the effective migration rate is impeded in regions containing loci involved in assortative mating, hybrid fitness depression, or other mechanisms of isolation—the so-called genetic barriers [12]. Consequently, gene flow is best identified by models explicitly accounting for among-locus heterogeneity in introgression rates, as demonstrated by a number of recent studies [13–16]. When homogeneous introgression rate across the genome is assumed, distant lineages that have accumulated a large number of genetic barriers can be inferred as currently isolated, whereas they actually exchange alleles at a minority of loci unlinked to barriers [14]. On the other hand, neglecting heterogeneity in introgression rates between closely related lineages can result in a failure to identify some regions of the genome that are already evolving independently [16,17]. Heterogeneous introgression models therefore appear necessary according to the genic view of speciation [18]. Importantly, introgression rates alone do not govern local patterns of genetic differentiation [19]. Linked selective processes, such as hitchhiking effects [20] or background selection [21], are expected to affect the landscape of population differentiation by lowering polymorphism levels at particular loci, especially in low-recombining or gene-dense genomic regions. Neglecting this confounding effect tends to inflate the proportion of false positives in statistical tests of ongoing gene flow [19] and to mislead inferences [22,23]. Linked directional selection is expected to locally increase the stochasticity of allele frequency evolution, a process sometimes coined genetic draft [24]. Its effect can therefore be modeled by assuming that the effective population size, Ne, which determines the strength of genetic drift, varies among loci [25]. Multilocus analyses of the process of population divergence have been achieved in various groups of animals [26,27] and plants [28–30] for which genome-wide data are available, revealing a diversity of patterns. These case studies, however, are limited in number and have taken different approaches, such that we still lack a unifying picture of the prevalence of gene flow during early divergence between gene pools. Here, we gathered a dataset of 61 pairs of populations/species of animals occupying a wide continuum of divergence level. Species were selected in order to sample the phylogenetic and ecological diversity of animals [31], irrespective of any aspect related to population structure or speciation. We investigated the effects of genomic divergence between populations on patterns of gene flow, paying attention to the ability of ABC methods to distinguish between competing models and the influence of model assumptions. Results Simulations: ABC as a Powerful Approach to Test for Current Introgression Five demographic models differing by the history of gene flow between two diverging populations were considered (Fig 1), namely strict isolation (SI), ancient migration (AM), isolation with migration (IM), secondary contact (SC), and panmixia (PAN). The latter three models involve ongoing gene flow between the two populations, whereas the former two do not. The five demographic models were subdivided into different genomic submodels that reflect alternative assumptions about the genomic distribution of indirect selective effects on the effective population size (homoN if homogeneous or heteroN if heterogeneous) and on the migration rate (homoM if homogeneous or heteroM if heterogeneous). Heterogeneous effective population size was considered in all the models, while heterogeneous migration rate was considered in models with gene flow (IM, AM, and SC). The SI and PAN models were divided into two submodels (homoN and heteroN), and the AM, IM, and SC models were divided into four submodels (homoN_homoM, homoN_heteroM, heteroN_homoM, and heteroN_heteroM). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Compared alternative models of speciation. SI = strict isolation: subdivision of an ancestral diploid panmictic population (of size Nanc) in two diploid populations (of constant sizes Npop1 and Npop2) at time Tsplit. AM = ancestral migration: the two newly formed populations continue to exchange alleles until time TAM. IM = isolation with migration: the two daughter populations continuously exchange alleles until present time. SC = secondary contact: the daughter populations first evolve in isolation (forward in time), then experience a secondary contact and start exchanging alleles at time TSC. PAN: panmictic model. All individuals are sampled from the same panmictic population. Red phylogenies represent possible gene trees under each alternative model. https://doi.org/10.1371/journal.pbio.2000234.g001 The dominant assumption in published demographic inferences is the homoN submodel, in which it is assumed that most of the genetic variation in the genome is unaffected (or equally affected) by selection at linked sites. Here, homoN was simulated using a single value of effective population size shared by all loci across the genome, but the effective population size differed among populations. The heteroN submodel accounts for local genomic effects of directional selection (background selection, selective sweeps) by considering a variable effective population size among loci, here assumed to follow a rescaled beta distribution. The homoM submodel assumes that all loci share the same probability to receive alleles from the sister population (i.e., posits the absence of species barriers or of adaptively introgressed loci). Alternatively, the heteroM submodel accounts for the existence of local barriers to gene flow, of variable strengths, and of variable levels of genetic linkage to the sampled loci. HeteroM was here simulated by assuming that the effective introgression rate is beta distributed across the genome, thus intending to account for the combined effects of selection, recombination, and gene density. In principle, one could explicitly include information on local recombination rates and gene density, but no such data was available in the species analyzed here. We explicitly tested the hypothesis of current gene flow by comparing the relative posterior probabilities of 16 models for 61 pairs of species distributed along a continuum of molecular divergence. In the ABC framework, the posterior probability of a model corresponds to its relative ability to theoretically produce datasets similar to the observed dataset, compared to a set of alternative models. Before analyzing datasets from the 61 pairs of animal species, we first assessed the power of the adopted ABC approach to correctly distinguish between models involving current isolation (SI + AM) versus ongoing migration (IM + SC + PAN). This was achieved by randomly simulating 116,000 datasets distributed over the 16 compared models and applying our ABC inference method to each of them. Specifically, we investigated which model had the highest posterior probability and assessed significance by estimating the associated robustness—the probability to correctly support a model given its posterior probability. A robustness greater than 0.95 can be interpreted as a p-value below 0.05 [32]. The analysis of simulated datasets allowed us to empirically measure a threshold value of 0.6419 for the posterior probability Pmigration (= PIM + PSC + PPAN), above which the robustness to support ongoing migration is greater than 0.95. Similarly, a posterior probability Pmigration below 0.1304 implied a statistical support for the current isolation model with a robustness greater than 0.95. Among the 58,000 simulated datasets in which current gene flow was assumed (IM, SC, and PAN; Fig 2A), 99.462% were true positives (Pmigration > Pisolation and robustness ≥ 0.95), 0.129% were false positives (Pmigration < Pisolation and robustness ≥ 0.95), and 0.409% were ambiguous cases for which ABC did not provide any robust conclusion (robustness < 0.95). Among the 58,000 simulated datasets in which current isolation was assumed (SI and AM; Fig 2B), 99.649% were true positives (Pisolation > Pmigration and robustness ≥ 0.95), 0.002% were false positives (Pisolation < Pmigration and robustness ≥ 0.95), and 0.34% were ambiguous cases (robustness < 0.95). When current gene flow was assumed, the rates of false positive and ambiguity were very low at every level of population divergence. When current isolation was assumed, a higher rate of ambiguity, but no elevation of the rate of false inference, was observed at low levels of divergence (Da < 0.01, Fig 2D). This contrasts with the recent suggestion that the full-likelihood method developed in the IMa2 software [33] might be biased towards supporting current gene flow when isolation is recent [19,34]—our approach appears to be immune from this bias. To specifically address this point, we repeated the exact same simulations as in [34] and confirmed that our ABC approach has a reduced power (i.e., more ambiguous cases with robustness <0.95) when the split is recent but still a very low rate of false positive in these conditions (see S1 text). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. ABC analysis of randomly simulated datasets. Posterior probability Pmigration to support ongoing migration was estimated for a total of 116,000 simulated datasets across 16 models. A. Pmigration as a function of the net synonymous divergence Da. Dots represent datasets simulated under the IM, SC, and PAN models. The colors show datasets for which gene flow is correctly supported (green) or wrongly rejected (red). Grey dots represent datasets for which the robustness of the ABC analysis is <0.95. B. Pmigration as a function of the net synonymous divergence Da. Dots represent datasets simulated under the SI or AM models. The colors show datasets for which gene flow is correctly rejected (green; robustness ≥ 0.95) or wrongly supported (red; robustness ≥ 0.95). C. Proportion of true positives (green), false positives (red), and ambiguous analyses (grey) for different ranges of Da across IM, SC, and PAN datasets. Horizontal red line shows 5%. D. Proportion of true positives (green), false positives (red), and ambiguous analyses (grey) for different ranges of Da across SI and AM datasets. https://doi.org/10.1371/journal.pbio.2000234.g002 In addition, the robustness of the ABC inference was only weakly dependent on the sample size when the number of loci was greater than 100: similar results were obtained when we simulated samples of size 2, 3, 25, or 50 diploid individuals (S1 Fig). Finally, and importantly, simulations showed that ABC is not accurate enough to discriminate between the IM and SC models. Datasets simulated under SC were assigned to SC with high confidence only when the period of isolation before secondary contact represents at least a proportion of about 60% of the total divergence time (S2A Fig). When shorter periods of isolation were simulated, the method either assigned the datasets to IM or did not provide an elevated posterior probability to any demographic model (S2B Fig). Dataset: Molecular Divergence and Population Differentiation in 61 Taxa The posterior probability of ongoing gene flow was estimated in 61 pairs of species/populations of animals (S1 Data) showing variable levels of molecular divergence (S1 Data). Fifty pairs were taken from a recent transcriptome-based population genomic study [31], with two individuals per population/species being analyzed here. The datasets for the other 11 species pairs were downloaded from the NCBI (S1 Data). They correspond to sequences from published studies using either ABC, Ima [33], or MIMAR [35], for which 3 to 78 diploid individuals were analyzed. We computed various measures of molecular divergence between species/populations: namely, Da, the relative average divergence, corrected for within-species diversity [36]; Dxy, the absolute average divergence; and FST, a classical measure of population differentiation. In our dataset, Da ranged from 5.10−5 (French versus Danish populations of Ostrea edulis) to 0.309 (Crepidula fornicata versus C. plana) and FST from 0 (between Anas crecca shemya and A. crecca attu) to 0.95 (between Camponotus ligniperdus and C. aethiops, S3 Fig). As expected, Da was strongly correlated to FST and less well to the absolute divergence Dxy (S3B Fig). The across-loci variance in FST was minimal for low and high values of Da (S3B Fig), which reflects an FST homogeneously low at early stages of divergence, homogeneously high at late stages of divergence, and heterogeneous among genes at intermediate levels of Da (S3 Fig). Statistical Analysis: Assessment of Ongoing Gene Flow For each of the 61 studied pairs of populations/species, we focused on synonymous positions and investigated the prevalence of ongoing gene flow by estimating the posterior probabilities of 16 different models under ABC. These 16 models represent the combinations of 5 demographic models (SI, AM, IM, SC, and panmixia) and four assumptions regarding the genomic heterogeneity in introgression (for AM, IM, and SC only) and drift rates (for all models; see above and Material and Methods). The posterior probability Pmigration that the two populations currently exchange migrants was estimated by summing the contributions of the PAN, IM, and SC models (Fig 1) and plotted against measures of molecular divergence (Fig 3). Da, which can be understood as the per-site amount of neutral derived mutations being fixed in the different lineages, provided the best relationship (Fig 3). Results with other measures of divergence and with the estimated age of the split (Tsplit parameter under the IM model) are also shown (S4–S7 Figs). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Probability of ongoing gene flow along a continuum of molecular divergence. Each dot is for one observed pair of populations/species. x-axis: net molecular divergence Da measured at synonymous positions (log10 scale) and averaged across sequenced loci. y-axis: relative posterior probability of ongoing gene flow (i.e., SC, IM, and PAN models) estimated by ABC. Red dots: pairs with a strong support for current isolation. Grey dots: pairs with no strong statistical support for any demographic model (robustness <0.95). Blue dots: pairs with strong statistical support for genome-homogeneous ongoing gene flow. Purple dots: pairs with strong statistical support for genome-heterogeneous ongoing gene flow. Filled symbols: pairs with a strong support for genome-heterogeneous Ne. Open symbols: genome-homogeneous Ne. The light grey rectangle spans the range of net synonymous divergence in which both currently isolated and currently connected pairs are found (see S1 Data). https://doi.org/10.1371/journal.pbio.2000234.g003 Over the continuum of divergence, the 22 pairs with Da lower than 0.5% received a support for ongoing gene flow with a robustness ≥0.95 (Fig 3). The first identified semipermeable barrier to gene flow was detected at Da ≈ 0.075%, a pair of Malurus (fairywren) species [37] for which ABC strongly supports heterogeneity in M. When the net divergence was between 0.5% and 2%, inferences about gene flow were variable and sometimes uncertain. In this grey zone, gene flow was strongly supported in 7 pairs, always with a strong support for genomic heterogeneity in introgression rates. Still, in the grey zone, ABC did not distinguish between isolation and introgression in 3 pairs of species and provided strong support for isolation in 2 other pairs. Finally, among the 27 most divergent pairs of species where Da was greater than 2%, we found 23 pairs with a strong support for current isolation and 4 ambiguous pairs (Fig 3). We investigated the impact of assumptions about genomic heterogeneity in Ne and M on the detection of current introgression (S4–S9 Figs). When both parameters were allowed to vary among loci, pairs of populations with Da exceeding 0.1% and showing strong statistical support for ongoing migration tended to obtain support for genomic heterogeneity in introgression rates. But when constant introgression rate was assumed (homoM_heteroN and homoM_homoN models), the importance of gene flow became underestimated in several divergent pairs of species, consistent with previous reports (e.g. [15]). When we compared models assuming homogeneous versus heterogeneous effective population size across loci, we found that the former tended to overestimate the prevalence of ongoing gene flow (S8 Fig), again in line with published analyses [19]. Analyses assuming homogeneous Ne and M in many cases failed to support either isolation or migration, producing the highest number of ambiguous pairs (S8 Fig). The detected genomic heterogeneity in gene flow increased with Da until 2% of divergence. Finally, across the whole continuum, there was no significant effect of the divergence on the probability of supporting genomic heterogeneity in effective population size in our dataset. No Effect of Habitat, Geography, Phylogeny, or Life History Traits We investigated the influence of a number of ecological, geographical, phylogenetic, and life history variables on the posterior probability of ongoing gene flow. This was achieved under the heteroM_heteroN model using data from [31]. We detected no significant effect of species longevity or log-transformed propagule size (size of the developmental stage that leaves the mother and disperses) on the log-transformed probability of ongoing gene flow. In the same vein, marine organisms (n = 25) did not exhibit a higher propensity for ongoing gene flow than terrestrial ones (n = 36; r2 below 0.01%). The log-transformed probability of ongoing gene flow was significantly higher (p-value = 0.002, r2 = 0.14) in vertebrates (n = 20) than in invertebrates (n = 41), but the effect disappeared when the level of divergence was controlled for (net synonymous divergence <0.04: 17 vertebrate pairs, 22 invertebrate pairs, p = 0.32, r2 = 0.03). This effect only reflects the paucity of pairs of vertebrate population/species with a high divergence in our dataset. Finally, we tested whether the current geographic distribution of species coincides with the establishment of genetic structure in our data by distinguishing pairs in which the two considered species/populations occur on the same versus distinct continents or oceans. We did not find any significant effect of this variable on the estimated values of Pmigration in either of the three divergence zones: Da < 0.5%, t test = –0.015269, df (degrees of freedom for the t-statistic) = 18.522, p-value = 0.988; 0.5% < Da < 2%, t test = –0.74229, df = 7.1996, p-value = 0.4814; 2% < Da, t test = 0.35512, df = 22.426, p-value = 0.7258. Ongoing Gene Flow and Taxonomic Status Finally, we verified whether our inferences confirmed or contradicted the current taxonomy (S1 table). Our dataset comprises 26 pairs of recognized species and 35 pairs of populations (or subspecies) sharing a common binomen. Twenty-one pairs of recognized species belonged to the high-divergence zone (Da > 0.02). Of these, 16 were inferred to be currently isolated, 4 produced ambiguous results and 1 pair, Eunicella cavolinii versus E. verrucosa (gorgonian), was found to be connected by heterogeneous gene flow. Among the 5 remaining recognized pairs of species (with Da < 0.02), 2 were inferred as being fully isolated and 3 were inferred to be connected species: 2 pairs of semi-isolated species with heterogeneous gene flow (Mytilus galloprovincialis versus M. edulis and Macaca mulatta versus M. fascicularis) and the Gorilla gorilla versus G. beringei pair, which was found to be connected by homogeneous gene flow. Of the 35 pairs of recognized populations from the same species, 6 with Da > 0.02 were inferred to be isolated cryptic species. Genetic isolation has been previously suspected between northern and southern populations of Pectinaria koreni (trumpet worms) [38], between the blue and purple morphs of Cystodytes dellechiajei (colonial ascidians) [39], and between the L1 and L2 lineages of Allolobophora chlorotica (earthworms) [40], but genetic isolation is here newly revealed between Morrocan and European populations of Melitaea cinxia (Glanville fritillary), between Spanish and French populations of A. chlorotica L2, and between Mediterranean and tropical populations of Culex pipiens. Simulations: ABC as a Powerful Approach to Test for Current Introgression Five demographic models differing by the history of gene flow between two diverging populations were considered (Fig 1), namely strict isolation (SI), ancient migration (AM), isolation with migration (IM), secondary contact (SC), and panmixia (PAN). The latter three models involve ongoing gene flow between the two populations, whereas the former two do not. The five demographic models were subdivided into different genomic submodels that reflect alternative assumptions about the genomic distribution of indirect selective effects on the effective population size (homoN if homogeneous or heteroN if heterogeneous) and on the migration rate (homoM if homogeneous or heteroM if heterogeneous). Heterogeneous effective population size was considered in all the models, while heterogeneous migration rate was considered in models with gene flow (IM, AM, and SC). The SI and PAN models were divided into two submodels (homoN and heteroN), and the AM, IM, and SC models were divided into four submodels (homoN_homoM, homoN_heteroM, heteroN_homoM, and heteroN_heteroM). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Compared alternative models of speciation. SI = strict isolation: subdivision of an ancestral diploid panmictic population (of size Nanc) in two diploid populations (of constant sizes Npop1 and Npop2) at time Tsplit. AM = ancestral migration: the two newly formed populations continue to exchange alleles until time TAM. IM = isolation with migration: the two daughter populations continuously exchange alleles until present time. SC = secondary contact: the daughter populations first evolve in isolation (forward in time), then experience a secondary contact and start exchanging alleles at time TSC. PAN: panmictic model. All individuals are sampled from the same panmictic population. Red phylogenies represent possible gene trees under each alternative model. https://doi.org/10.1371/journal.pbio.2000234.g001 The dominant assumption in published demographic inferences is the homoN submodel, in which it is assumed that most of the genetic variation in the genome is unaffected (or equally affected) by selection at linked sites. Here, homoN was simulated using a single value of effective population size shared by all loci across the genome, but the effective population size differed among populations. The heteroN submodel accounts for local genomic effects of directional selection (background selection, selective sweeps) by considering a variable effective population size among loci, here assumed to follow a rescaled beta distribution. The homoM submodel assumes that all loci share the same probability to receive alleles from the sister population (i.e., posits the absence of species barriers or of adaptively introgressed loci). Alternatively, the heteroM submodel accounts for the existence of local barriers to gene flow, of variable strengths, and of variable levels of genetic linkage to the sampled loci. HeteroM was here simulated by assuming that the effective introgression rate is beta distributed across the genome, thus intending to account for the combined effects of selection, recombination, and gene density. In principle, one could explicitly include information on local recombination rates and gene density, but no such data was available in the species analyzed here. We explicitly tested the hypothesis of current gene flow by comparing the relative posterior probabilities of 16 models for 61 pairs of species distributed along a continuum of molecular divergence. In the ABC framework, the posterior probability of a model corresponds to its relative ability to theoretically produce datasets similar to the observed dataset, compared to a set of alternative models. Before analyzing datasets from the 61 pairs of animal species, we first assessed the power of the adopted ABC approach to correctly distinguish between models involving current isolation (SI + AM) versus ongoing migration (IM + SC + PAN). This was achieved by randomly simulating 116,000 datasets distributed over the 16 compared models and applying our ABC inference method to each of them. Specifically, we investigated which model had the highest posterior probability and assessed significance by estimating the associated robustness—the probability to correctly support a model given its posterior probability. A robustness greater than 0.95 can be interpreted as a p-value below 0.05 [32]. The analysis of simulated datasets allowed us to empirically measure a threshold value of 0.6419 for the posterior probability Pmigration (= PIM + PSC + PPAN), above which the robustness to support ongoing migration is greater than 0.95. Similarly, a posterior probability Pmigration below 0.1304 implied a statistical support for the current isolation model with a robustness greater than 0.95. Among the 58,000 simulated datasets in which current gene flow was assumed (IM, SC, and PAN; Fig 2A), 99.462% were true positives (Pmigration > Pisolation and robustness ≥ 0.95), 0.129% were false positives (Pmigration < Pisolation and robustness ≥ 0.95), and 0.409% were ambiguous cases for which ABC did not provide any robust conclusion (robustness < 0.95). Among the 58,000 simulated datasets in which current isolation was assumed (SI and AM; Fig 2B), 99.649% were true positives (Pisolation > Pmigration and robustness ≥ 0.95), 0.002% were false positives (Pisolation < Pmigration and robustness ≥ 0.95), and 0.34% were ambiguous cases (robustness < 0.95). When current gene flow was assumed, the rates of false positive and ambiguity were very low at every level of population divergence. When current isolation was assumed, a higher rate of ambiguity, but no elevation of the rate of false inference, was observed at low levels of divergence (Da < 0.01, Fig 2D). This contrasts with the recent suggestion that the full-likelihood method developed in the IMa2 software [33] might be biased towards supporting current gene flow when isolation is recent [19,34]—our approach appears to be immune from this bias. To specifically address this point, we repeated the exact same simulations as in [34] and confirmed that our ABC approach has a reduced power (i.e., more ambiguous cases with robustness <0.95) when the split is recent but still a very low rate of false positive in these conditions (see S1 text). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. ABC analysis of randomly simulated datasets. Posterior probability Pmigration to support ongoing migration was estimated for a total of 116,000 simulated datasets across 16 models. A. Pmigration as a function of the net synonymous divergence Da. Dots represent datasets simulated under the IM, SC, and PAN models. The colors show datasets for which gene flow is correctly supported (green) or wrongly rejected (red). Grey dots represent datasets for which the robustness of the ABC analysis is <0.95. B. Pmigration as a function of the net synonymous divergence Da. Dots represent datasets simulated under the SI or AM models. The colors show datasets for which gene flow is correctly rejected (green; robustness ≥ 0.95) or wrongly supported (red; robustness ≥ 0.95). C. Proportion of true positives (green), false positives (red), and ambiguous analyses (grey) for different ranges of Da across IM, SC, and PAN datasets. Horizontal red line shows 5%. D. Proportion of true positives (green), false positives (red), and ambiguous analyses (grey) for different ranges of Da across SI and AM datasets. https://doi.org/10.1371/journal.pbio.2000234.g002 In addition, the robustness of the ABC inference was only weakly dependent on the sample size when the number of loci was greater than 100: similar results were obtained when we simulated samples of size 2, 3, 25, or 50 diploid individuals (S1 Fig). Finally, and importantly, simulations showed that ABC is not accurate enough to discriminate between the IM and SC models. Datasets simulated under SC were assigned to SC with high confidence only when the period of isolation before secondary contact represents at least a proportion of about 60% of the total divergence time (S2A Fig). When shorter periods of isolation were simulated, the method either assigned the datasets to IM or did not provide an elevated posterior probability to any demographic model (S2B Fig). Dataset: Molecular Divergence and Population Differentiation in 61 Taxa The posterior probability of ongoing gene flow was estimated in 61 pairs of species/populations of animals (S1 Data) showing variable levels of molecular divergence (S1 Data). Fifty pairs were taken from a recent transcriptome-based population genomic study [31], with two individuals per population/species being analyzed here. The datasets for the other 11 species pairs were downloaded from the NCBI (S1 Data). They correspond to sequences from published studies using either ABC, Ima [33], or MIMAR [35], for which 3 to 78 diploid individuals were analyzed. We computed various measures of molecular divergence between species/populations: namely, Da, the relative average divergence, corrected for within-species diversity [36]; Dxy, the absolute average divergence; and FST, a classical measure of population differentiation. In our dataset, Da ranged from 5.10−5 (French versus Danish populations of Ostrea edulis) to 0.309 (Crepidula fornicata versus C. plana) and FST from 0 (between Anas crecca shemya and A. crecca attu) to 0.95 (between Camponotus ligniperdus and C. aethiops, S3 Fig). As expected, Da was strongly correlated to FST and less well to the absolute divergence Dxy (S3B Fig). The across-loci variance in FST was minimal for low and high values of Da (S3B Fig), which reflects an FST homogeneously low at early stages of divergence, homogeneously high at late stages of divergence, and heterogeneous among genes at intermediate levels of Da (S3 Fig). Statistical Analysis: Assessment of Ongoing Gene Flow For each of the 61 studied pairs of populations/species, we focused on synonymous positions and investigated the prevalence of ongoing gene flow by estimating the posterior probabilities of 16 different models under ABC. These 16 models represent the combinations of 5 demographic models (SI, AM, IM, SC, and panmixia) and four assumptions regarding the genomic heterogeneity in introgression (for AM, IM, and SC only) and drift rates (for all models; see above and Material and Methods). The posterior probability Pmigration that the two populations currently exchange migrants was estimated by summing the contributions of the PAN, IM, and SC models (Fig 1) and plotted against measures of molecular divergence (Fig 3). Da, which can be understood as the per-site amount of neutral derived mutations being fixed in the different lineages, provided the best relationship (Fig 3). Results with other measures of divergence and with the estimated age of the split (Tsplit parameter under the IM model) are also shown (S4–S7 Figs). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Probability of ongoing gene flow along a continuum of molecular divergence. Each dot is for one observed pair of populations/species. x-axis: net molecular divergence Da measured at synonymous positions (log10 scale) and averaged across sequenced loci. y-axis: relative posterior probability of ongoing gene flow (i.e., SC, IM, and PAN models) estimated by ABC. Red dots: pairs with a strong support for current isolation. Grey dots: pairs with no strong statistical support for any demographic model (robustness <0.95). Blue dots: pairs with strong statistical support for genome-homogeneous ongoing gene flow. Purple dots: pairs with strong statistical support for genome-heterogeneous ongoing gene flow. Filled symbols: pairs with a strong support for genome-heterogeneous Ne. Open symbols: genome-homogeneous Ne. The light grey rectangle spans the range of net synonymous divergence in which both currently isolated and currently connected pairs are found (see S1 Data). https://doi.org/10.1371/journal.pbio.2000234.g003 Over the continuum of divergence, the 22 pairs with Da lower than 0.5% received a support for ongoing gene flow with a robustness ≥0.95 (Fig 3). The first identified semipermeable barrier to gene flow was detected at Da ≈ 0.075%, a pair of Malurus (fairywren) species [37] for which ABC strongly supports heterogeneity in M. When the net divergence was between 0.5% and 2%, inferences about gene flow were variable and sometimes uncertain. In this grey zone, gene flow was strongly supported in 7 pairs, always with a strong support for genomic heterogeneity in introgression rates. Still, in the grey zone, ABC did not distinguish between isolation and introgression in 3 pairs of species and provided strong support for isolation in 2 other pairs. Finally, among the 27 most divergent pairs of species where Da was greater than 2%, we found 23 pairs with a strong support for current isolation and 4 ambiguous pairs (Fig 3). We investigated the impact of assumptions about genomic heterogeneity in Ne and M on the detection of current introgression (S4–S9 Figs). When both parameters were allowed to vary among loci, pairs of populations with Da exceeding 0.1% and showing strong statistical support for ongoing migration tended to obtain support for genomic heterogeneity in introgression rates. But when constant introgression rate was assumed (homoM_heteroN and homoM_homoN models), the importance of gene flow became underestimated in several divergent pairs of species, consistent with previous reports (e.g. [15]). When we compared models assuming homogeneous versus heterogeneous effective population size across loci, we found that the former tended to overestimate the prevalence of ongoing gene flow (S8 Fig), again in line with published analyses [19]. Analyses assuming homogeneous Ne and M in many cases failed to support either isolation or migration, producing the highest number of ambiguous pairs (S8 Fig). The detected genomic heterogeneity in gene flow increased with Da until 2% of divergence. Finally, across the whole continuum, there was no significant effect of the divergence on the probability of supporting genomic heterogeneity in effective population size in our dataset. No Effect of Habitat, Geography, Phylogeny, or Life History Traits We investigated the influence of a number of ecological, geographical, phylogenetic, and life history variables on the posterior probability of ongoing gene flow. This was achieved under the heteroM_heteroN model using data from [31]. We detected no significant effect of species longevity or log-transformed propagule size (size of the developmental stage that leaves the mother and disperses) on the log-transformed probability of ongoing gene flow. In the same vein, marine organisms (n = 25) did not exhibit a higher propensity for ongoing gene flow than terrestrial ones (n = 36; r2 below 0.01%). The log-transformed probability of ongoing gene flow was significantly higher (p-value = 0.002, r2 = 0.14) in vertebrates (n = 20) than in invertebrates (n = 41), but the effect disappeared when the level of divergence was controlled for (net synonymous divergence <0.04: 17 vertebrate pairs, 22 invertebrate pairs, p = 0.32, r2 = 0.03). This effect only reflects the paucity of pairs of vertebrate population/species with a high divergence in our dataset. Finally, we tested whether the current geographic distribution of species coincides with the establishment of genetic structure in our data by distinguishing pairs in which the two considered species/populations occur on the same versus distinct continents or oceans. We did not find any significant effect of this variable on the estimated values of Pmigration in either of the three divergence zones: Da < 0.5%, t test = –0.015269, df (degrees of freedom for the t-statistic) = 18.522, p-value = 0.988; 0.5% < Da < 2%, t test = –0.74229, df = 7.1996, p-value = 0.4814; 2% < Da, t test = 0.35512, df = 22.426, p-value = 0.7258. Ongoing Gene Flow and Taxonomic Status Finally, we verified whether our inferences confirmed or contradicted the current taxonomy (S1 table). Our dataset comprises 26 pairs of recognized species and 35 pairs of populations (or subspecies) sharing a common binomen. Twenty-one pairs of recognized species belonged to the high-divergence zone (Da > 0.02). Of these, 16 were inferred to be currently isolated, 4 produced ambiguous results and 1 pair, Eunicella cavolinii versus E. verrucosa (gorgonian), was found to be connected by heterogeneous gene flow. Among the 5 remaining recognized pairs of species (with Da < 0.02), 2 were inferred as being fully isolated and 3 were inferred to be connected species: 2 pairs of semi-isolated species with heterogeneous gene flow (Mytilus galloprovincialis versus M. edulis and Macaca mulatta versus M. fascicularis) and the Gorilla gorilla versus G. beringei pair, which was found to be connected by homogeneous gene flow. Of the 35 pairs of recognized populations from the same species, 6 with Da > 0.02 were inferred to be isolated cryptic species. Genetic isolation has been previously suspected between northern and southern populations of Pectinaria koreni (trumpet worms) [38], between the blue and purple morphs of Cystodytes dellechiajei (colonial ascidians) [39], and between the L1 and L2 lineages of Allolobophora chlorotica (earthworms) [40], but genetic isolation is here newly revealed between Morrocan and European populations of Melitaea cinxia (Glanville fritillary), between Spanish and French populations of A. chlorotica L2, and between Mediterranean and tropical populations of Culex pipiens. Discussion We performed a comparative speciation genomics analysis in 61 pairs of populations/species from various phyla of animals. Our ABC analysis, which takes into account the confounding effect of linked selection heterogeneity, provides a first global picture of the prevalence of gene flow between diverging gene pools during the transition from one to two species. Accounting for Among-Locus Heterogeneity in Drift and Migration Rate Inferring the history of divergence and gene flow, which determines the rate of accumulation of species barriers, is of prime importance to understand the process of speciation [17]. This can be achieved by various methods, among which ABC approaches have proven particularly flexible and helpful to compare alternative evolutionary models. Our analysis of simulated datasets illustrates that ABC methods have the power to effectively discriminate recent introgression versus current isolation based on datasets of several hundreds of loci and a few individuals per species—typical of population genomic studies. Comparisons of alternative demographic models, however, can be strongly impacted by assumptions regarding the genomic distribution of effective population size (Ne) and introgression rate (M). Heterogeneities in Ne and M are common in natural populations as a result of selective processes applying either globally (background selection [19,41,42]) or specifically against migrants (genetic barriers [12,43]). Following [13], we here introduced a framework in which each of the two effects, or both, can be readily accounted for. In our analysis, the number of pairs of populations/species for which ambiguous conclusions were reached was maximal when genomic heterogeneities of both migration and drift were neglected. Incorporating within-genome variation in Ne tended to enhance the support for models with current isolation, as previously suggested [19]. The heteroN model makes a difference regarding inference of current gene flow between the highly divergent Ciona intestinalis and C. robusta species (see below). Conversely, incorporating heterogeneity in M doubled the number of pairs for which ongoing gene flow was supported when compared to analyses with homogenous M, in which most of these pairs exhibited ambiguous results. Our study therefore underlines the importance of accounting for genomic heterogeneities for both Ne and M when comparing alternative models of speciation [14,15,19] and calls for prudence regarding the conclusions to be drawn from the analysis of a single pair. However, it is important to recall here that the action of natural selection on its molecular target and neighborhood is more complex than a simple reduction in Ne. Our modeling of genomic heterogeneity in drift and selection by a beta distribution of Ne throughout the genome is an approximation which cannot replace an explicit modeling of these processes. In our modeling, we assumed that a given locus i is independently affected by drift and selection in all of the simulated populations including the ancestral one. Our choice was motivated by the generality of this model. An alternative approach to model genomic heterogeneity in Ne can be to assume that background selection is the main process shaping genomic landscapes of diversity. This can be approximated by assuming that a locus i is equally affected by drift and selection in all populations instead of assuming independent effects as in our study. Among models assuming ongoing gene flow, our ABC analysis of simulated and empirical data often failed to discriminate between the isolation with migration and secondary contact models. These two models yield similar signatures in genetic data, such that only relatively recent secondary contacts following long periods of interrupted gene flow can be detected with high confidence (S2D Fig) [44]. Similarly, among models excluding ongoing gene flow, distinguishing between strict isolation and ancient migration was not possible in a substantial number of cases. These are challenges for future methodological research in the field, with important implications regarding the debate about the requirement of geographic isolation to complete speciation [7,45]. Only two diploid individuals per population/species were used in this analysis for the sake of comparability between datasets (in many populations, no more than two individuals were available) and because of computational limitation. However, our evaluation of the effect of sample size on ABC-based demographic inference suggested that two diploid individuals per population were largely sufficient to capture the main signal when more than 100 loci are available (S1 Fig). Prevalent Gene Flow between Slightly Diverged Gene Pools Although ABC analyses of particular pairs of populations can be affected by the choice of model of genomic heterogeneity, the overall relationship between net molecular divergence and detected ongoing gene flow was qualitatively similar among analyses. Pairs of populations diverging by less than 0.5% were found to currently exchange migrants. This includes populations that form a single panmictic gene pool and pairs of diverging populations/species connected by gene flow. The low-divergence area contains pairs of populations showing conspicuous morphological differences, such as eastern versus western gorilla or the cuniculus and algirus subspecies of rabbit (Oryctolagus cuniculus). No pair of populations in this range of divergence was supported to be genetically isolated or yielded ambiguous results. Simulations indicate that our ABC approach is not expected to yield false inference of gene flow in recently isolated populations, contrary to what was suggested with the full-likelihood approach of IMa2 [34]. The main risk is rather a false inference of isolation despite gene flow (Fig 2), which can be explained by the fact that the SI model is less parameterized than models assuming gene flow (IM and SC). ABC had a low false positive rate even when we simulated very recent splits, as has been done in previous papers [19,34]. This is probably because in strict isolation, shared polymorphisms are quickly sorted into private polymorphisms and fixed differences after population split, such that Da can hardly be very small in the absence of gene flow [46]. Our analysis therefore identifies Da < 0.5% as a good synthetic proxy to attest for the existence of gene flow. Other measures of divergence, although producing a qualitatively similar pattern, did not predict the existence of current gene flow as nicely as Da did. Pairs in the low range of divergence must correspond to populations that did not accumulate sufficiently strong and numerous genetic barriers, such that gene flow currently occurs at important rates. The detection of significantly heterogeneous introgression rates in a number of low-diverged pairs (Da < 0.5%) demonstrates the ability of our ABC approach to detect semipermeable barriers quite efficiently at early stages of speciation and supports the rapid evolution of Dobzhansky–Muller incompatibilities [47,48]. A majority of the pairs from the low-divergence area, however, did not yield any evidence for among-locus heterogeneity of introgression rate. Some might correspond to effectively isolated backgrounds that are missed by our method by lack of power when the signal of heterogeneity is too tenuous. It is quite plausible, however, that some pairs of populations/species in the low-divergence zone have differentially fixed mutations with major effects on hybrid fitness, whereas others do not because of mutational stochasticity and/or across-taxa differences in the genetic architecture of barriers—i.e., simple (two locus) versus complex incompatibilities and strength of associated selective effects [49]. Suppressed Gene Flow at High Sequence Divergence At the other end of the continuum, it appears that above a divergence of a few percent, barriers are strong enough to completely suppress gene flow: almost all pairs of species with Da > 2% were found to have reached reproductive isolation with strong support. This might result from impaired homologous recombination because of improper pairing of dissimilar homologous chromosomes at meiosis, which would reduce the fecundity of hybrids [50,51]. Of note, the upper threshold for reproductive isolation (Da = 2%, Dx y = 5.5%) is of the order of magnitude of the maximal level of within-species genetic diversity reported in animals [31,52], somewhat consistent with the hypothesis of a physical constraint imposed by sequence divergence on the ability to reproduce sexually. Alternatively, the 2% figure may represent a threshold above which Dobzhansky–Muller incompatibilities are normally in sufficient number and strength to suppress introgression. The two hypotheses are not mutually exclusive but pertain to distinctive processes of genetic isolation; the former would be maximally expressed during F1 hybrid meiosis, while the latter would affect recombined, mosaic individuals carrying alleles from the two gene pools at a homozygous state. In the high-divergence area, no instance of among-locus heterogeneous migration was detected, indicating that introgression is blocked across the whole genome in these pairs of species. A number of highly divergent species pairs yielded support for among-locus heterogeneous Ne, suggesting that the same regions of the genome are under strong background selection in the two diverging entities—presumably regions of reduced recombination and/or high density in functional elements. Neglecting the genomic heterogeneity in Ne can lead to false inference of gene flow. For instance, allowing genomic heterogeneity in M but not in Ne led to strong statistical support for a secondary contact between the highly divergent Ciona intestinalis (formerly C. intestinalis B) and C. robusta (formerly C. intestinalis A) species (S4 and S5 Figs), consistent with [14], but accounting for heterogeneity in both M and Ne resulted in an ambiguous result without a sufficiently strong support for any models. The among-locus variance in differentiation between these two species, which was interpreted as mainly reflecting introgression at a few loci in [14], is shown here to possibly be the result of a more complex situation that our models failed to capture. Intermediate Divergence Levels: The Grey Zone of Speciation The area of intermediate divergence from 0.5% to 2% of net synonymous divergence unveils the grey zone of the speciation continuum. In this grey zone, isolated pairs of populations/species coexist with pairs connected by migration, and the latter are mainly composed of semi-isolated genetic backgrounds, the situation under which taxonomic conundrums flourish. Cases of ambiguous conclusions about the demographic history also tended to be found in this intermediate zone, perhaps reflecting instances of complex divergence models that are not well predicted by our demographic models. Researchers should be ready to face problems regarding demographic inference—and therefore parameter estimation—when conducting a project of speciation genomics falling in the grey zone. Accounting for genomic heterogeneity of introgression and drift rates appears to be crucial for detecting current gene flow in this range of divergence (S4–S7 Figs). For instance, the mussel species M. galloprovincialis versus M. edulis and the gorgonian species Eunicella cavolinii versus E. verrucosa are the two most divergent pairs for which ongoing introgression was detected, but this only appeared when the genomic variation in M was accounted for—the homoM_homoN and homoM_heteroN models yielded ambiguous conclusions about these pairs of species, in which the existence of semipermeable barriers has previously been demonstrated [53,54]. Our analysis revealed significant among-locus heterogeneous migration in as many as thirteen pairs of populations/species (Fig 3). This illustrates the commonness of semipermeable genomes at intermediate levels of speciation, when some, but not all, genomic regions are affected by barriers to gene flow. Besides mussels and gorgonians, heterogeneous gene flow was newly detected between American and European populations of Armadillidium vulgare (wood lice) and Artemia franciscana (brine shrimp), between Atlantic and Mediterranean populations of Sepia officinalis (cuttlefish), and between the closely related Eudyptes chrysolophus moseleyi versus E. c. filholi (penguins) and Macaca mulatta versus M. fascicularis (macaques)—in addition to the previously documented mouse [55], rabbit [56], and fairywren [57] cases. The grey zone, finally, includes populations between which unsuspected genetic isolation was here revealed, such as the Moroccan versus European populations of Melitaea cinxia (Glanville fritillary) and the Spanish versus French populations of A. chlorotica L2 (earthworm), which according to our analysis correspond to cryptic species. Our genome-wide approach and proper modeling of heterogeneous processes therefore clarified the status of a number of pairs from the grey zone, emphasizing the variety of situations and the conceptual difficulty with species delineation in this range of divergence. Implications for Speciation and Conservation Research Our dataset is composed of a large variety of taxa with deep phylogenetic relationships and diverse life history traits. In principle, the propensity to evolve prezygotic barriers might differ between groups of organisms (e.g., broadcast spawners versus copulating species [58]). We did not detect any significant effect of species biological and/or ecological features or taxonomy on the observed pattern. Highly polymorphic broadcast spawners and low-diversity large vertebrates with strong parental investment were equally likely to undergo current gene flow for a given divergence level. Whether the pace of accumulation of genetic barriers, the so-called speciation clock, varies among taxonomic group is a major challenge in speciation research and requires the dissection of the temporal establishment of barriers in many different taxa [59,60]. State-of-the-art ABC methods offer the opportunity to investigate the genome-wide effect of barriers to gene flow in natural populations but cannot provide answers about how and why barriers have evolved. However, our report of a strong and general relationship between molecular divergence and genetic isolation across a wide diversity of animals suggests that, at the genome level, speciation operates in a more or less similar fashion in distinct taxa, irrespective of biological and ecological particularities. Interestingly, we did not detect any significant effect of geographic range overlap. This result may appear as unexpected at first sight because one expects gene flow to be dependent on geography. One explanation could be that we used a too crude measure of range overlap. Alternatively, this result could support the idea that in many taxa, the observed genetic structure was established in the past in a geographic context different from the current one and only recently reshuffled by recent migration and/or colonization processes [61]. According to this hypothesis, genetic subdivision could have little to do with contemporary connectivity. The width of the grey zone indicates that a number of existing taxonomic debates regarding species definition and delineation are difficult by nature and unlikely to be resolved through the analysis of a limited number of loci. Most of the molecular ecology literature, however, is based on datasets consisting of mitochondrial DNA and rarely more than a dozen microsatellite loci. The time when genome-wide data will be available in most species of interest is approaching, though not yet reached. Since then, we have to accept that knowledge about the existence of gene flow between diverged entities could not be settled from genetic data alone in a substantial fraction of taxa. In addition, our study highlights the commonness of semi-isolated entities, between which gene flow can be demonstrated but only concerns a fraction of loci, further challenging the species concept. We should therefore be prepared to make decisions regarding conservation and management of biodiversity in absence of well-defined species boundaries. Accounting for Among-Locus Heterogeneity in Drift and Migration Rate Inferring the history of divergence and gene flow, which determines the rate of accumulation of species barriers, is of prime importance to understand the process of speciation [17]. This can be achieved by various methods, among which ABC approaches have proven particularly flexible and helpful to compare alternative evolutionary models. Our analysis of simulated datasets illustrates that ABC methods have the power to effectively discriminate recent introgression versus current isolation based on datasets of several hundreds of loci and a few individuals per species—typical of population genomic studies. Comparisons of alternative demographic models, however, can be strongly impacted by assumptions regarding the genomic distribution of effective population size (Ne) and introgression rate (M). Heterogeneities in Ne and M are common in natural populations as a result of selective processes applying either globally (background selection [19,41,42]) or specifically against migrants (genetic barriers [12,43]). Following [13], we here introduced a framework in which each of the two effects, or both, can be readily accounted for. In our analysis, the number of pairs of populations/species for which ambiguous conclusions were reached was maximal when genomic heterogeneities of both migration and drift were neglected. Incorporating within-genome variation in Ne tended to enhance the support for models with current isolation, as previously suggested [19]. The heteroN model makes a difference regarding inference of current gene flow between the highly divergent Ciona intestinalis and C. robusta species (see below). Conversely, incorporating heterogeneity in M doubled the number of pairs for which ongoing gene flow was supported when compared to analyses with homogenous M, in which most of these pairs exhibited ambiguous results. Our study therefore underlines the importance of accounting for genomic heterogeneities for both Ne and M when comparing alternative models of speciation [14,15,19] and calls for prudence regarding the conclusions to be drawn from the analysis of a single pair. However, it is important to recall here that the action of natural selection on its molecular target and neighborhood is more complex than a simple reduction in Ne. Our modeling of genomic heterogeneity in drift and selection by a beta distribution of Ne throughout the genome is an approximation which cannot replace an explicit modeling of these processes. In our modeling, we assumed that a given locus i is independently affected by drift and selection in all of the simulated populations including the ancestral one. Our choice was motivated by the generality of this model. An alternative approach to model genomic heterogeneity in Ne can be to assume that background selection is the main process shaping genomic landscapes of diversity. This can be approximated by assuming that a locus i is equally affected by drift and selection in all populations instead of assuming independent effects as in our study. Among models assuming ongoing gene flow, our ABC analysis of simulated and empirical data often failed to discriminate between the isolation with migration and secondary contact models. These two models yield similar signatures in genetic data, such that only relatively recent secondary contacts following long periods of interrupted gene flow can be detected with high confidence (S2D Fig) [44]. Similarly, among models excluding ongoing gene flow, distinguishing between strict isolation and ancient migration was not possible in a substantial number of cases. These are challenges for future methodological research in the field, with important implications regarding the debate about the requirement of geographic isolation to complete speciation [7,45]. Only two diploid individuals per population/species were used in this analysis for the sake of comparability between datasets (in many populations, no more than two individuals were available) and because of computational limitation. However, our evaluation of the effect of sample size on ABC-based demographic inference suggested that two diploid individuals per population were largely sufficient to capture the main signal when more than 100 loci are available (S1 Fig). Prevalent Gene Flow between Slightly Diverged Gene Pools Although ABC analyses of particular pairs of populations can be affected by the choice of model of genomic heterogeneity, the overall relationship between net molecular divergence and detected ongoing gene flow was qualitatively similar among analyses. Pairs of populations diverging by less than 0.5% were found to currently exchange migrants. This includes populations that form a single panmictic gene pool and pairs of diverging populations/species connected by gene flow. The low-divergence area contains pairs of populations showing conspicuous morphological differences, such as eastern versus western gorilla or the cuniculus and algirus subspecies of rabbit (Oryctolagus cuniculus). No pair of populations in this range of divergence was supported to be genetically isolated or yielded ambiguous results. Simulations indicate that our ABC approach is not expected to yield false inference of gene flow in recently isolated populations, contrary to what was suggested with the full-likelihood approach of IMa2 [34]. The main risk is rather a false inference of isolation despite gene flow (Fig 2), which can be explained by the fact that the SI model is less parameterized than models assuming gene flow (IM and SC). ABC had a low false positive rate even when we simulated very recent splits, as has been done in previous papers [19,34]. This is probably because in strict isolation, shared polymorphisms are quickly sorted into private polymorphisms and fixed differences after population split, such that Da can hardly be very small in the absence of gene flow [46]. Our analysis therefore identifies Da < 0.5% as a good synthetic proxy to attest for the existence of gene flow. Other measures of divergence, although producing a qualitatively similar pattern, did not predict the existence of current gene flow as nicely as Da did. Pairs in the low range of divergence must correspond to populations that did not accumulate sufficiently strong and numerous genetic barriers, such that gene flow currently occurs at important rates. The detection of significantly heterogeneous introgression rates in a number of low-diverged pairs (Da < 0.5%) demonstrates the ability of our ABC approach to detect semipermeable barriers quite efficiently at early stages of speciation and supports the rapid evolution of Dobzhansky–Muller incompatibilities [47,48]. A majority of the pairs from the low-divergence area, however, did not yield any evidence for among-locus heterogeneity of introgression rate. Some might correspond to effectively isolated backgrounds that are missed by our method by lack of power when the signal of heterogeneity is too tenuous. It is quite plausible, however, that some pairs of populations/species in the low-divergence zone have differentially fixed mutations with major effects on hybrid fitness, whereas others do not because of mutational stochasticity and/or across-taxa differences in the genetic architecture of barriers—i.e., simple (two locus) versus complex incompatibilities and strength of associated selective effects [49]. Suppressed Gene Flow at High Sequence Divergence At the other end of the continuum, it appears that above a divergence of a few percent, barriers are strong enough to completely suppress gene flow: almost all pairs of species with Da > 2% were found to have reached reproductive isolation with strong support. This might result from impaired homologous recombination because of improper pairing of dissimilar homologous chromosomes at meiosis, which would reduce the fecundity of hybrids [50,51]. Of note, the upper threshold for reproductive isolation (Da = 2%, Dx y = 5.5%) is of the order of magnitude of the maximal level of within-species genetic diversity reported in animals [31,52], somewhat consistent with the hypothesis of a physical constraint imposed by sequence divergence on the ability to reproduce sexually. Alternatively, the 2% figure may represent a threshold above which Dobzhansky–Muller incompatibilities are normally in sufficient number and strength to suppress introgression. The two hypotheses are not mutually exclusive but pertain to distinctive processes of genetic isolation; the former would be maximally expressed during F1 hybrid meiosis, while the latter would affect recombined, mosaic individuals carrying alleles from the two gene pools at a homozygous state. In the high-divergence area, no instance of among-locus heterogeneous migration was detected, indicating that introgression is blocked across the whole genome in these pairs of species. A number of highly divergent species pairs yielded support for among-locus heterogeneous Ne, suggesting that the same regions of the genome are under strong background selection in the two diverging entities—presumably regions of reduced recombination and/or high density in functional elements. Neglecting the genomic heterogeneity in Ne can lead to false inference of gene flow. For instance, allowing genomic heterogeneity in M but not in Ne led to strong statistical support for a secondary contact between the highly divergent Ciona intestinalis (formerly C. intestinalis B) and C. robusta (formerly C. intestinalis A) species (S4 and S5 Figs), consistent with [14], but accounting for heterogeneity in both M and Ne resulted in an ambiguous result without a sufficiently strong support for any models. The among-locus variance in differentiation between these two species, which was interpreted as mainly reflecting introgression at a few loci in [14], is shown here to possibly be the result of a more complex situation that our models failed to capture. Intermediate Divergence Levels: The Grey Zone of Speciation The area of intermediate divergence from 0.5% to 2% of net synonymous divergence unveils the grey zone of the speciation continuum. In this grey zone, isolated pairs of populations/species coexist with pairs connected by migration, and the latter are mainly composed of semi-isolated genetic backgrounds, the situation under which taxonomic conundrums flourish. Cases of ambiguous conclusions about the demographic history also tended to be found in this intermediate zone, perhaps reflecting instances of complex divergence models that are not well predicted by our demographic models. Researchers should be ready to face problems regarding demographic inference—and therefore parameter estimation—when conducting a project of speciation genomics falling in the grey zone. Accounting for genomic heterogeneity of introgression and drift rates appears to be crucial for detecting current gene flow in this range of divergence (S4–S7 Figs). For instance, the mussel species M. galloprovincialis versus M. edulis and the gorgonian species Eunicella cavolinii versus E. verrucosa are the two most divergent pairs for which ongoing introgression was detected, but this only appeared when the genomic variation in M was accounted for—the homoM_homoN and homoM_heteroN models yielded ambiguous conclusions about these pairs of species, in which the existence of semipermeable barriers has previously been demonstrated [53,54]. Our analysis revealed significant among-locus heterogeneous migration in as many as thirteen pairs of populations/species (Fig 3). This illustrates the commonness of semipermeable genomes at intermediate levels of speciation, when some, but not all, genomic regions are affected by barriers to gene flow. Besides mussels and gorgonians, heterogeneous gene flow was newly detected between American and European populations of Armadillidium vulgare (wood lice) and Artemia franciscana (brine shrimp), between Atlantic and Mediterranean populations of Sepia officinalis (cuttlefish), and between the closely related Eudyptes chrysolophus moseleyi versus E. c. filholi (penguins) and Macaca mulatta versus M. fascicularis (macaques)—in addition to the previously documented mouse [55], rabbit [56], and fairywren [57] cases. The grey zone, finally, includes populations between which unsuspected genetic isolation was here revealed, such as the Moroccan versus European populations of Melitaea cinxia (Glanville fritillary) and the Spanish versus French populations of A. chlorotica L2 (earthworm), which according to our analysis correspond to cryptic species. Our genome-wide approach and proper modeling of heterogeneous processes therefore clarified the status of a number of pairs from the grey zone, emphasizing the variety of situations and the conceptual difficulty with species delineation in this range of divergence. Implications for Speciation and Conservation Research Our dataset is composed of a large variety of taxa with deep phylogenetic relationships and diverse life history traits. In principle, the propensity to evolve prezygotic barriers might differ between groups of organisms (e.g., broadcast spawners versus copulating species [58]). We did not detect any significant effect of species biological and/or ecological features or taxonomy on the observed pattern. Highly polymorphic broadcast spawners and low-diversity large vertebrates with strong parental investment were equally likely to undergo current gene flow for a given divergence level. Whether the pace of accumulation of genetic barriers, the so-called speciation clock, varies among taxonomic group is a major challenge in speciation research and requires the dissection of the temporal establishment of barriers in many different taxa [59,60]. State-of-the-art ABC methods offer the opportunity to investigate the genome-wide effect of barriers to gene flow in natural populations but cannot provide answers about how and why barriers have evolved. However, our report of a strong and general relationship between molecular divergence and genetic isolation across a wide diversity of animals suggests that, at the genome level, speciation operates in a more or less similar fashion in distinct taxa, irrespective of biological and ecological particularities. Interestingly, we did not detect any significant effect of geographic range overlap. This result may appear as unexpected at first sight because one expects gene flow to be dependent on geography. One explanation could be that we used a too crude measure of range overlap. Alternatively, this result could support the idea that in many taxa, the observed genetic structure was established in the past in a geographic context different from the current one and only recently reshuffled by recent migration and/or colonization processes [61]. According to this hypothesis, genetic subdivision could have little to do with contemporary connectivity. The width of the grey zone indicates that a number of existing taxonomic debates regarding species definition and delineation are difficult by nature and unlikely to be resolved through the analysis of a limited number of loci. Most of the molecular ecology literature, however, is based on datasets consisting of mitochondrial DNA and rarely more than a dozen microsatellite loci. The time when genome-wide data will be available in most species of interest is approaching, though not yet reached. Since then, we have to accept that knowledge about the existence of gene flow between diverged entities could not be settled from genetic data alone in a substantial fraction of taxa. In addition, our study highlights the commonness of semi-isolated entities, between which gene flow can be demonstrated but only concerns a fraction of loci, further challenging the species concept. We should therefore be prepared to make decisions regarding conservation and management of biodiversity in absence of well-defined species boundaries. Materials and Methods All of the informatic codes, data and command lines used to produce the analysis are openly available online in the following GitHub repository: https://github.com/popgenomics/popPhylABC. Taxon Sampling A total of 61 pairs of populations/species of animals were analyzed (S1 Data). These include 10 pairs taken from the speciation literature and 51 pairs newly created here based on a recently published RNAseq dataset [31], which includes 96 species of animals from 31 distinct families and eight phyla, and 1 to 11 individuals per species. Twenty-nine of the newly created pairs corresponded to distinct populations within a named species. Populations were here defined based on a combination of geographic, ecotypic, and genetic criteria: we contrasted groups of individuals (i) living in allopatry and/or differing in terms of their ecology and (ii) clustering as distinct lineages in a neighbor-joining analysis of genetic distances between individuals. The 2 most covered individuals per population were selected for ABC analysis. In 4 species, 3 distinct populations were identified, in which case the three possible pairwise comparisons were performed. Results were qualitatively unchanged when we kept a single pair per species. Twenty-two of the newly created pairs consisted of individuals from 2 distinct named species that belonged to the same family. Again, the 2 most covered individuals per species were selected for analysis. In the case of species in which several populations had been identified, we chose to sample 2 individuals from the same population for between-species comparison. When more than 2 species from the same family were available, we selected a single pair based on a combination of sequencing coverage and genetic distance criteria, with comparisons between closely related species being favored. Raw and final datasets are available from the PopPhyl website (http://kimura.univ-montp2.fr/PopPhyl/). Sample sizes, number of loci, and source of data are listed in S1 Data. Transcriptome Assembly, Read Mapping, and Coding Sequence Prediction For the 51 recently obtained pairs, Illumina reads were mapped to predicted cDNAs (contigs) with the BWA program [62]. Contigs with a per-individual average coverage below ×2.5 were discarded. Open reading frames (ORFs) were predicted with the Trinity package [63]. Contigs carrying no ORF longer than 200 bp were discarded. In contigs including ORFs longer than 200 bp, 5ʹ and 3ʹ flanking noncoding sequences were deleted, thus producing predicted coding sequences that are hereafter referred to as loci. Calling Single Nucleotide Polymorphisms (SNPs) and Genotypes At each position of each locus and for each individual, diploid genotypes were called using the reads2snps program [64]. This method first estimates the sequencing error rate in the maximum-likelihood framework, calculates the posterior probability of each possible genotype, and retains genotypes supported at >95% if ten reads per position and per individual were detected. Possible hidden paralogs (duplicated genes) were filtered using a likelihood ratio test based on explicit modeling of paralogy. For our demographic inferences, only synonymous positions were retained. Synonymous length and positions were then computed for each loci using polydNdS [65]. Summary Statistics For all of the 61 pairs of populations/species, we calculated an array of 31 statistics widely used for demographic inferences [32,35,66,67]: the average and standard variation over loci for (1) the number of biallelic positions; (2) the number of fixed differences between the two gene pools; (3) the number of polymorphic sites specific to each gene pool; (4) the number of polymorphic sites existing in both gene pools; (5) Wald and Wolfowitz statistics [68]; (6) Tajima's pi [69]; (7) Watterson's theta [70]; Tajima's D for each gene pool [71]; (8) the gross divergence between the two gene pools (Dxy); (9) the net divergence between the two gene pools (Da); (10) FST measured by 1-pW/pT, where pW is the average allelic diversity based on the two gene pools and pT is the total allelic diversity over the two gene pools; and (11) the Pearson's R² correlation coefficient in p calculated between the two gene pools. Observed values of summary statistics are summarized for each species in S2 Data. Demographic Models Five distinct demographic models were considered: PAN, SI, AM, IM, and SC. (Fig 1). The PAN model assumes that the two investigated gene pools are sampled from a single panmictic population of size Ne sampled in the uniform prior [0–5,000,000] individuals. The SI model describes the subdivision of an ancestral panmictic population of size Nanc in two isolated gene pools of sizes Npop-1 and Npop-2. The two sister gene pools then evolve in absence of gene flow. Under the IM model, the two sister gene pools that split Tsplit (sampled in the uniform prior [0–10,000,000]) generations ago continuously exchange alleles as they diverge. Under the AM model, gene flow occurs between Tsplit and a more recent TAM date sampled from the uniform prior [0–Tsplit], after which the two gene pools evolve in strict isolation. The SC model assumes an early divergence in strict isolation followed by a period of gene flow that started TSC generations ago with TSC sampled from the uniform prior [0–Tsplit]. Heterogeneity in Introgression and Effective Population Size We assumed that the effects of selection on linked sites can be described in terms of heterogeneous effective population size (putatively affecting all demographic models) and/or migration rate (only affecting the IM, AM, and SC models). In the homoM setting, one gene flow parameter (M = N.m) is randomly sampled from a uniform prior distribution for each direction. M1 is the direction from gene pool 2 to gene pool 1 and M2 is the direction from gene pool 1 to gene pool 2. All loci share the same M1 and M2 values, but M1 and M2 are independently sampled. In the heteroM setting, a specific migration rate is attributed per locus and per direction of migration. Thus, for each direction, a hyperprior is first randomly designed as a beta distribution. A value of M1,i and M2,i is then drawn for each loci i from the two hyperpriors. In the homoN setting, the effective population sizes Nanc (ancestral population), Npop-1 (gene pool 1) and Npop-2 (gene pool 2) are independent but shared by all loci. In the heteroN setting, heterogeneity in effective population size is independently modeled for the three populations (ancestor, gene pool 1, and gene pool 2). For each population, a proportion a of loci is assumed to evolve neutrally and share a common value for Nanc, Npop-1, or Npop-2, a being sampled from the uniform prior [0–1]. The remaining loci, in proportion 1-a, are assumed to be affected by natural selection at linked loci. They are assigned independent values of N, which are sampled from beta distributions defined on the intervals [0–Nanc], [0–Npop-1], and [0–Npop-2]. In this setting, a and Ne differ between the three populations but are sampled from distributions sharing the same shape parameters. Approximate Bayesian Computation The combination of demographic models and genomic settings resulted in a total of 16 distinct models, namely the homoN and heteroN versions of PAN and SI and the homoM_homoN, homoM_heteroN, heteroM_homoN, heteroM_heteroN versions of IM, AM, and SC. Model fit assessment and parameter estimation were performed under the ABC framework. Under each model, 3,000,000 multilocus simulations were conducted using the coalescent simulator msnsam, a modified version of ms allowing variation across loci of the number of sampled individuals [66,72]. For each of the 61 pairs of populations/species, the posterior probability of each model was estimated using a feed-forward neural network implementing a nonlinear multivariate regression by considering the model itself as an additional parameter to be inferred under the ABC framework using the R package “abc” [73]. The 10,000 replicate simulations (out of 16 x 3,000,000) falling nearest to the observed values of summary statistics were selected, and these were weighted by an Epanechnikov kernel that peaks when Sobs = Ssim. Computations were performed using 50 trained neural networks and 10 hidden networks in the regression. The posterior probability of each model was obtained by averaging over ten replicated ABC analyses. Robustness Among a set of compared models, ABC returns a best-supported model M and its posterior probability PM. The returned model is validated when PM is above an arbitrary threshold X, corresponding to the posterior probability above which the statistical support for a model is considered as being significant. The robustness of the inference—i.e., the probability to correctly support model M if true—obviously depends on X. To assess the reliability of our approach, we randomly simulated 116,000 pseudo-observed datasets (PODs) distributed over the 16 compared models. Simulations were independent of the 3,000,000 x 16 reference simulations used for model comparisons in our main analysis, but their parameters share the same boundaries. For each simulated POD, we estimated the posterior probabilities Pi of the 16 compared models through ABC. The probability of correctly supporting M given X was calculated as: , where P(PM > X | i) is the probability that a dataset simulated under m will be supported by ABC as being M with a posterior probability above X [32]. This is the proportion (among simulated datasets inferred by ABC to correspond to M) of those actually generated under M. For the “ongoing gene flow” versus “current isolation” model comparison, we empirically measured that robustness to support gene flow starts to be above 0.95 if Pmigration ≥ 0.6419 and the robustness to support isolation is above 0.95 if Pmigration ≤ 0.1304. For datasets with Pmigration between 0.1304 and 0.6419, we did not attribute a best model but treated them as “ambiguous cases.” Taxon Sampling A total of 61 pairs of populations/species of animals were analyzed (S1 Data). These include 10 pairs taken from the speciation literature and 51 pairs newly created here based on a recently published RNAseq dataset [31], which includes 96 species of animals from 31 distinct families and eight phyla, and 1 to 11 individuals per species. Twenty-nine of the newly created pairs corresponded to distinct populations within a named species. Populations were here defined based on a combination of geographic, ecotypic, and genetic criteria: we contrasted groups of individuals (i) living in allopatry and/or differing in terms of their ecology and (ii) clustering as distinct lineages in a neighbor-joining analysis of genetic distances between individuals. The 2 most covered individuals per population were selected for ABC analysis. In 4 species, 3 distinct populations were identified, in which case the three possible pairwise comparisons were performed. Results were qualitatively unchanged when we kept a single pair per species. Twenty-two of the newly created pairs consisted of individuals from 2 distinct named species that belonged to the same family. Again, the 2 most covered individuals per species were selected for analysis. In the case of species in which several populations had been identified, we chose to sample 2 individuals from the same population for between-species comparison. When more than 2 species from the same family were available, we selected a single pair based on a combination of sequencing coverage and genetic distance criteria, with comparisons between closely related species being favored. Raw and final datasets are available from the PopPhyl website (http://kimura.univ-montp2.fr/PopPhyl/). Sample sizes, number of loci, and source of data are listed in S1 Data. Transcriptome Assembly, Read Mapping, and Coding Sequence Prediction For the 51 recently obtained pairs, Illumina reads were mapped to predicted cDNAs (contigs) with the BWA program [62]. Contigs with a per-individual average coverage below ×2.5 were discarded. Open reading frames (ORFs) were predicted with the Trinity package [63]. Contigs carrying no ORF longer than 200 bp were discarded. In contigs including ORFs longer than 200 bp, 5ʹ and 3ʹ flanking noncoding sequences were deleted, thus producing predicted coding sequences that are hereafter referred to as loci. Calling Single Nucleotide Polymorphisms (SNPs) and Genotypes At each position of each locus and for each individual, diploid genotypes were called using the reads2snps program [64]. This method first estimates the sequencing error rate in the maximum-likelihood framework, calculates the posterior probability of each possible genotype, and retains genotypes supported at >95% if ten reads per position and per individual were detected. Possible hidden paralogs (duplicated genes) were filtered using a likelihood ratio test based on explicit modeling of paralogy. For our demographic inferences, only synonymous positions were retained. Synonymous length and positions were then computed for each loci using polydNdS [65]. Summary Statistics For all of the 61 pairs of populations/species, we calculated an array of 31 statistics widely used for demographic inferences [32,35,66,67]: the average and standard variation over loci for (1) the number of biallelic positions; (2) the number of fixed differences between the two gene pools; (3) the number of polymorphic sites specific to each gene pool; (4) the number of polymorphic sites existing in both gene pools; (5) Wald and Wolfowitz statistics [68]; (6) Tajima's pi [69]; (7) Watterson's theta [70]; Tajima's D for each gene pool [71]; (8) the gross divergence between the two gene pools (Dxy); (9) the net divergence between the two gene pools (Da); (10) FST measured by 1-pW/pT, where pW is the average allelic diversity based on the two gene pools and pT is the total allelic diversity over the two gene pools; and (11) the Pearson's R² correlation coefficient in p calculated between the two gene pools. Observed values of summary statistics are summarized for each species in S2 Data. Demographic Models Five distinct demographic models were considered: PAN, SI, AM, IM, and SC. (Fig 1). The PAN model assumes that the two investigated gene pools are sampled from a single panmictic population of size Ne sampled in the uniform prior [0–5,000,000] individuals. The SI model describes the subdivision of an ancestral panmictic population of size Nanc in two isolated gene pools of sizes Npop-1 and Npop-2. The two sister gene pools then evolve in absence of gene flow. Under the IM model, the two sister gene pools that split Tsplit (sampled in the uniform prior [0–10,000,000]) generations ago continuously exchange alleles as they diverge. Under the AM model, gene flow occurs between Tsplit and a more recent TAM date sampled from the uniform prior [0–Tsplit], after which the two gene pools evolve in strict isolation. The SC model assumes an early divergence in strict isolation followed by a period of gene flow that started TSC generations ago with TSC sampled from the uniform prior [0–Tsplit]. Heterogeneity in Introgression and Effective Population Size We assumed that the effects of selection on linked sites can be described in terms of heterogeneous effective population size (putatively affecting all demographic models) and/or migration rate (only affecting the IM, AM, and SC models). In the homoM setting, one gene flow parameter (M = N.m) is randomly sampled from a uniform prior distribution for each direction. M1 is the direction from gene pool 2 to gene pool 1 and M2 is the direction from gene pool 1 to gene pool 2. All loci share the same M1 and M2 values, but M1 and M2 are independently sampled. In the heteroM setting, a specific migration rate is attributed per locus and per direction of migration. Thus, for each direction, a hyperprior is first randomly designed as a beta distribution. A value of M1,i and M2,i is then drawn for each loci i from the two hyperpriors. In the homoN setting, the effective population sizes Nanc (ancestral population), Npop-1 (gene pool 1) and Npop-2 (gene pool 2) are independent but shared by all loci. In the heteroN setting, heterogeneity in effective population size is independently modeled for the three populations (ancestor, gene pool 1, and gene pool 2). For each population, a proportion a of loci is assumed to evolve neutrally and share a common value for Nanc, Npop-1, or Npop-2, a being sampled from the uniform prior [0–1]. The remaining loci, in proportion 1-a, are assumed to be affected by natural selection at linked loci. They are assigned independent values of N, which are sampled from beta distributions defined on the intervals [0–Nanc], [0–Npop-1], and [0–Npop-2]. In this setting, a and Ne differ between the three populations but are sampled from distributions sharing the same shape parameters. Approximate Bayesian Computation The combination of demographic models and genomic settings resulted in a total of 16 distinct models, namely the homoN and heteroN versions of PAN and SI and the homoM_homoN, homoM_heteroN, heteroM_homoN, heteroM_heteroN versions of IM, AM, and SC. Model fit assessment and parameter estimation were performed under the ABC framework. Under each model, 3,000,000 multilocus simulations were conducted using the coalescent simulator msnsam, a modified version of ms allowing variation across loci of the number of sampled individuals [66,72]. For each of the 61 pairs of populations/species, the posterior probability of each model was estimated using a feed-forward neural network implementing a nonlinear multivariate regression by considering the model itself as an additional parameter to be inferred under the ABC framework using the R package “abc” [73]. The 10,000 replicate simulations (out of 16 x 3,000,000) falling nearest to the observed values of summary statistics were selected, and these were weighted by an Epanechnikov kernel that peaks when Sobs = Ssim. Computations were performed using 50 trained neural networks and 10 hidden networks in the regression. The posterior probability of each model was obtained by averaging over ten replicated ABC analyses. Robustness Among a set of compared models, ABC returns a best-supported model M and its posterior probability PM. The returned model is validated when PM is above an arbitrary threshold X, corresponding to the posterior probability above which the statistical support for a model is considered as being significant. The robustness of the inference—i.e., the probability to correctly support model M if true—obviously depends on X. To assess the reliability of our approach, we randomly simulated 116,000 pseudo-observed datasets (PODs) distributed over the 16 compared models. Simulations were independent of the 3,000,000 x 16 reference simulations used for model comparisons in our main analysis, but their parameters share the same boundaries. For each simulated POD, we estimated the posterior probabilities Pi of the 16 compared models through ABC. The probability of correctly supporting M given X was calculated as: , where P(PM > X | i) is the probability that a dataset simulated under m will be supported by ABC as being M with a posterior probability above X [32]. This is the proportion (among simulated datasets inferred by ABC to correspond to M) of those actually generated under M. For the “ongoing gene flow” versus “current isolation” model comparison, we empirically measured that robustness to support gene flow starts to be above 0.95 if Pmigration ≥ 0.6419 and the robustness to support isolation is above 0.95 if Pmigration ≤ 0.1304. For datasets with Pmigration between 0.1304 and 0.6419, we did not attribute a best model but treated them as “ambiguous cases.” Supporting Information S1 Fig. Effects of the number of sampled individuals on robustness of model comparisons when 100 loci are investigated. Analyses were made by simulating four different datasets: A-B: 100 loci sampled in two diploid individuals in each daughter species. C-D: 100 loci sampled in three diploid individuals in each daughter species. E-F: 100 loci sampled in 25 diploid individuals in each daughter species. G-H: 100 loci sampled in 50 diploid individuals in each daughter species. Panels on the left border show the distributions of P(current isolation | current isolation) (white bars) and P(current introgression | current introgression) (grey bars) measured after ABC analysis of 20,000 PODs simulated under each models. Panels on the right border show the distributions of P(SI | SI) (black lines), P(AM | AM) (red lines), P(IM | IM) (blue lines) and P(SC | SC) (green bars) measured after ABC analysis of 20,000 PODs simulated under each models. https://doi.org/10.1371/journal.pbio.2000234.s001 (TIF) S2 Fig. Effect of parameter combinations on the correct support of the SC model. A. Two-dimensional space of parameters of the SC model showing simulations leading to a correct support of SC (i.e P(SC | SC) > 0.8). X-axis represents the time since the ancestral split. Y-axis represents the relative time the two daughter species remained isolated before the secondary contact. Colors represent the density in simulations with P(SC | SC) > 0.8. B. Two-dimensional space of parameters of the SC model showing simulations leading to the absence of a robust conclusion using ABC. Colors represent the density in simulations with P(NA | SC). https://doi.org/10.1371/journal.pbio.2000234.s002 (TIF) S3 Fig. Relation between synonymous divergence and genetic differentiation. Each grey dot represents a pair of species/populations. Lepus (Spanish and Portuguese populations of Lepus granatensis), Eunicella (Eunicella cavolinii and E. verrucosa) and Crepidula (Crepidula fornicata and Bostrycapulus aculeatus) indicate representative pairs of poorly, intermediately and highly divergent species/populations. Effect of divergence on across-loci variance in FST. Genomic distribution of FST for the Lepus, Eunicella and Crepidula datasets (see S1 Data). https://doi.org/10.1371/journal.pbio.2000234.s003 (TIF) S4 Fig. Relation between net synonymous divergence Da and probability of ongoing gene flow. Net synonymous divergence is the average proportion of differences at synonymous positions between individuals sampled in the two compared species due to mutations occurring after the ancestral split. The “hetero M + Ne” analysis was made by assuming genomic variation for both M and Ne. The “hetero M” analysis solely takes into account genomic variation in introgression rates over the whole genome. The “hetero Ne” analysis solely takes into account genomic variation in Ne. The “homo M + Ne” analysis considers one value of M and one value of Ne shared by the whole genome. Red arrows indicate pairs of species inferred as ambiguous in heteroM (robustness < 0.95), heteroNe and homoM_homoN analysis but not in heteroM_heteroN (robustness ≥ 0.95). Green arrows indicate pairs of species with different and unambiguous inferences (robustness ≥ 0.95) made in heteroM, heteroNe and homoM_homoN when compared to heteroM_heteroN (see S1 Data). https://doi.org/10.1371/journal.pbio.2000234.s004 (TIF) S5 Fig. Relation between gross synonymous divergence Dxy and probability of ongoing gene flow. Gross synonymous divergence is the average proportion of differences at synonymous positions between individuals sampled in the two compared species, including differences present in the ancestral species. The “hetero M + Ne” analysis was made by assuming genomic variation for both M and Ne. The “hetero M” analysis solely takes into account genomic variation in introgression rates over the whole genome. The “hetero Ne” analysis solely takes into account genomic variation in Ne. The “homo M + Ne” analysis considers one value of M and one value of Ne shared by the whole genome. Red arrows indicate pairs of species inferred as ambiguous in heteroM (robustness < 0.95), heteroNe and homoM_homoN analysis but not in heteroM_heteroN (robustness ≥ 0.95). Green arrows indicate pairs of species with different and unambiguous inferences (robustness ≥ 0.95) made in heteroM, heteroNe and homoM_homoN when compared to heteroM_heteroN (see S1 Data). https://doi.org/10.1371/journal.pbio.2000234.s005 (TIF) S6 Fig. Relation between FST and probability of ongoing gene flow. The “hetero M + Ne” analysis was made by assuming genomic variation for both M and Ne. The “hetero M” analysis solely takes into account genomic variation in introgression rates over the whole genome. The “hetero Ne” analysis solely takes into account genomic variation in Ne. The “homo M + Ne” analysis considers one value of M and one value of Ne shared by the whole genome. Red arrows indicate pairs of species inferred as ambiguous in heteroM (robustness < 0.95), heteroNe and homoM_homoN analysis but not in heteroM_heteroN (robustness ≥ 0.95). Green arrows indicate pairs of species with different and unambiguous inferences (robustness ≥ 0.95) made in heteroM, heteroNe and homoM_homoN when compared to heteroM_heteroN (see S1 Data). https://doi.org/10.1371/journal.pbio.2000234.s006 (TIF) S7 Fig. Relation between the estimated Tsplit under the IM model and probability of ongoing gene flow. The “hetero M + Ne” analysis was made by assuming genomic variation for both M and Ne. The “hetero M” analysis solely takes into account genomic variation in introgression rates over the whole genome. The “hetero Ne” analysis solely takes into account genomic variation in Ne. The “homo M + Ne” analysis considers one value of M and one value of Ne shared by the whole genome. Red arrows indicate pairs of species inferred as ambiguous in heteroM (robustness < 0.95), heteroNe and homoM_homoN analysis but not in heteroM_heteroN (robustness ≥ 0.95). Green arrows indicate pairs of species with different and unambiguous inferences (robustness ≥ 0.95) made in heteroM, heteroNe and homoM_homoN when compared to heteroM_heteroN. https://doi.org/10.1371/journal.pbio.2000234.s007 (TIF) S8 Fig. Number of pair of species supporting current isolation, current introgression, or ambiguity in model choice. A pair of species is associated to “current isolation” if the sum of posterior probabilities P(SI) + P(AM) is associated to a robustness ≥ 0.95. A pair of species is associated to “current introgression” if the sum of posterior probabilities P(SC) + P(IM) is associated to a robustness ≥ 0.95. The ambiguous status is attributed to a pair of species when “current isolation” and “current introgression” are not strongly supported. The “homo M + N” analysis was made by assuming an unique genomic introgression rate and an unique Ne over the whole genome. The “hetero M” analysis takes into account genomic variation in introgression rates over the whole genome. The “hetero N” analysis takes into account genomic variation in Ne. The “hetero M + N” analysis takes into account genomic variation in introgression rates and in Ne (see S1 Data). https://doi.org/10.1371/journal.pbio.2000234.s008 (TIF) S9 Fig. Number of pair of species showing evidences for SI, AM, IM, SC, PAN, or ambiguity in model choice for three distinct ABC analyses. A pair of species is associated to SI or AM if its relative posterior probability is greater than 0.8696. A pair of species is associated to IM, SC or PAN tf its relative posterior probability is greater than 0.6419. The “homo M + N” analysis was made by assuming an unique genomic introgression rate and an unique Ne over the whole genome. The “hetero M” analysis takes into account genomic variation in introgression rates over the whole genome. The “hetero N” analysis takes into account genomic variation in Ne. The “hetero M + N” analysis takes into account genomic variation in introgression rates and in Ne (see S1 Data). https://doi.org/10.1371/journal.pbio.2000234.s009 (TIF) S10 Fig. Estimating α, the proportion of loci that introgress, under the IM model. 2,000 pseudo-observed datasets (PODs) were simulated under the IM model with heterogeneity in introgression rates. We estimated the parameters of this model by using the ABC approach described in the ‘Materials and Methods’ section. α is the proportion of the genome crossing the species barrier at a rate N.m > 0. x-axis: values of α used to produce the PODs; y-axis: values of α estimated by ABC from the simulated PODs. Solid line represents f(x) = x. Dotted lines represent f(x) = 2.x and f(x) = x/2 respectively. Estimated values of α for the observed pairs of population/species as a function of their net synonymous divergence. https://doi.org/10.1371/journal.pbio.2000234.s010 (TIF) S11 Fig. Estimating N.m, the effective migration rate, under the IM model. 2,000 pseudo-observed datasets (PODs) were simulated under the IM model with heterogeneity in introgression rates. A. x-axis: values of N.m used to produce the PODs; y-axis: values of N.m estimated by ABC from the simulated PODs. Solid line represents f(x) = x. Dotted lines represent f(x) = 2.x and f(x) = x/2 respectively. B. Estimated values of N.m for the observed pairs of population/species as a function of their net synonymous divergence. https://doi.org/10.1371/journal.pbio.2000234.s011 (TIF) S12 Fig. Estimating N, the effective population size of daughter populations, under the IM model. 2,000 pseudo-observed datasets (PODs) were simulated under the IM model with heterogeneity in introgression rates. A. x-axis: values of N used to produce the PODs; y-axis: current values of N estimated by ABC for all PODs. Solid line represents f(x) = x. Dotted lines represent f(x) = 2.x and f(x) = x/2 respectively. B. Estimated values of N for the observed pairs of population/species as a function of their net synonymous divergence. https://doi.org/10.1371/journal.pbio.2000234.s012 (TIF) S13 Fig. Estimating Nanc, the effective size of the ancestral population, under the IM model. 2,000 pseudo-observed datasets (PODs) were simulated under the IM model with heterogeneity in introgression rates. A. x-axis: values of Nanc used to produce the PODs; y-axis: estimated values of Nanc for all PODs. Solid line represents f(x) = x. Dotted lines represent f(x) = 2.x and f(x) = x/2 respectively. B. Estimated values of Nanc for the observed pairs of population/species as a function of their net synonymous divergence. https://doi.org/10.1371/journal.pbio.2000234.s013 (TIF) S14 Fig. Estimating Tsplit, the time of ancestral subdivision, under the IM model. 2,000 pseudo-observed datasets (PODs) were simulated under the IM model with heterogeneity in introgression rates. Tsplit is expressed in million of generations since the ancestral separation. A. x-axis: values of Tsplit used to produce the PODs; y-axis: estimated values of Tsplit for all PODs. Solid line represents f(x) = x. Dotted lines represent f(x) = 2.x and f(x) = x/2 respectively. B. Estimated values of Tsplit for the observed pairs of population/species as a function of their net synonymous divergence. https://doi.org/10.1371/journal.pbio.2000234.s014 (TIF) S1 Table. Number of populations and species inferred to be isolated or connected by ABC. https://doi.org/10.1371/journal.pbio.2000234.s015 (ODS) S1 Text. Simulation study to test the robustness of ABC in face of recent times of divergence. https://doi.org/10.1371/journal.pbio.2000234.s016 (PDF) S1 Data. Accessions of surveyed individuals, geographic locations and summary statistics. https://doi.org/10.1371/journal.pbio.2000234.s017 (XLSX) Acknowledgments We thank Aude Darracq, Vincent Castric, Pierre-Alexandre Gagnaire, Xavier Vekemans, and John Welch for insightful discussions. The computations were performed at the Vital-IT (http://www.vital-it.ch) Center for high-performance computing of the SIB Swiss Institute of Bioinformatics and the ISEM computing cluster at the platform Montpellier Bioinformatique et Biodiversité.

