Introduction

Balancing selection is a powerful mechanism to maintain genetic variation at selected loci, favoring rare alleles and disfavoring common ones1,2. Therefore, rare alleles are much less likely to go extinct than alleles not under selection. Balancing selection is responsible for the high level of polymorphisms observed at disease-related genes, like MHC in jawed vertebrates and R-genes in plants3,4,5, and at genes involved in protein–protein interaction, like Self-Incompatibility loci in plants and mating-type loci in fungi1. Loci under balancing selection display a genomic footprint that includes high, sometimes extraordinarily high, genetic variation around the target of selection and an excess of genetic polymorphisms segregating at intermediate frequencies (e.g. refs. 6,7,8,9). Of particular interest is the fact that some of these polymorphisms are shared by closely related species, suggesting that they have been maintained over time periods longer than the time since speciation occurred, and much longer than neutral polymorphisms could be expected to persist10. Indeed, some of these polymorphisms are independently maintained in reproductively isolated lineages that diverged millions of years ago11. Understanding such ancient polymorphisms opens a window into the past, helping us to see how selection acted many years ago. Evidence for such rare ancestral polymorphisms has been reported mostly for genes related to immune function, which is consistent with the idea put forth in the Red Queen model that host–parasite coevolution is a major force driving long-term balancing selection12. The Red Queen model of continuous host−parasite coevolution is named after a character of the same name from Lewis Carroll’s Alice in Wonderland, who needs to run constantly to maintain her position13. In evolutionary theory, the idea is that parasite lines with host-genotype-specific infection profiles will be most successful if they adapt primarily to common host genotypes. These common host genotypes will then be selected against, consequently declining in frequency along with the parasites that specialized on them, giving rare genotypes a selective advantage14,15,16,17. The Red Queen model of coevolution is thought to underlie the high level of genetic polymorphisms at loci associated with host defense, as observed in diverse natural systems5,6,18,19. To date, most studies on Red Queen coevolution have focused on the host, while the long-term coevolving parasites have remained unknown.

A system with strong evidence of Red Queen coevolution is the planktonic crustacean Daphnia magna and its obligate and highly virulent bacterial pathogen Pasteuria ramosa. In this system, coevolution has been associated with high diversity at disease loci20,21,22, with resistance being highly variable within and between host populations21,22,23. The same is true for parasite infectivity20,24. Genomic regions containing resistance loci in hosts and infectivity loci in the parasites show strongly elevated diversity levels compared to the genomic background22. Variation in hosts and parasites is linked by high specificity, such that infection can only result when host and parasite genotypes match22,25,26, an important assumption of the Red Queen model and long-term balancing selection27,28. The Daphnia–P. ramosa system represents a unique model to test for footprints of long-term balancing selection in the host genome. In a Eurasian sample of D. magna genomes, a group of P. ramosa resistance genes forming a cluster was identified, at which balancing selection could be inferred from the observed patterns of genomic variation21. The coalescence time of the haplotypes at this cluster is greater than average for the genome. When coalescent times exceed the time since two species diverged, these polymorphisms are believed to have existed in both species since the time before the species formed, a pattern known as trans-species polymorphisms (TSP)11. TSP is generally rare, but has been seen in some disease-related genes in plant and vertebrate hosts6,29,30,31,32,33, but the long-term coevolving parasites are unknown. TSP represents the strongest evidence of long-term balancing selection2,34.

In this work, we scan the genomes of three related host species for patterns of TSPs and test whether TSPs are observed in or near loci associated with polymorphisms for resistance to the virulent and widespread Pasteuria parasite. As part of the process, we analyze resistance phenotypes and whole genomes of 186 naturally collected Daphnia clones. One hundred fifty-seven clones belong to our focal species, D. magna, while 15 and 14 clones to two species, D. similis and D. sinensis, that together form a sister taxon to our focal species. D. similis and D. sinensis diverged about 7 Mya, while they share the most recent common ancestor with D. magna more than 15 Mya35. Our work provides rare evidence of TSPs for loci known to be functionally relevant in interactions with a widespread and highly specific parasite. Thus, our study is consistent with the theoretical prediction that specific antagonistic coevolution is able to maintain genetic diversity over millions of years.

