Since the early studies of evolutionary genetics, there has been no understanding of how and why genetic diversity levels vary between species. This old puzzle, considered four decades ago as ‘the central problem in population genetics’1, is still essentially unsolved in the genomic era2. Meanwhile, there is increasing evidence that genetic diversity is central to many conservation challenges, such as species response to environmental changes, ecosystem recovery, and the viability of recently endangered populations3,4,5,6,7. In this context, our ability to understand and predict this key aspect of biodiversity seems critical. But is it possible to quantify the contributory ecological and genetic factors? How predictable is the level of genetic diversity of a given species?

Population genetic theory states that neutral genetic polymorphism (that is, diversity) increases with effective population size, N e , which in a panmictic population is equal to the number of individuals contributing to reproduction. One would therefore expect the genetic diversity of a species to be linked to biological traits associated with abundance, such as body size or fecundity. However, this intuitive prediction has not yet been clearly confirmed by empirical data2,8,9,10. This is typically explained by invoking the many confounding factors potentially affecting genetic polymorphism, such as mutation rate, population structure, population bottlenecks, selective sweeps, and, more generally, ecological disturbances11,12. Whether demographic or adaptive, historical contingency is often considered to be the main driver of genetic diversity11. According to this viewpoint, polymorphism levels would be expected to fluctuate in time more or less randomly, irrespective of life-history traits.

In the absence of compelling empirical evidence, the relative importance of species biology and ecology (on the one hand) and historical, contingent factors (on the other) in shaping the genetic diversity of species is still highly contentious. Indeed, current knowledge on species genetic diversity is based on just a handful of model organisms, or small sets of molecular data2,8,13. Various animal taxa and lifestyles, particularly across the invertebrates, have yet to be explored. Here we fill this gap and present the first distribution of genome-wide polymorphism levels across the metazoan tree of life.

We focused on 31 families of animals spread across eight major animal phyla. In each family we produced high-coverage transcriptomic data (RNAseq) for about ten individuals of a particular species. In 25 of these families, we sampled one to three additional species of similar biology and ecology (two to seven individuals each), thus producing taxonomic replicates. The total data set consisted of 374 individual transcriptomes from 76 non-model species (Fig. 1, Extended Data Fig. 1 and Supplementary Tables 1 and 2), from which we predicted protein coding sequences14 and identified diploid genotypes and single nucleotide polymorphisms15,16 (Methods). Across species the number of analysed loci varied from 804 to 20,222 (median 5,347) and the number of polymorphic sites from 1,759 to ∼230,000 (median 17,924).

Figure 1: Genome-wide genetic diversity across the metazoan tree of life. Each branch of the tree represents a species (n = 76). The leftmost vertical coloured bar is the estimated genome-wide genetic diversity (π s ), the central bar is the prediction of π s based on a linear model with propagule size as the explanatory variable (P <10−14, r2 = 0.56), and the rightmost bar is the prediction of π s based on a linear model with average distance between GPS records, maximal distance between GPS records, average distance to Equator and invasive status as explanatory variables (P = 0.16). Each thumbnail corresponds to one metazoan family. Species are in the same order as in Supplementary Table 2 (from top to bottom). Full size image Download PowerPoint slide

Estimates of the synonymous nucleotide diversity (π s ) spanned two orders of magnitude across species, a range far wider than is usually observed in surveys restricted to a single taxonomic group. The extreme values of π s were observed in two invertebrate species: 0.1% in the subterranean termite Reticulitermes grassei; 8.3% in the slipper shell Bostrycapulus aculeatus. Figure 1 illustrates the patchy distribution of low-diversity (green) and high-diversity (red) species across the metazoan phylogeny. It also shows that species π s values tend to be similar within families, but distinct between families (analysis of variance; P < 10−12). Such a strong taxonomic effect would be unexpected if stochastic disturbances and contingent effects were the main drivers of genetic diversity, because species from a given family are not particularly expected to share a common demographic history. Testing this hypothesis more thoroughly, we detected no strong relationship between π s and any variable related to geography, such as the average distance between GPS records (regression test, P = 0.19, r2 = 0.02), maximum distance between GPS records (P = 0.02, r2 = 0.07), average distance to Equator (P = 0.87, r2 = 0.0003), population structure (measured as F it , P = 0.22, r2 = 0.02), invasive status (Student’s t-test, P = 0.14) and marine versus continental environment (Student’s t-test, P = 0.52).

To test whether species biology can explain variations in π s , we collected data for six life-history traits potentially related to N e : adult size, body mass, maximum longevity, adult dispersion ability, fecundity and propagule size (Supplementary Table 3). In contrast to the geographic variables, all these traits were significantly correlated with the nucleotide diversity (Extended Data Fig. 2) and collectively explained 73% of the variance in π s in a multiple linear regression test (P < 10−10). Propagule size, here defined as the size of the stage that leaves its parents and disperses (egg or juvenile depending on species), is by far the most predictive of these variables (linear regression test, r2 = 0.56; Fig. 2a). This is illustrated in Fig. 1 by the good agreement between the observed distribution of π s (leftmost coloured vertical bar) and the π s value predicted from propagule size (central bar). The predicted π s based on four demographic metrics is plotted alongside (rightmost bar) for visual comparison.