doi: 10.1371/journal.pbio.1002583pmid: 27977664

Introduction Lysosomal storage disorders (LSDs) represent some of the most difficult of medical challenges, with poorly understood pathologies and only rare treatment options. Despite having the common property of being caused by mutations in lysosomal enzymes, leading to accumulation of substances that would normally be degraded and to more generally compromised lysosomal function, the more than 40 different LSDs differ greatly in their primary tissue pathology, their severity, and in the specific substances that accumulate within compromised cells. The individuality of these diseases is mirrored by the dominant therapeutic strategies, which are generally focused on replacement of missing enzyme activity (by protein administration or gene expression) or on substrate reduction therapies that have the goal of decreasing availability of a precursor for the substance whose degradation is compromised by enzyme mutation [1–39]. Such therapies have proven useful in rare cases [40–43], but progress on therapeutic advances is infrequent and essentially nonexistent for LSDs exhibiting damage to the central nervous system (CNS) [44–46]. In addition, progress has tended to be disease specific rather than providing principles that may apply more broadly. Despite extensive study of LSDs, many critical questions remain unanswered about these diseases. For example, little is known about the biochemical linkage between any particular mutation and lysosomal dysfunction, or even whether there is a direct correlation between accumulation of particular substances and lysosomal dysfunction. In addition, although both lysosomal dysfunction and cellular dysfunctions occur in these diseases, it remains unclear how—or even if—these changes are functionally connected. Moreover, it is unclear whether principles that might be relevant to an individual disease are relevant to the pathology of diseases caused by different mutations. To attempt to discover principles that might be relevant to LSDs caused by different mutations, we have focused on diseases associated with accumulation of lipids that are able to cause a variety of cellular dysfunctions, up to and including cell death, when applied to cells in vitro. Such diseases include Krabbe disease (KD), metachromatic leukodystrophy (MLD), and Gaucher disease [22, 31, 47–55]. Although each of these diseases is associated with accumulation of a different lipid (or lipids) and with different disease pathologies, the effects of these lipids on cellular function are severe enough to suggest that such toxicities may contribute to disease pathogenesis. We now show that a structurally related subset of lipids that accumulate in KD, MLD, or Gaucher disease all induce multiple lysosomal dysfunctions (along with other cellular dysfunctions), thus providing a direct link between enzymatic mutations and lysosomal abnormalities. We further show that it is possible to use drug-repurposing assays to discover single compounds that block a wide range of lipid-induced toxicities. Analysis of the properties of toxic lipids and of protective compounds reveals a previously unsuspected role of lysosomal pH and re-acidification as a potentially valuable therapeutic target. We further provide proof of principle that selecting potential therapies based on their ability to improve lysosomal function without correcting a genetic defect can reveal compounds that offer clinically relevant benefits in a mouse model of a severe LSD. Results Psychosine Disrupts Multiple Cellular Functions in Oligodendrocyte Progenitor Cells We began our studies with an analysis of psychosine (Psy, also referred to as galactosylsphingosine), a lipid that is thought to be of central, and potentially causal, pathogenic importance in KD [56–60]. Psy is one of the most extensively studied of all the lipids known to accumulate in LSDs and is known to exhibit toxicity for multiple cell types in vitro [61–75] and to cause extensive damage when injected intracranially in wild-type (WT) mice [59]. Psy accumulates in tissues of individuals with KD due to galactocerebrosidase ([GALC], EC 3.2.1.46) mutations that cause abnormal processing of lipids that are important components of myelin, the insulating material that enwraps axons in the CNS and peripheral nervous systems (PNS), a primary target of damage in KD. Psy also accumulates in tissues of the naturally occurring, severe murine model of KD, the twitcher mouse [76–80], which also harbors GALC mutations and recapitulates most human pathologies. As a prelude to analyzing the ability of Psy to alter cellular function, we first determined which CNS cells were most vulnerable to this lipid and found that the most sensitive cells were primary oligodendrocyte (OL)/type-2 astrocyte progenitor cells ([O-2A/OPCs], also referred to as OL precursor cells). These progenitors, which give rise to the myelin-forming OLs of the CNS during development and in response to myelin damage, were killed by pathophysiologically relevant low-micromolar (3 μM) concentrations of Psy [77] that had no effect on hippocampal and cortical neuron survival (see also, e.g., [65, 81]) and were as toxic to OLs (Fig 1A). The vulnerability of primary O-2A/OPCs to Psy was also an order of magnitude greater than seen in immortalized CNS glial progenitor cell lines (e.g., [82]) and in Schwann cells of the PNS [83]. This level of sensitivity falls well within the reported Psy concentrations in the CNS of symptomatic (postnatal day [P]25) and moribund (P40) twitcher mice, which are between 15 μM and 34 μM, respectively [77]. Similar concentrations in the twitcher CNS have been reported by other investigators, ranging from 0.7 μM (P10), 4 μM (P16), and 4.5 μM (P20–P25) to as high as 27–50 μM (P30) [78–80]. Comparable concentrations have been reported in postmortem Krabbe patient CNS tissue, ranging from 2.7 μM to 45 μM in cortical grey and white matter, respectively [84, 85]. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Psy causes a diverse array of cellular and biochemical toxicities in cultured O-2A/OPCs. (A) Survival of Psy-treated purified rat O-2A/OPCs, OL, cortical, and hippocampal neurons for 5 d relative to untreated controls. (B) Quantification of the number of rat O-2A/OPCs per clone in the presence or absence of Psy over 5 d. (C) Quantification of the relative amount of neutral lipids and phospholipids in rat O-2A/OPCs exposed to positive controls cyclosporin A (10 μM) or propranalol (10 μM), 100 nM bafilomycin A (BafA), or 3.3 μM Psy for 48 h. (D) Quantification of time to half-maximal intensity for rat O-2A/OPCs exposed to Psy (1 μM) or vehicle (DMSO) for 24 h before addition of fluorescent nanobeads. A plot of relative intensity is also shown; lines indicate time to half-maximal intensity. (E) Quantification of relative Cathepsin B and D activities in rat O-2A/OPCs exposed to the indicated drugs or 1 μM Psy for 24 h. (F) Quantification of lysosomal pH in rat O-2A/OPCs exposed to 500 nM BafA, 10 μM chloroquine (CQN), 10 mM ammonium chloride (NH4Cl), or 1 μM Psy for 24 h or 48 h. Data for all graphs displayed as mean ± standard error of the mean (SEM); *p < 0.05, **p < 0.01, †p < 0.001 versus control. Data presented in this figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.1002583.g001 Further studies revealed that O-2A/OPCs exposed to still lower (1 μM) levels of Psy exhibited multiple abnormalities of potential relevance to understanding the decreased myelination and apparent failure to repair myelin damage seen in KD. In the absence of cell death, 1 μM Psy suppressed both cell division (Fig 1B) and differentiation of O-2A/OPCs into OLs (S1A Fig). It also disrupted cytoskeletal integrity and caused decreased cell migration (S1B and S1C Fig). Such sensitivity places these cells among those most sensitive to the effects of Psy exposure. Psy Alters Lysosomal Function in O-2A/OPCs We next discovered that exposure to 1 μM Psy has the previously unrecognized ability to cause multiple alterations in lysosomal function, indicating that this lipid may provide a direct link between enzymatic mutation and lysosomal dysfunction in KD. Exposure to Psy caused abnormalities in lipid homeostasis, endolysosomal transport, and cathepsin activity. Exposure to 1μM Psy disrupted lipid homeostasis, causing the intracellular accumulation of both neutral triglycerides and phospholipids (Fig 1C, S1D Fig). Endolysosomal transport was also compromised by exposure to 1 μM Psy, as shown by a decreased rate of endocytic import of fluorescently labeled polystyrene nanobeads (time to half-maximal staining intensity: 4.6 ± 1.0 min for vehicle-treated control versus 22.2 ± 5.7 min for Psy, p < 0.05; Fig 1D, S1E Fig). Psy exposure also increased the activity of resident lysosomal proteases cathepsin D and B, which can cause cellular damage or death upon export to the cytoplasm and the activities of which are known to be elevated in a number of LSDs (Fig 1E) [86–88]. Psy’s ability to disrupt lysosomal function was as great as that seen with bafilomycin A (BafA), which disrupts lysosomal function by antagonizing the lysosomal vacuolar-type H+-ATPase [89]. Exposure of O-2A/OPCs to 1 μM Psy significantly increased intralysosomal pH from 4.88 ± 0.04 to 5.62 ± 0.08 after 24 h, an increase maintained for at least 48 h after a single exposure (p < 0.001; Fig 1F). This elevation in lysosomal pH was observed in both fixed (Fig 1F) and live (S1F Fig) O-2A/OPCs. Psy exposure was as potent at increasing lysosomal pH as multiple compounds well known to exert such effects, including BafA, chloroquine, or the weak base NH4Cl (Fig 1F) [90]. This increase was evident within minutes of exposure to Psy and was comparable to treatment with BafA (S1G Fig, S1–S3 Movies), and the effects on lysosomal pH were sustained over 24–48 h after Psy exposure. Unbiased Screening Identifies Chemically Diverse Candidate Protective Agents That Prevent Psy-Induced Cellular and Lysosomal Dysfunctions To identify potential means of preventing Psy-induced toxicities that might be suitable for eventual clinical utilization, we conducted unbiased analysis of multiple concentrations of 1,040 mostly United States Food and Drug Administration (FDA)-approved small molecules [91] and a custom panel of 12 growth factors with known neuroprotective activity. We examined prevention of Psy-induced suppression of O-2A/OPC division in these analyses (Fig 2A). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Unbiased screening identifies chemically diverse candidate protective agents that reduce Psy toxicities. (A) Schematic depicting the work flow for unbiased screening with Celigo adherent cytometer (Nexcelom). (B) Representative plot of relative cell number over 5 d for rat O-2A/OPCs exposed to 1 μM Psy or vehicle (DMSO). Quantification of the relative proliferation rate for all 1,040 drugs at three concentrations over 5 d. Blue: “hit” drugs selected for further validation. A bar graph quantifying the relative proliferation rate of all vehicle and Psy controls is included. (C) Quantification of the relative proliferation rate of rat O-2A/OPCs exposed to 1.5 μM Psy, with and without verified pro-division “hits,” over 5 d. (D) Quantification as in (C) for cells exposed to Psy 48 h before exposure to the indicated drugs. (E) Summary of clinical metadata for lead “hits” as % of total. (* worldwide approval, including FDA. ** reported blood–brain barrier permeability; not all drugs have been examined/reported.) (F) Quantification of cell survival in cells exposed to 3.3 μM Psy for 5 d, with and without the indicated drugs, for 5 d. (G, H) Quantification of the (G) relative survival and (H) number of rat O-2A/OPCs per clone exposed to of Psy (1 μM in H), with and without 100 ng/mL insulin-like growth factor (IGF-1), for 5 d. (I–K) Quantification of (I) neutral lipid and phospholipid accumulation, (J) cathepsin B and D activities in rat O-2A/OPCs exposed to the indicated drugs, with and without 1 μM Psy, and (K) time to half-maximal intensity for endocytosis of fluorescent nanobeads for (I) 2 d or (J, K) 24 h. (L) Quantification of lysosomal pH for rat O-2A/OPCs exposed to 1 μM Psy, with and without the indicated “hits” (blue) or “non-hits” (gray), for 24 h. NT-3: 10 ng/mL; 1E04: 5 μM; 5F05, 9A06, 9H10: 1 μM. ap < 0.05, bp < 0.01 versus Psy-only. Data for all graphs displayed as mean ± SEM; ns = not significant; *p < 0.05, **p < 0.01, †p < 0.001 versus untreated; ap < 0.05, bp < 0.01, cp < 0.001 versus Psy-only treatment. See also S2 Fig, S1 and S2 Tables for drug names and concentrations. Data presented in this figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.1002583.g002 We found 16 structurally and functionally diverse compounds (S2A Fig, S1 and S2 Tables), in addition to 4 growth factors, that had the unexpected properties of rescuing cell division (Fig 2B and 2C, S2B Fig). Eight of the 9 most protective agents were effective at rescuing cell division even when their administration was delayed 48 h after Psy exposure (Fig 2D). Importantly, all small molecules were optimally protective at physiologically achievable concentrations (i.e., nanomolar to low micromolar), and most are approved for use in humans (94%) and are blood–brain barrier permeable (>80%) (Fig 2E). Five agents (chlorotrianisene [1G05], NKH-477 [(9C06), also known as colforsin], clofoctol [8D08], tulobuterol [9E07], and insulin-like growth factor [IGF-1]) revealed by our screens not only significantly rescued cell division but were also able to reduce cell death caused by exposure to higher concentrations of Psy (Fig 2F). With the exception of IGF-I, none of our compounds of interest were previously identified as being able to protect against toxicity of Psy (or of other lipids accumulating in LSDs). Even in the case of IGF-I, previous studies reported that supraphysiological (>10 μg/mL) concentrations decreased Psy-induced apoptosis in OLs [92] and in an O-2A/OPC cell line [82]. In our studies, by contrast, 100 ng/mL IGF-1 shifted Psy’s cytotoxicity curve by an order of magnitude and significantly deceased Psy-dependent suppression of self-renewal (Fig 2G and 2H). Further examination of three of the most protective agents—clofoctol, NKH-477, and IGF-I—demonstrated that these compounds also prevented Psy-induced alterations in lysosomal function. All three suppressed Psy-induced increases in lipid accumulation and cathepsin activity and restored normal endocytosis (Fig 2I–2K). Moreover, all three compounds significantly decreased Psy-dependent increases in pH (Fig 2L). In contrast, five randomly selected molecules that did not significantly reduce any Psy toxicities in our screens (caffeine [1E04], acetarsol [5F05], mepartricin [9A06], avobenzone [9H10], and neurotrophin-3; S1 Table) had no effect on Psy-induced increases in lysosomal pH. Protective Agents Converge on a Limited Number of Common Necessary Pathways for Their Activity As the protective compounds we discovered are structurally and functionally diverse, we next attempted to define regulatory pathways on which these agents might converge to confer their protective activity. To do this, we focused on multiple signaling pathways previously described to be antagonized by exposure of cells to Psy (e.g., mitogen-activated protein kinase [MAPK], phosphoinositide-3-kinase [PI3k]/Akt, protein kinase C [PKC], cyclic Amp (cAMP)-dependent signaling [51, 74, 82, 93–95]), as well as other proteins that have been implicated in mediating stress responses in O-2A/OPCs, including Jun N-terminal kinase (Jnk) [74], mammalian target of rapamycin (mTOR) [96–98], and estrogen receptor [99–101]. We next generated a secondary screen consisting of pharmacological inhibitors to components of these various signaling pathways. In these experiments, O-2A/OPCs were exposed to Psy; a combination of Psy and a protective agent; or a combination of Psy, a protective agent, and one of 15 pharmacological inhibitors targeting important signaling pathways and proteins (S3 Table). This allowed for the generation of a compound-specific “fingerprint of protection” that revealed putative signaling pathways used by the candidate compound to overcome Psy-induced suppression of division (e.g., Fig 3A and 3B). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Protective agents converge on a limited number of common necessary pathways for their activity. (A, B) Representative “fingerprint of protection” for candidate drug 1G05. Quantification of the relative proliferation rate of rat O-2A/OPCs exposed to 1.5 μM Psy, with and without 100 nM 1G05, and with and without pharmacological inhibitors targeting the indicated protein, for 5 d. To account for differences in their ability to reduce Psy-induced suppression of division between candidate agents, relative changes in cell division were normalized as in (B). (C) Unsupervised hierarchical clustering of “fingerprints” for the indicated drugs. Data for all graphs displayed as mean ± SEM. See also S1 and S2 Tables for drug names and concentrations, and S3 Table for details on the “fingerprinting” screen. Data presented in this figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.1002583.g003 We generated fingerprints for 12 of the most efficacious compounds in reducing Psy-induced suppression of division across 5 d using this approach. The results were then hierarchically clustered to identify similarities and dissimilarities between individual compound fingerprints and between the signaling pathways implicated in protection (Fig 3C). Despite structural and functional diversity among candidate protective agents, there was striking similarity in the signaling pathways needed for protection. The activity of the diverse protective agents was antagonized by pharmacological inhibition of the Ras/rapidly accelerated fibrosarcoma gene (Raf)/MAPK pathway, Akt, estrogen receptor, protein kinase A (PKA), and geranylgeranyl transferase ([GGT], which is needed for activation of small GTPases that are involved in cell division and migration). Despite their structural diversity, there was a surprisingly high degree of correlation between groups of small molecules; the cluster of structurally and functionally unrelated drugs 2G08 (ethopropazine, an antiparkinsonian drug), 2F11 (estradiol valerate, a synthetic estrogen), and 8D08 (clofoctol, an antibiotic), for example, showed the highest degree of similarity (correl. = 0.97) (S3 Fig). Protection Against Psy-Induced Lysosomal and Cellular Dysfunctions Can Be Provided by Promotion of Lysosomal Re-acidification Lysosomal ion homeostasis, maintained through the activity of several channels and transporters, is critical to the normal function of lysosomes. For example, H+ import is necessary for the maintenance of an acidic pH [102] and is achieved through the activity of the V-ATPase, Ca2+ is important for vesicle trafficking [103] and fusion [104], Na+ and K+ are required for the regulation of membrane potential [105, 106], and Cl− serves as a counterion to regulate lysosomal membrane potential and to facilitate the acidification of the lysosome lumen [107–109]. Although any of these may be potential therapeutic targets, we focused our attention on identifying those channels or transporters regulated by signaling pathways uncovered through our fingerprinting analysis. Of the several pathways that are required for activity of our protective agents, the one for which there is a clearly established linkage to at least one aspect of lysosomal function is the requirement for PKA activity. Previously, it has been shown that cAMP can promote lysosomal re-acidification [110], as can PKA, which is activated by cAMP [111]. In addition, we found that increases in cAMP not only normalized lysosomal pH but also prevented Psy-induced decreases in O-2A/OPC division (Fig 4A and 4B), raising the theoretical possibility that intervention at this point would provide additional benefits beyond that of pH restoration. We therefore focused attention on the role of lysosomal re-acidification as a potential therapeutic target. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. Promotion of lysosomal re-acidification is critical in protecting from multiple aspects of Psy toxicity. (A–D) Quantification of the lysosomal pH and proliferation rate in rat O-2A/OPCs exposed to (A, C) 1 μM Psy for 24 h, (B, D) 1.5 μM for 5 d, with and without (A, B) 1 mM cAMP (or 10 μM forskolin), or (C, D) 333 nM RP-107 and 1 μM cystic fibrosis transmembrane conductance regulator (CFTR) inhibitor 172 (CFTRi-172). (E) Quantification of neutral lipid and phospholipid accumulation in rat O-2A/OPCs exposed to the indicated drugs, with and without 1 μM Psy, for 2 d. (F, G) Quantification of the lysosomal pH and proliferation rate in rat O-2A/OPCs exposed to 1 μM Psy for (F) 24 h or (G) 5 d, with and without 100 nM KT-5720 or 3.3 μM H89 (cAMP: 1 mM). Data for all graphs displayed as mean ± SEM; *p < 0.05, **p < 0.01, †p < 0.001 versus untreated; ap < 0.05, bp < 0.01, cp < 0.001 versus Psy-only treatment. See also S1 and S2 Tables for drug names and concentrations. Data presented in this figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.1002583.g004 The most attractive explanation for how cAMP/PKA activity could restore lysosomal pH would be through activation of the cystic fibrosis transmembrane conductance regulator (CFTR), a PKA-activated transmembrane chloride channel that promotes lysosomal re-acidification [112]. Unlike the CLC-7 Cl−/H+ antiporter, another chloride channel that is localized to the lysosomal membrane and thought to play a role in the basal maintenance of lysosomal pH [113], the CFTR channel appears only to be critical for re-acidification. Moreover, although the CFTR can be activated by PKA and cAMP, there is no evidence for such activation of CLC-7. In addition, specific agonists and inhibitors exist for the CFTR, enabling a direct test of whether promoting re-acidification can prevent Psy-induced toxicity. As we predicted, treatment of cells with the cAMP-independent CFTR agonist RP-107 [114] restored lysosomal pH in cells exposed to Psy. Although control of lysosomal pH and/or lysosomal re-acidification has not been thought to have any upstream role in the multiple cellular dysfunctions caused by Psy exposure, we nonetheless found that RP-107 protected against Psy-induced suppression of division, as well as elevated storage of both neutral lipids and phospholipids (Fig 4C–4E). To test the hypothesis that these benefits were not due to off-target effects of RP-107, we also co-exposed cells to CFTR-inhibitor 172 (CFTRi-172) [115], which attenuated the protective effects of RP-107 treatment (Fig 4C–4E). The effects of RP-107 were CFTR dependent, as determined by knockdown of CFTR in O-2A/OPCs using small interfering RNA (siRNA) pools targeting rat CFTR, as well as a pool of nontargeting (NT) siRNAs as a control for transfection; the reduction in CFTR protein levels was confirmed by western blot analysis. Knockdown of CFTR did not significantly affect lysosomal pH when compared to cells exposed to NT controls (4.96 ± 0.13 versus 4.81 ± 0.11, respectively). Moreover, in the presence of Psy, in both NT and CFTR siRNA pools, there was a significant increase in lysosomal pH (5.47 ± 0.08 versus 5.61 ± 0.09, respectively), with no significant difference between these two treatment groups. However, when we tested the effect of RP-107, a specific CFTR agonist, we found that lysosomal pH was significantly reduced in cells exposed to NT siRNA but that CFTR knockdown attenuated RP-107’s protective effect (5.18 ± 0.09 versus 5.62 ± 0.03, p < 0.01; S4A Fig). Thus, as with our pharmacological experiments, genetic loss of CFTR does not appear to significantly affect basal lysosomal pH in untreated cells. However, the protective capacity of RP-107 is CFTR dependent. These results are consistent with the original studies demonstrating the role of the CFTR in control of lysosomal re-acidification [112]. Moreover, we found that the most effective protective agents identified in our studies did not themselves reduce the basal acidic pH of lysosomes in the absence of Psy (S4B Fig) but instead seemed to work to promote re-acidification. Indeed, their ability to normalize lysosomal pH, as well as rescue cell division, in cells exposed to Psy was blocked by co-exposure to inhibitors of PKA (Fig 4F and 4G). As these protective agents are able to rescue cells even when applied 48 h after Psy exposure (Fig 2D), it appears that their protective activity is not mediated simply by blocking lysosomal alkalization. Psy’s Free Amine Group Is Critical for Its Toxicity The observations that multiple Psy-induced lysosomal and cellular dysfunctions can be prevented by lysosomal re-acidification with RP-107 (Fig 4C–4E) and that Psy exposure causes rapid increases in lysosomal pH (Fig 1F, S1D Fig), raise complementary questions about how Psy causes such changes. One possibility is that Psy disrupts the function of particular proteins involved in lysosomal re-acidification, but another possibility is that structural features of Psy itself are directly relevant to understanding effects on lysosomal pH. Although multiple studies have attempted to understand the molecular mechanisms underlying Psy’s toxicity [61–75], we noted that Psy has unusual physicochemical features that might be of relevance to understanding its effects on lysosomes. Psy is unusual as a cationic, weakly basic lipid, carrying a net positive charge at physiological pH. With a pKa value of 7.18 [116], Psy is predicted to be 99.9% protonated in the acidic pH of the lysosome. If this aspect of Psy’s structure is important in altering lysosomal and cellular function, then the protonatable free amine group on Psy should be critical in mediating the changes in lysosomal pH that we observed. We therefore tested whether removing this free amine group altered effects on lysosomal pH and on other outcomes of Psy exposure. We found that the free amine group on Psy is critical in its ability not only to disrupt lysosomal pH but also to cause other toxic effects. We compared Psy toxicity to that of N-acetyl-Psy (N-AcPsy), a structural derivative containing an amide-linked acetyl group, rendering it no longer protonatable (Fig 5A). Unlike Psy, N-AcPsy did not induce cell death or alter O-2A/OPC self-renewal at similar concentrations (Fig 5B and 5C). Moreover, N-AcPsy did not elevate neutral lipid and phospholipid storage, increase endocytic transport time, increase cathepsin activity, or elevate lysosomal pH (Fig 5D–5G). Thus, the positively charged free amine group present on Psy was critical to increasing lysosomal pH and also to the subsequent lysosomal and cellular impairments observed after exposure in O-2A/OPCs. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. Psy’s free amine group is critical for its toxicity. (A) Structures of Psy and N-AcPsy. (B) Quantification of rat O-2A/OPC survival in response to Psy or N-AcPsy. (C–G) Quantification of rat O-2A/OPC self-renewal, lipid accumulation, endocytosis, cathepsin activity, and lysosomal pH in cells exposed to 1 μM Psy or N-AcPsy for (C) 5 d, (D) 2 d, or (E–G) 24 h. Data for all graphs displayed as mean ± SEM; ns = not significant; *p < 0.05, †p < 0.001 versus control. Data presented in this figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.1002583.g005 Structurally Related Lysosphingolipids from Multiple LSDs Alter Lysosomal Function To further test the hypothesis that the presence of free amine group on a cationic lipid is critical to lipid-induced toxicities, and that such lipids provide a direct link between enzymatic mutation and lysosomal disruption, we examined a series of lipids known to accumulate in other LSDs. Other lipids of potential interest include lyso-sulfatide (lyso-SF) (which accumulates in MLD [49]), glucosylsphingosine (GlcSph) and glucosylceramide (GlcCer) (which accumulate in Gaucher disease [47]), and lactosylsphingosine (LacSph) and lactosylceramide (LacCer), which accumulate in several LSDs (Fig 6A) [78, 117, 118]. Some of these lipids appear to have been only rarely studied for their effects on cell function in vitro (lyso-SF, GluSph, LacSph, LacCer) [119]. In the case of Gaucher disease, the majority of previous in vitro studies appears to have focused on GlcCer, and studies on both GlcCer and GlcSph often have required lipid concentrations severalfold greater than those at which Psy’s effects were observed (e.g., [22, 50, 51, 53, 55, 120–122]). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 6. Lysosphingolipids accumulating in other LSDs suppress critical O-2A/OPC behaviors and lysosomal function. (A) Chemical structures of the indicate lipids. Gal: galactose; Glc: glucose; R: variable hydrocarbon chain. (B–G) Quantification of (B) the relative survival, (C) the relative number of O-2A/OPCs per clone, (D) lipid accumulation, (E) endocytic import time, (F) cathepsin activity, and (G) lysosomal pH in rat O-2A/OPCs exposed to the indicated lipids for (B and C) 5 d, (D, E, G) 24 h, or (F) 2 d. Data for all graphs displayed as mean ± SEM; ns = not significant; *p < 0.05, **p < 0.01, †p < 0.001 versus untreated, unless otherwise indicated. S4 Table for lipid concentrations used in (C–G). Data presented in this figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.1002583.g006 In order to eliminate differences in cell types as potential contributors to different outcomes, we examined the survival and self-renewal of O-2A/OPCs exposed to lyso-SF, GlcSph, GlcCer, LacSph, and LacCer. Use of these cells also provided a test of the hypothesis that the structure of a lipid is of primary importance in determining toxicity. We also examined the effects of N-acetyl-sulfatide (N-AcSF) as a direct comparison with N-AcPsy. We found that sphingosine-derived lipids that accumulate in different LSDs and that contain a free amine group (and thus are structurally similar to Psy) caused significant cell death and suppression of self-renewal (Fig 6B and 6C) at similarly low lipid concentrations as we observed with Psy. In contrast, exposure to their ceramide-based counterparts GlcCer and LacCer, or to N-AcSF, did not cause cellular toxicities at comparable or 10-fold higher concentrations (Fig 6B and 6C). We also found that lysosphingolipids accumulating in other LSDs [22, 31, 47–55, 123–127] had similar effects as Psy on lysosomal function. Exposure to sublethal concentrations of GluSph, lyso-SF, or LacSph caused increases in neutral lipid and phospholipid accumulation, endocytic transport time, cathepsin activity, and lysosomal pH. In contrast, exposure to their non-lyso counterparts did not have such effects (Fig 6D–6G). If the hypotheses are correct that other toxic lysosphingolipids that accumulate in LSDs work through similar mechanisms as Psy, and that such mechanisms are relevant to understanding the efficacy of our protective agents, then our protective agents also should rescue cells from the toxic effects of lipids from other LSDs. If correct, such findings would provide both the first structural predictors of toxicity and the first example of protective agents of potential relevance in different LSDs. We found that our candidate protective agents also reduced the toxic effects of GlcSph, lyso-SF, and LacSph (Fig 7A). Three of our most effective agents—IGF-1, clofoctol (8D08), and NKH-477 (9C06)—prevented lipid-induced suppression of division and also attenuated increases in neutral lipid and phospholipid accumulation, endocytic import time, cathepsin activities, and lysosomal pH in rat O-2A/OPCs exposed to sublethal concentrations of GlcSph, lyso-SF, or LacSph (Fig 7B–7G, S5A Fig). These agents also rescued cell division in cells exposed to GlcSph or Lyso-SF for 48 h before addition of protective agents (S5B Fig). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 7. Candidate protective agents reduce multiple lysosphingolipid-induced lysosomal and cellular toxicities. (A) Venn diagram summarizing the number of protective drugs, including IGF-1, that reduce suppression of division in rat O-2A/OPCs exposed to the indicated lyso-lipids for 5 d. (B–E) Quantification of (B) lipid accumulation, (C) endocytic import time, (D) cathepsin activity, and (E) lysosomal pH in rat O-2A/OPCs exposed to the indicated lyso-lipids, with and without 100 ng/mL IGF-1, 100 nM 8D08, or 333 nM 9C06, for (B) 2 d or (C–E) 24 h. Data for all graphs displayed as mean ± SEM; ns = not significant; *p < 0.05, **p < 0.01, †p < 0.001 versus untreated; ap < 0.05, bp < 0.01 versus Psy-only treatment, unless otherwise indicated. See also S4 Fig and S1 and S2 Tables for drug names and concentrations and S4 Table for lipid concentrations. Data presented in this figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.1002583.g007 Lysosphingolipids Disrupt Human OL Progenitor Cell Behaviors, and Protective Compounds Rescue Human Cells We next examined the question of whether the principles revealed in our studies on cells derived from the CNS were applicable to human cells. In these experiments, we used an anti-CD140a (PDGFRα) antibody to enrich for a population of human O-2A/OPCs from the corpus callosal field of mid-gestation fetal tissue (S6 Fig) [128] and exposed cells to Psy and potential protective agents as for rat-derived cells. Exposure to lysosphingolipids caused death in human cells at concentrations comparable to those used in rat progenitor cells (S4 Table), as well as suppression of cell division and elevation of lysosomal pH at sublethal concentrations, whereas their non-lyso counterparts did not cause similar toxicities (Fig 8A–8C). Notably, cell division and normalization of lysosomal pH were restored in cells exposed to Psy with clofoctol, NKH-477, and IGF-I, as we observed for rat O-2A/OPCs (Fig 8D and 8E). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 8. Lysosphingolipids disrupt human O-2A/OPC behaviors. (A–C) Quantification of (A) the relative survival, (B) the relative proliferation rate, and (C) lysosomal pH of human O-2A/OPCs exposed to the indicated lipids for (A, B) 5 d or (C) 24 h. BafA: 100 nM. (D, E) Quantification of the relative proliferation rate and lysosomal pH in human O-2A/OPCs exposed to 1 μM Psy for (D) 5 d or (E) for 24 h, with and without the indicated drugs. ap < 0.05 versus Psy-only. Data for all graphs displayed as mean ± SEM; ns = not significant; *p < 0.05, **p < 0.01, †p < 0.001 versus untreated control; ap < 0.05, bp < 0.01 versus Psy-only treatment. See also S5 Fig and S1 and S2 Tables for drug names and concentrations, and S4 Table for lipid concentrations. Data presented in this figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.1002583.g008 Alterations in O-2A/OPC Biology in Twitcher Mice Are Like Those Induced By Psy Exposure In Vitro In the final section of our studies, we asked whether discoveries made on WT cells exposed exogenously to Psy in vitro revealed principles applicable to cells with an LSD-relevant mutation, both in respect to cellular pathologies and to rescue of lysosomal function. These studies were carried out using twitcher mice, a naturally occurring model of KD that recapitulates most human symptoms. Multiple studies have demonstrated that this mouse is a reliable model of KD in respect to enzymatic dysfunction and tissue pathology [77–80, 129, 130] and is also one of the most useful models for studying LSDs in general. In particular, twitcher mice progress from a lack of apparent pathology to severe disease over a relatively rapid time course, with function appearing to be normal at birth, followed by disease symptoms manifesting about 20 d after birth and with death ensuing at about 42 d after birth. This time course allows pathology and the effects of treatment to be studied at different stages of disease progression. In our studies on twitcher mice, we first determined that changes in O-2A/OPC function were like those induced by exposure to low doses of Psy in vitro. We found significant reductions in both myelin content and OL cell number (OLs; Olig2+/GST+) in the corpus callosum—the major myelinated tract of the CNS—at P40 when compared to age-matched WT littermates (Fig 9A and 9B), consistent with previous analyses of human and twitcher tissue [76, 131]. We also observed a significant reduction in the percentage of dividing (Ki67+) O-2A/OPCs (54.0% ± 1.9% of WT, p < 0.01; Fig 9C) at this late time point, during which time O-2A/OPCs should be undergoing rapid expansion through cell division to replace damaged OLs and myelin. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 9. NKH-477, a protective compound identified in vitro, protects against multiple toxicities in treated twitcher mice. (A–C) Quantification of (A) fluoromyelin intensity, (B) number of GSTpi+/Olig2+ OLs, and (C) the relative number of dividing Ki67+/Olig2+ O-2A/OPCs in the corpus callosa of P40 twitcher mice and age-matched WT littermates (n = 3 from different litters). (D) Analysis of clonal composition of P15 twi O-2A/OPCs and WT littermates across 5d. (E) Quantification of the relative number of dividing Ki67+/Olig2+ O-2A/OPCs in the corpus callosa of P15 twitcher mice and age-matched WT littermates (n = 3 from different litters). (F) Quantification of lysosomal pH of O-2A/OPCs acutely isolated from P17 twitcher and WT mice. (G) Overview of treatment paradigm and clinical course for twitcher mice. (H) Quantification of lysosomal pH of O-2A/OPCs isolated from P35 treated mice. (I) Quantification of the number of dividing callosal O-2A/OPCs in P35 treated mice. (J–L) Representative confocal images of fluoromyelin-stained corpus callosa of the indicated treatment groups at P40, in addition to quantification of staining intensity and the number of OLs. (M) Kaplan–Meyer survival curve for treated and untreated twi mice. Median survival of twi mice, with dotted lines indicating reported median survival of single-therapy treatments [15]. (N–Q) Quantification of travel speed, stance time, beam traverse time, and relative weights for P25 saline-treated WT and twi, as well as NKH-477–treated twi, mice. (R) Quantification of brain Psy levels at P35. Data for all graphs displayed as mean ± SEM; ns = not significant; *p < 0.05, †p < 0.001 versus WT; bp < 0.01 versus vehicle-treated twi. Data presented in this figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.1002583.g009 We additionally found that O-2A/OPC function was compromised in presymptomatic twitcher mice in ways similar to those induced by Psy exposure. We isolated O-2A/OPCs from presymptomatic P15 twitcher mice and examined their self-renewal capacity in vitro. These cells showed impaired self-renewal in comparison with cells of age-matched WT cells when maintained in vitro for 5 d (Fig 9D). Such findings were mirrored by significant reductions in the pool of dividing O-2A/OPCs in vivo at P15 (Fig 9E). To determine whether cells harboring a mutant GALC gene exhibit changes in lysosomal pH, we examined the endolysosomal pH of corpus callosal O-2A/OPCs acutely isolated from presymptomatic twitcher mice (P17). We found that the lysosomal pH was significantly less acidic than that of cells isolated from age-matched WT littermates (Fig 9F), similar to what was observed in vitro with exogenous Psy treatment (Fig 1F). Thus, O-2A/OPCs isolated at developmental time points in which symptoms are not obvious (prior to P18–20) show altered lysosomal pH and alterations in critical cellular behaviors like those induced by exposing WT cells to Psy in vitro. NKH-477 Protects against Multiple Toxicities in Twitcher Mice We next investigated whether the analytical approach employed in our in vitro studies could identify compounds able to provide clinically relevant benefits in vivo. We focused our studies on NKH-477 (9CO6), a water-soluble derivative of forskolin that is approved for treatment of acute heart failure in Japan [132], as this agent is known to be CNS penetrant and elevates cAMP levels (through direct activation of adenylyl cyclase) in brains of rats after systemic administration [133]. Moreover, unlike the other identified protective agents, the linkage of NKH-477 to PKA regulation (and thus to lysosomal re-acidification) is both defined and mediated through widely expressed proteins, consequentially not requiring cells to express specialized drug-targeted receptors in order to be responsive. We initiated treatment at P10, a time when CNS concentrations of Psy are already approaching the range at which we see effects on O-2A/OPCs [78–80], using once-daily intraperitoneal (IP) injections (1 mg/kg; Fig 9G). This is a point in time when disruptions in neuronal function can already be observed in twitcher mice [134], raising the possibility of initiating treatment only after subtle clinical changes are first observable. This delayed initiation of treatment is in marked contrast with the well-studied need to initiate the application of bone marrow transplantation and/or gene therapy in the first few days after birth in order to obtain benefit [1, 10, 16, 40, 130]. The primary endpoints of interest in our in vivo studies were whether we could rescue lysosomal and cellular function in O-2A/OPCs and whether once-daily treatment with NKH-477 is sufficient to provide benefit on both parameters. O-2A/OPCs were isolated at P35 to examine the effects of NKH-477 treatment on lysosomal pH, and we found a normalization of pH in cells isolated from treated twitcher mice when compared to vehicle-treated mice (Fig 9H). NKH-477–treated twitcher mice also showed an increase in the numbers of dividing O-2A/OPCs at P35 to near-normal levels, as well as increases in myelin content and increased OL cell numbers at moribund ages, when compared to vehicle-treated twitcher mice, again to levels not significantly different from WT mice (Fig 9I–9L). Remarkably, we also found that NKH-477 treatment provided significant lifespan extension that was comparable to published single-therapy treatments aimed at restoring GALC activity, including bone marrow transplantation (the current standard of care in patients) or viral-mediated gene therapy (Fig 9M) [1, 2, 10, 13, 15, 16]. Moreover, twitcher mice that received daily injections of NKH-477 also showed significantly improved locomotor and gait function (Fig 9N–9P, S7 Fig) and significantly improved weight gain throughout their lifespan when compared to vehicle-treated twitcher littermates (Fig 9Q). These benefits were observed despite the fact that we were not correcting the genetic defect; indeed, we did not find that NKH-477 treatment reduced the overall tissue burden of Psy in the CNS (Fig 9R). Psychosine Disrupts Multiple Cellular Functions in Oligodendrocyte Progenitor Cells We began our studies with an analysis of psychosine (Psy, also referred to as galactosylsphingosine), a lipid that is thought to be of central, and potentially causal, pathogenic importance in KD [56–60]. Psy is one of the most extensively studied of all the lipids known to accumulate in LSDs and is known to exhibit toxicity for multiple cell types in vitro [61–75] and to cause extensive damage when injected intracranially in wild-type (WT) mice [59]. Psy accumulates in tissues of individuals with KD due to galactocerebrosidase ([GALC], EC 3.2.1.46) mutations that cause abnormal processing of lipids that are important components of myelin, the insulating material that enwraps axons in the CNS and peripheral nervous systems (PNS), a primary target of damage in KD. Psy also accumulates in tissues of the naturally occurring, severe murine model of KD, the twitcher mouse [76–80], which also harbors GALC mutations and recapitulates most human pathologies. As a prelude to analyzing the ability of Psy to alter cellular function, we first determined which CNS cells were most vulnerable to this lipid and found that the most sensitive cells were primary oligodendrocyte (OL)/type-2 astrocyte progenitor cells ([O-2A/OPCs], also referred to as OL precursor cells). These progenitors, which give rise to the myelin-forming OLs of the CNS during development and in response to myelin damage, were killed by pathophysiologically relevant low-micromolar (3 μM) concentrations of Psy [77] that had no effect on hippocampal and cortical neuron survival (see also, e.g., [65, 81]) and were as toxic to OLs (Fig 1A). The vulnerability of primary O-2A/OPCs to Psy was also an order of magnitude greater than seen in immortalized CNS glial progenitor cell lines (e.g., [82]) and in Schwann cells of the PNS [83]. This level of sensitivity falls well within the reported Psy concentrations in the CNS of symptomatic (postnatal day [P]25) and moribund (P40) twitcher mice, which are between 15 μM and 34 μM, respectively [77]. Similar concentrations in the twitcher CNS have been reported by other investigators, ranging from 0.7 μM (P10), 4 μM (P16), and 4.5 μM (P20–P25) to as high as 27–50 μM (P30) [78–80]. Comparable concentrations have been reported in postmortem Krabbe patient CNS tissue, ranging from 2.7 μM to 45 μM in cortical grey and white matter, respectively [84, 85]. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Psy causes a diverse array of cellular and biochemical toxicities in cultured O-2A/OPCs. (A) Survival of Psy-treated purified rat O-2A/OPCs, OL, cortical, and hippocampal neurons for 5 d relative to untreated controls. (B) Quantification of the number of rat O-2A/OPCs per clone in the presence or absence of Psy over 5 d. (C) Quantification of the relative amount of neutral lipids and phospholipids in rat O-2A/OPCs exposed to positive controls cyclosporin A (10 μM) or propranalol (10 μM), 100 nM bafilomycin A (BafA), or 3.3 μM Psy for 48 h. (D) Quantification of time to half-maximal intensity for rat O-2A/OPCs exposed to Psy (1 μM) or vehicle (DMSO) for 24 h before addition of fluorescent nanobeads. A plot of relative intensity is also shown; lines indicate time to half-maximal intensity. (E) Quantification of relative Cathepsin B and D activities in rat O-2A/OPCs exposed to the indicated drugs or 1 μM Psy for 24 h. (F) Quantification of lysosomal pH in rat O-2A/OPCs exposed to 500 nM BafA, 10 μM chloroquine (CQN), 10 mM ammonium chloride (NH4Cl), or 1 μM Psy for 24 h or 48 h. Data for all graphs displayed as mean ± standard error of the mean (SEM); *p < 0.05, **p < 0.01, †p < 0.001 versus control. Data presented in this figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.1002583.g001 Further studies revealed that O-2A/OPCs exposed to still lower (1 μM) levels of Psy exhibited multiple abnormalities of potential relevance to understanding the decreased myelination and apparent failure to repair myelin damage seen in KD. In the absence of cell death, 1 μM Psy suppressed both cell division (Fig 1B) and differentiation of O-2A/OPCs into OLs (S1A Fig). It also disrupted cytoskeletal integrity and caused decreased cell migration (S1B and S1C Fig). Such sensitivity places these cells among those most sensitive to the effects of Psy exposure. Psy Alters Lysosomal Function in O-2A/OPCs We next discovered that exposure to 1 μM Psy has the previously unrecognized ability to cause multiple alterations in lysosomal function, indicating that this lipid may provide a direct link between enzymatic mutation and lysosomal dysfunction in KD. Exposure to Psy caused abnormalities in lipid homeostasis, endolysosomal transport, and cathepsin activity. Exposure to 1μM Psy disrupted lipid homeostasis, causing the intracellular accumulation of both neutral triglycerides and phospholipids (Fig 1C, S1D Fig). Endolysosomal transport was also compromised by exposure to 1 μM Psy, as shown by a decreased rate of endocytic import of fluorescently labeled polystyrene nanobeads (time to half-maximal staining intensity: 4.6 ± 1.0 min for vehicle-treated control versus 22.2 ± 5.7 min for Psy, p < 0.05; Fig 1D, S1E Fig). Psy exposure also increased the activity of resident lysosomal proteases cathepsin D and B, which can cause cellular damage or death upon export to the cytoplasm and the activities of which are known to be elevated in a number of LSDs (Fig 1E) [86–88]. Psy’s ability to disrupt lysosomal function was as great as that seen with bafilomycin A (BafA), which disrupts lysosomal function by antagonizing the lysosomal vacuolar-type H+-ATPase [89]. Exposure of O-2A/OPCs to 1 μM Psy significantly increased intralysosomal pH from 4.88 ± 0.04 to 5.62 ± 0.08 after 24 h, an increase maintained for at least 48 h after a single exposure (p < 0.001; Fig 1F). This elevation in lysosomal pH was observed in both fixed (Fig 1F) and live (S1F Fig) O-2A/OPCs. Psy exposure was as potent at increasing lysosomal pH as multiple compounds well known to exert such effects, including BafA, chloroquine, or the weak base NH4Cl (Fig 1F) [90]. This increase was evident within minutes of exposure to Psy and was comparable to treatment with BafA (S1G Fig, S1–S3 Movies), and the effects on lysosomal pH were sustained over 24–48 h after Psy exposure. Unbiased Screening Identifies Chemically Diverse Candidate Protective Agents That Prevent Psy-Induced Cellular and Lysosomal Dysfunctions To identify potential means of preventing Psy-induced toxicities that might be suitable for eventual clinical utilization, we conducted unbiased analysis of multiple concentrations of 1,040 mostly United States Food and Drug Administration (FDA)-approved small molecules [91] and a custom panel of 12 growth factors with known neuroprotective activity. We examined prevention of Psy-induced suppression of O-2A/OPC division in these analyses (Fig 2A). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Unbiased screening identifies chemically diverse candidate protective agents that reduce Psy toxicities. (A) Schematic depicting the work flow for unbiased screening with Celigo adherent cytometer (Nexcelom). (B) Representative plot of relative cell number over 5 d for rat O-2A/OPCs exposed to 1 μM Psy or vehicle (DMSO). Quantification of the relative proliferation rate for all 1,040 drugs at three concentrations over 5 d. Blue: “hit” drugs selected for further validation. A bar graph quantifying the relative proliferation rate of all vehicle and Psy controls is included. (C) Quantification of the relative proliferation rate of rat O-2A/OPCs exposed to 1.5 μM Psy, with and without verified pro-division “hits,” over 5 d. (D) Quantification as in (C) for cells exposed to Psy 48 h before exposure to the indicated drugs. (E) Summary of clinical metadata for lead “hits” as % of total. (* worldwide approval, including FDA. ** reported blood–brain barrier permeability; not all drugs have been examined/reported.) (F) Quantification of cell survival in cells exposed to 3.3 μM Psy for 5 d, with and without the indicated drugs, for 5 d. (G, H) Quantification of the (G) relative survival and (H) number of rat O-2A/OPCs per clone exposed to of Psy (1 μM in H), with and without 100 ng/mL insulin-like growth factor (IGF-1), for 5 d. (I–K) Quantification of (I) neutral lipid and phospholipid accumulation, (J) cathepsin B and D activities in rat O-2A/OPCs exposed to the indicated drugs, with and without 1 μM Psy, and (K) time to half-maximal intensity for endocytosis of fluorescent nanobeads for (I) 2 d or (J, K) 24 h. (L) Quantification of lysosomal pH for rat O-2A/OPCs exposed to 1 μM Psy, with and without the indicated “hits” (blue) or “non-hits” (gray), for 24 h. NT-3: 10 ng/mL; 1E04: 5 μM; 5F05, 9A06, 9H10: 1 μM. ap < 0.05, bp < 0.01 versus Psy-only. Data for all graphs displayed as mean ± SEM; ns = not significant; *p < 0.05, **p < 0.01, †p < 0.001 versus untreated; ap < 0.05, bp < 0.01, cp < 0.001 versus Psy-only treatment. See also S2 Fig, S1 and S2 Tables for drug names and concentrations. Data presented in this figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.1002583.g002 We found 16 structurally and functionally diverse compounds (S2A Fig, S1 and S2 Tables), in addition to 4 growth factors, that had the unexpected properties of rescuing cell division (Fig 2B and 2C, S2B Fig). Eight of the 9 most protective agents were effective at rescuing cell division even when their administration was delayed 48 h after Psy exposure (Fig 2D). Importantly, all small molecules were optimally protective at physiologically achievable concentrations (i.e., nanomolar to low micromolar), and most are approved for use in humans (94%) and are blood–brain barrier permeable (>80%) (Fig 2E). Five agents (chlorotrianisene [1G05], NKH-477 [(9C06), also known as colforsin], clofoctol [8D08], tulobuterol [9E07], and insulin-like growth factor [IGF-1]) revealed by our screens not only significantly rescued cell division but were also able to reduce cell death caused by exposure to higher concentrations of Psy (Fig 2F). With the exception of IGF-I, none of our compounds of interest were previously identified as being able to protect against toxicity of Psy (or of other lipids accumulating in LSDs). Even in the case of IGF-I, previous studies reported that supraphysiological (>10 μg/mL) concentrations decreased Psy-induced apoptosis in OLs [92] and in an O-2A/OPC cell line [82]. In our studies, by contrast, 100 ng/mL IGF-1 shifted Psy’s cytotoxicity curve by an order of magnitude and significantly deceased Psy-dependent suppression of self-renewal (Fig 2G and 2H). Further examination of three of the most protective agents—clofoctol, NKH-477, and IGF-I—demonstrated that these compounds also prevented Psy-induced alterations in lysosomal function. All three suppressed Psy-induced increases in lipid accumulation and cathepsin activity and restored normal endocytosis (Fig 2I–2K). Moreover, all three compounds significantly decreased Psy-dependent increases in pH (Fig 2L). In contrast, five randomly selected molecules that did not significantly reduce any Psy toxicities in our screens (caffeine [1E04], acetarsol [5F05], mepartricin [9A06], avobenzone [9H10], and neurotrophin-3; S1 Table) had no effect on Psy-induced increases in lysosomal pH. Protective Agents Converge on a Limited Number of Common Necessary Pathways for Their Activity As the protective compounds we discovered are structurally and functionally diverse, we next attempted to define regulatory pathways on which these agents might converge to confer their protective activity. To do this, we focused on multiple signaling pathways previously described to be antagonized by exposure of cells to Psy (e.g., mitogen-activated protein kinase [MAPK], phosphoinositide-3-kinase [PI3k]/Akt, protein kinase C [PKC], cyclic Amp (cAMP)-dependent signaling [51, 74, 82, 93–95]), as well as other proteins that have been implicated in mediating stress responses in O-2A/OPCs, including Jun N-terminal kinase (Jnk) [74], mammalian target of rapamycin (mTOR) [96–98], and estrogen receptor [99–101]. We next generated a secondary screen consisting of pharmacological inhibitors to components of these various signaling pathways. In these experiments, O-2A/OPCs were exposed to Psy; a combination of Psy and a protective agent; or a combination of Psy, a protective agent, and one of 15 pharmacological inhibitors targeting important signaling pathways and proteins (S3 Table). This allowed for the generation of a compound-specific “fingerprint of protection” that revealed putative signaling pathways used by the candidate compound to overcome Psy-induced suppression of division (e.g., Fig 3A and 3B). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Protective agents converge on a limited number of common necessary pathways for their activity. (A, B) Representative “fingerprint of protection” for candidate drug 1G05. Quantification of the relative proliferation rate of rat O-2A/OPCs exposed to 1.5 μM Psy, with and without 100 nM 1G05, and with and without pharmacological inhibitors targeting the indicated protein, for 5 d. To account for differences in their ability to reduce Psy-induced suppression of division between candidate agents, relative changes in cell division were normalized as in (B). (C) Unsupervised hierarchical clustering of “fingerprints” for the indicated drugs. Data for all graphs displayed as mean ± SEM. See also S1 and S2 Tables for drug names and concentrations, and S3 Table for details on the “fingerprinting” screen. Data presented in this figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.1002583.g003 We generated fingerprints for 12 of the most efficacious compounds in reducing Psy-induced suppression of division across 5 d using this approach. The results were then hierarchically clustered to identify similarities and dissimilarities between individual compound fingerprints and between the signaling pathways implicated in protection (Fig 3C). Despite structural and functional diversity among candidate protective agents, there was striking similarity in the signaling pathways needed for protection. The activity of the diverse protective agents was antagonized by pharmacological inhibition of the Ras/rapidly accelerated fibrosarcoma gene (Raf)/MAPK pathway, Akt, estrogen receptor, protein kinase A (PKA), and geranylgeranyl transferase ([GGT], which is needed for activation of small GTPases that are involved in cell division and migration). Despite their structural diversity, there was a surprisingly high degree of correlation between groups of small molecules; the cluster of structurally and functionally unrelated drugs 2G08 (ethopropazine, an antiparkinsonian drug), 2F11 (estradiol valerate, a synthetic estrogen), and 8D08 (clofoctol, an antibiotic), for example, showed the highest degree of similarity (correl. = 0.97) (S3 Fig). Protection Against Psy-Induced Lysosomal and Cellular Dysfunctions Can Be Provided by Promotion of Lysosomal Re-acidification Lysosomal ion homeostasis, maintained through the activity of several channels and transporters, is critical to the normal function of lysosomes. For example, H+ import is necessary for the maintenance of an acidic pH [102] and is achieved through the activity of the V-ATPase, Ca2+ is important for vesicle trafficking [103] and fusion [104], Na+ and K+ are required for the regulation of membrane potential [105, 106], and Cl− serves as a counterion to regulate lysosomal membrane potential and to facilitate the acidification of the lysosome lumen [107–109]. Although any of these may be potential therapeutic targets, we focused our attention on identifying those channels or transporters regulated by signaling pathways uncovered through our fingerprinting analysis. Of the several pathways that are required for activity of our protective agents, the one for which there is a clearly established linkage to at least one aspect of lysosomal function is the requirement for PKA activity. Previously, it has been shown that cAMP can promote lysosomal re-acidification [110], as can PKA, which is activated by cAMP [111]. In addition, we found that increases in cAMP not only normalized lysosomal pH but also prevented Psy-induced decreases in O-2A/OPC division (Fig 4A and 4B), raising the theoretical possibility that intervention at this point would provide additional benefits beyond that of pH restoration. We therefore focused attention on the role of lysosomal re-acidification as a potential therapeutic target. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. Promotion of lysosomal re-acidification is critical in protecting from multiple aspects of Psy toxicity. (A–D) Quantification of the lysosomal pH and proliferation rate in rat O-2A/OPCs exposed to (A, C) 1 μM Psy for 24 h, (B, D) 1.5 μM for 5 d, with and without (A, B) 1 mM cAMP (or 10 μM forskolin), or (C, D) 333 nM RP-107 and 1 μM cystic fibrosis transmembrane conductance regulator (CFTR) inhibitor 172 (CFTRi-172). (E) Quantification of neutral lipid and phospholipid accumulation in rat O-2A/OPCs exposed to the indicated drugs, with and without 1 μM Psy, for 2 d. (F, G) Quantification of the lysosomal pH and proliferation rate in rat O-2A/OPCs exposed to 1 μM Psy for (F) 24 h or (G) 5 d, with and without 100 nM KT-5720 or 3.3 μM H89 (cAMP: 1 mM). Data for all graphs displayed as mean ± SEM; *p < 0.05, **p < 0.01, †p < 0.001 versus untreated; ap < 0.05, bp < 0.01, cp < 0.001 versus Psy-only treatment. See also S1 and S2 Tables for drug names and concentrations. Data presented in this figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.1002583.g004 The most attractive explanation for how cAMP/PKA activity could restore lysosomal pH would be through activation of the cystic fibrosis transmembrane conductance regulator (CFTR), a PKA-activated transmembrane chloride channel that promotes lysosomal re-acidification [112]. Unlike the CLC-7 Cl−/H+ antiporter, another chloride channel that is localized to the lysosomal membrane and thought to play a role in the basal maintenance of lysosomal pH [113], the CFTR channel appears only to be critical for re-acidification. Moreover, although the CFTR can be activated by PKA and cAMP, there is no evidence for such activation of CLC-7. In addition, specific agonists and inhibitors exist for the CFTR, enabling a direct test of whether promoting re-acidification can prevent Psy-induced toxicity. As we predicted, treatment of cells with the cAMP-independent CFTR agonist RP-107 [114] restored lysosomal pH in cells exposed to Psy. Although control of lysosomal pH and/or lysosomal re-acidification has not been thought to have any upstream role in the multiple cellular dysfunctions caused by Psy exposure, we nonetheless found that RP-107 protected against Psy-induced suppression of division, as well as elevated storage of both neutral lipids and phospholipids (Fig 4C–4E). To test the hypothesis that these benefits were not due to off-target effects of RP-107, we also co-exposed cells to CFTR-inhibitor 172 (CFTRi-172) [115], which attenuated the protective effects of RP-107 treatment (Fig 4C–4E). The effects of RP-107 were CFTR dependent, as determined by knockdown of CFTR in O-2A/OPCs using small interfering RNA (siRNA) pools targeting rat CFTR, as well as a pool of nontargeting (NT) siRNAs as a control for transfection; the reduction in CFTR protein levels was confirmed by western blot analysis. Knockdown of CFTR did not significantly affect lysosomal pH when compared to cells exposed to NT controls (4.96 ± 0.13 versus 4.81 ± 0.11, respectively). Moreover, in the presence of Psy, in both NT and CFTR siRNA pools, there was a significant increase in lysosomal pH (5.47 ± 0.08 versus 5.61 ± 0.09, respectively), with no significant difference between these two treatment groups. However, when we tested the effect of RP-107, a specific CFTR agonist, we found that lysosomal pH was significantly reduced in cells exposed to NT siRNA but that CFTR knockdown attenuated RP-107’s protective effect (5.18 ± 0.09 versus 5.62 ± 0.03, p < 0.01; S4A Fig). Thus, as with our pharmacological experiments, genetic loss of CFTR does not appear to significantly affect basal lysosomal pH in untreated cells. However, the protective capacity of RP-107 is CFTR dependent. These results are consistent with the original studies demonstrating the role of the CFTR in control of lysosomal re-acidification [112]. Moreover, we found that the most effective protective agents identified in our studies did not themselves reduce the basal acidic pH of lysosomes in the absence of Psy (S4B Fig) but instead seemed to work to promote re-acidification. Indeed, their ability to normalize lysosomal pH, as well as rescue cell division, in cells exposed to Psy was blocked by co-exposure to inhibitors of PKA (Fig 4F and 4G). As these protective agents are able to rescue cells even when applied 48 h after Psy exposure (Fig 2D), it appears that their protective activity is not mediated simply by blocking lysosomal alkalization. Psy’s Free Amine Group Is Critical for Its Toxicity The observations that multiple Psy-induced lysosomal and cellular dysfunctions can be prevented by lysosomal re-acidification with RP-107 (Fig 4C–4E) and that Psy exposure causes rapid increases in lysosomal pH (Fig 1F, S1D Fig), raise complementary questions about how Psy causes such changes. One possibility is that Psy disrupts the function of particular proteins involved in lysosomal re-acidification, but another possibility is that structural features of Psy itself are directly relevant to understanding effects on lysosomal pH. Although multiple studies have attempted to understand the molecular mechanisms underlying Psy’s toxicity [61–75], we noted that Psy has unusual physicochemical features that might be of relevance to understanding its effects on lysosomes. Psy is unusual as a cationic, weakly basic lipid, carrying a net positive charge at physiological pH. With a pKa value of 7.18 [116], Psy is predicted to be 99.9% protonated in the acidic pH of the lysosome. If this aspect of Psy’s structure is important in altering lysosomal and cellular function, then the protonatable free amine group on Psy should be critical in mediating the changes in lysosomal pH that we observed. We therefore tested whether removing this free amine group altered effects on lysosomal pH and on other outcomes of Psy exposure. We found that the free amine group on Psy is critical in its ability not only to disrupt lysosomal pH but also to cause other toxic effects. We compared Psy toxicity to that of N-acetyl-Psy (N-AcPsy), a structural derivative containing an amide-linked acetyl group, rendering it no longer protonatable (Fig 5A). Unlike Psy, N-AcPsy did not induce cell death or alter O-2A/OPC self-renewal at similar concentrations (Fig 5B and 5C). Moreover, N-AcPsy did not elevate neutral lipid and phospholipid storage, increase endocytic transport time, increase cathepsin activity, or elevate lysosomal pH (Fig 5D–5G). Thus, the positively charged free amine group present on Psy was critical to increasing lysosomal pH and also to the subsequent lysosomal and cellular impairments observed after exposure in O-2A/OPCs. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. Psy’s free amine group is critical for its toxicity. (A) Structures of Psy and N-AcPsy. (B) Quantification of rat O-2A/OPC survival in response to Psy or N-AcPsy. (C–G) Quantification of rat O-2A/OPC self-renewal, lipid accumulation, endocytosis, cathepsin activity, and lysosomal pH in cells exposed to 1 μM Psy or N-AcPsy for (C) 5 d, (D) 2 d, or (E–G) 24 h. Data for all graphs displayed as mean ± SEM; ns = not significant; *p < 0.05, †p < 0.001 versus control. Data presented in this figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.1002583.g005 Structurally Related Lysosphingolipids from Multiple LSDs Alter Lysosomal Function To further test the hypothesis that the presence of free amine group on a cationic lipid is critical to lipid-induced toxicities, and that such lipids provide a direct link between enzymatic mutation and lysosomal disruption, we examined a series of lipids known to accumulate in other LSDs. Other lipids of potential interest include lyso-sulfatide (lyso-SF) (which accumulates in MLD [49]), glucosylsphingosine (GlcSph) and glucosylceramide (GlcCer) (which accumulate in Gaucher disease [47]), and lactosylsphingosine (LacSph) and lactosylceramide (LacCer), which accumulate in several LSDs (Fig 6A) [78, 117, 118]. Some of these lipids appear to have been only rarely studied for their effects on cell function in vitro (lyso-SF, GluSph, LacSph, LacCer) [119]. In the case of Gaucher disease, the majority of previous in vitro studies appears to have focused on GlcCer, and studies on both GlcCer and GlcSph often have required lipid concentrations severalfold greater than those at which Psy’s effects were observed (e.g., [22, 50, 51, 53, 55, 120–122]). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 6. Lysosphingolipids accumulating in other LSDs suppress critical O-2A/OPC behaviors and lysosomal function. (A) Chemical structures of the indicate lipids. Gal: galactose; Glc: glucose; R: variable hydrocarbon chain. (B–G) Quantification of (B) the relative survival, (C) the relative number of O-2A/OPCs per clone, (D) lipid accumulation, (E) endocytic import time, (F) cathepsin activity, and (G) lysosomal pH in rat O-2A/OPCs exposed to the indicated lipids for (B and C) 5 d, (D, E, G) 24 h, or (F) 2 d. Data for all graphs displayed as mean ± SEM; ns = not significant; *p < 0.05, **p < 0.01, †p < 0.001 versus untreated, unless otherwise indicated. S4 Table for lipid concentrations used in (C–G). Data presented in this figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.1002583.g006 In order to eliminate differences in cell types as potential contributors to different outcomes, we examined the survival and self-renewal of O-2A/OPCs exposed to lyso-SF, GlcSph, GlcCer, LacSph, and LacCer. Use of these cells also provided a test of the hypothesis that the structure of a lipid is of primary importance in determining toxicity. We also examined the effects of N-acetyl-sulfatide (N-AcSF) as a direct comparison with N-AcPsy. We found that sphingosine-derived lipids that accumulate in different LSDs and that contain a free amine group (and thus are structurally similar to Psy) caused significant cell death and suppression of self-renewal (Fig 6B and 6C) at similarly low lipid concentrations as we observed with Psy. In contrast, exposure to their ceramide-based counterparts GlcCer and LacCer, or to N-AcSF, did not cause cellular toxicities at comparable or 10-fold higher concentrations (Fig 6B and 6C). We also found that lysosphingolipids accumulating in other LSDs [22, 31, 47–55, 123–127] had similar effects as Psy on lysosomal function. Exposure to sublethal concentrations of GluSph, lyso-SF, or LacSph caused increases in neutral lipid and phospholipid accumulation, endocytic transport time, cathepsin activity, and lysosomal pH. In contrast, exposure to their non-lyso counterparts did not have such effects (Fig 6D–6G). If the hypotheses are correct that other toxic lysosphingolipids that accumulate in LSDs work through similar mechanisms as Psy, and that such mechanisms are relevant to understanding the efficacy of our protective agents, then our protective agents also should rescue cells from the toxic effects of lipids from other LSDs. If correct, such findings would provide both the first structural predictors of toxicity and the first example of protective agents of potential relevance in different LSDs. We found that our candidate protective agents also reduced the toxic effects of GlcSph, lyso-SF, and LacSph (Fig 7A). Three of our most effective agents—IGF-1, clofoctol (8D08), and NKH-477 (9C06)—prevented lipid-induced suppression of division and also attenuated increases in neutral lipid and phospholipid accumulation, endocytic import time, cathepsin activities, and lysosomal pH in rat O-2A/OPCs exposed to sublethal concentrations of GlcSph, lyso-SF, or LacSph (Fig 7B–7G, S5A Fig). These agents also rescued cell division in cells exposed to GlcSph or Lyso-SF for 48 h before addition of protective agents (S5B Fig). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 7. Candidate protective agents reduce multiple lysosphingolipid-induced lysosomal and cellular toxicities. (A) Venn diagram summarizing the number of protective drugs, including IGF-1, that reduce suppression of division in rat O-2A/OPCs exposed to the indicated lyso-lipids for 5 d. (B–E) Quantification of (B) lipid accumulation, (C) endocytic import time, (D) cathepsin activity, and (E) lysosomal pH in rat O-2A/OPCs exposed to the indicated lyso-lipids, with and without 100 ng/mL IGF-1, 100 nM 8D08, or 333 nM 9C06, for (B) 2 d or (C–E) 24 h. Data for all graphs displayed as mean ± SEM; ns = not significant; *p < 0.05, **p < 0.01, †p < 0.001 versus untreated; ap < 0.05, bp < 0.01 versus Psy-only treatment, unless otherwise indicated. See also S4 Fig and S1 and S2 Tables for drug names and concentrations and S4 Table for lipid concentrations. Data presented in this figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.1002583.g007 Lysosphingolipids Disrupt Human OL Progenitor Cell Behaviors, and Protective Compounds Rescue Human Cells We next examined the question of whether the principles revealed in our studies on cells derived from the CNS were applicable to human cells. In these experiments, we used an anti-CD140a (PDGFRα) antibody to enrich for a population of human O-2A/OPCs from the corpus callosal field of mid-gestation fetal tissue (S6 Fig) [128] and exposed cells to Psy and potential protective agents as for rat-derived cells. Exposure to lysosphingolipids caused death in human cells at concentrations comparable to those used in rat progenitor cells (S4 Table), as well as suppression of cell division and elevation of lysosomal pH at sublethal concentrations, whereas their non-lyso counterparts did not cause similar toxicities (Fig 8A–8C). Notably, cell division and normalization of lysosomal pH were restored in cells exposed to Psy with clofoctol, NKH-477, and IGF-I, as we observed for rat O-2A/OPCs (Fig 8D and 8E). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 8. Lysosphingolipids disrupt human O-2A/OPC behaviors. (A–C) Quantification of (A) the relative survival, (B) the relative proliferation rate, and (C) lysosomal pH of human O-2A/OPCs exposed to the indicated lipids for (A, B) 5 d or (C) 24 h. BafA: 100 nM. (D, E) Quantification of the relative proliferation rate and lysosomal pH in human O-2A/OPCs exposed to 1 μM Psy for (D) 5 d or (E) for 24 h, with and without the indicated drugs. ap < 0.05 versus Psy-only. Data for all graphs displayed as mean ± SEM; ns = not significant; *p < 0.05, **p < 0.01, †p < 0.001 versus untreated control; ap < 0.05, bp < 0.01 versus Psy-only treatment. See also S5 Fig and S1 and S2 Tables for drug names and concentrations, and S4 Table for lipid concentrations. Data presented in this figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.1002583.g008 Alterations in O-2A/OPC Biology in Twitcher Mice Are Like Those Induced By Psy Exposure In Vitro In the final section of our studies, we asked whether discoveries made on WT cells exposed exogenously to Psy in vitro revealed principles applicable to cells with an LSD-relevant mutation, both in respect to cellular pathologies and to rescue of lysosomal function. These studies were carried out using twitcher mice, a naturally occurring model of KD that recapitulates most human symptoms. Multiple studies have demonstrated that this mouse is a reliable model of KD in respect to enzymatic dysfunction and tissue pathology [77–80, 129, 130] and is also one of the most useful models for studying LSDs in general. In particular, twitcher mice progress from a lack of apparent pathology to severe disease over a relatively rapid time course, with function appearing to be normal at birth, followed by disease symptoms manifesting about 20 d after birth and with death ensuing at about 42 d after birth. This time course allows pathology and the effects of treatment to be studied at different stages of disease progression. In our studies on twitcher mice, we first determined that changes in O-2A/OPC function were like those induced by exposure to low doses of Psy in vitro. We found significant reductions in both myelin content and OL cell number (OLs; Olig2+/GST+) in the corpus callosum—the major myelinated tract of the CNS—at P40 when compared to age-matched WT littermates (Fig 9A and 9B), consistent with previous analyses of human and twitcher tissue [76, 131]. We also observed a significant reduction in the percentage of dividing (Ki67+) O-2A/OPCs (54.0% ± 1.9% of WT, p < 0.01; Fig 9C) at this late time point, during which time O-2A/OPCs should be undergoing rapid expansion through cell division to replace damaged OLs and myelin. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 9. NKH-477, a protective compound identified in vitro, protects against multiple toxicities in treated twitcher mice. (A–C) Quantification of (A) fluoromyelin intensity, (B) number of GSTpi+/Olig2+ OLs, and (C) the relative number of dividing Ki67+/Olig2+ O-2A/OPCs in the corpus callosa of P40 twitcher mice and age-matched WT littermates (n = 3 from different litters). (D) Analysis of clonal composition of P15 twi O-2A/OPCs and WT littermates across 5d. (E) Quantification of the relative number of dividing Ki67+/Olig2+ O-2A/OPCs in the corpus callosa of P15 twitcher mice and age-matched WT littermates (n = 3 from different litters). (F) Quantification of lysosomal pH of O-2A/OPCs acutely isolated from P17 twitcher and WT mice. (G) Overview of treatment paradigm and clinical course for twitcher mice. (H) Quantification of lysosomal pH of O-2A/OPCs isolated from P35 treated mice. (I) Quantification of the number of dividing callosal O-2A/OPCs in P35 treated mice. (J–L) Representative confocal images of fluoromyelin-stained corpus callosa of the indicated treatment groups at P40, in addition to quantification of staining intensity and the number of OLs. (M) Kaplan–Meyer survival curve for treated and untreated twi mice. Median survival of twi mice, with dotted lines indicating reported median survival of single-therapy treatments [15]. (N–Q) Quantification of travel speed, stance time, beam traverse time, and relative weights for P25 saline-treated WT and twi, as well as NKH-477–treated twi, mice. (R) Quantification of brain Psy levels at P35. Data for all graphs displayed as mean ± SEM; ns = not significant; *p < 0.05, †p < 0.001 versus WT; bp < 0.01 versus vehicle-treated twi. Data presented in this figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.1002583.g009 We additionally found that O-2A/OPC function was compromised in presymptomatic twitcher mice in ways similar to those induced by Psy exposure. We isolated O-2A/OPCs from presymptomatic P15 twitcher mice and examined their self-renewal capacity in vitro. These cells showed impaired self-renewal in comparison with cells of age-matched WT cells when maintained in vitro for 5 d (Fig 9D). Such findings were mirrored by significant reductions in the pool of dividing O-2A/OPCs in vivo at P15 (Fig 9E). To determine whether cells harboring a mutant GALC gene exhibit changes in lysosomal pH, we examined the endolysosomal pH of corpus callosal O-2A/OPCs acutely isolated from presymptomatic twitcher mice (P17). We found that the lysosomal pH was significantly less acidic than that of cells isolated from age-matched WT littermates (Fig 9F), similar to what was observed in vitro with exogenous Psy treatment (Fig 1F). Thus, O-2A/OPCs isolated at developmental time points in which symptoms are not obvious (prior to P18–20) show altered lysosomal pH and alterations in critical cellular behaviors like those induced by exposing WT cells to Psy in vitro. NKH-477 Protects against Multiple Toxicities in Twitcher Mice We next investigated whether the analytical approach employed in our in vitro studies could identify compounds able to provide clinically relevant benefits in vivo. We focused our studies on NKH-477 (9CO6), a water-soluble derivative of forskolin that is approved for treatment of acute heart failure in Japan [132], as this agent is known to be CNS penetrant and elevates cAMP levels (through direct activation of adenylyl cyclase) in brains of rats after systemic administration [133]. Moreover, unlike the other identified protective agents, the linkage of NKH-477 to PKA regulation (and thus to lysosomal re-acidification) is both defined and mediated through widely expressed proteins, consequentially not requiring cells to express specialized drug-targeted receptors in order to be responsive. We initiated treatment at P10, a time when CNS concentrations of Psy are already approaching the range at which we see effects on O-2A/OPCs [78–80], using once-daily intraperitoneal (IP) injections (1 mg/kg; Fig 9G). This is a point in time when disruptions in neuronal function can already be observed in twitcher mice [134], raising the possibility of initiating treatment only after subtle clinical changes are first observable. This delayed initiation of treatment is in marked contrast with the well-studied need to initiate the application of bone marrow transplantation and/or gene therapy in the first few days after birth in order to obtain benefit [1, 10, 16, 40, 130]. The primary endpoints of interest in our in vivo studies were whether we could rescue lysosomal and cellular function in O-2A/OPCs and whether once-daily treatment with NKH-477 is sufficient to provide benefit on both parameters. O-2A/OPCs were isolated at P35 to examine the effects of NKH-477 treatment on lysosomal pH, and we found a normalization of pH in cells isolated from treated twitcher mice when compared to vehicle-treated mice (Fig 9H). NKH-477–treated twitcher mice also showed an increase in the numbers of dividing O-2A/OPCs at P35 to near-normal levels, as well as increases in myelin content and increased OL cell numbers at moribund ages, when compared to vehicle-treated twitcher mice, again to levels not significantly different from WT mice (Fig 9I–9L). Remarkably, we also found that NKH-477 treatment provided significant lifespan extension that was comparable to published single-therapy treatments aimed at restoring GALC activity, including bone marrow transplantation (the current standard of care in patients) or viral-mediated gene therapy (Fig 9M) [1, 2, 10, 13, 15, 16]. Moreover, twitcher mice that received daily injections of NKH-477 also showed significantly improved locomotor and gait function (Fig 9N–9P, S7 Fig) and significantly improved weight gain throughout their lifespan when compared to vehicle-treated twitcher littermates (Fig 9Q). These benefits were observed despite the fact that we were not correcting the genetic defect; indeed, we did not find that NKH-477 treatment reduced the overall tissue burden of Psy in the CNS (Fig 9R). Discussion Our studies provide multiple novel findings related to the biology of LSDs. We found in both rodent and human cells that structurally related sphingolipids that accumulate in these disorders appear to directly cause multiple lysosomal dysfunctions. We also discovered multiple pharmacological agents, previously approved for other clinical purposes, that prevent all of the sphingolipid-induced lysosomal and cellular toxicities we analyzed, apparently by promoting lysosomal re-acidification. In vivo studies in the twitcher mouse model of KD demonstrated the ability of one of the agents we identified, which is known to be CNS penetrant, to correct lysosomal pH in O-2A/OPCs, as well as to provide multiple therapeutically relevant benefits in the absence of correcting the underlying genetic mutations implicated in the disease. The finding that pathophysiologically relevant low levels of four different sphingolipids known to accumulate in different LSDs are each sufficient to compromise multiple lysosomal functions appears to provide the first evidence that substances created due to mutations of lysosomal enzymes may be directly responsible for initiating the metabolic dysfunctions that characterize such diseases. Previous studies have speculated that lysosomal dysfunction is caused by such events as intralysosomal accumulation of substances that are not properly degraded (e.g., [19, 20, 24, 26, 28, 30, 32, 135–137]), but we could find no prior demonstration—or suggestion—that a specific substance known to accumulate in LSDs is able to simultaneously alter lysosomal pH, endolysosomal trafficking, lipid degradation, and cathepsin activation. Based on the comparative structures of toxic and nontoxic lipids, we hypothesize that toxicity is caused by disruption of lysosomal pH. All of the four toxic lipids we studied share the presence of a theoretically protonatable free amine group, raising the possibility that their accumulation increases the net positive charge in the lysosomal lumen, altering ion homeostasis and decreasing acidification by suppressing proton influx through the V-type H+ ATPase. In contrast, such effects were not caused by other lipids that accumulate in these disorders and that lack this free amine group (i.e., GlcCer, LacCer), or by lysosphingolipids with an acetylated amine group (N-AcPsy and N-AcSF) attenuated toxicity. The only remotely comparable studies we could find to our own were those of Sillence and colleagues [120, 121], who reported that exposure of the virally transformed tumorigenic RAW murine macrophage cell line to 40 μM GlcCer for 48 h altered trafficking of boron-dipyrromethene (BODIPY)-labeled LacCer to the lysosome, and that an unspecified concentration of GlcCer caused modest increases in lysosomal pH in these cells. However, such studies also demonstrated that exposure to 20 μM GlcCer or GlcSph decreased lysosomal pH in RAW cells exposed to a GlcCer synthase inhibitor, that such effects were not caused by exposure to Psy, and that these GlcCer and GlcSph concentrations caused negligible cell death [120, 121]. Thus, these previous results differ markedly from those obtained in our studies examining effects of 10- to 20-fold lower concentrations of Psy, GlcSph, lyso-SF, and LacSph and also do not indicate that lipids with similar structures cause similar lysosomal or cellular pathologies. In addition, although studies on sphingosine (applied at 10-μM concentrations) suggested the free amine group on this lipid is important for its toxicity, these studies considered the role of the amine group was to confer detergent-like properties on sphingosine and did not consider potential relevance to control of lysosomal pH [138, 139]. The possibility that changes in lysosomal pH may be of particular importance in understanding the effects of exposure to toxic sphingolipids, and that lysosomal neutralization may be upstream of multiple lysosomal and cellular dysfunctions and may provide a novel therapeutic target, was strongly supported by our findings that we rescued O-2A/OPCs from adverse effects of lipid exposure by activation of the CFTR (which promotes lysosomal re-acidification [112]). Exposure to RP-107, a chemical activator of CFTR [114], prevented alterations in lysosomal pH and also rescued cells from adverse effects on division in a CFTR-dependent manner. In addition to CFTR, there are likely several other lysosomal targets that may be relevant for treatment. Accumulation of undegraded sphingomyelin, for example, has been shown to alter membrane trafficking and lysosomal calcium homeostasis through the impairment of the TRPML1 channel [140]. We think it is also important not to interpret our findings as indicating that activation of chloride flux via the CFTR will be the sole mechanism available for promoting lysosomal re-acidification, or that regulation of chloride flux is the only possible way to promote restoration of a normally acidic pH. The CFTR provides a well-studied protein for which there is strong data indicating a role in re-acidification [112], for which useful experimental drugs are available, and for which a role of PKA in activation has been identified. But it seems likely there will be other proteins that offer potential entry points for promoting re-acidification. In respect to the much more studied problem of the control of basal lysosomal pH, there is strong disagreement on whether chloride ion flux through the CFTR or through CLC chloride channels or Cl−/H+ exchangers is central to controlling basal lysosomal acidification. Expression of mutant CFTR in alveolar macrophages was reported to be associated with a lack of proper acidification of their degradative compartments [141, 142]. In contrast, other authors found that the CFTR was not required for phagolysosomal acidification in macrophages [143] or respiratory epithelial cells [144], with other investigators also questioning the importance of CFTR in promoting lysosomal acidification [145]. These disagreements also extend to the CLC chloride channels or Cl−/H+ exchangers, and some investigators have reported that loss of CLC-7 is not associated with alterations in lysosomal acidification [107, 108, 146]. Moreover, there are also intriguing observations that cation transport may also be important in the regulation of basal lysosomal pH [102, 143]. Although it may be that some of these disagreements arise due to use of different techniques [113], it may also be the case that there are nuances of lysosomal regulation that differ in different cell types, and also that lysosomal re-acidification may be regulated by flux of cations or anions other than chloride. The extent to which controversies regarding control of basal lysosomal pH are pertinent to studies on control of lysosomal re-acidification is not yet known, however. We hope that the results of our present studies will further increase interest in this important problem and will lead to identification of other regulatory pathways of potential therapeutic relevance. Although some of the effects of individual toxic sphingolipids that we studied have been observed previously with other cell types (although usually at higher lipid concentrations than we utilized, e.g., [51, 61–75, 93, 95, 119–122, 147–154]), there is no previous indication that all of these forms of damage may ultimately be controlled by a single metabolic parameter or that such a parameter might control lysosomal pH. It is also worth noting that, although multiple mechanisms have been observed to contribute to particular effects of Psy or of other toxic lipids [51, 61–75, 82, 120, 123, 147–150, 152–187], none has demonstrated the ability to correct the multiple dysfunctions prevented by promotion of lysosomal re-acidification. Additional support for the hypothesis that control of lysosomal pH is of central importance in understanding the pathology of toxic lipid exposure was provided by the identification, by unbiased drug screening, of novel protective agents that show no known prior overlap in function but that all converged on promoting lysosomal re-acidification. We found that clofoctol, NKH-477, and IGF-1 all restored lysosomal pH in lipid-exposed cells, despite having no known common properties. Restoration of lysosomal pH appears to be due to promotion of re-acidification, as none of these protective agents acidified lysosomes in the absence of Psy. As the only known convergence of the protective substances we identified (including RP-107) is a common ability to promote lysosomal re-acidification, it currently is most likely that it is this aspect of their effects that is most important. The possibility that regulation of lysosomal pH and re-acidification could represent a convergence point of disease pathology and therapeutic intervention for LSDs appears to be novel. Interest is emerging in the possibility that promoting lysosomal re-acidification may offer therapeutic benefit in situations of lysosomal dysfunction, but studies thus far have been focused only on the possibility that restoring normal lysosomal pH will enhance normal protein degradation [32, 110, 111, 188–192]. Nonetheless, the possibility that regulation of lysosomal pH could represent a central mechanism in disease pathogenesis and treatment is consistent with the dependence of normal lysosomal function on an acidic pH (as summarized in Fig 10). For example, neutralization can cause release of lysosomal Ca2+ and cathepsins [32, 192]: Ca2+ release could compromise cytoskeletal function [193] and hence cell division, whereas cathepsin release and activation could initiate cell death [194]. In addition, increasing lysosomal pH would be predicted to decrease function of any lysosomal enzymes evolutionarily optimized for function in an acidic environment. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 10. Hierarchical targeting of lysosomal pH by candidate protective compounds reduces multiple lysosphingolipid-induced cellular dysfunctions. We propose a model in which genetic mutations in resident lysosomal enzymes result in the accumulation of lysosphingolipids, which increase endolysosomal pH and consequently disrupt lysosome-dependent cellular processes, including endocytosis, ion homeostasis, lipid metabolism, and enzymatic activities. Candidate protective drugs—as well as direct activation of CFTR—converge on their abilities to normalize lysosomal pH, apparently through re-acidification processes, and reduce disruptions of downstream cellular processes, including survival and cellular division. https://doi.org/10.1371/journal.pbio.1002583.g010 The question of whether it was possible to modify lysosomal and cellular function in an animal model of a severe LSD was studied by administering NKH-477 to twitcher mice, a severe murine model of an LSD that exhibits a pattern of disease progression similar to that seen in KD patients [195] and that is the most widely used model for studying possible disease interventions [1–17]. NKH-477 was the logical choice for such studies, as it was the one small molecule protective agent identified thus far with known CNS penetrance [133] and with a well-defined drug target. In contrast with studies on gene therapy and/or bone marrow transplantation, both of which would be constant in their effects and generally must be initiated shortly after birth when animals are presymptomatic to obtain benefit in experimental models [1, 2, 10, 13, 15, 16], we only administered NKH-477 once daily beginning 10 d after birth. This starting point was chosen both because this is a time when CNS concentrations of Psy first begin to approach those utilized in our in vitro studies and to test the hypothesis that our approaches could identify interventions able to provide benefit even when initiated after it might be possible to detect early changes in neuronal function [65–67]. The primary goal of our in vivo studies was to determine whether NKH-477 administration could be used to normalize lysosomal pH and improve O-2A/OPC function. After first confirming that O-2A/OPCs isolated from twitcher mice showed similar abnormalities as WT progenitor cells treated with Psy in vitro, we found that daily treatment with NKH-477 normalized lysosomal pH (as analyzed ex vivo in O-2A/OPCs isolated from treated and control twitcher mice) and rescued O-2A/OPC division in vivo. Benefits of daily NKH-477 treatment extended far beyond rescue of O-2A/OPC division and lysosomal pH and offered several clinically relevant improvements. At the cellular level, daily NKH-477 administration rescued OL numbers and myelin content even at a time when vehicle-treated littermates were moribund. Moreover, mice treated with NKH-477 showed improved motor behavior and weight gain (suggesting that cell types other than O-2A/OPCs also benefitted from this treatment). In addition, survival was significantly extended, even to the same degree previously reported with gene therapy alone (and exceeding that obtained with bone marrow transplants alone) [1, 2, 10, 13, 15, 16]. These multiple benefits were obtained even though we did not correct the underlying genetic defect nor decrease the overall Psy tissue burden, and there was no prior information on using this compound (or any related compounds) in the context of LSDs. Although NKH-477 increases cAMP levels in the CNS [133] and could thus have other effects beyond promotion of lysosomal re-acidification, the fact that NKH-477 shares the property of promoting re-acidification with the other compounds we identified, and was indeed able to rescue lysosomal pH and O-2A/OPC division in vivo, makes it seem likely that this aspect of drug action is at least partially relevant to the benefits observed. Even if some of the in vivo benefits we observed were due to other activities of NKH-477 than promotion of lysosomal re-acidification, this would not decrease interest in this agent as a potential candidate for further analysis. Recent reviews of the outcomes of implementing newborn screening (NBS) for detection of early infantile KD (EIKD) in New York state [196] have led to the conclusion that, “in addition to the potential harm to families receiving false-positive test results, NBS for EIKD appears to have resulted in a reduction in survival in individuals who have the disease. The data from the New York program suggest that NBS for EIKD should be abandoned, pending the development of improved screening or therapies shown to confer both survival and quality-of-life benefits over supportive care. The results of this experience suggest that research efforts should be focused on improving presymptomatic treatment outcomes in children identified by NBS prior to the redeployment of mandatory presymptomatic screening" [197]. As treatment with NKH-477 confers both survival and quality-of-life benefits in the established animal model for KD and already has been approved (in Japan) for use in humans [132], this may provide an attractive starting point for thinking about new approaches to some of the devastating LSDs with severe neuropathology. Moreover, the discovery of mechanisms and protective strategies that apply to distinct lipids accumulating in three different LSDs provides hope that these same general principles will apply to other LSDs characterized by lysosphingolipid accumulation, and perhaps also in other LSDs (such as the neuronal ceroid lipofuscinoses/Batten disease) in which lysosomal pH is abnormally more alkaline [198]. In addition, the ability of re-acidification to rescue a diverse range of lysosomal and cellular dysfunctions raises the question of whether similar strategies might provide broadly useful effects in important diseases in which lysosomal dysfunction has also been implicated, such as Alzheimer’s and Parkinson’s disease (e.g., [199–212]). Materials and Methods Ethics Statement The University of Rochester RSRB has reviewed this study and determined that based on federal (45 CFR 46.102) and University criteria the study does not qualify as human subjects research and has waived the need for consent (RSRB#00024759). All animal procedures were performed under guidelines of the National Institutes of Health and approved by the Institutional Animal Care and Utilization Committee (IACUC) of the University of Rochester Medical Center, Rochester, NY (UCAR#2001–140). Lipids Lipids used in this study were purchased from Matreya and were of highest purity. All lipids were resuspended in anhydrous dimethyl sulfoxide (DMSO) to 10 mM, stored at -20°C or -80°C, and resuspended in media before use. Comparable results for OPC division and survival were obtained with Psy purchased from Sigma-Aldrich and Santa Cruz Biotechnology. O-2A/OPC Isolation, Purification, and Culture Corpus callosa of P7 Sprague-Dawley rats (Charles River) were micro-dissected, finely minced with a sterile blade, and digested for 20 min in 2.2 mg/mL collagenase (Worthington #4189), 20 Kunitz/mL DNase (Sigma D4263) in HBSS (Gibco #114170–161) supplemented with Sato medium (see below). Collagenase-containing medium was then replaced with 20 Kunits/mL DNase and papain solution (1:40, activated per manufacturer’s directions; Worthington #LS003127) in HBSS/Sato for 20 min. Tissue was then sequentially triturated with 21-, 25-, and 26-guage needles in 35 Kunits/mL DNase in DMEM:F12 complete media (see below) before dissociated cells were plated on tissue culture plastic for 10 min (37°C, 7% CO2). Nonadherent cells were pelleted by centrifugation (5 min, 500 xg), resuspended in degassed HBSS/Sato supplemented with 1% BSA Fraction V and anti-A2B5 MACS beads (1:50; Miltenyi Biotec #130-093-388), and incubated on ice for 20 min. Cells were pelleted and sorted with MACS columns as per manufacturer’s directions (Miltenyi). A2B5+ cells were plated on tissue culture plastic coated with poly-L-lysine (1 μg/cm2 for 20 min; Sigma #P1274) in DMEM:F12 (Gibco #11330–057) supplemented with 10 μg/mL insulin (Sigma #I5500), 100 μg/mL holotransferrin (Sigma #T2252), Sato media (final concentration: 0.03% BSA Fraction V [Sigma #A7979-50ML], 10 μM putrescine [Sigma #P7505], 200 nM progesterone [Sigma #P0130], 235 nM sodium selenite [Sigma #S1382]), 50 μg/mL gentamycin (Gibco #15750–060), 10 ng/mL PDGF-AA (R&D #221-AA), and 5 ng/mL basic FGF (Miltenyi #130–093) and maintained at 37°C (7% CO2). Cells were passaged with 0.05% trypsin-EDTA (Gibco #2300), neutralized with 80 Kunitz/mL soybean trypsin inhibitor (Sigma #T9003), and replated in DMEM:F12 complete media supplemented with 10 ng/mL PDGF-AA (and without basic FGF). Purified O-2A/OPCs were passaged no more than once and were maintained in culture for at most 7–9 d in vitro for all experiments. To generate OLs, purified O-2A/OPCs were maintained in DMEM:F12 containing Sato components, transferrin, insulin, 100 pg/mL PDGF-AA, and 45 nM T3/T4 (Sigma #T6397/#0397) for 5 d before initiation of experiments. Human cells were isolated from corpus callosal fields of fetal week 18–21 de-identified tissue, purified with anti-CD140a-coupled magnetic beads (1:100; BD Biosciences #558774; Miltenyi), and maintained as above. Cortical and Hippocampal Neuron Isolation and Culture Embryonic neurons were isolated from E18 Sprague-Dawley rats (Charles River) and maintained as previously described [213]. Briefly, isolated tissue was digested in papain solution (1:50, activated per manufacturer’s directions; Worthington #LS003127) in HBSS/Sato for 20 min at 37°C (7% CO2). Pelleted tissue was then triturated with a pulled glass Pasteur pipet in 80 Kunitz/mL DNase (HBSS; Sigma #D4263), and dissociated cells were pelleted through a 0.5 M sucrose cushion (10 min, 500 xg). Immature neurons were plated on poly-L-lysine-coated tissue culture plastic (1 μg/cm2 for 20 min; Sigma #P1274) in NeuroBasal media (Gibco #21103–049) with 50 μg/mL gentamycin (Gibco #15750–060), 2 mM L-glutamine (Gibco #25030–081), and 1 X B27 serum-free supplement (Gibco #17504044). Neurons were allowed to mature for 7 d at 37°C (7% CO2) before initiation of experiments, with a 50% media change every third day. Analysis of cell survival was determined with calcein-AM/propidium iodide, as described below. Clonal Analysis Purified (A2B5+) O-2A/OPCs were isolated from P7 rat corpus callosa as above and plated at a density of 25 cells/cm2 in poly-L-lysine-coated 24-well plates in DMEM:F12 complete media supplemented with 10 ng/mL PDGF-AA immediately after purification. Experiments were initiated after 24 h. Cells were fixed after 5 d in 4% paraformaldehyde, stained with antibodies against A2B5 (1:4; in-house hybridoma, ATCC) and GalC (1:4; in-house hybridoma, ATCC), and counterstained with DAPI (1 μg/mL; Invitrogen #D1306), and the size and composition of each clone was then scored. Spontaneous generation of GalC+ OLs (i.e., in the absence of differentiation conditions) or Type 2 astrocytes was not detected in this paradigm, and so only the number of progenitor cells (A2B5+GalC−) per clone is reported. Cell Migration Cell migration with agarose drops was performed as previously reported [214, 215]. Briefly, purified (A2B5+) O-2A/OPCs were isolated from P7 rat corpus callosa as above, resuspended in 0.3% low-melt agarose (at 37°C; Sigma #A0701), and diluted in DMEM:F12 complete media supplemented with 10 ng/mL PDGF-AA at a density of 4 x 104 cells/μL, and 1.5 μL of the cell-agarose mixture was plated in the center of a poly-L-lysine-coated 24-well plate. The agarose was allowed to gel at 4°C for 10 min before DMEM:F12 complete media supplemented with 10 ng/mL PDGF-AA, with and without Psy. (PDGF-AA was omitted in some wells as controls for migration, as O-2A/OPC motility is stimulated in vitro by PDGF). Half of the media, with and without Psy, was replaced daily for 3 d, after which point the cells were loaded with calcein-AM (200 nM) and imaged. The distance migrated from the agarose drop to the leading edge was quantified as reported [215]. Immunocytochemistry Cells were fixed with 4% paraformaldehyde for 20 min before permeabilization for 10 min with 0.5% Triton X-100 (Sigma #X100) in blocking media (Earl’s Balanced Salt Solution [Gibco] with 5% calf serum and 1% BSA Fraction V). Permeabilized cells were then blocked for 1 h at 25°C before overnight incubation with primary antibodies (A2B5 hybridoma [IgM, 1:4], GalC hybridoma [IgG3, 1:10], Olig2 [1:500; Millipore #MABN50], Ki67 [1:1000, BD Pharmingen #550609], GFAP [1:2000; DAKO #Z0334], Tuj1 [1:2000; Abcam #14545]) diluted in blocking media at 4°C. After washing, cells were incubated with species- and isotype-matched Alexa Fluor-conjugated secondary antibodies (1:2000; Invitrogen) and counterstained with DAPI (1 μg/mL; Invitrogen #D1306) for 30 min at 25°C before final washes with PBS and ddH2O. Immunohistochemistry Mice were transcardially perfused with 4% paraformaldehyde/PBS. Isolated tissue was post-fixed for 24 h in 4% paraformaldehyde and normalized for 48 h in 20% sucrose. Brains were sectioned at 15-μm thickness in OCT (Tissue Tek) using a cryotome and immunostained with Ki67 (1:250; BD Pharmingen #550609), Olig2 (1:500; Millipore #MABN50), GST-pi (1:500; BD Biosciences #610718), Fluoromyelin (Invitrogen), and DAPI (Invitrogen). Mosaic images were acquired using a Leica TCS SP5 laser confocal microscope with a 40x oil immersion lens. Data represent analyses of the corpus callosa of at least three WT and three twitcher brains from separate litters. Analysis of Cell Survival Cells were incubated with 200 nM calcein-AM and 1 μg/mL prodium iodide for 30 min at 37°C to determine the proportion of live and dead cells, respectively, in an experimental condition. Single-cell analysis was performed using a Celigo cytometer (Nexcelom) and the %Live corrected values are reported (calcein+PI−). Analysis of Cell Proliferation Rate The total number of cells per well was determined using Brightfield analysis with a Celigo cytometer (Nexcelom) daily across 5 d, and cell numbers were normalized to the number of cells at the beginning of the experiment (Day 0). The proliferation rate was calculated as the fold change in cell number per unit time (days) using linear regression (0.95 < R2 < 1.0 over 5 d) for each well. Differentiation To induce differentiation, purified (A2B5+) O-2A/OPCs were exposed to DMEM:F12 complete media supplemented with 1 ng/mL PDGF-AA and 45 nM T3/T4 mixture (Sigma #T6397/#0397) and allowed to differentiate for 5 d. Cells were fixed in 4% paraformaldehyde, stained with antibodies against A2B5 (1:4; in-house hybridoma ATCC) and GalC (1:4; in-house hybridoma, ATCC), and counterstained with DAPI (1 μg/mL; Invitrogen #D1306), and the numbers of GalC+ OLs and A2B5+GalC− progenitor cells per condition were quantified. Small-Molecule Screen For analyses of cell division, purified (A2B5+) O-2A/OPCs were exposed to 1 μM Psy with each of the 1,040 compounds in the NINDS II Custom Collection library (Microsource) diluted to a final concentration of 0.2, 1, and 5 μM, each in duplicate, 15 compounds per plate. Each plate had control wells (in triplicate) of cells exposed to either vehicle (0.01% DMSO) or Psy (1 μM) alone. The proliferation rate was determined by daily cell counting using a Celigo cytometer (Nexcelom) across 5 d, as outlined above, and the increases in cell number within each well were internally normalized to the number of cells in that well at Day 0 (before the addition of Psy/compounds). The calculated proliferation rates were then normalized to the mean proliferation rate of in-plate vehicle-treated controls. Selected “hits” in both screens (15 for survival and 36 for proliferation) were rescreened at 0.2, 1, and 5 μM for their ability to significantly reduce Psy-induced suppression of division across 5 d. Those compounds that significantly reduced Psy toxicities with at least one of the three selected concentrations (22 in total) were rescreened using nine-point dose-response curves, ranging from 1 nM to 10 μM, to identify optimally protective concentrations to significantly reduce Psy-induced cell death at 5 d and/or suppression of division across 5 d. The list of 22 compounds that significantly reduced Psy-induced suppression of division was shortened to 15 by elimination of those minimally protective compounds for which commercial sources were cost prohibitive or unavailable. Growth Factor Screen For analyses of cell division, cells were exposed to 1 μM Psy with each growth factor (including BSA, with which all growth factors were diluted) diluted to a final concentration of 10, 33, and 100 ng/mL, each in triplicate, six growth factors per plate. Each plate had control wells (in triplicate) of cells exposed to either vehicle (0.01% DMSO) or Psy (1 μM) alone. The proliferation rate was determined by daily cell counting using a Celigo cytometer (Nexcelom) across 5 d, as outlined above, and the increases in cell number within each well were internally normalized to the number of cells in that well at Day 0 (before the addition of Psy/compounds). The calculated proliferation rates were then normalized to the mean proliferation rate of in-plate vehicle-treated controls. Fingerprint Analysis Concentrations of small-molecule inhibitors of signaling proteins were selected based on their ability to reduce phosphorylation of target proteins when analyzed by immunoblot when possible; the selected concentrations minimally enhanced—or did not alter—Psy’s effects on cell division when examined in the absence of protective agents. Purified (A2B5+) O-2A/OPCs were pretreated with the small-molecule inhibitors for 1 h before the addition of protective agent and Psy (each in triplicate). All plates had vehicle- and Psy-treated wells (each in triplicate), as well as Psy combined with the protective agent of interest, wells as controls. The proliferation rates were analyzed and normalized against vehicle-treated controls, as outlined above. Data were hierarchically clustered with an unweighted Euclidean distance similarity metric (complete linkage clustering) using Cluster 3.0 and visualized using TreeView. Delayed Administration of Protective Agents For experiments in which the administration of candidate protective agents was delayed, purified (A2B5+) O-2A/OPCs were exposed to the indicated concentrations of Psy for 2 d before each small molecule or IGF-1, prepared at a 10X concentration in DMEM:F12 complete media supplemented with 10 ng/mL PDGF-AA, was diluted to 1X so as to minimize dilution of Psy and perturbation of cells. An equal volume of DMEM:F12 complete media supplemented with 10 ng/mL PDGF-AA was added to untreated and Psy-only control wells. The proliferation rate was calculated from the daily change in cell number from time of administration (Day 2) for three days (Day 5), as outlined above. Lysosomal pH Measurements Purified (A2B5+) O-2A/OPCs were plated on poly-L-lysine-coated glass-bottom microwell dishes (MatTek Co., Ashland, MA; #P35G-1.5–14) in DMEM:F12 complete media supplemented with 10 ng/mL PDGF-AA after passaging. Cells were loaded with 500 μg/mL LysoSensor Yellow/Blue Dextran (Invitrogen) in complete media with PDGF for 24 h prior to treatment. After 24 h, the cells were fixed in 4% paraformaldehyde and were imaged using a Leica TCS SP5 laser confocal microscope with a 63X oil immersion lens. Using an excitation wavelength of 335 nm (405 diode), emission spectra at 450 nm (acidic) and 521 nm (alkaline) were quantified, and the ratio of these emissions was calculated using the Leica Advanced Fluorescence software. Live-cell imaging was performed as above, with the exception that cells were not fixed in PFA prior to imaging and analysis. To generate the lysosomal pH calibration curve, the pH of pre-loaded O2A/OPCs was measured as previously described [112]. Briefly, the cells were incubated in calibration buffers (20 mM MES, 110 mM KCl, and 20 mM NaCl containing 10 μM monensin and 20 μM nigericin; Sigma) adjusted to known pH values between 4.0 and 6.0 at 0.5 increments using HCl/NaOH for 1 h prior to imaging, and ratiometric quantification, as above. Calibration curves were generated using both fixed and live cells. Endocytosis Measurements Purified (A2B5+) O-2A/OPCs were exposed to indicated conditions for 24 h. Cells were trypsinized and resuspended at a density of 106/mL in conditioned (treated) medium and 1:1000 FluoSpheres polystyrene beads (Invitrogen). The cell:bead suspension was incubated at room temperature and gently inverted every 5 min. At indicated time points, 10 μL (10,000 cells) were transferred to 1 mL of ice-cold PBS and pelleted at 13,000 rpm at 4°C for 5 min. Cell pellets were resuspended in ice-cold 2% paraformaldehyde/PBS and transferred to PLL-coated 96-well dishes to adhere during fixation. The integrated fluorescence intensity per cell was measured using a Celigo cytometer (Nexcelom) and plotted over time; the time-to-half-maximal intensity was calculated using curve-fitting software (Prism). In Situ Cathepsin Activity Measurements Cathepsin B activity was measured as per manufacturer’s instructions (MagicRed Cathepsin B substrate, ICT). Briefly, purified (A2B5+) O-2A/OPCs were treated as indicated for 24 h before exposure to cell-permeant CathB substrate (1 μM); substrate cleavage occurred at 37°C for 1 h before the integrated fluorescent intensity per cell was quantified with a Celigo cytometer (Nexcelom). For cathepsin D measurements, fluorescently labeled, cell-permeant CathD active-site inhibitor (BODIPY-FL Pepstatin A; 10 μM) was added to O-2A/OPCs that had been treated as indicated for 24 h; active CathD labeling occurred at 37°C for 1 h before the integrated fluorescent intensity per cell was quantified with a Celigo cytometer (Nexcelom). In Situ Lipid Accumulation Neutral lipid and phospholipid accumulation were quantified with the HCS LipidTox Phospholipidosis and Steatosis Detection Kit (Invitrogen) as per the manufacturer’s directions. Positive controls cyclosporin A (10 μM) and propranolol (10 μM) were used, respectively. CFTR Knockdown Rat O-2A/OPCs were exposed to either 50-nM pools of four siRNA constructs targeting rat CFTR or 50-nM pools of four control siRNA constructs that do not target the rat genome with DharmaFECT-1 transfection reagent (1:1000), as per manufacturer’s directions (Dharmacon), for 24 h. Four d post transfection, cells were passaged and either lysed for western blot analysis or plated on glass-bottom dishes for analysis of lysosomal pH, as outlined above. Western blot analysis was performed as previously reported [216] using an anti-rabbit CFTR antibody (Cell Signaling) and HRP-conjugated beta-actin (Santa Cruz). Psy Quantification Mice were killed at P35 and transcardially perfused with ice-cold PBS. Isolated tissue was flash-frozen in liquid nitrogen and stored at -80°C until analysis. Analysis of sphingolipids was performed by Dr. Jacek Bielawski from the Lipidomics Core at the Medical University of South Carolina (MUSC) using liquid chromatography-mass spectrometry (LC-MS/MS) and supercritical fluid chromatography-mass spectrometry (SFC-MS/MS) methodologies, as described previously [217]. Animal Treatment Adult heterozygote (Galctwi/+) C57Bl/6J (B6.CE-Galctwi/J) mice were originally obtained from Dr. Ernesto Bongarzone (University of Illinois at Chicago, Chicago, IL) and used as breeder pairs to generate homozygous (twi; Galctwi/twi) twitcher mice and WT (Galc+/+) C57Bl/6J mice. All animal procedures were approved by the Institutional Animal Care and Use Committee (IACUC) at the University of Rochester School of Medicine and Dentistry and conformed to the requirements of the Animal Welfare Act. In total, three cohorts of aged-matched mice from different litters received daily IP injections beginning at P10: WT mice receiving saline (n = 6); twi receiving saline (n = 6); twi receiving 1 mg/kg NKH-477 (Tocris) in saline (n = 8). Mice were euthanized at P35 and tissue was isolated after transcardial perfusion with 4% paraformaldehyde. Immunohistochemical analyses were completed as outlined above. For survival analysis, animals were provided moistened chow and hydragel water packs and monitored daily for weight gain. Animals were euthanized when moribund, as assessed by when the animals could no longer ambulate to maintain food and water intake or exhibited clinical signs of pain such as hunched posture and ruffled fur, as determined on a daily basis. Animals were euthanized using CO2 exposure and cervical dislocation. Motor Behavior Testing Aged-matched mice from different litters received daily IP injections beginning at P10: WT mice receiving saline (n = 4); twi receiving saline (n = 3); twi receiving 1 mg/kg NKH-477 (Tocris) in saline (n = 4) were analyzed for locomotive ability and gait using the Phenoscan suite and Runwayscan software (CleverSys, Inc) at P25. Multiple parameters, including stance, stride, swing, brake, and propulsion time (milliseconds), stride length (millimeters), and average speed (millimeters/s) were collected for each animal over three compliant trials and averaged for both front and rear paws. Statistical Analyses Bar graphs are plotted as mean ± SEM and represent at minimum three independent biological replicates performed in triplicate, except where noted. Two-group comparisons were analyzed using a Student’s t test, and multiple-group comparisons were analyzed using an ANOVA with Bonferroni post-hoc test. Prism (v5.0; GraphPad) was used for data analysis and presentation. Ethics Statement The University of Rochester RSRB has reviewed this study and determined that based on federal (45 CFR 46.102) and University criteria the study does not qualify as human subjects research and has waived the need for consent (RSRB#00024759). All animal procedures were performed under guidelines of the National Institutes of Health and approved by the Institutional Animal Care and Utilization Committee (IACUC) of the University of Rochester Medical Center, Rochester, NY (UCAR#2001–140). Lipids Lipids used in this study were purchased from Matreya and were of highest purity. All lipids were resuspended in anhydrous dimethyl sulfoxide (DMSO) to 10 mM, stored at -20°C or -80°C, and resuspended in media before use. Comparable results for OPC division and survival were obtained with Psy purchased from Sigma-Aldrich and Santa Cruz Biotechnology. O-2A/OPC Isolation, Purification, and Culture Corpus callosa of P7 Sprague-Dawley rats (Charles River) were micro-dissected, finely minced with a sterile blade, and digested for 20 min in 2.2 mg/mL collagenase (Worthington #4189), 20 Kunitz/mL DNase (Sigma D4263) in HBSS (Gibco #114170–161) supplemented with Sato medium (see below). Collagenase-containing medium was then replaced with 20 Kunits/mL DNase and papain solution (1:40, activated per manufacturer’s directions; Worthington #LS003127) in HBSS/Sato for 20 min. Tissue was then sequentially triturated with 21-, 25-, and 26-guage needles in 35 Kunits/mL DNase in DMEM:F12 complete media (see below) before dissociated cells were plated on tissue culture plastic for 10 min (37°C, 7% CO2). Nonadherent cells were pelleted by centrifugation (5 min, 500 xg), resuspended in degassed HBSS/Sato supplemented with 1% BSA Fraction V and anti-A2B5 MACS beads (1:50; Miltenyi Biotec #130-093-388), and incubated on ice for 20 min. Cells were pelleted and sorted with MACS columns as per manufacturer’s directions (Miltenyi). A2B5+ cells were plated on tissue culture plastic coated with poly-L-lysine (1 μg/cm2 for 20 min; Sigma #P1274) in DMEM:F12 (Gibco #11330–057) supplemented with 10 μg/mL insulin (Sigma #I5500), 100 μg/mL holotransferrin (Sigma #T2252), Sato media (final concentration: 0.03% BSA Fraction V [Sigma #A7979-50ML], 10 μM putrescine [Sigma #P7505], 200 nM progesterone [Sigma #P0130], 235 nM sodium selenite [Sigma #S1382]), 50 μg/mL gentamycin (Gibco #15750–060), 10 ng/mL PDGF-AA (R&D #221-AA), and 5 ng/mL basic FGF (Miltenyi #130–093) and maintained at 37°C (7% CO2). Cells were passaged with 0.05% trypsin-EDTA (Gibco #2300), neutralized with 80 Kunitz/mL soybean trypsin inhibitor (Sigma #T9003), and replated in DMEM:F12 complete media supplemented with 10 ng/mL PDGF-AA (and without basic FGF). Purified O-2A/OPCs were passaged no more than once and were maintained in culture for at most 7–9 d in vitro for all experiments. To generate OLs, purified O-2A/OPCs were maintained in DMEM:F12 containing Sato components, transferrin, insulin, 100 pg/mL PDGF-AA, and 45 nM T3/T4 (Sigma #T6397/#0397) for 5 d before initiation of experiments. Human cells were isolated from corpus callosal fields of fetal week 18–21 de-identified tissue, purified with anti-CD140a-coupled magnetic beads (1:100; BD Biosciences #558774; Miltenyi), and maintained as above. Cortical and Hippocampal Neuron Isolation and Culture Embryonic neurons were isolated from E18 Sprague-Dawley rats (Charles River) and maintained as previously described [213]. Briefly, isolated tissue was digested in papain solution (1:50, activated per manufacturer’s directions; Worthington #LS003127) in HBSS/Sato for 20 min at 37°C (7% CO2). Pelleted tissue was then triturated with a pulled glass Pasteur pipet in 80 Kunitz/mL DNase (HBSS; Sigma #D4263), and dissociated cells were pelleted through a 0.5 M sucrose cushion (10 min, 500 xg). Immature neurons were plated on poly-L-lysine-coated tissue culture plastic (1 μg/cm2 for 20 min; Sigma #P1274) in NeuroBasal media (Gibco #21103–049) with 50 μg/mL gentamycin (Gibco #15750–060), 2 mM L-glutamine (Gibco #25030–081), and 1 X B27 serum-free supplement (Gibco #17504044). Neurons were allowed to mature for 7 d at 37°C (7% CO2) before initiation of experiments, with a 50% media change every third day. Analysis of cell survival was determined with calcein-AM/propidium iodide, as described below. Clonal Analysis Purified (A2B5+) O-2A/OPCs were isolated from P7 rat corpus callosa as above and plated at a density of 25 cells/cm2 in poly-L-lysine-coated 24-well plates in DMEM:F12 complete media supplemented with 10 ng/mL PDGF-AA immediately after purification. Experiments were initiated after 24 h. Cells were fixed after 5 d in 4% paraformaldehyde, stained with antibodies against A2B5 (1:4; in-house hybridoma, ATCC) and GalC (1:4; in-house hybridoma, ATCC), and counterstained with DAPI (1 μg/mL; Invitrogen #D1306), and the size and composition of each clone was then scored. Spontaneous generation of GalC+ OLs (i.e., in the absence of differentiation conditions) or Type 2 astrocytes was not detected in this paradigm, and so only the number of progenitor cells (A2B5+GalC−) per clone is reported. Cell Migration Cell migration with agarose drops was performed as previously reported [214, 215]. Briefly, purified (A2B5+) O-2A/OPCs were isolated from P7 rat corpus callosa as above, resuspended in 0.3% low-melt agarose (at 37°C; Sigma #A0701), and diluted in DMEM:F12 complete media supplemented with 10 ng/mL PDGF-AA at a density of 4 x 104 cells/μL, and 1.5 μL of the cell-agarose mixture was plated in the center of a poly-L-lysine-coated 24-well plate. The agarose was allowed to gel at 4°C for 10 min before DMEM:F12 complete media supplemented with 10 ng/mL PDGF-AA, with and without Psy. (PDGF-AA was omitted in some wells as controls for migration, as O-2A/OPC motility is stimulated in vitro by PDGF). Half of the media, with and without Psy, was replaced daily for 3 d, after which point the cells were loaded with calcein-AM (200 nM) and imaged. The distance migrated from the agarose drop to the leading edge was quantified as reported [215]. Immunocytochemistry Cells were fixed with 4% paraformaldehyde for 20 min before permeabilization for 10 min with 0.5% Triton X-100 (Sigma #X100) in blocking media (Earl’s Balanced Salt Solution [Gibco] with 5% calf serum and 1% BSA Fraction V). Permeabilized cells were then blocked for 1 h at 25°C before overnight incubation with primary antibodies (A2B5 hybridoma [IgM, 1:4], GalC hybridoma [IgG3, 1:10], Olig2 [1:500; Millipore #MABN50], Ki67 [1:1000, BD Pharmingen #550609], GFAP [1:2000; DAKO #Z0334], Tuj1 [1:2000; Abcam #14545]) diluted in blocking media at 4°C. After washing, cells were incubated with species- and isotype-matched Alexa Fluor-conjugated secondary antibodies (1:2000; Invitrogen) and counterstained with DAPI (1 μg/mL; Invitrogen #D1306) for 30 min at 25°C before final washes with PBS and ddH2O. Immunohistochemistry Mice were transcardially perfused with 4% paraformaldehyde/PBS. Isolated tissue was post-fixed for 24 h in 4% paraformaldehyde and normalized for 48 h in 20% sucrose. Brains were sectioned at 15-μm thickness in OCT (Tissue Tek) using a cryotome and immunostained with Ki67 (1:250; BD Pharmingen #550609), Olig2 (1:500; Millipore #MABN50), GST-pi (1:500; BD Biosciences #610718), Fluoromyelin (Invitrogen), and DAPI (Invitrogen). Mosaic images were acquired using a Leica TCS SP5 laser confocal microscope with a 40x oil immersion lens. Data represent analyses of the corpus callosa of at least three WT and three twitcher brains from separate litters. Analysis of Cell Survival Cells were incubated with 200 nM calcein-AM and 1 μg/mL prodium iodide for 30 min at 37°C to determine the proportion of live and dead cells, respectively, in an experimental condition. Single-cell analysis was performed using a Celigo cytometer (Nexcelom) and the %Live corrected values are reported (calcein+PI−). Analysis of Cell Proliferation Rate The total number of cells per well was determined using Brightfield analysis with a Celigo cytometer (Nexcelom) daily across 5 d, and cell numbers were normalized to the number of cells at the beginning of the experiment (Day 0). The proliferation rate was calculated as the fold change in cell number per unit time (days) using linear regression (0.95 < R2 < 1.0 over 5 d) for each well. Differentiation To induce differentiation, purified (A2B5+) O-2A/OPCs were exposed to DMEM:F12 complete media supplemented with 1 ng/mL PDGF-AA and 45 nM T3/T4 mixture (Sigma #T6397/#0397) and allowed to differentiate for 5 d. Cells were fixed in 4% paraformaldehyde, stained with antibodies against A2B5 (1:4; in-house hybridoma ATCC) and GalC (1:4; in-house hybridoma, ATCC), and counterstained with DAPI (1 μg/mL; Invitrogen #D1306), and the numbers of GalC+ OLs and A2B5+GalC− progenitor cells per condition were quantified. Small-Molecule Screen For analyses of cell division, purified (A2B5+) O-2A/OPCs were exposed to 1 μM Psy with each of the 1,040 compounds in the NINDS II Custom Collection library (Microsource) diluted to a final concentration of 0.2, 1, and 5 μM, each in duplicate, 15 compounds per plate. Each plate had control wells (in triplicate) of cells exposed to either vehicle (0.01% DMSO) or Psy (1 μM) alone. The proliferation rate was determined by daily cell counting using a Celigo cytometer (Nexcelom) across 5 d, as outlined above, and the increases in cell number within each well were internally normalized to the number of cells in that well at Day 0 (before the addition of Psy/compounds). The calculated proliferation rates were then normalized to the mean proliferation rate of in-plate vehicle-treated controls. Selected “hits” in both screens (15 for survival and 36 for proliferation) were rescreened at 0.2, 1, and 5 μM for their ability to significantly reduce Psy-induced suppression of division across 5 d. Those compounds that significantly reduced Psy toxicities with at least one of the three selected concentrations (22 in total) were rescreened using nine-point dose-response curves, ranging from 1 nM to 10 μM, to identify optimally protective concentrations to significantly reduce Psy-induced cell death at 5 d and/or suppression of division across 5 d. The list of 22 compounds that significantly reduced Psy-induced suppression of division was shortened to 15 by elimination of those minimally protective compounds for which commercial sources were cost prohibitive or unavailable. Growth Factor Screen For analyses of cell division, cells were exposed to 1 μM Psy with each growth factor (including BSA, with which all growth factors were diluted) diluted to a final concentration of 10, 33, and 100 ng/mL, each in triplicate, six growth factors per plate. Each plate had control wells (in triplicate) of cells exposed to either vehicle (0.01% DMSO) or Psy (1 μM) alone. The proliferation rate was determined by daily cell counting using a Celigo cytometer (Nexcelom) across 5 d, as outlined above, and the increases in cell number within each well were internally normalized to the number of cells in that well at Day 0 (before the addition of Psy/compounds). The calculated proliferation rates were then normalized to the mean proliferation rate of in-plate vehicle-treated controls. Fingerprint Analysis Concentrations of small-molecule inhibitors of signaling proteins were selected based on their ability to reduce phosphorylation of target proteins when analyzed by immunoblot when possible; the selected concentrations minimally enhanced—or did not alter—Psy’s effects on cell division when examined in the absence of protective agents. Purified (A2B5+) O-2A/OPCs were pretreated with the small-molecule inhibitors for 1 h before the addition of protective agent and Psy (each in triplicate). All plates had vehicle- and Psy-treated wells (each in triplicate), as well as Psy combined with the protective agent of interest, wells as controls. The proliferation rates were analyzed and normalized against vehicle-treated controls, as outlined above. Data were hierarchically clustered with an unweighted Euclidean distance similarity metric (complete linkage clustering) using Cluster 3.0 and visualized using TreeView. Delayed Administration of Protective Agents For experiments in which the administration of candidate protective agents was delayed, purified (A2B5+) O-2A/OPCs were exposed to the indicated concentrations of Psy for 2 d before each small molecule or IGF-1, prepared at a 10X concentration in DMEM:F12 complete media supplemented with 10 ng/mL PDGF-AA, was diluted to 1X so as to minimize dilution of Psy and perturbation of cells. An equal volume of DMEM:F12 complete media supplemented with 10 ng/mL PDGF-AA was added to untreated and Psy-only control wells. The proliferation rate was calculated from the daily change in cell number from time of administration (Day 2) for three days (Day 5), as outlined above. Lysosomal pH Measurements Purified (A2B5+) O-2A/OPCs were plated on poly-L-lysine-coated glass-bottom microwell dishes (MatTek Co., Ashland, MA; #P35G-1.5–14) in DMEM:F12 complete media supplemented with 10 ng/mL PDGF-AA after passaging. Cells were loaded with 500 μg/mL LysoSensor Yellow/Blue Dextran (Invitrogen) in complete media with PDGF for 24 h prior to treatment. After 24 h, the cells were fixed in 4% paraformaldehyde and were imaged using a Leica TCS SP5 laser confocal microscope with a 63X oil immersion lens. Using an excitation wavelength of 335 nm (405 diode), emission spectra at 450 nm (acidic) and 521 nm (alkaline) were quantified, and the ratio of these emissions was calculated using the Leica Advanced Fluorescence software. Live-cell imaging was performed as above, with the exception that cells were not fixed in PFA prior to imaging and analysis. To generate the lysosomal pH calibration curve, the pH of pre-loaded O2A/OPCs was measured as previously described [112]. Briefly, the cells were incubated in calibration buffers (20 mM MES, 110 mM KCl, and 20 mM NaCl containing 10 μM monensin and 20 μM nigericin; Sigma) adjusted to known pH values between 4.0 and 6.0 at 0.5 increments using HCl/NaOH for 1 h prior to imaging, and ratiometric quantification, as above. Calibration curves were generated using both fixed and live cells. Endocytosis Measurements Purified (A2B5+) O-2A/OPCs were exposed to indicated conditions for 24 h. Cells were trypsinized and resuspended at a density of 106/mL in conditioned (treated) medium and 1:1000 FluoSpheres polystyrene beads (Invitrogen). The cell:bead suspension was incubated at room temperature and gently inverted every 5 min. At indicated time points, 10 μL (10,000 cells) were transferred to 1 mL of ice-cold PBS and pelleted at 13,000 rpm at 4°C for 5 min. Cell pellets were resuspended in ice-cold 2% paraformaldehyde/PBS and transferred to PLL-coated 96-well dishes to adhere during fixation. The integrated fluorescence intensity per cell was measured using a Celigo cytometer (Nexcelom) and plotted over time; the time-to-half-maximal intensity was calculated using curve-fitting software (Prism). In Situ Cathepsin Activity Measurements Cathepsin B activity was measured as per manufacturer’s instructions (MagicRed Cathepsin B substrate, ICT). Briefly, purified (A2B5+) O-2A/OPCs were treated as indicated for 24 h before exposure to cell-permeant CathB substrate (1 μM); substrate cleavage occurred at 37°C for 1 h before the integrated fluorescent intensity per cell was quantified with a Celigo cytometer (Nexcelom). For cathepsin D measurements, fluorescently labeled, cell-permeant CathD active-site inhibitor (BODIPY-FL Pepstatin A; 10 μM) was added to O-2A/OPCs that had been treated as indicated for 24 h; active CathD labeling occurred at 37°C for 1 h before the integrated fluorescent intensity per cell was quantified with a Celigo cytometer (Nexcelom). In Situ Lipid Accumulation Neutral lipid and phospholipid accumulation were quantified with the HCS LipidTox Phospholipidosis and Steatosis Detection Kit (Invitrogen) as per the manufacturer’s directions. Positive controls cyclosporin A (10 μM) and propranolol (10 μM) were used, respectively. CFTR Knockdown Rat O-2A/OPCs were exposed to either 50-nM pools of four siRNA constructs targeting rat CFTR or 50-nM pools of four control siRNA constructs that do not target the rat genome with DharmaFECT-1 transfection reagent (1:1000), as per manufacturer’s directions (Dharmacon), for 24 h. Four d post transfection, cells were passaged and either lysed for western blot analysis or plated on glass-bottom dishes for analysis of lysosomal pH, as outlined above. Western blot analysis was performed as previously reported [216] using an anti-rabbit CFTR antibody (Cell Signaling) and HRP-conjugated beta-actin (Santa Cruz). Psy Quantification Mice were killed at P35 and transcardially perfused with ice-cold PBS. Isolated tissue was flash-frozen in liquid nitrogen and stored at -80°C until analysis. Analysis of sphingolipids was performed by Dr. Jacek Bielawski from the Lipidomics Core at the Medical University of South Carolina (MUSC) using liquid chromatography-mass spectrometry (LC-MS/MS) and supercritical fluid chromatography-mass spectrometry (SFC-MS/MS) methodologies, as described previously [217]. Animal Treatment Adult heterozygote (Galctwi/+) C57Bl/6J (B6.CE-Galctwi/J) mice were originally obtained from Dr. Ernesto Bongarzone (University of Illinois at Chicago, Chicago, IL) and used as breeder pairs to generate homozygous (twi; Galctwi/twi) twitcher mice and WT (Galc+/+) C57Bl/6J mice. All animal procedures were approved by the Institutional Animal Care and Use Committee (IACUC) at the University of Rochester School of Medicine and Dentistry and conformed to the requirements of the Animal Welfare Act. In total, three cohorts of aged-matched mice from different litters received daily IP injections beginning at P10: WT mice receiving saline (n = 6); twi receiving saline (n = 6); twi receiving 1 mg/kg NKH-477 (Tocris) in saline (n = 8). Mice were euthanized at P35 and tissue was isolated after transcardial perfusion with 4% paraformaldehyde. Immunohistochemical analyses were completed as outlined above. For survival analysis, animals were provided moistened chow and hydragel water packs and monitored daily for weight gain. Animals were euthanized when moribund, as assessed by when the animals could no longer ambulate to maintain food and water intake or exhibited clinical signs of pain such as hunched posture and ruffled fur, as determined on a daily basis. Animals were euthanized using CO2 exposure and cervical dislocation. Motor Behavior Testing Aged-matched mice from different litters received daily IP injections beginning at P10: WT mice receiving saline (n = 4); twi receiving saline (n = 3); twi receiving 1 mg/kg NKH-477 (Tocris) in saline (n = 4) were analyzed for locomotive ability and gait using the Phenoscan suite and Runwayscan software (CleverSys, Inc) at P25. Multiple parameters, including stance, stride, swing, brake, and propulsion time (milliseconds), stride length (millimeters), and average speed (millimeters/s) were collected for each animal over three compliant trials and averaged for both front and rear paws. Statistical Analyses Bar graphs are plotted as mean ± SEM and represent at minimum three independent biological replicates performed in triplicate, except where noted. Two-group comparisons were analyzed using a Student’s t test, and multiple-group comparisons were analyzed using an ANOVA with Bonferroni post-hoc test. Prism (v5.0; GraphPad) was used for data analysis and presentation. Supporting Information S1 Data. Quantification and analyses underlying the data summarized in all figures and Supporting Information figures. https://doi.org/10.1371/journal.pbio.1002583.s001 (XLSX) S1 Fig. Psy causes a diverse array of cellular and biochemical toxicities in O-2A/OPCs in vitro and in vivo. (A) Quantification of the percentage of GalC+ rat OLs derived from A2B5+ OPCs over 5 d in differentiation conditions, with and without 1 μM Psy. (B) Representative immunofluorescent images of rat O-2A/OPCs exposed to 1 μM Psy or vehicle (DMSO) for 1 d, showing cytoskeletal collapse, and stained with phalloidin (actin) and tubulin. (C) Representative images and quantification of rat O-2A/OPCs, with and without 3 μM Psy, migrating radially from an agarose drop after 3 d; calcein-AM (green) and prodium iodide (red) were used to identify live and dead cells, respectively. Note that PDGF stimulates O-2A/OPC migration. White dashed line: agarose drop border. (D) Representative immunofluorescent images of rat O-2A/OPCs exposed to 1 μM Psy or positive assay controls cyclosporine A (CysA; 10 μM) and propranolol (10 μM) for 2 d, stained for neutral lipid and phospholipid accumulation, respectively. (E) Representative brightfield and immunofluorescent images of rat O-2A/OPCs exposed to 1 μM Psy or vehicle (DMSO) for 1 d before the addition of fluorescently labeled nanobeads for the indicated times. (F) Quantification of lysosomal pH in live rat O-2A/OPCs exposed to vehicle (0.01% DMSO), 100 nM BafA, or 1 μM Psy for 24 h. (G) Representative immunofluorescent time-lapse images of rat O-2A/OPCs exposed to vehicle (0.01% DMSO), 100 nM BafA, or 1 μM Psy for 0–5 min. Data for all graphs displayed as mean ± SEM; *p < 0.05, †p < 0.001 versus control, unless otherwise indicated. See also S1, S2 and S3 Movies for time-lapse movies of lysosomal pH changes. Data presented in this figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.1002583.s002 (TIFF) S2 Fig. Unbiased screening identifies chemically diverse candidate protective agents that reduce Psy toxicities. (A) Physicochemical characterization of small molecules that reduce Psy-induced (D) cell death or (E) suppression of division, including atomic composition (% by mass), molecular weight (Daltons), logP partition coefficient, number of ring structures, and surface area (Å2). (B) Quantification of cell division of rat O-2A/OPCs exposed to 1.5 μM Psy for 5 d, with and without the indicated growth factors at 10, 33, or 100 ng/mL. Data for all graphs displayed as mean ± SEM; ap < 0.05, bp < 0.01, cp < 0.001 versus Psy-only treatment. See S1 and S2 Tables for drugs and concentrations used. Data presented in this figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.1002583.s003 (TIFF) S3 Fig. Protective agents converge on a limited number of common necessary pathways for their activity. Representative “fingerprints of protection” for the functionally and structurally unrelated candidate drugs 2G08, 2F11, and 8D08. Data represent mean ± SEM. See also See S1 and S2 Tables for drugs and concentrations, and S3 Table for details on the “fingerprinting” screen. Data presented in this figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.1002583.s004 (TIFF) S4 Fig. Candidate protective agents do not reduce basal lysosomal pH in the absence of Psy. (A) A representative western blot of CFTR knockdown versus NT controls in rat O-2A/OPCs, 4 d post transfection. Quantification of lysosomal pH in rat O-2A/OPCs, with or without CFTR knockdown (5 d post transfection), exposed to 1 μM Psy or 1 μM Psy and 333 nM RP-107 for 24 h. (B) Quantification of lysosomal pH of rat O-2A/OPCs exposed to the indicated drugs for 24 h in the absence of Psy. Data for all graphs displayed as mean ± SEM; *p < 0.05, **p < 0.01, †p < 0.001. See S1 and S2 Tables for drugs and concentrations used. Data presented in this figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.1002583.s005 (TIFF) S5 Fig. Protective agents rescue critical O-2A/OPC behaviors and lysosomal function in response to lysosphingolipids accumulating in other LSDs. (A) Proliferation analysis of rat O-2A/OPCs exposed to 1.5 μM Psy, 1 μM GlcSph, 3 μM Lyso-SF, or 12 μM LacSph for 5 d, with and without the indicated protective agents. (B) Proliferation analysis of rat O-2A/OPCs exposed to 1.5 μM Psy, 1 μM GlcSph, 3 μM Lyso-SF, or 12 μM LacSph for 5 d, with and without the indicated protective agents, which were administered 2 d after the indicated lyso-lipid. (C) Venn diagram summarizing (B) for all lyso-lipids. Data for all graphs displayed as mean ± SEM; ap < 0.05, bp < 0.01, cp < 0.001 versus lipid-only treatment. See S1 and S2 Tables for drugs and concentrations used. Data presented in this figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.1002583.s006 (TIFF) S6 Fig. Lysosphingolipids disrupt human O-2A/OPC behaviors. (A) Representative immunofluorescent images of human fetal O-2A/OPCs maintained in 10 ng/mL PDGF + 10 ng/mL bFGF (“PDGF+FGF”); 100 pg/mL PDGF (“- PDGF”); 1 ng/mL PDGF + 40 ng/mL T3/T4 (“+T3/T4”); and 1 ng/mL PDGF + 10 ng/mL BMP4 (“+BMP4”) for 5 d in mass culture. A2B5+: glial progenitor cells; GalC+: OLs; GFAP: astrocytes; A2B5+/GFAP+: Type-2 astrocytes; Ki67+: mitotically active cells; Olig2+: oligodendroglial-lineage cells. (B) Quantification of (A). (C) Quantification of human fetal O-2A/OPCs maintained as in (A) for 5 d, except at clonal density. The mean number of cells immunopositive for the indicated stain per clone, as well as the percentage of clones containing at least one immunopositive cell, are reported. Note that NeuN+ or Tuj1+ neurons were never detected in mass or clonal culture. Data for all graphs displayed as mean ± SD for one GW20 human sample. All experiments were repeated in four human GW19-21 samples with comparable results. Data presented in this figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.1002583.s007 (TIFF) S7 Fig. NKH-477 treatment reduces twitcher mouse gait abnormalities. Quantification of gait for P25 vehicle-treated WT (n = 3–4), vehicle-treated twitcher mice (n = 3), and NKH-treated twitcher mice (n = 4), including measurements of stance, break, propel, swing, and stride time, as well stride length, for front and rear paws. Data for all graphs displayed as mean ± SEM; *p < 0.05, **p < 0.01, †p < 0.001 versus WT; ap < 0.05, cp < 0.001 versus vehicle-treated twitcher. Data presented in this figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.1002583.s008 (TIFF) S1 Movie. Time-lapse movie of rat O-2A/OPCs loaded with a ratiometric lysosomal pH dye and exposed to vehicle (0.01% DMSO) for 5 min. https://doi.org/10.1371/journal.pbio.1002583.s009 (MP4) S2 Movie. Time-lapse movie of rat O-2A/OPCs loaded with a ratiometric lysosomal pH dye and exposed to 100 nM BafA for 5 min. https://doi.org/10.1371/journal.pbio.1002583.s010 (MP4) S3 Movie. Time-lapse movie of rat O-2A/OPCs loaded with a ratiometric lysosomal pH dye and exposed to 1 μM Psy for 5 min. https://doi.org/10.1371/journal.pbio.1002583.s011 (MP4) S1 Table. Chemical structures of lead protective compounds. https://doi.org/10.1371/journal.pbio.1002583.s012 (DOCX) S2 Table. List of lead protective compounds, optimal concentrations used in human and rat O-2A/OPCs, and meta-analyses of clinical usage. https://doi.org/10.1371/journal.pbio.1002583.s013 (DOCX) S3 Table. List of pharmacological inhibitors, their concentrations, and their protein targets used in the fingerprinting secondary screen. https://doi.org/10.1371/journal.pbio.1002583.s014 (DOCX) S4 Table. List of lipid concentrations used in human and rat O-2A/OPCs. https://doi.org/10.1371/journal.pbio.1002583.s015 (DOCX) Acknowledgments We thank Dr. Ernesto Bongarazone (University of Illinois at Chicago) for generously providing the twitcher mouse colony. We thank Dr. Hartmut Land for critical comments on the manuscript. We thank Ashley Chang, Michelle Lacagnina, and William Li for technical assistance, as well as Dr. Erhard Beiberich (Augusta University) and Dr. Jacek Bielawski at the MUSC sphingolipidomics core facility.