Results and discussion

Resistance profiles of Daphnia clones

Daphnia clones collected from the entire Holarctic (Fig. 1a and Supplementary Data 1) were tested for resistance to five strains of P. ramosa isolated from D. magna populations across Europe. Parasite attachment to the host gut is the crucial step in the infection process and correlates strongly with infection success36,37. Where parasites fail to attach, the hosts are resistant. Thus, we used the parasite attachment test to determine resistance. This test is conducted with fluorescent parasite spores that can be observed through the transparent Daphnia bodies adhering to the cuticle of the fore- or hindgut wall of susceptible hosts, but not of resistant hosts36,38. All three species showed variation among clones for parasite attachment, with 33% of all D. magna clones testing attachment positive (susceptibility), 12.4% for D. similis, and 46.9% for D. sinensis (Table 1 and Supplementary Datas 1 and 2). As previously reported36,39,40, all D. magna clones became infected after successful attachment of at least one P. ramosa isolate, while only about 64% and 74% of D. similis and D. sinensis clones, respectively, became infected after attachment (Supplementary Data 2). These reduced infection rates for D. similis and D. sinensis likely stem from the fact that our five P. ramosa isolates were isolated from D. magna and may show some degree of specific adaptation to this species8,21. Nevertheless, infected Daphnia of all three species displays the same, typical signs of the disease: the host body becomes darkish and nontransparent, and the bacterium castrates the host (Fig. 2)41 and by this, strongly impacts host fitness. Taken together, these results show that the three Daphnia species suffer in the same way from P. ramosa infections and exhibit substantial genetic variation for resistance to P. ramosa that has, presumably, been maintained over a long time.

Fig. 1: Sampling sites, principal component analysis, and phylogenetic relations of the samples of the three Daphnia species used in this study.
figure 1

a Map of the Northern hemisphere (20–70°N) with the Daphnia genotype sampling sites used in this study. b Principal component analysis (PCA) based on 92608 unlinked SNPs describing species diversification. c Simplified representation of the phylogenomic relationship among D. magna, D. similis, and D. sinensis from ref. 35. This tree is fossil-calibrated and based on 636 single-copy orthologs analyzed in ref. 35. Divergence times approximate the minimum estimates obtained when the most recent common ancestor of all Anomopoda (a sub-order including Daphnia) was dated in the Late Jurassic. Source data are provided as a source data file.

Full size image
Table 1 Percent of positive attachment tests for clones of D. magna, D. similis, and D. sinensis to five P. ramosa isolates
Full size table
Fig. 2: Photographs of P. ramosa infected and healthy females of the three Daphnia species used here.
figure 2

D. magna (a), D. similis (b), and D. sinensis (c) phenotypes of infection. On the right are animals infected with P. ramosa, while on the left are healthy animals. The body of infected hosts becomes non-transparent. The bacterium castrates the host, recognizable by the absence of eggs in the brood chamber.

Full size image

The three Daphnia species show no evidence of introgression

Next, we studied the genomes of all three Daphnia species to test for TSPs. First, we ruled out the horizontal transfer of genes by hybridization and introgression, which can mimic TSPs34, by using whole genome resequencing data for all the Daphnia clones to assess the species’ relationships, their genetic structure, and any evidence of admixture among them. A PCA approach showed high genetic differentiation among the three species, with the first PC separating D. magna from the other two species, which were clearly separated by the second PC (Fig. 1b and see also Supplementary Fig. 1). As previous population genomic studies have suggested21,42, differentiation exists because Daphnia magna inhabits most of the Holarctic, population. The PCA of the D. magna clones suggests the presence of three main clusters—a Western Eurasian (WE), an Eastern Asian (EA), and a Northern American (NA) cluster, which was included in the analyses (Supplementary Figs. 2–5). A mitogenome phylogeny supported the same population and species differentiation with five lineages (Supplementary Fig. 6). In addition to the PCA analysis, which indicates a deep split among the three sampled species, we estimated an ABBA-BABA statistic (Patterson’s D43;) to detect gene flow/introgression among our samples. The average value of D for our focal species D. magna in relation to D. similis and D. sinensis was 0.049 and not significantly different from zero (p value > 0.05), indicating no evidence of recent gene flow.

Genomic analysis of shared polymorphisms

For each of the five Daphnia lineages (D. similis, D. sinensis, D. magna WE, D. magna EA, and D. magna NA), we performed single nucleotide polymorphisms (SNPs) calling (Supplementary Table 1), overlapped the five lists of SNPs and obtained a total of 224 high-quality, biallelic, shared polymorphisms (Supplementary Data 3). More than 10% (32 of 224) of these SNPs fall into a large region of about 250 kb in contig 11F, a region previously identified as being highly variable39,40 and under-balancing selection in D. magna21. The 224 SNP candidate list included only polymorphisms observed in all five clades, excluding polymorphisms in D. magna related to clade-specific SNPs.

Balanced polymorphisms are expected to be surrounded by one or more neutral shared polymorphisms that are in strong linkage disequilibrium (LD, i.e., less than 1 kb from each other). Therefore, we applied a filter that required at least two shared SNPs in LD1,30, thereby identifying 35 genomic regions, including 131 SNPs, that satisfied this LD condition (Supplementary Data 4) and that we then considered for further analyses. As the further focus of our study is on the species level, we pooled the clones of the three D. magna lineages into one group to focus on species-level comparison in all following analyses.