Figure 2: Life-history traits and genetic diversity relationships. a, Relationship between propagule size and π s (P <10−14, r2 = 0.56, 76 species included; see Fig. 1). b, Relationship between adult size and π s (P <0.05, r2 = 0.07, 76 species included). The colour scale represents the degree of parental investment, here defined as the ratio of propagule size to adult size. c, Effect of fecundity per day (x axis) and propagule size (y axis) on genetic diversity (colour scale; P <10−6, r2 = 0.69, 29 family-averaged data points). d, Phylogenetic contrasts of family-averaged π s versus family-averaged propagule size (P <10−6, r2 = 0.62). Full size image Download PowerPoint slide

We explored in more detail the relative impact on π s of the various life-history traits of interest here (Extended Data Fig. 2). Figure 2b plots the relationship between π s and species adult size, a variable typically taken as a proxy for population size in some taxa9. Although significant, the correlation is not particularly strong (P = 0.018, r2 = 0.07). In particular, species with low genetic diversity cover a large range of body sizes, from less than 1 cm to more than 1 m. Low-polymorphism species include amniotes (turtles, mammals and birds), but also brooding marine species (seahorses, brooding urchins, nemerteans and brittle-stars), eusocial insects (ants, bees and termites) and cuttlefish. These phylogenetically unrelated species have in common a large parental investment in their offspring, as represented in Fig. 2b by the ratio of propagule size to adult size (red). In contrast, species with minimal parental investment (blue) tend to carry high genetic diversity given their size. This is typically the case of highly fecund, broadcast spawning sessile species (such as mussels, non-brooding urchins, nemerteans and brittle-stars, sea squirts and gorgonians). The trade-off between offspring quantity (fecundity) and quality (propagule size) seems to be the most relevant factor explaining variations in polymorphism between species in the animal kingdom (Fig. 2c). We shall for simplicity hereafter categorize as K-strategists the species that tend to invest in the quality of their progeny, and as r-strategists those that favour quantity17.

The correlation we report between life-history traits and π s is not due to phylogenetic non-independence of the sampled species: taking family averages from Fig. 1 increased the correlation coefficients (from r2 = 0.56 to r2 = 0.66 with propagule size alone, from r2 = 0.73 to r2 = 0.79 with the six life-history traits). When we took into account the between-family phylogenetic tree using independent contrasts, this still resulted in highly significant correlations between π s and life-history traits (r2 = 0.62 for propagule size; Fig. 2d and Extended Data Fig. 3). These relationships were also unaffected by sampling strategy, sequencing depth, gene expression levels or contaminants (Methods, Supplementary Table 4 and Extended Data Figs 4, 5, 6). Finally, our conclusions were unchanged when we included 14 previously published species of mammals10 or when we restricted the analysis to a subset of common orthologous genes (Supplementary Table 4).

The relationship between π s and life-history traits, however strong, could in principle be mediated by causative variables that were not included in the analysis. One of these potential confounding factors is the mutation rate: a higher average per-generation mutation rate in r-strategists than in K-strategists could explain our results irrespective of the population size effect. However, theoretical models and empirical measurements actually support the opposite; that is, an increased per-generation mutation rate in large, long-lived organisms due to a larger number of germline cell divisions per generation and a reduced efficacy of natural selection on the fidelity of polymerases18. Therefore, as far as we can tell, across-species variations of mutation rate are likely to oppose, not strengthen, the main effect we are reporting here.

We computed the non-synonymous nucleotide diversity, π n , and this was also found to be correlated with species life-history traits (Extended Data Fig. 2). We found substantial variation in π n /π s across metazoan species, and significant correlations with life-history traits, the best predictor in this case being longevity (Extended Data Fig. 7). This positive correlation is predicted by the nearly neutral theory of molecular evolution19: in small populations (long-lived species), the enhanced genetic drift counteracts purifying selection and promotes the segregation of weakly deleterious, non-synonymous mutations at high allele frequency. These results also confirm that the relationships we uncovered between life-history traits and diversity patterns are mediated in the first place by an effect of N e , not of the mutation rate; synonymous and non-synonymous positions being physically interspersed, the π n /π s ratio is unaffected by the mutation rate.

Our analysis reveals that polymorphism levels are well predicted by species biology, whereas historical and contingent factors are only minor determinants of the genetic diversity of a species. This unexpected result opens new questions. How can life-history traits be so predictive of π s in spite of the overwhelming evidence for the impact of ecological perturbations on patterns of genetic variation11,12? Why does the ‘r/K gradient’ affect genetic polymorphism so strongly?

In an attempt to resolve these paradoxes, we suggest that life-history strategies might influence the response of species to environmental perturbations. Because K-strategy species have been selected for survival and the optimization of offspring quality in complex, stable environments17, we speculate that they might experience fewer occasional disturbances (or be less sensitive to them), thus ensuring the long-term viability of even small populations. In contrast, only species with a large population-carrying capacity could sustain the ‘riskier’ r-strategy in the long term, thus buffering the frequent bottlenecks experienced in the context of high environmental sensitivity (see Supplementary Equations for a model formalizing these arguments). According to this hypothesis, environmental perturbations would be a common factor affecting every species, but their demographic impact would depend on the life-history strategy of each species.

This study highlights the importance of species life-history strategy when it comes to turning genetic diversity measures into conservation policy. So far, conservation efforts have mainly been focused on large-sized vertebrates. Here we show that these popular animals represent only a subset of the existing low-diversity, K-strategists. Invertebrate species with strong parental investment are probably equally vulnerable to genetic risks. Our results also indicate that r-strategists will typically show elevated amounts of genetic diversity irrespective of their current demography, which suggests that species of this kind might face significant extinction risks20 without any warning genetic signal.