doi: 10.1371/journal.pbio.2000705pmid: 27906976

A Laboratory CIRS for Academic Biomedical Research In clinical medicine, it is often necessary to clearly define and classify critical incidents. This is not the case in preclinical research, in which there is no standard set of terms for recording such incidents. However, the particularly open nature of the research process may lead to unanticipated and even unorthodox events that deserve reporting. We set out to improve academic biomedical research by systematically learning from errors and mistakes. We established a simple and scalable CIRS suitable for the preclinical biomedical research environment. We then implemented and tested it in a typical, multidisciplinary academic laboratory. The Department of Experimental Neurology, with approximately 100 students, researchers, and technicians, carries out multiprofessional academic research in preclinical biomedicine with such standard approaches and techniques as in vivo and in vitro modeling of disease, cell biology, molecular biology, and biochemistry, as well as imaging (from multiphoton microscopy to magnetic resonance imaging). Box 1 lists the essential features we expect from a CIRS in basic and preclinical biomedicine. Box 1. Features of a Laboratory CIR System for Experimental Biomedicine Easy to set up, run, and administer Easy to use, accessible, intuitive, and unambiguous Should be scalable so that it works in small single-investigator groups as well as in a large institute Allows anonymous reporting Allows free expression of “what actually happened”: the reporter’s own version of events Reports must be handled in a nonpunitive manner The incidents reported can be regularly analyzed by experts Learning points from such analyses need to be fed back promptly to those who need to know The reports are visible, and a clear path of action is communicated Feedback results in enhanced learning regarding the incident’s cause and systemic changes that will prevent its recurrence The flow of information and resultant activities in LabCIRS is summarized in Fig 1: Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Cartoon of how LabCIRS helps to prevent further mishaps and fosters an error culture. “Error”: a researcher mistook two faintly labeled reagents A and B, which ruined his experiment. “Reporting”: entry of the incident into LabCIRS. “Assessment”: a group of experts (scientists and technicians) reviews the error and takes preventive action by color labelling the reagents. “Feedback”: the errors as well as the measure to prevent it in the future are communicated to the entire laboratory. https://doi.org/10.1371/journal.pbio.2000705.g001 LabCIRS can be accessed from every computer logged into the intranet of the department. Incidents are reported anonymously, either in German or in English (see S1 and S2 Figs). A demo version is accessible at http://labcirs.charite.de (sign in as “reporter”). The LabCIRS “reviewer,” who could be a principal investigator (PI), a lab manager, or any other person with the skills to initially assess reports, is alerted to incoming reports via email. The reviewer assigns a risk category (low to high), determines responsibilities, initiates subsequent measures, and decides who is responsible for their implementation (see S3 Fig, sign in as “reviewer” in the demo version). All reported incidents are analyzed in a regular monthly quality assurance conference. Depending on the nature of the reported incident, additional expert members of the department may be invited to join the discussion. Time critical events are processed immediately. Agreement is reached on specific prevention measures, responsibilities, and an action plan. Relevant preventive measures are communicated to all members of the department at the weekly morning conferences, and a monthly email to all staff members summarizes events and countermeasures. These messages are accessible to everyone and are permanently archived in the LabCIRS. Errors and incidents are communicated anonymously, unless the incident report includes the name of the reporting person and consent to reveal his or her identity. Typically, the examples of incidents reported via the LabCIRS include injuries when working with a sharp object, mistaken labeling of solutions and chemicals, mix-ups in the randomization of experimental animals, and data loss due to instrument write failures. Box 1. Features of a Laboratory CIR System for Experimental Biomedicine Easy to set up, run, and administer Easy to use, accessible, intuitive, and unambiguous Should be scalable so that it works in small single-investigator groups as well as in a large institute Allows anonymous reporting Allows free expression of “what actually happened”: the reporter’s own version of events Reports must be handled in a nonpunitive manner The incidents reported can be regularly analyzed by experts Learning points from such analyses need to be fed back promptly to those who need to know The reports are visible, and a clear path of action is communicated Feedback results in enhanced learning regarding the incident’s cause and systemic changes that will prevent its recurrence Emergence of an Error Culture Motivated by the exceedingly high attrition rate of bench to bedside translation in the stroke research field, we began to establish a structured quality management (QM) system in our experimental laboratories in 2012. The aim of our QM system is to implement auditable standards for the planning, realization, evaluation, and publication of our experimental studies and to safeguard compliance to guidelines (such as [20]) and institutional rules and regulations of good scientific practice (GSP). We realized that for academic preclinical research there is a paucity, if not to say a lack, of systematic approaches to improve and maintain quality. We had to therefore design from scratch, implement, and refine effective and transparent procedures for quality control in experimental neuroscience research in the university setting. To the best of our knowledge, our QM system represents one of the first attempts to implement systematic QM in academic preclinical research in Germany and possibly worldwide. Since the ISO 9001:2008 norm [14], which we chose for our QM, requires the implementation of measures for identifying errors, we posted printed “error reporting sheets” in all our laboratories. Disappointingly, only a few errors were reported via this mode. Through discussions with scientists, students, and technicians, we realized that the main reason for the failure of this error reporting system was that it did not safeguard anonymous reporting; potential reporters feared punitive action. We therefore established the web-based system described here, which includes additional benefits, such as accessibility on every computer in the lab, uploading of photographs to describe the incident, automatic alerts of new reports to personnel responsible, and archiving. As most academic research laboratories operate on a frugal budget, and funding organizations might be reluctant to cover costs beyond specific research projects, it is important to know how resource intense the operation of LabCIRS is. In our department when an error is reported, the reviewer categorizes the incident and decides whether acute measures are necessary. If the reporter agreed to it, the report is communicated to the department. Error reports are analyzed in detail at monthly meetings by a group consisting of various users and experts (scientists, doctoral students, postdocs, and technicians) of the research groups. The group also considers preventive measures against recurrence of the error. Relevant errors and countermeasures are then communicated to all members of the department in a weekly joint lab meeting. Additionally, all error reports are sent out monthly to all members of the department via email. All in all, analyzing, discussing, and communicating one error report takes about 40 minutes. LabCIRS was immediately accepted by all members of the laboratory. Since its inception, approximately one to two incidents are reported per month (Fig 2). Interestingly, in the beginning about half the reported incidents were not only anonymous but also strictly confidential (i.e., the reporters ticked the option “I DO NOT agree that this report will be made public to people outside the quality management team even after copyedit.”). For more than a year now, all reporters tick the option “I agree that this report will be made public to people outside the quality management team after copyedit.” We interpret this as a sign of trust and an indicator of a mature error culture. Clearly, most members of the department have realized that while in the complex setting of the biomedical laboratory, errors and incidents may occur, they can be prevented in future through reflection on what happened and through the input of colleagues. All reported errors have led to actions and preventive measures. These include modifications of briefings, instructions, and responsibilities, changes in the way samples and chemicals are labelled, and modifications of standard operating procedures, among many other provisions. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Errors reported per quarter since system was initiated (I/2014) until June 2016. Grey: number of errors reported publicly; orange: number of errors reported confidentially, i.e., without allowing the report to be made public. https://doi.org/10.1371/journal.pbio.2000705.g002 Anonymous reporting is a key feature of any CIRS. LabCIRS does not collect any personal information from the reporter, as such information in a relatively small group of approximately 100 people could potentially reveal the identity of the reporter. In order to nevertheless address the question of whether the use of LabCIRS differs between professions, we conducted an anonymous online survey asking two questions: (1) Do you actively and/or passively use LabCIRS (i.e., have you reported errors, or do you only read about errors in LabCIRS)? and (2) What is your job profile or status within the lab? About half of the responding LabCIRS users stated that they actively report incidents, while the other half uses the CIRS to stay informed about reported errors and countermeasures. The largest groups of the active reporters are either technicians or lab managers, while students, postdocs, technicians, and group leaders are represented almost equally among the passive users (Fig 3). It is not surprising that the majority of active reporters belong to the groups which focus on practical laboratory work, but our survey demonstrates that members of all professions use LabCIRS either actively or at least passively. The only exception is undergraduate students, who do not work unsupervised in our institute and are guided directly by members of other professions. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Results of an anonymous survey to explore which professions and status groups use LabCIRS in an active or passive manner. https://doi.org/10.1371/journal.pbio.2000705.g003 We are convinced that LabCIRS has clearly improved the quality of our work and made the laboratory a safer and more communicative environment. Although desirable, it is unfortunately hard to quantify the effectiveness of such a measure. The use of a CIRS cannot be tested in a controlled experiment (one lab with and the other without CIRS). CIRS use is voluntary and anonymous, and reported incidents do not necessarily represent all incidents that may have happened. It should be noted that in many domains the use of CIRS is plausible and by now the legal standard (aviation, nuclear power plants, etc.), but rarely, if at all, has its efficiency been unequivocally proven in a controlled setting. Therefore, the efficacy of CIRS must often remain anecdotal, much like the notion that the Chernobyl disaster could have been prevented by critical incidence reporting [21]. Initial concerns by some members of the department that its implementation might lead to a “surveillance culture” that would stifle creativity turned out to be totally unfounded. Nevertheless, it needs to be acknowledged that reporting mistakes, mishaps, and errors is a sensitive issue in any work environment. Setting up a critical incidence reporting system must rely on an intense communication among all members of the lab about its purpose and nonpunitive nature. In addition, reporting of an incident is only the beginning of a sequence of events that include the search for remedies and preventive measures. This only works in a collaborative and quality-oriented environment. However, the administrative effort to maintain such a system is minimal and likely compensated by the savings made through error prevention and improved quality. We highly recommend the establishment of a systematic way of learning from errors and mistakes, whether in small single-investigator groups of a few researchers, students, and technicians, or in large research institutions with staffs of several hundred professionals. This practice will benefit the emergence of an error culture that will likely enhance the overall quality and safety of research. The open-source LabCIRS we provide here can help to start this process, but it needs to be stressed that the system lives with those who report, discuss, and disseminate the incidents and countermeasures. Supporting Information S1 Fig. Screenshot of LabCIRS login. https://doi.org/10.1371/journal.pbio.2000705.s001 (TIF) S2 Fig. Screenshot of LabCIRS incident reporting page. https://doi.org/10.1371/journal.pbio.2000705.s002 (TIF) S3 Fig. Screenshot of LabCIRS incident reviewer login. https://doi.org/10.1371/journal.pbio.2000705.s003 (TIF) S1 Text. Demo version and source code. https://doi.org/10.1371/journal.pbio.2000705.s004 (DOCX) Acknowledgments We thank Dr. Nikolas Offenhauser for his contributions in the early stages of this project.