To filter out from our list of shared SNPs those variants that are identical-by-state as a result of recurrent mutations, we produced allelic trees. A specific feature associated with the molecular signature of long-term balancing selection is that haplotypes from different species cluster by allele rather than by species. Thus, allelic trees do not resemble the species tree. We built trees based on fragments of variable length (i.e., from 100 bp to 1500 bp) surrounding, and centered on, each of the 131 candidate-shared SNPs. We excluded the focal (shared) SNPs from the fragments used in this analysis, as they were already a filter criterium in a previous step. For each tree, we estimated its probability of having haplotypes clustering by allele (Pallelic) following the method described by Teixeira et al. 31. Among the 131 SNPs for which the local trees were built, we identified 31 SNPs with probabilities higher than 85 in at least one of the generated trees and considered them putative TSPs (Supplementary Datas 4 and 5). These candidate TSPs were located in 11 regions with at least two shared SNPs in LD. In these 11 regions with at least one putative TSP (in total 31 SNPs), there were an additional 25 shared SNPs (a total of 56 shared SNPs, Supplementary Data 5).

Since long-term balancing selection is expected to maintain adaptive genetic diversity, we hypothesize that TSPs fall in coding regions that show excess polymorphism. For this reason, we calculated the ratio of polymorphism to divergence (PtoD) following Teixeira et al. 31 for all annotated genes in the D. magna reference genome, correcting for gene length. PtoD, which accounts for mutation rate, was obtained as p/(d + 1) where p is the number of polymorphisms in the samples of one species and d is the number of fixed SNPs between species. We then focused on the genes containing putative TSPs and calculated where the candidate genes fell relative to the empirical PtoD distribution of all genes. Each of the 11 regions with at least two shared SNPs in LD contained one gene with putative TSPs. Four of these genes showed exceptional variability, with the average of the four PtoD percentiles falling in the top 10% of PtoD distribution and no single value over the 15% percentile (Table 2 and Supplementary Data 6). Finally, the topologies of the gene trees built in the surrounding of the four remaining candidates suggest that haplotypes indeed cluster by allele and not by species (Fig. 3 and Supplementary Fig. 6), providing strong support for TSP.

Table 2 PtoD estimates and summary statistics indicating balancing selection for the four genes that included at least one putative TSP
Full size table
Fig. 3: Neighbor joining trees of haplotypes including the focal shared SNPs with evidence for TSP.
figure 3

The trees based on the sequences surrounding the best TSP candidates are reported: a 11F-21.35–100 bp, b 11F-27.73–100 bp, c 18F-16.99–200 bp, and d 28F-3.33–200 bp. In these trees, haplotypes cluster by allele and not by species, i.e., each cluster contains three colors: D. magna: blue, D. similis: yellow, and D. sinensis: green. In some cases, external branches were collapsed in order to facilitate tree representation. The alleles reported on the branches indicate the SNP variant present in all the haplotypes in the relative cluster. The circle area is proportional to the haplotype frequency, except when the haplotype frequency is too high for the purpose of the representation and a number is reported indicating the total of similar or identical haplotypes. Source data are provided as a source data file.

Full size image

Population genetic analysis of shared polymorphisms

Balancing selection maintains polymorphisms at selected loci, resulting in characteristic genomic signatures1,10 such as elevated nucleotide diversity (π) and Tajima’s D, which are indicative of increased genetic variation around the target of selection and excess polymorphisms segregating at intermediate frequencies, respectively. Balancing selection also reduces population or species differentiation as measured by Fst1,10. We therefore calculated π, Tajima’s D, and Fst for non-overlapping sliding windows of different sizes (i.e., 1 kb, 2.5 kb, and 5 kb) in the regions surrounding our candidate TSPs. The same was done for the 20 longest contigs and the contig where the TSP is located that functioned as background level. The target region was identified as a 100 kb genomic segment centered on a candidate TSP. For each species, we then assessed whether π and Tajima’s D were elevated and if Fst values were lower in the candidate regions compared to the background level, as theoretically expected for sites under long-term balancing selection (Supplementary Figs. 7–15 and Supplementary Tables 2–10). For our candidates, regardless of the size of the sliding window applied, we found a significant deviation from the background level in most of the statistics indicative of balancing selection (Table 2). These observations suggest that in each of the three reproductively isolated and highly divergent Daphnia species, the genomic regions around the putative TSPs display high genetic variability and that the species are less differentiated in these regions than expected from a genome-wide average. We further validated these results by applying the normalized ß statistics, a test specifically designed to detect clusters of polymorphisms segregating at similar frequencies, a pattern associated with signatures of long-term balancing selection44. Most comparisons showed significantly elevated ß scores in the target regions compared to the background level (Table 2, Supplementary Figs. 16–18, and Supplementary Tables 11–19), further supporting our hypothesis that the regions surrounding TSPs evolve under long-term balancing selection.