doi: 10.1371/journal.pbio.2000016pmid: 27977688

Introduction All mammalian telomeres (the ends of linear chromosomes) are composed of large tracts of repeated 5ʹ-TTAGGG sequences. Telomeres are well-conserved DNA end structures from yeast to mammals, and it is believed that the primary role of telomeres, in combination with shelterin proteins, is to provide protection of the linear chromosome ends from being recognized as damaged or broken DNA [1] and to facilitate the completion of DNA replication each cell cycle. Telomeres prevent DNA end-joining, DNA recombination, and loss of essential genetic information during DNA replication. Telomeres are maintained by many essential genes, including the six-component shelterin (TRF1, TRF2, POT1, TIN2, RAP1, and TPP1) and the CST (CTC1-STN1-TEN1) complexes [1,2]. Impairment of these genes is closely associated with age-related clinical pathology and defects in germ cell and stem cell maintenance [3–5]. It is well established that hTERT, the catalytic core reverse transcriptase component, protein levels are rate-limiting for telomerase activity and telomere length homeostasis [6]. Human embryonic stem cells and transit amplifying adult progenitor stem-like cells express hTERT and have active/functional telomerase that can fully or partially maintain telomeres during the substantial number of cell divisions required in fetal development [7]. While telomerase is present from the blastocyst stage in early human development, at approximately 16–18 wk of gestation, telomerase activity is silenced in the vast majority of somatic cells [8]. The molecular mechanisms (i.e., transcriptional regulation, alternative splicing changes, epigenetic modifications, or other regulatory processes) that trigger the silencing of telomerase at specific times during human development remain uncertain. Irrespective, telomerase largely remains silent throughout adult life except for tumor development. In ~90% of human tumors, telomerase is upregulated or reactivated for the maintenance of telomeres during the numerous rounds of cell divisions required for the emergence of malignant and metastatic disease [9]. Thus, tight regulation of telomerase and progressive telomere shortening are thought to be an initial barrier to the early onset of cancer. High resolution mapping of the three-dimensional chromatin interactome addresses many unanswered questions about the cis-regulatory long-range looping interactions important in gene regulation. The human genome is composed of continuous chromosome loops and TADs (topologically associating domains), forming gene territories [10,11]. Distal enhancers and/or insulators are believed to be responsible for the regulation of genes along the genome via chromatin folding. Recently, we demonstrated that telomeres also loop to specific loci to regulate gene expression, which we have termed TPE-OLD (telomere position effect—over long distance) [12–14]. In the examples characterized so far, genes close to telomeres are silenced in young cells (with long telomeres) and become expressed when telomeres are short. Importantly, re-elongation of cells with short telomeres by exogenous expression of the hTERT gene (active telomerase) results in expression patterns that mirror the expression of these genes in cells with long telomeres [12–15]. As we have observed genes between the TPE-OLD regulated genes that are not regulated by TPE-OLD, this mechanism is clearly distinct from classic TPE, which regulates genes proportional to the proximity to the telomeric repeats [15]. In the present study, we show that the expression of the hTERT gene itself is also regulated by TPE-OLD. The ability to regulate genes by telomere length without induction of a DNA damage signal from a single or a few critically short telomeres has potential explanatory value for what regulates the maximum length of human telomeres during fetal development and ways to regulate major age-associated transitions as well as to activate or repress genes as part of normal aging without requiring a DNA damage signal. Results Conserved TERT Loci in Higher Primates Long-ranged genomic interactions between telomeres and distal loci may play important roles in the regulation of gene expression, a phenomenon that we previously referred to as TPE-OLD [12,13]. Through previous microarray analyses [12], we identified the human CLPTM1L (cleft lip and palate-associated transmembrane protein 1-like) gene that is ~1.3 mega bases apart from the chromosome 5p telomere as a putative TPE-OLD candidate gene. CLPTM1L is frequently upregulated in cancer cells [16] and shows preserved colocalization with the TERT locus for a shared synteny in many species (Fig 1A). We analyzed mRNA expression of the genes at this locus, including CLPTM1L and hTERT, in BJ human fibroblast clones with long and short telomeres, to determine if the expression of this locus is regulated by TPE-OLD. CLPTM1L was expressed in normal young passaged cells but showed increased gene expression with progressive telomere shortening (S1A Fig). Historically, it is generally believed that hTERT is not actively transcribed in normal telomerase silent cells; however, expression of hTERT splice variants does occur [17]. The reason for this misconception is that most investigators use primer pairs designed to measure transcripts containing only the RT domain of TERT (exons 5–10), while exons outside of the RT domain are not measured (i.e., exons 1–4 and 11–16). It is now known that hTERT transcripts can be detected in a variety of telomerase-negative cells and tissues, but the mRNA produced is not full-length mRNA capable of producing active telomerase [17]. To test if replicative age or telomere length influenced hTERT expression, we measured hTERT gene expression using a primer pair targeting the 5ʹUTR to exon 1 of hTERT. We observed that hTERT is expressed at higher levels in two human fibroblast strains with short telomeres compared to the same cells with long telomeres (Fig 1B, S1 Fig). As previously described, we did not detect any transcripts that contain the RT domain of hTERT (Fig 1B); thus, transcripts that could code for active telomerase were not observed. We also analyzed protein expression of CLPTM1L (S1B Fig) and observed that the expression of CLPTM1L protein significantly increased during progressive telomere shortening, but the expression was greatly decreased when we re-introduced hTERT in old BJ cells and re-elongated telomeres (S1B Fig). We also examined mRNA expression of genes located between the 5p telomere and the hTERT-CLPTM1L locus (S1A Fig) in young and old BJ cells. The expression of the intermediate genes on chromosome 5p showed no significant increase in BJ cells with short telomeres (S1A Fig). We explored if telomere repeat containing RNA, TERRA, was also altered and potentially important in TPE-OLD. Consistent with previous reports [18,19] we observed an increase in TERRA expression from three subsets of chromosomes (1q-21q, 5p, and 9p-15q-Xq-Yq; Fig 1C) when telomeres were short compared to long. The TERRA data support our observations that the chromatin environment surrounding chromosome 5p and hTERT change when telomeres are short. Overall, this implies that the hTERT locus may be influenced by the length of telomeres through long-ranged chromatin interactions. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Conserved TERT loci in higher primates. (A) Conserved synteny map of TERT and CLPTM1L loci in mammals. Transcription orientation of TERT and location of telomere are shown in figure. (B) ddPCR analyses of the hTERT locus. mRNA expression in BJ cell lines at young population doublings (PD34) and old (PD74), BJ cell clones containing different telomere lengths, and IMR90 young (PD22) and old (PD50) were analyzed. RNA (1000ng) was reverse-transcribed, diluted, and 5ʹ hydrolysis probes (Roche UPL) were used to assess the number of mRNA molecules per reaction. Mean telomere length was analyzed by TRF in each of the analyzed cell lines. (C) Chromosome end-specific TERRA expression analysis was performed on BJ cell clones with long and short telomeres (same RNA as used above for TERT analysis). (D) Higher primates also retain the location of the TERT gene at the end of their chromosomes. Each bar represents an individual chromosome retaining the TERT locus. Location of the TERT gene is marked by green on the chromosome. Red bar represents location of telomeres. * = p < 0.05. kb = kilobases. Data are presented as means and standard errors of biological replicates and technical triplicates. Data associated with this figure can be found in the supplemental data file (S1 Data). https://doi.org/10.1371/journal.pbio.2000016.g001 Perhaps not surprising, but potentially significant, is that the location of the TERT gene is also evolutionarily conserved (Fig 1D). TERT genes are located at the very end of their chromosomes, near the telomere, in higher primates including humans and most other large long-lived mammals. However, the location of the TERT gene in rodents and many other smaller shorter-lived mammals is non-telomeric. The local genome structure around the TERT locus in rodents is quite different from primates, implying they may have developed different strategies to regulate telomerase expression [20,21]. Based on these observations, we decided to test if there is a functional role for TERT being localized at the end of human chromosome 5p. As the distance between the hTERT locus and the telomere is only ~1.3 mega bases, we postulated that hTERT might also be regulated in part by a long-ranged telomere looping mechanism in human cells. Three-Dimensional Interactions between the hTERT Locus and the Sub-telomeric 5p Region by Telomere Length We designed two specific BAC probes to visualize the hTERT locus and the sub-telomeric 5p region for three-dimensional fluorescence in situ hybridization (3D-FISH) (Fig 2A). We measured the distance between the hTERT locus and the sub-telomeric 5p region, and the pairs of alleles were divided into adjacent to (A) or separated (S) by the three-dimensional location (S2A and S2B Fig). We first stained the sub-telomeric BAC region, the hTERT locus and telomeres in old BJ cells, with short telomeres (Fig 2B). The telomere staining was detected at the hTERT locus with sub-telomere 5p in the adjacent allele pair. However, we observed at least one hTERT allele that was spatially separated from the sub-telomere 5p probe in old BJ cells without telomere staining. We measured the distance between the hTERT locus and the closest telomere (Fig 2C). The results showed that the hTERT locus colocalized with the telomere when it is adjacent to the sub-telomeric 5p region (Fig 2B and 2C). This implies that the telomere is likely to be adjacent to the hTERT locus for potential long-ranged looping interactions. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Three-dimensional interactions between the hTERT locus and the sub-telomeric 5p region by telomere length. (A) Schematic map of the human 5p chromosome containing the hTERT locus and design of probes. The sub-telomeric 5p probe (far, red, grey color) stains the end of chromosome 5p that is located 75 Kb from the 5p telomeric repeats (red color). The hTERT probe (green) stains a specific genomic region containing the hTERT gene, which is 1.1 Mb from the 5p telomere. (B) BJ human fibroblasts at PD60 was stained with indicated fluorescent probes against the hTERT locus, sub-telomeric region 5p, and the telomere. A representative deconvolved image was selected. Scale bar represents 5 uM (C) Graph shows the distance between the hTERT locus and the closest telomere depends on the distance between the hTERT locus and the sub-telomeric 5p. (D) Percentage of adjacent allele (A) pairs versus separated allele (S) pairs was determined by 3D-FISH in normal BJ cells. BJ fibroblasts at different PDs were analyzed as indicated in the Materials and Methods. Cas9-mediated transient perturbation at the sub-telomeric 5p region was performed in BJ cells at PD25. (E) Percentage of adjacent allele pairs (A) versus separated allele pairs (S) was determined by 3D-FISH in BJ fibroblasts. BJ clones were transfected with a floxable hTERT, followed by excision with Cre recombinase at different time points. Two clones with different lengths of telomeres were analyzed at the same number of population doublings in culture. Indicated telomere lengths were measured by TRF (terminal restriction fragment) Southern blot shown in Fig 1C. (F) Telomere-DNA damage induced foci (TIF) analysis performed in BJ long (13 kb) and short cells (9 kb) (TRF shown in Fig 1C for matched cells). Cells were stained with telomere PNA and gamma-H2AX antibody. At least 81 nuclei were quantified. Scale bar is 5 μM. Western blotting shows lack of induction of γ-H2AX in BJ cells with different telomere lengths. β-Actin was used as a loading control. (G) 3C analysis shows distal genomic interactions between the 5p telomere and hTERT. 3C libraries were generated from BJ cells at PD20 and PD70, followed by ddPCR amplification of genomic interactions between indicated regions. Proximity control amplification of 3C libraries from BJ cells at PD20 and PD70 was performed with ddPCR. t test revealed a significant effect. *p < 0.05, n.s. = no significant. Data associated with this figure can be found in the supplemental data file (S1 Data). https://doi.org/10.1371/journal.pbio.2000016.g002 We next tested if telomere looping close to the hTERT locus changes when telomeres became short. We measured and compared the distance between the hTERT locus and sub-telomeric 5p in young BJ fibroblasts at 20 population doublings (PDs) with long telomeres versus old BJ fibroblast at PD90 with short telomeres (Fig 2D). More than 70% of allele pairs were adjacent in BJ cells at PD20, implying that the telomeric heterochromatin might affect the expression of the hTERT locus in young BJ fibroblasts. BJ cells are telomerase-negative, but non-catalytic alternatively spliced variants are expressed, as shown in Fig 1 and as previously described [17]. This might explain why a small proportion of alleles is separated from the telomere in telomerase-negative young BJ cells with long telomeres, based on the assumption that the looping interactions suppress transcription. In old BJ cells at PD90, we found that the percentage of adjacent allele pairs was significantly reduced. Almost 60% of alleles were separated in the old cells with short telomeres, indicating that there is at least one hTERT locus spatially separated from the telomere in each cell. Importantly, we confirmed these 5p/TERT looping interactions in a second fibroblast cell strain, IMR90 (S2C and S2D Fig). We measured the number of separated and adjacent alleles in IMR90 cells young (PD 22) and old (PD 52) and show a shift from the majority of alleles being adjacent (76%) in young cells compared to the majority of alleles being separated (88%) in old cells. The looping data and the expression of hTERT are consistent. We suggest that old cells (with short telomeres) lose one control mechanism in regulating the hTERT locus (i.e., telomere chromatin looping) that helps repress the expression of hTERT. However, while we observed increased transcription of exon 1 of hTERT, there must be additional mechanisms preventing the inclusion of exons critical to produce active telomerase. There is substantial evidence that alternative splicing of hTERT may also have a major role in suppressing the production of active telomerase in old cells [22–24]. Furthermore, we performed 3D-FISH analysis in transformed SW26 and SW39 cells. SW cells are SV40 antigen expressing clones of IMR90 cells that have spontaneously immortalized using either telomerase (SW39) or an alternative lengthening of telomeres (ALT; SW26) mechanism to maintain telomeres (S2E Fig). In both cell lines, the majority of the alleles were separated (SW39 = 72%; SW26 = 66%), indicating that short telomeres due to replicative aging are likely responsible for the change in chromatin conformation and that a secondary change occurs to cause the production of full-length TERT or engage ALT. It has been suggested that hTERT shows mono-allelic expression in cancer, which is sufficient to preserve constant telomere length [25,26]. Our results support this assumption, as we observed that, on average, only one hTERT allele was generally in the open configuration during in vitro aging well before the onset of cancer. As controls for global conformational changes at chromosome 5p, we performed two additional FISH experiments. In the first experiment, we stained intermediate genomic region between the hTERT locus and the 5p telomere (S3A–S3C Fig). In addition, we also stained cells for two loci located 25.5 MB and 30.6 MB away from hTERT (S3D–S3F Fig). There were no changes in distances between the control loci in young and old cells, demonstrating that the conformation change occurs at the specific genomic region encompassing hTERT during in vitro aging, and this change is not due to classic TPE. To determine if we could artificially shorten telomeres and recapitulate the aging phenotype, we utilized CRISPR/Cas9 (clustered regularly interspaced short palindromic repeat-associated 9) to remove a large portion of the telomere and subtelomere region from chromosome 5p. This experiment allows testing the role of chromosome 5p’s telomere in regulating the looping observed in cells with short and long telomeres. As illustrated in Fig 2D, we also infected young BJ cells with a lentivirus expressing Cas9 protein and single guide-RNA targeting the sub-telomeric region of 5p to specifically disturb telomeres at chromosome 5p for a short period of time [27]. We also added an NHEJ inhibitor, SCR7, simultaneously during the infection to suppress repair of the double strand breaks induced by the Cas9 protein [28]. The targeted cells showed an unstable end structure of chromosome 5p (S5 Fig), and the specific disturbance of the 5p telomere significantly diminished telomere looping at the end of the chromosome 5p. We further examined if the proposed mechanism was present in BJ cell clones in which both young and old cells were passaged the same amount of time in culture. This approach was necessary to eliminate the possibility that young and old cells that were in culture for vastly differing times could introduce artifacts. To accomplish this, we expressed a floxable hTERT in BJ clones, followed by excision at different time points in order to make isogenic cells with different length of telomeres but passaged similar times in cell culture [12,29]. Telomere length of the early-excision clone was 9 kb, and this was extended up to 13 kb in the late-excision clone. The telomere length (terminal restriction fragment [TRF]) results are presented in Figure 1B. Population doublings were evenly matched between clones (to avoid confounding effects of passage of time in culture), and we also analyzed telomere looping. Similar to our observations in normally passaged BJ cells, the isogenic clones also showed decreased levels of telomere looping with telomere shortening (Fig 2E). Importantly, there were only background levels of DNA damage signaling during telomere shortening (Fig 2F) indicating that the change in genome structure occurred before initiation of DNA damage responses from critically short telomeres. To ensure that our staining protocol was robust, we induced DNA damage (double strand breaks) by treating long and short telomere BJ cells with zeocin and assaying for DNA damage (S2F Fig). These data can be interpreted to indicate that our staining protocol is robust and that we are analyzing cells before telomere-DNA damage induced foci are present or significant DNA damage occurs in the cells. We next performed droplet digital 3C (chromatin conformation capture) to detect the genomic interactions between the 5p telomere and the hTERT locus in young and old BJ cells (Fig 2G, left side). The results showed that the hTERT locus has specific genomic interactions with the 5p telomere, and the interaction was reduced during in vitro aging and telomere shortening. A proximity control primer which is 10kb away from the fixed primer at the hTERT locus was selected for normalization of 3C results (Fig 2G, right side). Taken together, telomere looping exists between the hTERT locus and the sub-telomeric 5p in normal human cells, and this looping is greatly reduced by gradual telomere shortening. Telomere Looping Determines Permissiveness of the hTERT Locus It has been shown that cis-elements upstream of the hTERT locus may play important roles in the tight regulation of human telomerase [30]. Thus, we decided to test if telomere looping could affect the epigenetic status of the hTERT proximal promoter region. We first analyzed DNA methylation of the region from -720bp to +90bp of the hTERT promoter in isogenic BJ cells with different lengths of telomeres but similar times in cell culture (Fig 3A). The relationship between DNA methylation and transcription in the hTERT promoter remains controversial in normal and cancer cells [31,32], but the transcription start site of hTERT retains little or no methylation in telomerase-active cancer cells for active transcription [33]. We found that the level of DNA methylation is significantly higher in BJ cells with long telomeres at several regions associated with hTERT and the hTERT region in comparison to cells with shorter telomeres. The largest differences were observed at -580bp, -250bp, -30bp, and +20bp of the hTERT promoter, including the E-box motif (a putative Myc binding sequence). It has also been reported that the proximal region of the hTERT promoter, including exon 1 and 2, regulates the activity of the hTERT promoter and that the methylation of this region is responsible for binding of several proteins [34,35]. Therefore, our results can be interpreted to indicate that telomere length-associated changes in methylation levels of the hTERT proximal promoter might affect transcriptional regulation of this locus. We next analyzed active and inactive histone marks on the hTERT proximal promoter using chromatin immunoprecipitation combined with droplet digital polymerase chain reaction (ChIP-ddPCR; [12]) (Fig 3B). We measured two histone marks associated with active chromatin H3K4 trimethylation (H3K4me3) and H3K9 acetylation (H3K4ac) and two histone marks associated with repressed chromatin H3K27 trimethylation (H3K27me3) and H3K9 trimethylation (H3K9me3), which have key roles in regulating gene expression [36]. We observed an increase in both H3K4me3 and H3K9ac across the TERT promoter in aged cells with short telomeres (Fig 3B). We also observed an increase in the repressive histone mark H3K27me3, but did not observe any significant differences in young or old BJ cells for the repressive histone mark H3K9me3. Collectively, this shows that the chromatin status of the hTERT promoter in old BJ cells with short telomeres is different and may be more transcriptionally permissive compared to young BJ cells with long telomeres. These data correlate well with the increased hTERT transcription we observed in cells with short telomeres. Furthermore, we analyzed chromatin at the promoters of three genes surrounding TERT that could also be affected by the altered chromatin environment with aging. We analyzed the proximal promoter regions of CLPTM1L, SCL6A18, and SCL6A19 for the same histone marks described above in the same cells and preparations used for TERT ChIP. At the CLPTM1L promoter we observed significant increases in histone marks indicating active transcription (Fig 3B). These data correlate well with an increase in CLPTM1L transcripts and protein levels (S1 Fig). We also observed significant changes in the chromatin surrounding the solute/amino acid transporter genes (SCL6A18 and SCL6A19), even though these genes are not expressed above basal/background levels in old/short telomere BJ cells. Specifically, we observed that both the repressive histone marks were increased in old cells (short telomeres) compared to young cells (long telomere). However, there was an increase in the activation marks as well. This indicates an intricate balance between chromatin modifications, methylation status, telomere length, and the expression of tissue-specific transcription and splicing factors that dictates the activation or repression of genes with replicative aging (telomere shortening—TPE-OLD). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Chromatin looping and epigenetic modifications (i.e., histone modifications and DNA methylation) determine permissiveness of the hTERT locus. (A) Bisulfite DNA methylation sequencing analysis of the hTERT proximal promoter region from -720bp to +90bp. Genomic DNA of BJ clones with different lengths of telomeres was modified and PCR-amplified. Each amplicon was TA-cloned for bacterial amplification and sequenced. Percentage of CpG methylation of the hTERT promoter is indicated in two BJ cell clones with different lengths of telomeres. (B) Illustration of genomic locus containing TERT. Black arrows indicate approximate location of primers in the promoters of the indicated genes. Chromatin immunoprecipitation was performed with BJ cells at PD34 and PD74. Six antibodies against H3K4me3, H3K9ac, H3K9me3, H3K27me3, H3 total, and IgG were used to pull down chromatin extracts, and the promoter regions of hTERT, SLC6A18, SLC6A19 and CLPTM1L were analyzed by ddPCR. Data are presented as means and standard errors of biological and technical duplicates. Student’s paired t tests comparing young and old determined significance (* = p < 0.05). (C) BJ cells at different PDs were infected with shRNAs against p21 (CDKN1A) and selected using puromycin to generate stable knockdown clones. hTERT mRNA expression was analyzed by ddPCR in sh-p21 cells and controls. The number of full-length and total hTERT mRNAs was assessed by amplifying the hTERT exon 7/8 junction and exon 15/16 junction, respectively. Knockdown efficiency of p21 was determined by western blotting. β-actin was used as a loading control. ANOVA revealed a significant effect. *p < 0.001. Data associated with this figure can be found in the supplemental data file (S1 Data). https://doi.org/10.1371/journal.pbio.2000016.g003 While we demonstrated that telomere shortening induced conformation changes between the hTERT locus and the sub-telomeric 5p resulting in up regulation of exon 1, presumably containing spliced hTERT transcripts in normal BJ cells (see Fig 1B), it did not result in full-length telomerase activity competent transcripts. Thus, we suggest that telomere shortening may render the hTERT locus more permissive and under oncogenic stress may lead to the production of full-length hTERT mRNA transcripts that could in turn produce telomerase activity. To test this, we simulated a step in spontaneous cancer transformation by knocking down p21 (CDKN1A) and analyzing mRNA expression level of hTERT (Fig 3C). The knockdown of p21 was previously shown to de-repress hTERT expression [37]. Thus, we tested if the knockdown of p21 would increase the expression of hTERT mRNAs and result in the inclusion of exons 7/8 in the short-telomere old BJ cells but not in the young BJ cells with long telomeres. We measured the expression level of hTERT transcripts in young and old BJ cells with and without p21 stable knockdown; mRNA containing exons 7/8 (exons coding for critical residues in the reverse transcriptase domain of TERT) and exon 15/16 (most splice variants of hTERT contain exons 15 and 16), responsible for putative active hTERT and total hTERT variants respectively. Both the active and the total hTERT transcript variants significantly increased with the knockdown of p21 in old BJ but not in young BJ cells; however, we did not detect telomerase activity (S6 Fig). While we observed an increased portion of transcripts that contain exons 7/8 of the TERT RT domain, other critical regions such as exon 2 may be spliced out [38]. Further work into the regulation of hTERT splicing is necessary to more fully understand the complex regulatory network surrounding hTERT and why the majority of transcripts are inactive splice variants as opposed to full length. While this result does not prove a causal role during cancer development, this series of experiments does demonstrate that telomere shortening in cells that bypass replicative senescence leads to the hTERT locus entering into a more permissive state (e.g., increased hTERT mRNA expression) in the presence of oncogenic stresses, consistent with the disengagement of telomere looping. TRF2 as a Mediator of Telomere Looping between the hTERT and Sub-telomeric 5p in Cells with Long Telomeres Characterization of cis- or trans-acting factors responsible for telomere looping will be important to understand this novel mechanism for telomerase regulation. A recent report showed that TRF2 (telomeric repeat-binding factor 2) protein is essential for the functional organization of chromosome ends, including human fibroblasts [39,40]. There is also mounting evidence for off-telomere functions of the shelterin components [41]. While a recent whole genome sequencing study found 2,920 interstitial TTAGGG repeats throughout the human genome [39], we also found frequent internal (interstitial) telomeric sequences (ITS) near the TERT locus in higher primates but not in rodent cells (Fig 4A). Thus, we first checked for a putative role of TRF2 in telomere looping in BJ cells as a candidate approach. We knocked down TRF2 by siRNA and performed 3C to directly assess the genomic interactions between the telomere and the hTERT locus (Fig 4B). The knockdown of TRF2 significantly reduced the genomic interactions between the telomere and the hTERT locus in young PD30 BJ cells, implying TRF2 may have a role in telomere looping interaction on hTERT locus. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. TRF2 as a mediator of telomere looping between the hTERT and sub-telomeric 5p in cells with long telomeres. (A) The number of (TTAGGG)2 repeats was analyzed on a small series of mammalian genomes. The TERT locus flanked by 350kb was examined. The (TTAGGG)2 repeats are depicted as a single black box. Multiple (TTAGGG)2 repeats at a locus are depicted by stacked black boxes. Transcription orientation of TERT and location of telomere are shown in figure. (B) 3C analyses show distal genomic interactions between the 5p telomere and hTERT. 3C libraries were generated from BJ cells treated with siRNA against TRF2, followed by ddPCR amplification to determine genomic interactions. t test revealed a significant effect. *p < 0.05, **p < 0.01 (C) ChIP shows enrichment of TRF2 protein on hTERT-ITS. The 1.153 Mbp to 1.154 Mbp region from the end of the chromosome 5p was examined. Chromatin extracts from BJ cell at PD25 and PD50 were prepared and pulled down with an antibody against TRF2. TTAGGG repeats are depicted as a red box. (D) A schematic map of the human chromosome 5 at the 1,153,167 bp to 1,353,167 bp region from the p arm terminus. Arrows indicate location of HindIII restriction enzyme sites. Red boxes indicate location of TTAGGGTTAGGG repeats along the genome. (E) 3C analysis shows distal genomic interactions between 5`end of hTERT locus and each HindIII site. 3C libraries were generated from BJ cells at PD25 and PD70, followed by ddPCR amplification to determine genomic interactions. t test revealed a significant effect. *p < 0.05, **p < 0.01. (F) ChIP shows enrichment of TRF2 protein on the hTERT promoter. Chromatin extracts from BJ cell at PD25 and PD50 were prepared and pulled down with an antibody against TRF2. The hTERT promoter region from -500bp to +200bp was analyzed by real-time-qPCR. t test revealed a significant effect. *p < 0.05, **p < 0.05, n.s. = no significance. (G) Model of genomic folding at hTERT locus based on 3C analyses. Blue boxes indicate the location of genes on chromosome 5p. Red boxes indicate the location of hTERT-ITS. Blue and red dashes indicate chromosome 5p and telomere of chromosome 5p, respectively. (H) BJ cells at PD 17 were transfected in individual experiments with a siRNA against TRF2, CTCF, and LDB1. Three days after transfection, cells were fixed for 3D-FISH analysis. Percentage of adjacent allele (A) pairs versus separated allele (S) pairs was determined by 3D-FISH. (I) Western blotting analysis shows knockdown efficiency of the siRNA against TRF2, CTCF, and LDB1. Histone H3 was used as a loading control. Data associated with this figure can be found in the supplemental data file (S1 Data). https://doi.org/10.1371/journal.pbio.2000016.g004 As shown in Fig 4A, a region 100 kb downstream of the hTERT (Chr5: 1,154,047–1,154,347) contains a series of internal telomeric sequences that may recruit TRF2 shelterin protein (hereafter termed hTERT-ITS). Thus, we reasoned that this region would be a putative binding site for TRF2 and may be responsible for the telomere looping interaction between the telomere and the hTERT locus in cells with long but not short telomeres. ChIP-qPCR analysis showed that the TRF2 protein associates with the hTERT-ITS region in young and old BJ cells as proposed (Fig 4C). We next performed 3C to further clarify that hTERT-ITS interact with the hTERT promoter by genome folding to affect transcriptional permissiveness as shown in Figs 1 and 3. Within 200kb, we found more than 20 HindIII restriction enzyme sites were in the hTERT/CLPTM1L locus (Fig 4D). Droplet digital PCR (ddPCR)-mediated amplification showed specific interactions between the 5ʹ end of hTERT and the hTERT-ITS (Fig 4E). Moreover, the interaction was weakened in old BJ cells, implying there might be a transition from a more repressive state to a more active state of this TAD location during in vitro aging, consistent with the increased hTERT mRNA, altered methylation, and chromatin. This result also shows that there is an additional genome folding between the hTERT locus and the hTERT-ITS at an intermediate region between the SLC6A18 and SLC6A19 loci. The hTERT promoter is not close to the hTERT-ITS on a linear genome map, but the unique genome folding at this region potentially positions the hTERT promoter close to the ITS, followed by putative TRF2-mediated telomere recruitment to the hTERT promoter only in cells with long telomeres. In Fig 4F, we demonstrate that TRF2 protein is also enriched in the hTERT promoter region using ChIP-qPCR approaches. While TRF2 protein was enriched at proximal regions on the hTERT promoter, the interaction was significantly decreased in old BJ cells at the genomic regions containing -350bp to -50bp of the hTERT promoter. This shows TRF2 protein can occupy the hTERT promoter region, but the interaction is weakened during in vitro aging and telomere shortening. Together, we interpret these experiments to indicate that TRF2, and perhaps upregulated TERRA, may have at least a partial mechanistic role in telomere looping at the hTERT locus through interaction with the conserved interstitial telomeric repeats. Because we have shown the interaction between the 5p telomere and the hTERT locus, we modeled one possibility for the detailed local genome structure of this locus (Fig 4G). In this model, the hTERT promoter is close to the hTERT-ITS by genome folding in young cells with long telomeres. In addition, this model shows that TRF2 protein is recruited to hTERT locus and hTERT-ITS, which makes this interaction potentially dependent on telomere length. In summary, the hTERT promoter has specific interactions with the hTERT-ITS through gene looping, which may also recruit telomere length-dependent looping (TPE-OLD) mechanisms through TRF2 protein. We next performed 3D-FISH to visualize the genomic structure changes between the hTERT locus and the sub-telomeric 5p (Fig 4H). Control PD17 BJ cells showed that 89% of the hTERT and sub-telomeric 5p allele pairs were adjacent, but knockdown of TRF2 reduced this down to 34%. We also knocked down CTCF (CCCTC-binding factor) and LDB1 (LIM domain-binding protein 1), which are proposed to be essential proteins in global gene looping maintenance [42,43]. CTCF and LDB1 knockdown also significantly reduced the adjacent allele pairs, implying that the general gene looping mechanisms may also be involved in telomere looping. Western blotting was also performed to show knockdown efficiency (Fig 4I). Taken together, TRF2, part of the shelterin complex, may be mechanistically involved in the establishment of telomere looping near the hTERT locus through ITS together with general chromosome looping mechanisms. Telomere Length Affects Expression of hTERT in Telomerase-Positive Cancer Cells In almost all primary human cancers, telomere length is very short compared to adjacent normal tissues [44]. It is likely that short telomeres, in combination with oncogenic alterations, result in the hTERT gene becoming more permissive for protein expression and enzyme activity. Thus, we next investigated how telomere length affects hTERT expression in telomerase-active cancer cells. We first infected hTERT and hTR (hTERC) into the SW39 cell line (SV40 immortalized human telomerase expressing fibroblasts) and analyzed mRNA expression of the endogenous hTERT by examining the 3ʹ untranslated region. We observed that the extended telomere length reduced endogenous expression of hTERT mRNA in qPCR analysis (Fig 5A) implying TPE-OLD remains engaged at least in this tumor cell line. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. Telomere length affects expression of hTERT in telomerase-positive cancer cells. (A) SW39 cells were infected with viruses for ectopic expression of hTERT and hTR (hTERC). RNA was purified, and the 3ʹUTR was amplified to analyze endogenous level of hTERT mRNA. ddPCR was performed to assess relative expression levels. t test revealed a significant effect. *p < 0.005 (B) HeLa cells were cloned and infected with a floxable hTERT. After treatment of adeno-cre recombinase at different time points, HeLa clones with long and short telomeres at same number of population doublings in cell culture were established. The number of full-length, total, and minus beta hTERT mRNAs were assessed by amplifying exon7/8 junction, exon 15/16 junction, and the exon 6/9 junction respectively. Telomere lengths are determined by TRF and indicated under the graph. t test revealed a significant effect between splice variants. *p < 0.001 (C) Percentage of adjacent allele (A) pairs versus separated allele (S) pairs was determined by 3D-FISH in HeLa clones. HeLa clones with long and short telomeres at the same PD in culture were analyzed. (D) Simplified model of how TPE-OLD regulates hTERT expression in human cells during aging and cancer progression. See text for details. Data associated with this figure can be found in the supplemental data file (S1 Data). https://doi.org/10.1371/journal.pbio.2000016.g005 We further established isogenic HeLa cell clones with different telomere lengths by excising a floxable hTERT cDNA at different time points. We examined expression of splice variants of hTERT mRNA containing total, full-length (indicative of telomerase activity), and minus beta alternative spliced forms through ddPCR analysis (Fig 5B). All three splice variants showed significantly decreased expression in the long-telomere HeLa clone. We also performed 3D-FISH to analyze changes in genomic structure between the hTERT locus and the sub-telomeric 5p after the extension of telomeres in HeLa cells (Fig 5C). The long-telomere HeLa clone showed a higher percentage of adjacent allele pairs compared with the short-telomere HeLa clone. This indicates that the expression of hTERT may also be influenced by the length of telomeres through TPE-OLD in telomerase-positive cancer cells. Conserved TERT Loci in Higher Primates Long-ranged genomic interactions between telomeres and distal loci may play important roles in the regulation of gene expression, a phenomenon that we previously referred to as TPE-OLD [12,13]. Through previous microarray analyses [12], we identified the human CLPTM1L (cleft lip and palate-associated transmembrane protein 1-like) gene that is ~1.3 mega bases apart from the chromosome 5p telomere as a putative TPE-OLD candidate gene. CLPTM1L is frequently upregulated in cancer cells [16] and shows preserved colocalization with the TERT locus for a shared synteny in many species (Fig 1A). We analyzed mRNA expression of the genes at this locus, including CLPTM1L and hTERT, in BJ human fibroblast clones with long and short telomeres, to determine if the expression of this locus is regulated by TPE-OLD. CLPTM1L was expressed in normal young passaged cells but showed increased gene expression with progressive telomere shortening (S1A Fig). Historically, it is generally believed that hTERT is not actively transcribed in normal telomerase silent cells; however, expression of hTERT splice variants does occur [17]. The reason for this misconception is that most investigators use primer pairs designed to measure transcripts containing only the RT domain of TERT (exons 5–10), while exons outside of the RT domain are not measured (i.e., exons 1–4 and 11–16). It is now known that hTERT transcripts can be detected in a variety of telomerase-negative cells and tissues, but the mRNA produced is not full-length mRNA capable of producing active telomerase [17]. To test if replicative age or telomere length influenced hTERT expression, we measured hTERT gene expression using a primer pair targeting the 5ʹUTR to exon 1 of hTERT. We observed that hTERT is expressed at higher levels in two human fibroblast strains with short telomeres compared to the same cells with long telomeres (Fig 1B, S1 Fig). As previously described, we did not detect any transcripts that contain the RT domain of hTERT (Fig 1B); thus, transcripts that could code for active telomerase were not observed. We also analyzed protein expression of CLPTM1L (S1B Fig) and observed that the expression of CLPTM1L protein significantly increased during progressive telomere shortening, but the expression was greatly decreased when we re-introduced hTERT in old BJ cells and re-elongated telomeres (S1B Fig). We also examined mRNA expression of genes located between the 5p telomere and the hTERT-CLPTM1L locus (S1A Fig) in young and old BJ cells. The expression of the intermediate genes on chromosome 5p showed no significant increase in BJ cells with short telomeres (S1A Fig). We explored if telomere repeat containing RNA, TERRA, was also altered and potentially important in TPE-OLD. Consistent with previous reports [18,19] we observed an increase in TERRA expression from three subsets of chromosomes (1q-21q, 5p, and 9p-15q-Xq-Yq; Fig 1C) when telomeres were short compared to long. The TERRA data support our observations that the chromatin environment surrounding chromosome 5p and hTERT change when telomeres are short. Overall, this implies that the hTERT locus may be influenced by the length of telomeres through long-ranged chromatin interactions. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Conserved TERT loci in higher primates. (A) Conserved synteny map of TERT and CLPTM1L loci in mammals. Transcription orientation of TERT and location of telomere are shown in figure. (B) ddPCR analyses of the hTERT locus. mRNA expression in BJ cell lines at young population doublings (PD34) and old (PD74), BJ cell clones containing different telomere lengths, and IMR90 young (PD22) and old (PD50) were analyzed. RNA (1000ng) was reverse-transcribed, diluted, and 5ʹ hydrolysis probes (Roche UPL) were used to assess the number of mRNA molecules per reaction. Mean telomere length was analyzed by TRF in each of the analyzed cell lines. (C) Chromosome end-specific TERRA expression analysis was performed on BJ cell clones with long and short telomeres (same RNA as used above for TERT analysis). (D) Higher primates also retain the location of the TERT gene at the end of their chromosomes. Each bar represents an individual chromosome retaining the TERT locus. Location of the TERT gene is marked by green on the chromosome. Red bar represents location of telomeres. * = p < 0.05. kb = kilobases. Data are presented as means and standard errors of biological replicates and technical triplicates. Data associated with this figure can be found in the supplemental data file (S1 Data). https://doi.org/10.1371/journal.pbio.2000016.g001 Perhaps not surprising, but potentially significant, is that the location of the TERT gene is also evolutionarily conserved (Fig 1D). TERT genes are located at the very end of their chromosomes, near the telomere, in higher primates including humans and most other large long-lived mammals. However, the location of the TERT gene in rodents and many other smaller shorter-lived mammals is non-telomeric. The local genome structure around the TERT locus in rodents is quite different from primates, implying they may have developed different strategies to regulate telomerase expression [20,21]. Based on these observations, we decided to test if there is a functional role for TERT being localized at the end of human chromosome 5p. As the distance between the hTERT locus and the telomere is only ~1.3 mega bases, we postulated that hTERT might also be regulated in part by a long-ranged telomere looping mechanism in human cells. Three-Dimensional Interactions between the hTERT Locus and the Sub-telomeric 5p Region by Telomere Length We designed two specific BAC probes to visualize the hTERT locus and the sub-telomeric 5p region for three-dimensional fluorescence in situ hybridization (3D-FISH) (Fig 2A). We measured the distance between the hTERT locus and the sub-telomeric 5p region, and the pairs of alleles were divided into adjacent to (A) or separated (S) by the three-dimensional location (S2A and S2B Fig). We first stained the sub-telomeric BAC region, the hTERT locus and telomeres in old BJ cells, with short telomeres (Fig 2B). The telomere staining was detected at the hTERT locus with sub-telomere 5p in the adjacent allele pair. However, we observed at least one hTERT allele that was spatially separated from the sub-telomere 5p probe in old BJ cells without telomere staining. We measured the distance between the hTERT locus and the closest telomere (Fig 2C). The results showed that the hTERT locus colocalized with the telomere when it is adjacent to the sub-telomeric 5p region (Fig 2B and 2C). This implies that the telomere is likely to be adjacent to the hTERT locus for potential long-ranged looping interactions. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Three-dimensional interactions between the hTERT locus and the sub-telomeric 5p region by telomere length. (A) Schematic map of the human 5p chromosome containing the hTERT locus and design of probes. The sub-telomeric 5p probe (far, red, grey color) stains the end of chromosome 5p that is located 75 Kb from the 5p telomeric repeats (red color). The hTERT probe (green) stains a specific genomic region containing the hTERT gene, which is 1.1 Mb from the 5p telomere. (B) BJ human fibroblasts at PD60 was stained with indicated fluorescent probes against the hTERT locus, sub-telomeric region 5p, and the telomere. A representative deconvolved image was selected. Scale bar represents 5 uM (C) Graph shows the distance between the hTERT locus and the closest telomere depends on the distance between the hTERT locus and the sub-telomeric 5p. (D) Percentage of adjacent allele (A) pairs versus separated allele (S) pairs was determined by 3D-FISH in normal BJ cells. BJ fibroblasts at different PDs were analyzed as indicated in the Materials and Methods. Cas9-mediated transient perturbation at the sub-telomeric 5p region was performed in BJ cells at PD25. (E) Percentage of adjacent allele pairs (A) versus separated allele pairs (S) was determined by 3D-FISH in BJ fibroblasts. BJ clones were transfected with a floxable hTERT, followed by excision with Cre recombinase at different time points. Two clones with different lengths of telomeres were analyzed at the same number of population doublings in culture. Indicated telomere lengths were measured by TRF (terminal restriction fragment) Southern blot shown in Fig 1C. (F) Telomere-DNA damage induced foci (TIF) analysis performed in BJ long (13 kb) and short cells (9 kb) (TRF shown in Fig 1C for matched cells). Cells were stained with telomere PNA and gamma-H2AX antibody. At least 81 nuclei were quantified. Scale bar is 5 μM. Western blotting shows lack of induction of γ-H2AX in BJ cells with different telomere lengths. β-Actin was used as a loading control. (G) 3C analysis shows distal genomic interactions between the 5p telomere and hTERT. 3C libraries were generated from BJ cells at PD20 and PD70, followed by ddPCR amplification of genomic interactions between indicated regions. Proximity control amplification of 3C libraries from BJ cells at PD20 and PD70 was performed with ddPCR. t test revealed a significant effect. *p < 0.05, n.s. = no significant. Data associated with this figure can be found in the supplemental data file (S1 Data). https://doi.org/10.1371/journal.pbio.2000016.g002 We next tested if telomere looping close to the hTERT locus changes when telomeres became short. We measured and compared the distance between the hTERT locus and sub-telomeric 5p in young BJ fibroblasts at 20 population doublings (PDs) with long telomeres versus old BJ fibroblast at PD90 with short telomeres (Fig 2D). More than 70% of allele pairs were adjacent in BJ cells at PD20, implying that the telomeric heterochromatin might affect the expression of the hTERT locus in young BJ fibroblasts. BJ cells are telomerase-negative, but non-catalytic alternatively spliced variants are expressed, as shown in Fig 1 and as previously described [17]. This might explain why a small proportion of alleles is separated from the telomere in telomerase-negative young BJ cells with long telomeres, based on the assumption that the looping interactions suppress transcription. In old BJ cells at PD90, we found that the percentage of adjacent allele pairs was significantly reduced. Almost 60% of alleles were separated in the old cells with short telomeres, indicating that there is at least one hTERT locus spatially separated from the telomere in each cell. Importantly, we confirmed these 5p/TERT looping interactions in a second fibroblast cell strain, IMR90 (S2C and S2D Fig). We measured the number of separated and adjacent alleles in IMR90 cells young (PD 22) and old (PD 52) and show a shift from the majority of alleles being adjacent (76%) in young cells compared to the majority of alleles being separated (88%) in old cells. The looping data and the expression of hTERT are consistent. We suggest that old cells (with short telomeres) lose one control mechanism in regulating the hTERT locus (i.e., telomere chromatin looping) that helps repress the expression of hTERT. However, while we observed increased transcription of exon 1 of hTERT, there must be additional mechanisms preventing the inclusion of exons critical to produce active telomerase. There is substantial evidence that alternative splicing of hTERT may also have a major role in suppressing the production of active telomerase in old cells [22–24]. Furthermore, we performed 3D-FISH analysis in transformed SW26 and SW39 cells. SW cells are SV40 antigen expressing clones of IMR90 cells that have spontaneously immortalized using either telomerase (SW39) or an alternative lengthening of telomeres (ALT; SW26) mechanism to maintain telomeres (S2E Fig). In both cell lines, the majority of the alleles were separated (SW39 = 72%; SW26 = 66%), indicating that short telomeres due to replicative aging are likely responsible for the change in chromatin conformation and that a secondary change occurs to cause the production of full-length TERT or engage ALT. It has been suggested that hTERT shows mono-allelic expression in cancer, which is sufficient to preserve constant telomere length [25,26]. Our results support this assumption, as we observed that, on average, only one hTERT allele was generally in the open configuration during in vitro aging well before the onset of cancer. As controls for global conformational changes at chromosome 5p, we performed two additional FISH experiments. In the first experiment, we stained intermediate genomic region between the hTERT locus and the 5p telomere (S3A–S3C Fig). In addition, we also stained cells for two loci located 25.5 MB and 30.6 MB away from hTERT (S3D–S3F Fig). There were no changes in distances between the control loci in young and old cells, demonstrating that the conformation change occurs at the specific genomic region encompassing hTERT during in vitro aging, and this change is not due to classic TPE. To determine if we could artificially shorten telomeres and recapitulate the aging phenotype, we utilized CRISPR/Cas9 (clustered regularly interspaced short palindromic repeat-associated 9) to remove a large portion of the telomere and subtelomere region from chromosome 5p. This experiment allows testing the role of chromosome 5p’s telomere in regulating the looping observed in cells with short and long telomeres. As illustrated in Fig 2D, we also infected young BJ cells with a lentivirus expressing Cas9 protein and single guide-RNA targeting the sub-telomeric region of 5p to specifically disturb telomeres at chromosome 5p for a short period of time [27]. We also added an NHEJ inhibitor, SCR7, simultaneously during the infection to suppress repair of the double strand breaks induced by the Cas9 protein [28]. The targeted cells showed an unstable end structure of chromosome 5p (S5 Fig), and the specific disturbance of the 5p telomere significantly diminished telomere looping at the end of the chromosome 5p. We further examined if the proposed mechanism was present in BJ cell clones in which both young and old cells were passaged the same amount of time in culture. This approach was necessary to eliminate the possibility that young and old cells that were in culture for vastly differing times could introduce artifacts. To accomplish this, we expressed a floxable hTERT in BJ clones, followed by excision at different time points in order to make isogenic cells with different length of telomeres but passaged similar times in cell culture [12,29]. Telomere length of the early-excision clone was 9 kb, and this was extended up to 13 kb in the late-excision clone. The telomere length (terminal restriction fragment [TRF]) results are presented in Figure 1B. Population doublings were evenly matched between clones (to avoid confounding effects of passage of time in culture), and we also analyzed telomere looping. Similar to our observations in normally passaged BJ cells, the isogenic clones also showed decreased levels of telomere looping with telomere shortening (Fig 2E). Importantly, there were only background levels of DNA damage signaling during telomere shortening (Fig 2F) indicating that the change in genome structure occurred before initiation of DNA damage responses from critically short telomeres. To ensure that our staining protocol was robust, we induced DNA damage (double strand breaks) by treating long and short telomere BJ cells with zeocin and assaying for DNA damage (S2F Fig). These data can be interpreted to indicate that our staining protocol is robust and that we are analyzing cells before telomere-DNA damage induced foci are present or significant DNA damage occurs in the cells. We next performed droplet digital 3C (chromatin conformation capture) to detect the genomic interactions between the 5p telomere and the hTERT locus in young and old BJ cells (Fig 2G, left side). The results showed that the hTERT locus has specific genomic interactions with the 5p telomere, and the interaction was reduced during in vitro aging and telomere shortening. A proximity control primer which is 10kb away from the fixed primer at the hTERT locus was selected for normalization of 3C results (Fig 2G, right side). Taken together, telomere looping exists between the hTERT locus and the sub-telomeric 5p in normal human cells, and this looping is greatly reduced by gradual telomere shortening. Telomere Looping Determines Permissiveness of the hTERT Locus It has been shown that cis-elements upstream of the hTERT locus may play important roles in the tight regulation of human telomerase [30]. Thus, we decided to test if telomere looping could affect the epigenetic status of the hTERT proximal promoter region. We first analyzed DNA methylation of the region from -720bp to +90bp of the hTERT promoter in isogenic BJ cells with different lengths of telomeres but similar times in cell culture (Fig 3A). The relationship between DNA methylation and transcription in the hTERT promoter remains controversial in normal and cancer cells [31,32], but the transcription start site of hTERT retains little or no methylation in telomerase-active cancer cells for active transcription [33]. We found that the level of DNA methylation is significantly higher in BJ cells with long telomeres at several regions associated with hTERT and the hTERT region in comparison to cells with shorter telomeres. The largest differences were observed at -580bp, -250bp, -30bp, and +20bp of the hTERT promoter, including the E-box motif (a putative Myc binding sequence). It has also been reported that the proximal region of the hTERT promoter, including exon 1 and 2, regulates the activity of the hTERT promoter and that the methylation of this region is responsible for binding of several proteins [34,35]. Therefore, our results can be interpreted to indicate that telomere length-associated changes in methylation levels of the hTERT proximal promoter might affect transcriptional regulation of this locus. We next analyzed active and inactive histone marks on the hTERT proximal promoter using chromatin immunoprecipitation combined with droplet digital polymerase chain reaction (ChIP-ddPCR; [12]) (Fig 3B). We measured two histone marks associated with active chromatin H3K4 trimethylation (H3K4me3) and H3K9 acetylation (H3K4ac) and two histone marks associated with repressed chromatin H3K27 trimethylation (H3K27me3) and H3K9 trimethylation (H3K9me3), which have key roles in regulating gene expression [36]. We observed an increase in both H3K4me3 and H3K9ac across the TERT promoter in aged cells with short telomeres (Fig 3B). We also observed an increase in the repressive histone mark H3K27me3, but did not observe any significant differences in young or old BJ cells for the repressive histone mark H3K9me3. Collectively, this shows that the chromatin status of the hTERT promoter in old BJ cells with short telomeres is different and may be more transcriptionally permissive compared to young BJ cells with long telomeres. These data correlate well with the increased hTERT transcription we observed in cells with short telomeres. Furthermore, we analyzed chromatin at the promoters of three genes surrounding TERT that could also be affected by the altered chromatin environment with aging. We analyzed the proximal promoter regions of CLPTM1L, SCL6A18, and SCL6A19 for the same histone marks described above in the same cells and preparations used for TERT ChIP. At the CLPTM1L promoter we observed significant increases in histone marks indicating active transcription (Fig 3B). These data correlate well with an increase in CLPTM1L transcripts and protein levels (S1 Fig). We also observed significant changes in the chromatin surrounding the solute/amino acid transporter genes (SCL6A18 and SCL6A19), even though these genes are not expressed above basal/background levels in old/short telomere BJ cells. Specifically, we observed that both the repressive histone marks were increased in old cells (short telomeres) compared to young cells (long telomere). However, there was an increase in the activation marks as well. This indicates an intricate balance between chromatin modifications, methylation status, telomere length, and the expression of tissue-specific transcription and splicing factors that dictates the activation or repression of genes with replicative aging (telomere shortening—TPE-OLD). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Chromatin looping and epigenetic modifications (i.e., histone modifications and DNA methylation) determine permissiveness of the hTERT locus. (A) Bisulfite DNA methylation sequencing analysis of the hTERT proximal promoter region from -720bp to +90bp. Genomic DNA of BJ clones with different lengths of telomeres was modified and PCR-amplified. Each amplicon was TA-cloned for bacterial amplification and sequenced. Percentage of CpG methylation of the hTERT promoter is indicated in two BJ cell clones with different lengths of telomeres. (B) Illustration of genomic locus containing TERT. Black arrows indicate approximate location of primers in the promoters of the indicated genes. Chromatin immunoprecipitation was performed with BJ cells at PD34 and PD74. Six antibodies against H3K4me3, H3K9ac, H3K9me3, H3K27me3, H3 total, and IgG were used to pull down chromatin extracts, and the promoter regions of hTERT, SLC6A18, SLC6A19 and CLPTM1L were analyzed by ddPCR. Data are presented as means and standard errors of biological and technical duplicates. Student’s paired t tests comparing young and old determined significance (* = p < 0.05). (C) BJ cells at different PDs were infected with shRNAs against p21 (CDKN1A) and selected using puromycin to generate stable knockdown clones. hTERT mRNA expression was analyzed by ddPCR in sh-p21 cells and controls. The number of full-length and total hTERT mRNAs was assessed by amplifying the hTERT exon 7/8 junction and exon 15/16 junction, respectively. Knockdown efficiency of p21 was determined by western blotting. β-actin was used as a loading control. ANOVA revealed a significant effect. *p < 0.001. Data associated with this figure can be found in the supplemental data file (S1 Data). https://doi.org/10.1371/journal.pbio.2000016.g003 While we demonstrated that telomere shortening induced conformation changes between the hTERT locus and the sub-telomeric 5p resulting in up regulation of exon 1, presumably containing spliced hTERT transcripts in normal BJ cells (see Fig 1B), it did not result in full-length telomerase activity competent transcripts. Thus, we suggest that telomere shortening may render the hTERT locus more permissive and under oncogenic stress may lead to the production of full-length hTERT mRNA transcripts that could in turn produce telomerase activity. To test this, we simulated a step in spontaneous cancer transformation by knocking down p21 (CDKN1A) and analyzing mRNA expression level of hTERT (Fig 3C). The knockdown of p21 was previously shown to de-repress hTERT expression [37]. Thus, we tested if the knockdown of p21 would increase the expression of hTERT mRNAs and result in the inclusion of exons 7/8 in the short-telomere old BJ cells but not in the young BJ cells with long telomeres. We measured the expression level of hTERT transcripts in young and old BJ cells with and without p21 stable knockdown; mRNA containing exons 7/8 (exons coding for critical residues in the reverse transcriptase domain of TERT) and exon 15/16 (most splice variants of hTERT contain exons 15 and 16), responsible for putative active hTERT and total hTERT variants respectively. Both the active and the total hTERT transcript variants significantly increased with the knockdown of p21 in old BJ but not in young BJ cells; however, we did not detect telomerase activity (S6 Fig). While we observed an increased portion of transcripts that contain exons 7/8 of the TERT RT domain, other critical regions such as exon 2 may be spliced out [38]. Further work into the regulation of hTERT splicing is necessary to more fully understand the complex regulatory network surrounding hTERT and why the majority of transcripts are inactive splice variants as opposed to full length. While this result does not prove a causal role during cancer development, this series of experiments does demonstrate that telomere shortening in cells that bypass replicative senescence leads to the hTERT locus entering into a more permissive state (e.g., increased hTERT mRNA expression) in the presence of oncogenic stresses, consistent with the disengagement of telomere looping. TRF2 as a Mediator of Telomere Looping between the hTERT and Sub-telomeric 5p in Cells with Long Telomeres Characterization of cis- or trans-acting factors responsible for telomere looping will be important to understand this novel mechanism for telomerase regulation. A recent report showed that TRF2 (telomeric repeat-binding factor 2) protein is essential for the functional organization of chromosome ends, including human fibroblasts [39,40]. There is also mounting evidence for off-telomere functions of the shelterin components [41]. While a recent whole genome sequencing study found 2,920 interstitial TTAGGG repeats throughout the human genome [39], we also found frequent internal (interstitial) telomeric sequences (ITS) near the TERT locus in higher primates but not in rodent cells (Fig 4A). Thus, we first checked for a putative role of TRF2 in telomere looping in BJ cells as a candidate approach. We knocked down TRF2 by siRNA and performed 3C to directly assess the genomic interactions between the telomere and the hTERT locus (Fig 4B). The knockdown of TRF2 significantly reduced the genomic interactions between the telomere and the hTERT locus in young PD30 BJ cells, implying TRF2 may have a role in telomere looping interaction on hTERT locus. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. TRF2 as a mediator of telomere looping between the hTERT and sub-telomeric 5p in cells with long telomeres. (A) The number of (TTAGGG)2 repeats was analyzed on a small series of mammalian genomes. The TERT locus flanked by 350kb was examined. The (TTAGGG)2 repeats are depicted as a single black box. Multiple (TTAGGG)2 repeats at a locus are depicted by stacked black boxes. Transcription orientation of TERT and location of telomere are shown in figure. (B) 3C analyses show distal genomic interactions between the 5p telomere and hTERT. 3C libraries were generated from BJ cells treated with siRNA against TRF2, followed by ddPCR amplification to determine genomic interactions. t test revealed a significant effect. *p < 0.05, **p < 0.01 (C) ChIP shows enrichment of TRF2 protein on hTERT-ITS. The 1.153 Mbp to 1.154 Mbp region from the end of the chromosome 5p was examined. Chromatin extracts from BJ cell at PD25 and PD50 were prepared and pulled down with an antibody against TRF2. TTAGGG repeats are depicted as a red box. (D) A schematic map of the human chromosome 5 at the 1,153,167 bp to 1,353,167 bp region from the p arm terminus. Arrows indicate location of HindIII restriction enzyme sites. Red boxes indicate location of TTAGGGTTAGGG repeats along the genome. (E) 3C analysis shows distal genomic interactions between 5`end of hTERT locus and each HindIII site. 3C libraries were generated from BJ cells at PD25 and PD70, followed by ddPCR amplification to determine genomic interactions. t test revealed a significant effect. *p < 0.05, **p < 0.01. (F) ChIP shows enrichment of TRF2 protein on the hTERT promoter. Chromatin extracts from BJ cell at PD25 and PD50 were prepared and pulled down with an antibody against TRF2. The hTERT promoter region from -500bp to +200bp was analyzed by real-time-qPCR. t test revealed a significant effect. *p < 0.05, **p < 0.05, n.s. = no significance. (G) Model of genomic folding at hTERT locus based on 3C analyses. Blue boxes indicate the location of genes on chromosome 5p. Red boxes indicate the location of hTERT-ITS. Blue and red dashes indicate chromosome 5p and telomere of chromosome 5p, respectively. (H) BJ cells at PD 17 were transfected in individual experiments with a siRNA against TRF2, CTCF, and LDB1. Three days after transfection, cells were fixed for 3D-FISH analysis. Percentage of adjacent allele (A) pairs versus separated allele (S) pairs was determined by 3D-FISH. (I) Western blotting analysis shows knockdown efficiency of the siRNA against TRF2, CTCF, and LDB1. Histone H3 was used as a loading control. Data associated with this figure can be found in the supplemental data file (S1 Data). https://doi.org/10.1371/journal.pbio.2000016.g004 As shown in Fig 4A, a region 100 kb downstream of the hTERT (Chr5: 1,154,047–1,154,347) contains a series of internal telomeric sequences that may recruit TRF2 shelterin protein (hereafter termed hTERT-ITS). Thus, we reasoned that this region would be a putative binding site for TRF2 and may be responsible for the telomere looping interaction between the telomere and the hTERT locus in cells with long but not short telomeres. ChIP-qPCR analysis showed that the TRF2 protein associates with the hTERT-ITS region in young and old BJ cells as proposed (Fig 4C). We next performed 3C to further clarify that hTERT-ITS interact with the hTERT promoter by genome folding to affect transcriptional permissiveness as shown in Figs 1 and 3. Within 200kb, we found more than 20 HindIII restriction enzyme sites were in the hTERT/CLPTM1L locus (Fig 4D). Droplet digital PCR (ddPCR)-mediated amplification showed specific interactions between the 5ʹ end of hTERT and the hTERT-ITS (Fig 4E). Moreover, the interaction was weakened in old BJ cells, implying there might be a transition from a more repressive state to a more active state of this TAD location during in vitro aging, consistent with the increased hTERT mRNA, altered methylation, and chromatin. This result also shows that there is an additional genome folding between the hTERT locus and the hTERT-ITS at an intermediate region between the SLC6A18 and SLC6A19 loci. The hTERT promoter is not close to the hTERT-ITS on a linear genome map, but the unique genome folding at this region potentially positions the hTERT promoter close to the ITS, followed by putative TRF2-mediated telomere recruitment to the hTERT promoter only in cells with long telomeres. In Fig 4F, we demonstrate that TRF2 protein is also enriched in the hTERT promoter region using ChIP-qPCR approaches. While TRF2 protein was enriched at proximal regions on the hTERT promoter, the interaction was significantly decreased in old BJ cells at the genomic regions containing -350bp to -50bp of the hTERT promoter. This shows TRF2 protein can occupy the hTERT promoter region, but the interaction is weakened during in vitro aging and telomere shortening. Together, we interpret these experiments to indicate that TRF2, and perhaps upregulated TERRA, may have at least a partial mechanistic role in telomere looping at the hTERT locus through interaction with the conserved interstitial telomeric repeats. Because we have shown the interaction between the 5p telomere and the hTERT locus, we modeled one possibility for the detailed local genome structure of this locus (Fig 4G). In this model, the hTERT promoter is close to the hTERT-ITS by genome folding in young cells with long telomeres. In addition, this model shows that TRF2 protein is recruited to hTERT locus and hTERT-ITS, which makes this interaction potentially dependent on telomere length. In summary, the hTERT promoter has specific interactions with the hTERT-ITS through gene looping, which may also recruit telomere length-dependent looping (TPE-OLD) mechanisms through TRF2 protein. We next performed 3D-FISH to visualize the genomic structure changes between the hTERT locus and the sub-telomeric 5p (Fig 4H). Control PD17 BJ cells showed that 89% of the hTERT and sub-telomeric 5p allele pairs were adjacent, but knockdown of TRF2 reduced this down to 34%. We also knocked down CTCF (CCCTC-binding factor) and LDB1 (LIM domain-binding protein 1), which are proposed to be essential proteins in global gene looping maintenance [42,43]. CTCF and LDB1 knockdown also significantly reduced the adjacent allele pairs, implying that the general gene looping mechanisms may also be involved in telomere looping. Western blotting was also performed to show knockdown efficiency (Fig 4I). Taken together, TRF2, part of the shelterin complex, may be mechanistically involved in the establishment of telomere looping near the hTERT locus through ITS together with general chromosome looping mechanisms. Telomere Length Affects Expression of hTERT in Telomerase-Positive Cancer Cells In almost all primary human cancers, telomere length is very short compared to adjacent normal tissues [44]. It is likely that short telomeres, in combination with oncogenic alterations, result in the hTERT gene becoming more permissive for protein expression and enzyme activity. Thus, we next investigated how telomere length affects hTERT expression in telomerase-active cancer cells. We first infected hTERT and hTR (hTERC) into the SW39 cell line (SV40 immortalized human telomerase expressing fibroblasts) and analyzed mRNA expression of the endogenous hTERT by examining the 3ʹ untranslated region. We observed that the extended telomere length reduced endogenous expression of hTERT mRNA in qPCR analysis (Fig 5A) implying TPE-OLD remains engaged at least in this tumor cell line. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. Telomere length affects expression of hTERT in telomerase-positive cancer cells. (A) SW39 cells were infected with viruses for ectopic expression of hTERT and hTR (hTERC). RNA was purified, and the 3ʹUTR was amplified to analyze endogenous level of hTERT mRNA. ddPCR was performed to assess relative expression levels. t test revealed a significant effect. *p < 0.005 (B) HeLa cells were cloned and infected with a floxable hTERT. After treatment of adeno-cre recombinase at different time points, HeLa clones with long and short telomeres at same number of population doublings in cell culture were established. The number of full-length, total, and minus beta hTERT mRNAs were assessed by amplifying exon7/8 junction, exon 15/16 junction, and the exon 6/9 junction respectively. Telomere lengths are determined by TRF and indicated under the graph. t test revealed a significant effect between splice variants. *p < 0.001 (C) Percentage of adjacent allele (A) pairs versus separated allele (S) pairs was determined by 3D-FISH in HeLa clones. HeLa clones with long and short telomeres at the same PD in culture were analyzed. (D) Simplified model of how TPE-OLD regulates hTERT expression in human cells during aging and cancer progression. See text for details. Data associated with this figure can be found in the supplemental data file (S1 Data). https://doi.org/10.1371/journal.pbio.2000016.g005 We further established isogenic HeLa cell clones with different telomere lengths by excising a floxable hTERT cDNA at different time points. We examined expression of splice variants of hTERT mRNA containing total, full-length (indicative of telomerase activity), and minus beta alternative spliced forms through ddPCR analysis (Fig 5B). All three splice variants showed significantly decreased expression in the long-telomere HeLa clone. We also performed 3D-FISH to analyze changes in genomic structure between the hTERT locus and the sub-telomeric 5p after the extension of telomeres in HeLa cells (Fig 5C). The long-telomere HeLa clone showed a higher percentage of adjacent allele pairs compared with the short-telomere HeLa clone. This indicates that the expression of hTERT may also be influenced by the length of telomeres through TPE-OLD in telomerase-positive cancer cells. Discussion The local genome structure around the hTERT locus may be important for the tight regulation of human telomerase. For example, introduction of proximal cis-elements of the hTERT promoter sufficiently inhibits the activity of the TERT promoter [45]. In addition, chemicals perturbing chromatin structure, including trichostatin A and 5-aza-2ʹ-deoxycytidine, induce changes in hTERT expression [46]. Moreover, chromosomal translocation and gene duplication of the hTERT locus can occur as part of the immortalization process in primary cultured cells [47,48]. Here, we reasoned that the hTERT locus might recruit telomeric heterochromatin to regulate its own gene expression, especially in large, long-lived mammals where tumor suppression mechanisms are perhaps more important. We showed that telomere looping exists in long-telomere young fibroblasts and that telomere looping was reduced by in vitro aging. This is one possible explanation for why higher primates preserved the location of the TERT gene at the end of one of their chromosomes. We speculate that, in addition to other conserved tumor suppressor mechanisms, higher primates also developed a mechanism to suppress the undesired expression of TERT. For example, it is well established that during human fetal development, full-length telomerase transcription is repressed and correlates with increases in nonfunctional alternative splicing changes in hTERT [8]. Thus, during early human development, when telomerase is active, telomeres elongate. Our current results are consistent with the idea that longer telomeres can fold back on the TERT locus and repress or significantly reduce transcription. Our results also show that replicative senescence, while initially a tumor suppressor mechanism, may paradoxically impinge on the predisposition to cancer through telomerase transcriptional de-repression. While still preliminary, the hTERT locus is arranged in a local chromatin domain that is regulated by telomere length and the interstitial telomere sequences in the vicinity of the hTERT locus. We showed that expression of the CLPTM1L gene (adjacent to hTERT) is also regulated by the length of telomeres and predicts transcriptional permissiveness of this locus. However, because hTERT re-activation is an extremely rare event, there may be additional levels of regulation. We propose that, upon telomere shortening, the hTERT region becomes permissive (as indicated by increased transcription of exon 1 containing RNAs), but this first step is not sufficient to support full-length hTERT transcripts at an adequate level to produce telomerase enzyme activity. We further propose that there is another biological role for telomere looping at this locus during development to repress telomerase when telomere length homoeostasis is reached (i.e., suggesting that having too-long telomeres may be detrimental). Here, we demonstrated a novel epigenetic mechanism regulating hTERT expression during in vitro aging (Fig 5D). Cells with long telomeres at the end of chromosome 5p in young passaged cells form a chromatin loop in the region of the hTERT locus. Importantly, we demonstrated that the chromatin loop is disengaged in cells with short telomeres, leading to partial increased expression of hTERT mRNA during in vitro aging and in response to p21 knockdown; however, telomerase activity was not detected, and, alternatively, spliced variants were likely produced [17,22–24,38]. Finally, we demonstrated that, in old cells with short telomeres, re-introduction of hTERT and elongation of telomeres results in a re-engagement of TPE-OLD. We found that DNA methylation and histone modifications in the hTERT promoter region showed significant changes as cells developed shorter telomeres, and that TRF2 and, perhaps, TERRA, may have important roles in these age-dependent genomic changes. These observations offer a model and a partial explanation for how age-dependent changes in the genome structure affect the regulation of hTERT without initiating a DNA damage response from a critically shortened telomere. Materials and Methods Cell Culture BJ, SW39, HeLa, HEK293FT, IMR90, and Phoenix A cells were maintained in a 4:1 ratio of Dulbecco’s modified Eagle’s medium to Medium 199 containing 10% of fetal bovine serum (Hyclone, Logan, UT, USA) under 5% CO2 in a humidified incubator. Retrovirus containing human TERT cDNA was infected into BJ cells and HeLa cells, followed by adenoviral infection for transient expression of Cre recombinase at different time points to produce cells with different lengths of telomeres that had been passaged in vitro for similar times [12,29]. Retrovirus was prepared by transfecting viral vectors into Phoenix A cells for 48 h. Medium containing virus was filtered through a 0.45 μm pore and provided to cells in the presence of 2 μg/ml polybrene. Lentivirus was prepared by transfecting viral vectors into HEK293FT cells with two packaging vectors (pMD2 and psPAX2) for 48 h. Medium containing virus was filtered through a 0.45 μm pore and cells exposed to lentivirus in the presence of 2 μg/ml polybrene. Selection for hygromycin was performed using 100 μg/ml and for puromycin using 1 μg/ml. CRISPR-Cas9 introduction for 5p genomic editing was performed by infecting cells with lentivirus carrying sgRNA target sequence of 5ʹ-GCCTCACTCCTTACGGAGTG-3ʹ. Three-Dimensional Fluorescence In Situ Hybridization (3D-FISH) 3D-FISH was performed as described previously [12]. 104 BJ cells were seeded into 4-chambered slides. Cells on slides were fixed with 4% paraformaldehyde, followed by permeabilization with 0.1% Triton X-100 in PBS. Repeated liquid nitrogen freezing-thawing cycles were performed for further permeabilization with preservation of intact nuclear structure under 20% glycerol in PBS. After 5 d of incubation of with 50% formamide in 2X SSC, cells were stained with indicated probes at 37°C for overnight. Slides were washed with 0.1% SDS in 0.5X SSC at 70°C for 5 min, followed by 2 rounds of PBST (Phosphate-buffered saline with Triton X-100) washing for 10 min. Images were acquired using a LSM780 confocal microscope (Carl Zeiss, Jena, Germany), and analyzed by Imaris deconvolution software (Bitplane, Zurich, Switzerland). The proximity of allele pairs was determined visually and quantitated. At least 30 nuclei were counted for the statistical analyses. We used the following criteria for the analyses: adjacent ~0.5 μm space or less between probes, separated ~1.0 μm between probes or more. The length was determined by calculating the 3D distance between each center of deconvolved fluorescent spots. Probes were prepared using nick translation kits (Abbott Laboratories, Abbott Park, IL, USA) from each BAC following manufacturer’s instructions. BAC plasmids were purchased from CHORI (Children’s Hospital Oakland Research Institute, Oakland, CA, USA); RP11-990A6 for hTERT locus staining and RP11-44H14 for sub-telomeric region 5p staining. Quality of probes was assessed by metaphase spread analyses and PCR. Droplet Digital PCR (ddPCR) and TRAP (ddTRAP) DdPCR and ddTRAP were performed as previously described [12,49]. Messenger RNA was prepared from RNeasy plus mini kit (Qiagen, Valencia, CA, USA) following the manufacturer’s instructions. 100 ng of RNA was reverse-transcribed from cDNA synthesis kit (Bio-Rad, Hercules, CA, USA) by following the manufacturer’s instructions. Ten percent of synthesized cDNA was used for the ddPCR reaction. For ddTRAP, harvested cells were lysed in NP40 lysis buffer (1mM Tris-Cl pH8.0, 1mM MgCl2, 1mM EDTA, 1% NP40, 0.25mM sodium deoxycholate, 10% glycerol, 0.15M NaCl, 0.05% 2-ME) for ddTRAP. Lysate was used for TS extension, and the extended products were analyzed with ddTRAP. Endogenous levels of 3ʹUTR were assessed with EvaGreen dye (Bio-Rad, Hercules, CA, USA). Probes were purchased from Roche (Basel, Switzerland), and the primer sequences are described below: 5ʹUTR-Exon 1 hTERT 5ʹ-AGCCCCTCCCCTTCCTTT 5ʹ-TGCGTCCCAGGGCACGCACACCAGGCACTG Full-length hTERT (exon7/8) 5ʹ-GCGTAGGAAGACGTCGAAGA-3ʹ 5ʹ-ACAGTTCGTGGCTCACCTG-3ʹ Probe-UPL #52 Total hTERT (exon15/16) 5ʹ-GGGTCACTCAGGACAGCCCAG-3ʹ 5ʹ-GGGCGGGTGGCCATCAGT-3ʹ Probe-UPL #37 Minus beta hTERT (exon6/9) 5ʹ-CAAGAGCCACGTCCTACGTC-3ʹ 5ʹ-CAAGAAATCATCCACCAAACG-3ʹ Probe-UPL #58 3ʹUTR of endogenous hTERT 5ʹ-CAGCTTTTCCTCACCAGGAG-3ʹ 5ʹ-GGTCACTCCAAATTCCCAGA-3ʹ Bisulfite Sequencing 100 ng of gDNA was modified using the EpiTect Bisulfite kit by following the manufacturer’s instructions (Qiagen, Valencia, CA, USA). Modified DNA was PCR-amplified and cloned into the T vector system (Promega, Madison, WI, USA). 7~10 bacterial clones were sequenced for methylation analysis. Primers for the hTERT promoter region amplification were designed as previously described [33]. dd3C (Droplet Digital Chromatin Conformation Capture) Chromatin conformation capture (3C) was performed as previously described [12]. Five million cells were washed with PBS and fixed with 25 ml of medium containing 1% formaldehyde for 10 min at room temperatures. To quench the crosslinking reaction, 1.5 ml of 2.5 M glycine was added and incubated for 10 min at room temperature, followed by an additional 15 min of incubation at 4°C. Cells were washed with PBS and harvested into 1 ml of cold-PBS with protease inhibitor. Cells were next lysed by homogenization, and the nuclear pellet was collected by centrifugation. The nuclear pellet was washed and resuspended in 500 μl of ice-cold NEBuffer 2 (NEB, Ipswich, MA, USA). 15 μl of 10% SDS was added and incubated at 37°C for 1 h, followed by addition of 46.35 μl of 20% Triton X-100 for 1 h on a shaking incubator. HindIII (400U) was added and incubated overnight. Enzyme reaction was stopped by adding 88 μl of 10% SDS at 65°C for 20 min. Samples were next transferred to DNA ligation mix containing 50 mM Tris-Cl, pH 7.5, 10 mM MgCl2, 1 mM ATP, 10 mM DTT, and 50 μg/ml BSA. 372 μl of 20% Triton X-100 was added and incubated at 37°C for 1 h. 2,000 U of ligase (NEB, Ipswich, MA, USA) was added and incubated for 5 h at 16°C. 40 μl of 20 mg/ml Proteinase K was added to the ligation mix at 65°C overnight. DNA extraction was performed by phenol-chloroform extraction and ethanol precipitation. Quality of the libraries were determined by checking for a single DNA band under agarose gel electrophoresis. Taq-man probe and 5ʹ primers were selected to amplify constant regions at the 5p telomere regardless of genome conformation. 3ʹ primers were selected to amplify the genomic interaction between 5p telomere and subtelomeric genes up to 1.3 mega base pairs from 5p containing hTERT. Primer binding regions are 100 base pairs apart from a HindIII recognizing motif. Primer and probe sequences are described below; Probe 5ʹ-[6FAM] GCCAACACAGGAATGAATTG [BHQ1]-3ʹ Constant 5ʹ primer at 5p 5ʹ-GCCAATAAAAACAGCTACCGATG-3ʹ 3ʹ primers 5ʹ-GAGATAACTCACTACCTTCAGACCA-3ʹ (PLEKHG4B) 5ʹ-TTGAAGACATTCCTCACATCCC-3ʹ (SDHA) 5ʹ-TGTTTCCGTGATTCCTGGCAC-3ʹ (PDCD6) 5ʹ-TGGACTGTGTTGTGGGTCCTC-3ʹ (AHRR) 5ʹ-ATGCACCCACAGGTGGGTG-3ʹ (SLC9A3) 5ʹ-CTGCTGAGAAGTGTTGCCTTCT-3ʹ (CEP72) 5ʹ-ATTAGGATCACCCATCGCAG-3ʹ (TPPP) 5ʹ-TGCGCAGCATTTTGCACATG-3ʹ (ZDHHC11) 5ʹ-ATGTCGGCTTGGCCTAGAAG-3ʹ (BRD9) 5ʹ-AGGTCACTGCTGGCCTGG-3ʹ (NKD2) 5ʹ-ATGCTGGTGCCAGCTCTGAG-3ʹ (SLC12A7) 5ʹ-AGGGCTCTGGGATGTGCTG-3ʹ (SLC6A19) 5ʹ-CATTTGGAGTCCATGGAGTGAG-3ʹ (hTERT) 5ʹ-CCAGCTGTTCAGTTCAGCAGC-3ʹ (CLPTM1L) 5ʹ-TAATAGGAAGTTAACGTGCTTTGGC-3ʹ (SLC6A3) Chromatin Immunoprecipitation (ChIP) Analysis Chromatin immunoprecipitation was performed as previously described [12]. Antibodies against total histone H3 and a 1:1 mixture of rabbit and mouse IgG isotypes were used as pulldown positive and negative controls of ChIP analyses, respectively. Relative occupancy was determined by first normalizing the target results with amplification signals from total H3 and then dividing by 1% input chromatin extracts. Antibodies against H3K4me3, H3K27me3, H3K9me3, H3K9ac, and LDB1 were purchased from Abcam (Cambridge, MA, USA). Antibody against TRF2 was purchased from Novus biologicals (Littleton, CO, USA). Antibody against CTCF and histone H3 was purchased from Cell signaling (Cell signaling technology, Danvers, MA, USA). Primers for hTERT promoter amplification were described in a previous study [33]. Primers for detection of CLPTML1, SLC6A18, CLC6A19, and the hTERT-ITS are described below; CLPTM1L promoter ChIP 5ʹ- TGGGTTTGTACTGGGGAAAA 5ʹ- GAGCCTGGTGGAAGGTGATA SLC6A18 promoter ChIP 5ʹ- CCTGGTGTCTGCAACAAAAA 5ʹ- GCCCCACTGCAGTTGTATTT SLC6A19 promoter ChIP 5ʹ- TCTGGGTCCTGAACCTATGG 5ʹ- GATGTGGCCTGAATCAACCT TERT ITS primer #1 5ʹ-GGAGCTGTGGTCTGTGTCTC-3ʹ 5ʹ-ACGCTAACCCTAACCCACAG-3ʹ TERT ITS primer #2 5ʹ-GGGTTAGGGACACAAGCCTG-3ʹ 5ʹ-TAGAAGGGCAGGTGTCTCGT-3ʹ TERT ITS primer #3 5ʹ-AGCAGACACCTGCCCTTCTA-3ʹ 5ʹ-ACTTTGTGTGCATCTGGGGA-3ʹ Telomere Dysfunction Induced Foci (TIF) Assay The TIF assay is based on the co-localization detection of DNA damage by an antibody against gamma-H2AX and telomeres using FITC-conjugated telomere sequence (TTAGGG)3-specific peptide nucleic acid (PNA) probe. Briefly, BJ cells with long and short telomeres (100,000 cells) were seeded to four-well chamber slides, and, after the cells attached to the surface (next day), slides were rinsed twice with 1xPBS and fixed in 4% formaldehyde (ThermoScientific, IL) in PBS for 10 min. Then, cells were washed twice with PBS and permeabilized in 0.5% Triton X-100 in PBS for 10 min. Following permeabilization, cells were washed three times with PBS. Cells were blocked with 10% goat serum in 0.1% PBST (TritonX-100) for 1 h. Gamma-H2AX (mouse) (Millipore, Billerica, MA) was diluted 1:1,000 in blocking solution and incubated on cells for 2 h. Following three washes with PBST (1x PBS in 0.1% Triton) and three washes with PBS, cells were incubated with Alexaflour 568 conjugated goat anti mouse (1:500) (Invitrogen, Grand Island, NY) for 40 min, then washed five times with 0.1% PBST. Cells were fixed in 4% formaldehyde in PBS for 20 min at room temperature. The slides were sequentially dehydrated with 70%, 90%, and 100% ethanol. Following dehydration, denaturation was conducted with hybridization buffer containing FITC-conjugated telomere sequence (TTAGGG)3-specific peptide nucleic acid (PNA) probe (PNA Bio, Thousand Oaks, CA), 70% formamide, 30% 2xSSC, 10% (w/v) MgCl2.6*H20 (Fisher Sci), 0.25% (w/v) blocking reagent for nucleic acid hybridization and detection (Roche) for 7 min at 80°C on heat block, followed by overnight incubation at room temperature. Slides were washed sequentially with 70% formamide (Ambion, Life Technologies, Grand Island, NY) / 0.6 x SSC (Invitrogen) (2 x 1 h), 2 x SSC (1 x 15 min), PBS (1 x 5 min), and sequentially dehydrated with 70%, 90%, and 100% ethanol, then mounted with Vectashield mounting medium with DAPI (Vector Laboratories, Burlingame, CA). Images were captured with Deltavision wide-field microscope using the 60X objective. TIFs were quantified using Image J and representative pictures were prepared in Imaris software after deconvolution using Autoquant X3. Terminal Restriction Fragmental (TRF) Length Analysis for Mean Telomere Length The average length of telomeres (terminal restriction fragment lengths) was measured as described in [50] with the following modifications. DNA was transferred to Hybond-N+ membranes (GE Healthcare, Piscataway, NJ) using vacuum transfer. The membrane was air-dried and DNA was fixed by UV-crosslinking. Membranes were then probed for telomeres using a DIG-labeled telomere probe [51], detected with an HRP-linked anti-DIG antibody (Roche), and exposed with CDP-star (Roche). Cell Culture BJ, SW39, HeLa, HEK293FT, IMR90, and Phoenix A cells were maintained in a 4:1 ratio of Dulbecco’s modified Eagle’s medium to Medium 199 containing 10% of fetal bovine serum (Hyclone, Logan, UT, USA) under 5% CO2 in a humidified incubator. Retrovirus containing human TERT cDNA was infected into BJ cells and HeLa cells, followed by adenoviral infection for transient expression of Cre recombinase at different time points to produce cells with different lengths of telomeres that had been passaged in vitro for similar times [12,29]. Retrovirus was prepared by transfecting viral vectors into Phoenix A cells for 48 h. Medium containing virus was filtered through a 0.45 μm pore and provided to cells in the presence of 2 μg/ml polybrene. Lentivirus was prepared by transfecting viral vectors into HEK293FT cells with two packaging vectors (pMD2 and psPAX2) for 48 h. Medium containing virus was filtered through a 0.45 μm pore and cells exposed to lentivirus in the presence of 2 μg/ml polybrene. Selection for hygromycin was performed using 100 μg/ml and for puromycin using 1 μg/ml. CRISPR-Cas9 introduction for 5p genomic editing was performed by infecting cells with lentivirus carrying sgRNA target sequence of 5ʹ-GCCTCACTCCTTACGGAGTG-3ʹ. Three-Dimensional Fluorescence In Situ Hybridization (3D-FISH) 3D-FISH was performed as described previously [12]. 104 BJ cells were seeded into 4-chambered slides. Cells on slides were fixed with 4% paraformaldehyde, followed by permeabilization with 0.1% Triton X-100 in PBS. Repeated liquid nitrogen freezing-thawing cycles were performed for further permeabilization with preservation of intact nuclear structure under 20% glycerol in PBS. After 5 d of incubation of with 50% formamide in 2X SSC, cells were stained with indicated probes at 37°C for overnight. Slides were washed with 0.1% SDS in 0.5X SSC at 70°C for 5 min, followed by 2 rounds of PBST (Phosphate-buffered saline with Triton X-100) washing for 10 min. Images were acquired using a LSM780 confocal microscope (Carl Zeiss, Jena, Germany), and analyzed by Imaris deconvolution software (Bitplane, Zurich, Switzerland). The proximity of allele pairs was determined visually and quantitated. At least 30 nuclei were counted for the statistical analyses. We used the following criteria for the analyses: adjacent ~0.5 μm space or less between probes, separated ~1.0 μm between probes or more. The length was determined by calculating the 3D distance between each center of deconvolved fluorescent spots. Probes were prepared using nick translation kits (Abbott Laboratories, Abbott Park, IL, USA) from each BAC following manufacturer’s instructions. BAC plasmids were purchased from CHORI (Children’s Hospital Oakland Research Institute, Oakland, CA, USA); RP11-990A6 for hTERT locus staining and RP11-44H14 for sub-telomeric region 5p staining. Quality of probes was assessed by metaphase spread analyses and PCR. Droplet Digital PCR (ddPCR) and TRAP (ddTRAP) DdPCR and ddTRAP were performed as previously described [12,49]. Messenger RNA was prepared from RNeasy plus mini kit (Qiagen, Valencia, CA, USA) following the manufacturer’s instructions. 100 ng of RNA was reverse-transcribed from cDNA synthesis kit (Bio-Rad, Hercules, CA, USA) by following the manufacturer’s instructions. Ten percent of synthesized cDNA was used for the ddPCR reaction. For ddTRAP, harvested cells were lysed in NP40 lysis buffer (1mM Tris-Cl pH8.0, 1mM MgCl2, 1mM EDTA, 1% NP40, 0.25mM sodium deoxycholate, 10% glycerol, 0.15M NaCl, 0.05% 2-ME) for ddTRAP. Lysate was used for TS extension, and the extended products were analyzed with ddTRAP. Endogenous levels of 3ʹUTR were assessed with EvaGreen dye (Bio-Rad, Hercules, CA, USA). Probes were purchased from Roche (Basel, Switzerland), and the primer sequences are described below: 5ʹUTR-Exon 1 hTERT 5ʹ-AGCCCCTCCCCTTCCTTT 5ʹ-TGCGTCCCAGGGCACGCACACCAGGCACTG Full-length hTERT (exon7/8) 5ʹ-GCGTAGGAAGACGTCGAAGA-3ʹ 5ʹ-ACAGTTCGTGGCTCACCTG-3ʹ Probe-UPL #52 Total hTERT (exon15/16) 5ʹ-GGGTCACTCAGGACAGCCCAG-3ʹ 5ʹ-GGGCGGGTGGCCATCAGT-3ʹ Probe-UPL #37 Minus beta hTERT (exon6/9) 5ʹ-CAAGAGCCACGTCCTACGTC-3ʹ 5ʹ-CAAGAAATCATCCACCAAACG-3ʹ Probe-UPL #58 3ʹUTR of endogenous hTERT 5ʹ-CAGCTTTTCCTCACCAGGAG-3ʹ 5ʹ-GGTCACTCCAAATTCCCAGA-3ʹ Bisulfite Sequencing 100 ng of gDNA was modified using the EpiTect Bisulfite kit by following the manufacturer’s instructions (Qiagen, Valencia, CA, USA). Modified DNA was PCR-amplified and cloned into the T vector system (Promega, Madison, WI, USA). 7~10 bacterial clones were sequenced for methylation analysis. Primers for the hTERT promoter region amplification were designed as previously described [33]. dd3C (Droplet Digital Chromatin Conformation Capture) Chromatin conformation capture (3C) was performed as previously described [12]. Five million cells were washed with PBS and fixed with 25 ml of medium containing 1% formaldehyde for 10 min at room temperatures. To quench the crosslinking reaction, 1.5 ml of 2.5 M glycine was added and incubated for 10 min at room temperature, followed by an additional 15 min of incubation at 4°C. Cells were washed with PBS and harvested into 1 ml of cold-PBS with protease inhibitor. Cells were next lysed by homogenization, and the nuclear pellet was collected by centrifugation. The nuclear pellet was washed and resuspended in 500 μl of ice-cold NEBuffer 2 (NEB, Ipswich, MA, USA). 15 μl of 10% SDS was added and incubated at 37°C for 1 h, followed by addition of 46.35 μl of 20% Triton X-100 for 1 h on a shaking incubator. HindIII (400U) was added and incubated overnight. Enzyme reaction was stopped by adding 88 μl of 10% SDS at 65°C for 20 min. Samples were next transferred to DNA ligation mix containing 50 mM Tris-Cl, pH 7.5, 10 mM MgCl2, 1 mM ATP, 10 mM DTT, and 50 μg/ml BSA. 372 μl of 20% Triton X-100 was added and incubated at 37°C for 1 h. 2,000 U of ligase (NEB, Ipswich, MA, USA) was added and incubated for 5 h at 16°C. 40 μl of 20 mg/ml Proteinase K was added to the ligation mix at 65°C overnight. DNA extraction was performed by phenol-chloroform extraction and ethanol precipitation. Quality of the libraries were determined by checking for a single DNA band under agarose gel electrophoresis. Taq-man probe and 5ʹ primers were selected to amplify constant regions at the 5p telomere regardless of genome conformation. 3ʹ primers were selected to amplify the genomic interaction between 5p telomere and subtelomeric genes up to 1.3 mega base pairs from 5p containing hTERT. Primer binding regions are 100 base pairs apart from a HindIII recognizing motif. Primer and probe sequences are described below; Probe 5ʹ-[6FAM] GCCAACACAGGAATGAATTG [BHQ1]-3ʹ Constant 5ʹ primer at 5p 5ʹ-GCCAATAAAAACAGCTACCGATG-3ʹ 3ʹ primers 5ʹ-GAGATAACTCACTACCTTCAGACCA-3ʹ (PLEKHG4B) 5ʹ-TTGAAGACATTCCTCACATCCC-3ʹ (SDHA) 5ʹ-TGTTTCCGTGATTCCTGGCAC-3ʹ (PDCD6) 5ʹ-TGGACTGTGTTGTGGGTCCTC-3ʹ (AHRR) 5ʹ-ATGCACCCACAGGTGGGTG-3ʹ (SLC9A3) 5ʹ-CTGCTGAGAAGTGTTGCCTTCT-3ʹ (CEP72) 5ʹ-ATTAGGATCACCCATCGCAG-3ʹ (TPPP) 5ʹ-TGCGCAGCATTTTGCACATG-3ʹ (ZDHHC11) 5ʹ-ATGTCGGCTTGGCCTAGAAG-3ʹ (BRD9) 5ʹ-AGGTCACTGCTGGCCTGG-3ʹ (NKD2) 5ʹ-ATGCTGGTGCCAGCTCTGAG-3ʹ (SLC12A7) 5ʹ-AGGGCTCTGGGATGTGCTG-3ʹ (SLC6A19) 5ʹ-CATTTGGAGTCCATGGAGTGAG-3ʹ (hTERT) 5ʹ-CCAGCTGTTCAGTTCAGCAGC-3ʹ (CLPTM1L) 5ʹ-TAATAGGAAGTTAACGTGCTTTGGC-3ʹ (SLC6A3) Chromatin Immunoprecipitation (ChIP) Analysis Chromatin immunoprecipitation was performed as previously described [12]. Antibodies against total histone H3 and a 1:1 mixture of rabbit and mouse IgG isotypes were used as pulldown positive and negative controls of ChIP analyses, respectively. Relative occupancy was determined by first normalizing the target results with amplification signals from total H3 and then dividing by 1% input chromatin extracts. Antibodies against H3K4me3, H3K27me3, H3K9me3, H3K9ac, and LDB1 were purchased from Abcam (Cambridge, MA, USA). Antibody against TRF2 was purchased from Novus biologicals (Littleton, CO, USA). Antibody against CTCF and histone H3 was purchased from Cell signaling (Cell signaling technology, Danvers, MA, USA). Primers for hTERT promoter amplification were described in a previous study [33]. Primers for detection of CLPTML1, SLC6A18, CLC6A19, and the hTERT-ITS are described below; CLPTM1L promoter ChIP 5ʹ- TGGGTTTGTACTGGGGAAAA 5ʹ- GAGCCTGGTGGAAGGTGATA SLC6A18 promoter ChIP 5ʹ- CCTGGTGTCTGCAACAAAAA 5ʹ- GCCCCACTGCAGTTGTATTT SLC6A19 promoter ChIP 5ʹ- TCTGGGTCCTGAACCTATGG 5ʹ- GATGTGGCCTGAATCAACCT TERT ITS primer #1 5ʹ-GGAGCTGTGGTCTGTGTCTC-3ʹ 5ʹ-ACGCTAACCCTAACCCACAG-3ʹ TERT ITS primer #2 5ʹ-GGGTTAGGGACACAAGCCTG-3ʹ 5ʹ-TAGAAGGGCAGGTGTCTCGT-3ʹ TERT ITS primer #3 5ʹ-AGCAGACACCTGCCCTTCTA-3ʹ 5ʹ-ACTTTGTGTGCATCTGGGGA-3ʹ Telomere Dysfunction Induced Foci (TIF) Assay The TIF assay is based on the co-localization detection of DNA damage by an antibody against gamma-H2AX and telomeres using FITC-conjugated telomere sequence (TTAGGG)3-specific peptide nucleic acid (PNA) probe. Briefly, BJ cells with long and short telomeres (100,000 cells) were seeded to four-well chamber slides, and, after the cells attached to the surface (next day), slides were rinsed twice with 1xPBS and fixed in 4% formaldehyde (ThermoScientific, IL) in PBS for 10 min. Then, cells were washed twice with PBS and permeabilized in 0.5% Triton X-100 in PBS for 10 min. Following permeabilization, cells were washed three times with PBS. Cells were blocked with 10% goat serum in 0.1% PBST (TritonX-100) for 1 h. Gamma-H2AX (mouse) (Millipore, Billerica, MA) was diluted 1:1,000 in blocking solution and incubated on cells for 2 h. Following three washes with PBST (1x PBS in 0.1% Triton) and three washes with PBS, cells were incubated with Alexaflour 568 conjugated goat anti mouse (1:500) (Invitrogen, Grand Island, NY) for 40 min, then washed five times with 0.1% PBST. Cells were fixed in 4% formaldehyde in PBS for 20 min at room temperature. The slides were sequentially dehydrated with 70%, 90%, and 100% ethanol. Following dehydration, denaturation was conducted with hybridization buffer containing FITC-conjugated telomere sequence (TTAGGG)3-specific peptide nucleic acid (PNA) probe (PNA Bio, Thousand Oaks, CA), 70% formamide, 30% 2xSSC, 10% (w/v) MgCl2.6*H20 (Fisher Sci), 0.25% (w/v) blocking reagent for nucleic acid hybridization and detection (Roche) for 7 min at 80°C on heat block, followed by overnight incubation at room temperature. Slides were washed sequentially with 70% formamide (Ambion, Life Technologies, Grand Island, NY) / 0.6 x SSC (Invitrogen) (2 x 1 h), 2 x SSC (1 x 15 min), PBS (1 x 5 min), and sequentially dehydrated with 70%, 90%, and 100% ethanol, then mounted with Vectashield mounting medium with DAPI (Vector Laboratories, Burlingame, CA). Images were captured with Deltavision wide-field microscope using the 60X objective. TIFs were quantified using Image J and representative pictures were prepared in Imaris software after deconvolution using Autoquant X3. Terminal Restriction Fragmental (TRF) Length Analysis for Mean Telomere Length The average length of telomeres (terminal restriction fragment lengths) was measured as described in [50] with the following modifications. DNA was transferred to Hybond-N+ membranes (GE Healthcare, Piscataway, NJ) using vacuum transfer. The membrane was air-dried and DNA was fixed by UV-crosslinking. Membranes were then probed for telomeres using a DIG-labeled telomere probe [51], detected with an HRP-linked anti-DIG antibody (Roche), and exposed with CDP-star (Roche). Supporting Information S1 Fig. Expression of intermediate genes between 5p telomere and hTERT-CLPTM1L locus. (A) Schematic map of relative location of intermediate genes are depicted. mRNA expression in BJ cells at PD34 and PD74, BJ cells with long telomeres (13 kb) and short telomeres (9kb) and IMR 90 cells at PD20 and PD51 were analyzed. RNA (1000 ng) was reverse-transcribed and diluted 1:4 prior to ddPCR and specific probes were used to assess the number of mRNA molecules per reaction. Data are presented as means and standard errors of biological replicates and technical triplicates (6 data points). Student’s paired T test determined significance. *p<0.05. (B) Western blot of CLPTM1L with total histone H3 as loading control at various population doublings in BJ fibroblasts and with expression of hTERT cDNA at different doublings post addition of hTERT. (C) RT-PCR analysis of TERT 5’UTR/exon 1 and exons 5 through 9 in young (long telomere) and old (short telomere) fibroblasts. We included human H9 stem cells as a telomerase positive control. This is a qualitative analysis only as 55 cycles of PCR were performed to detect adequate levels of hTERT transcripts in young BJ cells so we could visualize them on a gel. Quantification was performed using droplet digital PCR shown in Fig 1. Data associated with this figure can be found in the supplemental data file (S1 Data). https://doi.org/10.1371/journal.pbio.2000016.s001 (TIFF) S2 Fig. Location of 3D-FISH probes against hTERT. Intermediate, and sub-telomere loci are also described on the map. Box-and-whisker plots showing single allele representation of distance between probes in 3D-FISH experiments. (A) Average distance between probes against hTERT locus and sub-telomeric region 5p was assessed in normal BJ cells at PD20 and PD90. Adjacent allele (A) and separated allele (S) were visually determined, and the distance was assessed by Imaris software. (B) Average distance between probes against hTERT locus and sub-telomeric region 5p was also assessed in cloned BJ cells with different telomere lengths. The proximity of allele pairs was determined visually and quantitated. (C) IMR90 young cell 3D FISH quantification as above with representative micrograph, scale bar = 2 microns. (D) IMR90 old cell 3D FISH quantification with representative micrograph, scale bar = 3 microns. (E) SW39 and SW26 SV40 large T antigen transformed cell 3D FISH quantification with representative micrograph, scale bar = 3 microns. (F) Long and short telomere BJ cells stained with telomere probe (green), nuclear DNA probe (DAPI, blue) and DNA damage (gH2A.X, red) in cells that were treated with 100 mg/mL of zeocin for 48 hrs or not (control). Scale bar = 5 microns. Data associated with this figure can be found in the supplemental data file (S1 Data). https://doi.org/10.1371/journal.pbio.2000016.s002 (TIFF) S3 Fig. Difference of conformation are restricted between the hTERT and the sub-telomeric 5p. (A) Green fluorescent probe against intermediate region (RP11-846K3) between the sub-telomeric 5p and the hTERT locus was selected as a control. Red fluorescent probe stained sub-telomeric 5p region. (B) Representative deconvolved image shows no conformation change in genome structure between sub-telomeric 5p and RP11-846K3. (C) Box-and-whisker plots showing average distance between two probes assessed by Imaris software. (D) Two fluorescent probes against intermediate region on chromosome 5p (RP11-162J5: Green, RP11-5H14: Red) were selected as a control. Green and red probes are 25.5MB and 30.6MB apart from telomere respectively. (E) Representative deconvolved image shows no conformation change in genome structure between two control loci. (F) Box-and-whisker plots showing average distance between two probes assessed by Imaris software. Data associated with this figure can be found in the supplemental data file (S1 Data). https://doi.org/10.1371/journal.pbio.2000016.s003 (TIFF) S4 Fig. ChIP analysis of TERT promoter. ChIP was performed as in Fig 3. Data are presented as means and standard errors of biological and technical duplicates. Student’s paired T test determine significance. *p<0.05. Data associated with this figure can be found in the supplemental data file (S1 Data). https://doi.org/10.1371/journal.pbio.2000016.s004 (TIFF) S5 Fig. Validation of genome editing at chromosome 5p. (A) A Taq-man probe was designed to bind next to sgRNA target sequence. PCR amplification of flanking sequences hydrolyzes the probe to emit positive signals. (B) ddPCR amplification of 5p end region was performed with genomic DNA from Cas9-infected cells. The number of positive signals shows the approximate level of intact 5p end structure. (C) Metaphase spread analysis of Cas9-infected cells shows telomere signals at the end of chromosome 5p. 21% of chromosomes showed two telomere signals at 5p ends, while 79% of chromosomes lost at least one signal in Cas9-infected cells. Data associated with this figure can be found in the supplemental data file (S1 Data). https://doi.org/10.1371/journal.pbio.2000016.s005 (TIFF) S6 Fig. Lack of telomerase activity in BJ cells with p21 knockdown. (A) BJ cells with long and short telomeres had p21 knocked-down with shRNAs. Telomerase enzyme activity (TRAP- gel based) and (B) Droplet-digital TRAP were performed and no telomerase activity was detected above background. ITAS = internal telomerase activity standard. Data associated with this figure can be found in the supplemental data file (S1 Data). https://doi.org/10.1371/journal.pbio.2000016.s006 (TIFF) S1 Data. Data. https://doi.org/10.1371/journal.pbio.2000016.s007 (XLSX) Acknowledgments We thank for S.B. Kim, L. Zhang, and J. Peters-Hall for plasmids and cell lines; B. Holohan, and K. Batten for critical discussions; C. Cornelius, S. Barron, and M. Coquelin for technical assistance; the Live Cell Imaging Facility at UT Southwestern Medical Center for imaging analyses; the McDermott DNA sequencing core at UT Southwestern Medical Center. We thank Dr. Jay Schneider and Dr. Sean Goetsch for the kind gift of RNA from human stem cells.