To ensure that our observations were not biased by the occurrence of paralogs, which can cause read mismapping, or by other artifacts, we Sanger-sequenced the flanking regions of the identified TSPs. We included this control step because sequence duplication is known to determine pseudo-heterozygosity45. In addition, we wanted to ensure that there were no inconsistencies between the genotypes obtained with Illumina sequencing and the ones obtained with Sanger sequencing (details in Supplementary Tables 20). We were able to confirm the Illumina genotypes and the presence of a plausible number of heterozygous positions surrounding all our candidates (examples of chromatograms in Supplementary Figs. 19–24).

Functional analysis of the shared polymorphisms

The four candidate regions revealed interesting associations. Two of them are located on contig 11F, in close association with a large cluster of resistance loci for the virulent parasite P. ramosa. This cluster includes four known resistance genes, A, B, C, and F39,40, and has been shown to be under balancing selection in a Euasian D. magna sample21. A, B, and C form a cluster with strongly reduced recombination, while the F locus is directly adjacent to it. One of the TSPs is situated about 48 kb before the ABC cluster, while the other is located within the F locus. The TSP observed on contig 18F is located in a region rich in immunity-related genes (e.g., multiple lectins, TRIMs, clip serine protease), about 101 kb before another Pasteuria resistance locus (the D locus)38, while contig 28 F includes multiple genes with immune functions (e.g., lectins, clip serine proteases, complement related elements). These observations suggest that long-term balancing selection maintains, independently in three Daphnia species, ancestral alleles in genomic regions that have a functional link with P. ramosa resistance

Our best TSP candidates, three of which showed at least a non-synonymous TSP, can be classified as immune-related genes or genes involved in immunity pathways (Table 2). The gene 11F-21.35, which is in close proximity with the ABC,F resistance loci against P. ramosa, encodes a C1q domain-containing protein. C1q are immune molecules responsible for initiating the complement pathway in vertebrates. Their role in invertebrate immunity has been demonstrated in mollusks and their presence is documented in crustaceans46. This gene was shown to be upregulated in the D. magna transcriptome 8 days and 16 days after Pasteuria exposure47. Another gene, 11F-23.73 contained in the Pasteuria resistance F locus, is a member of a large “Cladoceran-specific protein” family40. Genes in this family, 11F-23.73 included, have features suggesting they play a role in host−pathogen interaction, such as signaling capacity via a putative STAT binding site and differential expression between susceptible and resistant clones40. The two remaining candidate genes on contigs 18 and 28 have no known direct immune-related function. Candidate 18F-16.99 is a Na+/K+-ATPase that can function as a signal transducer/integrator to regulate the production of reactive oxygen species, which are involved in a large number of processes including macrophage-mediated immunity48. Gene 28F-3.33 is a metalloendopeptidase, which is considered to play a major role in regulating numerous physiological and pathological processes, as well as immune responses in invertebrates49. While our candidates are enriched for immune function-related genes, we need to keep in mind that also other evolutionary processes, might be responsible for the maintenance of TSPs.

In conclusion, our study provides genomic evidence of TSPs in or close to known resistance loci (ABC, D, and F loci) in the focal species of this study, D. magna. The loci are involved in the coevolutionary interactions between the Daphnia host and P. ramosa. In order to be effective, the selective agent underlying the maintenance of the TSPs should be geographically widespread, should be present over an extended period of time, and should show similar tropism towards related host species50. The parasite P. ramosa fulfills all these criteria: it has been shown to have a wide geographic distribution51, a long history of coevolution with its obligate hosts21,52,53, and—as demonstrated here—phenotypic variation in resistance and a similar infection phenotype among evolutionary-related species (Table 1 and Fig. 2). Together, these results suggest that long-term balancing selection for P. ramosa resistance is a determinant in the maintenance of ancestral alleles in D. magna and the sister clade that shared a common ancestor at least 15 million years ago. Examples of TSPs, as a signature of ancient coevolution, are thought to be extremely rare30, and indeed, we found only a few instances of identical-by-descent TSPs in our study. In contrast to earlier reports, our study provides compelling evidence for the causal agent driving coevolution in this system, and thus, corroborates the hypothesis that specific—one host–one parasite—antagonistic coevolution can be an important factor in the generation and maintenance of genetic diversity2. It represents to our knowledge the first evidence of long-term balancing selection concerning a specific infectious disease and the loci directly underlying resistance to this disease.

Methods

Samples and phenotyping

Daphnia are cyclical parthenogens that can be maintained as stable clones under lab conditions in an asexual mode of reproduction. These clonal cultures produce sufficient material for whole genome resequencing. We sequenced 186 clones from three Daphnia species: D. magna (n = 157 clones), D. similis (n = 15), and D. sinensis (n = 14). These lines were collected and isolated from natural populations as adult females or resting eggs and maintained under laboratory conditions as clonal lines35,42,54. D. similis and D. sinensis are suggested to be sister species; together, they are close to D. magna. Another newly identified sister species of D. magna is only known from one population and was therefore not used here35. A phylogenomic analysis has suggested that D. magna diverged from D. similis/D. sinensis at least 15 Mya, while D. similis and D. sinensis share the most recent common ancestor about 7 Mya35. In our study, we aimed to include the widest geographic spread for Daphnia clones, which encompasses the entire Holarctic for D. magna, and much of Eurasia for the other two species (Fig. 1a). Details of the clones included in the study are reported in Supplementary Datas 1 and 2.

We tested all Daphnia clones for resistance to different isolates of the bacterial parasite P. ramosa. For each Daphnia genotype, we estimated the resistance phenotype (resistotype) using five Pasteuria strains (C1, C19, P15, P20, and P21) isolated from natural D. magna populations across Europe. Since the attachment of parasite spores to the host is the most important step in the infection procedure and explains about 90% of the overall variation in infection success36,41, resistance is defined as the ability of the host to prevent attachment to a given attachment site. The attachment test described in ref. 36 allows us to score resistotypes by exposing the host to fluorescently labeled Pasteuria spores and then observing whether these spores attach to the cuticle of the Daphnia’s fore- or hindgut, where the parasite penetrates the host’s cuticle and enters its body cavity36,41. Specifically, we assessed whether labeled spores attached to the host’s foregut (all five Pasteuria strains) and/or hindgut (two Pasteuria strains: P15 and P21). For each Daphnia genotype, six replicates were performed, and a genotype was considered susceptible to the bacterial strain when more than half of the replicates showed attachment (Supplementary Data 1). In D. magna, resistance and susceptible phenotypes were found to be uniformly distributed across the entire species distribution (Supplementary Data 1) as previously shown in ref. 21, including the newly analyzed North American clones.

Attachment of Pasteuria spores in the native host species from which it was isolated leads to successful infection8,36. Attachment polymorphisms are also seen in related non-native host species; however, in these cases, the correlation between attachment and successful infection is reduced8,36. Besides attachment tests, we performed infection trials by experimentally exposing D. similis and D. sinensis clones, that were positive to spore attachment to the five previously mentioned Pasteuria strains, following the protocols of Luijckx et al. 8. To determine infection, we observed crushed D. magna host bodies under a phase contrast microscope (400×); individual Daphnia were considered infected when the presence of Pasteuria spores was detected. When at least one replicate of a clone became infected, we considered the clone infectable.

DNA isolation and sequencing

Before genomic DNA extraction, we reduced bacterial and other non-target DNA. For this, all animals were kept for three days in a solution of Ampicillin, Streptomycin, and Tetracycline (Sigma) at a concentration of 50 mg/l each, and transferred daily into a fresh antibiotic solution. To reduce gut content, the animals were not fed during this three-day treatment, instead receiving 5 mg of superfine beads of the gel filtration resin Sephadex ® G-25 (Sigma-Aldrich) twice a day in their medium. Sephadex causes gut evacuation when ingested by the animals. We extracted DNA from 20 to 50 animals of each clone using the QIAGEN Gentra Puregene Tissue Kit, including the RNaseA (100 mg/ml; Sigma) digestion step. Whole-genome Illumina paired-end sequencing (read length 125 bp) was performed by the Genomics Facility service platform at the Department of Biosystem Science and Engineering (D-BSSE, ETH) in Basel, Switzerland, on an Illumina HiSeq 2500.

SNP calling

Raw reads were trimmed to remove adapters using Trimmomatic 0.3655. We aligned the trimmed reads to a newly assembled high-quality reference genome of a three-times selfed Finnish D. magna clone (the same genotype for which a previous genome assembly has been released (NCBI accession LRGB00000000.1)). This reference genome, assembled using the FALCON assembler56, was obtained using high-coverage PacBio long-read sequencing (see Data availability). In order to increase the accuracy of read mapping, different approaches were used for the reads obtained from the different species. For D. magna clones, BWA v0.7.757, a software package for mapping low-divergent sequences against a reference genome, was used to align the reads. To align the reads from D. similis and D. sinensis clones, we used Stampy v1.0.3258, an algorithm specifically designed to account for the divergence between the reference genome and the mapping reads. Polymorphisms were called using the HaplotypeCaller function in GATK 4.059. Using VCFTOOLS 0.1.1660, we retained only high-quality biallelic SNPs, satisfying the conditions of minimum genotype depth of 10, maximum genotype depth of 100, minimum genotype quality of 30, maximum amount of missing data of 50%, and minor allele frequency of 0.05.