doi: 10.1371/journal.pbio.2000631pmid: 28027306

Introduction Antibiotics are indispensable for fighting bacterial infections. Yet the rapid emergence of resistance during the last decades renders current drugs increasingly ineffective and poses a serious threat to human health [1]. Drug action and bacterial resistance mechanisms are well understood in population assays of isogenic cultures in vitro. However, ecological factors and cell physiological parameters in natural environments influence the impact of antibiotics [2,3]. Streptococcus pneumoniae (pneumococcus) is an important human pathogen that resides in complex and dynamic host environments. The bacterium primarily populates the nasopharynx of healthy individuals, together with numerous commensal microbiota, and often alongside disease-associated species, including Staphylococcus aureus, Moraxella catarrhalis, and Haemophilus influenzae [4–6]. While an individual pneumococcal cell competes for limited resources with all other bacteria present in the niche, it may also benefit from a community setting. In a collective effort, bacteria become recalcitrant to antibiotics when forming biofilms that represent a physical constraint for drug accessibility [7,8]. Additional population-based survival strategies involve the phenotypic diversification of an isogenic population, either to preadapt for environmental changes (bet-hedging) or to enable division of labor [9]. Because the impact of most antibiotics is growth rate dependent [10–12], a bifurcation into growing and nongrowing cells increases the drug tolerance for the latter fraction, commonly referred to as persisters [13,14]. Cell-to-cell communication represents another way to react to antibiotic inhibition by allowing bacteria to coordinate a common response; S. pneumoniae, for example, activates the developmental process of competence whereupon it may acquire resistance [15–17]. A quorum-sensing mechanism that compromises antibiotic effectiveness was also found in evolved Escherichia coli cultures, in which cells of increased resistance induce drug efflux pumps in susceptible cells via the signaling molecule indole [18]. As an alternative to reduced drug susceptibility, bacteria can also clear lethal doses of antibiotics from their environment. High cell densities and thus the presence of many drug target sites may be sufficient to lower the concentration of active compound by titration of free drug molecules [19]. Furthermore, antibiotic degradation via β-lactamase enables growth not only of resistant cells but also of susceptible cells in their vicinity [20–22], even across species, as demonstrated for amoxicillin-resistant H. influenzae and susceptible S. pneumoniae [23,24]. This mechanism is of direct relevance to clinical medicine and is alternatively referred to as passive or indirect resistance (from the perspective of susceptible cells) or collective resistance (from the perspective of mixed populations) [25]. Here, we describe another mechanism by which bacteria survive antibiotic therapy without obtaining genetic resistance, with the example of the bacteriostatic antibiotic chloramphenicol (Cm) and the opportunistic human pathogen S. pneumoniae. We show that Cm-resistant pneumococci expressing the resistance factor Cm acetyltransferase (CAT) can provide passive resistance for Cm-susceptible pneumococci by intracellular antibiotic deactivation. CAT covalently attaches an acetyl group from acetyl coenzyme A (acetyl-CoA) to Cm [26,27] and thus prevents the drug from binding to bacterial ribosomes [28]. Intracellular CAT in resistant bacteria can potently detoxify an entire environment in growth culture, semisolid surfaces of microscopy slides, or in a mouse infection model, supporting the survival and growth of genetically susceptible bacteria in the presence of initially effective Cm concentrations. Our results expand recent findings on the basis of E. coli growth cultures and indicate a potential clinical relevance of passive Cm resistance [29,30]. Results Antibiotic Resistance of the Pneumococcus Resistances to all currently prescribed antibiotics have been identified in clinical isolate strains of S. pneumoniae [31]. Genes that transfer antibiotic resistance can be classified according to their mode of action [32]. One class keeps the cytoplasmic drug level low by preventing drug entry or by exporting drug molecules. Another class alters the targeted enzymes by modifying their drug binding sites or by replacing the entire functional unit. A third class alters the drug molecules themselves. Only members of the latter group are potential candidates for establishing passive resistance. In the pneumococcus, resistance genes that deactivate antibiotics include aminoglycoside phosphor- or acetyltransferases and cat. To date, β-lactam antibiotic-degrading enzymes have not been reported in S. pneumoniae genomes or plasmids [33]. Standard therapy of pneumococcal infections does not include aminoglycosides because of the relatively high intrinsic resistance of S. pneumoniae to members of this antibiotic family. In contrast, Cm, a member of the World Health Organization Model List of Essential Medicines [34], is regularly prescribed throughout low-income countries for infections with S. pneumoniae and other Gram-positive pathogens due to its broad spectrum, oral availability, and excellent tissue distribution, including the central nervous system. Recently, the antibiotic was also discussed as candidate for a comeback in developed nations due to spreading resistances against first-line agents [35–37]. To test whether passive resistance emerges from antibiotic-deactivating resistance markers with S. pneumoniae, we used the drug-susceptible clinical isolate D39 [38]. We constructed an antibiotic-susceptible reporter strain expressing firefly luciferase (luc) and antibiotic-resistant strains expressing single-copy genomic integrated kanamycin 3′-phosphotransferase (aphA1), gentamicin 3′-acetyltransferase (aacC1), and chloramphenicol acetyltransferase (cat). Resistant and susceptible cells were grown at a one-to-one ratio, and optical density (both strains) and bioluminescence (emitted by susceptible cells only) were measured (Fig 1). Expression of cat, but not aphA1 or aacCI, conferred passive resistance to susceptible cells (as observed by increased luminescence in mixed populations compared with assays of susceptible cells only; S1 Fig), mirroring prior investigations of antibiotic deactivation by resistant isolates of S. pneumoniae [39]. Aminoglycosides permeate the bacterial cell only at low frequency [40]; high permeability, however, was recently shown to represent an important precondition for the establishment of passive resistance, explaining why the phenomenon could not be observed with aphA1 and aacCI expression [29]. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Experimental setup to determine passive resistance. Antibiotic-susceptible cells (AbS) constitutively expressing luc are grown together with antibiotic-resistant cells (AbR, which do not express luc). Only when the concentration of the antibiotic in the medium is reduced by enzymatic deactivation of resistant cells will the genetically antibiotic-susceptible cells be able to grow and produce light. https://doi.org/10.1371/journal.pbio.2000631.g001 Collective Resistance to Cm In Vitro To characterize the observed Cm collective resistance in more detail, we used the Cm-susceptible strain D-PEP2K1 (from here on CmS), which constitutively expresses luc and the kanamycin resistance marker aphA1 [41], and the Cm-resistant strain D-PEP1-pJS5 (from here on CmR), which expresses cat from plasmid pJS5 [42] (see Methods). Luminescence allowed for the real-time estimation of growth (or inhibition) of the CmS population, and kanamycin resistance allowed for the monitoring of their viable cell count by plating assays in the presence of kanamycin. Cm represses the growth of susceptible pneumococci at a minimal inhibitory concentration (MIC) of 2.2 μg ml−1, and during Cm exposure, luminescence from luc expression of susceptible pneumococci was previously shown to decrease at a rate that depends on the applied Cm concentration [12]. However, when CmS was co-inoculated with CAT-expressing CmR, luminescence (indicative for growth or inhibition of the CmS cell fraction) recovered, both for a Cm concentration slightly above the MIC (3 μg ml−1; Fig 2A) and even for a Cm concentration of more than two times the MIC (5 μg ml−1; S2 Fig). Luminescence recovery in mixed population assays (CmR + CmS) exceeded the values measured with CmS monoculture by up to 10-fold (Fig 2A and S2 Fig), and plating assays (with kanamycin) revealed that the difference in viable cell count was 1,000-fold greater after 8 h of cocultivation (Fig 2B and S2 Fig). Although Cm is commonly regarded as bacteriostatic, bactericidal activity has also been demonstrated against S. pneumoniae [43], explaining the observed decrease in viability of CmS monoculture (Fig 2B and S2 Fig). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Cm deactivation during mixed population assays. (a) Plate reader assay sets in quadruplicate (average and standard error of the mean [s.e.m.]) measuring luminescence (symbols with color outline) and cell density (corresponding grey symbols) of S. pneumoniae CmS growing in the presence of 3 μg ml−1 Cm, in presence (+) or absence (−) of CmR cells. (b) Development of the count of viable CmS cells (colony-forming units per ml [CFUs ml−1]) during the cultivation assay presented in a, determined via plating in the presence of kanamycin; average values of duplicates are shown. (c) Culture supernatant (S) samples after 0, 1, 2, and 4 h of CmR cultivation (inoculation at optical density OD 0.001) in the presence of 5 μg ml−1 Cm, analyzed for Cm content by high-performance liquid chromatography (HPLC) separation and ultraviolet (UV) detection at 278 nm. (d) Luminescence and cell density profiles of CmS cells treated with 3 μg ml−1 Cm (inoculation at OD 0.001) in dependency of the inoculum size of CmR cells. (e, f), CmS luminescence and growth analysis (e) in Cm-supplemented medium (3 μg ml−1) that was pretreated with CmR cell pellet (P), S, and culture lysate (L), and controls without (C−) and with Cm (C+); (f) schematic overview of the assay (see also Methods and S1 Data). https://doi.org/10.1371/journal.pbio.2000631.g002 To confirm that CmR cells actually deactivate Cm in the growth medium, we analyzed culture supernatant (S) by high-performance liquid chromatography (HPLC) [44]. As shown in Fig 2C, within 4 h of growth, CmR cells entirely converted an initial Cm concentration of 5 μg ml−1, as evidenced by the disappearance of the corresponding Cm peak at wavelength 278 nm. New peaks (at later elution times) appeared and gradually increased in HPLC profiles of S collected after 1, 2, and 4 h of cultivation; these peaks were previously shown to correspond to acetylated Cm derivates (1- and 3-acetylchloramphenicol) [44]. Next, we focused on whether the initial amount of CAT-expressing CmR cells was important for the survival and growth of CmS cells during drug treatment. To test this, we inoculated microtiter plate wells with a fixed number of CmS cells (inoculation at optical density [OD] 0.001, corresponding to ~1.5 × 106 colony-forming units per ml [CFUs ml−1]) while varying the number of CmR cells (Fig 2D). High inoculation densities of CmR cells (OD 0.01) resulted in a fast recovery of luminescence activity of CmS cells; however, the peak of luminescence was lower compared to intermediate CmR inoculation densities. This difference can be explained by cells reaching the carrying capacity of the growth medium before the pool of Cm is completely deactivated; luciferase expression activity was previously shown to slow down when cultures reach high cell densities (above ~OD 0.05) [41]. Relatively low CmR inoculation densities (OD 0.0001) also limited luminescence recovery of CmS cells during cocultivation. This finding likely reflects fewer CmR cells requiring more time to deactivate Cm, resulting in increased time spans of CmS drug exposure. Prolonged drug exposure of susceptible pneumococci was previously shown to result in increasing lag periods after drug removal, indicating a more severe perturbation of cell homeostasis [12]. The time span before outgrowth of CmS cells consequently consists of both the period required for drug clearance (by CmR cells) and the period required to reestablish intracellular conditions allowing for cell division. Intracellular Deactivation of Cm To test whether Cm processing by CAT is an intracellular process, or if it takes place after secretion or cell lysis, we examined the potential of the S and the cytosolic content of CmR cells to deactivate Cm (assay scheme in Fig 2F). Precultured CmR cells were diluted to OD 0.02 and translation activity was blocked by adding 1 μg ml−1 tetracycline ([Tc]; S. pneumoniae D39 MIC: 0.26 μg ml−1) [12] for 1 h at 37°C to prevent ongoing protein synthesis and thus CAT expression. Next, the Tc-treated culture was split into three fractions: cell pellet (P) and S, separated via centrifugation, and cell culture lysate (L), obtained by sonication. The P was resuspended in C+Y medium containing 3 μg ml−1 Cm (and 1 μg ml−1 Tc), and 3 μg ml−1 Cm was added to the S and the L, followed by incubation at 37°C. After 2 h, the remaining cells and cell debris were removed by centrifugation and filtration, and the treated medium was used to test cell growth of a Tc-resistant variant of the CmS strain. Neither the S nor the L could support growth of CmS, whereas medium preincubated with the P did (Fig 2E). Together, these experiments indicate that CAT is only active inside living cells, in which acetyl-CoA is present [26,27]. Single-Cell Observations of Collective Resistance Because the abovementioned experiments were performed in bulk assays, we wondered whether CAT-expressing bacteria would also efficiently deactivate Cm, and thus support the growth of susceptible cells, in a more complex environment, such as on semi-solid surfaces. To do so, we spotted CmR cells together with Cm-susceptible D-PEP33 cells expressing green fluorescent protein (GFP) on a matrix of 10% polyacrylamide C+Y medium containing 3 μg ml−1 Cm. Indeed Cm-susceptible D-PEP33 cells were able to grow and divide under these conditions (S3 Fig). S. pneumoniae cohabitates the human nasopharynx with other bacteria, such as S. aureus [6]. Therefore, we investigated whether CAT-expressing S. aureus could also support growth of Cm-susceptible S. pneumoniae in environments containing Cm. As shown in Fig 3 and S1 Movie, all S. aureus cells grew and divided from the starting point of the experiment, whereas S. pneumoniae CmS cells did not grow initially. However, after 8 h, a fraction of CmS cells grew out to form microcolonies. Note that CmS cells spotted in the absence of S. aureus did not grow under these conditions (S2 Movie). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Interspecies collective resistance. Still images (overlay of phase contrast and fluorescence microscopy) of a time-lapse experiment of S. pneumoniae CmS, cocultivated with a strain of the pneumococcal niche competitor S. aureus (strain LAC pCM29) that expresses CAT and GFP, growing on a semi-solid surface supplemented with 3 μg ml−1 Cm. Scale bar 10 μm. https://doi.org/10.1371/journal.pbio.2000631.g003 Requirements for Stable Coexistence The observation that CmS cells grow only when Cm-deactivating cells are present in their close vicinity (Fig 3) suggests that the establishment of collective resistance requires CmS and CmR bacteria to be present in the same niche. However, such coexistence is subject to ecological constraints (e.g., the competitive exclusion principle) [45], particularly if susceptible and resistant strains compete for the same limiting resource. We therefore developed an ecological model to assess the scope for coexistence between CAT-producing bacteria and an antibiotic-susceptible strain (S1 Text). Consistent with this objective, we employed a minimalist modeling strategy and disentangled the qualitative effects of different factors (antibiotic stress, relative cost of Cm degradation and density regulation by ecological resource competition) from the interaction between CmS and CmR bacteria rather than aiming for a precise quantitative reconstruction of the experimental conditions. In fact, in contrast to natural environments (such as the human nasopharynx) that provide ample opportunities for coexistence because of spatial structure and concentration gradients of multiple resources, the model considers a worst-case scenario for coexistence: the two populations are assumed to grow in a well-mixed, homogeneous chemostat environment and are limited by the same resource. Nonetheless, we found that coexistence between CmR and CmS bacteria was feasible (Fig 4A and 4B), albeit under a restricted range of conditions (Fig 4C and S4 Fig). A mathematical analysis of the model (S1 Text) indicates that resistant and susceptible bacteria can establish a stable coexistence when CAT expression has a modest fitness cost. Without such a cost, the CmR strain is predicted to outcompete the CmS strain in the presence of antibiotics. Conversely, if the cost of expressing resistance is too high, the CmS strain will be the superior competitor. Interestingly, the model furthermore predicts parameter ranges that result in the extinction of mixed populations during drug treatment, while CmR populations on their own could survive (S4 and S5 Figs). A second condition for coexistence demands that the CmR population has a significant impact on the extracellular Cm concentration in its ecological niche. This requires that the population density reached at steady state must be high, so that coexistence can be stabilized by frequency-dependent selection, generated by a negative feedback loop between the relative abundance of drug-deactivating cells and the level of antibiotic stress in the environment. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. Population dynamics of bacterial communities. (a) Simulated growth trajectories for CmR and CmS populations subject to antibiotic stress and resource competition. (b) Dynamic of intracellular Cm (yr and ys) and growth-limiting resource (z). Simulation time is scaled relative to the mean residence time of cells in a chemostat, which is equal to the generation time at steady state. At low population densities, the CmR strain can grow, whereas CmS cannot, due to a high concentration of Cm. However, the invasion of CmR lowers antibiotic stress, generating permissive conditions for the growth of CmS cells. The chemostat is then rapidly colonized by both strains (shortly after t = 180) until the resource becomes limiting. From that moment onwards, total cell density changes little, while the relative frequencies of the two strains continue to shift. Eventually, a stable equilibrium is reached, at which the cost and benefit of CAT expression (i.e., reduced growth rate efficiency for CmR cells versus their lower intracellular Cm concentration) balance out. Inset (c), The dark-red dot pinpoints the parameter set used in the simulation shown in a and b: r = 20.0, η = 0.9, kz = 4.0, c = 1.0, p = 50.0, hY = 0.25/Y0, kY = 2.5/Y0, d = 30.0/Y0 and Y0 = 0.8. These parameters were selected to lie in a restricted area of parameter space (highlighted in red) where stable coexistence between CmS and CmR cells is observed Alternative model outcomes, which were identified by a numerical bifurcation analysis (see S1 Text and S4 Fig), include establishment of CmS only (area S), establishment of CmR only (area R), no bacterial growth (area N), and competition-induced extinction (area E, where CmS bacteria first outcompete CmR bacteria and subsequently are cleared by the antibiotic; see S5 Fig). https://doi.org/10.1371/journal.pbio.2000631.g004 We note that competitive exclusion acts at a local scale in structured environments, where the presence of spatial gradients in Cm and resources may help to create refuges in which either strain can escape competition from the other. In addition, we expect that coexistence between resistant and susceptible bacteria would be promoted in vivo by previously evolved ecological niche partitioning between co-occurring species. In Vivo Collective Cm Resistance A general prediction from our mathematical model (S1 Text) is that coexistence of CmS and CmR in the presence of the antibiotic is precluded when the production of CAT carries no fitness cost; we expect this prediction to apply likewise in more complex environments with spatial and/or temporal heterogeneity in Cm concentrations. However, in vitro, in short-term experiments, we did not observe any obvious fitness cost for CAT expression (such as reduced growth rates or a reduced maximum cell density; Fig 2). Nevertheless, in vivo, a fitness cost might come into existence because resistant cells that grow rapidly might be preferentially targeted by the host innate immune system, as previously shown for commensal and pathogenic bacteria, including E. coli and S. aureus [46]. We tested the activity of the human antimicrobial peptide LL-37 in dependency of Cm treatment and found, indeed, increased killing efficiency against CAT-expressing S. pneumoniae (S6 Fig). Furthermore, although collective resistance could be successfully demonstrated in vitro, the phenomenon might not occur in more complex environments in vivo, such as in an animal infection model, because of a different flux balance between local Cm deactivation and restoration of effective drug concentrations via diffusion from surrounding tissues. To examine whether a coexistence between CmS and CmR is possible under therapy in vivo, we performed intratracheal infection of 8-wk-old female CD-1 mice with CmS alone and the combination of CmS + CmR. In the absence of Cm treatment, we observed no significant difference in the amount of viable bacteria recovered from the lungs 24 h after infection with CmS alone versus CmS + CmR at a one-to-one ratio (Fig 5A). When mice were given three doses of 75 mg kg−1 Cm once every 5 h, mice infected with CmS alone demonstrated a significant drop of one log-fold versus the untreated control. This is in contrast to mice coinfected with CmS + CmR, in which Cm treatment did not significantly reduce the number of viable bacteria recovered from the lung versus the untreated control (Fig 5A). In the one-to-one mixed infection, approximately equal numbers of CmS and CmR cells were recovered in the absence of Cm treatment: 46% CmS and 54% CmR (Fig 5B). Surprisingly, with Cm treatment, 6 out of 14 animals had a dramatic increase in the percentage of CmS cells versus CmR cells. No pneumococcal colonies recovered could grow in both Cm- and kanamycin-containing media, excluding the possibility that horizontal gene transfer of the cat gene occurred during coinfection. Together, these results show that passive Cm resistance and the coexistence of resistant and susceptible cells also occur in vivo, associated with a fitness cost to the CmR niche members benefiting the CmS subpopulation. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. Cross-protection in a mouse pneumonia model. (a) Eight-wk-old female CD1 mice were infected intratracheally with CmS pneumococci or an equivalent amount of CmS + CmR pneumococci in a one-to-one ratio. One h post infection, mice were treated with one intraperitoneal injection of Cm 75 mg kg−1 followed by two additional doses spaced 5 h apart. Control mice received an injection of the vehicle alone. n = 14 for CmS control; 13 for CmS Cm-treated; 13 for CmS + CmR control; and 14 for CmS + CmR Cm-treated. Data plotted as average and s.e.m. of two independent experiments combined. Dashed line ‘inoc’ denotes the initial inoculum. *p < 0.05; one-way ANOVA with Tukey's multiple comparison post-test. (b) Bacterial colonies recovered from the CmS + CmR control and CmS + CmR Cm-treated mice were individually picked and used to inoculate THY media in 96-well plates. These 96-well plates were then used to inoculate 96-well plates with THY media containing either 15 μg ml−1 Cm or 100 μg ml−1 kanamycin to determine whether or not the original bacterial colony was CmS or CmR. n = 9 for CmS + CmR control and 14 for CmS + CmR Cm-treated. Data plotted as average and s.e.m. of two independent experiments combined. *p = 0.04; Mann–Whitney U test (see S1 Data). https://doi.org/10.1371/journal.pbio.2000631.g005 Antibiotic Resistance of the Pneumococcus Resistances to all currently prescribed antibiotics have been identified in clinical isolate strains of S. pneumoniae [31]. Genes that transfer antibiotic resistance can be classified according to their mode of action [32]. One class keeps the cytoplasmic drug level low by preventing drug entry or by exporting drug molecules. Another class alters the targeted enzymes by modifying their drug binding sites or by replacing the entire functional unit. A third class alters the drug molecules themselves. Only members of the latter group are potential candidates for establishing passive resistance. In the pneumococcus, resistance genes that deactivate antibiotics include aminoglycoside phosphor- or acetyltransferases and cat. To date, β-lactam antibiotic-degrading enzymes have not been reported in S. pneumoniae genomes or plasmids [33]. Standard therapy of pneumococcal infections does not include aminoglycosides because of the relatively high intrinsic resistance of S. pneumoniae to members of this antibiotic family. In contrast, Cm, a member of the World Health Organization Model List of Essential Medicines [34], is regularly prescribed throughout low-income countries for infections with S. pneumoniae and other Gram-positive pathogens due to its broad spectrum, oral availability, and excellent tissue distribution, including the central nervous system. Recently, the antibiotic was also discussed as candidate for a comeback in developed nations due to spreading resistances against first-line agents [35–37]. To test whether passive resistance emerges from antibiotic-deactivating resistance markers with S. pneumoniae, we used the drug-susceptible clinical isolate D39 [38]. We constructed an antibiotic-susceptible reporter strain expressing firefly luciferase (luc) and antibiotic-resistant strains expressing single-copy genomic integrated kanamycin 3′-phosphotransferase (aphA1), gentamicin 3′-acetyltransferase (aacC1), and chloramphenicol acetyltransferase (cat). Resistant and susceptible cells were grown at a one-to-one ratio, and optical density (both strains) and bioluminescence (emitted by susceptible cells only) were measured (Fig 1). Expression of cat, but not aphA1 or aacCI, conferred passive resistance to susceptible cells (as observed by increased luminescence in mixed populations compared with assays of susceptible cells only; S1 Fig), mirroring prior investigations of antibiotic deactivation by resistant isolates of S. pneumoniae [39]. Aminoglycosides permeate the bacterial cell only at low frequency [40]; high permeability, however, was recently shown to represent an important precondition for the establishment of passive resistance, explaining why the phenomenon could not be observed with aphA1 and aacCI expression [29]. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Experimental setup to determine passive resistance. Antibiotic-susceptible cells (AbS) constitutively expressing luc are grown together with antibiotic-resistant cells (AbR, which do not express luc). Only when the concentration of the antibiotic in the medium is reduced by enzymatic deactivation of resistant cells will the genetically antibiotic-susceptible cells be able to grow and produce light. https://doi.org/10.1371/journal.pbio.2000631.g001 Collective Resistance to Cm In Vitro To characterize the observed Cm collective resistance in more detail, we used the Cm-susceptible strain D-PEP2K1 (from here on CmS), which constitutively expresses luc and the kanamycin resistance marker aphA1 [41], and the Cm-resistant strain D-PEP1-pJS5 (from here on CmR), which expresses cat from plasmid pJS5 [42] (see Methods). Luminescence allowed for the real-time estimation of growth (or inhibition) of the CmS population, and kanamycin resistance allowed for the monitoring of their viable cell count by plating assays in the presence of kanamycin. Cm represses the growth of susceptible pneumococci at a minimal inhibitory concentration (MIC) of 2.2 μg ml−1, and during Cm exposure, luminescence from luc expression of susceptible pneumococci was previously shown to decrease at a rate that depends on the applied Cm concentration [12]. However, when CmS was co-inoculated with CAT-expressing CmR, luminescence (indicative for growth or inhibition of the CmS cell fraction) recovered, both for a Cm concentration slightly above the MIC (3 μg ml−1; Fig 2A) and even for a Cm concentration of more than two times the MIC (5 μg ml−1; S2 Fig). Luminescence recovery in mixed population assays (CmR + CmS) exceeded the values measured with CmS monoculture by up to 10-fold (Fig 2A and S2 Fig), and plating assays (with kanamycin) revealed that the difference in viable cell count was 1,000-fold greater after 8 h of cocultivation (Fig 2B and S2 Fig). Although Cm is commonly regarded as bacteriostatic, bactericidal activity has also been demonstrated against S. pneumoniae [43], explaining the observed decrease in viability of CmS monoculture (Fig 2B and S2 Fig). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Cm deactivation during mixed population assays. (a) Plate reader assay sets in quadruplicate (average and standard error of the mean [s.e.m.]) measuring luminescence (symbols with color outline) and cell density (corresponding grey symbols) of S. pneumoniae CmS growing in the presence of 3 μg ml−1 Cm, in presence (+) or absence (−) of CmR cells. (b) Development of the count of viable CmS cells (colony-forming units per ml [CFUs ml−1]) during the cultivation assay presented in a, determined via plating in the presence of kanamycin; average values of duplicates are shown. (c) Culture supernatant (S) samples after 0, 1, 2, and 4 h of CmR cultivation (inoculation at optical density OD 0.001) in the presence of 5 μg ml−1 Cm, analyzed for Cm content by high-performance liquid chromatography (HPLC) separation and ultraviolet (UV) detection at 278 nm. (d) Luminescence and cell density profiles of CmS cells treated with 3 μg ml−1 Cm (inoculation at OD 0.001) in dependency of the inoculum size of CmR cells. (e, f), CmS luminescence and growth analysis (e) in Cm-supplemented medium (3 μg ml−1) that was pretreated with CmR cell pellet (P), S, and culture lysate (L), and controls without (C−) and with Cm (C+); (f) schematic overview of the assay (see also Methods and S1 Data). https://doi.org/10.1371/journal.pbio.2000631.g002 To confirm that CmR cells actually deactivate Cm in the growth medium, we analyzed culture supernatant (S) by high-performance liquid chromatography (HPLC) [44]. As shown in Fig 2C, within 4 h of growth, CmR cells entirely converted an initial Cm concentration of 5 μg ml−1, as evidenced by the disappearance of the corresponding Cm peak at wavelength 278 nm. New peaks (at later elution times) appeared and gradually increased in HPLC profiles of S collected after 1, 2, and 4 h of cultivation; these peaks were previously shown to correspond to acetylated Cm derivates (1- and 3-acetylchloramphenicol) [44]. Next, we focused on whether the initial amount of CAT-expressing CmR cells was important for the survival and growth of CmS cells during drug treatment. To test this, we inoculated microtiter plate wells with a fixed number of CmS cells (inoculation at optical density [OD] 0.001, corresponding to ~1.5 × 106 colony-forming units per ml [CFUs ml−1]) while varying the number of CmR cells (Fig 2D). High inoculation densities of CmR cells (OD 0.01) resulted in a fast recovery of luminescence activity of CmS cells; however, the peak of luminescence was lower compared to intermediate CmR inoculation densities. This difference can be explained by cells reaching the carrying capacity of the growth medium before the pool of Cm is completely deactivated; luciferase expression activity was previously shown to slow down when cultures reach high cell densities (above ~OD 0.05) [41]. Relatively low CmR inoculation densities (OD 0.0001) also limited luminescence recovery of CmS cells during cocultivation. This finding likely reflects fewer CmR cells requiring more time to deactivate Cm, resulting in increased time spans of CmS drug exposure. Prolonged drug exposure of susceptible pneumococci was previously shown to result in increasing lag periods after drug removal, indicating a more severe perturbation of cell homeostasis [12]. The time span before outgrowth of CmS cells consequently consists of both the period required for drug clearance (by CmR cells) and the period required to reestablish intracellular conditions allowing for cell division. Intracellular Deactivation of Cm To test whether Cm processing by CAT is an intracellular process, or if it takes place after secretion or cell lysis, we examined the potential of the S and the cytosolic content of CmR cells to deactivate Cm (assay scheme in Fig 2F). Precultured CmR cells were diluted to OD 0.02 and translation activity was blocked by adding 1 μg ml−1 tetracycline ([Tc]; S. pneumoniae D39 MIC: 0.26 μg ml−1) [12] for 1 h at 37°C to prevent ongoing protein synthesis and thus CAT expression. Next, the Tc-treated culture was split into three fractions: cell pellet (P) and S, separated via centrifugation, and cell culture lysate (L), obtained by sonication. The P was resuspended in C+Y medium containing 3 μg ml−1 Cm (and 1 μg ml−1 Tc), and 3 μg ml−1 Cm was added to the S and the L, followed by incubation at 37°C. After 2 h, the remaining cells and cell debris were removed by centrifugation and filtration, and the treated medium was used to test cell growth of a Tc-resistant variant of the CmS strain. Neither the S nor the L could support growth of CmS, whereas medium preincubated with the P did (Fig 2E). Together, these experiments indicate that CAT is only active inside living cells, in which acetyl-CoA is present [26,27]. Single-Cell Observations of Collective Resistance Because the abovementioned experiments were performed in bulk assays, we wondered whether CAT-expressing bacteria would also efficiently deactivate Cm, and thus support the growth of susceptible cells, in a more complex environment, such as on semi-solid surfaces. To do so, we spotted CmR cells together with Cm-susceptible D-PEP33 cells expressing green fluorescent protein (GFP) on a matrix of 10% polyacrylamide C+Y medium containing 3 μg ml−1 Cm. Indeed Cm-susceptible D-PEP33 cells were able to grow and divide under these conditions (S3 Fig). S. pneumoniae cohabitates the human nasopharynx with other bacteria, such as S. aureus [6]. Therefore, we investigated whether CAT-expressing S. aureus could also support growth of Cm-susceptible S. pneumoniae in environments containing Cm. As shown in Fig 3 and S1 Movie, all S. aureus cells grew and divided from the starting point of the experiment, whereas S. pneumoniae CmS cells did not grow initially. However, after 8 h, a fraction of CmS cells grew out to form microcolonies. Note that CmS cells spotted in the absence of S. aureus did not grow under these conditions (S2 Movie). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Interspecies collective resistance. Still images (overlay of phase contrast and fluorescence microscopy) of a time-lapse experiment of S. pneumoniae CmS, cocultivated with a strain of the pneumococcal niche competitor S. aureus (strain LAC pCM29) that expresses CAT and GFP, growing on a semi-solid surface supplemented with 3 μg ml−1 Cm. Scale bar 10 μm. https://doi.org/10.1371/journal.pbio.2000631.g003 Requirements for Stable Coexistence The observation that CmS cells grow only when Cm-deactivating cells are present in their close vicinity (Fig 3) suggests that the establishment of collective resistance requires CmS and CmR bacteria to be present in the same niche. However, such coexistence is subject to ecological constraints (e.g., the competitive exclusion principle) [45], particularly if susceptible and resistant strains compete for the same limiting resource. We therefore developed an ecological model to assess the scope for coexistence between CAT-producing bacteria and an antibiotic-susceptible strain (S1 Text). Consistent with this objective, we employed a minimalist modeling strategy and disentangled the qualitative effects of different factors (antibiotic stress, relative cost of Cm degradation and density regulation by ecological resource competition) from the interaction between CmS and CmR bacteria rather than aiming for a precise quantitative reconstruction of the experimental conditions. In fact, in contrast to natural environments (such as the human nasopharynx) that provide ample opportunities for coexistence because of spatial structure and concentration gradients of multiple resources, the model considers a worst-case scenario for coexistence: the two populations are assumed to grow in a well-mixed, homogeneous chemostat environment and are limited by the same resource. Nonetheless, we found that coexistence between CmR and CmS bacteria was feasible (Fig 4A and 4B), albeit under a restricted range of conditions (Fig 4C and S4 Fig). A mathematical analysis of the model (S1 Text) indicates that resistant and susceptible bacteria can establish a stable coexistence when CAT expression has a modest fitness cost. Without such a cost, the CmR strain is predicted to outcompete the CmS strain in the presence of antibiotics. Conversely, if the cost of expressing resistance is too high, the CmS strain will be the superior competitor. Interestingly, the model furthermore predicts parameter ranges that result in the extinction of mixed populations during drug treatment, while CmR populations on their own could survive (S4 and S5 Figs). A second condition for coexistence demands that the CmR population has a significant impact on the extracellular Cm concentration in its ecological niche. This requires that the population density reached at steady state must be high, so that coexistence can be stabilized by frequency-dependent selection, generated by a negative feedback loop between the relative abundance of drug-deactivating cells and the level of antibiotic stress in the environment. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. Population dynamics of bacterial communities. (a) Simulated growth trajectories for CmR and CmS populations subject to antibiotic stress and resource competition. (b) Dynamic of intracellular Cm (yr and ys) and growth-limiting resource (z). Simulation time is scaled relative to the mean residence time of cells in a chemostat, which is equal to the generation time at steady state. At low population densities, the CmR strain can grow, whereas CmS cannot, due to a high concentration of Cm. However, the invasion of CmR lowers antibiotic stress, generating permissive conditions for the growth of CmS cells. The chemostat is then rapidly colonized by both strains (shortly after t = 180) until the resource becomes limiting. From that moment onwards, total cell density changes little, while the relative frequencies of the two strains continue to shift. Eventually, a stable equilibrium is reached, at which the cost and benefit of CAT expression (i.e., reduced growth rate efficiency for CmR cells versus their lower intracellular Cm concentration) balance out. Inset (c), The dark-red dot pinpoints the parameter set used in the simulation shown in a and b: r = 20.0, η = 0.9, kz = 4.0, c = 1.0, p = 50.0, hY = 0.25/Y0, kY = 2.5/Y0, d = 30.0/Y0 and Y0 = 0.8. These parameters were selected to lie in a restricted area of parameter space (highlighted in red) where stable coexistence between CmS and CmR cells is observed Alternative model outcomes, which were identified by a numerical bifurcation analysis (see S1 Text and S4 Fig), include establishment of CmS only (area S), establishment of CmR only (area R), no bacterial growth (area N), and competition-induced extinction (area E, where CmS bacteria first outcompete CmR bacteria and subsequently are cleared by the antibiotic; see S5 Fig). https://doi.org/10.1371/journal.pbio.2000631.g004 We note that competitive exclusion acts at a local scale in structured environments, where the presence of spatial gradients in Cm and resources may help to create refuges in which either strain can escape competition from the other. In addition, we expect that coexistence between resistant and susceptible bacteria would be promoted in vivo by previously evolved ecological niche partitioning between co-occurring species. In Vivo Collective Cm Resistance A general prediction from our mathematical model (S1 Text) is that coexistence of CmS and CmR in the presence of the antibiotic is precluded when the production of CAT carries no fitness cost; we expect this prediction to apply likewise in more complex environments with spatial and/or temporal heterogeneity in Cm concentrations. However, in vitro, in short-term experiments, we did not observe any obvious fitness cost for CAT expression (such as reduced growth rates or a reduced maximum cell density; Fig 2). Nevertheless, in vivo, a fitness cost might come into existence because resistant cells that grow rapidly might be preferentially targeted by the host innate immune system, as previously shown for commensal and pathogenic bacteria, including E. coli and S. aureus [46]. We tested the activity of the human antimicrobial peptide LL-37 in dependency of Cm treatment and found, indeed, increased killing efficiency against CAT-expressing S. pneumoniae (S6 Fig). Furthermore, although collective resistance could be successfully demonstrated in vitro, the phenomenon might not occur in more complex environments in vivo, such as in an animal infection model, because of a different flux balance between local Cm deactivation and restoration of effective drug concentrations via diffusion from surrounding tissues. To examine whether a coexistence between CmS and CmR is possible under therapy in vivo, we performed intratracheal infection of 8-wk-old female CD-1 mice with CmS alone and the combination of CmS + CmR. In the absence of Cm treatment, we observed no significant difference in the amount of viable bacteria recovered from the lungs 24 h after infection with CmS alone versus CmS + CmR at a one-to-one ratio (Fig 5A). When mice were given three doses of 75 mg kg−1 Cm once every 5 h, mice infected with CmS alone demonstrated a significant drop of one log-fold versus the untreated control. This is in contrast to mice coinfected with CmS + CmR, in which Cm treatment did not significantly reduce the number of viable bacteria recovered from the lung versus the untreated control (Fig 5A). In the one-to-one mixed infection, approximately equal numbers of CmS and CmR cells were recovered in the absence of Cm treatment: 46% CmS and 54% CmR (Fig 5B). Surprisingly, with Cm treatment, 6 out of 14 animals had a dramatic increase in the percentage of CmS cells versus CmR cells. No pneumococcal colonies recovered could grow in both Cm- and kanamycin-containing media, excluding the possibility that horizontal gene transfer of the cat gene occurred during coinfection. Together, these results show that passive Cm resistance and the coexistence of resistant and susceptible cells also occur in vivo, associated with a fitness cost to the CmR niche members benefiting the CmS subpopulation. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. Cross-protection in a mouse pneumonia model. (a) Eight-wk-old female CD1 mice were infected intratracheally with CmS pneumococci or an equivalent amount of CmS + CmR pneumococci in a one-to-one ratio. One h post infection, mice were treated with one intraperitoneal injection of Cm 75 mg kg−1 followed by two additional doses spaced 5 h apart. Control mice received an injection of the vehicle alone. n = 14 for CmS control; 13 for CmS Cm-treated; 13 for CmS + CmR control; and 14 for CmS + CmR Cm-treated. Data plotted as average and s.e.m. of two independent experiments combined. Dashed line ‘inoc’ denotes the initial inoculum. *p < 0.05; one-way ANOVA with Tukey's multiple comparison post-test. (b) Bacterial colonies recovered from the CmS + CmR control and CmS + CmR Cm-treated mice were individually picked and used to inoculate THY media in 96-well plates. These 96-well plates were then used to inoculate 96-well plates with THY media containing either 15 μg ml−1 Cm or 100 μg ml−1 kanamycin to determine whether or not the original bacterial colony was CmS or CmR. n = 9 for CmS + CmR control and 14 for CmS + CmR Cm-treated. Data plotted as average and s.e.m. of two independent experiments combined. *p = 0.04; Mann–Whitney U test (see S1 Data). https://doi.org/10.1371/journal.pbio.2000631.g005 Discussion This work elucidates that CAT, which is commonly found as a resistance marker in the human microbiome [47–49], can effectively protect Cm-susceptible pneumococci from the activity of the drug within local environments occupied by CAT-expressing cells. Because of its potency, long shelf life, and low cost, Cm remains a mainstay of broad-spectrum antibiotic therapy in several countries in Africa, the Indian subcontinent, and China [50]. The rise of multidrug resistance among human pathogens has also provoked interest in reevaluating Cm for certain serious infections in developed countries [35–37]. This work points out some caveats in using Cm to target human pathogens on mucosal surfaces because CAT-expressing commensals might provide passive resistance. CAT can only deactivate Cm inside living cells (Fig 2E and 2F), presumably because it needs acetyl-CoA to acetylate and deactivate the target drug [26,27]. We show that Cm deactivation and collective resistance via CAT is not limited to S. pneumoniae, because CAT-expressing S. aureus can also support the local growth of pneumococci in the presence of initially effective Cm concentrations (Fig 3). Collective resistance by CAT does not only occur in vitro but also in vivo in a mouse pneumonia model (Fig 5). Strikingly, when Cm-treated mice were coinfected with CAT-expressing and Cm-susceptible pneumococci, the susceptible bacteria outcompeted the resistant ones (Fig 5). We previously showed that the susceptibility of bacteria towards antimicrobial peptides, produced by the host innate immune system, is markedly diminished in the presence of bacteriostatic antibiotics; Cm-inhibited E. coli, for example, are less efficiently cleared by the human peptide LL-37 [46]. We could demonstrate that this mechanism also takes place in S. pneumoniae (S6 Fig). When Cm was added to LL-37 treatment of pneumococci, the number of CmS cells recovered was one log-fold higher compared with CmR cells (S6 Fig). As shown before, this effect occurs because bacteriostatic antibiotics, such as Cm, inhibit the growth of susceptible bacteria and thereby reduce the susceptibility to host antimicrobial peptides that target bacterial division; Cm-resistant bacteria, in contrast, maintain fast growth in the presence of Cm and are therefore more rapidly killed by host antimicrobial peptides in vivo [46]. This phenomenon may therefore represent a contributing factor underlying our findings of the mouse pneumonia model. In this framework, rapidly growing CmR cells would suffer immune clearance, while the initially nongrowing CmS are less efficiently targeted by host defense factors. Once the Cm concentration has dropped sufficiently, CmS cells can outgrow and outcompete the diminished CmR population. Our work with CAT and pneumococcus extends the known phenomenon of passive resistance via β-lactamase expression and expands on recent findings of collective resistance of bacterial communities [29,30]. Intracellular antibiotic deactivation requires a high drug permeability, and it is worth noting that this—in general desired—drug characteristic can also represent a risk factor for the effectiveness of an antibiotic therapy. Passive resistance could also appear with other antibiotic-degrading resistance factors in other bacteria [29] and may even emerge for synthetic antibiotic compounds [51]. In light of numerous reports of prevalence of drug resistance in pathogens, successful antibiotic therapy might become increasingly complicated with the occurrence of collective resistance. The phenomena could furthermore give rise to multidrug resistance of bacterial communities, in which individual resistances are expressed in different bacterial community residents [2,18,30]. Our mathematical model, however, predicts that collective resistance is only sustainable when resistance expression comes at a (modest) fitness cost (S4 Fig), and competitive exclusion is avoided by strong ecological feedback or alternative mechanisms (such as spatiotemporal structure or previously evolved niche partitioning). Nevertheless, even if coexistence is limited, the prolonged survival of susceptible cells within resistant communities may already represent an issue by increasing the opportunity for horizontal gene transfer during antibiotic selection pressure. Passive resistance might consequently represent an important factor towards the development of genetic multidrug resistance. Methods Strains and Growth Conditions S. pneumoniae CmS, a Cm-susceptible D39 derivate strain that constitutively expresses luc and a kanamycin resistance marker was used throughout. The Tc-resistant variant of this strain contained the Tc resistance gene tetM integrated at the bgaA locus, obtained via transformation with pPP1 [52]. luc has a reported half-life of 3 min in S. pneumoniae, and luminescence therefore gives real-time information on the level of gene expression activity [41,53]. S. pneumoniae CmR, expressing CAT from plasmid pJS5, was used as standard for a Cm-resistant strain. Initial experiments were carried out with the Cm-resistant strain D-PEP1C3 that expresses CAT from a strong synthetic promoter. S. pneumoniae D-PEP33 expressing GFP was used as a Cm-susceptible strain in time-lapse microscopy experiments [41]. S. aureus experiments were performed with strain LAC pCM29 [54] that constitutively expresses CAT and GFP. S. pneumoniae and S. aureus cells were grown in C+Y medium (pH 6.8), supplemented with 0.5 μg ml−1 D-luciferin for luminescence measurements, at 37°C [55]. Pre-cultures for all experiments were obtained by a standardized protocol, in which previously exponentially growing cells from −80°C stocks were diluted to OD (600 nm, path length 10 mm) 0.005 and grown until OD 0.1 in a volume of 2 ml medium inside tubes that allow for direct in-tube OD measurements. To determine the number of colony-forming units (CFUs), S. pneumoniae cells were plated inside Columbia agar supplemented with 3% (v v−1) sheep blood and incubated overnight at 37°C. Microtiter Plate Reader Assays Costar 96-well plates (white, clear bottom) with a total assay volume of 300 μl per well were inoculated to the designated starting OD value. Microtiter plate reader experiments were performed using a TECAN infinite pro 200 plate reader (Tecan Group) by measuring every 10 min with the following protocol: 5 s shaking, OD (595 nm, path length 10 mm) measurement with 25 flashes, luminescence measurement with an integration time of 1 s. In mixed population assays (shown in Fig 2A), all cultures were inoculated with CmS cells to an initial cell density of OD 0.001. CmR cells were inoculated to the same density, and control cultures without CmR cells contained equal amounts of Cm-sensitive D39 wild-type cells to correct for unspecific effects such as drug-titration via cellular Cm binding. HPLC Analysis S were obtained by CmR cultivation (inoculation at OD 0.001) in the presence of 5 μg ml−1 Cm in microtiter plates (as described above). Four wells were sampled and pooled per time point (combined volume of 1.2 ml), centrifuged to remove cells, and filtered through a 0.2 μm filter. HPLC analysis was carried out using an Agilent 1260 Infinity system (Agilent Technologies) with ultraviolet (UV) detection at 278 nm (maximum absorbance of Cm) [44]. An Aeris Peptide XB-C18 column (Phenomenex) with 3.6 μm particle and a size of 250 × 4.60 mm was used. Reversed-phase chromatography was carried out at a constant flow rate of 1 ml min−1, with the mobile phase consisting of solution A: 10 mM sodium acetate buffer (pH 6.0) containing 5% acetonitrile (v v−1) and solution B: acetonitrile 0.1% TFA, according to the following protocol: 100 μl sample loading, 3 min 10% B, 20 min gradient 10% to 50% B, 1 min gradient 50% to 95% B, 3 min 95% B, 1 min gradient 95% to 10% B, 6 min 10% B. Microscopy A Nikon Ti-E microscope equipped with a CoolsnapHQ2 camera and an Intensilight light source was used. Time-lapse microscopy was carried out by spotting pre-cultured cells on 10% polyacrylamide slides inside a Gene Frame (Thermo Fisher Scientific) that was sealed with the cover glass to guarantee stable conditions during microscopy. The polyacrylamide slide was prepared with growth medium containing 3 μg ml−1 Cm. Images of fluorescing cells were taken with the following protocol and filter settings: 0.3 s exposure for phase contrast, 0.5 s exposure for fluorescence at 440–490 nm excitation via a dichroic mirror of 495 nm, and an emission filter of 500–550 nm. Temperature during microscopy was controlled by an Okolab climate incubator, and images were taken every 10 min during 20 h at 37°C. Mouse Infection Model The murine pneumonia model was performed with slight modifications as previously described [56]. Based on pilot experiments, we estimated that the number of animals required to observe a statistical difference between the groups would exceed the technical limit of animals that could be inoculated and treated per day. Therefore, the experiment was split into 2 d with the original pool of animals randomized to each group at the start of the multi-day experiment. Prior to statistical analysis, the data were combined. Note that all intratracheal infections were performed in a blinded fashion with respect to Cm or vehicle treatment. Eight-wk-old female CD1 mice (Charles River Laboratories) with an average body weight of 28 g were used. Fresh cultures of CmS and CmR were started in 10 ml of Todd-Hewitt broth containing 2% yeast extract (THY) and 10 ml of THY supplemented with 5 μg ml−1 Cm, respectively. Cultures were grown at 37°C in a 5% CO2 incubator until OD (600 nm) 0.6. Bacteria were washed twice with PBS via centrifugation at 3,220 × g at room temperature and concentrated in PBS to yield 3.5 × 107 CFU in the inoculation volume of 40 μl. For mixed infections, an equal volume of concentrated CmS and CmR pneumococci were combined. Mice were anesthetized with 100 mg kg−1 ketamine and 10 mg kg−1 xylazine. Once sedated, the vocal chords were visualized using an operating otoscope (Welch Allyn), and 40 μl of bacteria was instilled into the trachea during inspiration using a plastic gel loading pipette tip. Mice were placed on a warmed pad for recovery. After 1 h, one intraperitoneal injection of Cm 75 mg kg−1 or vehicle controls was given, followed by two additional doses spaced 5 h apart. Mice were sacrificed with CO2 24 h after infection. To enumerate total surviving bacteria in the lungs, both lung lobes were removed and placed in a 2 ml sterile micro tube (Sarstedt) containing 1 ml of PBS and 1 mm silica beads (Biospec). Lungs were homogenized by shaking twice at 6,000 rpm for 1 min using a MagNA Lyser (Roche), with the specimens placed on ice as soon as they were harvested. Aliquots from each tube were serially diluted for CFU enumeration on THY plates. To determine whether or not a colony was CmS or CmR, individual colonies from the THY plates were picked and transferred into 100 μl of THY media in 96-well plates. The 96-well plates were incubated overnight at 37°C in a 5% CO2 incubator. After overnight incubation, wells were mixed, and 5 μl of media from each well was transferred into 100 μl of THY containing 15 μg ml−1 Cm or 100 μg ml−1 kanamycin. The 96-well plates were once again incubated overnight at 37°C in a 5% CO2 incubator, and wells were finally assessed for the presence or absence of a bacterial P. Cm (≥98% purity; Sigma) for animal injection was prepared as follows: 40 mg ml−1 of Cm was dissolved in 800 μl of 70% ethanol in PBS to make a 50 mg ml−1 stock solution. This stock solution was diluted in PBS to 3.75 mg ml−1 for intraperitoneal injection into mice at 75 mg kg−1. Ethics Statement This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The corresponding protocol entitled “Mouse Models of Bacterial Infection and Innate Immunity” (#S00227M) was approved by the Institutional Animal Care and Use Committee of the University of California, San Diego (Animal Welfare Assurance Number: A3033-01). All efforts were made to minimize suffering of animals employed in this study. Modeling The model describes the dynamic of a coculture of CAT-expressing CmR and CmS bacterial cells growing in the presence of Cm in a chemostat environment. The two strains, with population densities xr and xs, respectively, compete for a growth-limiting resource, z. Cm is assumed to inhibit growth; we separately keep track of the intracellular concentrations of Cm (ys in susceptible cells and yr in resistant cells) and its concentration in the extracellular medium ym. The equations for the growth of the two bacterial populations and the growth-limiting resource are given by (1) where r is the maximum growth rate of CmS cells, η is the relative growth efficiency of CmR cells, c is the resource consumption rate, and kz and hy, respectively, are the half-saturation and inhibitory constants of the growth function. Time, resource concentration, and cell densities have been scaled relative to the flow rate of the chemostat, the resource concentration in the inflow medium, and the number of cells that fit in the chemostat volume, respectively, in order to reduce the number of free parameters (see S1 Text for details). The concentrations of Cm in the different compartments, which have been scaled relative to the Cm concentration in the inflow medium, change according to the equations: (2) The processes described by the terms on the right-hand side include inflow of Cm into the medium, passive transport of Cm between compartments at rate p, outflow from the chemostat, degradation of Cm by CAT in CmR cells (according to Michaelis–Menten kinetics with maximum rate d and half-saturation constant ky), and concentration changes due to fluctuations in the volume of the compartments. Eqs (1) and (2) were solved numerically using Mathematica (Wolfram) or simulation software written in C++ (used for the numerical bifurcation analysis, based on a Runge–Kutta integration algorithm with adaptive step-size control). Strains and Growth Conditions S. pneumoniae CmS, a Cm-susceptible D39 derivate strain that constitutively expresses luc and a kanamycin resistance marker was used throughout. The Tc-resistant variant of this strain contained the Tc resistance gene tetM integrated at the bgaA locus, obtained via transformation with pPP1 [52]. luc has a reported half-life of 3 min in S. pneumoniae, and luminescence therefore gives real-time information on the level of gene expression activity [41,53]. S. pneumoniae CmR, expressing CAT from plasmid pJS5, was used as standard for a Cm-resistant strain. Initial experiments were carried out with the Cm-resistant strain D-PEP1C3 that expresses CAT from a strong synthetic promoter. S. pneumoniae D-PEP33 expressing GFP was used as a Cm-susceptible strain in time-lapse microscopy experiments [41]. S. aureus experiments were performed with strain LAC pCM29 [54] that constitutively expresses CAT and GFP. S. pneumoniae and S. aureus cells were grown in C+Y medium (pH 6.8), supplemented with 0.5 μg ml−1 D-luciferin for luminescence measurements, at 37°C [55]. Pre-cultures for all experiments were obtained by a standardized protocol, in which previously exponentially growing cells from −80°C stocks were diluted to OD (600 nm, path length 10 mm) 0.005 and grown until OD 0.1 in a volume of 2 ml medium inside tubes that allow for direct in-tube OD measurements. To determine the number of colony-forming units (CFUs), S. pneumoniae cells were plated inside Columbia agar supplemented with 3% (v v−1) sheep blood and incubated overnight at 37°C. Microtiter Plate Reader Assays Costar 96-well plates (white, clear bottom) with a total assay volume of 300 μl per well were inoculated to the designated starting OD value. Microtiter plate reader experiments were performed using a TECAN infinite pro 200 plate reader (Tecan Group) by measuring every 10 min with the following protocol: 5 s shaking, OD (595 nm, path length 10 mm) measurement with 25 flashes, luminescence measurement with an integration time of 1 s. In mixed population assays (shown in Fig 2A), all cultures were inoculated with CmS cells to an initial cell density of OD 0.001. CmR cells were inoculated to the same density, and control cultures without CmR cells contained equal amounts of Cm-sensitive D39 wild-type cells to correct for unspecific effects such as drug-titration via cellular Cm binding. HPLC Analysis S were obtained by CmR cultivation (inoculation at OD 0.001) in the presence of 5 μg ml−1 Cm in microtiter plates (as described above). Four wells were sampled and pooled per time point (combined volume of 1.2 ml), centrifuged to remove cells, and filtered through a 0.2 μm filter. HPLC analysis was carried out using an Agilent 1260 Infinity system (Agilent Technologies) with ultraviolet (UV) detection at 278 nm (maximum absorbance of Cm) [44]. An Aeris Peptide XB-C18 column (Phenomenex) with 3.6 μm particle and a size of 250 × 4.60 mm was used. Reversed-phase chromatography was carried out at a constant flow rate of 1 ml min−1, with the mobile phase consisting of solution A: 10 mM sodium acetate buffer (pH 6.0) containing 5% acetonitrile (v v−1) and solution B: acetonitrile 0.1% TFA, according to the following protocol: 100 μl sample loading, 3 min 10% B, 20 min gradient 10% to 50% B, 1 min gradient 50% to 95% B, 3 min 95% B, 1 min gradient 95% to 10% B, 6 min 10% B. Microscopy A Nikon Ti-E microscope equipped with a CoolsnapHQ2 camera and an Intensilight light source was used. Time-lapse microscopy was carried out by spotting pre-cultured cells on 10% polyacrylamide slides inside a Gene Frame (Thermo Fisher Scientific) that was sealed with the cover glass to guarantee stable conditions during microscopy. The polyacrylamide slide was prepared with growth medium containing 3 μg ml−1 Cm. Images of fluorescing cells were taken with the following protocol and filter settings: 0.3 s exposure for phase contrast, 0.5 s exposure for fluorescence at 440–490 nm excitation via a dichroic mirror of 495 nm, and an emission filter of 500–550 nm. Temperature during microscopy was controlled by an Okolab climate incubator, and images were taken every 10 min during 20 h at 37°C. Mouse Infection Model The murine pneumonia model was performed with slight modifications as previously described [56]. Based on pilot experiments, we estimated that the number of animals required to observe a statistical difference between the groups would exceed the technical limit of animals that could be inoculated and treated per day. Therefore, the experiment was split into 2 d with the original pool of animals randomized to each group at the start of the multi-day experiment. Prior to statistical analysis, the data were combined. Note that all intratracheal infections were performed in a blinded fashion with respect to Cm or vehicle treatment. Eight-wk-old female CD1 mice (Charles River Laboratories) with an average body weight of 28 g were used. Fresh cultures of CmS and CmR were started in 10 ml of Todd-Hewitt broth containing 2% yeast extract (THY) and 10 ml of THY supplemented with 5 μg ml−1 Cm, respectively. Cultures were grown at 37°C in a 5% CO2 incubator until OD (600 nm) 0.6. Bacteria were washed twice with PBS via centrifugation at 3,220 × g at room temperature and concentrated in PBS to yield 3.5 × 107 CFU in the inoculation volume of 40 μl. For mixed infections, an equal volume of concentrated CmS and CmR pneumococci were combined. Mice were anesthetized with 100 mg kg−1 ketamine and 10 mg kg−1 xylazine. Once sedated, the vocal chords were visualized using an operating otoscope (Welch Allyn), and 40 μl of bacteria was instilled into the trachea during inspiration using a plastic gel loading pipette tip. Mice were placed on a warmed pad for recovery. After 1 h, one intraperitoneal injection of Cm 75 mg kg−1 or vehicle controls was given, followed by two additional doses spaced 5 h apart. Mice were sacrificed with CO2 24 h after infection. To enumerate total surviving bacteria in the lungs, both lung lobes were removed and placed in a 2 ml sterile micro tube (Sarstedt) containing 1 ml of PBS and 1 mm silica beads (Biospec). Lungs were homogenized by shaking twice at 6,000 rpm for 1 min using a MagNA Lyser (Roche), with the specimens placed on ice as soon as they were harvested. Aliquots from each tube were serially diluted for CFU enumeration on THY plates. To determine whether or not a colony was CmS or CmR, individual colonies from the THY plates were picked and transferred into 100 μl of THY media in 96-well plates. The 96-well plates were incubated overnight at 37°C in a 5% CO2 incubator. After overnight incubation, wells were mixed, and 5 μl of media from each well was transferred into 100 μl of THY containing 15 μg ml−1 Cm or 100 μg ml−1 kanamycin. The 96-well plates were once again incubated overnight at 37°C in a 5% CO2 incubator, and wells were finally assessed for the presence or absence of a bacterial P. Cm (≥98% purity; Sigma) for animal injection was prepared as follows: 40 mg ml−1 of Cm was dissolved in 800 μl of 70% ethanol in PBS to make a 50 mg ml−1 stock solution. This stock solution was diluted in PBS to 3.75 mg ml−1 for intraperitoneal injection into mice at 75 mg kg−1. Ethics Statement This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The corresponding protocol entitled “Mouse Models of Bacterial Infection and Innate Immunity” (#S00227M) was approved by the Institutional Animal Care and Use Committee of the University of California, San Diego (Animal Welfare Assurance Number: A3033-01). All efforts were made to minimize suffering of animals employed in this study. Modeling The model describes the dynamic of a coculture of CAT-expressing CmR and CmS bacterial cells growing in the presence of Cm in a chemostat environment. The two strains, with population densities xr and xs, respectively, compete for a growth-limiting resource, z. Cm is assumed to inhibit growth; we separately keep track of the intracellular concentrations of Cm (ys in susceptible cells and yr in resistant cells) and its concentration in the extracellular medium ym. The equations for the growth of the two bacterial populations and the growth-limiting resource are given by (1) where r is the maximum growth rate of CmS cells, η is the relative growth efficiency of CmR cells, c is the resource consumption rate, and kz and hy, respectively, are the half-saturation and inhibitory constants of the growth function. Time, resource concentration, and cell densities have been scaled relative to the flow rate of the chemostat, the resource concentration in the inflow medium, and the number of cells that fit in the chemostat volume, respectively, in order to reduce the number of free parameters (see S1 Text for details). The concentrations of Cm in the different compartments, which have been scaled relative to the Cm concentration in the inflow medium, change according to the equations: (2) The processes described by the terms on the right-hand side include inflow of Cm into the medium, passive transport of Cm between compartments at rate p, outflow from the chemostat, degradation of Cm by CAT in CmR cells (according to Michaelis–Menten kinetics with maximum rate d and half-saturation constant ky), and concentration changes due to fluctuations in the volume of the compartments. Eqs (1) and (2) were solved numerically using Mathematica (Wolfram) or simulation software written in C++ (used for the numerical bifurcation analysis, based on a Runge–Kutta integration algorithm with adaptive step-size control). Supporting Information S1 Fig. Antibiotic degradation in the pneumococcus. (a–c), Plate reader assay sets in quadruplicate (average and s.e.m.) measuring luminescence (symbols with color outline) and cell density (corresponding grey symbols) of antibiotic-resistant (AbR), antibiotic-susceptible (AbS) and a mixture of resistant and susceptible (AbR+AbS) S. pneumoniae cells growing in the presence of 200 μg ml−1 kanamycin (a), 20 μg ml−1 gentamycin (b), and 3 μg ml−1 chloramphenicol (c). Assays with resistant cells (AbR) were inoculated to a density of OD 0.002, mixed populations (AbR+AbS) to a density of OD 0.001 each, and susceptible cells-only (AbS) also to a density of OD 0.001 with the addition of equal amounts of D39 wild type cells to correct for unspecific effects such as cellular drug binding. D-PEP22 that constitutively expresses firefly luciferase was used throughout as susceptible strain. Resistant strains expressed aphA1 (a), aacCI (b), and cat(c). Note that in aminoglycoside-inhibited cultures (a and b) luminescence of AbR+AbS assays decreased more rapidly compared with AbS assays. This can be explained by reduced luciferase expression rates when cultures exceed OD 0.05; AbScultures, in contrast to AbR+AbS cultures, do not reach OD 0.05 and consequently continue to express luciferase at a higher rate [41] (see S1 Data). https://doi.org/10.1371/journal.pbio.2000631.s001 (TIF) S2 Fig. Chloramphenicol deactivation assay at a concentration of two times the MIC. (a), Plate reader assay sets in quadruplicate (average and s.e.m.) measuring luminescence (symbols with color outline) and cell density (corresponding grey symbols) of chloramphenicol-susceptible S. pneumoniae D-PEP2K1 (CmS) growing in the presence of 5 μg ml−1 chloramphenicol (Cm), in presence (+) or absence (−) of resistant D-PEP1-pJS5 (CmR) cells. (b), Development of the count of viable CmS cells (CFUs ml−1, colony-forming units per ml) during the cultivation assay presented in a, determined via plating in the presence of kanamycin; average values of duplicates are shown (see S1 Data). https://doi.org/10.1371/journal.pbio.2000631.s002 (TIF) S3 Fig. Single-cell analysis of pneumococcal collective resistance. (a,b), Still images (overlay of phase contrast and fluorescence microscopy) of a time-lapse experiment of chloramphenicol-susceptible S. pneumoniae D-PEP33 cells that constitutively express GFP, either co-cultivated with the CAT-expressing S. pneumoniae D-PEP1-pJS5 (a) or in monoculture (b), growing on a semi-solid surface supplemented with 3 μg ml−1 chloramphenicol. High inoculation densities were spotted, resulting in rapid chloramphenicol deactivation in the co-cultivation assay. Note that GFP, which allows for the distinction between chloramphenicol-susceptible and -resistant cells at the beginning of the time-lapse experiment (fluorescent versus non-fluorescent), bleaches quickly in the course of the assay; inhibited susceptible cells, even after (partial) Cm clearance, do not express sufficient levels of GFP (counteracting photobleaching) to allow for a continuous detection. Scale bar, 10 μm. https://doi.org/10.1371/journal.pbio.2000631.s003 (TIF) S4 Fig. Numerical model analysis. (a–c), Colored areas indicate qualitatively different outcomes of competition between CmS and CmR cells in model simulations, as a function of two key parameters: the growth rate efficiency of CmR cells (1 – η quantifies the cost of CAT expression), and the concentration of Cm in the inflow medium (Y0; ‘antibiotic stress’). (a), An orange line borders the region in which the CmS strain can grow from low initial density. This is below a critical level of antibiotic stress, or in a narrow range of η values when the resistant cells are present. The CmR strain can grow from low initial density in the area bordered by a solid blue line, but can maintain high population densities over a larger area of parameter space (i.e., in the area bordered by a dashed blue line). Stable coexistence of both strains is maintained in a narrow parameter region (red area). When η is close to 1, CmR is always a superior competitor (dark blue area), in line with the analytical result that coexistence cannot be maintained unless CAT expression is costly. When CAT expression costs are high, CmS tends to outcompete CmR. However, this process leads to an elevation of Cm in the medium, which may eventually cause both strains to go extinct (light blue area; here, CmR can survive on its own, but not when CmS is also present; see S5 Fig). Alternatively, CmS can persist on its own after driving CmR to extinction (orange area). Parameters are: r = 20.0, kz = 4.0, c = 1.0, p = 50.0, hY = 0.25/Y0, kY = 2.5/Y0 and d = 30.0/Y0. In (b), the relative benefit of CAT degradation is larger, due to a slower diffusion of Cm across the cell membrane (p = 25.0; other parameters as in a). (c), This panel illustrates the effect of a change in the resource consumption rate c which affects the equilibrium population densities (c = 2.0; other parameters as in a). In this case, CmR and CmS reach lower equilibrium densities, weakening the effect of CmR on the environment. As a result, the conditions for coexistence become more stringent. Throughout, we performed multiple simulations per parameter condition to search for boundary and interior equilibria, and classified the dynamics based on the stability properties of the equilibria. Color saturation within each area gives an indication of the total cell density at equilibrium. https://doi.org/10.1371/journal.pbio.2000631.s004 (TIF) S5 Fig. Extinction induced by competition. (a), Simulated growth trajectories for CmR and CmS populations subject to antibiotic stress and competition for a limiting resource. Here, the CmR strain is an inferior competitor that is driven to extinction by the invasion of CmS cells, even though the growth conditions are not permissive for the survival of CmS on its own. Extinction is caused by a bistability in the growth dynamic of CmR cells: a critical cell density is required to lower the concentration of Cm below the level that permits population growth. The initial CmR cell density in the simulation was just above this critical level (indicated by a dotted gray line); the CmR cells are not able to invade if their initial density lies below the threshold (shown by the dashed blue trajectory). However, after successful invasion (solid blue trajectory), the CmR cells can still be pushed below their critical density by competition with the CmS strain, triggering the collapse of both populations. (b), Dynamics of intracellular Cm concentrations and resource. Parameters are: r = 20.0, η = 0.85, kz = 4.0, c = 1.0, p = 50.0, hY = 0.25, kY = 2.5 and d = 30.0. https://doi.org/10.1371/journal.pbio.2000631.s005 (TIF) S6 Fig. LL-37 activity in dependency on chloramphenicol. Killing of Cm-susceptible S. neumoniae D-PEP2K1 (CmS) and Cm-resistant D-PEP1C3 (CmR) by the human antimicrobial peptide LL-37 at a concentration of 50 μg ml−1, in absence (−) or presence (+) of 5 μg ml−1 chloramphenicol (Cm); average and s.e.m. of duplicates are shown. *P< 0.05; two-tailed t-test (see S1 Data). https://doi.org/10.1371/journal.pbio.2000631.s006 (TIF) S1 Text. Derivation of the mathematical model and model analysis. https://doi.org/10.1371/journal.pbio.2000631.s007 (PDF) S1 Data. Numerical values underlying the data presented in the figures. https://doi.org/10.1371/journal.pbio.2000631.s008 (XLSX) S1 Movie. Development of interspecies collective resistance. Time-lapse microscopy experiment of S. pneumoniae D-PEP2K1 (CmS), co-cultivated with a strain of the pneumococcal niche competitor Staphylococcus aureus that expresses CAT and GFP (strain LAC pCM29), growing on a semi-solid surface containing 3 μg ml−1 chloramphenicol. The first still frame of the time-lapse experiment is annotated as one hour into the cultivation start (01:00); one hour was the time required for reaching stable conditions inside the microscopy slide that allow for automated recording. Note that GFP expression, in the case of the Cm-resistant S. aureus LAC pCM29, is not inhibited by the Cm treatment, and GFP is consequently continuously produced (counteracting photobleaching and dilution). The observed high fluorescence is the result of GFP expression from a multi-copy plasmid (in contrast to the single-copy genomic integration in S. pneumoniae D-PEP33 shown in S3 Fig). https://doi.org/10.1371/journal.pbio.2000631.s009 (MP4) S2 Movie. Chloramphenicol-treated susceptible pneumococci. Time-lapse microscopy experiment of chloramphenicol-susceptible S. pneumoniae D-PEP2K1 (CmS) monoculture growing on a semi-solid surface containing 3 μg ml−1 chloramphenicol. https://doi.org/10.1371/journal.pbio.2000631.s010 (MP4) Acknowledgments We thank M. Espinosa for plasmid pJS5, R. Nijland for S. aureus strain LAC pCM29, M. Montalban and M. Bartholomae for assistance with the HPLC system, and Lingjun He for support with statistical analysis.

doi: 10.1371/journal.pbio.2000598pmid: 27911892

Introduction Reproducibility is a fundamental principle of the scientific method and distinguishes scientific evidence from mere anecdote. The advancement of basic as well as applied research depends on the reproducibility of the findings, and can be seriously hampered if reproducibility is poor. However, accumulating evidence indicates that reproducibility is poor in many disciplines across the life sciences [1]. For example, in a study on microarray gene expression, only 8 out of 18 studies could be reproduced [2]; Prinz and colleagues [3] found large inconsistencies (65%) between published and in-house data in the fields of oncology, women’s health, and cardiovascular diseases; oncologists from Amgen could confirm only 6 out of 53 published findings [4]; and, of more than 100 compounds that showed promising effects on amyotrophic lateral sclerosis (ALS) in preclinical trials, none displayed the same effect when retested by the ALS Therapy Development Institute in Cambridge [5]. Besides a waste of time and resources for inconclusive research [6–8], however, poor reproducibility also entails serious ethical problems. In clinical research, irreproducibility of preclinical research may expose patients to unnecessary risks [9,10], while in basic and preclinical animal research, it may cause unjustified harm to experimental animals [11]. Reproducibility critically depends on experimental design and conduct, which together account for the internal and external validity of experimental results [12]. External validity refers to how applicable results are to other environmental conditions, experimenters, study populations, and even to other strains or species of animals (including humans) [12]. Thus, it also determines reproducibility of the results across replicate studies (i.e., across different labs, different experimenters, different study populations, etc.) [11,13,14]. Internal validity refers to the extent to which a causal relation between experimental treatment and outcome is warranted, and critically depends on scientific rigor, i.e., the extent to which experimental design and conduct minimize systematic bias [12,15]. It has been suggested that poor internal validity due to a lack of scientific rigor may also be a major cause of poor reproducibility in animal research [16–18]. There are various sources of bias (e.g., selection bias, performance bias, detection bias), and specific measures exist to mitigate them (e.g., randomization, blinding, sample-size calculation; [12,15,19,20]). To assess the internal validity of studies, e.g., in the peer review process, and to facilitate replication of studies, publications must contain sufficiently detailed information about experimental design and conduct, including measures taken against risks of bias [20,21]. However, systematic reviews generally found a low prevalence of reporting of measures against risks of bias (further referred to as reporting) in animal research publications. Thus, reporting ranged from 8% to 55.6% for allocation concealment, from 3% to 61% for blinded outcome assessment, from 7% to 55% for randomization, and from 0% to 3% for sample size calculation [19,22–29]. Low rates of reporting have been interpreted as evidence for a lack of scientific rigor (e.g., [20]). Indeed, several systematic reviews found correlations between poor reporting and overstated treatment effects [19,29–31]. Reporting guidelines have thus become a major weapon in the fight against risks of bias in animal research [32]. However, although the ARRIVE guidelines (Animal Research: Reporting of In Vivo Experiments) by the United Kingdom-based organization NC3Rs (National Centre for the Replacement, Refinement & Reduction of Animals in Research) have been endorsed by over 1,000 journals, this did not lead to a substantial improvement of reporting in animal studies [33]. Nevertheless, awareness seems to rise, as Macleod and colleagues [28] recently found that reporting increased over the past decades, although there is still considerable scope for improvement. Research on the internal validity of animal experiments has focused mainly on reporting in scientific publications. However, most published research has undergone peer review when submitted for funding, and in some countries (e.g., Switzerland, Germany), individual animal experiments are licensed by national or regional authorities. For example, in Switzerland, the licensing of animal experiments is based on an explicit harm–benefit analysis, whereby any harm imposed on the animals is gauged against the expected benefit (gain of knowledge) of the experiment. Because the gain of knowledge critically depends on the scientific validity of the findings, risks of bias may affect the weight attributed to the expected benefit of a study in the harm–benefit analysis. An accurate harm–benefit analysis thus depends on information regarding risks of bias and measures used to mitigate them. In the present study, we therefore screened applications for animal experiments submitted to the cantonal authorities in Switzerland (n = 1,277) for evidence of the use of measures to avoid risks of bias, and compared the rates at which these measures were described in applications (for reasons of simplicity hereafter also referred to as reporting) with the rates of reporting of the same measures in a representative sub-sample of publications (n = 50) resulting from experiments described in these applications. This allowed us, for the first time, to compare evidence of scientific rigor available to the authorities when licensing animal experiments with the evidence reported in scientific publications, and to assess whether poor reporting in the scientific literature is predicted by poor reporting in applications for experiments. Results Our database included a final sample of 1,277 applications for animal experiments approved by the cantonal authorities of Switzerland in the years 2008, 2010, and 2012, respectively. Evidence of scientific rigor was assessed based on seven common measures against risks of bias: allocation concealment, randomization, blinded outcome assessment, sample size calculation, inclusion and exclusion criteria, primary outcome, and a statistical analysis plan (S2 and S3 Texts). Besides analyzing each item separately, we also calculated an internal validity score (IVS; see Eq 1), which served as the primary outcome variable for the statistical analysis of effects of various study descriptors on rates of reporting. In addition, we calculated an accuracy score (AS; see Eq 2) based on six items of information explicitly asked for on the application form as a measure of how accurately the applicants had filled out the application form to control for effects of accuracy on the IVS. Reporting Rates Reporting rates were generally very low (Table 1); on average, less than one out of the seven items were reported in applications for animal experiments, with reporting rates varying among the seven items, ranging from 2.4% for the statistical analysis plan to 18.5% for the primary outcome variable (Table 1). However, reporting rates greatly differed between individual applications, with the IVS ranging from 0 (i.e., 0/7 items reported) to 0.857 (i.e., 6/7 items reported), whereby 711 out of the 1,277 applications (55.68%) scored 0 (S1 Fig). Download: PPT PowerPoint slide PNG larger image TIFF original image Table 1. Reporting rates (in %) of measures against risks of bias in applications for animal experiments in Switzerland depending on year of authorization, type of institution, use of genetically modified animals, authorizing canton, language of the application, species category, and degree of severity of the planned procedures. https://doi.org/10.1371/journal.pbio.2000598.t001 Influence of Study Descriptors We hypothesized that reporting rates and, thus, the IVS might depend on various characteristics of the studies, including the year of authorization (Year), the types of animals used (Species), the severity of the experimental procedures (Severity), the institution conducting the study (Institution), the canton authorizing the study (Canton), and the language in which the application was written (Language), as well as the AS of the application. Generalized linear models in a Bayesian information criterion selection process were used to identify which of the study descriptors best described our data, indicating that they were most likely to have influenced the IVS. The best fitting model included Year, Canton, Language, Institution, and the interaction between Species and AS (see Eq 4). According to the model output (S1 Data), however, none of the individual descriptors had a significant effect on the IVS except Language, as applications written in German had a significantly higher IVS compared to applications written in English (odds ratio [OR] = 0.79, 95% confidence interval [CI] = 0.64–0.98) and applications written in French (OR = 0.46, CI = 0.32–0.65), and the interaction between farm animals and AS (OR = 168.24, CI = 1.17–2,5571.31). Thus, below we report trends that were observed regarding effects of the descriptors that were included in the final model on the IVS. The IVS was similar across all three years of authorization: 2012 (median = 0.0, range: 0–0.71), 2010 (0.0, 0–0.71), and 2008 (0.0, 0–0.85). At the level of individual items, trends of improvement across years were observed in the reporting rates of blinding, sample size calculation, and statistical analysis plan (Fig 1B–1E). While there was some variation in IVS across cantons, canton did not seem to have a strong effect (Fig 2). Among the different research institutions, academic institutions (i.e., universities, federal institutes of technology, or university hospitals) accounted by far for the largest part of applications, with 972 (76%) applications compared to 87 (7%) from industry, 56 (4%) from governmental institutions, and 162 (13%) from other private institutions. Overall, academic institutions (0.0, 0–0.86) tended to score lower on IVS than institutions from industry (0.14, 0–0.57), governmental institutions (0.14, 0–0.71), and other private institutions (0.14, 0–0.57; Fig 3A). At the level of individual items, similar trends were observed in the reporting rates of randomization and sample size calculation (Fig 3B–3E). There was also variation in IVS depending on the species of animals used (Fig 4A). Thus, applications for experiments on “higher” mammals (i.e., cats, dogs, rabbits, and primates [CDRP]) tended to score higher (0.17, 0–0.71) compared to experiments on farm animals (0.14, 0–0.86), other mammals (0.15, 0–0.29), laboratory rodents (0.0, 0–0.71), and non-mammals (0.0, 0–0.6), respectively. A similar trend was observed in the reporting rates of blinding, randomization, sample size calculation, and statistical analysis (Fig 4B–4E). Thus, applications for experiments on CDRP as well as farm animals scored higher compared to those involving laboratory rodents and non-mammals, while data from applications for experiments involving other mammals varied widely due to the small sample size (n = 8). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Internal validity score of applications depending on the year of authorization. (A) Boxplot of the IVS for the three years of authorization. The dashed blue line represents the overall mean IVS for the entire sample. The red squares represent the mean IVS for each year. (B–E) Barplots (with binomial confidence intervals) representing reporting rates per year of authorization for blinding (B), randomization (C), sample size calculation (D), and statistical analysis (E). Individual data are shown in https://figshare.com/s/bc48ed5dff9e6ebd2000 (Figure 1). https://doi.org/10.1371/journal.pbio.2000598.g001 Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Internal validity score of applications depending on the authorizing canton. Boxplot of the IVS for the six largest cantons (1–6) and the group of small cantons. The dashed blue line represents the overall mean IVS for the entire sample. The red squares represent the mean IVS for each canton or group of cantons. Individual data are shown in https://figshare.com/s/bc48ed5dff9e6ebd2000 (Figure 2). https://doi.org/10.1371/journal.pbio.2000598.g002 Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Internal validity score of applications depending on institutions. (A) Boxplot of the IVS for the four categories of institutions. The dashed blue line represents the overall mean IVS for the entire sample. The red squares represent the mean IVS for each category of institutions. (B–E) Barplots (with binomial confidence intervals) representing the reporting rates for each category of institutions (A: academia; G: governmental institutions; I: industry; O: other) for blinding (B), randomization (C), sample size calculation (D), and statistical analysis (E). Individual data are shown in https://figshare.com/s/bc48ed5dff9e6ebd2000 (Figure 3). https://doi.org/10.1371/journal.pbio.2000598.g003 Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. Internal validity score of applications depending on the species of animals. (A) Boxplot of the IVS for the five categories of animal species (CDRP: cats, dogs, rabbits and, primates; Farm: farm animals; O_mamm: other mammals; Rodents: laboratory rodents; N_mamm: non-mammals). The dashed blue line represents the overall mean IVS for the entire sample. The red squares represent the mean IVS for each category of animal species. (B–E) Barplots (with binomial confidence intervals) representing the reporting rates for each category of species (C: cats, dogs, rabbits and, primates; F: farm animals; O: other mammals; R: laboratory rodents; N: non-mammals) for blinding (B), randomization (C), sample size calculation (D), and statistical analysis (E). Individual data are shown in https://figshare.com/s/bc48ed5dff9e6ebd2000 (Figure 4). https://doi.org/10.1371/journal.pbio.2000598.g004 In contrast to the IVS, the AS was generally high, with a median score of 0.8, ranging from 0.11 to 1.00. Despite the low IVS and more than half of the applications scoring 0, there was a weak but positive correlation between AS and IVS (Spearman’s rho = 0.17, p < 0.001; Fig 5). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. Relationship between AS and IVS. Scatter plots of IVS in relation to AS. Individual data are shown in https://figshare.com/s/bc48ed5dff9e6ebd2000 (Figure 5). https://doi.org/10.1371/journal.pbio.2000598.g005 Reliability In order to ensure reliability of the data between the two investigators (TSR, LV) as well as across time, inter-rater and intra-rater reliability tests were conducted at regular intervals. Inter-rater reliability scores (see Eq 3) of the IVS ranged from 91.4% to 97.1%, while the respective intra-rater reliability scores ranged from 87.1% to 95.7% for TSR and from 94.3% to 97.1% for LV. Similarly, inter-rater reliability scores of the AS ranged from 91.3% to 96.3%, while the respective intra-rater reliability scores ranged from 87.5% to 97.5% for TSR and from 92.5% to 98.8% for LV (see S2 Data). Comparison between Applications and Publications In order to relate the reporting rates obtained from applications for animal experiments to reporting rates found in the scientific literature, we selected 50 publications originating from 50 independent applications in our sample, screened them for the same seven internal validity criteria, and calculated the IVS for each publication using the same method. Similar to what we found for applications, reporting rates in the 50 publications were generally low, albeit slightly higher than in the applications (Fig 6), resulting in a median IVS of 0.14. Reporting rates for the seven items ranged from 0% for sample size calculation to 34% for the statistical analysis plan. Again, reporting rates differed greatly between individual publications, with IVS ranging from 0 to 0.6, whereby 23 out of 50 publications (46%) scored 0. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 6. Reporting rates of the seven internal validity criteria in applications and publications. Barplot (with binomial confidence intervals) representing reporting rates of the seven internal validity criteria (Alloc: allocation concealment; Blind: blinding; Random: randomization; Sample: sample size calculation; Inc/Exc: inclusion and exclusion criteria; Outcome: primary outcome; Stats: statistical analysis). Individual data are shown in https://figshare.com/s/bc48ed5dff9e6ebd2000 (Sample Applications and Sample Publications). https://doi.org/10.1371/journal.pbio.2000598.g006 Except for sample size calculation and the primary outcome variable, reporting rates for individual items were higher in publications than in applications (see Fig 6). Whereas IVS of applications and publications were the same in 27 cases (of which 21 scored 0), it was higher in 18 pairs (which was due to a statistical analysis plan in 12 cases) and lower in five cases. This increase was corroborated by a weak positive correlation between the IVS of applications and that of publications (Spearman’s rho = 0.34, p = 0.014). Influence of Study Descriptors Due to the smaller sample size, not all descriptors assessed for their effects on the IVS of applications could be analyzed here. Instead, we analyzed publication-specific descriptors, namely whether or not the journal in which the study was published had endorsed the ARRIVE guidelines and the impact factor of the journal (IF). There was no significant effect of ARRIVE on IVS (yes: median = 0.14, range: 0 to 0.57; no: median = 0, range: 0 to 0.60; p = 0.69; Fig 7A). In contrast, IF had a significant negative effect on IVS (Spearman’s rho = -0.49, p < 0.001; Fig 7B). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 7. Internal validity score of publications depending on endorsement of the ARRIVE guidelines by the journal and the journal’s impact factor. (A) IVS depending on endorsement of the ARRIVE guidelines. The dashed blue line represents the overall mean IVS for the entire sample of publications. The red squares represent the mean IVS for each group. (B) Scatter plot of the IVS depending on the impact factor of the journal. Individual data are shown in https://figshare.com/s/bc48ed5dff9e6ebd2000 (Sample Publications). https://doi.org/10.1371/journal.pbio.2000598.g007 Reporting Rates Reporting rates were generally very low (Table 1); on average, less than one out of the seven items were reported in applications for animal experiments, with reporting rates varying among the seven items, ranging from 2.4% for the statistical analysis plan to 18.5% for the primary outcome variable (Table 1). However, reporting rates greatly differed between individual applications, with the IVS ranging from 0 (i.e., 0/7 items reported) to 0.857 (i.e., 6/7 items reported), whereby 711 out of the 1,277 applications (55.68%) scored 0 (S1 Fig). Download: PPT PowerPoint slide PNG larger image TIFF original image Table 1. Reporting rates (in %) of measures against risks of bias in applications for animal experiments in Switzerland depending on year of authorization, type of institution, use of genetically modified animals, authorizing canton, language of the application, species category, and degree of severity of the planned procedures. https://doi.org/10.1371/journal.pbio.2000598.t001 Influence of Study Descriptors We hypothesized that reporting rates and, thus, the IVS might depend on various characteristics of the studies, including the year of authorization (Year), the types of animals used (Species), the severity of the experimental procedures (Severity), the institution conducting the study (Institution), the canton authorizing the study (Canton), and the language in which the application was written (Language), as well as the AS of the application. Generalized linear models in a Bayesian information criterion selection process were used to identify which of the study descriptors best described our data, indicating that they were most likely to have influenced the IVS. The best fitting model included Year, Canton, Language, Institution, and the interaction between Species and AS (see Eq 4). According to the model output (S1 Data), however, none of the individual descriptors had a significant effect on the IVS except Language, as applications written in German had a significantly higher IVS compared to applications written in English (odds ratio [OR] = 0.79, 95% confidence interval [CI] = 0.64–0.98) and applications written in French (OR = 0.46, CI = 0.32–0.65), and the interaction between farm animals and AS (OR = 168.24, CI = 1.17–2,5571.31). Thus, below we report trends that were observed regarding effects of the descriptors that were included in the final model on the IVS. The IVS was similar across all three years of authorization: 2012 (median = 0.0, range: 0–0.71), 2010 (0.0, 0–0.71), and 2008 (0.0, 0–0.85). At the level of individual items, trends of improvement across years were observed in the reporting rates of blinding, sample size calculation, and statistical analysis plan (Fig 1B–1E). While there was some variation in IVS across cantons, canton did not seem to have a strong effect (Fig 2). Among the different research institutions, academic institutions (i.e., universities, federal institutes of technology, or university hospitals) accounted by far for the largest part of applications, with 972 (76%) applications compared to 87 (7%) from industry, 56 (4%) from governmental institutions, and 162 (13%) from other private institutions. Overall, academic institutions (0.0, 0–0.86) tended to score lower on IVS than institutions from industry (0.14, 0–0.57), governmental institutions (0.14, 0–0.71), and other private institutions (0.14, 0–0.57; Fig 3A). At the level of individual items, similar trends were observed in the reporting rates of randomization and sample size calculation (Fig 3B–3E). There was also variation in IVS depending on the species of animals used (Fig 4A). Thus, applications for experiments on “higher” mammals (i.e., cats, dogs, rabbits, and primates [CDRP]) tended to score higher (0.17, 0–0.71) compared to experiments on farm animals (0.14, 0–0.86), other mammals (0.15, 0–0.29), laboratory rodents (0.0, 0–0.71), and non-mammals (0.0, 0–0.6), respectively. A similar trend was observed in the reporting rates of blinding, randomization, sample size calculation, and statistical analysis (Fig 4B–4E). Thus, applications for experiments on CDRP as well as farm animals scored higher compared to those involving laboratory rodents and non-mammals, while data from applications for experiments involving other mammals varied widely due to the small sample size (n = 8). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Internal validity score of applications depending on the year of authorization. (A) Boxplot of the IVS for the three years of authorization. The dashed blue line represents the overall mean IVS for the entire sample. The red squares represent the mean IVS for each year. (B–E) Barplots (with binomial confidence intervals) representing reporting rates per year of authorization for blinding (B), randomization (C), sample size calculation (D), and statistical analysis (E). Individual data are shown in https://figshare.com/s/bc48ed5dff9e6ebd2000 (Figure 1). https://doi.org/10.1371/journal.pbio.2000598.g001 Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Internal validity score of applications depending on the authorizing canton. Boxplot of the IVS for the six largest cantons (1–6) and the group of small cantons. The dashed blue line represents the overall mean IVS for the entire sample. The red squares represent the mean IVS for each canton or group of cantons. Individual data are shown in https://figshare.com/s/bc48ed5dff9e6ebd2000 (Figure 2). https://doi.org/10.1371/journal.pbio.2000598.g002 Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Internal validity score of applications depending on institutions. (A) Boxplot of the IVS for the four categories of institutions. The dashed blue line represents the overall mean IVS for the entire sample. The red squares represent the mean IVS for each category of institutions. (B–E) Barplots (with binomial confidence intervals) representing the reporting rates for each category of institutions (A: academia; G: governmental institutions; I: industry; O: other) for blinding (B), randomization (C), sample size calculation (D), and statistical analysis (E). Individual data are shown in https://figshare.com/s/bc48ed5dff9e6ebd2000 (Figure 3). https://doi.org/10.1371/journal.pbio.2000598.g003 Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. Internal validity score of applications depending on the species of animals. (A) Boxplot of the IVS for the five categories of animal species (CDRP: cats, dogs, rabbits and, primates; Farm: farm animals; O_mamm: other mammals; Rodents: laboratory rodents; N_mamm: non-mammals). The dashed blue line represents the overall mean IVS for the entire sample. The red squares represent the mean IVS for each category of animal species. (B–E) Barplots (with binomial confidence intervals) representing the reporting rates for each category of species (C: cats, dogs, rabbits and, primates; F: farm animals; O: other mammals; R: laboratory rodents; N: non-mammals) for blinding (B), randomization (C), sample size calculation (D), and statistical analysis (E). Individual data are shown in https://figshare.com/s/bc48ed5dff9e6ebd2000 (Figure 4). https://doi.org/10.1371/journal.pbio.2000598.g004 In contrast to the IVS, the AS was generally high, with a median score of 0.8, ranging from 0.11 to 1.00. Despite the low IVS and more than half of the applications scoring 0, there was a weak but positive correlation between AS and IVS (Spearman’s rho = 0.17, p < 0.001; Fig 5). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. Relationship between AS and IVS. Scatter plots of IVS in relation to AS. Individual data are shown in https://figshare.com/s/bc48ed5dff9e6ebd2000 (Figure 5). https://doi.org/10.1371/journal.pbio.2000598.g005 Reliability In order to ensure reliability of the data between the two investigators (TSR, LV) as well as across time, inter-rater and intra-rater reliability tests were conducted at regular intervals. Inter-rater reliability scores (see Eq 3) of the IVS ranged from 91.4% to 97.1%, while the respective intra-rater reliability scores ranged from 87.1% to 95.7% for TSR and from 94.3% to 97.1% for LV. Similarly, inter-rater reliability scores of the AS ranged from 91.3% to 96.3%, while the respective intra-rater reliability scores ranged from 87.5% to 97.5% for TSR and from 92.5% to 98.8% for LV (see S2 Data). Comparison between Applications and Publications In order to relate the reporting rates obtained from applications for animal experiments to reporting rates found in the scientific literature, we selected 50 publications originating from 50 independent applications in our sample, screened them for the same seven internal validity criteria, and calculated the IVS for each publication using the same method. Similar to what we found for applications, reporting rates in the 50 publications were generally low, albeit slightly higher than in the applications (Fig 6), resulting in a median IVS of 0.14. Reporting rates for the seven items ranged from 0% for sample size calculation to 34% for the statistical analysis plan. Again, reporting rates differed greatly between individual publications, with IVS ranging from 0 to 0.6, whereby 23 out of 50 publications (46%) scored 0. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 6. Reporting rates of the seven internal validity criteria in applications and publications. Barplot (with binomial confidence intervals) representing reporting rates of the seven internal validity criteria (Alloc: allocation concealment; Blind: blinding; Random: randomization; Sample: sample size calculation; Inc/Exc: inclusion and exclusion criteria; Outcome: primary outcome; Stats: statistical analysis). Individual data are shown in https://figshare.com/s/bc48ed5dff9e6ebd2000 (Sample Applications and Sample Publications). https://doi.org/10.1371/journal.pbio.2000598.g006 Except for sample size calculation and the primary outcome variable, reporting rates for individual items were higher in publications than in applications (see Fig 6). Whereas IVS of applications and publications were the same in 27 cases (of which 21 scored 0), it was higher in 18 pairs (which was due to a statistical analysis plan in 12 cases) and lower in five cases. This increase was corroborated by a weak positive correlation between the IVS of applications and that of publications (Spearman’s rho = 0.34, p = 0.014). Influence of Study Descriptors Due to the smaller sample size, not all descriptors assessed for their effects on the IVS of applications could be analyzed here. Instead, we analyzed publication-specific descriptors, namely whether or not the journal in which the study was published had endorsed the ARRIVE guidelines and the impact factor of the journal (IF). There was no significant effect of ARRIVE on IVS (yes: median = 0.14, range: 0 to 0.57; no: median = 0, range: 0 to 0.60; p = 0.69; Fig 7A). In contrast, IF had a significant negative effect on IVS (Spearman’s rho = -0.49, p < 0.001; Fig 7B). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 7. Internal validity score of publications depending on endorsement of the ARRIVE guidelines by the journal and the journal’s impact factor. (A) IVS depending on endorsement of the ARRIVE guidelines. The dashed blue line represents the overall mean IVS for the entire sample of publications. The red squares represent the mean IVS for each group. (B) Scatter plot of the IVS depending on the impact factor of the journal. Individual data are shown in https://figshare.com/s/bc48ed5dff9e6ebd2000 (Sample Publications). https://doi.org/10.1371/journal.pbio.2000598.g007 Discussion Based on the low reporting rates in publications of animal research and evidence suggesting that poor reporting may reflect a lack of scientific rigor [19,29–31], this study examined whether poor reporting in the scientific literature is predicted by poor reporting in applications for animal experiments, that is before the studies have actually been conducted. The study was restricted to animal experiments authorized in Switzerland for two reasons. First, Switzerland has an authorization system for animal experiments that requires detailed description of study protocols for every planned study. These study protocols form the basis of the harm–benefit analysis upon which the decision for or against authorization of individual studies is based. Second, the study was facilitated by the Swiss Federal Food Safety and Veterinary Office (FSVO) providing access to all applications for animal experiments via their online platform (e-tierversuche) through which scientists communicate with the authorities and submit their applications for animal experiments. Such unlimited access to application forms for animal experiments is unprecedented, and it is laudable that the FSVO supported this meta-research. This kind of support has notoriously proven difficult to obtain for reasons of confidentiality, as highlighted by Chan et al. [34], with respect to clinical trial protocols for meta-research. As described in the Materials and Methods, access to the application forms was possible without violating confidentiality. Low Reporting Rates We evaluated 1,277 applications for animal experiments and 50 publications derived thereof and found very low reporting rates in both applications and publications (Fig 6). Reporting rates in publications were within the range reported in previous studies (e.g., [19,20]). That reporting rates in applications were similar—even slightly lower—indicates that the authorities approving animal experiments are lacking important information about experimental conduct that may be critical for evaluating the expected benefit in a harm–benefit analysis. Risks of bias question the scientific validity of the results, which is a precondition for a study to achieve the expected benefit. Whether the authorities are unaware of risks of bias and measures to avoid them or whether they consider them as unimportant for the benefit of the research is unknown and warrants further study. As a result, however, animal experiments are authorized based on implicit confidence rather than explicit evidence of scientific rigor. Similarly, poor reporting in publications means that manuscripts are often accepted for publication in the absence of evidence of scientific rigor. This “trust me model” of science has been criticized before [1,35,36]. It sheds serious doubts on the current authorization procedure for animal experiments as well as the peer-review process for scientific publications, which in the long run may compromise the credibility of the research. Relationship between Reporting in Applications and Publications We found a weak positive correlation between the IVS of applications and that of the corresponding publications. This suggests that the reporting of bias avoidance measures in applications predicted, at least to some extent, the reporting of such measures in publications. If this reflects a consistent relationship, asking for more detailed information on experimental conduct in applications for animal experiments might help to promote better experimental conduct as well as better reporting in publications. Asking for more detailed information at the planning stage of the research might also reduce the danger of normative responses, whereby scientists simply satisfy the guidelines (e.g., ARRIVE) at a time when it is too late to take corrective actions on experimental conduct. The increase in the IVS of publications compared to applications was largely due to better reporting of the statistical analysis plan (S2 Fig). This is likely due to the fact that journals (and reviewers) generally insist on a detailed description of the statistical analysis. It indicates that reporting guidelines (such as ARRIVE) could potentially increase scientific quality of animal research, if editors and reviewers helped to enforce them. However, as shown by Baker et al. [33] and confirmed by the present study (Fig 7A), this has not been the case so far; publications in journals having endorsed the ARRIVE guidelines did not score higher than publications in other journals. We also found a weak positive correlation between the accuracy of completing the application forms (AS) and the IVS. Thus, applicants who answered questions in the application form more accurately had a higher IVS. As shown by Minnerup et al. [37], this further confirms that enforcement of guidelines may be important in view of improving reporting standards. Effects of Study Characteristics on the IVS of Applications and Publications In the final statistical model, language was the only descriptor having a significant effect on IVS of applications for animal experiments. Applications written in German had significantly higher IVS than applications written in English or French. Several explanations may account for this result. For example, the proportion of German native speakers may have been higher among authors of German applications; German may have been mostly used by native German speakers, while English may have been used by many non-native English speakers. Similarly, French may have been used by many non-native French speakers because, apparently, authorities in French-speaking cantons of Switzerland strongly encourage submission of applications in French (own observation). However, one might not necessarily expect language skills to affect such standardized terminology (randomization, blinding, etc.), but because these items are not explicitly asked for, applicants writing in their native language might be more likely to provide unsolicited detail. Alternatively, differences in regional policies of authorities between French- and German-speaking cantons, as well as the fact that all French applications were scored by only one experimenter (LV), may have contributed to this effect, but our data do not allow us to examine these explanations further. Apart from language, all other explanatory variables in the final model had only weak effects on IVS that did not reach statistical significance (S1 Data). For example, there was a weak tendency for the reporting rates of blinding, sample size calculation, and statistical analysis to be higher in 2012 compared to those from previous years (Fig 1). This trend might reflect increasing awareness by both researchers and authorities of the importance of reporting, and it is consistent with recent evidence from a random sample of life sciences publications [28]. However, despite the many systematic reviews revealing flaws in experimental design and conduct since Ioannidis’ seminal opinion paper [38], and the wealth of solutions that have since been proposed [2,5,32,39], little progress has been made. Like Baker et al. in 2014 [33], we did not find convincing evidence that reporting had increased from applications authorized before (2008) to those authorized after (2012) publication of the ARRIVE guidelines. Again, the main reason for this might be a lack of enforcement of these guidelines by authorities as well as journal editors. However, our sample was mostly based on studies designed and authorized before the ARRIVE guidelines became widely known. That the endorsement of the ARRIVE guidelines had no effect on the IVS of publications may thus reflect the delay in such a change taking effect. Recent evidence indicated that industry-sponsored research is less biased than academic research [40]. We therefore predicted higher rates of reporting of measures against risks of bias in applications from private compared to academic institutions. Although there was a weak tendency for applications from academic institutions to score lower on IVS compared to governmental and private institutions, we cannot exclude random variation as the source of this trend. If true, however, it might reflect the different incentives between institutions, favoring more conservative approaches in non-academic institutions [41]. An interesting tendency was found in relation to the type of animals being used. Thus, applications for experiments on CDRP, farm animals, and other mammals had slightly higher IVS than those for experiments on lab rodents and non-mammals. CDRP and, to a lesser extent, farm animals and other mammals may benefit from the attribution of a higher moral status, e.g., because they are close relatives (primates), social partners (dogs, cats, rabbits), or otherwise elicit more compassion (farm animals, other mammals) than lab rodents (that are also considered as “pest” species) and non-mammals (mostly fish; e.g., [42,43,44]). On the one hand, this might indicate that applications are assessed more carefully when the stakes are perceived as morally high, although it would remain unclear whether this effect is due to the applicants providing more information or to the authorities asking for more. On the other hand, IVS was low throughout, and the difference between species categories was not significant. In addition, there was no such trend with increasing degree of severity of studies. Importantly, however, the Swiss Animal Welfare Act does not provide a legal basis for such “speciesism” among vertebrates, and both authors and authorities should treat all vertebrates equally. Finally, we found a weak but significant negative relationship between the IVS of publications and the IF of the journal in which it was published. That the journal IF does not necessarily reflect the quality of research has long been known (e.g., [45]), and a systematic review of a random sample of life sciences publications recently found no evidence for a positive relationship between IF and reporting [28]. Across the whole range of journal IF in our sample of publications, IVS of 0 clearly prevails, confirming that poor reporting of measures against risk of bias is common throughout the scientific literature. General Conclusions According to the Animal Protection Index (API) by World Animal Protection, Switzerland (together with the United Kingdom, Austria, and New Zealand) ranked top in an international comparison of animal protection policy among 50 countries (http://api.worldanimalprotection.org/). In particular, authorization of animal experiments is based on a harm–benefit analysis, and authorization is denied if, in relation to the anticipated gain in knowledge, they inflict disproportionate harm on the animals (Article 19(4), [46]). Because the anticipated gain in knowledge critically depends on experimental design and conduct, the lack of information on measures against risks of bias in applications means that, in Switzerland, authorization of animal experiments is based on implicit confidence rather than explicit evidence of scientific rigor. Several arguments may be held against this interpretation of our results, namely (i) that the measures against risks of bias assessed here are not important determinants of scientific validity, (ii) that they are not explicitly asked for on the application form for animal experiments, (iii) that, as the system currently works, it is not the authorities’ duty to assess the scientific validity of the experiments, and (iv) that the authorities’ confidence in scientific rigor is well justified. First, it is certainly the case that the authorities assess the scientific rationale underlying the proposed studies, thereby assessing several important aspects of scientific validity, although these are not specified explicitly. Also, there may be other, even more important risks of bias (e.g., use of inappropriate control group) that were not included in our evaluation. However, all seven items included here are considered as relevant measures against risks of bias that may compromise scientific validity in important ways; they have therefore been included in reporting guidelines such as the ARRIVE guidelines. Second, while it is also true that the application form does not explicitly ask for allocation concealment, randomization, blinding, and inclusion or exclusion criteria, it does ask explicitly for the primary and secondary outcome variables, sample size calculation, and a detailed statistical analysis plan. Moreover, the first example of how to describe procedures presented in the explanatory notes to the application form by the FSVO starts with “The dogs are divided randomly into 3 groups,” indicating that randomization is also considered a relevant aspect of the description of procedures. Even if only those measures explicitly asked for on the application form were enforced, all applications would score IVS ≥ 0.42 (i.e., 3/7). Third, authorities may argue that it is the peers’ duty to assess and guarantee scientific rigor, while the authorities’ duties (and those of their advisory committees) should be limited to assessing the scope for applying the 3Rs (replacement of animal experiments, reduction of animal use, and refinement of procedures) and whether the expected benefits (as declared by the applicants) outweigh the harms inflicted on the animals. However, it is important to note that not all experiments are based on project proposals that have undergone scientific peer review (e.g., most applications from the private sector), and that peer review does not seem to guarantee good scientific practice [47]. Finally, whether the authorities’ implicit confidence in the scientific validity of the results of licensed experiments is justified is an empirical question. Concerns that such confidence may not be warranted is largely based on studies showing a negative relationship between reporting of measures against risks of bias and inflation of treatment effect size in preclinical studies (e.g., [19,25]). Together with accumulating evidence of poor reproducibility of in vivo research, these findings have shed doubts on the quality of experimental design and conduct. However, there is clearly a need for more research on the actual implementation of measures against risks of bias in experimental animal research. We have recently conducted an online survey amongst all Swiss animal researchers to elucidate actual implementation of the same seven measures against risks of bias assessed here. Our findings suggest that although reporting rates found in the literature tend to underestimate actual implementation of these measures, there is considerable scope for improvement [48]. Lack of scientific rigor in experimental conduct is widely considered to be an important determinant of poor reproducibility of in vivo research [16,17,18]. However, this assumption is based on the indirect evidence outlined above, and has never been tested directly. Randomization, blinding, sample size calculation, and all the other measures against risks of bias assessed here mainly affect the internal validity of experiments. Although the reproducibility of results can be affected by the internal validity of studies, reproducibility depends more on the external validity of studies [11–13]. Reproducibility may thus be enhanced mainly by using design features aimed to increase the external validity of results, such as more heterogeneous study populations, independent replicate cohorts, or multicenter study designs [14,49,50]. Thus, there is also a need for more research on the relative contribution of experimental conduct and experimental design, respectively, to the reproducibility of results. Last, but not least, besides experimental design and experimental conduct, several other factors introduce bias into the scientific literature, in particular “hypothesizing after results are known” (HARKing, [51]), p-hacking [52], selective reporting [53], and publication bias [54]. The most effective way of eliminating all of these biases would be prospective registration of preclinical animal experiments similar to preregistration of clinical trials [55]. Further research is certainly needed on how to facilitate practical implementation of preregistration in the face of several contentious issues such as confidentiality, property rights, and theft of ideas. However, the authorization procedure for animal experiments already in place in Switzerland (and other countries, e.g., Germany), provides an ideal basis for implementing preregistration of animal experiments, which would not only benefit the scientific validity of results from animal experiments but also minimize unnecessary harm to animals for inconclusive research. By this, Switzerland could consolidate its position as a leader in animal protection as well as extend its leadership to scientific rigor. Low Reporting Rates We evaluated 1,277 applications for animal experiments and 50 publications derived thereof and found very low reporting rates in both applications and publications (Fig 6). Reporting rates in publications were within the range reported in previous studies (e.g., [19,20]). That reporting rates in applications were similar—even slightly lower—indicates that the authorities approving animal experiments are lacking important information about experimental conduct that may be critical for evaluating the expected benefit in a harm–benefit analysis. Risks of bias question the scientific validity of the results, which is a precondition for a study to achieve the expected benefit. Whether the authorities are unaware of risks of bias and measures to avoid them or whether they consider them as unimportant for the benefit of the research is unknown and warrants further study. As a result, however, animal experiments are authorized based on implicit confidence rather than explicit evidence of scientific rigor. Similarly, poor reporting in publications means that manuscripts are often accepted for publication in the absence of evidence of scientific rigor. This “trust me model” of science has been criticized before [1,35,36]. It sheds serious doubts on the current authorization procedure for animal experiments as well as the peer-review process for scientific publications, which in the long run may compromise the credibility of the research. Relationship between Reporting in Applications and Publications We found a weak positive correlation between the IVS of applications and that of the corresponding publications. This suggests that the reporting of bias avoidance measures in applications predicted, at least to some extent, the reporting of such measures in publications. If this reflects a consistent relationship, asking for more detailed information on experimental conduct in applications for animal experiments might help to promote better experimental conduct as well as better reporting in publications. Asking for more detailed information at the planning stage of the research might also reduce the danger of normative responses, whereby scientists simply satisfy the guidelines (e.g., ARRIVE) at a time when it is too late to take corrective actions on experimental conduct. The increase in the IVS of publications compared to applications was largely due to better reporting of the statistical analysis plan (S2 Fig). This is likely due to the fact that journals (and reviewers) generally insist on a detailed description of the statistical analysis. It indicates that reporting guidelines (such as ARRIVE) could potentially increase scientific quality of animal research, if editors and reviewers helped to enforce them. However, as shown by Baker et al. [33] and confirmed by the present study (Fig 7A), this has not been the case so far; publications in journals having endorsed the ARRIVE guidelines did not score higher than publications in other journals. We also found a weak positive correlation between the accuracy of completing the application forms (AS) and the IVS. Thus, applicants who answered questions in the application form more accurately had a higher IVS. As shown by Minnerup et al. [37], this further confirms that enforcement of guidelines may be important in view of improving reporting standards. Effects of Study Characteristics on the IVS of Applications and Publications In the final statistical model, language was the only descriptor having a significant effect on IVS of applications for animal experiments. Applications written in German had significantly higher IVS than applications written in English or French. Several explanations may account for this result. For example, the proportion of German native speakers may have been higher among authors of German applications; German may have been mostly used by native German speakers, while English may have been used by many non-native English speakers. Similarly, French may have been used by many non-native French speakers because, apparently, authorities in French-speaking cantons of Switzerland strongly encourage submission of applications in French (own observation). However, one might not necessarily expect language skills to affect such standardized terminology (randomization, blinding, etc.), but because these items are not explicitly asked for, applicants writing in their native language might be more likely to provide unsolicited detail. Alternatively, differences in regional policies of authorities between French- and German-speaking cantons, as well as the fact that all French applications were scored by only one experimenter (LV), may have contributed to this effect, but our data do not allow us to examine these explanations further. Apart from language, all other explanatory variables in the final model had only weak effects on IVS that did not reach statistical significance (S1 Data). For example, there was a weak tendency for the reporting rates of blinding, sample size calculation, and statistical analysis to be higher in 2012 compared to those from previous years (Fig 1). This trend might reflect increasing awareness by both researchers and authorities of the importance of reporting, and it is consistent with recent evidence from a random sample of life sciences publications [28]. However, despite the many systematic reviews revealing flaws in experimental design and conduct since Ioannidis’ seminal opinion paper [38], and the wealth of solutions that have since been proposed [2,5,32,39], little progress has been made. Like Baker et al. in 2014 [33], we did not find convincing evidence that reporting had increased from applications authorized before (2008) to those authorized after (2012) publication of the ARRIVE guidelines. Again, the main reason for this might be a lack of enforcement of these guidelines by authorities as well as journal editors. However, our sample was mostly based on studies designed and authorized before the ARRIVE guidelines became widely known. That the endorsement of the ARRIVE guidelines had no effect on the IVS of publications may thus reflect the delay in such a change taking effect. Recent evidence indicated that industry-sponsored research is less biased than academic research [40]. We therefore predicted higher rates of reporting of measures against risks of bias in applications from private compared to academic institutions. Although there was a weak tendency for applications from academic institutions to score lower on IVS compared to governmental and private institutions, we cannot exclude random variation as the source of this trend. If true, however, it might reflect the different incentives between institutions, favoring more conservative approaches in non-academic institutions [41]. An interesting tendency was found in relation to the type of animals being used. Thus, applications for experiments on CDRP, farm animals, and other mammals had slightly higher IVS than those for experiments on lab rodents and non-mammals. CDRP and, to a lesser extent, farm animals and other mammals may benefit from the attribution of a higher moral status, e.g., because they are close relatives (primates), social partners (dogs, cats, rabbits), or otherwise elicit more compassion (farm animals, other mammals) than lab rodents (that are also considered as “pest” species) and non-mammals (mostly fish; e.g., [42,43,44]). On the one hand, this might indicate that applications are assessed more carefully when the stakes are perceived as morally high, although it would remain unclear whether this effect is due to the applicants providing more information or to the authorities asking for more. On the other hand, IVS was low throughout, and the difference between species categories was not significant. In addition, there was no such trend with increasing degree of severity of studies. Importantly, however, the Swiss Animal Welfare Act does not provide a legal basis for such “speciesism” among vertebrates, and both authors and authorities should treat all vertebrates equally. Finally, we found a weak but significant negative relationship between the IVS of publications and the IF of the journal in which it was published. That the journal IF does not necessarily reflect the quality of research has long been known (e.g., [45]), and a systematic review of a random sample of life sciences publications recently found no evidence for a positive relationship between IF and reporting [28]. Across the whole range of journal IF in our sample of publications, IVS of 0 clearly prevails, confirming that poor reporting of measures against risk of bias is common throughout the scientific literature. General Conclusions According to the Animal Protection Index (API) by World Animal Protection, Switzerland (together with the United Kingdom, Austria, and New Zealand) ranked top in an international comparison of animal protection policy among 50 countries (http://api.worldanimalprotection.org/). In particular, authorization of animal experiments is based on a harm–benefit analysis, and authorization is denied if, in relation to the anticipated gain in knowledge, they inflict disproportionate harm on the animals (Article 19(4), [46]). Because the anticipated gain in knowledge critically depends on experimental design and conduct, the lack of information on measures against risks of bias in applications means that, in Switzerland, authorization of animal experiments is based on implicit confidence rather than explicit evidence of scientific rigor. Several arguments may be held against this interpretation of our results, namely (i) that the measures against risks of bias assessed here are not important determinants of scientific validity, (ii) that they are not explicitly asked for on the application form for animal experiments, (iii) that, as the system currently works, it is not the authorities’ duty to assess the scientific validity of the experiments, and (iv) that the authorities’ confidence in scientific rigor is well justified. First, it is certainly the case that the authorities assess the scientific rationale underlying the proposed studies, thereby assessing several important aspects of scientific validity, although these are not specified explicitly. Also, there may be other, even more important risks of bias (e.g., use of inappropriate control group) that were not included in our evaluation. However, all seven items included here are considered as relevant measures against risks of bias that may compromise scientific validity in important ways; they have therefore been included in reporting guidelines such as the ARRIVE guidelines. Second, while it is also true that the application form does not explicitly ask for allocation concealment, randomization, blinding, and inclusion or exclusion criteria, it does ask explicitly for the primary and secondary outcome variables, sample size calculation, and a detailed statistical analysis plan. Moreover, the first example of how to describe procedures presented in the explanatory notes to the application form by the FSVO starts with “The dogs are divided randomly into 3 groups,” indicating that randomization is also considered a relevant aspect of the description of procedures. Even if only those measures explicitly asked for on the application form were enforced, all applications would score IVS ≥ 0.42 (i.e., 3/7). Third, authorities may argue that it is the peers’ duty to assess and guarantee scientific rigor, while the authorities’ duties (and those of their advisory committees) should be limited to assessing the scope for applying the 3Rs (replacement of animal experiments, reduction of animal use, and refinement of procedures) and whether the expected benefits (as declared by the applicants) outweigh the harms inflicted on the animals. However, it is important to note that not all experiments are based on project proposals that have undergone scientific peer review (e.g., most applications from the private sector), and that peer review does not seem to guarantee good scientific practice [47]. Finally, whether the authorities’ implicit confidence in the scientific validity of the results of licensed experiments is justified is an empirical question. Concerns that such confidence may not be warranted is largely based on studies showing a negative relationship between reporting of measures against risks of bias and inflation of treatment effect size in preclinical studies (e.g., [19,25]). Together with accumulating evidence of poor reproducibility of in vivo research, these findings have shed doubts on the quality of experimental design and conduct. However, there is clearly a need for more research on the actual implementation of measures against risks of bias in experimental animal research. We have recently conducted an online survey amongst all Swiss animal researchers to elucidate actual implementation of the same seven measures against risks of bias assessed here. Our findings suggest that although reporting rates found in the literature tend to underestimate actual implementation of these measures, there is considerable scope for improvement [48]. Lack of scientific rigor in experimental conduct is widely considered to be an important determinant of poor reproducibility of in vivo research [16,17,18]. However, this assumption is based on the indirect evidence outlined above, and has never been tested directly. Randomization, blinding, sample size calculation, and all the other measures against risks of bias assessed here mainly affect the internal validity of experiments. Although the reproducibility of results can be affected by the internal validity of studies, reproducibility depends more on the external validity of studies [11–13]. Reproducibility may thus be enhanced mainly by using design features aimed to increase the external validity of results, such as more heterogeneous study populations, independent replicate cohorts, or multicenter study designs [14,49,50]. Thus, there is also a need for more research on the relative contribution of experimental conduct and experimental design, respectively, to the reproducibility of results. Last, but not least, besides experimental design and experimental conduct, several other factors introduce bias into the scientific literature, in particular “hypothesizing after results are known” (HARKing, [51]), p-hacking [52], selective reporting [53], and publication bias [54]. The most effective way of eliminating all of these biases would be prospective registration of preclinical animal experiments similar to preregistration of clinical trials [55]. Further research is certainly needed on how to facilitate practical implementation of preregistration in the face of several contentious issues such as confidentiality, property rights, and theft of ideas. However, the authorization procedure for animal experiments already in place in Switzerland (and other countries, e.g., Germany), provides an ideal basis for implementing preregistration of animal experiments, which would not only benefit the scientific validity of results from animal experiments but also minimize unnecessary harm to animals for inconclusive research. By this, Switzerland could consolidate its position as a leader in animal protection as well as extend its leadership to scientific rigor. Materials and Methods Sampling Process Applications for animal experiments (Form A, S1 Text) were selected from an anonymized database obtained from the FSVO, containing all applications submitted in Switzerland since 1983. Access to applications archived by the FSVO was based on a contract between the FSVO and the authors of this study, which guaranteed confidentiality to the applicants. Applications were selected based on predefined inclusion and exclusion criteria. Thus, only new applications submitted during the years 2008, 2010, and 2012 were included, of which applications related to (i) diagnosis of disease, (ii) education and training, and (iii) the protection of humans, animals, and the environment by toxicological or other safety tests required by law were excluded a priori (S3 Fig). A total of 1590 applications met these criteria and were subjected to formal screening. Checklist In order to assess risks of bias in the experiments described in the applications, a checklist was elaborated (S2 Text) based on checklists used in previous studies assessing the use of measures to reduce risks of biases as reported in the published literature [19,20,56]. We restricted our checklist to items that (i) are essentially applicable to all kinds of experimental studies and (ii) can be assessed objectively without specific expertise of the research topic, and included those seven items that we encountered most often in the literature: (1) allocation concealment, (2) blinded outcome assessment, (3) randomization, (4) formal sample size calculation, (5) inclusion and exclusion criteria, (6) a primary outcome variable, and (7) a statistical analysis plan. These seven items were also used to calculate an IVS based on the number of items that were reported in the application divided by the total number of items applicable to the study (max = 7). (1) Additional items were assessed that were, however, not included in the IVS. These included additional aspects of study conduct (blinded conduct of study, randomized conduct of study, termination criteria, references for the sample size, and general statements on statistical analysis; S2 Text). In addition, we assessed the accuracy with which the application forms (Form A) were filled out, using items that were explicitly asked for on Form A, and for which the content to be filled in was explicitly specified in the accompanying guidelines to Form A on the FSVO webpage (https://www.blv.admin.ch/dam/blv/en/dokumente/tiere/publikationen-und-forschung/tierversuche/erlaeuterungen-form-a.pdf.download.pdf/erlaeuterungen-form-a.pdf). Furthermore, we chose items that are relevant for the harm–benefit analysis and could be determined with high reliability. The following six items were included: (1) description and justification of the methods used (e.g., by indicating references, previous results, or results from a pilot study); (2) information about the identification of individual animals; (3) the total number of animals used, the number of treatment groups, and the number of animals per treatment group; (4) reference to a score sheet for the assessment of animal welfare; (5) the degrees of severity for all animals involved in the experiments; and (6) the fate of the animals at the end of the experiments. These six items were used to calculate an AS based on the number of items reported divided by the total number of items applicable to the study (max = 6). (2) The AS was constructed as a control measure, to control for variation in IVS induced by variation in the accuracy with which the form was filled out. Both IVS and AS were assessed by scoring whether or not the respective items were reported in any of the experiments included in an application form. Thus, a “YES” was recorded if an item was reported in at least one of the described experiments and a “NO” if an item was either not reported at all or if it was unclear. If an item was not applicable to the experiment described in the application form, “NA” was recorded (more details are given in the S3 Text). Data Collection The 1590 applications were randomly allocated to two investigators (LV, TSR) for formal screening (leading to two lists of 795 applications each, one for each investigator). During screening, 94 applications were excluded because they were either incomplete or not available in the archives of the FSVO. A further 36 applications were excluded because they met one or more of the exclusion criteria reported above. This left 1,460 applications that were deemed suitable for screening. Applications written in French (n = 423) or Italian (n = 5) were screened by the investigator with better knowledge of these languages (LV), regardless of their assignment to the two investigators, while applications written in German (n = 430) or English (n = 602) were screened according to their assignments to the two investigators. Therefore, a total sample of n = 935 was screened by investigator LV while a total sample of n = 525 applications was screened by investigator TSR. To restrict analysis to experimental in vivo studies, a further 183 applications were excluded in the course of the screening process because they referred to in vitro studies (if the animals were killed before the experimental treatment was applied; n = 106), monitoring studies (if the animals were observed in the wild; n = 28), or other exceptions (e.g., breeding studies, post-mortem studies; n = 49), resulting in a final sample size of n = 1,277 applications used for analysis (see S3 Fig). Based on information provided by the applicants on Form A and used for the annual statistics of animal use by the FSVO, we also recorded several descriptors that might influence the reporting of internal validity items; these included (i) year of authorization (2008, 2010, 2012), (ii) language (English, German, French), (iii) canton (the six largest cantons of Basel, Bern, Freiburg, Geneva, Vaud, Zurich, and the group of the remaining small cantons), (iv) type of institution (academic institutions [i.e., universities, federal institutes of technology, hospitals], industry, governmental institutions [national and cantonal], other [e.g., private institutions, foundations]), (v) animal species (laboratory rodents, higher mammals [CDRP], farm animals, other mammals, non-mammals), (vi) genetically modified animals (yes, no), and (vii) the prospective degree of severity of the planned procedures as defined by the FSVO (0, 1, 2, 3). Inter-rater Reliability Prior to the screening of the selected Form A, two pilot studies on separate applications (i.e., applications authorized in 2009) were conducted to ensure the applicability of the checklist and to ensure consistency of scoring within and between investigators. To ensure consistent scoring of applications between the two investigators, both investigators screened the same 10 applications, and discrepancies were checked at the end of the day. Inter-rater reliability (Eq 3) was assessed at regular intervals (on day 1 and then after the 100th, 300th, 500th, and 700th application on the investigators’ list, respectively) by assessing the proportion of agreement between the two investigators. For this, the first five applications on each investigator’s list were screened by both investigators. (3) Only applications written in either German or English were used for inter-rater reliability tests. Overall, 50 applications were screened twice in the course of these inter-rater reliability tests. Inter-rater reliability never dropped below 85% (S2 Data). Intra-rater Reliability To ensure that both investigators scored applications consistently over time, samples of 10 applications were re-scored at regular intervals (after 50, 150, 350, and 550 listed applications, respectively). In addition, each investigator conducted a final intra-rater reliability test on 10 randomly chosen applications from the whole list after completing the screening procedure. If systematic discrepancies would have occurred, the applications previously scored would have been re-scored. However, as in the case of inter-rater reliability, intra-rater reliability never dropped below 85% (S2 Data). Sample Size Calculation No a priori sample size calculation was performed, as all applications were included in our sample that fulfilled the inclusion/exclusion criteria. However, once the sample size was determined, we verified that it was suitable for the planned statistical analysis (see model description below). Statistics The screening data from the checklists were transferred to a tabulating program (Microsoft Excel 2010.Ink, Redmond, WA, USA) and imported into the statistical software R [57]. We used descriptive statistics to represent reporting rates for individual criteria of internal validity (allocation concealment, blinded outcome assessment, randomization, sample size calculation, inclusion and exclusion criteria, primary outcome, and statistical analysis). Furthermore, influences of relevant descriptors (year, canton, institution, and animal species) were represented graphically, with median and mean IVS of the group, and overall mean IVS. For the statistical analysis of the overall internal validity score of applications, we used generalized linear models to evaluate the influence of the a priori stated descriptors on the internal validity score. The analyses were performed in R [57] using the built in function glm with a binomial error distribution to account for the data structure (primary outcome as proportions). As a first step, we compared univariate models (model with one descriptor) with an intercept-only model (modelling the intercept of the internal validity score) based on significant (p < 0.05) likelihood ratio test of the package lmtest [58] in order to identify descriptors to be included in the further modelling process. The descriptors to be retained were language, canton, species category, institutions, authorization year, and accuracy of the application. In a second step, by means of an information theoretic approach to model selection using the Bayesian Information Criterion (BIC), we identified the model that best fit our data. For an automated model selection procedure, the package MuMln [59] with the function dredge was used to compare all models with all possible combinations of the retained descriptors (full model included also the interaction term for species category and accuracy; see Eq 4). (4) The dredge function ranks all descriptor combinations according to their BIC; the model with the lowest BIC was assumed to be the one representing our data best. The final model included the following main effects (descriptors): language (3 levels), cantons (7 levels), species category (5 levels), accuracy (continuous), institution (4 levels), and authorization year (3 levels). In addition to these main effects, the candidate model included the two-way interaction between species category and accuracy (corresponds to full model, cf. Eq 4). The model parameters were retrieved after correction for over dispersion (see S1 Data). Publications Sampling Process and Screening In order to relate the reporting rates of internal validity criteria assessed here by scoring applications for animal experiments with the reporting rates of such criteria in the published literature [19,22–28,60], we also scored a sub-sample of publications originating from studies based on applications in our study sample. These were identified by searching through grant numbers mentioned in the applications and references listed as output in the annual reports to the FSVO. For 155 applications (12.1%) we identified one or more corresponding publications. This number was reduced to 139 after excluding reviews and publications that were clearly unrelated to the study described in the applications (mismatch in animal species, general topic, or methods). This low number can be explained by the fact that studies licensed in 2012 and also many of those licensed in 2010 were not yet published, and that the search for publications had to rely on grant numbers mentioned in both application and publication (often grant numbers were not mentioned on applications) or on publications listed in the final reports required by the authorities upon completion of licensed studies (for most studies licensed in 2012 and also many of those licensed in 2010, final reports were not yet available). For the comparison of the internal validity scores between applications and publications, we aimed to detect a medium effect size (0.3) with a statistical power of 0.8 at a significance level of p < 0.05. Based on this, we chose a sample size of n = 50, which allowed us to detect an effect size of 0.276 (G*power for correlations, bivariate normal model) [61]. A stratified random sampling procedure was used to select 50 publications from the 139 available publications, so as to select publications derived from a representative sample of all applications with respect to canton and type of animals used. Because this sample of publications was biased towards older applications, we compared the IVS of the sub-sample of 50 applications from which these 50 publications originated with the IVS of the entire sample of applications and found no significant difference; median IVS of the entire sample of applications (n = 1,277) was 0.0 (range 0 to 0.857), compared to 0.0 (range 0 to 0.714) for the sub-sample of applications (n = 50) from which the 50 publications were derived. The publications were screened for reporting of internal validity criteria with a checklist containing the same seven internal validity criteria as were used for applications. The screening of all 50 publications was performed by one single investigator (LV). Publications were randomly allocated to one of the 10 d of screening (five publications per day). Days of screening were separated by two non-screening days. For the publications, descriptors were impact factor of the journal and endorsement of the ARRIVE guidelines by the journal. To determine the descriptors, the impact factor for the year of the publication as well as the ARRIVE status of the journal were assessed. If it was not possible to determine the ARRIVE status of a journal for the date of publication, given that all publications were published in 2012 or later, we used the ARRIVE status of the journal in 2015. Whether or not the ARRIVE status affected the internal validity score of publications was tested with a univariate generalized linear model (binomial error distribution), with IVS as dependent and the descriptor (endorsement of ARRIVE yes or no) as independent variables. (5) Whether or not the internal validity score of publications was correlated with the impact factor of the journal was investigated using a spearman rank correlation test. To ensure that the investigator scored the publications constantly over time, an independent person randomly chose one publication per five publications screened (i.e., one per day of screening) for an intra-rater reliability test. The chosen publication was re-screened on the second following day. The reliability (Eq 3) never dropped below the threshold of 85%. Sampling Process Applications for animal experiments (Form A, S1 Text) were selected from an anonymized database obtained from the FSVO, containing all applications submitted in Switzerland since 1983. Access to applications archived by the FSVO was based on a contract between the FSVO and the authors of this study, which guaranteed confidentiality to the applicants. Applications were selected based on predefined inclusion and exclusion criteria. Thus, only new applications submitted during the years 2008, 2010, and 2012 were included, of which applications related to (i) diagnosis of disease, (ii) education and training, and (iii) the protection of humans, animals, and the environment by toxicological or other safety tests required by law were excluded a priori (S3 Fig). A total of 1590 applications met these criteria and were subjected to formal screening. Checklist In order to assess risks of bias in the experiments described in the applications, a checklist was elaborated (S2 Text) based on checklists used in previous studies assessing the use of measures to reduce risks of biases as reported in the published literature [19,20,56]. We restricted our checklist to items that (i) are essentially applicable to all kinds of experimental studies and (ii) can be assessed objectively without specific expertise of the research topic, and included those seven items that we encountered most often in the literature: (1) allocation concealment, (2) blinded outcome assessment, (3) randomization, (4) formal sample size calculation, (5) inclusion and exclusion criteria, (6) a primary outcome variable, and (7) a statistical analysis plan. These seven items were also used to calculate an IVS based on the number of items that were reported in the application divided by the total number of items applicable to the study (max = 7). (1) Additional items were assessed that were, however, not included in the IVS. These included additional aspects of study conduct (blinded conduct of study, randomized conduct of study, termination criteria, references for the sample size, and general statements on statistical analysis; S2 Text). In addition, we assessed the accuracy with which the application forms (Form A) were filled out, using items that were explicitly asked for on Form A, and for which the content to be filled in was explicitly specified in the accompanying guidelines to Form A on the FSVO webpage (https://www.blv.admin.ch/dam/blv/en/dokumente/tiere/publikationen-und-forschung/tierversuche/erlaeuterungen-form-a.pdf.download.pdf/erlaeuterungen-form-a.pdf). Furthermore, we chose items that are relevant for the harm–benefit analysis and could be determined with high reliability. The following six items were included: (1) description and justification of the methods used (e.g., by indicating references, previous results, or results from a pilot study); (2) information about the identification of individual animals; (3) the total number of animals used, the number of treatment groups, and the number of animals per treatment group; (4) reference to a score sheet for the assessment of animal welfare; (5) the degrees of severity for all animals involved in the experiments; and (6) the fate of the animals at the end of the experiments. These six items were used to calculate an AS based on the number of items reported divided by the total number of items applicable to the study (max = 6). (2) The AS was constructed as a control measure, to control for variation in IVS induced by variation in the accuracy with which the form was filled out. Both IVS and AS were assessed by scoring whether or not the respective items were reported in any of the experiments included in an application form. Thus, a “YES” was recorded if an item was reported in at least one of the described experiments and a “NO” if an item was either not reported at all or if it was unclear. If an item was not applicable to the experiment described in the application form, “NA” was recorded (more details are given in the S3 Text). Data Collection The 1590 applications were randomly allocated to two investigators (LV, TSR) for formal screening (leading to two lists of 795 applications each, one for each investigator). During screening, 94 applications were excluded because they were either incomplete or not available in the archives of the FSVO. A further 36 applications were excluded because they met one or more of the exclusion criteria reported above. This left 1,460 applications that were deemed suitable for screening. Applications written in French (n = 423) or Italian (n = 5) were screened by the investigator with better knowledge of these languages (LV), regardless of their assignment to the two investigators, while applications written in German (n = 430) or English (n = 602) were screened according to their assignments to the two investigators. Therefore, a total sample of n = 935 was screened by investigator LV while a total sample of n = 525 applications was screened by investigator TSR. To restrict analysis to experimental in vivo studies, a further 183 applications were excluded in the course of the screening process because they referred to in vitro studies (if the animals were killed before the experimental treatment was applied; n = 106), monitoring studies (if the animals were observed in the wild; n = 28), or other exceptions (e.g., breeding studies, post-mortem studies; n = 49), resulting in a final sample size of n = 1,277 applications used for analysis (see S3 Fig). Based on information provided by the applicants on Form A and used for the annual statistics of animal use by the FSVO, we also recorded several descriptors that might influence the reporting of internal validity items; these included (i) year of authorization (2008, 2010, 2012), (ii) language (English, German, French), (iii) canton (the six largest cantons of Basel, Bern, Freiburg, Geneva, Vaud, Zurich, and the group of the remaining small cantons), (iv) type of institution (academic institutions [i.e., universities, federal institutes of technology, hospitals], industry, governmental institutions [national and cantonal], other [e.g., private institutions, foundations]), (v) animal species (laboratory rodents, higher mammals [CDRP], farm animals, other mammals, non-mammals), (vi) genetically modified animals (yes, no), and (vii) the prospective degree of severity of the planned procedures as defined by the FSVO (0, 1, 2, 3). Inter-rater Reliability Prior to the screening of the selected Form A, two pilot studies on separate applications (i.e., applications authorized in 2009) were conducted to ensure the applicability of the checklist and to ensure consistency of scoring within and between investigators. To ensure consistent scoring of applications between the two investigators, both investigators screened the same 10 applications, and discrepancies were checked at the end of the day. Inter-rater reliability (Eq 3) was assessed at regular intervals (on day 1 and then after the 100th, 300th, 500th, and 700th application on the investigators’ list, respectively) by assessing the proportion of agreement between the two investigators. For this, the first five applications on each investigator’s list were screened by both investigators. (3) Only applications written in either German or English were used for inter-rater reliability tests. Overall, 50 applications were screened twice in the course of these inter-rater reliability tests. Inter-rater reliability never dropped below 85% (S2 Data). Intra-rater Reliability To ensure that both investigators scored applications consistently over time, samples of 10 applications were re-scored at regular intervals (after 50, 150, 350, and 550 listed applications, respectively). In addition, each investigator conducted a final intra-rater reliability test on 10 randomly chosen applications from the whole list after completing the screening procedure. If systematic discrepancies would have occurred, the applications previously scored would have been re-scored. However, as in the case of inter-rater reliability, intra-rater reliability never dropped below 85% (S2 Data). Sample Size Calculation No a priori sample size calculation was performed, as all applications were included in our sample that fulfilled the inclusion/exclusion criteria. However, once the sample size was determined, we verified that it was suitable for the planned statistical analysis (see model description below). Statistics The screening data from the checklists were transferred to a tabulating program (Microsoft Excel 2010.Ink, Redmond, WA, USA) and imported into the statistical software R [57]. We used descriptive statistics to represent reporting rates for individual criteria of internal validity (allocation concealment, blinded outcome assessment, randomization, sample size calculation, inclusion and exclusion criteria, primary outcome, and statistical analysis). Furthermore, influences of relevant descriptors (year, canton, institution, and animal species) were represented graphically, with median and mean IVS of the group, and overall mean IVS. For the statistical analysis of the overall internal validity score of applications, we used generalized linear models to evaluate the influence of the a priori stated descriptors on the internal validity score. The analyses were performed in R [57] using the built in function glm with a binomial error distribution to account for the data structure (primary outcome as proportions). As a first step, we compared univariate models (model with one descriptor) with an intercept-only model (modelling the intercept of the internal validity score) based on significant (p < 0.05) likelihood ratio test of the package lmtest [58] in order to identify descriptors to be included in the further modelling process. The descriptors to be retained were language, canton, species category, institutions, authorization year, and accuracy of the application. In a second step, by means of an information theoretic approach to model selection using the Bayesian Information Criterion (BIC), we identified the model that best fit our data. For an automated model selection procedure, the package MuMln [59] with the function dredge was used to compare all models with all possible combinations of the retained descriptors (full model included also the interaction term for species category and accuracy; see Eq 4). (4) The dredge function ranks all descriptor combinations according to their BIC; the model with the lowest BIC was assumed to be the one representing our data best. The final model included the following main effects (descriptors): language (3 levels), cantons (7 levels), species category (5 levels), accuracy (continuous), institution (4 levels), and authorization year (3 levels). In addition to these main effects, the candidate model included the two-way interaction between species category and accuracy (corresponds to full model, cf. Eq 4). The model parameters were retrieved after correction for over dispersion (see S1 Data). Publications Sampling Process and Screening In order to relate the reporting rates of internal validity criteria assessed here by scoring applications for animal experiments with the reporting rates of such criteria in the published literature [19,22–28,60], we also scored a sub-sample of publications originating from studies based on applications in our study sample. These were identified by searching through grant numbers mentioned in the applications and references listed as output in the annual reports to the FSVO. For 155 applications (12.1%) we identified one or more corresponding publications. This number was reduced to 139 after excluding reviews and publications that were clearly unrelated to the study described in the applications (mismatch in animal species, general topic, or methods). This low number can be explained by the fact that studies licensed in 2012 and also many of those licensed in 2010 were not yet published, and that the search for publications had to rely on grant numbers mentioned in both application and publication (often grant numbers were not mentioned on applications) or on publications listed in the final reports required by the authorities upon completion of licensed studies (for most studies licensed in 2012 and also many of those licensed in 2010, final reports were not yet available). For the comparison of the internal validity scores between applications and publications, we aimed to detect a medium effect size (0.3) with a statistical power of 0.8 at a significance level of p < 0.05. Based on this, we chose a sample size of n = 50, which allowed us to detect an effect size of 0.276 (G*power for correlations, bivariate normal model) [61]. A stratified random sampling procedure was used to select 50 publications from the 139 available publications, so as to select publications derived from a representative sample of all applications with respect to canton and type of animals used. Because this sample of publications was biased towards older applications, we compared the IVS of the sub-sample of 50 applications from which these 50 publications originated with the IVS of the entire sample of applications and found no significant difference; median IVS of the entire sample of applications (n = 1,277) was 0.0 (range 0 to 0.857), compared to 0.0 (range 0 to 0.714) for the sub-sample of applications (n = 50) from which the 50 publications were derived. The publications were screened for reporting of internal validity criteria with a checklist containing the same seven internal validity criteria as were used for applications. The screening of all 50 publications was performed by one single investigator (LV). Publications were randomly allocated to one of the 10 d of screening (five publications per day). Days of screening were separated by two non-screening days. For the publications, descriptors were impact factor of the journal and endorsement of the ARRIVE guidelines by the journal. To determine the descriptors, the impact factor for the year of the publication as well as the ARRIVE status of the journal were assessed. If it was not possible to determine the ARRIVE status of a journal for the date of publication, given that all publications were published in 2012 or later, we used the ARRIVE status of the journal in 2015. Whether or not the ARRIVE status affected the internal validity score of publications was tested with a univariate generalized linear model (binomial error distribution), with IVS as dependent and the descriptor (endorsement of ARRIVE yes or no) as independent variables. (5) Whether or not the internal validity score of publications was correlated with the impact factor of the journal was investigated using a spearman rank correlation test. To ensure that the investigator scored the publications constantly over time, an independent person randomly chose one publication per five publications screened (i.e., one per day of screening) for an intra-rater reliability test. The chosen publication was re-screened on the second following day. The reliability (Eq 3) never dropped below the threshold of 85%. Supporting Information S1 Fig. Distribution of internal validity score. Individual data are shown in https://figshare.com/s/bc48ed5dff9e6ebd2000 (Sample Applications). https://doi.org/10.1371/journal.pbio.2000598.s001 (TIFF) S2 Fig. Comparison of reporting of internal validity criteria between applications and resulting publications. AC: Allocation concealment, BL: Blinding, RA: Randomization,SS: Sample size calculation, IE: Inlcusion/exclusion criteria, PO: Primary outcome, SA: Statistical analysis. Individual data are shown in https://figshare.com/s/bc48ed5dff9e6ebd2000 (Sample Applications and Sample Publications). https://doi.org/10.1371/journal.pbio.2000598.s002 (TIF) S3 Fig. Criteria of inclusion of the application in our study. https://doi.org/10.1371/journal.pbio.2000598.s003 (TIF) S1 Text. Form A. https://doi.org/10.1371/journal.pbio.2000598.s004 (PDF) S2 Text. Checklist. https://doi.org/10.1371/journal.pbio.2000598.s005 (PDF) S3 Text. Annexe to the checklist. https://doi.org/10.1371/journal.pbio.2000598.s006 (PDF) S1 Data. Outcome of the generalized linear model. Output from the generalized linear model used to identify factors influencing IVS of applications. Data are presented with estimate, odds ratios, and the values for the 2.50% quartile and the 97.5% quartile. For more information about the equation we refer to Materials and Methods. https://doi.org/10.1371/journal.pbio.2000598.s007 (XLSX) S2 Data. Outcome of the reliability tests. Outcome of the reliability tests (inter-rater reliability and intra-rater reliability). Percentage agreement and number of discrepancies are available for each item composing the IVS, as well as for the IVS. For more information about the equation we refer to Materials and Methods. https://doi.org/10.1371/journal.pbio.2000598.s008 (XLSX) Acknowledgments The authors wish to thank Heinrich Binder, Sven Süptitz, and Michel Lehmann from the Swiss Food Safety and Veterinary Office (FSVO) for help and support with access to the applications for animal experiments. They are also grateful to the authorities from the veterinary offices of the cantons of Aargau, Luzern, Freiburg, and Basel Stadt for providing additional information about specific applications, and to the authorities of the veterinary office of the canton of Zürich for providing access to their archives. Special thanks go to Emily S. Sena for helpful advice on the scoring of measures against risks of bias and review of the study protocol, and to Beatriz Vidondo for support with statistics.

doi: 10.1371/journal.pbio.2000618pmid: 27918584

Introduction Gene duplications have long been considered to be rare (although some studies contradicted this perception, e.g. [1]), neutral events providing raw genetic material for long-term evolution. However, next-generation sequencing (NGS) technologies have revealed that copy-number variations (CNVs), such as deletions and duplications of genetic material, are widespread in natural populations (review in [2]). Increasing numbers of studies also suggest that CNVs may play a role in adaptation to environmental changes at the micro-evolutionary scale (for a review, [3]; see also [4,5]). In homogeneous gene duplications (also referred to as gene amplifications in cases of successive repeats), the gene copies are identical. They can confer a quantitative advantage in situations in which having larger amounts of the corresponding protein is advantageous. Examples of such coding-gene duplications abound for proteins involved in functions directly related to the environment (e.g., resistance to xenobiotics through higher levels of detoxification, [6–8]; greater amylase production providing adaptation to a starch-rich diet in humans and dogs, [9,10]; higher hexose transporter levels for adaptation to an environment in which resources are limited, [11]). However, homogeneous duplications are also often associated with deleterious pleiotropic effects (or selective costs), probably due to the disruption of biochemical balance or overproduction costs ([12], e.g., in Culex pipiens, esterases may account for up to 12% of total protein, by weight, [13]). By contrast, heterogeneous duplications comprise two different copies of the same gene; they can, thus, provide a more qualitative advantage, through the simultaneous production of two different proteins. It has been suggested that heterogeneous duplications are advantageous in contexts in which the heterozygote genotype is the fittest (i.e., overdominance). Heterogeneous duplications are not affected by the segregation burden carried by standard heterozygotes, and this allows the fixation of the heterozygote phenotype [14,15]. However, empirical evidence of heterogeneous duplications remains scarce, and the role of such duplications in adaptive processes is poorly documented. The few examples described to date concern genes targeted by insecticides: rdl in Drosophila melanogaster [16] and the parallel evolution of the ace-1 locus in the West Nile mosquito C. pipiens and the malaria mosquito An. gambiae, in response to the use of organophosphate (OP) and carbamate (CX) insecticides [17–19]. OPs and CXs target acetylcholinesterase (AChE1), a synaptic enzyme encoded by ace-1 in mosquitoes. The inhibition of this enzyme impairs hydroxylation of the neurotransmitter acetylcholine, inducing death through tetany [20]. Resistance is the consequence of a single-base substitution in the ace-1 gene (ace-1R allele, or R allele), resulting in an amino-acid substitution (G119S) in AChE1 that limits the insecticide binding [21]. This substitution has been selected in several mosquito species exposed to OPs and CXs [21–24]. However, in Cx. pipiens and An. gambiae s. l., the G119S substitution has also been shown to decrease the affinity of the resistant enzyme for its substrate by more than 60% relative to the susceptible version [25,26]. This lower affinity probably underlies the high selective cost of the R allele in both species [27–32]. Several heterogeneous duplications associating a susceptible S and a resistant R ace-1 copy on the same chromosome (D alleles) have been described in natural populations of Cx. pipiens [17,18,33–35]. These duplications restore protein activity whilst maintaining substantial resistance, thus conferring a phenotype similar to that of a standard heterozygote (RS) [36]. A heterogeneous ace-1 duplication has also recently been described in An. gambiae s. l. This duplication occurs in Anopheles coluzzii and An. gambiae s.s. and has spread over a large geographic area [19,37]. Its phenotypic consequences have been shown to be similar to those in Cx. pipiens: an intermediate level of resistance and a large decrease in the selective cost associated with the G119S mutation [32]. It has, therefore, been proposed that heterozygotes (RS) and individuals carrying heterogeneous ace-1 duplications (DD, DS, or DR) may be selected in mosaics of treated and untreated areas, because they probably represent the best resistance/cost trade-off, outperforming cost-free SS susceptible or highly resistant but costly RR homozygotes [17,32,36,38]. Surprisingly, duplications involving several copies of an R allele per chromosome (hereafter Rx, with x the number of copies) have also recently been reported in natural populations of An. gambiae, suggesting that relationships between ace-1 CNVs and resistance to OPs and CXs may be more complex than previously thought [39,40]. However, the impact on fitness of these Rx duplications and the basis of their selection over single-copy alleles and/or heterogeneous duplications according to environmental conditions remain unclear. Given the increasing use of OP and CX insecticides to control malaria mosquitoes, there is an urgent need to determine the roles of the various ace-1 duplications (D and Rx) in resistance. Furthermore, from an evolutionary point of view, this situation provides us with a rare opportunity to determine how the use of different genetic architectures enables an organism to cope with different environmental variations. We present here an integrative study, from the genomic to the phenotypic level, of the role of ace-1 duplications in An. gambiae resistance to OPs and CXs. We first determined the genomic structure of ace-1 duplications, with base-resolution of the breakpoints. We found that the same 203 kb genomic region, encompassing the ace-1 gene and 11 other genes, was amplified in all resistant mosquitoes, through heterogeneous (D) or homogeneous (Rx) duplications. We then investigated the influence of the architecture of duplications and gene-dosage on mosquito fitness. We considered the implications of the results obtained in terms of both potential applications in resistance management, and fundamental evolutionary aspects such as adaptation to a changing environment. Results A Large Chromosomal Segment Encompassing ace-1 Is Recurrently Duplicated in All Resistant An. gambiae Mosquitoes ace-1 is part of a 203 kb tandem heterogeneous duplication containing 12 genes. We characterized the genomic structure of the ace-1 heterogeneous duplication (D allele, carrying one copy of the S allele and one copy of the R allele) by comparing the genomes of two strains, Acerduplikis (DD) and KisumuP (SS). Illumina-generated 250 bp spaced paired-end reads from both strains were first mapped onto the reference An. gambiae PEST genome (VectorBase; AgamP4; [41]). We then calculated the DD to SS ratio of the read depths of coverage (DOC; see Materials and Methods). We identified a clear 2-fold increase in coverage, of ~200 kb, on the 2R chromosomal arm encompassing the ace-1 locus. This observation confirms the duplication of the ace-1 locus but also reveals that this locus is part of a much larger duplicated fragment (Fig 1A). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Genomic structure of the 202.91 kb amplicon encompassing the ace-1 gene. (A) Acerduplikis strain (DD) unique segmental duplication event based on the DD/SS depth of coverage (DOC) ratio (SS corresponds to the KisumuP strain). (B) Acerkis strain (RR) multiple segmental duplication events based on the RR/SS DOC ratio. The pink box corresponds to a decrease in DOC by one third within the amplicon, suggesting that there has been a deletion in only one of the three copies. (C) Gbrowse view of the duplicated region in VectorBase (https://www.vectorbase.org). The duplicated region contains 12 genes: the ace-1 locus (large red arrow) lies about 50 kb from its 5ʹ end, and a gene encoding a transposase overlaps its 3ʹ end (AGAP001368, large black arrow). Genes used as markers to delineate the amplicon are shown as red boxes, with their names underlined in red. Underlying data can be found in NCBI http://www.ncbi.nlm.nih.gov/bioproject/348825. https://doi.org/10.1371/journal.pbio.2000618.g001 This finding was further confirmed by assessing the insert size distribution for read pairs mapping close to the putative duplication breakpoints defined by the DOC ratio, for both the DD and SS strains. As expected, discordant pairs (i.e. with reads mapping ~200 kb apart) were found only in the DD strain (S1A Fig). We investigated whether the two amplicons were in tandem by looking for soft-clipped reads, defined as reads encompassing the junction between amplicons, which would therefore map only partially onto the reference non-duplicated genome. As expected, soft-clipped reads were found only in the DD strain (S1B and S1C Fig). A multiple-sequence alignment of soft-clipped reads and discordant paired-end reads showed that the duplication was 202.91 kb in size (positions 3,436,927 to 3,639,836; Fig 1). This alignment also made it possible to reconstruct the sequences of the breakpoints and, thus, the sequence of the junction between the two amplicons (S1C Fig). The Illumina junction sequence was then confirmed by Sanger sequencing (S1D Fig), which revealed a strict tandem duplication event. Finally, from the VectorBase reference genome, the duplication appeared to contain 12 genes (S1 Table), beginning with ace-1 itself (~50 kb from the 5' breakpoint), whereas the AGAP001368 locus encoding the Harbinger transposase overlapped the 3' breakpoint (Fig 1C). [RR] individuals, including those from the Acerkis reference strain, also display duplication. We used the junction sequence to develop a diagnostic PCR duplication test: each of the primers used (Agduplispedir2 and AgduplispeRev1, S1D Fig and S2 Table) binds to a different amplicon. This PCR amplifies a 460 bp fragment overlapping the junction in the Acerduplikis (DD) strain. It is therefore specific for individuals carrying the duplication. Ten An. gambiae field populations from Benin, Burkina Faso, Togo, and Ivory Coast (Table 1) were screened. All mosquitoes were first typed by ace-1 PCR-RLFP, which distinguishes between the [SS], [RS], and [RR] phenotypes (it does not, however, distinguish between RS, DD, DS, and DR genotypes, which all provide [RS] phenotype) [22]. They were then tested with the diagnostic duplication test. Download: PPT PowerPoint slide PNG larger image TIFF original image Table 1. Screening for the presence (+) or absence (-) of the ace-1 duplication in West African An. gambiae populations. https://doi.org/10.1371/journal.pbio.2000618.t001 No amplicons were obtained in diagnostic duplication tests on [SS] individuals (Table 1), consistent with the presence of the S allele as a single copy in susceptible individuals. However, a 460 bp fragment similar to that detected in the DD strain was amplified from all [RS] individuals and, unexpectedly, from the 53 [RR] individuals from Baguida and Natitingou (Table 1). Even more surprisingly, this fragment was also amplified from the 20 individuals of the Acerkis [RR] resistant strain tested. Thus, all [RR] phenotypes in An. gambiae seem to have at least two R copies within tandem amplicons. We estimated the number of ace-1 copies in Acerkis [RR] mosquitoes by analyzing 32 mosquitoes from this strain by real-time quantitative PCR (qPCR). These mosquitoes had three times as many ace-1 copies (3.21 ± 0.18) as individuals from the KisumuP (SS) strain (1.03 ± 0.09; LM, t = 61, df = 62, p < 0.001, Fig 2). They thus had three ace-1 R copies per chromosome. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Number of ace-1 copies in KisumuP, AcerkisR3, AgRR5 strains and in the Baguida [RR] population. Boxes indicate ace-1 to Rps7 gene concentration ratios (advanced relative quantification method, LightCycler 480 software 1.5.0). The significance of ratio differences between strains is indicated (***, p < 0.001). Underlying data can be found in DRYAD http://dx.doi.org/10.5061/dryad.4f7qg. https://doi.org/10.1371/journal.pbio.2000618.g002 Similar structures for homogeneous R duplications and heterogeneous D duplications. Acerkis [RR] individuals were sequenced with Illumina technology (see Methods), for the precise characterization of their amplicons. We analyzed the breakpoints and junctions, as described above. As expected, the DOC RR/SS ratio around the ace-1 locus was about three times that of the adjacent regions (Fig 1B). Only one sequence was retrieved for the ace-1 locus itself. These ace-1R duplications were, thus, homogeneous: the same R copy was repeated. The allele carried by these Acerkis individuals consisted of three R copies, within amplicons organized strictly in tandem (as in the D allele). We will therefore refer to this allele as R3, and, as Acerkis individuals are homozygous, their genotype will hereafter be indicated by R3R3 and the strain used here renamed AcerKisR3 (the number of ace-1 copies when the strain was isolated cannot be determined; S3 Table). Discordant (relative to the PEST reference genome) read pairs were found at exactly the same locations as for the D allele. Thus, all the amplicons, regardless of whether they carried the R (in R3 or D) or S (in D) allele, presented identical breakpoints (Fig 1). However, the RR/SS DOC ratio was not uniform along the amplicons: an internal area not encompassing the ace-1 locus, in which this ratio was one third lower than elsewhere in the amplicon, suggested that a deletion had occurred in one of the three amplicons. Estimation on the basis of the discordant paired-end reads suggested that this deletion covered 97 kb (from 3,502,148–3,598,855, Fig 1B and 1C). Detailed analysis of the intra-amplicon deletion showed that it disrupted two genes (AGAP001357 and AGAP001367) and completely deleted eight others (AGAP001358; AGAP001360-1366); only ace-1 remained complete in the amplicon with the deletion (the twelfth gene, AGAP001368, was disrupted during the amplification process, Fig 1C). Finally, a comparison of the Illumina amplicon sequences of R3R3 and DD individuals revealed the presence of several single-nucleotide polymorphisms (SNPs) at the 3ʹ-end of the DD amplicons, whereas the R3R3 amplicons were monomorphic at these positions. All these DD SNPs were a mixture of two bases, one of which was identical to that found in R3R3. Assuming that the polymorphism in DD was due to the presence of the S amplicon, we designed a pair of PCR primers (AgRDdir1 and AgRDrev1) to amplify a fragment containing this polymorphic region and the junction, to make it possible to determine the orientation of the duplication (S2 Table). The Sanger sequences of this fragment were strictly identical in R3R3 and DD individuals. Thus, in the D allele, the amplicon containing the R copy is positioned upstream from both the junction and the amplicon containing the S copy (S1B Fig). Variable numbers of ace-1 copies in the field. We used real-time qPCR to determine the number of ace-1 copies in 39 [RR] mosquitoes from the Baguida (Togo) field population (Fig 2). There were 3.3 to 9.1 ace-1 copies per individual. In a diploid, this corresponds to 1.67 to 4.57 copies per chromosome, with a median of about three R copies (median = 3.1, Fig 2). However, the distribution of copies per chromosome is unknown. We investigated whether the ace-1 amplicons segregating in field populations were similar to those identified in our strains, by quantifying, in individuals from Baguida (Togo) and from Tiassale and Bouake (Ivory Coast), the numbers of copies for two loci located just outside the amplified region (AGAP001355 and AGAP001369, referred to as 5'out and 3'out, respectively, see Methods, Fig 1C). All the individuals tested had only one copy of each of these genes, suggesting that their amplicons were similar in size to those of the DD and R3R3 strains. Fitness associated with homogeneous R duplications The fitness impact of the D allele, the heterogeneous duplication pairing a susceptible, and a resistant copy of the ace-1 gene has been assessed elsewhere [32]. The D allele was found to confer a lower level of resistance than the R allele, but it almost completely resorbed the high fitness cost associated with the G119S mutation for the various traits studied [32]. Surprisingly, we show here that the R allele used in this previous study (from the AcerkisR3 strain) is actually a homogeneous duplication of three copies of R. Consistent with other recent studies [39,40], we also found that the number of copies of R in [RR] individuals was variable in natural populations (Fig 2). We thus investigated the phenotypic and fitness consequences of different amplicon architectures and gene dosages. AChE1R activity and insecticide resistance increase with the number of R copies. We first investigated the relationships between AChE1 activity and the number of R and S copies in various ace-1 genotypes. We measured the activities of the resistant (AR) and susceptible (AS) forms of the enzyme in 40 mosquitoes (20 of each sex) of the KisumuP (SS), AcerkisR3 (R3R3), and Acerduplikis (DD) strains, and their F1 offspring (R3S, DS, and DR3, see S3 Table). AChE1 activities were consistently lower in females than in males (t = 3.7, df = 219, p < 0.001), so we analyzed the two sexes separately. The AChE1 activity corresponding to one resistant copy (measured in R3R3 individuals as total activity divided by six) was one-quarter to one-fifth (0.21 ± 0.07 and 0.23 ± 0.07, for females and males, respectively) that of a single susceptible copy (measured in SS individuals), confirming the decrease in AChE1 activity associated with the G119S mutation [26]. In both sexes, highly significant correlations were found between AS and the number of S copies in the genotype (Pearson’s correlation coefficient: r = 0.82, t = 16, df = 117, p < 0.001 and r = 0.84, t = 15, df = 96, p < 0.001, for females and males, respectively), and between AR and the number of R copies in the genotype (r = 0.96, t = 37, df = 117, p < 0.001 and r = 0.93, t = 26, df = 96, p < 0.001, for females and males, respectively; S2 Fig). We assessed the relationships between ace-1 R copy-number variations and mosquito AChE1R activity (Ar) on the one hand, and insecticide resistance on the other, by exposing a first batch of larvae from the Baguida field population to 2 x 10−2 mg/l chlorpyrifos methyl (leading to 30% mortality) and another batch of larvae to 4 x 10−2 mg/l (leading to 80% mortality). We analyzed 30 emerging adults from each batch and showed that AChE1R activity was significantly higher (1.5 ± 0.4 times) in larvae surviving exposure to 4 x 10−2 mg/l chlorpyrifos methyl than in larvae surviving 2 x 10−2 mg/l (Student’s t test: t = -5.8, df = 53, p < 0.001, Fig 3A). We then measured the number of R copies (Nc) in 30 [RR] survivors (15 from each batch) for which AChE1R activity had previously been measured. There was a significant positive correlation between individual AChE1R activity and the number of R copies (GLM: Ar = Nc + ε, with ε the error parameter (Gaussian distribution), t = 2.7, df = 28, p = 0.01, Fig 3B). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Relationship between AChE1 activity and insecticide resistance, or the number of R copies. (A) Boxplots present the distribution of AChE1R activity for [RR] individuals selected at low (0.02 mg/l, n = 30) and high (0.04 mg/l, n = 30) doses of chlorpyrifos methyl. ***: Student’s t test, p < 0.001. (B) Regression analysis showing a significant positive relationship (GLM, *: p < 0.05) between AChE1R activity and the number of R copies (as the distribution by chromosome is unknown, the total number of R copies is given for each individual). Underlying data can be found in DRYAD http://dx.doi.org/10.5061/dryad.4f7qg. https://doi.org/10.1371/journal.pbio.2000618.g003 More precise investigations of the effects of the number of R copies on fitness required a strain carrying a homogeneous duplicated allele with more R copies than R3 (AcerkisR3). Baguida larvae were selected by exposure to 2 x 10−2 mg/l chlorpyrifos methyl. Surviving females were crossed with SS males and allowed to lay eggs individually. We used the PCR-RFLP test [22] to phenotype these females for the ace-1 locus, to identify [RR] females. We determined the number of ace-1 copies in these females by qPCR. For females carrying more ace-1 copies than AcerkisR3 individuals, we determined the number of ace-1 copies in six second-instar larvae from their progenies. We found identical copy numbers in all six larvae (i.e., indicating that the mother was homozygous) in only one of the 70 progenies tested. These larvae all had six ace-1 copies; as they were [RS], they therefore carried five ace-1 copies on the same chromosome (i.e., an R5 allele). The rest of this progeny was backcrossed five times successively with SS males (KisumuP) to homogenize the genetic background. This strain thus had a genetic background similar to those of KisumuP (SS) and AcerkisR3 (R3R3). Mosquitoes were then crossed with each other, to fix the homozygous resistant phenotype [RR], to generate the AgRR5 strain. Quantitative PCR was used to confirm that the individuals of this strain were all of genotype R5R5 (n = 30 individuals, 4.93 ± 0.42 R copies per chromosome, Fig 2). A comparison between the two [RR] strains showed that R5R5 individuals had AChE1R activity levels 1.5 ± 0.1 times those of R3R3, close to the expected value assuming strict additivity according to the number of copies present (1.7, S3 Fig). Bioassays were then carried out on larvae from the three strains (SS, R3R3 and R5R5), with different insecticides: one CX (bendiocarb), one OP (chlorpyrifos methyl) and one pyrethroid (PYR, permethrin). Mortality in the controls never exceeded 5%. SS, R3R3 (RR50 = 0.97, p > 0.05), and R5R5 (RR50 = 0.99, p > 0.05) mosquitoes were all susceptible to permethrin, confirming that the kdr alleles present in the Baguida field population had been eliminated during the backcrosses (Table 2). For the two insecticides targeting AChE1, bendiocarb and chlorpyriphos-methyl, the R5R5 strain displayed significantly higher resistance than R3R3 (RR50 = 290 versus 207, p < 0.001 and RR50 = 14 versus 12, p < 0.001, respectively) (Table 2 and S4 Fig). Download: PPT PowerPoint slide PNG larger image TIFF original image Table 2. Dose-mortality responses to various insecticides observed in the various Anopheles gambiae s. s. strains. https://doi.org/10.1371/journal.pbio.2000618.t002 Increasing the number of R copies increases selective costs. We assessed the fitness cost associated with the different ace-1 genotypes by comparing several life history traits between KisumuP (SS), AcerkisR3 (R3R3), and AgRR5 (R5R5). We assessed pre-imaginal mortality from egg hatching to adult emergence. We recorded the number of dead larvae at each developmental stage, to assess differences between the strains in overall mortality and mortality dynamics. Overall mortality was highest for R5R5 individuals (mR5R5 = 0.73 [0.63–0.82]; the 95% confidence intervals are given in brackets), followed by R3R3 (mR3R3 = 0.64 [0.53–0.73]), with much higher survival rates for SS (mSS = 0.22 [0.15–0.32]). The SS genotype had a significantly different mortality pattern (Cox model: SS versus R3R3, z = 5.3, p < 0.001; SS versus R5R5, z = 6.4, p < 0.001), with a lower mortality at each larval stage, the patterns being similar for the R3R3 and R5R5 genotypes (Cox model: z = 1.5, p = 0.14; Fig 4A). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. Life history traits for the KisumuP (SS), AcerkisR3 (R3R3) and AgRR5 (R5R5) strains. (A) Larval mortality: the proportion of larvae surviving at each development stage is presented, from hatching to emergence (Li is the larval stage i); "+" indicates the proportion of emerged adults. (B) Development time: the proportion of emerged adults on each day after the start of the experiment is presented for each genotype; arrows indicate the mean development time of each genotype. (C) Female fecundity: the mean number of larvae per female in each strain is presented with its standard error; the significance of differences in fertility is indicated (n.s., p > 0.05; **, p < 0.01). Underlying data can be found in DRYAD http://dx.doi.org/10.5061/dryad.4f7qg. https://doi.org/10.1371/journal.pbio.2000618.g004 Development time was recorded as the number of days required for a first-instar larva to reach adulthood (i.e., the time until emergence). We detected no interaction between sex and genotype (GLM, likelihood ratio test [LRT]: χ2 = 2.64, Δdf = 2, p = 0.26) and no sex effect (8.8 ± 1.5 d for males and 8.98 ± 1.7 d for females; LRT: χ2 = 0.4, Δdf = 1, p = 0.55). However, R3R3 and R5R5 individuals had similar development times (10.11 ± 1.38 and 10.53 ± 1.36 days, respectively; Cox model: z = -0.9, p = 0.38), with both developing significantly more slowly than SS individuals (7.74 ± 0.43 days; Cox model: z = -7.78, p < 0.001 and z = -8, p < 0.001; Fig 4B). We assessed the influence of R copy-number variation on female reproductive success by allowing 40 females of each genotype (SS, R3R3 and R5R5) to lay eggs. Overall reproductive success did not differ significantly between R3R3 and R5R5 females (mean ± SE = 27 ± 3.2 and 22 ± 4.3 larvae per female, respectively; GLM: F = 0.64, Δdf = 1, p = 0.43) but was significantly lower for both these genotypes than for SS females (43 ± 5.2 larvae per female; GLM: F = 6.8, Δdf = 2, p < 0.01; Fig 4C). The observed differences were due solely to R3R3 and R5R5 females laying fewer eggs than SS females (GLM: F = 12.2, Δdf = 2, p < 0.001; S5A Fig), as neither the proportion of females laying eggs nor the hatching rate per female differed significantly between the three genotypes (GLM: χ2 = 1.5, Δdf = 2, p = 0.47, S5B Fig, and F = 0.93, Δdf = 2, p = 0.40, S5C Fig, respectively). Overall, the performance of R5R5 mosquitoes did not differ significantly from those of R3R3 mosquitoes for any of the development, mortality, or fecundity traits analyzed. Nevertheless, the mean performances of R5R5 mosquitoes were always slightly lower than those of R3R3 mosquitoes for all these traits (Fig 4), suggesting that R5R5 individuals may actually be subjected to slightly higher costs than R3R3 individuals. We carried out an experimental evolution study, taking the whole life cycle into account, to confirm this trend. Competition between the R3 and R5 alleles was established by crossing 250 females of the AcerkisR3 strain (R3R3) with 250 males from the AgRR5 strain (R5R5). Their F1 progeny (all R3R5) was reared in standard conditions (27 ± 2°C, 80 ± 2% humidity and 12h:12h light/dark cycle), in the absence of insecticide. After emergence, the adults were released into a new cage and allowed to reproduce freely for five discrete generations. Three replicates (C1, C2 and C3) were set up. Total AChE1 activity, which is correlated with ace-1 copy number (S3 Fig), was used to assess the change in the proportions of the R3 and R5 alleles in the cages. For each replicate, we measured total AChE1 activity for 32 individuals from each generation: two measurements were made per individual, to limit measurement error, and only females were analyzed to avoid a sex effect. Five individuals each of the two reference strains, AcerkisR3 (R3R3) and AgRR5 (R5R5), were used as controls. For each individual, an activity index (AI) was constructed as follows: AI = (Ax - AR3) / (AR5 - AR3), where Ax is the mean total activity of individual x, and AR3 and AR5 are the mean total activities estimated for the control genotypes. An AI close to 0 corresponds to an AChE1 activity similar to that of R3R3 individuals, whereas a value close to 1 corresponds to an AChE1 activity similar to that of R5R5 individuals. We followed the change in mean AI index in each replicate across the five discrete generations. We used the following GLM to determine whether AI changed over generations: AI = Gen + ε, where Gen is a five-level factor corresponding to the generations and ε is a Gaussian error parameter. In all replicates, the mean activity index (AI) decreased significantly, from 0.53 ± 0.07 to 0.23 ± 0.10, between G1 and G5 (GLM: t = -2.13, df = 163, p < 0.05; t = -6.21, df = 163, p < 0.001 and t = -3.15, df = 163, p < 0.01; for C1, C2, and C3, respectively). This result suggests that the R5 allele tends to be eliminated by R3 (S6 Fig), consistent with the trend previously observed for individual life history traits. A Large Chromosomal Segment Encompassing ace-1 Is Recurrently Duplicated in All Resistant An. gambiae Mosquitoes ace-1 is part of a 203 kb tandem heterogeneous duplication containing 12 genes. We characterized the genomic structure of the ace-1 heterogeneous duplication (D allele, carrying one copy of the S allele and one copy of the R allele) by comparing the genomes of two strains, Acerduplikis (DD) and KisumuP (SS). Illumina-generated 250 bp spaced paired-end reads from both strains were first mapped onto the reference An. gambiae PEST genome (VectorBase; AgamP4; [41]). We then calculated the DD to SS ratio of the read depths of coverage (DOC; see Materials and Methods). We identified a clear 2-fold increase in coverage, of ~200 kb, on the 2R chromosomal arm encompassing the ace-1 locus. This observation confirms the duplication of the ace-1 locus but also reveals that this locus is part of a much larger duplicated fragment (Fig 1A). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Genomic structure of the 202.91 kb amplicon encompassing the ace-1 gene. (A) Acerduplikis strain (DD) unique segmental duplication event based on the DD/SS depth of coverage (DOC) ratio (SS corresponds to the KisumuP strain). (B) Acerkis strain (RR) multiple segmental duplication events based on the RR/SS DOC ratio. The pink box corresponds to a decrease in DOC by one third within the amplicon, suggesting that there has been a deletion in only one of the three copies. (C) Gbrowse view of the duplicated region in VectorBase (https://www.vectorbase.org). The duplicated region contains 12 genes: the ace-1 locus (large red arrow) lies about 50 kb from its 5ʹ end, and a gene encoding a transposase overlaps its 3ʹ end (AGAP001368, large black arrow). Genes used as markers to delineate the amplicon are shown as red boxes, with their names underlined in red. Underlying data can be found in NCBI http://www.ncbi.nlm.nih.gov/bioproject/348825. https://doi.org/10.1371/journal.pbio.2000618.g001 This finding was further confirmed by assessing the insert size distribution for read pairs mapping close to the putative duplication breakpoints defined by the DOC ratio, for both the DD and SS strains. As expected, discordant pairs (i.e. with reads mapping ~200 kb apart) were found only in the DD strain (S1A Fig). We investigated whether the two amplicons were in tandem by looking for soft-clipped reads, defined as reads encompassing the junction between amplicons, which would therefore map only partially onto the reference non-duplicated genome. As expected, soft-clipped reads were found only in the DD strain (S1B and S1C Fig). A multiple-sequence alignment of soft-clipped reads and discordant paired-end reads showed that the duplication was 202.91 kb in size (positions 3,436,927 to 3,639,836; Fig 1). This alignment also made it possible to reconstruct the sequences of the breakpoints and, thus, the sequence of the junction between the two amplicons (S1C Fig). The Illumina junction sequence was then confirmed by Sanger sequencing (S1D Fig), which revealed a strict tandem duplication event. Finally, from the VectorBase reference genome, the duplication appeared to contain 12 genes (S1 Table), beginning with ace-1 itself (~50 kb from the 5' breakpoint), whereas the AGAP001368 locus encoding the Harbinger transposase overlapped the 3' breakpoint (Fig 1C). [RR] individuals, including those from the Acerkis reference strain, also display duplication. We used the junction sequence to develop a diagnostic PCR duplication test: each of the primers used (Agduplispedir2 and AgduplispeRev1, S1D Fig and S2 Table) binds to a different amplicon. This PCR amplifies a 460 bp fragment overlapping the junction in the Acerduplikis (DD) strain. It is therefore specific for individuals carrying the duplication. Ten An. gambiae field populations from Benin, Burkina Faso, Togo, and Ivory Coast (Table 1) were screened. All mosquitoes were first typed by ace-1 PCR-RLFP, which distinguishes between the [SS], [RS], and [RR] phenotypes (it does not, however, distinguish between RS, DD, DS, and DR genotypes, which all provide [RS] phenotype) [22]. They were then tested with the diagnostic duplication test. Download: PPT PowerPoint slide PNG larger image TIFF original image Table 1. Screening for the presence (+) or absence (-) of the ace-1 duplication in West African An. gambiae populations. https://doi.org/10.1371/journal.pbio.2000618.t001 No amplicons were obtained in diagnostic duplication tests on [SS] individuals (Table 1), consistent with the presence of the S allele as a single copy in susceptible individuals. However, a 460 bp fragment similar to that detected in the DD strain was amplified from all [RS] individuals and, unexpectedly, from the 53 [RR] individuals from Baguida and Natitingou (Table 1). Even more surprisingly, this fragment was also amplified from the 20 individuals of the Acerkis [RR] resistant strain tested. Thus, all [RR] phenotypes in An. gambiae seem to have at least two R copies within tandem amplicons. We estimated the number of ace-1 copies in Acerkis [RR] mosquitoes by analyzing 32 mosquitoes from this strain by real-time quantitative PCR (qPCR). These mosquitoes had three times as many ace-1 copies (3.21 ± 0.18) as individuals from the KisumuP (SS) strain (1.03 ± 0.09; LM, t = 61, df = 62, p < 0.001, Fig 2). They thus had three ace-1 R copies per chromosome. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Number of ace-1 copies in KisumuP, AcerkisR3, AgRR5 strains and in the Baguida [RR] population. Boxes indicate ace-1 to Rps7 gene concentration ratios (advanced relative quantification method, LightCycler 480 software 1.5.0). The significance of ratio differences between strains is indicated (***, p < 0.001). Underlying data can be found in DRYAD http://dx.doi.org/10.5061/dryad.4f7qg. https://doi.org/10.1371/journal.pbio.2000618.g002 Similar structures for homogeneous R duplications and heterogeneous D duplications. Acerkis [RR] individuals were sequenced with Illumina technology (see Methods), for the precise characterization of their amplicons. We analyzed the breakpoints and junctions, as described above. As expected, the DOC RR/SS ratio around the ace-1 locus was about three times that of the adjacent regions (Fig 1B). Only one sequence was retrieved for the ace-1 locus itself. These ace-1R duplications were, thus, homogeneous: the same R copy was repeated. The allele carried by these Acerkis individuals consisted of three R copies, within amplicons organized strictly in tandem (as in the D allele). We will therefore refer to this allele as R3, and, as Acerkis individuals are homozygous, their genotype will hereafter be indicated by R3R3 and the strain used here renamed AcerKisR3 (the number of ace-1 copies when the strain was isolated cannot be determined; S3 Table). Discordant (relative to the PEST reference genome) read pairs were found at exactly the same locations as for the D allele. Thus, all the amplicons, regardless of whether they carried the R (in R3 or D) or S (in D) allele, presented identical breakpoints (Fig 1). However, the RR/SS DOC ratio was not uniform along the amplicons: an internal area not encompassing the ace-1 locus, in which this ratio was one third lower than elsewhere in the amplicon, suggested that a deletion had occurred in one of the three amplicons. Estimation on the basis of the discordant paired-end reads suggested that this deletion covered 97 kb (from 3,502,148–3,598,855, Fig 1B and 1C). Detailed analysis of the intra-amplicon deletion showed that it disrupted two genes (AGAP001357 and AGAP001367) and completely deleted eight others (AGAP001358; AGAP001360-1366); only ace-1 remained complete in the amplicon with the deletion (the twelfth gene, AGAP001368, was disrupted during the amplification process, Fig 1C). Finally, a comparison of the Illumina amplicon sequences of R3R3 and DD individuals revealed the presence of several single-nucleotide polymorphisms (SNPs) at the 3ʹ-end of the DD amplicons, whereas the R3R3 amplicons were monomorphic at these positions. All these DD SNPs were a mixture of two bases, one of which was identical to that found in R3R3. Assuming that the polymorphism in DD was due to the presence of the S amplicon, we designed a pair of PCR primers (AgRDdir1 and AgRDrev1) to amplify a fragment containing this polymorphic region and the junction, to make it possible to determine the orientation of the duplication (S2 Table). The Sanger sequences of this fragment were strictly identical in R3R3 and DD individuals. Thus, in the D allele, the amplicon containing the R copy is positioned upstream from both the junction and the amplicon containing the S copy (S1B Fig). Variable numbers of ace-1 copies in the field. We used real-time qPCR to determine the number of ace-1 copies in 39 [RR] mosquitoes from the Baguida (Togo) field population (Fig 2). There were 3.3 to 9.1 ace-1 copies per individual. In a diploid, this corresponds to 1.67 to 4.57 copies per chromosome, with a median of about three R copies (median = 3.1, Fig 2). However, the distribution of copies per chromosome is unknown. We investigated whether the ace-1 amplicons segregating in field populations were similar to those identified in our strains, by quantifying, in individuals from Baguida (Togo) and from Tiassale and Bouake (Ivory Coast), the numbers of copies for two loci located just outside the amplified region (AGAP001355 and AGAP001369, referred to as 5'out and 3'out, respectively, see Methods, Fig 1C). All the individuals tested had only one copy of each of these genes, suggesting that their amplicons were similar in size to those of the DD and R3R3 strains. ace-1 is part of a 203 kb tandem heterogeneous duplication containing 12 genes. We characterized the genomic structure of the ace-1 heterogeneous duplication (D allele, carrying one copy of the S allele and one copy of the R allele) by comparing the genomes of two strains, Acerduplikis (DD) and KisumuP (SS). Illumina-generated 250 bp spaced paired-end reads from both strains were first mapped onto the reference An. gambiae PEST genome (VectorBase; AgamP4; [41]). We then calculated the DD to SS ratio of the read depths of coverage (DOC; see Materials and Methods). We identified a clear 2-fold increase in coverage, of ~200 kb, on the 2R chromosomal arm encompassing the ace-1 locus. This observation confirms the duplication of the ace-1 locus but also reveals that this locus is part of a much larger duplicated fragment (Fig 1A). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Genomic structure of the 202.91 kb amplicon encompassing the ace-1 gene. (A) Acerduplikis strain (DD) unique segmental duplication event based on the DD/SS depth of coverage (DOC) ratio (SS corresponds to the KisumuP strain). (B) Acerkis strain (RR) multiple segmental duplication events based on the RR/SS DOC ratio. The pink box corresponds to a decrease in DOC by one third within the amplicon, suggesting that there has been a deletion in only one of the three copies. (C) Gbrowse view of the duplicated region in VectorBase (https://www.vectorbase.org). The duplicated region contains 12 genes: the ace-1 locus (large red arrow) lies about 50 kb from its 5ʹ end, and a gene encoding a transposase overlaps its 3ʹ end (AGAP001368, large black arrow). Genes used as markers to delineate the amplicon are shown as red boxes, with their names underlined in red. Underlying data can be found in NCBI http://www.ncbi.nlm.nih.gov/bioproject/348825. https://doi.org/10.1371/journal.pbio.2000618.g001 This finding was further confirmed by assessing the insert size distribution for read pairs mapping close to the putative duplication breakpoints defined by the DOC ratio, for both the DD and SS strains. As expected, discordant pairs (i.e. with reads mapping ~200 kb apart) were found only in the DD strain (S1A Fig). We investigated whether the two amplicons were in tandem by looking for soft-clipped reads, defined as reads encompassing the junction between amplicons, which would therefore map only partially onto the reference non-duplicated genome. As expected, soft-clipped reads were found only in the DD strain (S1B and S1C Fig). A multiple-sequence alignment of soft-clipped reads and discordant paired-end reads showed that the duplication was 202.91 kb in size (positions 3,436,927 to 3,639,836; Fig 1). This alignment also made it possible to reconstruct the sequences of the breakpoints and, thus, the sequence of the junction between the two amplicons (S1C Fig). The Illumina junction sequence was then confirmed by Sanger sequencing (S1D Fig), which revealed a strict tandem duplication event. Finally, from the VectorBase reference genome, the duplication appeared to contain 12 genes (S1 Table), beginning with ace-1 itself (~50 kb from the 5' breakpoint), whereas the AGAP001368 locus encoding the Harbinger transposase overlapped the 3' breakpoint (Fig 1C). [RR] individuals, including those from the Acerkis reference strain, also display duplication. We used the junction sequence to develop a diagnostic PCR duplication test: each of the primers used (Agduplispedir2 and AgduplispeRev1, S1D Fig and S2 Table) binds to a different amplicon. This PCR amplifies a 460 bp fragment overlapping the junction in the Acerduplikis (DD) strain. It is therefore specific for individuals carrying the duplication. Ten An. gambiae field populations from Benin, Burkina Faso, Togo, and Ivory Coast (Table 1) were screened. All mosquitoes were first typed by ace-1 PCR-RLFP, which distinguishes between the [SS], [RS], and [RR] phenotypes (it does not, however, distinguish between RS, DD, DS, and DR genotypes, which all provide [RS] phenotype) [22]. They were then tested with the diagnostic duplication test. Download: PPT PowerPoint slide PNG larger image TIFF original image Table 1. Screening for the presence (+) or absence (-) of the ace-1 duplication in West African An. gambiae populations. https://doi.org/10.1371/journal.pbio.2000618.t001 No amplicons were obtained in diagnostic duplication tests on [SS] individuals (Table 1), consistent with the presence of the S allele as a single copy in susceptible individuals. However, a 460 bp fragment similar to that detected in the DD strain was amplified from all [RS] individuals and, unexpectedly, from the 53 [RR] individuals from Baguida and Natitingou (Table 1). Even more surprisingly, this fragment was also amplified from the 20 individuals of the Acerkis [RR] resistant strain tested. Thus, all [RR] phenotypes in An. gambiae seem to have at least two R copies within tandem amplicons. We estimated the number of ace-1 copies in Acerkis [RR] mosquitoes by analyzing 32 mosquitoes from this strain by real-time quantitative PCR (qPCR). These mosquitoes had three times as many ace-1 copies (3.21 ± 0.18) as individuals from the KisumuP (SS) strain (1.03 ± 0.09; LM, t = 61, df = 62, p < 0.001, Fig 2). They thus had three ace-1 R copies per chromosome. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Number of ace-1 copies in KisumuP, AcerkisR3, AgRR5 strains and in the Baguida [RR] population. Boxes indicate ace-1 to Rps7 gene concentration ratios (advanced relative quantification method, LightCycler 480 software 1.5.0). The significance of ratio differences between strains is indicated (***, p < 0.001). Underlying data can be found in DRYAD http://dx.doi.org/10.5061/dryad.4f7qg. https://doi.org/10.1371/journal.pbio.2000618.g002 Similar structures for homogeneous R duplications and heterogeneous D duplications. Acerkis [RR] individuals were sequenced with Illumina technology (see Methods), for the precise characterization of their amplicons. We analyzed the breakpoints and junctions, as described above. As expected, the DOC RR/SS ratio around the ace-1 locus was about three times that of the adjacent regions (Fig 1B). Only one sequence was retrieved for the ace-1 locus itself. These ace-1R duplications were, thus, homogeneous: the same R copy was repeated. The allele carried by these Acerkis individuals consisted of three R copies, within amplicons organized strictly in tandem (as in the D allele). We will therefore refer to this allele as R3, and, as Acerkis individuals are homozygous, their genotype will hereafter be indicated by R3R3 and the strain used here renamed AcerKisR3 (the number of ace-1 copies when the strain was isolated cannot be determined; S3 Table). Discordant (relative to the PEST reference genome) read pairs were found at exactly the same locations as for the D allele. Thus, all the amplicons, regardless of whether they carried the R (in R3 or D) or S (in D) allele, presented identical breakpoints (Fig 1). However, the RR/SS DOC ratio was not uniform along the amplicons: an internal area not encompassing the ace-1 locus, in which this ratio was one third lower than elsewhere in the amplicon, suggested that a deletion had occurred in one of the three amplicons. Estimation on the basis of the discordant paired-end reads suggested that this deletion covered 97 kb (from 3,502,148–3,598,855, Fig 1B and 1C). Detailed analysis of the intra-amplicon deletion showed that it disrupted two genes (AGAP001357 and AGAP001367) and completely deleted eight others (AGAP001358; AGAP001360-1366); only ace-1 remained complete in the amplicon with the deletion (the twelfth gene, AGAP001368, was disrupted during the amplification process, Fig 1C). Finally, a comparison of the Illumina amplicon sequences of R3R3 and DD individuals revealed the presence of several single-nucleotide polymorphisms (SNPs) at the 3ʹ-end of the DD amplicons, whereas the R3R3 amplicons were monomorphic at these positions. All these DD SNPs were a mixture of two bases, one of which was identical to that found in R3R3. Assuming that the polymorphism in DD was due to the presence of the S amplicon, we designed a pair of PCR primers (AgRDdir1 and AgRDrev1) to amplify a fragment containing this polymorphic region and the junction, to make it possible to determine the orientation of the duplication (S2 Table). The Sanger sequences of this fragment were strictly identical in R3R3 and DD individuals. Thus, in the D allele, the amplicon containing the R copy is positioned upstream from both the junction and the amplicon containing the S copy (S1B Fig). Variable numbers of ace-1 copies in the field. We used real-time qPCR to determine the number of ace-1 copies in 39 [RR] mosquitoes from the Baguida (Togo) field population (Fig 2). There were 3.3 to 9.1 ace-1 copies per individual. In a diploid, this corresponds to 1.67 to 4.57 copies per chromosome, with a median of about three R copies (median = 3.1, Fig 2). However, the distribution of copies per chromosome is unknown. We investigated whether the ace-1 amplicons segregating in field populations were similar to those identified in our strains, by quantifying, in individuals from Baguida (Togo) and from Tiassale and Bouake (Ivory Coast), the numbers of copies for two loci located just outside the amplified region (AGAP001355 and AGAP001369, referred to as 5'out and 3'out, respectively, see Methods, Fig 1C). All the individuals tested had only one copy of each of these genes, suggesting that their amplicons were similar in size to those of the DD and R3R3 strains. Fitness associated with homogeneous R duplications The fitness impact of the D allele, the heterogeneous duplication pairing a susceptible, and a resistant copy of the ace-1 gene has been assessed elsewhere [32]. The D allele was found to confer a lower level of resistance than the R allele, but it almost completely resorbed the high fitness cost associated with the G119S mutation for the various traits studied [32]. Surprisingly, we show here that the R allele used in this previous study (from the AcerkisR3 strain) is actually a homogeneous duplication of three copies of R. Consistent with other recent studies [39,40], we also found that the number of copies of R in [RR] individuals was variable in natural populations (Fig 2). We thus investigated the phenotypic and fitness consequences of different amplicon architectures and gene dosages. AChE1R activity and insecticide resistance increase with the number of R copies. We first investigated the relationships between AChE1 activity and the number of R and S copies in various ace-1 genotypes. We measured the activities of the resistant (AR) and susceptible (AS) forms of the enzyme in 40 mosquitoes (20 of each sex) of the KisumuP (SS), AcerkisR3 (R3R3), and Acerduplikis (DD) strains, and their F1 offspring (R3S, DS, and DR3, see S3 Table). AChE1 activities were consistently lower in females than in males (t = 3.7, df = 219, p < 0.001), so we analyzed the two sexes separately. The AChE1 activity corresponding to one resistant copy (measured in R3R3 individuals as total activity divided by six) was one-quarter to one-fifth (0.21 ± 0.07 and 0.23 ± 0.07, for females and males, respectively) that of a single susceptible copy (measured in SS individuals), confirming the decrease in AChE1 activity associated with the G119S mutation [26]. In both sexes, highly significant correlations were found between AS and the number of S copies in the genotype (Pearson’s correlation coefficient: r = 0.82, t = 16, df = 117, p < 0.001 and r = 0.84, t = 15, df = 96, p < 0.001, for females and males, respectively), and between AR and the number of R copies in the genotype (r = 0.96, t = 37, df = 117, p < 0.001 and r = 0.93, t = 26, df = 96, p < 0.001, for females and males, respectively; S2 Fig). We assessed the relationships between ace-1 R copy-number variations and mosquito AChE1R activity (Ar) on the one hand, and insecticide resistance on the other, by exposing a first batch of larvae from the Baguida field population to 2 x 10−2 mg/l chlorpyrifos methyl (leading to 30% mortality) and another batch of larvae to 4 x 10−2 mg/l (leading to 80% mortality). We analyzed 30 emerging adults from each batch and showed that AChE1R activity was significantly higher (1.5 ± 0.4 times) in larvae surviving exposure to 4 x 10−2 mg/l chlorpyrifos methyl than in larvae surviving 2 x 10−2 mg/l (Student’s t test: t = -5.8, df = 53, p < 0.001, Fig 3A). We then measured the number of R copies (Nc) in 30 [RR] survivors (15 from each batch) for which AChE1R activity had previously been measured. There was a significant positive correlation between individual AChE1R activity and the number of R copies (GLM: Ar = Nc + ε, with ε the error parameter (Gaussian distribution), t = 2.7, df = 28, p = 0.01, Fig 3B). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Relationship between AChE1 activity and insecticide resistance, or the number of R copies. (A) Boxplots present the distribution of AChE1R activity for [RR] individuals selected at low (0.02 mg/l, n = 30) and high (0.04 mg/l, n = 30) doses of chlorpyrifos methyl. ***: Student’s t test, p < 0.001. (B) Regression analysis showing a significant positive relationship (GLM, *: p < 0.05) between AChE1R activity and the number of R copies (as the distribution by chromosome is unknown, the total number of R copies is given for each individual). Underlying data can be found in DRYAD http://dx.doi.org/10.5061/dryad.4f7qg. https://doi.org/10.1371/journal.pbio.2000618.g003 More precise investigations of the effects of the number of R copies on fitness required a strain carrying a homogeneous duplicated allele with more R copies than R3 (AcerkisR3). Baguida larvae were selected by exposure to 2 x 10−2 mg/l chlorpyrifos methyl. Surviving females were crossed with SS males and allowed to lay eggs individually. We used the PCR-RFLP test [22] to phenotype these females for the ace-1 locus, to identify [RR] females. We determined the number of ace-1 copies in these females by qPCR. For females carrying more ace-1 copies than AcerkisR3 individuals, we determined the number of ace-1 copies in six second-instar larvae from their progenies. We found identical copy numbers in all six larvae (i.e., indicating that the mother was homozygous) in only one of the 70 progenies tested. These larvae all had six ace-1 copies; as they were [RS], they therefore carried five ace-1 copies on the same chromosome (i.e., an R5 allele). The rest of this progeny was backcrossed five times successively with SS males (KisumuP) to homogenize the genetic background. This strain thus had a genetic background similar to those of KisumuP (SS) and AcerkisR3 (R3R3). Mosquitoes were then crossed with each other, to fix the homozygous resistant phenotype [RR], to generate the AgRR5 strain. Quantitative PCR was used to confirm that the individuals of this strain were all of genotype R5R5 (n = 30 individuals, 4.93 ± 0.42 R copies per chromosome, Fig 2). A comparison between the two [RR] strains showed that R5R5 individuals had AChE1R activity levels 1.5 ± 0.1 times those of R3R3, close to the expected value assuming strict additivity according to the number of copies present (1.7, S3 Fig). Bioassays were then carried out on larvae from the three strains (SS, R3R3 and R5R5), with different insecticides: one CX (bendiocarb), one OP (chlorpyrifos methyl) and one pyrethroid (PYR, permethrin). Mortality in the controls never exceeded 5%. SS, R3R3 (RR50 = 0.97, p > 0.05), and R5R5 (RR50 = 0.99, p > 0.05) mosquitoes were all susceptible to permethrin, confirming that the kdr alleles present in the Baguida field population had been eliminated during the backcrosses (Table 2). For the two insecticides targeting AChE1, bendiocarb and chlorpyriphos-methyl, the R5R5 strain displayed significantly higher resistance than R3R3 (RR50 = 290 versus 207, p < 0.001 and RR50 = 14 versus 12, p < 0.001, respectively) (Table 2 and S4 Fig). Download: PPT PowerPoint slide PNG larger image TIFF original image Table 2. Dose-mortality responses to various insecticides observed in the various Anopheles gambiae s. s. strains. https://doi.org/10.1371/journal.pbio.2000618.t002 Increasing the number of R copies increases selective costs. We assessed the fitness cost associated with the different ace-1 genotypes by comparing several life history traits between KisumuP (SS), AcerkisR3 (R3R3), and AgRR5 (R5R5). We assessed pre-imaginal mortality from egg hatching to adult emergence. We recorded the number of dead larvae at each developmental stage, to assess differences between the strains in overall mortality and mortality dynamics. Overall mortality was highest for R5R5 individuals (mR5R5 = 0.73 [0.63–0.82]; the 95% confidence intervals are given in brackets), followed by R3R3 (mR3R3 = 0.64 [0.53–0.73]), with much higher survival rates for SS (mSS = 0.22 [0.15–0.32]). The SS genotype had a significantly different mortality pattern (Cox model: SS versus R3R3, z = 5.3, p < 0.001; SS versus R5R5, z = 6.4, p < 0.001), with a lower mortality at each larval stage, the patterns being similar for the R3R3 and R5R5 genotypes (Cox model: z = 1.5, p = 0.14; Fig 4A). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. Life history traits for the KisumuP (SS), AcerkisR3 (R3R3) and AgRR5 (R5R5) strains. (A) Larval mortality: the proportion of larvae surviving at each development stage is presented, from hatching to emergence (Li is the larval stage i); "+" indicates the proportion of emerged adults. (B) Development time: the proportion of emerged adults on each day after the start of the experiment is presented for each genotype; arrows indicate the mean development time of each genotype. (C) Female fecundity: the mean number of larvae per female in each strain is presented with its standard error; the significance of differences in fertility is indicated (n.s., p > 0.05; **, p < 0.01). Underlying data can be found in DRYAD http://dx.doi.org/10.5061/dryad.4f7qg. https://doi.org/10.1371/journal.pbio.2000618.g004 Development time was recorded as the number of days required for a first-instar larva to reach adulthood (i.e., the time until emergence). We detected no interaction between sex and genotype (GLM, likelihood ratio test [LRT]: χ2 = 2.64, Δdf = 2, p = 0.26) and no sex effect (8.8 ± 1.5 d for males and 8.98 ± 1.7 d for females; LRT: χ2 = 0.4, Δdf = 1, p = 0.55). However, R3R3 and R5R5 individuals had similar development times (10.11 ± 1.38 and 10.53 ± 1.36 days, respectively; Cox model: z = -0.9, p = 0.38), with both developing significantly more slowly than SS individuals (7.74 ± 0.43 days; Cox model: z = -7.78, p < 0.001 and z = -8, p < 0.001; Fig 4B). We assessed the influence of R copy-number variation on female reproductive success by allowing 40 females of each genotype (SS, R3R3 and R5R5) to lay eggs. Overall reproductive success did not differ significantly between R3R3 and R5R5 females (mean ± SE = 27 ± 3.2 and 22 ± 4.3 larvae per female, respectively; GLM: F = 0.64, Δdf = 1, p = 0.43) but was significantly lower for both these genotypes than for SS females (43 ± 5.2 larvae per female; GLM: F = 6.8, Δdf = 2, p < 0.01; Fig 4C). The observed differences were due solely to R3R3 and R5R5 females laying fewer eggs than SS females (GLM: F = 12.2, Δdf = 2, p < 0.001; S5A Fig), as neither the proportion of females laying eggs nor the hatching rate per female differed significantly between the three genotypes (GLM: χ2 = 1.5, Δdf = 2, p = 0.47, S5B Fig, and F = 0.93, Δdf = 2, p = 0.40, S5C Fig, respectively). Overall, the performance of R5R5 mosquitoes did not differ significantly from those of R3R3 mosquitoes for any of the development, mortality, or fecundity traits analyzed. Nevertheless, the mean performances of R5R5 mosquitoes were always slightly lower than those of R3R3 mosquitoes for all these traits (Fig 4), suggesting that R5R5 individuals may actually be subjected to slightly higher costs than R3R3 individuals. We carried out an experimental evolution study, taking the whole life cycle into account, to confirm this trend. Competition between the R3 and R5 alleles was established by crossing 250 females of the AcerkisR3 strain (R3R3) with 250 males from the AgRR5 strain (R5R5). Their F1 progeny (all R3R5) was reared in standard conditions (27 ± 2°C, 80 ± 2% humidity and 12h:12h light/dark cycle), in the absence of insecticide. After emergence, the adults were released into a new cage and allowed to reproduce freely for five discrete generations. Three replicates (C1, C2 and C3) were set up. Total AChE1 activity, which is correlated with ace-1 copy number (S3 Fig), was used to assess the change in the proportions of the R3 and R5 alleles in the cages. For each replicate, we measured total AChE1 activity for 32 individuals from each generation: two measurements were made per individual, to limit measurement error, and only females were analyzed to avoid a sex effect. Five individuals each of the two reference strains, AcerkisR3 (R3R3) and AgRR5 (R5R5), were used as controls. For each individual, an activity index (AI) was constructed as follows: AI = (Ax - AR3) / (AR5 - AR3), where Ax is the mean total activity of individual x, and AR3 and AR5 are the mean total activities estimated for the control genotypes. An AI close to 0 corresponds to an AChE1 activity similar to that of R3R3 individuals, whereas a value close to 1 corresponds to an AChE1 activity similar to that of R5R5 individuals. We followed the change in mean AI index in each replicate across the five discrete generations. We used the following GLM to determine whether AI changed over generations: AI = Gen + ε, where Gen is a five-level factor corresponding to the generations and ε is a Gaussian error parameter. In all replicates, the mean activity index (AI) decreased significantly, from 0.53 ± 0.07 to 0.23 ± 0.10, between G1 and G5 (GLM: t = -2.13, df = 163, p < 0.05; t = -6.21, df = 163, p < 0.001 and t = -3.15, df = 163, p < 0.01; for C1, C2, and C3, respectively). This result suggests that the R5 allele tends to be eliminated by R3 (S6 Fig), consistent with the trend previously observed for individual life history traits. AChE1R activity and insecticide resistance increase with the number of R copies. We first investigated the relationships between AChE1 activity and the number of R and S copies in various ace-1 genotypes. We measured the activities of the resistant (AR) and susceptible (AS) forms of the enzyme in 40 mosquitoes (20 of each sex) of the KisumuP (SS), AcerkisR3 (R3R3), and Acerduplikis (DD) strains, and their F1 offspring (R3S, DS, and DR3, see S3 Table). AChE1 activities were consistently lower in females than in males (t = 3.7, df = 219, p < 0.001), so we analyzed the two sexes separately. The AChE1 activity corresponding to one resistant copy (measured in R3R3 individuals as total activity divided by six) was one-quarter to one-fifth (0.21 ± 0.07 and 0.23 ± 0.07, for females and males, respectively) that of a single susceptible copy (measured in SS individuals), confirming the decrease in AChE1 activity associated with the G119S mutation [26]. In both sexes, highly significant correlations were found between AS and the number of S copies in the genotype (Pearson’s correlation coefficient: r = 0.82, t = 16, df = 117, p < 0.001 and r = 0.84, t = 15, df = 96, p < 0.001, for females and males, respectively), and between AR and the number of R copies in the genotype (r = 0.96, t = 37, df = 117, p < 0.001 and r = 0.93, t = 26, df = 96, p < 0.001, for females and males, respectively; S2 Fig). We assessed the relationships between ace-1 R copy-number variations and mosquito AChE1R activity (Ar) on the one hand, and insecticide resistance on the other, by exposing a first batch of larvae from the Baguida field population to 2 x 10−2 mg/l chlorpyrifos methyl (leading to 30% mortality) and another batch of larvae to 4 x 10−2 mg/l (leading to 80% mortality). We analyzed 30 emerging adults from each batch and showed that AChE1R activity was significantly higher (1.5 ± 0.4 times) in larvae surviving exposure to 4 x 10−2 mg/l chlorpyrifos methyl than in larvae surviving 2 x 10−2 mg/l (Student’s t test: t = -5.8, df = 53, p < 0.001, Fig 3A). We then measured the number of R copies (Nc) in 30 [RR] survivors (15 from each batch) for which AChE1R activity had previously been measured. There was a significant positive correlation between individual AChE1R activity and the number of R copies (GLM: Ar = Nc + ε, with ε the error parameter (Gaussian distribution), t = 2.7, df = 28, p = 0.01, Fig 3B). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Relationship between AChE1 activity and insecticide resistance, or the number of R copies. (A) Boxplots present the distribution of AChE1R activity for [RR] individuals selected at low (0.02 mg/l, n = 30) and high (0.04 mg/l, n = 30) doses of chlorpyrifos methyl. ***: Student’s t test, p < 0.001. (B) Regression analysis showing a significant positive relationship (GLM, *: p < 0.05) between AChE1R activity and the number of R copies (as the distribution by chromosome is unknown, the total number of R copies is given for each individual). Underlying data can be found in DRYAD http://dx.doi.org/10.5061/dryad.4f7qg. https://doi.org/10.1371/journal.pbio.2000618.g003 More precise investigations of the effects of the number of R copies on fitness required a strain carrying a homogeneous duplicated allele with more R copies than R3 (AcerkisR3). Baguida larvae were selected by exposure to 2 x 10−2 mg/l chlorpyrifos methyl. Surviving females were crossed with SS males and allowed to lay eggs individually. We used the PCR-RFLP test [22] to phenotype these females for the ace-1 locus, to identify [RR] females. We determined the number of ace-1 copies in these females by qPCR. For females carrying more ace-1 copies than AcerkisR3 individuals, we determined the number of ace-1 copies in six second-instar larvae from their progenies. We found identical copy numbers in all six larvae (i.e., indicating that the mother was homozygous) in only one of the 70 progenies tested. These larvae all had six ace-1 copies; as they were [RS], they therefore carried five ace-1 copies on the same chromosome (i.e., an R5 allele). The rest of this progeny was backcrossed five times successively with SS males (KisumuP) to homogenize the genetic background. This strain thus had a genetic background similar to those of KisumuP (SS) and AcerkisR3 (R3R3). Mosquitoes were then crossed with each other, to fix the homozygous resistant phenotype [RR], to generate the AgRR5 strain. Quantitative PCR was used to confirm that the individuals of this strain were all of genotype R5R5 (n = 30 individuals, 4.93 ± 0.42 R copies per chromosome, Fig 2). A comparison between the two [RR] strains showed that R5R5 individuals had AChE1R activity levels 1.5 ± 0.1 times those of R3R3, close to the expected value assuming strict additivity according to the number of copies present (1.7, S3 Fig). Bioassays were then carried out on larvae from the three strains (SS, R3R3 and R5R5), with different insecticides: one CX (bendiocarb), one OP (chlorpyrifos methyl) and one pyrethroid (PYR, permethrin). Mortality in the controls never exceeded 5%. SS, R3R3 (RR50 = 0.97, p > 0.05), and R5R5 (RR50 = 0.99, p > 0.05) mosquitoes were all susceptible to permethrin, confirming that the kdr alleles present in the Baguida field population had been eliminated during the backcrosses (Table 2). For the two insecticides targeting AChE1, bendiocarb and chlorpyriphos-methyl, the R5R5 strain displayed significantly higher resistance than R3R3 (RR50 = 290 versus 207, p < 0.001 and RR50 = 14 versus 12, p < 0.001, respectively) (Table 2 and S4 Fig). Download: PPT PowerPoint slide PNG larger image TIFF original image Table 2. Dose-mortality responses to various insecticides observed in the various Anopheles gambiae s. s. strains. https://doi.org/10.1371/journal.pbio.2000618.t002 Increasing the number of R copies increases selective costs. We assessed the fitness cost associated with the different ace-1 genotypes by comparing several life history traits between KisumuP (SS), AcerkisR3 (R3R3), and AgRR5 (R5R5). We assessed pre-imaginal mortality from egg hatching to adult emergence. We recorded the number of dead larvae at each developmental stage, to assess differences between the strains in overall mortality and mortality dynamics. Overall mortality was highest for R5R5 individuals (mR5R5 = 0.73 [0.63–0.82]; the 95% confidence intervals are given in brackets), followed by R3R3 (mR3R3 = 0.64 [0.53–0.73]), with much higher survival rates for SS (mSS = 0.22 [0.15–0.32]). The SS genotype had a significantly different mortality pattern (Cox model: SS versus R3R3, z = 5.3, p < 0.001; SS versus R5R5, z = 6.4, p < 0.001), with a lower mortality at each larval stage, the patterns being similar for the R3R3 and R5R5 genotypes (Cox model: z = 1.5, p = 0.14; Fig 4A). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. Life history traits for the KisumuP (SS), AcerkisR3 (R3R3) and AgRR5 (R5R5) strains. (A) Larval mortality: the proportion of larvae surviving at each development stage is presented, from hatching to emergence (Li is the larval stage i); "+" indicates the proportion of emerged adults. (B) Development time: the proportion of emerged adults on each day after the start of the experiment is presented for each genotype; arrows indicate the mean development time of each genotype. (C) Female fecundity: the mean number of larvae per female in each strain is presented with its standard error; the significance of differences in fertility is indicated (n.s., p > 0.05; **, p < 0.01). Underlying data can be found in DRYAD http://dx.doi.org/10.5061/dryad.4f7qg. https://doi.org/10.1371/journal.pbio.2000618.g004 Development time was recorded as the number of days required for a first-instar larva to reach adulthood (i.e., the time until emergence). We detected no interaction between sex and genotype (GLM, likelihood ratio test [LRT]: χ2 = 2.64, Δdf = 2, p = 0.26) and no sex effect (8.8 ± 1.5 d for males and 8.98 ± 1.7 d for females; LRT: χ2 = 0.4, Δdf = 1, p = 0.55). However, R3R3 and R5R5 individuals had similar development times (10.11 ± 1.38 and 10.53 ± 1.36 days, respectively; Cox model: z = -0.9, p = 0.38), with both developing significantly more slowly than SS individuals (7.74 ± 0.43 days; Cox model: z = -7.78, p < 0.001 and z = -8, p < 0.001; Fig 4B). We assessed the influence of R copy-number variation on female reproductive success by allowing 40 females of each genotype (SS, R3R3 and R5R5) to lay eggs. Overall reproductive success did not differ significantly between R3R3 and R5R5 females (mean ± SE = 27 ± 3.2 and 22 ± 4.3 larvae per female, respectively; GLM: F = 0.64, Δdf = 1, p = 0.43) but was significantly lower for both these genotypes than for SS females (43 ± 5.2 larvae per female; GLM: F = 6.8, Δdf = 2, p < 0.01; Fig 4C). The observed differences were due solely to R3R3 and R5R5 females laying fewer eggs than SS females (GLM: F = 12.2, Δdf = 2, p < 0.001; S5A Fig), as neither the proportion of females laying eggs nor the hatching rate per female differed significantly between the three genotypes (GLM: χ2 = 1.5, Δdf = 2, p = 0.47, S5B Fig, and F = 0.93, Δdf = 2, p = 0.40, S5C Fig, respectively). Overall, the performance of R5R5 mosquitoes did not differ significantly from those of R3R3 mosquitoes for any of the development, mortality, or fecundity traits analyzed. Nevertheless, the mean performances of R5R5 mosquitoes were always slightly lower than those of R3R3 mosquitoes for all these traits (Fig 4), suggesting that R5R5 individuals may actually be subjected to slightly higher costs than R3R3 individuals. We carried out an experimental evolution study, taking the whole life cycle into account, to confirm this trend. Competition between the R3 and R5 alleles was established by crossing 250 females of the AcerkisR3 strain (R3R3) with 250 males from the AgRR5 strain (R5R5). Their F1 progeny (all R3R5) was reared in standard conditions (27 ± 2°C, 80 ± 2% humidity and 12h:12h light/dark cycle), in the absence of insecticide. After emergence, the adults were released into a new cage and allowed to reproduce freely for five discrete generations. Three replicates (C1, C2 and C3) were set up. Total AChE1 activity, which is correlated with ace-1 copy number (S3 Fig), was used to assess the change in the proportions of the R3 and R5 alleles in the cages. For each replicate, we measured total AChE1 activity for 32 individuals from each generation: two measurements were made per individual, to limit measurement error, and only females were analyzed to avoid a sex effect. Five individuals each of the two reference strains, AcerkisR3 (R3R3) and AgRR5 (R5R5), were used as controls. For each individual, an activity index (AI) was constructed as follows: AI = (Ax - AR3) / (AR5 - AR3), where Ax is the mean total activity of individual x, and AR3 and AR5 are the mean total activities estimated for the control genotypes. An AI close to 0 corresponds to an AChE1 activity similar to that of R3R3 individuals, whereas a value close to 1 corresponds to an AChE1 activity similar to that of R5R5 individuals. We followed the change in mean AI index in each replicate across the five discrete generations. We used the following GLM to determine whether AI changed over generations: AI = Gen + ε, where Gen is a five-level factor corresponding to the generations and ε is a Gaussian error parameter. In all replicates, the mean activity index (AI) decreased significantly, from 0.53 ± 0.07 to 0.23 ± 0.10, between G1 and G5 (GLM: t = -2.13, df = 163, p < 0.05; t = -6.21, df = 163, p < 0.001 and t = -3.15, df = 163, p < 0.01; for C1, C2, and C3, respectively). This result suggests that the R5 allele tends to be eliminated by R3 (S6 Fig), consistent with the trend previously observed for individual life history traits. Discussion Recent NGS studies have revealed that CNVs are pervasive in natural populations. Altering the number of copies of a gene is generally thought to be deleterious, although some CNVs have been shown to be adaptive. For example, contemporary duplications of the ace-1 gene underlying resistance to organophosphate (OP) and carbamate (CX) insecticides have been observed in several mosquito species (Cx. pipiens, An. gambiae, and An. albimanus, [18,37,42]), in one moth species (Plutella xylostella, [43]), and in two spider-mite species (Tetranychus urticae and Tetranychus evansi, [44,45]). The number of ace-1 copies is variable (up to five in T. urticae and T. evansi), but duplications are usually heterogeneous, involving at least one susceptible copy (S) and one resistant copy (R). An. gambiae is unusual in that both heterogeneous and homogeneous (with only R copies) duplications have been selected independently and segregate in natural populations ([19,39], this study). ace-1 Gene Duplications Are Pervasive in An. gambiae Populations By resolving the genomic structure of these duplications, we were able to design a diagnostic test for duplications that revealed that ace-1 was systematically duplicated in resistant mosquitoes, but never in susceptible mosquitoes (there were 173 resistant individuals among the 398 mosquitoes from four sub-Saharan African countries tested). The 200-kb amplicon is also found in specimens from Burkina Faso and Guinea sequenced by the An. gambiae 1,000 Genomes Consortium (manuscript in preparation) and occurs only in specimens carrying the resistance allele. This confirms that ace-1 duplications are adaptive, whether heterogeneous (with one S and one R copy) or homogeneous (with multiple R copies), and pervasive in natural populations of An. gambiae. Genomic analysis also revealed that the ace-1 locus was part of the same 203 kb amplicon in both kinds of duplications. Amplicons are arranged strictly in tandem (i.e., contiguous to each other), with identical breakpoints and junctions, down to individual base level. Heterogeneous duplications contained only one S copy and one R copy, whereas the number of R copies was more variable in homogeneous duplications (up to five copies were detected). OPs and CXs were introduced relatively recently (over the last 50 y) for the control of malaria vector populations. The number and diversity of the duplicated alleles suggest that these duplications are relatively common events. This conclusion is consistent with recent studies showing that the rate of gene duplication per gene and per generation ranges from a value similar to the substitution rate to a value four orders of magnitude higher [46–49]. It has been suggested that such duplications are promoted by the presence of repeated elements [50,51], which favor unequal crossing-over events by inducing chromatin mismatching. Indeed, several studies have reported the presence of transposable elements at the breakpoints of large adaptive segmental duplications [4,16,52]. We also identified a transposable element (the Harbinger transposase) at the 3ʹ end of the amplicons. The non-allelic homologous recombination model (NAHR [53]) is, thus, probably the most parsimonious mechanism explaining this high frequency of duplication events with conserved breakpoints around the ace-1 amplicons. All Resistance Alleles Share the Same 203 kb Amplicon, Precluding Specific Genotyping It was not possible to design a PCR-based molecular test specific for the ace-1 R copy to differentiate between heterogeneous and homozygous duplications, as the Rx and D alleles shared the same R copy, identical to the only R copy described to date [19,21]. Genomic structure was also conserved among homogeneous and heterogeneous duplications, precluding the development of specific DNA markers to distinguish between amplicons containing Rx and D alleles on the basis of breakpoints or junction sequences. The diagnostic duplication test developed in this study can thus be used only to determine whether a mosquito carries a duplication of some kind. Furthermore, while the amplicons carrying the S or R copy in the D allele had several diagnostic SNPs at their 3' ends, they were identical at their 5ʹ ends, as the amplicon carrying the R copy was located upstream from that carrying the S copy, making it impossible to develop a PCR test specific for the R-S association. Finally, while it is possible to evaluate the number of ace-1 copies with high precision by qPCR, there is currently no method for determining the distribution of copies between homologous chromosomes. Thus, if qPCR and PCR-RFLP ace-1 tests on a particular [RS] individual indicate the presence of four ace-1 copies, for example, it is not possible to determine whether the genotype of this individual is DD, DR2, or R3S. Indirect methods, such as coupling molecular tests with extensive crossing experiments, thus remain the only way to determine the precise genotype, and such methods are unsuitable for use in large field screenings. In our study, we used such indirect methods to genotype the offspring of individuals crossed with the reference SS strain. The quantification of ace-1 copy number by this approach was precise enough to determine the exact genotype when one of the chromosomes was known (here, the chromosome carrying S). This made it possible to establish strains with various numbers of R copies, to investigate the effect of ace-1 CNVs on the phenotype and fitness of mosquitoes. Different Genomic Architectures Result in Different Fitness Trade-Offs We previously investigated the phenotypic and fitness impacts of heterogeneous duplications of ace-1 in An. gambiae [32]. We showed that the D allele confers a phenotype resembling that of standard (RS) heterozygotes, with an intermediate level of resistance and a lower fitness cost. In this study, we showed that acetylcholinesterase activity, regardless of the form of the enzyme (susceptible [AChE1S] or resistant [AChE1R]), is globally proportional to the number of S or R copies, respectively, in the genotype (S2 Fig). Consequently, the D allele has a higher level of activity than a single R copy, because the G119S mutation greatly decreases AChE1 activity. The partial restoration of AChE1 activity induced by the combination of R and S on a single chromosome probably accounts for the lower fitness cost and the intermediate level of resistance of the D allele [32]. The D allele thus provides a new and intermediate alternative to the otherwise irreducible trade-off between protein activity and resistance level (S: no cost, no resistance; R: high cost, high resistance). By contrast to heterogeneous duplications, which combine qualitatively different alleles, homogeneous duplications usually provide a quantitative advantage through repeats of the same allele (see the review by [3]). In the case of ace-1, the total AChE1 activity provided by Rx alleles appears to be positively correlated with the number of R copies (i.e., the number of amplicons). However, no quantitative advantage in terms of resistance was anticipated; resistance was thought to be a binary trait conditioned solely by target protein conformation [21]. The finding that resistance level increased with the number of R copies (R5R5 individuals clearly displayed higher resistance than R3R3 individuals, S4 Fig) was therefore surprising, indicating that resistance is more quantitative than was previously thought (Fig 3). However, the relationship between cost and copy number may not be monotonic. First, life history traits and experimental evolution studies comparing R5 and R3 alleles suggested that a larger number of R copies was associated with a higher cost (Fig 4 and S6 Fig). Conversely, this relationship may not hold if there are fewer than three R copies: (i) three R copies per chromosome were found in the AcerkisR3 strain [RR], despite the absence of insecticide selection over a number of generations (at least 4 y; if carrying fewer copies was less costly, this number would be expected to decrease with time in the absence of selection); (ii) the median number of copies in the Baguida field population (Togo) was three, although only a few [RR] individuals with different numbers of copies were found. It thus seems that three R copies on the same chromosome may be optimal in terms of the associated cost, with carrying fewer or more copies entailing higher fitness costs. So how can we explain this surprising situation? Where do these costs come from? It is possible that total cost is a combination of at least two different, opposite functions correlated with amplification level (Fig 5). First, the cost associated with the low level of AChE1R activity (cost of the G119S mutation) probably decreases with increasing numbers of R copies (e.g., the number of amplicons). The paucity of individuals with only one- or two-copy alleles in the field suggests a high cost; conversely, the total activity provided by the R5 allele is higher than that of the R3 allele, but less than that of the susceptible allele (no overshoot cost). The second cost function would increase with the number of amplicons: R5 appears to be more costly overall than R3. As the two alleles have the same amplicon and structure, this cost is probably metabolic rather than structural in nature. First, as AChE1 is involved in numerous functions at different stages of development (for a review see [54]), the production of larger amounts of AChE1 may have a deleterious effect on these other functions (i.e., a pleiotropic effect). The second possibility, which is probably more likely, is that this cost is not directly related to ace-1 itself, instead being due to changes in the dosage of one or several of the other genes encompassed by the duplication (i.e., gene dosage cost, Fig 5). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. Hypothetical model to explain the non-monotonous relationship between fitness cost and the number of R copies, and the minimum cost centered on three copies per chromosome. Under this hypothesis, the overall fitness cost (red triangles) is the result of two opposite functions: the cost associated with G119S mutation (light gray circles) and the cost due to the change in gene dosage (dark gray squares). The first of these costs is negatively correlated with the number of R copies. Thus, larger numbers of R copies are associated with lower costs. By contrast, the second cost is positively correlated with the number of R copies. Note that the shapes of each function and their interaction are unknown. https://doi.org/10.1371/journal.pbio.2000618.g005 The two functions are probably not symmetric, as the activity of the R5 allele was similar to the expected value under an additive model, whereas the increase in cost relative to R3 was barely visible in terms of life history traits. This hypothetical combination of two cost functions is consistent with the observations and suggests that the R3 allele may be optimal (Fig 5). Internal Deletions May Reduce the Gene-Dosage Disruption Cost Genome analyses revealed that the 203-kb amplicon contained 12 genes, only one of which (ace-1) is involved in resistance to OPs and CXs (S2 Table). Amplicon size may be constrained by the presence of the Harbinger element, favoring the occurrence of large duplications encompassing the other 11 loci. However, these genes are probably only hitchhiking, so their duplication is unlikely to be adaptive. By contrast to ace-1, an increase in their gene-dosage may actually be deleterious: (i) it could alter biochemical equilibria between the duplicated genes and the single-copy genes with which they interact [55,56]; (ii) the function of these duplicated genes may be disrupted because optimal levels of protein may be overshot [57,58]; or (iii) the superfluous production of excess protein from the duplicated genes may be energetically costly [59]. These three hypotheses are not mutually exclusive. We report here observations supporting the notion that duplication costs are due, at least in part, to gene-dosage imbalance between the co-amplified genes rather than directly to ace-1. In R3R3 individuals (AcerKis), DOC decreased within some of the amplicons (Fig 1B), revealing 97 kb intra-amplicon deletions affecting the 11 loci other than ace-1. In these individuals, only ace-1 remains fully amplified, the number of copies of the other genes having decreased towards their initial dosage (Fig 1B). Moreover, this internal deletion was recurrent and also detected in mosquitoes from Burkina Faso and Guinea sequenced by the An. gambiae 1,000 Genomes Consortium (manuscript in preparation). These observations suggest that the intra-amplicon deletions occurred after the duplication events. More importantly, they suggest that these secondary deletions are probably adaptive and were selected to reduce the cost of dosage imbalance for the 11 genes concerned. These deletions would alter the shape of the gene-dosage cost function (Fig 5). This process is probably still ongoing, and it should be possible to test this adaptive hypothesis by comparing fitness between individuals carrying the same number of R copies with and without these deletions. The Genomic Architecture Selected for Resistance Alleles Probably Depends on the Intensity of Selective Pressure, and This Must Be Taken into Account in Vector Control The key finding of this study is that the various architectures of the ace-1 amplicon correspond to different evolutionary trade-offs between resistance and costs, and that they all segregate in An. gambiae field populations. The reason for this coexistence of different alleles with different fitness trade-offs probably lies in the diversity of treatment practices for vector control. This diversity often results in mosaic environments, composed of patches covering the whole gradient from untreated to intensely treated areas, leading to the selection of different alleles. Moderate treatments or fine-grained alternation of treated and untreated areas (over time and/or space) should favor the heterozygous phenotype, with moderate resistance and costs (overdominance and marginal overdominance, respectively). As the D allele does not bear the segregation burden of standard heterozygotes, this heterogeneous duplication should, thus, be favored in this context [17,32,36]. This may be the case in An. gambiae populations from the Ivory Coast, which contained only [RS] individuals, all displaying amplification (Tiassalé and Bouaké, n = 43 and 38, respectively), suggesting that D is present at high frequency, if not already fixed (Table 1). By contrast, higher insecticide doses and coarse-grained environments should select more specialist phenotypes, and, thus, homogeneous duplications with higher levels of resistance (despite their higher costs). Moreover, higher doses should select for larger numbers of copies (R5 confers stronger resistance than R3). In this respect, variations in the level of amplification in homogeneous duplications should allow a more finely tuned response to variations in selection intensity, as the quantitative effects of such duplications on fitness seem to be more subtle than those of heterogeneous duplication. The demonstration of this versatility in the trade-offs available through duplication represents a major development in our understanding of the evolutionary processes of adaptation. However, the consequences, in terms of vector control, are much more negative. Following the massive increase in PYR insecticide resistance in African populations of An. gambiae, several African countries have tried to preserve the efficacy of vector control by switching from PYRs to CXs or OPs, in accordance with the American President’s Malaria Initiative [60], in collaboration with the National Malaria Control Program [61]. However, duplicated ace-1 resistance alleles are already widespread. Some populations contained only resistant individuals, with either 100% [RS] individuals (suggesting the presence of the D allele, Tiassalé and Bouaké, Ivory Coast) or with mostly [RR] individuals (83% carrying RX alleles, Baguida, Togo). Very careful management of the doses of insecticide used, and of their spatial and temporal application, will be required to control resistance, as treatment could rapidly lead to the selection of different types of ace-1 alleles, hampering mosquito control. Such fine-tuning of treatment may prove difficult in a context in which insecticides are used for vector control and various compounds used in agriculture can also select ace-1 resistance alleles (for a review [62]). This issue is particularly pressing because OPs and CXs have been recommended in the context of a high frequency of PYR resistance alleles, but recent studies have reported alarming synergic effects between resistance alleles specific for each insecticide class [63,64]. The high adaptability conferred by ace-1 duplications may have a major impact on An. gambiae vector control in Africa, potentially impeding the control of malaria transmission. ace-1 Gene Duplications Are Pervasive in An. gambiae Populations By resolving the genomic structure of these duplications, we were able to design a diagnostic test for duplications that revealed that ace-1 was systematically duplicated in resistant mosquitoes, but never in susceptible mosquitoes (there were 173 resistant individuals among the 398 mosquitoes from four sub-Saharan African countries tested). The 200-kb amplicon is also found in specimens from Burkina Faso and Guinea sequenced by the An. gambiae 1,000 Genomes Consortium (manuscript in preparation) and occurs only in specimens carrying the resistance allele. This confirms that ace-1 duplications are adaptive, whether heterogeneous (with one S and one R copy) or homogeneous (with multiple R copies), and pervasive in natural populations of An. gambiae. Genomic analysis also revealed that the ace-1 locus was part of the same 203 kb amplicon in both kinds of duplications. Amplicons are arranged strictly in tandem (i.e., contiguous to each other), with identical breakpoints and junctions, down to individual base level. Heterogeneous duplications contained only one S copy and one R copy, whereas the number of R copies was more variable in homogeneous duplications (up to five copies were detected). OPs and CXs were introduced relatively recently (over the last 50 y) for the control of malaria vector populations. The number and diversity of the duplicated alleles suggest that these duplications are relatively common events. This conclusion is consistent with recent studies showing that the rate of gene duplication per gene and per generation ranges from a value similar to the substitution rate to a value four orders of magnitude higher [46–49]. It has been suggested that such duplications are promoted by the presence of repeated elements [50,51], which favor unequal crossing-over events by inducing chromatin mismatching. Indeed, several studies have reported the presence of transposable elements at the breakpoints of large adaptive segmental duplications [4,16,52]. We also identified a transposable element (the Harbinger transposase) at the 3ʹ end of the amplicons. The non-allelic homologous recombination model (NAHR [53]) is, thus, probably the most parsimonious mechanism explaining this high frequency of duplication events with conserved breakpoints around the ace-1 amplicons. All Resistance Alleles Share the Same 203 kb Amplicon, Precluding Specific Genotyping It was not possible to design a PCR-based molecular test specific for the ace-1 R copy to differentiate between heterogeneous and homozygous duplications, as the Rx and D alleles shared the same R copy, identical to the only R copy described to date [19,21]. Genomic structure was also conserved among homogeneous and heterogeneous duplications, precluding the development of specific DNA markers to distinguish between amplicons containing Rx and D alleles on the basis of breakpoints or junction sequences. The diagnostic duplication test developed in this study can thus be used only to determine whether a mosquito carries a duplication of some kind. Furthermore, while the amplicons carrying the S or R copy in the D allele had several diagnostic SNPs at their 3' ends, they were identical at their 5ʹ ends, as the amplicon carrying the R copy was located upstream from that carrying the S copy, making it impossible to develop a PCR test specific for the R-S association. Finally, while it is possible to evaluate the number of ace-1 copies with high precision by qPCR, there is currently no method for determining the distribution of copies between homologous chromosomes. Thus, if qPCR and PCR-RFLP ace-1 tests on a particular [RS] individual indicate the presence of four ace-1 copies, for example, it is not possible to determine whether the genotype of this individual is DD, DR2, or R3S. Indirect methods, such as coupling molecular tests with extensive crossing experiments, thus remain the only way to determine the precise genotype, and such methods are unsuitable for use in large field screenings. In our study, we used such indirect methods to genotype the offspring of individuals crossed with the reference SS strain. The quantification of ace-1 copy number by this approach was precise enough to determine the exact genotype when one of the chromosomes was known (here, the chromosome carrying S). This made it possible to establish strains with various numbers of R copies, to investigate the effect of ace-1 CNVs on the phenotype and fitness of mosquitoes. Different Genomic Architectures Result in Different Fitness Trade-Offs We previously investigated the phenotypic and fitness impacts of heterogeneous duplications of ace-1 in An. gambiae [32]. We showed that the D allele confers a phenotype resembling that of standard (RS) heterozygotes, with an intermediate level of resistance and a lower fitness cost. In this study, we showed that acetylcholinesterase activity, regardless of the form of the enzyme (susceptible [AChE1S] or resistant [AChE1R]), is globally proportional to the number of S or R copies, respectively, in the genotype (S2 Fig). Consequently, the D allele has a higher level of activity than a single R copy, because the G119S mutation greatly decreases AChE1 activity. The partial restoration of AChE1 activity induced by the combination of R and S on a single chromosome probably accounts for the lower fitness cost and the intermediate level of resistance of the D allele [32]. The D allele thus provides a new and intermediate alternative to the otherwise irreducible trade-off between protein activity and resistance level (S: no cost, no resistance; R: high cost, high resistance). By contrast to heterogeneous duplications, which combine qualitatively different alleles, homogeneous duplications usually provide a quantitative advantage through repeats of the same allele (see the review by [3]). In the case of ace-1, the total AChE1 activity provided by Rx alleles appears to be positively correlated with the number of R copies (i.e., the number of amplicons). However, no quantitative advantage in terms of resistance was anticipated; resistance was thought to be a binary trait conditioned solely by target protein conformation [21]. The finding that resistance level increased with the number of R copies (R5R5 individuals clearly displayed higher resistance than R3R3 individuals, S4 Fig) was therefore surprising, indicating that resistance is more quantitative than was previously thought (Fig 3). However, the relationship between cost and copy number may not be monotonic. First, life history traits and experimental evolution studies comparing R5 and R3 alleles suggested that a larger number of R copies was associated with a higher cost (Fig 4 and S6 Fig). Conversely, this relationship may not hold if there are fewer than three R copies: (i) three R copies per chromosome were found in the AcerkisR3 strain [RR], despite the absence of insecticide selection over a number of generations (at least 4 y; if carrying fewer copies was less costly, this number would be expected to decrease with time in the absence of selection); (ii) the median number of copies in the Baguida field population (Togo) was three, although only a few [RR] individuals with different numbers of copies were found. It thus seems that three R copies on the same chromosome may be optimal in terms of the associated cost, with carrying fewer or more copies entailing higher fitness costs. So how can we explain this surprising situation? Where do these costs come from? It is possible that total cost is a combination of at least two different, opposite functions correlated with amplification level (Fig 5). First, the cost associated with the low level of AChE1R activity (cost of the G119S mutation) probably decreases with increasing numbers of R copies (e.g., the number of amplicons). The paucity of individuals with only one- or two-copy alleles in the field suggests a high cost; conversely, the total activity provided by the R5 allele is higher than that of the R3 allele, but less than that of the susceptible allele (no overshoot cost). The second cost function would increase with the number of amplicons: R5 appears to be more costly overall than R3. As the two alleles have the same amplicon and structure, this cost is probably metabolic rather than structural in nature. First, as AChE1 is involved in numerous functions at different stages of development (for a review see [54]), the production of larger amounts of AChE1 may have a deleterious effect on these other functions (i.e., a pleiotropic effect). The second possibility, which is probably more likely, is that this cost is not directly related to ace-1 itself, instead being due to changes in the dosage of one or several of the other genes encompassed by the duplication (i.e., gene dosage cost, Fig 5). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. Hypothetical model to explain the non-monotonous relationship between fitness cost and the number of R copies, and the minimum cost centered on three copies per chromosome. Under this hypothesis, the overall fitness cost (red triangles) is the result of two opposite functions: the cost associated with G119S mutation (light gray circles) and the cost due to the change in gene dosage (dark gray squares). The first of these costs is negatively correlated with the number of R copies. Thus, larger numbers of R copies are associated with lower costs. By contrast, the second cost is positively correlated with the number of R copies. Note that the shapes of each function and their interaction are unknown. https://doi.org/10.1371/journal.pbio.2000618.g005 The two functions are probably not symmetric, as the activity of the R5 allele was similar to the expected value under an additive model, whereas the increase in cost relative to R3 was barely visible in terms of life history traits. This hypothetical combination of two cost functions is consistent with the observations and suggests that the R3 allele may be optimal (Fig 5). Internal Deletions May Reduce the Gene-Dosage Disruption Cost Genome analyses revealed that the 203-kb amplicon contained 12 genes, only one of which (ace-1) is involved in resistance to OPs and CXs (S2 Table). Amplicon size may be constrained by the presence of the Harbinger element, favoring the occurrence of large duplications encompassing the other 11 loci. However, these genes are probably only hitchhiking, so their duplication is unlikely to be adaptive. By contrast to ace-1, an increase in their gene-dosage may actually be deleterious: (i) it could alter biochemical equilibria between the duplicated genes and the single-copy genes with which they interact [55,56]; (ii) the function of these duplicated genes may be disrupted because optimal levels of protein may be overshot [57,58]; or (iii) the superfluous production of excess protein from the duplicated genes may be energetically costly [59]. These three hypotheses are not mutually exclusive. We report here observations supporting the notion that duplication costs are due, at least in part, to gene-dosage imbalance between the co-amplified genes rather than directly to ace-1. In R3R3 individuals (AcerKis), DOC decreased within some of the amplicons (Fig 1B), revealing 97 kb intra-amplicon deletions affecting the 11 loci other than ace-1. In these individuals, only ace-1 remains fully amplified, the number of copies of the other genes having decreased towards their initial dosage (Fig 1B). Moreover, this internal deletion was recurrent and also detected in mosquitoes from Burkina Faso and Guinea sequenced by the An. gambiae 1,000 Genomes Consortium (manuscript in preparation). These observations suggest that the intra-amplicon deletions occurred after the duplication events. More importantly, they suggest that these secondary deletions are probably adaptive and were selected to reduce the cost of dosage imbalance for the 11 genes concerned. These deletions would alter the shape of the gene-dosage cost function (Fig 5). This process is probably still ongoing, and it should be possible to test this adaptive hypothesis by comparing fitness between individuals carrying the same number of R copies with and without these deletions. The Genomic Architecture Selected for Resistance Alleles Probably Depends on the Intensity of Selective Pressure, and This Must Be Taken into Account in Vector Control The key finding of this study is that the various architectures of the ace-1 amplicon correspond to different evolutionary trade-offs between resistance and costs, and that they all segregate in An. gambiae field populations. The reason for this coexistence of different alleles with different fitness trade-offs probably lies in the diversity of treatment practices for vector control. This diversity often results in mosaic environments, composed of patches covering the whole gradient from untreated to intensely treated areas, leading to the selection of different alleles. Moderate treatments or fine-grained alternation of treated and untreated areas (over time and/or space) should favor the heterozygous phenotype, with moderate resistance and costs (overdominance and marginal overdominance, respectively). As the D allele does not bear the segregation burden of standard heterozygotes, this heterogeneous duplication should, thus, be favored in this context [17,32,36]. This may be the case in An. gambiae populations from the Ivory Coast, which contained only [RS] individuals, all displaying amplification (Tiassalé and Bouaké, n = 43 and 38, respectively), suggesting that D is present at high frequency, if not already fixed (Table 1). By contrast, higher insecticide doses and coarse-grained environments should select more specialist phenotypes, and, thus, homogeneous duplications with higher levels of resistance (despite their higher costs). Moreover, higher doses should select for larger numbers of copies (R5 confers stronger resistance than R3). In this respect, variations in the level of amplification in homogeneous duplications should allow a more finely tuned response to variations in selection intensity, as the quantitative effects of such duplications on fitness seem to be more subtle than those of heterogeneous duplication. The demonstration of this versatility in the trade-offs available through duplication represents a major development in our understanding of the evolutionary processes of adaptation. However, the consequences, in terms of vector control, are much more negative. Following the massive increase in PYR insecticide resistance in African populations of An. gambiae, several African countries have tried to preserve the efficacy of vector control by switching from PYRs to CXs or OPs, in accordance with the American President’s Malaria Initiative [60], in collaboration with the National Malaria Control Program [61]. However, duplicated ace-1 resistance alleles are already widespread. Some populations contained only resistant individuals, with either 100% [RS] individuals (suggesting the presence of the D allele, Tiassalé and Bouaké, Ivory Coast) or with mostly [RR] individuals (83% carrying RX alleles, Baguida, Togo). Very careful management of the doses of insecticide used, and of their spatial and temporal application, will be required to control resistance, as treatment could rapidly lead to the selection of different types of ace-1 alleles, hampering mosquito control. Such fine-tuning of treatment may prove difficult in a context in which insecticides are used for vector control and various compounds used in agriculture can also select ace-1 resistance alleles (for a review [62]). This issue is particularly pressing because OPs and CXs have been recommended in the context of a high frequency of PYR resistance alleles, but recent studies have reported alarming synergic effects between resistance alleles specific for each insecticide class [63,64]. The high adaptability conferred by ace-1 duplications may have a major impact on An. gambiae vector control in Africa, potentially impeding the control of malaria transmission. Materials and Methods Mosquito Strains and Collections Mosquito strains. Three An. gambiae laboratory strains that were already available were used in this study: KisumuP, Acerkis, and Acerduplikis. KisumuP was derived from the susceptible reference strain Kisumu [65] and was rendered homozygous for a single susceptible ace-1S allele (or S, genotype SS; [32]). Acerkis and Acerduplikis are resistant to both OPs and CXs [32,66]. Acerkis is homozygous for the G119S mutation in the ace-1 gene (ace-1R allele or R, genotype RR). Acerduplikis is homozygous for the ace-1 heterogeneous duplicated allele (ace-1D allele or D, genotype DD), associating one R copy and one S copy on the same chromosome. The three strains mostly share the same KisumuP genetic background (>99.6% similarity [32]). Mosquito collection. Larvae from ten An. gambiae field populations were collected and reared until adulthood in the laboratory: two from Benin, four from Burkina Faso, one from Togo, and three from Ivory Coast (Table 1). Adults were assigned to members of the An. gambiae cryptic-species complex on the basis of morphological tests and molecular analyses [67,68]; their ace-1 phenotype (susceptible [SS], homozygous resistant [RR], or heterozygous [RS]) was assessed through the ace-1 PCR-RLFP test [22]. Genomic DNA Preparation and Sequencing Genomic DNA was extracted from individual mosquitoes of the three strains with the Qiagen DNeasy kit and was treated with RNase A to remove residual RNA. DNA concentration was assessed in the Qbit dsDNA BR Assay (LifeTechnologies). Illumina whole-genome sequencing libraries were constructed with the Nextera DNA sample preparation kit (Illumina), in accordance with the manufacturer’s instructions. KisumuP (SS, two individuals) and Acerduplikis (DD, eight individuals) samples were sequenced by the Wellcome Trust Sanger Institute, as part of the Malaria genome project (https://www.malariagen.net/projects/vector/ag1000g). FASTQ format libraries were generated from 100 bp-read pairs separated by a 250 bp insert. These reads were mapped to the An. gambiae PEST reference genome assembly, downloaded from VectorBase (https://www.vectorbase.org; AgamP4; [41]) with a Wellcome Trust pipeline including Picard tools (https://broadinstitute.github.io/picard/) to format the data, bwa (0.7.5a-r405 [69]) for mapping, and other analysis tools providing optimized mapping data in SAM format. The Acerkis (RR, two individuals) samples were specifically sequenced for this project with the same sequencing technology (Illumina). The sequencing data consisted of 125 bp-read pairs separated by a 500 bp insert. The reads were then trimmed and mapped directly, using bwa, onto the same PEST reference genome. For the genomic analysis, we used the gene annotations from the AgamP4.3 (https://www.vectorbase.org/organisms/anopheles-gambiae/pest/AgamP4.3). Duplication Detection and Characterization We used a two-step approach to detect and characterize the structure of the duplications containing the ace-1 locus. We first analyzed the variations in read depth of coverage (DOC) for short-read mappings on the PEST reference genome. This made it possible to detect CNVs; duplications induce a local increase in DOC, whereas deletions induce a local decrease in DOC. We specifically focused on the genomic region in which the ace-1 locus is located, from the 2 Mb to 5 Mb positions on chromosomal arm 2R (AgamP4.3 release; https://www.vectorbase.org/organisms/anopheles-gambiae/pest/AgamP4.3; Fig 1). Rather than calculating the DOC for each base, we determined the number of reads falling into adjacent 100 bp windows. This made it possible to minimize the computer resources required without losing DOC information. We detected DOC shifts, by calculating the DOC ratio for each resistant strain (i.e., the DOC of the considered strain, RR, or DD, over the DOC of the SS strain, which has no ace-1 gene duplication Fig 1A and 1B). We merged all the sequencing data for individuals of the same strain to intensify the signal and its resolution. This made it possible to confirm and refine the location of the duplication by resolving its breakpoints. In particular, we analyzed the insert size distribution of the paired-end reads (S1A Fig) close to the previously defined putative breakpoints (± 1 kb; S1B Fig). Breakpoint sequences can be determined precisely by analyzing soft-clipped reads (i.e., partially mapped reads indicative of tandem duplication events as they overlap two copies) and discordant read pairs (i.e., pairs with a mapping span and/or orientation inconsistent with the expected insert size). The soft-clipped reads and discordant pairs were selected by parsing the mapping data in SAM format https://samtools.gihub.io/hts-spes/SAMv1.pdf. We then aligned them with MUSCLE, a multiple alignment tool suitable for short sequences available through SeaView alignment editor/viewer version 4 [70,71]. The consensus sequences of both the breakpoints (i.e., 5ʹ and 3ʹ flanking sequences of the duplication) and the junction (i.e., sequences overlapping the two amplicons; S1C Fig) were then constructed. We determined the 5'–3' orientations and relative positions of the different amplicons by amplifying a 2,465 bp PCR fragment (AgRDdir1 and AgRDrev1 primers) overlapping the junction from the genomic DNA of individuals (S1 Table). PCR products were purified with the QIAquick Gel Extraction Kit (Qiagen) and directly sequenced with an ABI Prism 310 sequencer (BigDye Terminator Kit, Applied Biosystems, Foster City, CA). Quick Diagnostic PCR Test for Duplications A PCR primer pair was designed (Agduplispedir2 and AgduplispeRev1), with each primer binding to a different amplicon, on either side of the junction (S1D Fig and S1 Table). The resulting 460 bp fragment overlaps the junction, and is therefore amplified only in individuals carrying multiple copies of the ace-1 locus. This quick, simple diagnostic PCR test thus reveals the presence or absence of ace-1 duplications. Gene Copy-Number Quantification We estimated the number of copies present for several target loci relative to a reference locus AGAP010592) present as a single copy in the VectorBase PEST genome (Rps7, AgS7Ex5qtidir and AgS7Ex5qtirev primers; https://www.vectorbase.org/, S1 Table), by real-time quantitative PCR (qPCR, LC480 Light Cycler, Roche). Three genes were targeted: ace-1 (AGAP001356, AgAce1qtidir2, and AgAce1qtirev2 primers) and two loci flanking the amplified region: AGAP001355 on the 5ʹ-flanking side ("5'out", Ag5ʹoutdir and Ag5ʹoutrev primers), and AGAP001369 on the 3ʹ-flanking side ("3'out", Ag3ʹoutdir and Ag3ʹoutrev primers; Fig 1 and S1 Table). These flanking loci were used as qPCR markers of the amplified zone. As they are located outside the amplicon, they should not be amplified. We dispensed 0.5 μl of genomic DNA and 1.5 μl of reaction mixture containing specific primers, each at a concentration of 0.8 μM and 0.75 μl of Master Mix (LightCycler 480 SYBR Green I Master, Roche) into the wells of a 384-well plate, with a Labcyte Echo525 dispenser. We performed qPCR as follows: activation at 95°C for 8 min, followed by 45 cycles of 95°C for 4 s, 67°C for 13 s, and 72°C for 19 s. Melting curves were generated by a post-amplification melting step between 70°C and 95°C, for Tm analysis. All quantifications were replicated four times for each DNA template. Standard curves were constructed with 10-fold dilutions of a PCR product previously amplified with specific primers for each locus from KisumuP (SS) strain DNA. The ace-1, 5ʹout, and 3'out concentration ratios over RpS7 were determined by the advanced relative quantification method (LightCycler 480 software v.1.5.0). Resistance Measure Bioassays were used to assess mosquito resistance to three insecticides: one CX (bendiocarb, technical grade, 99.5% purity), one OP (chlorpyrifos methyl, 99.9% purity), and one PYR (permethrin, 98.3% purity). We incubated 25 late third-instar larvae for 24 h at 27°C ± 2°C in plastic cups containing 99 ml of distilled water, to which we added 1 ml of insecticide solution at the required concentration (1 ml of ethanol in controls). Four replicates were performed for each concentration. Larval mortality was recorded after 24 h of exposure. Dose-mortality responses were analyzed with the BioRssay R script (v.6.2 [72]) freely available from the ISEM website (http://www.isem.univ-montp2.fr/recherche/teams/genomic-adaptation/staff/labbepierrick/?lang=en). This script computes the 50% or 95% lethal concentrations (LC50 and LC95, i.e., insecticide doses killing 50% and 95% of the tested population or strain) and their confidence intervals, and assesses the linearity of the dose-mortality response (χ2 test). Finally, it compares the dose-mortality responses of two or more strains (or populations) and calculates their resistance ratios (RR50 or RR95, = LC50 or LC95 of the tested strain/population over LC50 or LC95 of the reference strain, respectively) and 95% confidence intervals. AChE1 Activity Measure Adult mosquitoes were decapitated, and each head was individually homogenized in 400 μl phosphate buffer (0.25 M, pH7) supplemented with 1% Triton X-100. Homogenates were centrifuged (9.3 g for 3 min) and 100 μl of the supernatant was dispensed into each of two wells of a 96-well microtitration plate. We added 10 μl ethanol (95%) to the first well and 10 μl propoxur (a CX insecticide, at 10-1M, diluted in ethanol) to the second. The plate was incubated for 15 min at room temperature. We then added 100 μl of substrate solution (25 mM sodium phosphate, pH 7.0, 0.2 mM DTNB, 0.35 mM sodium bicarbonate, 2.5 mM acetylthiocholine) to each well. AChE1 activity was estimated by measuring the change in optical density following the cleavage of acetylthiocholine, as described by Ellman et al. [73]. Optical density at 412 nm was recorded every minute for 15 min with an EL 800 microplate reader (Bio-Tek Instruments, Inc.). The mean slope of each reaction was calculated with KCjunior v1.41.4 analysis software (Bio-Tek Instruments, Inc.) and was used as a measurement of AChE1 activity. The first well (ethanol) was used to assess total AChE1 activity (ATOT), whereas the second (propoxur) provided information about AChE1R activity (AR) only. AChE1S activity (AS) was deduced as = ATOT—AR. Note that activity in the second well was never equal to 0: a very shallow slope was observed even for susceptible individuals, due to the spontaneous degradation of DTNB. For comparisons of different strains or populations, samples of each were distributed on the same plate and analyzed simultaneously to avoid experimental artifacts. Proxies for Fitness Cost Larval mortality and development time. We assessed the development time and pre-imaginal mortality associated with different ace-1 copy numbers by performing larval mortality assays on different strains, as described by Assogba et al. [32]. Briefly, first-instar larvae were individually reared in Drosophila tubes, in 1 ml of mineral water with TetraMin powdered fish food (2 g/l). Dead larvae were counted daily to assess the mortality rate at each development stage. The timing of adult emergence was also recorded. Female fecundity and fertility. All strains were reared under the same soft environmental conditions (relatively low densities, no food limitation). In each strain, 200 males were crossed with 200 females. Females were blood-fed after 3 d, and 40 gravid females from each strain were allowed to oviposit individually in plastic cups containing 70 ml dechlorinated water. Three days after blood feeding, the number of egg-laying females and the number of eggs per female were recorded. Two days later, we determined the number of hatching larvae per female. Statistical Analyses ace-1 copy-number variation. The numbers of ace-1 copies for individuals from different strains were estimated by qPCR. The significance of the differences observed was assessed with the following generalized linear model (GLM): Cn = Geno + ε, where Cn is the number of copies, Geno is a multi-level factor corresponding to genotype and ε is the error parameter, which follows a Gaussian distribution. We checked the normality of the model residuals in a Shapiro-Wilk test [74]. Fitness costs. Larval mortality was analyzed by calculating the following Cox proportional hazards regression model (Cox model): Surv = Geno + ε, where Surv is the proportion of dead larvae at each developmental stage, Geno is a three-level factor corresponding to the different genotypes tested, and ε is the error parameter, which follows a binomial distribution to take any overdispersion into account. Emerging adults were censored in the analyses. Differences in development time between genotypes and/or sexes were assessed with the following Cox model: Dev = Geno + Sex + Geno.Sex+ ε, where Dev is the number of adults emerging on a given day, Geno is a three-level factor corresponding to the different genotypes tested, Sex is a two-level factor (male or female), Geno.Sex is the interaction between these two factors, and ε is the error parameter (binomial distribution). The other cost proxies (Cp) were analyzed with GLMs in the form Cp = Geno + ε, where Geno is a three-level factor corresponding to the different genotypes tested and ε is the error parameter, which follows a binomial distribution for the proportion of females laying eggs and the hatching rate, and a Gaussian distribution for the numbers of eggs and larvae per female. All calculations were performed with free R software (v.3.1.1, http://www.r-project.org). LM, GLM, and Cox models were simplified as follows: the significance of the different terms was assessed, beginning with the higher-order terms, in likelihood ratio tests (LRTs), and non-significant terms (p > 0.05) were removed. The factor levels of qualitative variables that were not significantly different (in LRT) were grouped [75]. Data Availability Data deposited in the Dryad repository: http://dx.doi.org/10.5061/dryad.4f7qg [76]. Sequence data deposited in the NCBI repository: http://www.ncbi.nlm.nih.gov/bioproject/348825. Mosquito Strains and Collections Mosquito strains. Three An. gambiae laboratory strains that were already available were used in this study: KisumuP, Acerkis, and Acerduplikis. KisumuP was derived from the susceptible reference strain Kisumu [65] and was rendered homozygous for a single susceptible ace-1S allele (or S, genotype SS; [32]). Acerkis and Acerduplikis are resistant to both OPs and CXs [32,66]. Acerkis is homozygous for the G119S mutation in the ace-1 gene (ace-1R allele or R, genotype RR). Acerduplikis is homozygous for the ace-1 heterogeneous duplicated allele (ace-1D allele or D, genotype DD), associating one R copy and one S copy on the same chromosome. The three strains mostly share the same KisumuP genetic background (>99.6% similarity [32]). Mosquito collection. Larvae from ten An. gambiae field populations were collected and reared until adulthood in the laboratory: two from Benin, four from Burkina Faso, one from Togo, and three from Ivory Coast (Table 1). Adults were assigned to members of the An. gambiae cryptic-species complex on the basis of morphological tests and molecular analyses [67,68]; their ace-1 phenotype (susceptible [SS], homozygous resistant [RR], or heterozygous [RS]) was assessed through the ace-1 PCR-RLFP test [22]. Mosquito strains. Three An. gambiae laboratory strains that were already available were used in this study: KisumuP, Acerkis, and Acerduplikis. KisumuP was derived from the susceptible reference strain Kisumu [65] and was rendered homozygous for a single susceptible ace-1S allele (or S, genotype SS; [32]). Acerkis and Acerduplikis are resistant to both OPs and CXs [32,66]. Acerkis is homozygous for the G119S mutation in the ace-1 gene (ace-1R allele or R, genotype RR). Acerduplikis is homozygous for the ace-1 heterogeneous duplicated allele (ace-1D allele or D, genotype DD), associating one R copy and one S copy on the same chromosome. The three strains mostly share the same KisumuP genetic background (>99.6% similarity [32]). Mosquito collection. Larvae from ten An. gambiae field populations were collected and reared until adulthood in the laboratory: two from Benin, four from Burkina Faso, one from Togo, and three from Ivory Coast (Table 1). Adults were assigned to members of the An. gambiae cryptic-species complex on the basis of morphological tests and molecular analyses [67,68]; their ace-1 phenotype (susceptible [SS], homozygous resistant [RR], or heterozygous [RS]) was assessed through the ace-1 PCR-RLFP test [22]. Genomic DNA Preparation and Sequencing Genomic DNA was extracted from individual mosquitoes of the three strains with the Qiagen DNeasy kit and was treated with RNase A to remove residual RNA. DNA concentration was assessed in the Qbit dsDNA BR Assay (LifeTechnologies). Illumina whole-genome sequencing libraries were constructed with the Nextera DNA sample preparation kit (Illumina), in accordance with the manufacturer’s instructions. KisumuP (SS, two individuals) and Acerduplikis (DD, eight individuals) samples were sequenced by the Wellcome Trust Sanger Institute, as part of the Malaria genome project (https://www.malariagen.net/projects/vector/ag1000g). FASTQ format libraries were generated from 100 bp-read pairs separated by a 250 bp insert. These reads were mapped to the An. gambiae PEST reference genome assembly, downloaded from VectorBase (https://www.vectorbase.org; AgamP4; [41]) with a Wellcome Trust pipeline including Picard tools (https://broadinstitute.github.io/picard/) to format the data, bwa (0.7.5a-r405 [69]) for mapping, and other analysis tools providing optimized mapping data in SAM format. The Acerkis (RR, two individuals) samples were specifically sequenced for this project with the same sequencing technology (Illumina). The sequencing data consisted of 125 bp-read pairs separated by a 500 bp insert. The reads were then trimmed and mapped directly, using bwa, onto the same PEST reference genome. For the genomic analysis, we used the gene annotations from the AgamP4.3 (https://www.vectorbase.org/organisms/anopheles-gambiae/pest/AgamP4.3). Duplication Detection and Characterization We used a two-step approach to detect and characterize the structure of the duplications containing the ace-1 locus. We first analyzed the variations in read depth of coverage (DOC) for short-read mappings on the PEST reference genome. This made it possible to detect CNVs; duplications induce a local increase in DOC, whereas deletions induce a local decrease in DOC. We specifically focused on the genomic region in which the ace-1 locus is located, from the 2 Mb to 5 Mb positions on chromosomal arm 2R (AgamP4.3 release; https://www.vectorbase.org/organisms/anopheles-gambiae/pest/AgamP4.3; Fig 1). Rather than calculating the DOC for each base, we determined the number of reads falling into adjacent 100 bp windows. This made it possible to minimize the computer resources required without losing DOC information. We detected DOC shifts, by calculating the DOC ratio for each resistant strain (i.e., the DOC of the considered strain, RR, or DD, over the DOC of the SS strain, which has no ace-1 gene duplication Fig 1A and 1B). We merged all the sequencing data for individuals of the same strain to intensify the signal and its resolution. This made it possible to confirm and refine the location of the duplication by resolving its breakpoints. In particular, we analyzed the insert size distribution of the paired-end reads (S1A Fig) close to the previously defined putative breakpoints (± 1 kb; S1B Fig). Breakpoint sequences can be determined precisely by analyzing soft-clipped reads (i.e., partially mapped reads indicative of tandem duplication events as they overlap two copies) and discordant read pairs (i.e., pairs with a mapping span and/or orientation inconsistent with the expected insert size). The soft-clipped reads and discordant pairs were selected by parsing the mapping data in SAM format https://samtools.gihub.io/hts-spes/SAMv1.pdf. We then aligned them with MUSCLE, a multiple alignment tool suitable for short sequences available through SeaView alignment editor/viewer version 4 [70,71]. The consensus sequences of both the breakpoints (i.e., 5ʹ and 3ʹ flanking sequences of the duplication) and the junction (i.e., sequences overlapping the two amplicons; S1C Fig) were then constructed. We determined the 5'–3' orientations and relative positions of the different amplicons by amplifying a 2,465 bp PCR fragment (AgRDdir1 and AgRDrev1 primers) overlapping the junction from the genomic DNA of individuals (S1 Table). PCR products were purified with the QIAquick Gel Extraction Kit (Qiagen) and directly sequenced with an ABI Prism 310 sequencer (BigDye Terminator Kit, Applied Biosystems, Foster City, CA). Quick Diagnostic PCR Test for Duplications A PCR primer pair was designed (Agduplispedir2 and AgduplispeRev1), with each primer binding to a different amplicon, on either side of the junction (S1D Fig and S1 Table). The resulting 460 bp fragment overlaps the junction, and is therefore amplified only in individuals carrying multiple copies of the ace-1 locus. This quick, simple diagnostic PCR test thus reveals the presence or absence of ace-1 duplications. Gene Copy-Number Quantification We estimated the number of copies present for several target loci relative to a reference locus AGAP010592) present as a single copy in the VectorBase PEST genome (Rps7, AgS7Ex5qtidir and AgS7Ex5qtirev primers; https://www.vectorbase.org/, S1 Table), by real-time quantitative PCR (qPCR, LC480 Light Cycler, Roche). Three genes were targeted: ace-1 (AGAP001356, AgAce1qtidir2, and AgAce1qtirev2 primers) and two loci flanking the amplified region: AGAP001355 on the 5ʹ-flanking side ("5'out", Ag5ʹoutdir and Ag5ʹoutrev primers), and AGAP001369 on the 3ʹ-flanking side ("3'out", Ag3ʹoutdir and Ag3ʹoutrev primers; Fig 1 and S1 Table). These flanking loci were used as qPCR markers of the amplified zone. As they are located outside the amplicon, they should not be amplified. We dispensed 0.5 μl of genomic DNA and 1.5 μl of reaction mixture containing specific primers, each at a concentration of 0.8 μM and 0.75 μl of Master Mix (LightCycler 480 SYBR Green I Master, Roche) into the wells of a 384-well plate, with a Labcyte Echo525 dispenser. We performed qPCR as follows: activation at 95°C for 8 min, followed by 45 cycles of 95°C for 4 s, 67°C for 13 s, and 72°C for 19 s. Melting curves were generated by a post-amplification melting step between 70°C and 95°C, for Tm analysis. All quantifications were replicated four times for each DNA template. Standard curves were constructed with 10-fold dilutions of a PCR product previously amplified with specific primers for each locus from KisumuP (SS) strain DNA. The ace-1, 5ʹout, and 3'out concentration ratios over RpS7 were determined by the advanced relative quantification method (LightCycler 480 software v.1.5.0). Resistance Measure Bioassays were used to assess mosquito resistance to three insecticides: one CX (bendiocarb, technical grade, 99.5% purity), one OP (chlorpyrifos methyl, 99.9% purity), and one PYR (permethrin, 98.3% purity). We incubated 25 late third-instar larvae for 24 h at 27°C ± 2°C in plastic cups containing 99 ml of distilled water, to which we added 1 ml of insecticide solution at the required concentration (1 ml of ethanol in controls). Four replicates were performed for each concentration. Larval mortality was recorded after 24 h of exposure. Dose-mortality responses were analyzed with the BioRssay R script (v.6.2 [72]) freely available from the ISEM website (http://www.isem.univ-montp2.fr/recherche/teams/genomic-adaptation/staff/labbepierrick/?lang=en). This script computes the 50% or 95% lethal concentrations (LC50 and LC95, i.e., insecticide doses killing 50% and 95% of the tested population or strain) and their confidence intervals, and assesses the linearity of the dose-mortality response (χ2 test). Finally, it compares the dose-mortality responses of two or more strains (or populations) and calculates their resistance ratios (RR50 or RR95, = LC50 or LC95 of the tested strain/population over LC50 or LC95 of the reference strain, respectively) and 95% confidence intervals. AChE1 Activity Measure Adult mosquitoes were decapitated, and each head was individually homogenized in 400 μl phosphate buffer (0.25 M, pH7) supplemented with 1% Triton X-100. Homogenates were centrifuged (9.3 g for 3 min) and 100 μl of the supernatant was dispensed into each of two wells of a 96-well microtitration plate. We added 10 μl ethanol (95%) to the first well and 10 μl propoxur (a CX insecticide, at 10-1M, diluted in ethanol) to the second. The plate was incubated for 15 min at room temperature. We then added 100 μl of substrate solution (25 mM sodium phosphate, pH 7.0, 0.2 mM DTNB, 0.35 mM sodium bicarbonate, 2.5 mM acetylthiocholine) to each well. AChE1 activity was estimated by measuring the change in optical density following the cleavage of acetylthiocholine, as described by Ellman et al. [73]. Optical density at 412 nm was recorded every minute for 15 min with an EL 800 microplate reader (Bio-Tek Instruments, Inc.). The mean slope of each reaction was calculated with KCjunior v1.41.4 analysis software (Bio-Tek Instruments, Inc.) and was used as a measurement of AChE1 activity. The first well (ethanol) was used to assess total AChE1 activity (ATOT), whereas the second (propoxur) provided information about AChE1R activity (AR) only. AChE1S activity (AS) was deduced as = ATOT—AR. Note that activity in the second well was never equal to 0: a very shallow slope was observed even for susceptible individuals, due to the spontaneous degradation of DTNB. For comparisons of different strains or populations, samples of each were distributed on the same plate and analyzed simultaneously to avoid experimental artifacts. Proxies for Fitness Cost Larval mortality and development time. We assessed the development time and pre-imaginal mortality associated with different ace-1 copy numbers by performing larval mortality assays on different strains, as described by Assogba et al. [32]. Briefly, first-instar larvae were individually reared in Drosophila tubes, in 1 ml of mineral water with TetraMin powdered fish food (2 g/l). Dead larvae were counted daily to assess the mortality rate at each development stage. The timing of adult emergence was also recorded. Female fecundity and fertility. All strains were reared under the same soft environmental conditions (relatively low densities, no food limitation). In each strain, 200 males were crossed with 200 females. Females were blood-fed after 3 d, and 40 gravid females from each strain were allowed to oviposit individually in plastic cups containing 70 ml dechlorinated water. Three days after blood feeding, the number of egg-laying females and the number of eggs per female were recorded. Two days later, we determined the number of hatching larvae per female. Larval mortality and development time. We assessed the development time and pre-imaginal mortality associated with different ace-1 copy numbers by performing larval mortality assays on different strains, as described by Assogba et al. [32]. Briefly, first-instar larvae were individually reared in Drosophila tubes, in 1 ml of mineral water with TetraMin powdered fish food (2 g/l). Dead larvae were counted daily to assess the mortality rate at each development stage. The timing of adult emergence was also recorded. Female fecundity and fertility. All strains were reared under the same soft environmental conditions (relatively low densities, no food limitation). In each strain, 200 males were crossed with 200 females. Females were blood-fed after 3 d, and 40 gravid females from each strain were allowed to oviposit individually in plastic cups containing 70 ml dechlorinated water. Three days after blood feeding, the number of egg-laying females and the number of eggs per female were recorded. Two days later, we determined the number of hatching larvae per female. Statistical Analyses ace-1 copy-number variation. The numbers of ace-1 copies for individuals from different strains were estimated by qPCR. The significance of the differences observed was assessed with the following generalized linear model (GLM): Cn = Geno + ε, where Cn is the number of copies, Geno is a multi-level factor corresponding to genotype and ε is the error parameter, which follows a Gaussian distribution. We checked the normality of the model residuals in a Shapiro-Wilk test [74]. Fitness costs. Larval mortality was analyzed by calculating the following Cox proportional hazards regression model (Cox model): Surv = Geno + ε, where Surv is the proportion of dead larvae at each developmental stage, Geno is a three-level factor corresponding to the different genotypes tested, and ε is the error parameter, which follows a binomial distribution to take any overdispersion into account. Emerging adults were censored in the analyses. Differences in development time between genotypes and/or sexes were assessed with the following Cox model: Dev = Geno + Sex + Geno.Sex+ ε, where Dev is the number of adults emerging on a given day, Geno is a three-level factor corresponding to the different genotypes tested, Sex is a two-level factor (male or female), Geno.Sex is the interaction between these two factors, and ε is the error parameter (binomial distribution). The other cost proxies (Cp) were analyzed with GLMs in the form Cp = Geno + ε, where Geno is a three-level factor corresponding to the different genotypes tested and ε is the error parameter, which follows a binomial distribution for the proportion of females laying eggs and the hatching rate, and a Gaussian distribution for the numbers of eggs and larvae per female. All calculations were performed with free R software (v.3.1.1, http://www.r-project.org). LM, GLM, and Cox models were simplified as follows: the significance of the different terms was assessed, beginning with the higher-order terms, in likelihood ratio tests (LRTs), and non-significant terms (p > 0.05) were removed. The factor levels of qualitative variables that were not significantly different (in LRT) were grouped [75]. ace-1 copy-number variation. The numbers of ace-1 copies for individuals from different strains were estimated by qPCR. The significance of the differences observed was assessed with the following generalized linear model (GLM): Cn = Geno + ε, where Cn is the number of copies, Geno is a multi-level factor corresponding to genotype and ε is the error parameter, which follows a Gaussian distribution. We checked the normality of the model residuals in a Shapiro-Wilk test [74]. Fitness costs. Larval mortality was analyzed by calculating the following Cox proportional hazards regression model (Cox model): Surv = Geno + ε, where Surv is the proportion of dead larvae at each developmental stage, Geno is a three-level factor corresponding to the different genotypes tested, and ε is the error parameter, which follows a binomial distribution to take any overdispersion into account. Emerging adults were censored in the analyses. Differences in development time between genotypes and/or sexes were assessed with the following Cox model: Dev = Geno + Sex + Geno.Sex+ ε, where Dev is the number of adults emerging on a given day, Geno is a three-level factor corresponding to the different genotypes tested, Sex is a two-level factor (male or female), Geno.Sex is the interaction between these two factors, and ε is the error parameter (binomial distribution). The other cost proxies (Cp) were analyzed with GLMs in the form Cp = Geno + ε, where Geno is a three-level factor corresponding to the different genotypes tested and ε is the error parameter, which follows a binomial distribution for the proportion of females laying eggs and the hatching rate, and a Gaussian distribution for the numbers of eggs and larvae per female. All calculations were performed with free R software (v.3.1.1, http://www.r-project.org). LM, GLM, and Cox models were simplified as follows: the significance of the different terms was assessed, beginning with the higher-order terms, in likelihood ratio tests (LRTs), and non-significant terms (p > 0.05) were removed. The factor levels of qualitative variables that were not significantly different (in LRT) were grouped [75]. Data Availability Data deposited in the Dryad repository: http://dx.doi.org/10.5061/dryad.4f7qg [76]. Sequence data deposited in the NCBI repository: http://www.ncbi.nlm.nih.gov/bioproject/348825. Supporting Information S1 Fig. Resolution of the ace-1 duplication structure. (A) Distribution of the paired-end (PE) insert size in the vicinity of the breakpoints (± 1 kb). For each strain, we recorded the insert size of each read and its paired read; for each 200 bp insert size class, we calculated the number of reads, which was then normalized relative to the 2R chromosome mean DOC (between 2 Mb and 5 Mb, excluding the duplicated region). Discordant PEs presented insert sizes distributed around 202 kb, and were identified only for the Acerduplikis (DD, mean insert size 250 bp) and AcerkisR3 (R3R3, mean insert size 500 bp) strains. (B) Duplication breakpoints and junction resolution. The top figure shows the relative positions of the amplicons (i.e. carrying the resistant R or susceptible S copy) of the DD strain duplication; the 5’ end of the amplified region is crosshatched in blue and the 3’ end is crosshatched in red. The bottom figure shows the expected mapping of the reads from DD strain onto the reference genomes: PE reads surrounding the duplication junction result in discordant pairs (i.e. pairs with reads mapping in opposite orientations, with an insert size different from the expected 250 bp); reads overlapping the duplication junction result in soft-clipped reads (i.e. partially mapped reads). These features were used to estimate the duplication length and to reconstitute the junction and breakpoint sequences. (C) Alignment of breakpoints and junction sequences. The 5’ and 3’ breakpoint sequences are aligned with the junction sequence. (D) Junction sequences for the Acerduplikis (DD) and AcerkisR3 (R3R3) strains. The two sequences are strictly identical; the junction position is indicated in the red box. The Agduplispedir2 and Agduplisperev1 primers used for sequencing and for the diagnostic test for duplications are highlighted in gray (see S2 Table). Underlying data can be found in DRYAD http://dx.doi.org/10.5061/dryad.4f7qg. https://doi.org/10.1371/journal.pbio.2000618.s001 (PDF) S2 Fig. Relative AChE1 activities of the various genotypes. Relative AChE1R activities (scaled by the mean AChE1R activity of the R3R3genotype, top panels) and relative AChE1S activities (scaled by the mean AChE1S activity of the SS genotype, bottom panels) are shown for various genotypes, as a function of their number of R or S ace-1 copies. The linear regression is plotted as a dotted line. Underlying data can be found in DRYAD http://dx.doi.org/10.5061/dryad.4f7qg. https://doi.org/10.1371/journal.pbio.2000618.s002 (PDF) S3 Fig. Relative AChE1R activity in R3R3 and R5R5 individuals. Boxplots representing the relative AChE1R activity distribution measured on 20 males from the AcerkisR3(R3R3, red) and AgRR5 (R5R5, blue) strains. Differences in activity were assessed with the following GLM: Activity = Geno + ε, where Geno is a two-level factor corresponding to the genotype and ε is the error parameter, which follows a Gaussian distribution (***, p < 0.001). Underlying data can be found in DRYAD http://dx.doi.org/10.5061/dryad.4f7qg. https://doi.org/10.1371/journal.pbio.2000618.s003 (PDF) S4 Fig. Resistance to bendiocarb (CX) and chlorpyrifos-methyl (OP) insecticides. Mortality (probit scale) is presented as a function of insecticide dose (log10) for the three strains: KisumuP (SS; green squares), AcerkisR3(R3R3, red triangles) and AgRR5 (R5R5, blue dots). Linear regressions between the two factors (solid lines) are indicated, together with the associated 95% confidence intervals (dotted lines). Underlying data can be found in DRYAD http://dx.doi.org/10.5061/dryad.4f7qg. https://doi.org/10.1371/journal.pbio.2000618.s004 (PDF) S5 Fig. Female fertility and fecundity in susceptible (SS) and resistant (R3R3and R5R5) homozygotes. For each genotype, SS (green), R3R3 (red) and R5R5 (blue), we present the following: (A) the mean oviposition rate (i.e. the number of females laying eggs over the number of females studied) and its standard error (SEM), (B) the mean number of eggs laid per female and its SEM, and (C) the mean hatching rate (i.e. the number of larvae produced over the number of eggs) and its SEM. The significance of the differences between the various genotypes is indicated (n.s., p < 0.05; ***, p < 0.001). Underlying data can be found in DRYAD http://dx.doi.org/10.5061/dryad.4f7qg. https://doi.org/10.1371/journal.pbio.2000618.s005 (PDF) S6 Fig. Dynamics of AChE1 activity index (AI) over generations in the experimental evolution assay. For each replicate (C1, C2 and C3), boxplots represent the distribution of activity index (AI) for each generation. Blue and red lines correspond to the expected AI of R5R5 and R3R3 homozygotes, respectively. For each replicate, the green line corresponds to the following GLM: AI = Gen + ε, where Gen is a five-level factor corresponding to generation and ε is the error parameter, which follows a Gaussian distribution. Underlying data can be found in DRYAD http://dx.doi.org/10.5061/dryad.4f7qg. https://doi.org/10.1371/journal.pbio.2000618.s006 (PDF) S1 Table. List of the 12 genes present within the duplicated region and their function (from VectorBase AgamP4 Anopheles gambiae genome). https://doi.org/10.1371/journal.pbio.2000618.s007 (PDF) S2 Table. List of the primers used in this study. https://doi.org/10.1371/journal.pbio.2000618.s008 (PDF) S3 Table. Nature and number of ace-1 copies in different mosquito genotypes. https://doi.org/10.1371/journal.pbio.2000618.s009 (PDF) Acknowledgments We would like to thank Nicole Pasteur for her helpful comments on the manuscript. Benoît S. Assogba was supported by a fellowship from the Institut pour la Recherche et le Développement (IRD). The Sanger sequences used in this work were generated by the technical facilities of the Environmental Genomic Platform of the LabEx Centre “Méditerranéen Environnement Biodiversité” and GenSeq ISEM Platform. The genome sequences of the AcerKis strain were generated by MGX-Montpellier GenomiX. The genome sequences of the Acerduplikis and KisumuP strains were obtained thanks to Pr. Martin Donnelly through the Wellcome Trust Sanger Institute. Contribution number 2016–246 of the Institut des Sciences de l’Evolution de Montpellier, UMR 5554 CNRS-UM-IRD-EPHE.