Genotype clustering and test for introgression

Working with the dataset of all 186 Daphnia clones, we used a Principal Component Analysis (PCA) implemented in the R package SNPRelate61 to detect possible evidence of recent introgression among the three species. In order to account for the linkage between markers, we analyzed only SNPs at least 1 kb distant from each other. As expected, the PCA approach showed high genetic differentiation among the three species, with the first PC separating D. magna from D. similis and D. sinensis (Fig. 1b and Supplementary Figs. 1 and 2). Since PCA can be biased by unequal sample size62, we repeated the PCA analysis with subsamples of only 15 D. magna clones. The result remained unchanged (Supplementary Fig. 1). A PCA analysis was also used to assess evidence of population structure and recent introgression events among lineages within D. magna that inhabit most of the Northern Hemisphere. It has already been shown that population structure exists in D. magna42,63. Here, we increased the number of D. magna clones and included clones from North America for the first time (Fig. 1). For this reason, a detailed assessment of population differentiation among the D. magna clones was necessary. The PCA results obtained for the D. magna clones confirmed the presence of three main clusters, namely a WE, an EA, and an NA cluster (Supplementary Fig. 4).

To test for recent introgression among the three Daphnia species, we estimated an ABBA-BABA statistic (or Patterson’s D43;), which detects geneflow among samples. We used the same genotypes as for our PCA analysis, as well as the program Dsuite64 to estimate D. Significant elevation of D above zero was assessed by applying a block-jackknife resampling procedure65,66 with 20 subsamples.

Mitochondrial genome assembly and mitogenome phylogeny

We assembled the mitochondrial genome for a subset of Daphnia clones following the protocol described in ref. 35. The mitogenomes were then annotated using the MITOS web server67. We aligned the protein-coding genes with MUSCLE v3.8.3168 and concatenated them to generate a mitochondrial phylogeny using the software RAxML v8.1.2069. The best ML tree was obtained by assuming a general time reversal model of sequence evolution with a gamma-distributed model of rate heterogeneity, taking into account gene partitioning. In order to test for tree reliability, we performed a bootstrap approach by generating 100 pseudo-replicates. Daphnia hispanica was used as an outgroup (ENA accession number: LS991493), as this species is known to be closely related to the species considered here35.

The best ML mitochondrial phylogeny resembled the previously reported topology35 with D. similis and D. sinensis being sisters as part of the similis group described in ref. 70. The similis group, in turn, is sister to D. magna. The obtained tree was highly supported with almost all internal nodes, showing a bootstrap value of 100. Although their evolutionary relationship cannot be fully resolved, the three D. magna lineages highlighted by genome-wide polymorphisms were also highly supported by the mitogenomes (Supplementary Fig. 5). Overall, the mitogenomes suggest, as shown by SNP data, the occurrence of five main lineages among the Daphnia analyzed in this study.

Finding shared SNPs

After defining the five main lineages of the Daphnia clones in this study, we performed SNP calling within each of these lineages. We used VCFTOOLS 0.1.1660 and the requisites specified above to obtain lists of high-quality polymorphisms for each of the five Daphnia lineages (D. magna from North America (13), D. magna from Eastern Asia (18), D. magna from Western Europe (126), D. similis (15), and D. sinensis (14). We then overlapped the lists of SNPs to identify shared polymorphisms among species/lineages. We obtained a total of 227 polymorphisms shared among the five lineages. In order to obtain a reliable list of shared SNPs, we ensured that the identified polymorphisms did not lie in low-complexity regions (i.e., regions rich in repeats), where were identified in our reference genomes using the software DustMasker 1.071. The three identified shared SNPs that were found to occur in low-complexity regions were excluded from our set of candidate polymorphisms.

Distinguishing between shared and TSP

In order to exclude variants from our list of shared SNPs that are identical-by-state as a result of recurrent mutations, we used, as a reference, features associated with the molecular signature of long-term balancing selection. The main feature is that haplotypes from different species cluster by allele, rather than species, i.e., they produce allelic trees, and not trees resembling the species tree. These haplotypes include polymorphisms that predate species splits, while neutral recurrent shared polymorphisms are expected to generate species trees72. Because of the long-term effects of recombination, we expect this signature to be restricted to a short genomic region around the putative TSP1. In addition, this segment is likely to contain at least two shared polymorphisms that arose in the ancestral populations of D. magna-D. similis-D. sinensis and are in strong or complete LD with the selected site30. Thus, we used these properties to exclude SNPs from our candidate list that resulted from recurrent mutations. Specifically, we required shared SNPs to be in LD (i.e., within 1 kb) with at least one other shared SNP. Among the 224 shared SNPs retained after the previous filtering (listed in Supplementary Data 3), we identified 35 regions (including 131 SNPs) having at least two shared SNPs in LD and were considered for further analyses (Supplementary Data 4).