doi: 10.1371/journal.pbio.2001221pmid: 28002462

Box 1. Examples of Interests in Biomedical Research Personal, religious, or political beliefs Personal experiences Advocacy or policy positions of the researcher or organization with which they are affiliated Intellectual, theoretical, or school of thought commitments Type of training; professional or academic education Profession or discipline Academic competition or rivalry Career advancement or promotion Glory seeking or desire for fame Dominant researcher in area of research Personal experience with subject of the research Personal relationship with someone who has the disease or condition under study Role as investigator on study included in a systematic review Published opinion essay or commentary on topic of research Institutional affiliation or academic associations Refocusing Definitions of Conflict of Interest The Institute of Medicine (IOM) defines a conflict of interest as “a set of circumstances that creates a risk that professional judgment or actions regarding a primary interest will be unduly influenced by a secondary interest” [15]. Conflicts of interest are a problem for those who must make expert judgments on behalf of others; thus, their primary interest is the well-being of those who rely upon these judgments. For researchers, this is the scientific community and public, who make decisions on the basis of the outcomes of research. Conflicts of interest are distinct from ethical dilemmas in that one interest has a claim to priority—the primary or professional interest—and efforts are directed at ensuring that secondary interests do not dominate, or appear to dominate, the primary interest [14]. Being a researcher means having particular education and training, typically a personal interest in a topic or field, usually employment with some kind of academic or scientific institution, and a whole host of experiences that make up a research career. These interests are part of the primary roles and responsibilities that come with being a scientist [13]. However, interests are distinct from conflicts of interest (Box 2). Box 2. Is an Interest a Conflict of Interest? The individual or institution is in a position where others rely on their decision making. One of the conflicting interests has an ethical claim to priority. It is theoretically possible to eliminate the conflict of interest. (If the only solution is recusal, this is not a conflict of interest.) The direction of bias produced by the conflict of interest is consistent within a set of circumstances. The scope of influence may extend beyond an individual and immediate set of circumstances, as in the case of sponsorship. We propose three rules of thumb to distinguish conflicts of interest from “interests” more broadly. First, it is theoretically possible, though not always necessary, to eliminate a conflict of interest. For example, an investigator can divest themselves from shares in the company that commercializes their research product, whereas they cannot possibly separate themselves from their disciplinary training. Similarly, if the only solution for a particular type of interest is recusal because the interest cannot be eliminated, this is not a conflict of interest but rather part of the researcher’s professional role or personal identity. Second, the direction of the bias produced by a conflict of interest is consistent within a set of circumstances. Evidence examining the influence of industry sponsorship on research suggests that the existence of these conflicts may systematically distort the outcomes, effect sizes, or conclusions of research in a direction which consistently favors the sponsor [7–11]. Third, conflicts of interest can be widespread, and their scope of influence may extend far beyond an individual. People participate in research as individuals, and each bring their experiences and personal and professional interests to the process. However, sponsorship serves to amplify a particular viewpoint, ensuring its widespread dissemination and representation in decision making. For example, corporate sponsorship of research is capable of distorting an entire body of published research, as was the case with the tobacco industry and research on secondhand smoke [16]. Conflicts of interest are not exclusively financial: for example, conflicts of interest can arise from personal relationships. First, a person theoretically can eliminate the conflict by eliminating the relationship (which in most cases is an extreme measure) or, more commonly, by reorganizing roles and responsibilities or seeking oversight. For example, spouses working in the same lab commonly require special permission from the institution and are not allowed to directly supervise each other. Recusal is a common strategy but, importantly, not the only option. For example, individuals disclose the relationship and typically recuse themselves from reviewing grant applications written by their students, postdocs, collaborators, rivals, or family members. Second, the direction of the bias can almost always be predicted with conflicts of interest arising from personal relationships—individuals will typically help their friends and harm their enemies. However, while personal relationships can lead to conflicts of interest, unlike conflicts of interest arising from financial ties, their effects rarely extend beyond the immediate situation. That said, conflicts of interest stemming from relationships may be more damaging for individuals in positions of power (such as supervisory or leadership roles), as their scope of influence is greater. Box 2. Is an Interest a Conflict of Interest? The individual or institution is in a position where others rely on their decision making. One of the conflicting interests has an ethical claim to priority. It is theoretically possible to eliminate the conflict of interest. (If the only solution is recusal, this is not a conflict of interest.) The direction of bias produced by the conflict of interest is consistent within a set of circumstances. The scope of influence may extend beyond an individual and immediate set of circumstances, as in the case of sponsorship. Why Financial Conflicts of Interest are so Problematic Commercial sponsorship of research and investigator financial conflicts of interest are two forms of conflict of interest with the strongest evidence base, suggesting that they are a widespread and harmful source of bias. Pharmaceutical, tobacco, food, or chemical industry funding biases human research studies towards outcomes that are favorable to the sponsor’s product, even when controlling for other biases in the methods [7–8,10–11,17]. Thus, even when the methods meet high standards for internal validity, financial conflicts of interest may influence research results through other mechanisms, such as the framing of the question, how the study is actually conducted, and whether it is fully and accurately reported. While everyone comes to the table with an identity, past experiences, and professional interests, industry sponsorship or investigator payments serve as a megaphone, amplifying and multiplying a set of interests that align with the sponsor’s and thereby create a widespread platform of influence for the sponsor. The tobacco industry for decades used sponsorship of research and payments to scientists to fund research that supported their interests, to suppress research that did not, and to disseminate interest group data to the lay press and policymakers [16]. The influence of industry sponsorship on research has created a crisis within biomedicine—not only of confidence in the evidence that underlies clinical practice and public policy, but the integrity of the research itself. For example, Coca-Cola’s sponsorship of public health research resulted in the reshaping of an entire field of research to focus on physical activity to reduce obesity, to the exclusion of nutrition-related research [18]. Not All Interests are Conflicts of Interest The growing concern about “nonfinancial” interests may be part of the recognition that the social context in which research is conducted influences the results. For example, researchers conducted replications of 100 experimental and correlational studies in psychology and found that over half of the replications produced weaker evidence for the original findings, despite attempts to use the author’s original materials [19]. This suggests that something other than the research design produced such variation. In science, there has been a common pretext that the researcher should approach the research without any desire to influence the outcome and that any conclusions should be driven solely by the data. Social scientists, however, have long argued that it is not possible to be impartial, disinterested, or value-neutral and that it is essential to acknowledge this as a means of being answerable for what science claims to know about the world [20–21]. Feminist scholars in particular have argued that it is not possible to be neutral in research as it is neither possible nor even desirable for scientists to be interest-free [21–22]. While it is essential to systematically examine all of the social values shaping a research process, these cannot possibly be eliminated but must instead be made visible and open to critical interpretation [23]. However, this does not mean that one cannot be objective or that all interests and values are merely relative. Instead, scholars have advocated for a “passionate detachment,” which emphasizes fairness, honesty, and recognition of the social positions and interests that influence how evidence is produced and shared [20–21]. Recognizing that scientists possess an interested view—grounded in personal experiences and beliefs, education and disciplinary training, and intellectual commitments—is crucial to rigorous research and healthy scientific debate, but it is distinct from conflict of interest. Scientists cannot be separated from their interests or their social position in the world. Recusal of scientists based on their personal beliefs and experience can serve exclusionary purposes and falsely identify certain individuals, who also possess personal beliefs and experiences, as “objective,” narrowing the diversity of perspectives involved in decision making. For example, a recent analysis of recusals from Food and Drug Administration (FDA) advisory committees on the basis of “intellectual conflicts of interest” found that each instance led to a decision that favored industry interests and that no expert had been excluded because he or she supported a particular drug or device [24]. Similarly, the effect of interests on research is isolated to the sphere of influence of the individual, and the direction of the bias created cannot be predicted. A scientist seeking academic glory may do so on either side of a research agenda—for example, climate scientists and climate change-deniers have achieved equal renown. It could be argued that researchers’ personal and intellectual interests are fundamental to good research and that these interests provide the momentum for scientific ideas to persist long enough to be scrutinized and, ultimately, to be useful. Managing Interests and Dealing with Conflicts Confusing “interests” with “conflicts of interests” makes conflicts of interest appear so pervasive that they cannot be avoided but only disclosed. Rather, there are precedents for disclosing and managing financial conflicts of interest, which should be adopted and enforced widely within biomedical research institutions. Disclosure is only a first step: requirements for disclosing financial conflicts of interest should be enforced, clear, and should list examples of financial conflicts of interest, following the example of the ICMJE [25]. Most disclosure policies currently rely on self-report; however, regulation such as the United States Physician Payments Sunshine Act could be expanded to include researchers and institutions to provide a more comprehensive source of data on financial conflicts of interest. While disclosure is essential in order to understand the scope of the problem and its effects, it is necessary that policy action go beyond disclosure to ensure that disclosures do not provide a moral license to dispense with their actual management [26]. Organizations, such as research institutions, scientific journals, and grant review panels should have policies for reviewing, mitigating, or eliminating financial conflicts of interest, and these should be enforced. Before engaging with a commercial entity, the proposed interaction should be subject to a thorough risk–benefit assessment by the research institution that includes the risk for reputational damage or loss of trust in the research activity [27]. For example, the Charles Perkins Centre at the University of Sydney has guidelines that require a committee to assess the alignment of a sponsor with the Centre’s mission; potential influence on the design, conduct, and publication of research; and the reputational risk to the institution [27]. Researchers should not engage with industry when the commercial interests are not aligned with improved public health or when the sponsor has any control over the design, conduct, or dissemination of the project [27]. Scientific institutions should consider prohibiting the acceptance of funding or publication of research when industry sponsorship or investigator conflicts of interest pose too high a risk. For example, several biomedical journals will not publish tobacco industry–funded research [28], and some universities do not accept tobacco industry funding for research [29]. The biomedical community needs new tools to account for the influence of interests and identities more generally on research. Reflexivity is a tool that could be borrowed from the social sciences and adapted to biomedical research that makes transparent and accounts for researchers’ professional and personal identities. It can be implemented at the level of the individual scientist but perhaps can be more effective when built into existing institutional processes. For example, if a university wished to set up a center on reproductive technology using CRISPR to genetically modify embryos, the steering committee would be well served to identify personal interests and past experience with the issue. Experts and those with particular opinions on the matter would not be recused, but perhaps opposing perspectives would be purposely included and processes put in place to ensure fair representation and an evidence-informed approach [30]. In lab settings, reflexivity has been used to make day-to-day research decisions transparent and to evaluate these decisions in terms of the interests at play [31–32]. Box 3 outlines key questions that are characteristic of reflexive processes [33]. Box 3. Key Questions for Reflexivity [33] Who is the researcher? What are their professional identities? What is their discipline, educational background, or training? Where are they employed? What is their career stage, and are they in a position of power or influence? What is their area of research or theoretical perspective? What are their advocacy positions? What are their relevant personal identities, including age, race/ethnicity, gender, religious or political affiliations, and life experience? How could who they are affect the design, conduct, or reporting of research? Who or what is the focus of the research? For whom does this have consequences? What are these consequences? Who or what is placed at risk by this research? How? Who or what is advantaged by this research? How? What are the ethical, social, political, or economic implications of this research? We use a grant funding review panel as an example in Table 1 to show how two parallel processes can be used simultaneously: reflexive processes to address relevant interests and comprehensive disclosure and management for conflicts of interest (Table 1). These practices could also be useful at the research agenda–setting phase, in determining funding priorities, or in peer review processes. The goal is to hold researchers accountable without discrediting scientific findings and claims as mere “personal biases” [33] and to ensure fair representation rather than excluding certain individuals based on their personal characteristics, beliefs, or expertise. Download: PPT PowerPoint slide PNG larger image TIFF original image Table 1. A grant proposal review panel: a hypothetical example of reflexivity and conflict of interest management. https://doi.org/10.1371/journal.pbio.2001221.t001 Box 3. Key Questions for Reflexivity [33] Who is the researcher? What are their professional identities? What is their discipline, educational background, or training? Where are they employed? What is their career stage, and are they in a position of power or influence? What is their area of research or theoretical perspective? What are their advocacy positions? What are their relevant personal identities, including age, race/ethnicity, gender, religious or political affiliations, and life experience? How could who they are affect the design, conduct, or reporting of research? Who or what is the focus of the research? For whom does this have consequences? What are these consequences? Who or what is placed at risk by this research? How? Who or what is advantaged by this research? How? What are the ethical, social, political, or economic implications of this research? Conclusion The muddying of the waters around conflicts of interest serves to erode the political will to identify and manage conflicts of interest in scientific research, as they are portrayed as increasingly ubiquitous and unsolvable. A substantial evidence base supports the need for continued policy action, particularly around the disclosure and management of commercial sponsorship of research and investigator financial conflicts of interest. Different strategies are needed to manage interests more broadly to ensure fair representation and accountability.