For the polymorphisms that satisfied the above condition, we generated haplotype trees based on haplotype alignments of putative TSPs flanking regions. We selected regions of variable length (100 bp, 150 bp, 200 bp, 400 bp, 600 bp, 800 bp, 1000 bp, and 1500 bp) centered on the putative TSPs and inferred the chromosomal phase for each fragment using the software SHAPEIT473. Phasing was done for all species and samples simultaneously, including the use of individual BAM files to incorporate read-based phasing. Then we built local trees excluding the putative TSPs from the sequences. For this, we used a subset of the Daphnia clones included in the study consisting of all the 15 D. similis clones, all the 14 D. sinensis clones, all 13 D. magna clones from North America, all 18 D. magna genotypes from Eastern Asia and 20 randomly selected D. magna clones from Western Eurasia. We performed this subsampling procedure so as to have similar sample sizes among species/lineages. Because our focus was at the species level in this and all subsequent analyses, we pooled the clones of all three D. magna lineages into one group. For each allele tree, which included 160 haplotypes, we estimated the probability of having haplotypes clustering by allele (Pallelic) following the method described in ref. 31. Specifically, for each generated tree, we applied a resampling strategy with six randomly selected haplotypes (one haplotype per allele and per species) 1000 times. With these haplotypes, we built neighbor-joining trees that were considered allelic trees if the three closest tips were haplotypes with the same allele of the three species. In this way, for each of the local trees surrounding a putative TSP, we obtained a Pallelic that was estimated as the proportion of trees having haplotypes clustering by allele over the 1000 resampled trees. A shared SNP was considered a candidate TSP when it showed a Pallelic > 85 in at least one of the generated trees (i.e., trees based on regions of variable length).

Annotation of TSPs and gene function identification

The retained SNPs from the previous step were annotated and their effect was predicted using the software SnpEff74 (Supplementary Data 5). Among those, 47 (84%) were located in coding sequences, 17 of which were non-synonymous and 28 synonymous variants. One substitution determined the gain of a stop codon (Supplementary Data 5). The 11 identified regions overlapped with 11 different genes, which are reported alongside their function in Supplementary Data 6.

Since D. magna genome annotation is not as accurate as the D. pulex annotation, we used a blast approach to obtain the D. pulex orthologs of the 11 genes containing TSPs in order to infer a functional classification. Specifically, we used blastp75 and the fasta sequences of the D. magna proteins as queries against a D. pulex protein database (from: http://wfleabase.org/genome/Daphnia_pulex/dpulex_genes2017/genes/). We used an e-value threshold of 1e-40 and considered the most likely ortholog of the hit showing the lowest e-value. For one gene (genemark-000017F-processed-gene-17.81-mRNA-1), we did not identify sequence similarities in D. pulex even when the e-value threshold was increased to 1e-10. Since this suggested there is no D. pulex ortholog for this gene, it was therefore classified as an “uncharacterized protein” in the D. magna annotation. We then used Panther (v16.0)76,77 to obtain a functional classification for each of the remaining ten genes using Daphnia pulex, included in the Panther database, as a reference organism. For six of the 11 genes, we could not find a defined function. These genes may represent cases of Daphnia-specific gene families (accounting for about 36% of D. pulex genes) that do not have homologs78, or they might be the result of some drawbacks of our blast approach. All genes were further inspected with InterProScan (v5.48)79 to provide additional validation of gene function and assess the presence of specific domains. Overall, of the 11 genes containing TSPs, two could be associated with an immune-related function (genemark-000011F-processed-gene-21.35-mRNA-1: C1q domain-containing protein, maker-000011F-snap-gene-23.73-mRNA-1: cladocera-specific protein).

Polymorphism-to-divergence ratio

Since long-term balancing selection is expected to maintain adaptive genetic diversity, we hypothesized that TSPs fall into coding regions that show an excess of polymorphism. We thus calculated the ratio of PtoD following31 for all genes in our reference genome. PtoD, which accounts for mutation rate, was obtained as p/(d + 1) where p is the number of polymorphisms in one species and d is the number of fixed SNPs between species. We calculated d with vcftools as the number of SNPs showing an Fst = 1 in the following species pairs: D. magna vs D. sinensis for D. sinensis; D. magna vs D. similis for D. similis; for D. magna we made two comparisons: D. sinensis vs D. magna and D. similis vs D. magna. We then corrected for a slightly positive correlation (R = 0.06, p < 0.05) observed between PtoD and gene length by dividing the calculated PtoD by the gene length; the obtained value was multiplied by 1000 to make it more readable. We then focused on the corrected PtoD values observed in the genes containing putative TSPs and determined where the candidate genes fell into the empirical PtoD distribution of all genes. Haplotype trees obtained in the flanking regions of the TSPs identified in our top candidate genes are shown in Fig. 3.

Molecular signature of balancing selection in the candidate regions

A molecular signature of (long-term) balancing selection is high genetic diversity in correspondence with TSPs, since polymorphisms near these loci should be more ancient than the average genome-wide coalescent time72. We, therefore, calculated nucleotide diversity (π) for the regions surrounding the TSPs in non-overlapping sliding windows of 5000 bp, 2500 bp, and 1000 bp using VCFTOOLS 0.1.1660 and assessed, within each species, if π was elevated compared to the background level of genetic diversity (i.e., the contig where the TSP is located, excluding the candidate region, and the 20 longest contigs in our reference genome without loci known to be associated with Pasteuria resistance). We used multiple sliding window sizes in order to test if the size of the window influences the results, given that balancing selection is expected to leave its molecular footprints in relatively short genomic fragments1. Furthermore, because the allele frequency distribution for sites under balancing selection is expected to exhibit a trend towards intermediate frequencies, we tested this hypothesis with the neutrality test Tajima’s D80, which summarizes polymorphism data based on the difference between the average pairwise diversity (k) and the number of segregating sites (S). When there is an excess of polymorphisms at intermediate frequency, k increases more than S, resulting in a positive Tajima’s D. Within each lineage, we also assessed, using the same approach as above (i.e., calculating Tajima’s D in non-overlapping sliding windows of 5000 bp, 2500 bp, and 1000 bp), if Tajima’s D displayed significantly elevated values in the regions surrounding TSPs compared to the background level. Moreover, theory predicts that balancing selection reduces population differentiation, as measured by Fst1,30; consequently, Fst values lower than the background level, are expected in the proximity of ancestral polymorphisms. We tested this prediction by comparing Fst obtained in non-overlapping sliding windows of 5000 bp, 2500 bp, and 1000 bp in the candidate regions vs. the background.

In order to validate these candidate regions, we also performed a test specifically designed to detect signatures of long-term balancing selection44. We calculated, using the Betascan program44, the ß score that quantifies allele frequency correlations between SNPs. We specified the length of the window surrounding each SNP following the formula in ref. 44. For window size calculation we used Daphnia-specific parameters (T, time since balancing selection, and ρ, the recombination rate) as in ref. 21, which resulted in a window size of 250 bp (option-w 250). For the other population genetic indexes, we averaged ß scores in non-overlapping sliding windows of 5000 bp, 2500 bp, and 1000 bp that were then compared between candidate regions and backgrounds. These analyses were repeated three times using different minor allele frequency thresholds (i.e., 0.05, 0.1, and 0.17, following44) to explore the effects of this parameter on the results. The results proved to be very consistent (Supplementary Tables 11–19), and in the main text, as well as in Table 2, we report the ß score results obtained with a 5000 bp sliding window and a minor allele frequency of 0.05.

Contig 11F, where two genes including TSPs were identified and where the ABC and the F Pasteuria resistant loci are located39, is about 3 Mb long. The ABC locus is characterized by a large (i.e., about 150 kb) nonhomologous region (NHR) segregating in susceptible and resistant D. magna clones39. Therefore, we excluded the polymorphisms in this NHR in our comparisons. We plotted the obtained results and performed statistical analyses using Wilcoxon rank sum test) with ggplot2 and R.

Functional annotation

Our study highlighted four genes as the most likely candidates to have evolved under long-term balancing selection and to contain TSPs. However, as we did not obtain a precise functional annotation using a blast approach for one of the four genes, we initially classified it as an “uncharacterized protein” (Supplementary Data 6). In fact, this “uncharacterized protein” (gene 11F.23.73) had already been identified as a member of a very large immunity-related gene family (the Cladoceran-specific protein” family with over 100 members) and was also an outlier in a study that focused on the P. ramosa resistant locus F40. This gene, 11F-21.35, has several features suggesting involvement in the interaction between Daphnia and Pasteuria. It encodes a protein with two domains: a COLFI domain (the fibrillar collagen C-terminal domain) and a globular C1q domain at the C terminus. Both domains are known to have binding capacities. The members of the C1q family can be involved in host defense, inflammation, apoptosis, and autoimmunity81. As a pattern recognition molecule, C1q can engage a variety of ligands via its globular domain and ultimately modulate cells of the immune system. Its role in invertebrate immunity has been demonstrated in several mollusks82, and its presence is documented in crustaceans46.

The gene 18F-16.99 encodes a Na+/K+-ATPase. Na+/K+-ATPase can function as a signal transducer/integrator to regulate the production of reactive oxygen species.

Reactive oxygen species are involved in a large number of processes including macrophage-mediated immunity48. They can engage in direct antimicrobial activity against bacteria and parasites, as well as redox-regulation of immune signaling and induction of inflammasome activation.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.