doi: 10.1371/journal.pbio.2000322pmid: 27923039

Introduction Human activities have increased the concentrations of CO2 and harmful air pollutants such as ozone in the troposphere. During the last 200 y, the CO2 concentration has increased from 280 to 400 ppm, and it is predicted to double relative to the preindustrial level by 2050 [1]. Elevated CO2 is likely to have complex effects on plant productivity, since CO2 is not only a driver of climate change but also the main substrate for photosynthesis. Altered atmospheric chemistry is not limited to CO2; the concentration of tropospheric ozone has more than doubled within the past 100 y [2]. Ozone is a notorious air pollutant causing severe damage to crops; present day global yield reductions caused by ozone range from 8.5%–14% for soybean, 3.9%–15% for wheat, and 2.2%–5.5% for maize [3]. Both CO2 and ozone enter the plant through stomata—small pores on the surfaces of plants that are formed by pairs of guard cells. Guard cells also regulate plant water balance since plants with more open stomata allow faster water evaporation. Water availability is the most limiting factor for agricultural production, and insufficient water supply can cause large reductions in crop yields [4]. Thus, plants are constantly facing a dilemma; assimilation of CO2 requires stomatal opening but also opens the gates for entrance of harmful air pollutants and leads to excessive water loss. A consequence of increased atmospheric CO2 concentration can be higher biomass production [5], but at the same time, plants adjust to elevated CO2 by partial closure of stomata [5,6] and expressing an altered developmental program that leads to reduced stomatal number [7]. CO2-induced stomatal closure reduces water loss; hence, it can directly modify plant water use efficiency (WUE)—carbon assimilated through photosynthesis versus water lost through stomata. Natural variation among Arabidopsis thaliana accessions provides a rich genetic resource for addressing plant function and adaptation to diverse environmental conditions. The Arabidopsis accession Cvi-0 from the Cape Verde Islands has impaired CO2 responses, more open stomata than Col-0, and is extremely sensitive to ozone treatment [8,9]. A single amino acid change in Cvi-0 MITOGEN-ACTIVATED PROTEIN KINASE 12 (MPK12) was recently shown to affect water use efficiency as well as stomatal size and to impair abscisic acid (ABA)-induced inhibition of stomatal opening [10]. MPK12 also regulates auxin signaling in roots [11]. However, the involvement of MPK12 in the CO2 signaling pathway in guard cells has not been addressed thus far. Among the important components of A. thaliana guard cell CO2 signaling are carbonic anhydrases (βCA1 and βCA4) that catalyze the conversion of CO2 to bicarbonate and the protein kinase HIGH LEAF TEMPERATURE 1 (HT1) that has been suggested to function as a negative regulator of CO2-induced stomatal movements [12,13]. Ultimately, for stomata to close, a signal from the bicarbonate has to activate protein kinases such as OPEN STOMATA 1 (OST1) that in turn activate plasma membrane anion channels, including SLOW ANION CHANNEL 1 (SLAC1), followed by extrusion of ions and water that causes stomatal closure [14–17]. Isolation of a dominant HT1 allele, ht1-8D, revealed that HT1 may directly inhibit OST1- and GUARD CELL HYDROGEN PEROXIDE-RESISTANT 1 (GHR1)-induced activation of SLAC1 [18]. Bicarbonate-induced activation of SLAC1 has been reconstituted in Xenopus laevis oocytes [19,20]. The pathway was shown to consist of RESISTANT TO HIGH CARBON DIOXIDE 1 (RHC1), HT1, OST1, and SLAC1 [19], while more recently the importance of CARBONIC ANHYDRASE 4 (βCA4), aquaporin PIP2;1, OST1, and SLAC1 was demonstrated [20]. Although guard cells are perhaps the best characterized single cell signaling system in the plant kingdom, there are still large gaps in our understanding of how CO2 signaling in guard cells is regulated and by which mechanism CO2 might regulate plant water management and WUE [5,21,22]. Here, we present the results of quantitative trait loci (QTL) mapping and sequencing of near-isogenic lines (NILs) of Cvi-0 ozone sensitivity. In a parallel approach, we mapped more open stomata and CO2-insensitivity phenotypes of a mutant cis (CO2 insensitive). A single amino acid change (G53R) in MPK12 and complete deletion of MPK12 are the causes of more open stomata and altered CO2 responses of Cvi-0 and cis, respectively. Based on kinase activity assays, we conclude that MPK12 acts as an inhibitor of the HT1 kinase, which represents a crucial step in the regulation of plant stomatal CO2 responses. Results Mapping of Cvi-0 Ozone Sensitivity Phenotypes Our initial QTL mapping of ozone sensitivity in Cvi-0 placed the two major contributing loci on the lower ends of chromosomes 2 and 3 [8]. To identify the causative loci related to the extreme ozone sensitivity and more open stomata of Cvi-0, we created a NIL termed Col-S (for Col-0 ozone sensitive) through eight generations of backcrossing of Cvi-0 with Col-0 (Fig 1A, S1A Fig, and S1 Video). In parallel, ozone tolerance from Col-0 was introgressed to Cvi-0 by six generations of backcrossing, which generated the ozone-tolerant Cvi-T (S1 Video). Using these accessions, NILs, and recombinant inbred lines (RILs), we mapped the causative ozone QTLs to a region of 90 kb on chromosome 2 and 17.70–18.18 Mbp on chromosome 3 (S1B Fig). We have previously shown that the QTL on chromosome 2 also controls plant water loss and stomatal function [8]. We isolated both QTLs by backcrossing Col-S with Col-0 and obtained the NILs Col-S2 and Col-S3. Both of these were less sensitive to ozone than Col-S (S1A Fig), indicating that these QTLs act additively to regulate ozone sensitivity. Col-S2 (but not Col-S3) showed much higher daytime stomatal conductance than Col-0 (Fig 1B). The mapping resolution on chromosome 3 was not sufficient to identify the causative gene. Hence, we focused on Col-S2 and its role in stomatal function. Within the 90-kb mapping region on chromosome 2, one gene, At2g46070, encoding a MAP kinase MPK12, shows strong preferential guard cell expression [23]. A single point mutation was found in Cvi-0 MPK12, leading to a glycine to arginine substitution at position 53 of the protein. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Mapping Cvi-0 ozone sensitivity and Cvi-0 and cis stomatal phenotypes. (A) Tissue damage after 6 h of O3 exposure (350 ppb). Visual damage of plant rosettes (upper images) and cell death visualized with trypan blue staining (lower images). Scale bars 1 cm. (B) Stomatal conductance of Col-0, Cvi-0, and NILs (mean ± standard error of the mean [SEM], n = 7–12). (C) Elevated CO2 (800 ppm) induced stomatal closure in intact whole plants (n = 9–10, except cas-2 [n = 3]). Experiment was repeated at least three times with similar results. (D) Stomatal half-response times to elevated CO2 (800 ppm). Error bars indicate ± SEM (n = 13). Pooled data from two experimental series are shown. (E) Gene model of MPK12 (At2g46070) and BYPASS2 (At2g46080). The deletion mutant cis (renamed as mpk12-4) has a 4,772 bp deletion (end and start indicated). Col-S2 has a G to C missense mutation at position 157 of MPK12, which leads to G53R substitution in MPK12. The mpk12-3 mutant has a Syngenta Arabidopsis Insertion Library (SAIL) transfer DNA (T-DNA) insertion in the second exon of MPK12. White boxes refer to exons, grey boxes to introns, and black lines to intergenic regions. Small letters (B) and asterisks (D) denote statistically significant differences according to one-way ANOVA with Tukey honest significant difference (HSD) for unequal sample size (Spjotvoll & Stoline test) or Tukey HSD post hoc test, respectively. The raw data for panels B–D can be found in S1 Data. https://doi.org/10.1371/journal.pbio.2000322.g001 Stomata-Related Phenotypes of Cvi-0 and cis are Caused by Mutations in MPK12 CAS (calcium-sensing receptor) is a chloroplast-localized protein important for proper stomatal responses to external Ca2+ [24,25]. While testing stomatal phenotypes in cas mutants, we observed phenotypic discrepancy between different alleles of cas. Whereas the cas-2 (GABI-665G12) line had more open stomata and impaired CO2 responses, this was neither observed in cas-1 nor in cas-3 (Fig 1C and S1C Fig). Further experiments showed that the T-DNA insert in the CAS gene was not linked to the CO2-insensitive phenotype of cas-2. In a backcross with Col-0, the T-DNA insert in cas-2 was removed, thereby generating the mutant cis (CO2 insensitive). Both cis and Col-S2 had impaired responses to high CO2 (800 ppm), leading to longer half-response times, but a residual CO2 response could still be observed (Fig 1D and S1E Fig). In order to identify the causative mutation in cis, mapping and whole genome sequencing of cis × C24 population was performed, which revealed a complete deletion of the MPK12 gene and its neighbor BYPASS2 in cis (Fig 1E and S1D Fig). Thus, cis was renamed mpk12-4. A second mutant (gdsl3-1) from the GABI-Kat collection (GABI-492D11) contained an identical deletion of BYPASS2 and MPK12 (S2 Fig). We also identified a line with a T-DNA insert in exon 2 of MPK12 from the SAIL collection (Fig 1E), which was recently named mpk12-3 [26]. No full-length transcript was found in mpk12-3 (S3 Fig). SALK T-DNA insertion lines of MPK12 were previously described as lethal [11,23]; similarly, we were unable to retrieve homozygous plants of the same alleles, possibly indicating the presence of an additional T-DNA in an essential gene. The new mpk12 deletion, SAIL T-DNA insertion, and Col-S2 point mutation alleles allowed a detailed characterization of the role of MPK12 in stomatal regulation. Stomatal conductance was higher throughout the day in all three lines (Col-S2, mpk12-3, and mpk12-4) (Fig 2A), suggesting that the amino acid substitution in Cvi-0 MPK12 leads to loss of function. Furthermore, Col-0 transformed with MPK12 from Cvi-0 showed stomatal conductance similar to Col-0, which excludes the option that the G53R substitution in MPK12 would lead to gain of function (S1F Fig). Moreover, the wild-type (Col-0) stomatal phenotype was observed in heterozygous F1 plants from a cross of Col-S2 and Col-gl1, in which the gl1 mutation that gives a trichome-less phenotype was used as a noninvasive method for selecting successfully crossed plants in the first generation (S1G Fig). Increased stomatal conductance may result from an increased number of stomata, larger stomata, or more open stomata. However, the stomatal index, length, and density did not differ between the lines, indicating that MPK12 regulates a function related to the stomatal aperture (S4 Fig). Because of the higher degree of stomatal opening, the instantaneous WUE was lower in mpk12-3, mpk12-4, and Col-S2 (Fig 2B). Altered WUE was previously also seen in mpk12-1 and a NIL with Cvi-0 MPK12 in Ler [10]. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Stomatal conductance of the NIL Col-S2, mpk12 mutants, and complementation lines. (A) Diurnal pattern of stomatal conductance with 12 h/12 h light–dark periods (n = 13–16). (B) Instantaneous water use efficiency (WUE) measured as an average of daytime light period from 09:00 to 17:00 (n = 13–16). (C) Stomatal conductance of Cvi-0 transformed with Col-0 MPK12 driven by its native promoter in T2 generation (n = 9). (D) Stomatal conductance of Col-S2 complementation line in T2 generation transformed with Col-0 MPK12, driven by its native promoter (n = 5–8). (E) Stomatal conductance of T3 transformants in the mpk12-4 background transformed with either the Col-0 or Cvi-0 version of MPK12, driven by its respective native promoter (n = 5–6). All graphs present mean ± SEM. Small letters denote statistically significant differences according to one-way ANOVA with Tukey HSD post hoc test for either unequal (B, D, E) or equal sample size (C). The raw data for panels A–E can be found in S1 Data. https://doi.org/10.1371/journal.pbio.2000322.g002 Cvi-0 and Col-S2 were complemented by expression of MPK12 from Col-0 (Fig 2C and 2D). Similarly, mpk12-4 was complemented by expression of Col-0 MPK12 but not by Cvi-0 MPK12 (Fig 2E). We conclude that MPK12 is a crucial regulator of stomatal conductance, and a single amino acid substitution (G53R) in Cvi-0 leads to loss of function of MPK12. MPK12 Functions in Guard Cell CO2 Signaling Reduction of CO2 levels inside the leaf [27] is a signal that indicates a shortage of substrate for photosynthesis and triggers stomatal opening. The rate of stomatal opening in response to low CO2 was severely impaired in mpk12 and Col-S2 (Fig 3A and S5A Fig). Another signal for stomatal opening is light; this response was intact in plants with impaired or absence of MPK12 (S5B and S5C Fig). The hormone ABA has dual roles in stomatal regulation; it induces stomatal closure but also inhibits light-induced stomatal opening. The latter response was impaired in mpk12 mutants and Col-S2 (Fig 3B and S5C Fig). Stomata close in response to several signals, including darkness, reduced air humidity, ozone pulse, elevated CO2, and ABA. Of these, only the response to elevated CO2 was impaired in mpk12 and Col-S2 (Fig 3C and 3D, S5D–S5H and S6 Figs). CO2 signaling is impaired in the carbonic anhydrase double mutant ca1 ca4 [13], and the product of carbonic anhydrase, bicarbonate, activates S-type anion currents [15]. In Col-S2 and mpk12-4, bicarbonate-induced S-type anion currents were strongly impaired (Fig 3E). Collectively, these data indicated that MPK12 has an important role in the regulation of CO2-induced stomatal movements in Arabidopsis. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Responsiveness of the NIL Col-S2 and mpk12 mutants to stomatal opening and closing stimuli. (A) Stomatal opening induced by 100 ppm CO2 in whole plants (58 min after induction; n = 12–13). (B) Light-induced stomatal opening inhibited by 2.5 μM ABA in whole plants (24 min after induction; n = 16–18). (C) Stomatal closure induced by 800 ppm CO2 in whole plants (10 min after induction; n = 12–13). (D) Stomatal closure induced by spraying whole plants with 5 μM ABA solution (24 min after induction; n = 12–14). (E) MPK12 is required for the bicarbonate (HCO3-)-induced slow type anion channel activation in guard cell protoplasts. Upper panels show typical whole guard cell protoplast recordings with 11.5 mM free HCO3- added to the pipette solution, and lower panels show average steady-state current-voltage relationships for wild-type (Col-0), NIL Col-S2, and mpk12-4 after treatment with mock or 11.5 mM HCO3- (n = 4–8 per line and treatment). Small letters (A, C) and asterisks (B, D) indicate statistically significant differences according to one-way ANOVA and two-way ANOVA with Tukey HSD for unequal sample size post hoc tests (p < 0.05), respectively. Error bars mark ± SEM. The raw data for panels A–E can be found in S1 Data. https://doi.org/10.1371/journal.pbio.2000322.g003 MPK12 Interacts with the Protein Kinase HT1 Only a few regulators of stomatal CO2 signaling in Arabidopsis have been identified. These include the protein kinases HT1 and OST1 [12,15,16]. To find the interaction partners of MPK12, we conducted pairwise split-ubiquitin yeast two-hybrid (Y2H) assays against several kinases and phosphatases involved in stomatal signaling (Fig 4A and 4B and S7A and S7B Fig). A strong interaction was observed between MPK12 and HT1 in yeast. The MPK12–HT1 interaction was also confirmed in Nicotiana benthamiana with bimolecular fluorescence complementation (BiFC) (Fig 4C–4E) and split luciferase complementation assays (S7C Fig). Strong interaction between MPK12 and HT1 was observed in the cell periphery (Fig 4C). Recently, HT1 was shown to be a plasma membrane–associated protein [28]. In contrast, Col-0 and Cvi-0 MPK12-YFP were located inside the cell (S8A–S8D Fig). Hence, it is likely that the interaction with HT1 brings MPK12 to the plasma membrane. HT1 interacted with both the Col-0 and Cvi-0 versions of MPK12, but the interaction with Cvi-0 MPK12 (G53R) was weaker both in quantitative BiFC and Y2H assays (Fig 4B and 4D). MPK11, an MPK from the same group as MPK12 [29], did not interact with HT1 (Fig 4C). INDOLE-3-BUTYRIC ACID RESPONSE 5 (IBR5) is a MPK phosphatase that regulates auxin signaling in roots and has been shown to interact with and regulate the activity of MPK12 [11]. We confirmed the interaction between MPK12 and IBR5 (S7A Fig). However, the ibr5-1 mutant exhibited wild-type stomatal phenotypes in response to CO2 changes (Fig 3A and 3C, and S5A and S5E Fig), suggesting that IBR5 is not required in stomatal CO2 signaling. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. MPK12 interacts with HT1. (A) Split-ubiquitin yeast two-hybrid assay on the SD-LeuTrp plate (left and middle panels) indicates the presence of both bait and prey plasmids; X-gal overlay assay (middle) and growth assay on the SD-LeuTrpHisAde plate (right) show HT1 interaction with MPK12 that is similar to the positive control (pAI-Alg5). Only weak or no interaction was detected with MPK12 G53R and MPK11, similar to the negative control (pDL2-Alg5). (B) Quantitative β-galactosidase assay from pools of ten colonies each. Activities are shown as the percentage of the positive control (± SEM; n = 3). (C) High-magnification (63x objective) BiFC images from a single infiltrated N. benthamiana leaf with identical confocal microscopy acquisition settings. Scale bar = 50 μm. (D) Ratiometric BiFC shows weaker interaction of MPK12 G53R than MPK12 with HT1, while MPK11 exhibits a weak interaction with HT1. The plasma membrane–localized SLAC1-CFP was used as an internal control. Eighteen images (from three leaves) of each construct set were analyzed. (E) Western blot together with Coomassie staining of proteins extracted from BiFC samples used for confocal imaging and controls with single construct shows expression of all fusion proteins. (F) Steady-state stomatal conductance of Col-S2 ht1-2, mpk12-4 ht1-2, and Col-S2 abi1-1 (ABA insensitive 1–1) double mutants (mean ± SEM, n = 11–13). Experiments were repeated at least three times. Letters in B, D, and F denote statistically significant differences with one-way ANOVA and Tukey HSD post hoc test for equal B, D, or unequal F sample size. The raw data for panels B, D, and F can be found in S1 Data. https://doi.org/10.1371/journal.pbio.2000322.g004 MPK12 Inhibits HT1 Activity The function of MPK12 in ABA and CO2 signaling was further explored through genetic analysis. A strong loss-of-function allele, ht1-2, that has low stomatal conductance [12] was used to evaluate the relationship between mpk12 and ht1-2. The Col-S2 ht1-2 and mpk12-4 ht1-2 double mutants had a more closed stomata phenotype similar to ht1-2 (Fig 4F), suggesting that HT1 is epistatic to MPK12. The strong impairment of stomatal function in abi1-1 (ABA insensitive1-1) was additive to Col-S2 in the double mutant Col-S2 abi1-1 (Fig 4F). Hence, signaling through MPK12 seems to act—at least to some extent—independently of the core ABA signaling pathway. Taken together, the MPK12-HT1 interaction (Fig 4A–4E) and the epistasis between ht1-2 and mpk12-4 (Fig 4F) suggest that MPK12 functions upstream of HT1 and could regulate the activity of HT1. To test this directly, we performed in vitro kinase assays with casein as the substrate for HT1 (Fig 5A). HT1 displayed strong autophosphorylation and phosphorylated casein efficiently. Addition of the Col-0 version of MPK12 and a hyperactive version (MPK12 Y122C) efficiently inhibited HT1 activity (Fig 5A and quantified in Fig 5B). A point-mutated version (MPK12 K70R) designed to remove the kinase activity of MPK12 also inhibited the autophosphorylation activity of HT1 and phosphorylation of casein by HT1, although it was less efficient than the wild-type (Fig 5B). Importantly, the Cvi-0 version of MPK12 (G53R) displayed strongly suppressed inhibition of HT1 activity (Fig 5A and 5B). MPK12 did not phosphorylate the kinase-dead version of HT1 (K113M) (Fig 5C). The kinase-dead version of HT1 (K113M) was used as a substrate, since the strong autophosphorylation activity of HT1 would otherwise have obscured the result. Wild-type MPK12 and hyperactive MPK12 (Y122C) displayed autophosphorylation, whereas MPK12 (G53R) as well as MPK12 (K70R) had lost their autophosphorylation activity, indicating that the G53R substitution in Cvi-0 MPK12 disrupts the kinase activity of the protein (Fig 5C). The inhibition of HT1 by MPK12 was specific, as MPK11, which belongs to the same group as MPK12, was not able to affect HT1 kinase activity (S9 Fig). We conclude that the stomatal phenotypes of mpk12 mutants and Cvi-0 can be explained by a lack of inhibition of HT1 activity by MPK12, which leads to more open stomata and impaired CO2 responses (Figs 2A, 3 and 5A and 5C). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. Regulation of HT1 by MPK12. (A) Inhibition of HT1 kinase activity in vitro by different versions of MPK12 (MPK12 G53R—Cvi-0 version of MPK12; MPK12 K70R—inactive kinase; MPK12 Y122C—hyperactive kinase). Upper panel: autoradiography of the SDS PAGE gel; lower panel: Coomassie-stained SDS PAGE. Reaction mixture was incubated for 30 min. (B) Casein phosphorylation by HT1 with different MPK12 concentrations (mean ± SEM; n = 3). The raw data can be found in S1 Data. (C) Kinase-dead HT1 K113M was not in vitro phosphorylated by different versions of MPK12, and only MPK12 and MPK12 (Y122C) display clear autophosphorylation activities. https://doi.org/10.1371/journal.pbio.2000322.g005 Both MPK12 and MPK4 Regulate the CO2 Signaling Pathway MPK12 belongs to the same group of MPKs as MPK4, a crucial regulator of pathogen and stress responses [29]. In tobacco, the silencing of MPK4 impaired CO2-induced stomatal closure [30]. Since Arabidopsis MPK4 and MPK12 are highly similar [31], it is possible that both MPK4 and MPK12 could regulate stomatal CO2 responses. Indeed, in an Y2H screen to identify HT1 interacting proteins, one prominent interactor was MPK4 [18]. The Arabidopsis mpk4 mutant is severely dwarfed, and measurements of accurate stomatal conductance with these plants are not feasible [32]. However, the impaired stomatal response to CO2 in mpk12-4 (Fig 3) was further enhanced by guard cell–specific silencing of MPK4 [18]; hence, in guard cells MPK4 is acting redundantly with MPK12 in stomatal CO2 signaling. Furthermore, MPK4 could also inhibit HT1 kinase activity [18]. The G53 residue in MPK12 is conserved in all Arabidopsis MPKs [10]. Since the G53R mutation blocked MPK12 function, we tested whether a similar mutation would impair MPK4 function. This experiment showed that MPK4-induced inhibition of HT1 activity was blocked by the introduction of a G55R mutation in MPK4; this mutation corresponds to G53R in Cvi-0 MPK12 (Fig 6A). Since MPK11 did not inhibit HT1 activity (S9 Fig), the function of MPKs as kinase inhibitors in Arabidopsis may be restricted to MPK12 and its closest homologue MPK4. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 6. A conserved glycine is important for MPK4 and MPK12 function. (A) Inhibition of HT1 kinase activity in vitro by MPK4 and MPK4 G55R. Upper panel: autoradiography of the SDS PAGE gel; lower panel: Coomassie-stained SDS PAGE. Reaction mixture was incubated for 30 min. (B) Whole protein (left) and close-up (right) view of the superposition of models for MPK12 wild-type (secondary structure and surface in white) and MPK12 G53R (secondary structure in green). There is a close structural similarity between the structures except where the arginine at position 53 protrudes from the mutant protein surface and changes the loop region for the mutant. (C) Whole protein (left) and close-up (right) view of the superposition of models for MPK4 wild-type (secondary structure and surface in white) and MPK4 G55R (secondary structure in yellow). Similar to MPK12 G53R, the arginine at position 55 in MPK4 protrudes from the mutant protein surface and changes the loop region. https://doi.org/10.1371/journal.pbio.2000322.g006 The Arabidopsis MPK6 crystal structure [33] was used to model the structure of MPK4 and MPK12 and to address the role of the G55R and G53R mutations that were shown to be crucial for the function of these proteins (Fig 6B and 6C). The mutation of Gly to Arg in position 53 in MPK12 caused the protrusion of the arginine sidechain on the surface of the protein, which could affect its binding affinity for other proteins in addition to an altered structure of the loop region. Similarly, the Arg in position 55 of MPK4 protruded from the surface as compared to the wild-type. Thus, the MPK12 G53R and MPK4 G55R amino acid substitutions may alter protein binding affinities of these MPKs to other proteins. Collectively, the presented experiments suggest that the CO2 signal leading to stomatal movements is transmitted through MPK12 and MPK4, leading to inhibition of HT1, and this enables SLAC1 activation by its activators, including OST1 and GHR1. Neither MPK12 G53R from Cvi-0 nor MPK4 G55R can fully inhibit HT1 (Fig 7). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 7. Schematic model of molecular events during elevated CO2-induced stomatal closure. (Left guard cell) CO2 enters guard cells through the PIP2;1 aquaporin [20] and is converted to bicarbonate by carbonic anhydrases βCA4 and βCA1. The mechanism by which bicarbonate is sensed in guard cells still needs to be resolved; nevertheless, it is likely that in elevated CO2 conditions, activation of MPK12 and MPK4 leads to inhibition of HT1, and this enables activation of slow-type anion channel SLAC1 by OST1 [15]. Additionally, GHR1 participates in the regulation of SLAC1 activity and is involved in CO2-induced stomatal closure [18]. Bicarbonate-induced inhibition of HT1 by RHC1 has also been shown [19]. (Right guard cell) The G53R mutation in Cvi-0 MPK12, as well as G55R mutation in MPK4, decreases the ability of these MPKs to inhibit HT1 kinase activity, which results in enhanced inhibition of SLAC1 activity by HT1 and decreased sensitivity to CO2 in stomatal closure. https://doi.org/10.1371/journal.pbio.2000322.g007 Mapping of Cvi-0 Ozone Sensitivity Phenotypes Our initial QTL mapping of ozone sensitivity in Cvi-0 placed the two major contributing loci on the lower ends of chromosomes 2 and 3 [8]. To identify the causative loci related to the extreme ozone sensitivity and more open stomata of Cvi-0, we created a NIL termed Col-S (for Col-0 ozone sensitive) through eight generations of backcrossing of Cvi-0 with Col-0 (Fig 1A, S1A Fig, and S1 Video). In parallel, ozone tolerance from Col-0 was introgressed to Cvi-0 by six generations of backcrossing, which generated the ozone-tolerant Cvi-T (S1 Video). Using these accessions, NILs, and recombinant inbred lines (RILs), we mapped the causative ozone QTLs to a region of 90 kb on chromosome 2 and 17.70–18.18 Mbp on chromosome 3 (S1B Fig). We have previously shown that the QTL on chromosome 2 also controls plant water loss and stomatal function [8]. We isolated both QTLs by backcrossing Col-S with Col-0 and obtained the NILs Col-S2 and Col-S3. Both of these were less sensitive to ozone than Col-S (S1A Fig), indicating that these QTLs act additively to regulate ozone sensitivity. Col-S2 (but not Col-S3) showed much higher daytime stomatal conductance than Col-0 (Fig 1B). The mapping resolution on chromosome 3 was not sufficient to identify the causative gene. Hence, we focused on Col-S2 and its role in stomatal function. Within the 90-kb mapping region on chromosome 2, one gene, At2g46070, encoding a MAP kinase MPK12, shows strong preferential guard cell expression [23]. A single point mutation was found in Cvi-0 MPK12, leading to a glycine to arginine substitution at position 53 of the protein. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Mapping Cvi-0 ozone sensitivity and Cvi-0 and cis stomatal phenotypes. (A) Tissue damage after 6 h of O3 exposure (350 ppb). Visual damage of plant rosettes (upper images) and cell death visualized with trypan blue staining (lower images). Scale bars 1 cm. (B) Stomatal conductance of Col-0, Cvi-0, and NILs (mean ± standard error of the mean [SEM], n = 7–12). (C) Elevated CO2 (800 ppm) induced stomatal closure in intact whole plants (n = 9–10, except cas-2 [n = 3]). Experiment was repeated at least three times with similar results. (D) Stomatal half-response times to elevated CO2 (800 ppm). Error bars indicate ± SEM (n = 13). Pooled data from two experimental series are shown. (E) Gene model of MPK12 (At2g46070) and BYPASS2 (At2g46080). The deletion mutant cis (renamed as mpk12-4) has a 4,772 bp deletion (end and start indicated). Col-S2 has a G to C missense mutation at position 157 of MPK12, which leads to G53R substitution in MPK12. The mpk12-3 mutant has a Syngenta Arabidopsis Insertion Library (SAIL) transfer DNA (T-DNA) insertion in the second exon of MPK12. White boxes refer to exons, grey boxes to introns, and black lines to intergenic regions. Small letters (B) and asterisks (D) denote statistically significant differences according to one-way ANOVA with Tukey honest significant difference (HSD) for unequal sample size (Spjotvoll & Stoline test) or Tukey HSD post hoc test, respectively. The raw data for panels B–D can be found in S1 Data. https://doi.org/10.1371/journal.pbio.2000322.g001 Stomata-Related Phenotypes of Cvi-0 and cis are Caused by Mutations in MPK12 CAS (calcium-sensing receptor) is a chloroplast-localized protein important for proper stomatal responses to external Ca2+ [24,25]. While testing stomatal phenotypes in cas mutants, we observed phenotypic discrepancy between different alleles of cas. Whereas the cas-2 (GABI-665G12) line had more open stomata and impaired CO2 responses, this was neither observed in cas-1 nor in cas-3 (Fig 1C and S1C Fig). Further experiments showed that the T-DNA insert in the CAS gene was not linked to the CO2-insensitive phenotype of cas-2. In a backcross with Col-0, the T-DNA insert in cas-2 was removed, thereby generating the mutant cis (CO2 insensitive). Both cis and Col-S2 had impaired responses to high CO2 (800 ppm), leading to longer half-response times, but a residual CO2 response could still be observed (Fig 1D and S1E Fig). In order to identify the causative mutation in cis, mapping and whole genome sequencing of cis × C24 population was performed, which revealed a complete deletion of the MPK12 gene and its neighbor BYPASS2 in cis (Fig 1E and S1D Fig). Thus, cis was renamed mpk12-4. A second mutant (gdsl3-1) from the GABI-Kat collection (GABI-492D11) contained an identical deletion of BYPASS2 and MPK12 (S2 Fig). We also identified a line with a T-DNA insert in exon 2 of MPK12 from the SAIL collection (Fig 1E), which was recently named mpk12-3 [26]. No full-length transcript was found in mpk12-3 (S3 Fig). SALK T-DNA insertion lines of MPK12 were previously described as lethal [11,23]; similarly, we were unable to retrieve homozygous plants of the same alleles, possibly indicating the presence of an additional T-DNA in an essential gene. The new mpk12 deletion, SAIL T-DNA insertion, and Col-S2 point mutation alleles allowed a detailed characterization of the role of MPK12 in stomatal regulation. Stomatal conductance was higher throughout the day in all three lines (Col-S2, mpk12-3, and mpk12-4) (Fig 2A), suggesting that the amino acid substitution in Cvi-0 MPK12 leads to loss of function. Furthermore, Col-0 transformed with MPK12 from Cvi-0 showed stomatal conductance similar to Col-0, which excludes the option that the G53R substitution in MPK12 would lead to gain of function (S1F Fig). Moreover, the wild-type (Col-0) stomatal phenotype was observed in heterozygous F1 plants from a cross of Col-S2 and Col-gl1, in which the gl1 mutation that gives a trichome-less phenotype was used as a noninvasive method for selecting successfully crossed plants in the first generation (S1G Fig). Increased stomatal conductance may result from an increased number of stomata, larger stomata, or more open stomata. However, the stomatal index, length, and density did not differ between the lines, indicating that MPK12 regulates a function related to the stomatal aperture (S4 Fig). Because of the higher degree of stomatal opening, the instantaneous WUE was lower in mpk12-3, mpk12-4, and Col-S2 (Fig 2B). Altered WUE was previously also seen in mpk12-1 and a NIL with Cvi-0 MPK12 in Ler [10]. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Stomatal conductance of the NIL Col-S2, mpk12 mutants, and complementation lines. (A) Diurnal pattern of stomatal conductance with 12 h/12 h light–dark periods (n = 13–16). (B) Instantaneous water use efficiency (WUE) measured as an average of daytime light period from 09:00 to 17:00 (n = 13–16). (C) Stomatal conductance of Cvi-0 transformed with Col-0 MPK12 driven by its native promoter in T2 generation (n = 9). (D) Stomatal conductance of Col-S2 complementation line in T2 generation transformed with Col-0 MPK12, driven by its native promoter (n = 5–8). (E) Stomatal conductance of T3 transformants in the mpk12-4 background transformed with either the Col-0 or Cvi-0 version of MPK12, driven by its respective native promoter (n = 5–6). All graphs present mean ± SEM. Small letters denote statistically significant differences according to one-way ANOVA with Tukey HSD post hoc test for either unequal (B, D, E) or equal sample size (C). The raw data for panels A–E can be found in S1 Data. https://doi.org/10.1371/journal.pbio.2000322.g002 Cvi-0 and Col-S2 were complemented by expression of MPK12 from Col-0 (Fig 2C and 2D). Similarly, mpk12-4 was complemented by expression of Col-0 MPK12 but not by Cvi-0 MPK12 (Fig 2E). We conclude that MPK12 is a crucial regulator of stomatal conductance, and a single amino acid substitution (G53R) in Cvi-0 leads to loss of function of MPK12. MPK12 Functions in Guard Cell CO2 Signaling Reduction of CO2 levels inside the leaf [27] is a signal that indicates a shortage of substrate for photosynthesis and triggers stomatal opening. The rate of stomatal opening in response to low CO2 was severely impaired in mpk12 and Col-S2 (Fig 3A and S5A Fig). Another signal for stomatal opening is light; this response was intact in plants with impaired or absence of MPK12 (S5B and S5C Fig). The hormone ABA has dual roles in stomatal regulation; it induces stomatal closure but also inhibits light-induced stomatal opening. The latter response was impaired in mpk12 mutants and Col-S2 (Fig 3B and S5C Fig). Stomata close in response to several signals, including darkness, reduced air humidity, ozone pulse, elevated CO2, and ABA. Of these, only the response to elevated CO2 was impaired in mpk12 and Col-S2 (Fig 3C and 3D, S5D–S5H and S6 Figs). CO2 signaling is impaired in the carbonic anhydrase double mutant ca1 ca4 [13], and the product of carbonic anhydrase, bicarbonate, activates S-type anion currents [15]. In Col-S2 and mpk12-4, bicarbonate-induced S-type anion currents were strongly impaired (Fig 3E). Collectively, these data indicated that MPK12 has an important role in the regulation of CO2-induced stomatal movements in Arabidopsis. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Responsiveness of the NIL Col-S2 and mpk12 mutants to stomatal opening and closing stimuli. (A) Stomatal opening induced by 100 ppm CO2 in whole plants (58 min after induction; n = 12–13). (B) Light-induced stomatal opening inhibited by 2.5 μM ABA in whole plants (24 min after induction; n = 16–18). (C) Stomatal closure induced by 800 ppm CO2 in whole plants (10 min after induction; n = 12–13). (D) Stomatal closure induced by spraying whole plants with 5 μM ABA solution (24 min after induction; n = 12–14). (E) MPK12 is required for the bicarbonate (HCO3-)-induced slow type anion channel activation in guard cell protoplasts. Upper panels show typical whole guard cell protoplast recordings with 11.5 mM free HCO3- added to the pipette solution, and lower panels show average steady-state current-voltage relationships for wild-type (Col-0), NIL Col-S2, and mpk12-4 after treatment with mock or 11.5 mM HCO3- (n = 4–8 per line and treatment). Small letters (A, C) and asterisks (B, D) indicate statistically significant differences according to one-way ANOVA and two-way ANOVA with Tukey HSD for unequal sample size post hoc tests (p < 0.05), respectively. Error bars mark ± SEM. The raw data for panels A–E can be found in S1 Data. https://doi.org/10.1371/journal.pbio.2000322.g003 MPK12 Interacts with the Protein Kinase HT1 Only a few regulators of stomatal CO2 signaling in Arabidopsis have been identified. These include the protein kinases HT1 and OST1 [12,15,16]. To find the interaction partners of MPK12, we conducted pairwise split-ubiquitin yeast two-hybrid (Y2H) assays against several kinases and phosphatases involved in stomatal signaling (Fig 4A and 4B and S7A and S7B Fig). A strong interaction was observed between MPK12 and HT1 in yeast. The MPK12–HT1 interaction was also confirmed in Nicotiana benthamiana with bimolecular fluorescence complementation (BiFC) (Fig 4C–4E) and split luciferase complementation assays (S7C Fig). Strong interaction between MPK12 and HT1 was observed in the cell periphery (Fig 4C). Recently, HT1 was shown to be a plasma membrane–associated protein [28]. In contrast, Col-0 and Cvi-0 MPK12-YFP were located inside the cell (S8A–S8D Fig). Hence, it is likely that the interaction with HT1 brings MPK12 to the plasma membrane. HT1 interacted with both the Col-0 and Cvi-0 versions of MPK12, but the interaction with Cvi-0 MPK12 (G53R) was weaker both in quantitative BiFC and Y2H assays (Fig 4B and 4D). MPK11, an MPK from the same group as MPK12 [29], did not interact with HT1 (Fig 4C). INDOLE-3-BUTYRIC ACID RESPONSE 5 (IBR5) is a MPK phosphatase that regulates auxin signaling in roots and has been shown to interact with and regulate the activity of MPK12 [11]. We confirmed the interaction between MPK12 and IBR5 (S7A Fig). However, the ibr5-1 mutant exhibited wild-type stomatal phenotypes in response to CO2 changes (Fig 3A and 3C, and S5A and S5E Fig), suggesting that IBR5 is not required in stomatal CO2 signaling. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. MPK12 interacts with HT1. (A) Split-ubiquitin yeast two-hybrid assay on the SD-LeuTrp plate (left and middle panels) indicates the presence of both bait and prey plasmids; X-gal overlay assay (middle) and growth assay on the SD-LeuTrpHisAde plate (right) show HT1 interaction with MPK12 that is similar to the positive control (pAI-Alg5). Only weak or no interaction was detected with MPK12 G53R and MPK11, similar to the negative control (pDL2-Alg5). (B) Quantitative β-galactosidase assay from pools of ten colonies each. Activities are shown as the percentage of the positive control (± SEM; n = 3). (C) High-magnification (63x objective) BiFC images from a single infiltrated N. benthamiana leaf with identical confocal microscopy acquisition settings. Scale bar = 50 μm. (D) Ratiometric BiFC shows weaker interaction of MPK12 G53R than MPK12 with HT1, while MPK11 exhibits a weak interaction with HT1. The plasma membrane–localized SLAC1-CFP was used as an internal control. Eighteen images (from three leaves) of each construct set were analyzed. (E) Western blot together with Coomassie staining of proteins extracted from BiFC samples used for confocal imaging and controls with single construct shows expression of all fusion proteins. (F) Steady-state stomatal conductance of Col-S2 ht1-2, mpk12-4 ht1-2, and Col-S2 abi1-1 (ABA insensitive 1–1) double mutants (mean ± SEM, n = 11–13). Experiments were repeated at least three times. Letters in B, D, and F denote statistically significant differences with one-way ANOVA and Tukey HSD post hoc test for equal B, D, or unequal F sample size. The raw data for panels B, D, and F can be found in S1 Data. https://doi.org/10.1371/journal.pbio.2000322.g004 MPK12 Inhibits HT1 Activity The function of MPK12 in ABA and CO2 signaling was further explored through genetic analysis. A strong loss-of-function allele, ht1-2, that has low stomatal conductance [12] was used to evaluate the relationship between mpk12 and ht1-2. The Col-S2 ht1-2 and mpk12-4 ht1-2 double mutants had a more closed stomata phenotype similar to ht1-2 (Fig 4F), suggesting that HT1 is epistatic to MPK12. The strong impairment of stomatal function in abi1-1 (ABA insensitive1-1) was additive to Col-S2 in the double mutant Col-S2 abi1-1 (Fig 4F). Hence, signaling through MPK12 seems to act—at least to some extent—independently of the core ABA signaling pathway. Taken together, the MPK12-HT1 interaction (Fig 4A–4E) and the epistasis between ht1-2 and mpk12-4 (Fig 4F) suggest that MPK12 functions upstream of HT1 and could regulate the activity of HT1. To test this directly, we performed in vitro kinase assays with casein as the substrate for HT1 (Fig 5A). HT1 displayed strong autophosphorylation and phosphorylated casein efficiently. Addition of the Col-0 version of MPK12 and a hyperactive version (MPK12 Y122C) efficiently inhibited HT1 activity (Fig 5A and quantified in Fig 5B). A point-mutated version (MPK12 K70R) designed to remove the kinase activity of MPK12 also inhibited the autophosphorylation activity of HT1 and phosphorylation of casein by HT1, although it was less efficient than the wild-type (Fig 5B). Importantly, the Cvi-0 version of MPK12 (G53R) displayed strongly suppressed inhibition of HT1 activity (Fig 5A and 5B). MPK12 did not phosphorylate the kinase-dead version of HT1 (K113M) (Fig 5C). The kinase-dead version of HT1 (K113M) was used as a substrate, since the strong autophosphorylation activity of HT1 would otherwise have obscured the result. Wild-type MPK12 and hyperactive MPK12 (Y122C) displayed autophosphorylation, whereas MPK12 (G53R) as well as MPK12 (K70R) had lost their autophosphorylation activity, indicating that the G53R substitution in Cvi-0 MPK12 disrupts the kinase activity of the protein (Fig 5C). The inhibition of HT1 by MPK12 was specific, as MPK11, which belongs to the same group as MPK12, was not able to affect HT1 kinase activity (S9 Fig). We conclude that the stomatal phenotypes of mpk12 mutants and Cvi-0 can be explained by a lack of inhibition of HT1 activity by MPK12, which leads to more open stomata and impaired CO2 responses (Figs 2A, 3 and 5A and 5C). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. Regulation of HT1 by MPK12. (A) Inhibition of HT1 kinase activity in vitro by different versions of MPK12 (MPK12 G53R—Cvi-0 version of MPK12; MPK12 K70R—inactive kinase; MPK12 Y122C—hyperactive kinase). Upper panel: autoradiography of the SDS PAGE gel; lower panel: Coomassie-stained SDS PAGE. Reaction mixture was incubated for 30 min. (B) Casein phosphorylation by HT1 with different MPK12 concentrations (mean ± SEM; n = 3). The raw data can be found in S1 Data. (C) Kinase-dead HT1 K113M was not in vitro phosphorylated by different versions of MPK12, and only MPK12 and MPK12 (Y122C) display clear autophosphorylation activities. https://doi.org/10.1371/journal.pbio.2000322.g005 Both MPK12 and MPK4 Regulate the CO2 Signaling Pathway MPK12 belongs to the same group of MPKs as MPK4, a crucial regulator of pathogen and stress responses [29]. In tobacco, the silencing of MPK4 impaired CO2-induced stomatal closure [30]. Since Arabidopsis MPK4 and MPK12 are highly similar [31], it is possible that both MPK4 and MPK12 could regulate stomatal CO2 responses. Indeed, in an Y2H screen to identify HT1 interacting proteins, one prominent interactor was MPK4 [18]. The Arabidopsis mpk4 mutant is severely dwarfed, and measurements of accurate stomatal conductance with these plants are not feasible [32]. However, the impaired stomatal response to CO2 in mpk12-4 (Fig 3) was further enhanced by guard cell–specific silencing of MPK4 [18]; hence, in guard cells MPK4 is acting redundantly with MPK12 in stomatal CO2 signaling. Furthermore, MPK4 could also inhibit HT1 kinase activity [18]. The G53 residue in MPK12 is conserved in all Arabidopsis MPKs [10]. Since the G53R mutation blocked MPK12 function, we tested whether a similar mutation would impair MPK4 function. This experiment showed that MPK4-induced inhibition of HT1 activity was blocked by the introduction of a G55R mutation in MPK4; this mutation corresponds to G53R in Cvi-0 MPK12 (Fig 6A). Since MPK11 did not inhibit HT1 activity (S9 Fig), the function of MPKs as kinase inhibitors in Arabidopsis may be restricted to MPK12 and its closest homologue MPK4. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 6. A conserved glycine is important for MPK4 and MPK12 function. (A) Inhibition of HT1 kinase activity in vitro by MPK4 and MPK4 G55R. Upper panel: autoradiography of the SDS PAGE gel; lower panel: Coomassie-stained SDS PAGE. Reaction mixture was incubated for 30 min. (B) Whole protein (left) and close-up (right) view of the superposition of models for MPK12 wild-type (secondary structure and surface in white) and MPK12 G53R (secondary structure in green). There is a close structural similarity between the structures except where the arginine at position 53 protrudes from the mutant protein surface and changes the loop region for the mutant. (C) Whole protein (left) and close-up (right) view of the superposition of models for MPK4 wild-type (secondary structure and surface in white) and MPK4 G55R (secondary structure in yellow). Similar to MPK12 G53R, the arginine at position 55 in MPK4 protrudes from the mutant protein surface and changes the loop region. https://doi.org/10.1371/journal.pbio.2000322.g006 The Arabidopsis MPK6 crystal structure [33] was used to model the structure of MPK4 and MPK12 and to address the role of the G55R and G53R mutations that were shown to be crucial for the function of these proteins (Fig 6B and 6C). The mutation of Gly to Arg in position 53 in MPK12 caused the protrusion of the arginine sidechain on the surface of the protein, which could affect its binding affinity for other proteins in addition to an altered structure of the loop region. Similarly, the Arg in position 55 of MPK4 protruded from the surface as compared to the wild-type. Thus, the MPK12 G53R and MPK4 G55R amino acid substitutions may alter protein binding affinities of these MPKs to other proteins. Collectively, the presented experiments suggest that the CO2 signal leading to stomatal movements is transmitted through MPK12 and MPK4, leading to inhibition of HT1, and this enables SLAC1 activation by its activators, including OST1 and GHR1. Neither MPK12 G53R from Cvi-0 nor MPK4 G55R can fully inhibit HT1 (Fig 7). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 7. Schematic model of molecular events during elevated CO2-induced stomatal closure. (Left guard cell) CO2 enters guard cells through the PIP2;1 aquaporin [20] and is converted to bicarbonate by carbonic anhydrases βCA4 and βCA1. The mechanism by which bicarbonate is sensed in guard cells still needs to be resolved; nevertheless, it is likely that in elevated CO2 conditions, activation of MPK12 and MPK4 leads to inhibition of HT1, and this enables activation of slow-type anion channel SLAC1 by OST1 [15]. Additionally, GHR1 participates in the regulation of SLAC1 activity and is involved in CO2-induced stomatal closure [18]. Bicarbonate-induced inhibition of HT1 by RHC1 has also been shown [19]. (Right guard cell) The G53R mutation in Cvi-0 MPK12, as well as G55R mutation in MPK4, decreases the ability of these MPKs to inhibit HT1 kinase activity, which results in enhanced inhibition of SLAC1 activity by HT1 and decreased sensitivity to CO2 in stomatal closure. https://doi.org/10.1371/journal.pbio.2000322.g007 Discussion Natural variation within a species holds great potential to identify regulatory mechanisms that are not easily uncovered through mutant screens. The Cvi-0 accession originates from the southern border of the Arabidopsis distribution area, the Cape Verde Islands. The Ler × Cvi RIL population was one of the first RILs produced, and it has been phenotyped for multiple traits [34]. Despite this, only a few QTLs from Cvi-0 have been identified at the molecular level. Our earlier research identified a locus related to ozone sensitivity and more open stomata phenotype of Cvi-0 in chromosome 2 [8]. Recently, the G53R substitution in MPK12 that affects plant water use efficiency was identified by using the Ler × Cvi populations, but the biochemical function of MPK12 in stomatal regulation was not further investigated [10]. Here, we generated NILs by backcrossing Cvi-0 eight times to Col-0 and show that the same natural mutation in Cvi-0 and lack of MPK12 in cis are the causes of ozone sensitivity, more open stomata, and altered CO2 responses of Arabidopsis plants. Furthermore, we showed that MPK12 regulates the activity of the protein kinase HT1, a major component of the CO2 signaling pathway in guard cells. The regulators of HT1 have remained largely unknown, despite the exceptionally strong CO2-insensitivity phenotype of plants with impaired HT1 function [12,15]. Our findings provide the first evidence for the role of MPK12 in guard cell CO2 signaling and provide a mechanistic insight for the MPK12 function in the regulation of plant water management. The role of MPKs in Arabidopsis guard cell signaling has concentrated on MPK9 and MPK12, which are preferentially expressed in guard cells. Plants with point mutations in MPK9 (mpk9-1, L295F) and MPK12 (mpk12-1, T220I) had wild-type ABA responses, but mpk12-1 has decreased WUE [10]. The mpk9-1, mpk12-1, and mpk12-2 alleles are Tilling (Targeting Induced Local Lesions IN Genomes) lines in the Col-erecta background and the previously characterized MPK12-Cvi NIL is in the Ler background [10,23]. Mutations in ERECTA modify transpiration efficiency and stomatal density, which may have influenced some of the previously described mpk12-1 phenotypes [10,35]. In contrast, the full knockout alleles described here, mpk12-3 and mpk12-4, are in Col-0 and imply a major function for MPK12 in CO2 signaling. Additional roles for MPK12 in stomatal responses have been inferred through the use of the double mutant mpk9-1 mpk12-1 that has impaired stomatal closure responses to ABA and H2O2 treatment and has impaired S-type anion channel activation in response to ABA and Ca2+ [23]. It is also highly susceptible to Pseudomonas syringae infection and impaired in yeast elicitor-, chitosan-, and methyl jasmonate–induced stomatal closure [36]. Since the mpk9 mpk12 double mutant appears to be more severely impaired in abiotic and biotic stomatal responses and S-type anion channel activation than the loss of function MPK12 alleles (Fig 3), it is possible that MPK12 together with MPK9 regulates stomatal aperture in response to various signals. MPK12 also regulates auxin responses in the root [11,26]. However, beyond the observation that plants with impaired MPK12 are hypersensitive to auxin inhibition of root growth, no details about the targets of MPK12 in roots are known. HT1 was the first component shown to be specifically associated with stomatal CO2 signaling, and the ht1-2 mutant has more closed stomata displaying constitutive high CO2 response at ambient CO2 levels (Fig 4D [12]). The opposite phenotypes of mpk12 and ht1-2 allowed us to use genetic analysis to position MPK12 in the guard cell signaling network. The stomata of mpk12 ht1-2 were more closed, thus positioning MPK12 upstream of HT1 and possibly as a direct regulator of HT1 (Fig 4D). CO2 signaling in guard cells is initiated through the production of bicarbonate by carbonic anhydrases, and bicarbonate initiates signaling leading to activation of S-type anion channels [13,15]. In mpk12, the bicarbonate-dependent activation of S-type anion channels was impaired, as was previously found for the plants with impaired OST1 and SLAC1 (Fig 3E) [15]. The combined evidence from mpk12 phenotypes, genetic analysis, and measurements of S-type anion currents all pointed towards MPK12 as a crucial regulator of CO2 signaling acting through HT1. Indeed, HT1 kinase activity was inhibited in the presence of Col-0 MPK12 but not by the Cvi-0 version of MPK12 (Fig 5A). Thus, the inhibitory function of MPK12 was impaired by the G53R amino acid substitution, probably by its weaker interaction with HT1 (Figs 4 and 5A). This explains the similar phenotypes of the NIL Col-S2, mpk12-3, and mpk12-4; they all display lack of inhibition of the negative regulator HT1, leading to higher stomatal conductance. Further support for the regulatory interplay between HT1 and MPK12 is provided by the isolation of a dominant mutation in HT1, ht1-8D, which in contrast to ht1-2 has constitutively more open stomata and is biochemically resistant to inhibition by MPK12 [18]. Cvi-0 has altered phenotypes in many traits, including drought and pathogen resistance [34,37,38]. All of these traits are regulated through stomatal function; thus, the MPK12-HT1 regulatory module identified here may influence many of the previously observed phenotypes of Cvi-0. Recently, two independent studies used X. laevis oocytes as a heterologous expression system to reconstitute bicarbonate-induced activation of the SLAC1 anion channel [19,20]. Tian et al. [19] reported that a multidrug and toxic compound extrusion (MATE)-type transporter RHC1 functions as a bicarbonate-sensing component that inactivates HT1 and promotes SLAC1 activation by OST1. More recently, it was demonstrated that expression of RHC1 alone was sufficient to activate ion currents in oocytes; these currents were independent of bicarbonate, calling into question the role of RHC1 as a bicarbonate sensor [20]. Furthermore, it was shown that SLAC1 activation can be reconstituted by extracellular bicarbonate in the presence of aquaporin PIP2;1, carbonic anhydrase CA4, and the protein kinases OST1, CPK6, and CPK23 [20]. However, in the guard cell, any proposed CO2 signaling pathway should include HT1, since plants with mutations in HT1 completely lack CO2-induced stomatal responses [12,18,28]. We showed that bicarbonate-induced S-type anion currents were strongly impaired in guard cell protoplasts, which lacked functional MPK12 (Fig 3E). Thus, MPK12, MPK4, and possibly other MPKs that are expressed in guard cells play a role in controlling the activity of HT1, and future research should identify the signaling pathway upstream of MPK12 (Fig 7). Dissection of different domains in SLAC1 revealed that the CO2 signal may involve the transmembrane region of SLAC1, whereas ABA activation of SLAC1 requires an intact N- and C-terminus [39]. Hence, ABA and CO2 regulation of SLAC1 could use different signaling pathways, and this may explain the lack of strong ABA phenotypes in plants with mutations in MPK12. We propose that stomatal movements triggered by changes in CO2 concentration are regulated by MPK12- and MPK4-induced inhibition of HT1 activity (Fig 7). The MPK12 glycine 53 is conserved in all Arabidopsis MPKs [10] and is located on the protein surface in the glycine-rich loop that coordinates the gamma-phosphate of ATP (Fig 6B and 6C). Thus, this glycine may also be important for the function of other Arabidopsis MPKs. Further studies into the mechanisms controlling activation of MPKs in guard cells will help to identify molecular switches that function in plant acclimation to environmental stress and modulate the overall plant water use efficiency. Such information may allow the designing of molecular targets that can be used for breeding crops with improved water management. Materials and Methods Plant Material and Growth Conditions Col-0, Col-gl, Cvi-0, gdsl3-1 (GABI-492D11; CS447183), cas-1 (SALK_070416), cas-2 (GABI-665G12), and cas-3 (SAIL_1157_C10) were from the European Arabidopsis Stock Centre (www.arabidopsis.info). Seeds of ht1-2 were a gift from Dr. Koh Iba. Col-0 × Cvi-0 RILs were obtained from INRA Versailles. The abi1-1 allele used was in Col-0 accession. Double mutants and other crosses were made through standard techniques and genotyped with PCR-based markers (S1 Table). For ozone screening, seeds were sown at high density on a 1:1 v/v mixture of vermiculite and peat (type B2, Kekkilä, Finland), and kept for 2 d at 4°C for stratification. The plants were grown in controlled growth chambers (Bio 1300, Weiss Umwelttechnik, Germany) under a 12 h photoperiod, with a 23°C/19°C day/night temperature and a 70%/90% relative humidity or in growth rooms with equivalent growth conditions. The average photosynthetic photon flux density during the light period was 200 μmol m-2 s-1. When seedlings were 1 wk old, they were transplanted into 8 × 8 cm pots at a density of five plants per pot. Three-week-old plants were exposed to ozone in growth chambers under the same conditions as they were grown until the experiments. Ozone exposure was acute (300–350 ppb for 6 h) and started 2 h after light was switched on. Ozone damage was visualized with trypan blue stain or quantified as electrolyte leakage. Mapping of Cvi-0 Ozone Sensitivity QTLs NILs were created by crossing Col-0 with Cvi-0 and selecting the most ozone-sensitive plant in F2 and backcrossing to Col-0 for eight generations (generating Col-S) or selecting the most tolerant plant and backcrossing to Cvi-0 for six generations (generating Cvi-T). The genomes of Cvi-0 and Cvi-T were sequenced at BGI Tech Solutions (Hong Kong) with Illumina technology, and the genomes of Col-S and Cvi-T were sequenced at the DNA Sequencing and Genomics lab of the University of Helsinki with SOLiD technology. Genome sequence data is available from the NCBI BioProject database with the accession number PRJNA345097. The 90-bp-long Illumina paired end sequencing library reads were mapped onto the Col-0 reference genome (TAIR10) with using the Bowtie2 aligner (version 2.0.0-beta7; [40]) in “end-to-end” alignment mode, yielding an average genomic sequence coverage of 45-fold. Variation calling and haplotype phasing was performed with the help of samtools (tools for alignments in the SAM format, Version: 0.1.18; [41]). Based on the aligned sequences, various PCR-based markers (S1 Table) were designed to genotype Cvi-0 versus Col-0 in the NILs and informative RILs from the INRA Versailles Col-0 × Cvi-0 RIL population. The markers were also used to genotype ozone-sensitive individuals from segregating F2 populations. Mapping of cis Mutation Mapping population was created by crossing cis (Col-0) and C24 as an Arabidopsis accession with low stomatal conductance. High water loss from excised leaves and decreased responses to high CO2 were used as a selective trait. Rough mapping with 22 markers using 59 F2 samples showed linkage to the bottom of chromosome 2, at the marker UPSC_2–18415 at 18.4 Mbp. Pooled genomic DNA from 66 selected F3 plants was used for sequencing. Whole genome sequencing was conducted with Illumina HiSeq 2000, and the reads were mapped against Col-0 genome (release TAIR10) by BGI Tech Solutions (Hong Kong). Genome sequence data is available from the NCBI BioProject database with the accession number PRJNA345097 and PRJNA343292. For mapping the genomic area of the mutation, the Next Generation Mapping tool was used [42], which positioned the mutation on chromosome 2 between 18,703,644 –19,136,098 bp. The deletion mutation in cis was verified by PCR to be 4,770 bp (at the position 18,945,427–18,950,196 bp). Complementation Lines MPK12 and its promoter were amplified from Col-0 or Cvi-0 genomic DNA using Phusion (Thermo Fisher Scientific) and Gateway (Invitrogen) cloned into entry vector pDONR-Zeo. Subsequently, the genes were cloned into pGWB13 and pMCD100. Plants were transformed with floral dipping [43]. Southern Blotting Analyses Total DNAs from different genotyping plants were extracted by CTAB method, and 12 micrograms of total DNA was digested by HindIII or EcoRI. The DNAs were running on the gel and transformed onto Nylon membrane. Hybridization was performed with digoxigenin-labeled specific genomic DNA amplified by primers F3 and R4 for 12 h. The membrane was washed several times by washing buffer and Maleic acid buffer. The membrane was blocked by blocking solution for 1 h at room temperature and washed and incubated with anti-DIG-AP for 30 min. Detection was performed using substrate DIG CSPD. Plant Growth and Experimental Settings for Gas Exchange Measurements Seeds were planted on a soil mixture consisting of 2:1 (v:v) peat:vermiculite and grown through a hole in a glass plate covering the pot as described previously [44]. Plants were grown in growth chambers (MCA1600, Snijders Scientific, Drogenbos, Belgium) at 12 h/12 h day/night cycle, 23°C/20°C temperature, 100 μmol m-2 s-1 light, and 70% relative humidity (RH). For gas exchange experiments, 24- to 30-d-old plants were used. Stomatal conductance of intact plants was measured using a rapid-response gas exchange measurement device consisting of eight through-flow whole-rosette cuvettes [44]. The unit of stomatal conductance mmol m-2 s-1 reflects the amount of H2O moles that exits the plant through stomata per one m2 of leaf area per second. Prior to the experiment, plants were acclimated in the measurement cuvettes in ambient CO2 concentration (~400 ppm), 100 μmol m-2 s-1 light (if not stated otherwise), and ambient humidity (RH 65%–80%) for at least 1 h or until stomatal conductance was stable. Thereafter, the following stimuli were applied: decrease or increase in CO2 concentration, darkness, reduced air humidity, and ozone. CO2 concentration was decreased to 100 ppm by filtering air through a column of granular potassium hydroxide. In CO2 enrichment experiments, CO2 was increased by adding it to the air inlet to achieve a concentration of 800 ppm. Darkness was applied by covering the measurement cuvettes. In blue light experiments, dark-adapted plants were exposed to blue light (50 μmol m-2 s-1) from an LED light source (B42180, Seoul Semiconductor, Ansan, South Korea). The decreased or increased CO2 concentration, darkness, and blue light were applied for 58 min. In the long-term elevated CO2 experiment (Fig 1D and S1E Fig), CO2 concentration was increased from 400 ppm to 800 ppm for 2.5 h. To calculate stomatal half-response time, the whole 2.5-h stomatal response to elevated CO2 was scaled to a range from 0% to 100%, and the time when 50% of stomatal closure had occurred was calculated. Humidity was decreased by a thermostat system to 30%–40% RH, and stomatal conductance was monitored for another 56 min. In ozone experiments, the plants were exposed to 350–450 ppb of ozone for 3 min and stomatal conductance was measured for 60 min after the start of the exposure. In ABA-induced stomatal closure experiments, 5 μM ABA solution was applied by spraying as described in [45]. At time point 0, plants were removed from cuvettes and sprayed with either 5 μM ABA solution (5 μM ABA, 0.012% Silwet L-77 [PhytoTechnology Laboratories], and 0.05% ethanol) or control solution (0.012% Silwet L-77 and 0.05% ethanol). Thereafter, plants were returned to the cuvettes and stomatal conductance was monitored for 56 min. In ABA-induced inhibition of stomatal opening experiments, plants were acclimated in measurement cuvettes in darkness. At time point 0, plants were removed from cuvettes and sprayed with 2.5 μM ABA solution (2.5 μM ABA, 0.012% Silwet L-77 [PhytoTechnology Laboratories], and 0.05% ethanol) or control solution (0.012% Silwet L-77 and 0.05% ethanol). Thereafter, plants were returned to the cuvettes, dark covers were removed, and stomatal conductance was monitored in light for 56 min. Prior to the measurement of the diurnal pattern of stomatal conductance, plants were preincubated in the measurement cuvette for at least 12 h in respective light and humidity conditions. Plants were measured in 16-min intervals. WUE was calculated based on the data of diurnal experiments as an average of daytime light period (from 9:00 to 17:00). CO2-induced stomatal conductance in S2 Fig was measured as following. Five-week-old healthy plants growing in a growth chamber with 70% humidity and a 16 h light/8 h dark condition were used for stomatal conductance analyses at different CO2 concentrations by a LiCOR-6400XT, as previously described [13]. Relative stomatal conductance values were normalized relative to the last data point preceding the [CO2] transitions (400 to 800 or 1,000 ppm). Stomatal Aperture The MPK12 deletion mutant mpk12-4 and wild-type plants were grown in a growth chamber at 70% humidity, 75 μmolm-2 s-1 light intensity, 21°C, and 16 h light/8 h dark regime. Leaf epidermal layers from 2-wk-old plants of both genotypes were preincubated in an opening buffer (10 mM MES, 10 mM KCl, and 50 mM CaCl2 at pH 6.15) for 2 h, and stomata were individually imaged and tracked for measurement before treatment. After that, the leaf epidermal layers were incubated with buffers containing 10 μM ABA for 30 min and the individually tracked stomata were imaged. Stomatal apertures were measured by ImageJ software and genotype-blind analyses were used. The data presented are means and SEM n = 3 experiments, with 30 stomata per experiment and condition. Stomatal Index and Density Plants at the age of 28–30 d were used for stomatal index and density measurements. Rosette leaves of equal size were excised, and the abaxial side was covered with the dental resin (Xantopren M mucosa, Heraeus Kulzer, Germany). Transparent nail varnish was applied onto the dried impressions after the removal of the leaves. The hardened nail varnish imprints were attached onto a microscope glass slide with a transparent tape and imaged under a Zeiss SteREO Discovery.V20 stereomicroscope. For quantification, an image with the coverage of 0.12 mm2 was taken from the middle of the leaf, next to the middle vein. In total, 81–84 plants per line from two independent biological repeats were analyzed—one leaf from each plant, one image from each leaf. Stomatal index was calculated with the following formula: SI = Stomatal density / (Density of other epidermal cells + Stomatal density). Stomatal Complex Length For the stomatal complex length measurements, plants at the age of 28–35 d were used. Whole leaves were preincubated for 4 h abaxial side down in the buffer (10 mM MES, 5 mM KCl, 50 μM CaCl2, pH 6.15 [with TRIS]) in the light. Four to six plants per genotype and one leaf per plant were analyzed, and altogether 84–126 stomatal complexes per genotype were measured. Y2H Interaction Tests Interactions between MPK12 and selected protein kinases and phosphatases were tested in pairwise split-ubiquitin Y2H assays using the DUALhunter and DUALmembrane 3 kits (Dualsystems Biotech). For bait construction, the coding sequences of MPK12 were PCR-amplified from total cDNAs from Col-0 and Cvi-0. Other MPK12 variants with point mutations (K70R, Y122C, and D196G+E200A) were created by two-step overlap PCR using the Col-0 MPK12 as a template. HT1 was also PCR-amplified from Col-0 cDNA. All MPK12s and HT1 were digested with SfiI and cloned to the corresponding site in pDHB1, which contained the Cub-LexA-VP16 fusion. For prey constructs, coding sequences of each selected gene were amplified from total Col-0 cDNAs, digested with SfiI, and cloned into either pPR3-N (HT1, OST1, BLUS1, IBR5, MKP2, MPK12, MPK12G53R, MPK11) or pPR3-STE (SnRK2.2, SnRK3.11, ABI1, ABI2, HAB1, HAB2), which contained a mutated NubG. All primers used are listed in Table S1. The pAI-Alg5 with a native NubI was used as a positive prey control, whereas the pDL2-Alg5 containing NubG served as a negative control. For pairwise Y2H assays, the yeast strain NMY51 was cotransformed with bait and prey plasmids and grown on SD-Leu-Trp plates to select for presence of both plasmids. At least ten colonies from each transformation were pooled and resuspended in water to an OD600 of 0.5, from which 100, 1,000, and 10,000x serial dilutions were prepared and spotted on SD-Leu-Trp and SD-Leu-Trp-His-Ade plates. SD-Leu-Trp plates were incubated at 30°C for 2 d, photographed, and used for β-galatosidase overlay assays. SD-Leu-Trp-His-Ade plates were incubated for 2–4 d and photographed. The quantitative β-galactosidase assay was performed with three pools of ten independent colonies from each pairwise combination using the Yeast β-galactosidase assay kit (Thermo Fisher Scientific) by the nonstop quantitative method. Ratiometric BiFC Assay Binary constructs containing split YFPs were designed and generated for cloning genes of interest by the ligation independent cloning (LIC) method as described in [18]. Each gene of interest was amplified by two consecutive PCR reactions: first with gene-specific primers and later with a pair of universal primers designed specifically for the LIC method. All primers used are listed in S1 Table. To prepare vectors for LIC, plasmids of 35S:YFPn and 35S:YFPc were linearized by PmlI digestion, followed by T4 DNA polymerase treatment with dGTP to create 15–16 nucleotide 5ʹ-overhangs. For insert preparation, the final PCR products of target genes were incubated with T4 DNA polymerase in the presence of dCTP to create the complementary overhangs with the vectors. Both vector and insert were mixed at room temperature and proceeded with Escherichia coli transformation after 5 min. The final constructs were sequence verified and transformed to Agrobacterium tumefaciens GV3101 for agro-infiltration experiments. For the ratiometric BiFC assays, four different agrobacterial clones—each harboring a YFPn fusion, a YFPc fusion, the SLAC1-CFP internal control, or the gene silencing suppressor P19—were co-infiltrated to the leaves of N. benthamiana at an OD600 of 0.02 for each clone in the infiltration buffer (10mM MES, 10mM MgCl2, 200 μM acetosyringone). Images were acquired at 3 dpi with a Zeiss LSM710 confocal microscope using a 63x objective (for high magnification images) or a 20x objective (for fluorescence quantification). The YFP signals were excited by a 514 nm laser, and emission between 518–564 nm was collected. The CFP signals were excited by a 405 nm laser, and emission at 460–530 was collected. Z-stack images of approximately 15 μm thickness were collected, and all images were acquired at the 16-bit depth for a higher dynamic range. The fluorescence intensity was measured by the ImageJ software. The leaf samples used for imaging were collected and used for protein extraction followed by western blot analysis. Western Blot Analysis The leaf samples (30–40 mg) were ground under liquid nitrogen and boiled for 10 min in 100 μL of 6X Laemmli buffer. 12 μL of each sample were separated on 10% SDS polyacrylamide gel. After SDS-PAGE, proteins were transferred onto nitrocellulose membrane. Immunodetection of HA-tagged proteins was performed with a monoclonal anti-HA antibody. Split Luciferase Complementation Assay The MPK12 cDNA was cloned into a vector containing the N-terminal half of luciferase (nLUC) and HT1 was cloned into the cLUC. The constructs in the A. tumefaciens strain GV3101 were co-infiltrated into N. benthamiana leaves with P19 at an OD600 of 0.8. The infiltrated leaves after 3 d of infiltration were harvested for bioluminescence detection. Images were captured with a CCD camera. Measurement of S-type Anion Currents Arabidopsis guard cell protoplasts were isolated as described previously [46]. Guard cell protoplasts were washed twice with a washing solution containing 1 mM MgCl2, 1 mM CaCl2, 5 mM MES, and 500 mM D-sorbitol (pH 5.6 with Tris). During patch clamp recordings of S-type anion currents, the membrane voltage started at +35 to –145 mV for 7 s with –30 mV decrements, and the holding potential was +30 mV. The bath solutions contained 30 mM CsCl, 2 mM MgCl2, 10 mM MES (Tris, pH 5.6), and 1 mM CaCl2, with an osmolality of 485 mmol/kg. The pipette solutions contained 5.86 mM CaCl2, 6.7 mM EGTA, 2 mM MgCl2, 10 mM Hepes-Tris (pH 7.1), and 150 mM CsCl, with an osmolality of 500 mmol/kg. The free calcium concentration was 2 μM. The final osmolalities in both bath and pipette solutions were adjusted with D-sorbitol. Mg-ATP (5 mM) was added to the pipette solution before use. 13.5 mM CsHCO3 (11.5 mM free [HCO3-] and 2 mM free [CO2]) was freshly dissolved in the pipette solution before patch clamp experiments. The concentrations of free bicarbonate and free CO2 were calculated using the Henderson–Hasselbalch equation (pH = pK1 + log [HCO3-] / [CO2]). pK1 = 6.352 was used for the calculation. [HCO3-] represents the free bicarbonate concentration and [CO2] represents the free CO2 concentration. Protein Expression and Purification For in vitro kinase assays, the respective sequences of HT1, HT1 K113M, MPK11, MPK12, MPK12 G53R, MPK12 K70R, and MPK12 Y122C were cloned into a pET28a vector (Novagen, Merck Millipore) using primers listed in S1 Table. Point mutations corresponding to K113M in HT1, K70R in MPK12, and Y122C in MPK12 were created with two-step PCR using primers listed in S1 Table. MPK4 was cloned as previously described [18]. 6xHis-HT1WT, 6xHis-HT1 K113M, 6xHis-MPK12, 6xHis-MPK12 G53R, 6xHis-MPK12 K70R, 6xHis-MPK12 Y122C, 6xHis-MPK11, 6xHis-MPK4 WT, and 6xHis-MPK4 G55R were expressed in E. coli BL21(DE3) cells. A 2 mL aliquot of an overnight culture was transferred to a fresh 1 L 2xYT medium and grown at 37°C to an absorbance of ~0.6 at OD600. The cultures were chilled to 16°C and recombinant protein expression was induced by 0.3 mM isopropyl b-D-thiogalactopyranoside for 16 h. The cells were harvested by centrifugation (5,000 rpm, 10 min, 4°C) and stored at –80°C until use. All purification procedures were carried out at 4°C. The cells were resuspended in 30 mL of lysis buffer (50 mM Tris-HCl [pH 7.4], 300 mM NaCl, 5% [v/v] glycerol, 1% [v/v] Triton X-100, 1 mM PMSF, 1 μg/ml aprotinin, 1 μg/ml pepstatin A, 1 μg/ml leupeptin) and lysed using an Emulsiflex C3 Homogenizer. Cell debris was removed by centrifugation at 20,000 rpm for 30 min. The protein-containing supernatant was mixed for 1 h at 4°C with 0.20 mL of Chelating Sepharose Fast Flow resin (GE Healthcare), charged with 200 mM NiSO4 and pre-equilibrated in the lysis buffer. The protein–resin complex was packed into a column, and the beads were washed with 5x10 column volumes (CV) of a wash buffer I (50 mM Tris-HCl [pH 7.4], 600 mM NaCl, 5% [v/v] glycerol, 1% [v/v] Triton X-100), 5x10 CV of a wash buffer II (50 mM Tris-HCl [pH 7.4], 300 mM NaCl, 5% [v/v] glycerol, 0.1% [v/v] NP-40), and 2x10 CV of a wash buffer III (50 mM Tris-HCl [pH 7.4], 150 mM NaCl, 5% [v/v] glycerol, 0.1% [v/v] NP-40). The protein was eluted by incubating the beads for 5 min at room temperature with an imidazole-containing elution buffer (50 mM Tris-HCl, 150 mM NaCl, 5% [v/v] glycerol, 0.1% [v/v] NP-40, 300 mM imidazole). MPK12 proteins were concentrated and imidazole was removed by Millipore Amicon Ultra-0.5 Centrifugal Filter Concentrators (NMWL 3000). Glycerol was added to a final concentration of 20% (v/v), and 20 μL aliquots of the eluted protein were snap-frozen in liquid nitrogen and stored at –80°C. In Vitro Kinase Assays Protein concentrations were estimated on 10% SDS-polyacrylamide gel using BSA as a standard. HT1 kinase activity assay was performed by incubating a constant amount of purified recombinant HT1 and 0–30 μM MPK12, 0–20 μM MPK4, or 0–10 μM MPK11 in a reaction buffer (50 mM Tris-HCl [pH 7.4], 150 mM NaCl, 20 mM MgCl2, 60 mM imidazole, 1 mM DTT, 0.2 mg/ml insulin) at room temperature for 10 min. Then, casein (1 mg/ml), 500 μM ATP, and 100 μCi/ml 32P-γ-ATP were added and reaction aliquots were taken at the 30 min time point. Reactions were stopped by the addition of SDS loading buffer. Proteins were separated on a 10% SDS-polyacrylamide gel and visualized by Coomassie brilliant blue R-250 (Sigma) staining. HT1 activity was determined by autoradiography and quantified by ImageQuant TL Software. Model of MPK12 and MPK4 Sequence searches and alignments were conducted with SWISS-MODEL [47]. The crystal structure with the best sequence identity and resolution was selected for building homology models. Arabidopsis MPK12 and MPK4 have sequence identity to the 3 Ångstrom resolution Arabidopsis MPK6 structure (5CI6; [33]) of 64.61% and 70.67%, respectively. This structure was then used to construct models for the wild-type and mutant structures. The RMSD from aligning the structures for MPK12 and MPK12 G53R was 0.324 Ångstroms (i.e., a close structural similarity). Structures were checked for clashes and with quality controls and were then superposed. Statistical Analysis Statistical analyses were performed with Statistica, version 7.1 (StatSoft Inc., Tulsa, Oklahoma, United States). All effects were considered significant at p < 0.05. Plant Material and Growth Conditions Col-0, Col-gl, Cvi-0, gdsl3-1 (GABI-492D11; CS447183), cas-1 (SALK_070416), cas-2 (GABI-665G12), and cas-3 (SAIL_1157_C10) were from the European Arabidopsis Stock Centre (www.arabidopsis.info). Seeds of ht1-2 were a gift from Dr. Koh Iba. Col-0 × Cvi-0 RILs were obtained from INRA Versailles. The abi1-1 allele used was in Col-0 accession. Double mutants and other crosses were made through standard techniques and genotyped with PCR-based markers (S1 Table). For ozone screening, seeds were sown at high density on a 1:1 v/v mixture of vermiculite and peat (type B2, Kekkilä, Finland), and kept for 2 d at 4°C for stratification. The plants were grown in controlled growth chambers (Bio 1300, Weiss Umwelttechnik, Germany) under a 12 h photoperiod, with a 23°C/19°C day/night temperature and a 70%/90% relative humidity or in growth rooms with equivalent growth conditions. The average photosynthetic photon flux density during the light period was 200 μmol m-2 s-1. When seedlings were 1 wk old, they were transplanted into 8 × 8 cm pots at a density of five plants per pot. Three-week-old plants were exposed to ozone in growth chambers under the same conditions as they were grown until the experiments. Ozone exposure was acute (300–350 ppb for 6 h) and started 2 h after light was switched on. Ozone damage was visualized with trypan blue stain or quantified as electrolyte leakage. Mapping of Cvi-0 Ozone Sensitivity QTLs NILs were created by crossing Col-0 with Cvi-0 and selecting the most ozone-sensitive plant in F2 and backcrossing to Col-0 for eight generations (generating Col-S) or selecting the most tolerant plant and backcrossing to Cvi-0 for six generations (generating Cvi-T). The genomes of Cvi-0 and Cvi-T were sequenced at BGI Tech Solutions (Hong Kong) with Illumina technology, and the genomes of Col-S and Cvi-T were sequenced at the DNA Sequencing and Genomics lab of the University of Helsinki with SOLiD technology. Genome sequence data is available from the NCBI BioProject database with the accession number PRJNA345097. The 90-bp-long Illumina paired end sequencing library reads were mapped onto the Col-0 reference genome (TAIR10) with using the Bowtie2 aligner (version 2.0.0-beta7; [40]) in “end-to-end” alignment mode, yielding an average genomic sequence coverage of 45-fold. Variation calling and haplotype phasing was performed with the help of samtools (tools for alignments in the SAM format, Version: 0.1.18; [41]). Based on the aligned sequences, various PCR-based markers (S1 Table) were designed to genotype Cvi-0 versus Col-0 in the NILs and informative RILs from the INRA Versailles Col-0 × Cvi-0 RIL population. The markers were also used to genotype ozone-sensitive individuals from segregating F2 populations. Mapping of cis Mutation Mapping population was created by crossing cis (Col-0) and C24 as an Arabidopsis accession with low stomatal conductance. High water loss from excised leaves and decreased responses to high CO2 were used as a selective trait. Rough mapping with 22 markers using 59 F2 samples showed linkage to the bottom of chromosome 2, at the marker UPSC_2–18415 at 18.4 Mbp. Pooled genomic DNA from 66 selected F3 plants was used for sequencing. Whole genome sequencing was conducted with Illumina HiSeq 2000, and the reads were mapped against Col-0 genome (release TAIR10) by BGI Tech Solutions (Hong Kong). Genome sequence data is available from the NCBI BioProject database with the accession number PRJNA345097 and PRJNA343292. For mapping the genomic area of the mutation, the Next Generation Mapping tool was used [42], which positioned the mutation on chromosome 2 between 18,703,644 –19,136,098 bp. The deletion mutation in cis was verified by PCR to be 4,770 bp (at the position 18,945,427–18,950,196 bp). Complementation Lines MPK12 and its promoter were amplified from Col-0 or Cvi-0 genomic DNA using Phusion (Thermo Fisher Scientific) and Gateway (Invitrogen) cloned into entry vector pDONR-Zeo. Subsequently, the genes were cloned into pGWB13 and pMCD100. Plants were transformed with floral dipping [43]. Southern Blotting Analyses Total DNAs from different genotyping plants were extracted by CTAB method, and 12 micrograms of total DNA was digested by HindIII or EcoRI. The DNAs were running on the gel and transformed onto Nylon membrane. Hybridization was performed with digoxigenin-labeled specific genomic DNA amplified by primers F3 and R4 for 12 h. The membrane was washed several times by washing buffer and Maleic acid buffer. The membrane was blocked by blocking solution for 1 h at room temperature and washed and incubated with anti-DIG-AP for 30 min. Detection was performed using substrate DIG CSPD. Plant Growth and Experimental Settings for Gas Exchange Measurements Seeds were planted on a soil mixture consisting of 2:1 (v:v) peat:vermiculite and grown through a hole in a glass plate covering the pot as described previously [44]. Plants were grown in growth chambers (MCA1600, Snijders Scientific, Drogenbos, Belgium) at 12 h/12 h day/night cycle, 23°C/20°C temperature, 100 μmol m-2 s-1 light, and 70% relative humidity (RH). For gas exchange experiments, 24- to 30-d-old plants were used. Stomatal conductance of intact plants was measured using a rapid-response gas exchange measurement device consisting of eight through-flow whole-rosette cuvettes [44]. The unit of stomatal conductance mmol m-2 s-1 reflects the amount of H2O moles that exits the plant through stomata per one m2 of leaf area per second. Prior to the experiment, plants were acclimated in the measurement cuvettes in ambient CO2 concentration (~400 ppm), 100 μmol m-2 s-1 light (if not stated otherwise), and ambient humidity (RH 65%–80%) for at least 1 h or until stomatal conductance was stable. Thereafter, the following stimuli were applied: decrease or increase in CO2 concentration, darkness, reduced air humidity, and ozone. CO2 concentration was decreased to 100 ppm by filtering air through a column of granular potassium hydroxide. In CO2 enrichment experiments, CO2 was increased by adding it to the air inlet to achieve a concentration of 800 ppm. Darkness was applied by covering the measurement cuvettes. In blue light experiments, dark-adapted plants were exposed to blue light (50 μmol m-2 s-1) from an LED light source (B42180, Seoul Semiconductor, Ansan, South Korea). The decreased or increased CO2 concentration, darkness, and blue light were applied for 58 min. In the long-term elevated CO2 experiment (Fig 1D and S1E Fig), CO2 concentration was increased from 400 ppm to 800 ppm for 2.5 h. To calculate stomatal half-response time, the whole 2.5-h stomatal response to elevated CO2 was scaled to a range from 0% to 100%, and the time when 50% of stomatal closure had occurred was calculated. Humidity was decreased by a thermostat system to 30%–40% RH, and stomatal conductance was monitored for another 56 min. In ozone experiments, the plants were exposed to 350–450 ppb of ozone for 3 min and stomatal conductance was measured for 60 min after the start of the exposure. In ABA-induced stomatal closure experiments, 5 μM ABA solution was applied by spraying as described in [45]. At time point 0, plants were removed from cuvettes and sprayed with either 5 μM ABA solution (5 μM ABA, 0.012% Silwet L-77 [PhytoTechnology Laboratories], and 0.05% ethanol) or control solution (0.012% Silwet L-77 and 0.05% ethanol). Thereafter, plants were returned to the cuvettes and stomatal conductance was monitored for 56 min. In ABA-induced inhibition of stomatal opening experiments, plants were acclimated in measurement cuvettes in darkness. At time point 0, plants were removed from cuvettes and sprayed with 2.5 μM ABA solution (2.5 μM ABA, 0.012% Silwet L-77 [PhytoTechnology Laboratories], and 0.05% ethanol) or control solution (0.012% Silwet L-77 and 0.05% ethanol). Thereafter, plants were returned to the cuvettes, dark covers were removed, and stomatal conductance was monitored in light for 56 min. Prior to the measurement of the diurnal pattern of stomatal conductance, plants were preincubated in the measurement cuvette for at least 12 h in respective light and humidity conditions. Plants were measured in 16-min intervals. WUE was calculated based on the data of diurnal experiments as an average of daytime light period (from 9:00 to 17:00). CO2-induced stomatal conductance in S2 Fig was measured as following. Five-week-old healthy plants growing in a growth chamber with 70% humidity and a 16 h light/8 h dark condition were used for stomatal conductance analyses at different CO2 concentrations by a LiCOR-6400XT, as previously described [13]. Relative stomatal conductance values were normalized relative to the last data point preceding the [CO2] transitions (400 to 800 or 1,000 ppm). Stomatal Aperture The MPK12 deletion mutant mpk12-4 and wild-type plants were grown in a growth chamber at 70% humidity, 75 μmolm-2 s-1 light intensity, 21°C, and 16 h light/8 h dark regime. Leaf epidermal layers from 2-wk-old plants of both genotypes were preincubated in an opening buffer (10 mM MES, 10 mM KCl, and 50 mM CaCl2 at pH 6.15) for 2 h, and stomata were individually imaged and tracked for measurement before treatment. After that, the leaf epidermal layers were incubated with buffers containing 10 μM ABA for 30 min and the individually tracked stomata were imaged. Stomatal apertures were measured by ImageJ software and genotype-blind analyses were used. The data presented are means and SEM n = 3 experiments, with 30 stomata per experiment and condition. Stomatal Index and Density Plants at the age of 28–30 d were used for stomatal index and density measurements. Rosette leaves of equal size were excised, and the abaxial side was covered with the dental resin (Xantopren M mucosa, Heraeus Kulzer, Germany). Transparent nail varnish was applied onto the dried impressions after the removal of the leaves. The hardened nail varnish imprints were attached onto a microscope glass slide with a transparent tape and imaged under a Zeiss SteREO Discovery.V20 stereomicroscope. For quantification, an image with the coverage of 0.12 mm2 was taken from the middle of the leaf, next to the middle vein. In total, 81–84 plants per line from two independent biological repeats were analyzed—one leaf from each plant, one image from each leaf. Stomatal index was calculated with the following formula: SI = Stomatal density / (Density of other epidermal cells + Stomatal density). Stomatal Complex Length For the stomatal complex length measurements, plants at the age of 28–35 d were used. Whole leaves were preincubated for 4 h abaxial side down in the buffer (10 mM MES, 5 mM KCl, 50 μM CaCl2, pH 6.15 [with TRIS]) in the light. Four to six plants per genotype and one leaf per plant were analyzed, and altogether 84–126 stomatal complexes per genotype were measured. Y2H Interaction Tests Interactions between MPK12 and selected protein kinases and phosphatases were tested in pairwise split-ubiquitin Y2H assays using the DUALhunter and DUALmembrane 3 kits (Dualsystems Biotech). For bait construction, the coding sequences of MPK12 were PCR-amplified from total cDNAs from Col-0 and Cvi-0. Other MPK12 variants with point mutations (K70R, Y122C, and D196G+E200A) were created by two-step overlap PCR using the Col-0 MPK12 as a template. HT1 was also PCR-amplified from Col-0 cDNA. All MPK12s and HT1 were digested with SfiI and cloned to the corresponding site in pDHB1, which contained the Cub-LexA-VP16 fusion. For prey constructs, coding sequences of each selected gene were amplified from total Col-0 cDNAs, digested with SfiI, and cloned into either pPR3-N (HT1, OST1, BLUS1, IBR5, MKP2, MPK12, MPK12G53R, MPK11) or pPR3-STE (SnRK2.2, SnRK3.11, ABI1, ABI2, HAB1, HAB2), which contained a mutated NubG. All primers used are listed in Table S1. The pAI-Alg5 with a native NubI was used as a positive prey control, whereas the pDL2-Alg5 containing NubG served as a negative control. For pairwise Y2H assays, the yeast strain NMY51 was cotransformed with bait and prey plasmids and grown on SD-Leu-Trp plates to select for presence of both plasmids. At least ten colonies from each transformation were pooled and resuspended in water to an OD600 of 0.5, from which 100, 1,000, and 10,000x serial dilutions were prepared and spotted on SD-Leu-Trp and SD-Leu-Trp-His-Ade plates. SD-Leu-Trp plates were incubated at 30°C for 2 d, photographed, and used for β-galatosidase overlay assays. SD-Leu-Trp-His-Ade plates were incubated for 2–4 d and photographed. The quantitative β-galactosidase assay was performed with three pools of ten independent colonies from each pairwise combination using the Yeast β-galactosidase assay kit (Thermo Fisher Scientific) by the nonstop quantitative method. Ratiometric BiFC Assay Binary constructs containing split YFPs were designed and generated for cloning genes of interest by the ligation independent cloning (LIC) method as described in [18]. Each gene of interest was amplified by two consecutive PCR reactions: first with gene-specific primers and later with a pair of universal primers designed specifically for the LIC method. All primers used are listed in S1 Table. To prepare vectors for LIC, plasmids of 35S:YFPn and 35S:YFPc were linearized by PmlI digestion, followed by T4 DNA polymerase treatment with dGTP to create 15–16 nucleotide 5ʹ-overhangs. For insert preparation, the final PCR products of target genes were incubated with T4 DNA polymerase in the presence of dCTP to create the complementary overhangs with the vectors. Both vector and insert were mixed at room temperature and proceeded with Escherichia coli transformation after 5 min. The final constructs were sequence verified and transformed to Agrobacterium tumefaciens GV3101 for agro-infiltration experiments. For the ratiometric BiFC assays, four different agrobacterial clones—each harboring a YFPn fusion, a YFPc fusion, the SLAC1-CFP internal control, or the gene silencing suppressor P19—were co-infiltrated to the leaves of N. benthamiana at an OD600 of 0.02 for each clone in the infiltration buffer (10mM MES, 10mM MgCl2, 200 μM acetosyringone). Images were acquired at 3 dpi with a Zeiss LSM710 confocal microscope using a 63x objective (for high magnification images) or a 20x objective (for fluorescence quantification). The YFP signals were excited by a 514 nm laser, and emission between 518–564 nm was collected. The CFP signals were excited by a 405 nm laser, and emission at 460–530 was collected. Z-stack images of approximately 15 μm thickness were collected, and all images were acquired at the 16-bit depth for a higher dynamic range. The fluorescence intensity was measured by the ImageJ software. The leaf samples used for imaging were collected and used for protein extraction followed by western blot analysis. Western Blot Analysis The leaf samples (30–40 mg) were ground under liquid nitrogen and boiled for 10 min in 100 μL of 6X Laemmli buffer. 12 μL of each sample were separated on 10% SDS polyacrylamide gel. After SDS-PAGE, proteins were transferred onto nitrocellulose membrane. Immunodetection of HA-tagged proteins was performed with a monoclonal anti-HA antibody. Split Luciferase Complementation Assay The MPK12 cDNA was cloned into a vector containing the N-terminal half of luciferase (nLUC) and HT1 was cloned into the cLUC. The constructs in the A. tumefaciens strain GV3101 were co-infiltrated into N. benthamiana leaves with P19 at an OD600 of 0.8. The infiltrated leaves after 3 d of infiltration were harvested for bioluminescence detection. Images were captured with a CCD camera. Measurement of S-type Anion Currents Arabidopsis guard cell protoplasts were isolated as described previously [46]. Guard cell protoplasts were washed twice with a washing solution containing 1 mM MgCl2, 1 mM CaCl2, 5 mM MES, and 500 mM D-sorbitol (pH 5.6 with Tris). During patch clamp recordings of S-type anion currents, the membrane voltage started at +35 to –145 mV for 7 s with –30 mV decrements, and the holding potential was +30 mV. The bath solutions contained 30 mM CsCl, 2 mM MgCl2, 10 mM MES (Tris, pH 5.6), and 1 mM CaCl2, with an osmolality of 485 mmol/kg. The pipette solutions contained 5.86 mM CaCl2, 6.7 mM EGTA, 2 mM MgCl2, 10 mM Hepes-Tris (pH 7.1), and 150 mM CsCl, with an osmolality of 500 mmol/kg. The free calcium concentration was 2 μM. The final osmolalities in both bath and pipette solutions were adjusted with D-sorbitol. Mg-ATP (5 mM) was added to the pipette solution before use. 13.5 mM CsHCO3 (11.5 mM free [HCO3-] and 2 mM free [CO2]) was freshly dissolved in the pipette solution before patch clamp experiments. The concentrations of free bicarbonate and free CO2 were calculated using the Henderson–Hasselbalch equation (pH = pK1 + log [HCO3-] / [CO2]). pK1 = 6.352 was used for the calculation. [HCO3-] represents the free bicarbonate concentration and [CO2] represents the free CO2 concentration. Protein Expression and Purification For in vitro kinase assays, the respective sequences of HT1, HT1 K113M, MPK11, MPK12, MPK12 G53R, MPK12 K70R, and MPK12 Y122C were cloned into a pET28a vector (Novagen, Merck Millipore) using primers listed in S1 Table. Point mutations corresponding to K113M in HT1, K70R in MPK12, and Y122C in MPK12 were created with two-step PCR using primers listed in S1 Table. MPK4 was cloned as previously described [18]. 6xHis-HT1WT, 6xHis-HT1 K113M, 6xHis-MPK12, 6xHis-MPK12 G53R, 6xHis-MPK12 K70R, 6xHis-MPK12 Y122C, 6xHis-MPK11, 6xHis-MPK4 WT, and 6xHis-MPK4 G55R were expressed in E. coli BL21(DE3) cells. A 2 mL aliquot of an overnight culture was transferred to a fresh 1 L 2xYT medium and grown at 37°C to an absorbance of ~0.6 at OD600. The cultures were chilled to 16°C and recombinant protein expression was induced by 0.3 mM isopropyl b-D-thiogalactopyranoside for 16 h. The cells were harvested by centrifugation (5,000 rpm, 10 min, 4°C) and stored at –80°C until use. All purification procedures were carried out at 4°C. The cells were resuspended in 30 mL of lysis buffer (50 mM Tris-HCl [pH 7.4], 300 mM NaCl, 5% [v/v] glycerol, 1% [v/v] Triton X-100, 1 mM PMSF, 1 μg/ml aprotinin, 1 μg/ml pepstatin A, 1 μg/ml leupeptin) and lysed using an Emulsiflex C3 Homogenizer. Cell debris was removed by centrifugation at 20,000 rpm for 30 min. The protein-containing supernatant was mixed for 1 h at 4°C with 0.20 mL of Chelating Sepharose Fast Flow resin (GE Healthcare), charged with 200 mM NiSO4 and pre-equilibrated in the lysis buffer. The protein–resin complex was packed into a column, and the beads were washed with 5x10 column volumes (CV) of a wash buffer I (50 mM Tris-HCl [pH 7.4], 600 mM NaCl, 5% [v/v] glycerol, 1% [v/v] Triton X-100), 5x10 CV of a wash buffer II (50 mM Tris-HCl [pH 7.4], 300 mM NaCl, 5% [v/v] glycerol, 0.1% [v/v] NP-40), and 2x10 CV of a wash buffer III (50 mM Tris-HCl [pH 7.4], 150 mM NaCl, 5% [v/v] glycerol, 0.1% [v/v] NP-40). The protein was eluted by incubating the beads for 5 min at room temperature with an imidazole-containing elution buffer (50 mM Tris-HCl, 150 mM NaCl, 5% [v/v] glycerol, 0.1% [v/v] NP-40, 300 mM imidazole). MPK12 proteins were concentrated and imidazole was removed by Millipore Amicon Ultra-0.5 Centrifugal Filter Concentrators (NMWL 3000). Glycerol was added to a final concentration of 20% (v/v), and 20 μL aliquots of the eluted protein were snap-frozen in liquid nitrogen and stored at –80°C. In Vitro Kinase Assays Protein concentrations were estimated on 10% SDS-polyacrylamide gel using BSA as a standard. HT1 kinase activity assay was performed by incubating a constant amount of purified recombinant HT1 and 0–30 μM MPK12, 0–20 μM MPK4, or 0–10 μM MPK11 in a reaction buffer (50 mM Tris-HCl [pH 7.4], 150 mM NaCl, 20 mM MgCl2, 60 mM imidazole, 1 mM DTT, 0.2 mg/ml insulin) at room temperature for 10 min. Then, casein (1 mg/ml), 500 μM ATP, and 100 μCi/ml 32P-γ-ATP were added and reaction aliquots were taken at the 30 min time point. Reactions were stopped by the addition of SDS loading buffer. Proteins were separated on a 10% SDS-polyacrylamide gel and visualized by Coomassie brilliant blue R-250 (Sigma) staining. HT1 activity was determined by autoradiography and quantified by ImageQuant TL Software. Model of MPK12 and MPK4 Sequence searches and alignments were conducted with SWISS-MODEL [47]. The crystal structure with the best sequence identity and resolution was selected for building homology models. Arabidopsis MPK12 and MPK4 have sequence identity to the 3 Ångstrom resolution Arabidopsis MPK6 structure (5CI6; [33]) of 64.61% and 70.67%, respectively. This structure was then used to construct models for the wild-type and mutant structures. The RMSD from aligning the structures for MPK12 and MPK12 G53R was 0.324 Ångstroms (i.e., a close structural similarity). Structures were checked for clashes and with quality controls and were then superposed. Statistical Analysis Statistical analyses were performed with Statistica, version 7.1 (StatSoft Inc., Tulsa, Oklahoma, United States). All effects were considered significant at p < 0.05. Supporting Information S1 Fig. Identification of Col-S2 and cis mutation. (A) Ion leakage after 6 h of ozone exposure (350 ppb ozone). Experiment was repeated three times (mean ± SD; 1-way ANOVA of ozone treated plants). (B) Scheme of mapping the ozone sensitive trait of Col-S2. (C) CO2-induced changes in stomatal conductance in cas mutants (mean ± SEM; n = 5–6 plants). (D) Mapping scheme of cis mutation obtained from cas-2 T-DNA line. (E) Stomatal response to elevation of the atmospheric CO2 concentration from 400 ppm to 800 ppm at time point 0. Data are given as average stomatal conductance, ± SEM of Col-0 (n = 13), Col-S2 (n = 13) and mpk12-4 (n = 13). The data were pooled from two experimental series. (F) Stomatal conductance of Col-0 plants transformed with MPK12-Cvi in T1 generation (mean ± SEM; 1-way ANOVA, Tukey HSD post hoc test for unequal sample size; n = 4–16 plants). (G) Stomatal conductance of F1 generation of Col-S2 x gl1 (mean ± SEM; 1-way ANOVA, Tukey HSD post hoc test for unequal sample size). Experiment was repeated two times (n = 10–60 plants). The raw data for panels (A), (C), (E-G) can be found in S1 Data file. https://doi.org/10.1371/journal.pbio.2000322.s001 (TIF) S2 Fig. Mutations in MPK12 are causing impaired CO2-responses in both cas-2 and gdsl3-1 mutants. (A) The originally described T-DNA insertions were confirmed in cas-2 and gdsl3-1 (GABI_492D11) plants by genotyping analyses. PCR product for the CAS gene was amplified by primers CASLP and CASRP. PCR product for the cas-2 insert was amplified by primers CASRP and GABILb. PCR product for the GDSL3 gene was amplified by primers GDSL3RP and GDSL3LP. PCR product for the gdsl3-1 insert was amplified by primers GDSL3LP and GABILb. Lane 9: DNA marker. (B) Time-resolved relative stomatal conductance analyses showed that CO2-induced stomatal closing was greatly impaired in CAS mutant allele cas-2, but not in cas-1 allele. Data present are means ± SEM, n = 3 leaves for wild type and n = 4 leaves for CAS alleles. (C) Time-resolved relative stomatal conductance analyses showed that CO2-induced stomatal closure was greatly impaired in gdsl3-1. Data present are means ± SEM, n = 3 leaves for each genotype. (D) Thermal imaging showed that cas-2 and gdsl3-1 have much lower leaf temperature compared to cas-1 and Col-0 plants. (E, F) Southern blotting confirmed that a 4770 bp region containing MPK12 and BYPASS2 between R3 and F21 was deleted in cas-2 and gdsl3-1 mutants. Genomic DNAs extracted from cas-2, gdsl3-1 and Col-0 were digested with HindIII and BamHI. The probe was set in the region F3 and F2 marked in E. (G) mpk12-4 mutant from gdsl3-1×Col-0 backcross F2 offsprings, in which MPK12-BYPASS2 was deleted but contained GDSL3, displayed similar responses to CO2 changes as gdsl3-1 by gas exchange analyses. Data present are means ± SEM, n = 3 leaves for each genotype. (H) Time-resolved relative stomatal conductance analyses showed that expression of MPK12 under the control of UBQ10 promoter in gdsl3-1 complemented the insensitive stomatal CO2 responses. Data present are means ± SEM, n = 3 leaves for each genotype. The raw data for panels (B-C), (G-H) can be found in S1 Data file. https://doi.org/10.1371/journal.pbio.2000322.s002 (TIF) S3 Fig. RT-PCR analysis of full length MPK12 transcript in Col-0, mpk12-3, and mpk12-4 plants. ACTIN2 was amplified as a control. https://doi.org/10.1371/journal.pbio.2000322.s003 (TIF) S4 Fig. Stomatal index, length, and density in mpk12. (A) Stomatal index of studied lines (mean ± SEM; 1-way ANOVA, Tukey HSD post hoc test). Experiment was repeated twice (n = 81–84 plants). (B) Stomatal complex length of mpk12 lines (mean ± SEM; 1-way ANOVA). Sample size was 4–6 plants, altogether 84–126 stomatal complexes per line were measured. (C) Stomatal density of studied lines (mean ± SEM; 1-way ANOVA). Experiment was repeated twice (n = 81–84 plants). The raw data for panels (A-C) can be found in S1 Data file. https://doi.org/10.1371/journal.pbio.2000322.s004 (TIF) S5 Fig. Time-dependent changes in stomatal conductance. Various stimuli were applied as indicated by the bars or arrows in the legends of each panel. Stomatal opening induced by 100 ppm CO2 (A) and 50 μmol m-2s-1 blue light (B). ABA inhibited light-induced stomatal opening (C). Stomatal closure in response to darkness (D), 800 ppm CO2 (E), decrease in air humidity (F), a 3-minute O3 pulse (G) and spraying the rosette with 5 μM ABA solution (H). The data in all the figures is represented as mean ± SEM. All experiments were repeated at least three times (n = 11–18). The raw data for panels (A-H) can be found in S1 Data file. https://doi.org/10.1371/journal.pbio.2000322.s005 (TIF) S6 Fig. Deletion of MPK12 did not affect ABA-induced stomatal closure. The stomata in the MPK12 deletion mutant mpk12-4 closed after treatment with 10 μM ABA for 30 min, similar as in wild type. Data are average of 3 experiments, 10 stomata per experiment and condition. Small letters denote statistically significant differences according to 2-way ANOVA with Tukey HSD post hoc test. The raw data for the figure can be found in S1 Data file. https://doi.org/10.1371/journal.pbio.2000322.s006 (TIF) S7 Fig. MPK12 interacts with HT1 and IBR5. Split-ubiquitin yeast two-hybrid assays with MPK12 and different versions of MPK12 with amino acid substitutions; MPK12 G53R with the same point mutation as in Cvi-0, MPK12 K70R kinase inactive version, MPK12 Y122C and MPK12 D196G, E200A constitutively active kinase versions. (A) Yeast growth observed on SD-leu-trp plate without 3-amino-1,2,4-triazole (3-AT), 24 hours of X-Gal incubation. (B) Yeast growth observed on SD-leu-trp-his-ade plate with 20 mM 3-AT. (C) Split luciferase complementation assays showed that MPK12 interacts with HT1 in tobacco leaves. MPK12:nLUC with only cLUC was used as negative control, and showed no luciferase bioluminescence signal. https://doi.org/10.1371/journal.pbio.2000322.s007 (TIF) S8 Fig. Stable expression of YFP-labelled MPK12 in intact leaves of Arabidopsis thaliana and transient expression in leaves of Nicotiana benthamiana. Expression of MPK12-YFP (A) and MPK12 G53R-YFP (B) under native MPK12 promoter in A. thaliana Col-0. Transient expression under the CaMV35S promoter was also shown for MPK12-YFP (C) and MPK12 G53R-YFP in N. bethamiana (D). Scale bar = 50 μm. https://doi.org/10.1371/journal.pbio.2000322.s008 (TIF) S9 Fig. MPK11 does not inhibit the activity of HT1. MPK11, an MPK from the same group as MPK12, was not able to inhibit HT1 showing that not all the Arabidopsis MPKs are inhibitors of HT1. This experiment was repeated four times. https://doi.org/10.1371/journal.pbio.2000322.s009 (TIF) S1 Table. Primers used in this study. https://doi.org/10.1371/journal.pbio.2000322.s010 (DOCX) S1 Data. Raw data for all figures and supplemental figures. https://doi.org/10.1371/journal.pbio.2000322.s011 (XLSX) S1 Video. A time course of Col-0, Cvi-0, Col-S and Cvi-T exposed to 350 ppb ozone. https://doi.org/10.1371/journal.pbio.2000322.s012 (WMV) Acknowledgments Tuomas Puukko provided excellent technical assistance. We thank Aleksia Vaattovaara for comments on the manuscript.