Key words

1 Introduction

In-depth investigations on genome organization and evolution are increasing and have revealed marked contrasts between species, e.g., evolutionary rates, nucleotide composition, and gene repertoires. However, little is still known on how to link this “genomic diversity” to the diversity of life history traits or ecological forms. Synthesizing previous works in a provocative and exciting book, M. Lynch asserts that variations in fundamental population genetic processes are essential for explaining the diversity of genome architectures while emphasizing the role of the effective population size (N e) and nonadaptive processes [1]. Life history and ecological traits may influence population genetic parameters, including N e, making it possible to link species’ biology and their genomic organization and evolution (e.g., [2,3,4,5,6,7])

Among life history traits affecting population genetic processes, breeding systems are pivotal as they determine the way genes are transmitted to the next generation (Fig. 1). Outcrossing, sexual species (outcrossers) reproduce through the alternation of syngamy (from haploid to diploid) and meiosis (from diploid to haploid), with random mating of gametes from distinct individuals at each generation. Outcrossing is a common breeding system that is predominant in vertebrates, arthropods, and many plants, especially perennials, etc. [8, 9]. Selfing species (selfers) also undergo meiosis, but fertilization only occurs between gametes produced by the same hermaphrodite individual. Consequently, diploid individuals from selfing species are highly homozygous (FIS ~ 1; see, for instance, ref. 10)—heterozygosity is divided by two at each generation, and the two gene copies carried by an individual have a high probability of being identical by descent. Selfing is common in various plant families (e.g., Arabidopsis thaliana), mollusks, nematodes (e.g., Caenorhabditis elegans), and platyhelminthes, among others [8, 9]. Note that many sexual species have intermediate systems in which inbreeding and outbreeding coexist. In organisms with a prolonged haploid phase (such as mosses, ferns, or many algae and fungi), a more extreme form of selfing can occur by taking place during the haploid phase (haploid selfing or intra-gametophytic selfing), leading instantaneously to genome-wide homozygosity [11]. Clonal asexual species, finally, only reproduce via mitosis, so that daughters are genetically identical to mothers unless a mutation occurs. In diploid asexuals, homologous chromosomes associated in a given zygote do not segregate in distinct gametes—they are co-transmitted to the next generation in the absence of any haploid phase. In contrast to selfing species, individuals from asexual diploid species tend to be highly heterozygous (FIS ~ −1, [12]), since any new mutation will remain at the heterozygote stage forever, unless the same mutation occurs in the homologous chromosome. Clonality is documented in insects (e.g., aphids), crustaceans (e.g., daphnia), mollusks, vertebrates, and angiosperms, among others [13,14,15,16]. As for selfing, clonality can also be partial, with sexual reproduction occurring in addition or in alternation with asexual reproduction. In addition to this common form of asexuality, other forms such as automixis imply a modified meiosis in females where unfertilized diploid eggs produce offspring potentially diverse and distinct from their mother, leading to different levels of heterozygosity [13]. This diversity of reproductive systems should be kept in mind, but for clarity we will mainly compare outcrossing, diploid selfing, and clonality.

Fig. 1
figure 1

Reproduction and genotype transmission in outcrossing, selfing, and asexual species. In outcrossers, parental and recombinant (dotted lines) gametes from distinct zygotes are shuffled at generation n + 1. In selfers, only gametes produced by a given zygote can mate, which quickly increases homozygosity and reduces the recombination efficacy. Asexuals do not undergo meiosis or syngamy. They reproduce clonally

Full size image

Through the occurrence, or not, of syngamy, recombination, and segregation, breeding systems affect population genetic parameters (effective population size, recombination rate, efficacy of natural selection; Fig. 2) and thus, potentially, genomic patterns. A large corpus of population genetic theory has been developed to study the causes and consequences of the evolution of breeding systems (Table 1). Thanks to the exponentially growing amount of genomic data, and especially data from closely related species with contrasted breeding systems, it is now possible to test these theoretical predictions. Conversely, genomic data may help in understanding the evolution of breeding systems. Genomes should record the footprints of transitions in breeding systems and help in testing the theory of breeding system evolution in the long run, e.g., the “dead-end hypothesis,” which posits that selfers and asexuals are doomed to extinction because of their inefficient selection and low adaptive potential [17, 18]. Since the first edition of this book, several theoretical developments have clarified the population genetics consequences of the different breeding systems, and empirical evidences have been accumulating, partly changing our view of breeding system evolution and consequences, especially for asexual organisms. We first review and update the consequences of breeding systems on genome evolution and then discuss and re-evaluate how evolutionary genomics shed new light on the old question of breeding system evolution.

Fig. 2
figure 2

A schematic representation of the effect of breeding systems on population genetic parameters

Full size image
Table 1 Summary of the major theoretical predictions regarding breeding systems and evolutionary genomic variables, with outcrossing being taken as reference
Full size table

2 Contrasted Genomic Consequences of Breeding Systems

2.1 Consequences of Breeding Systems on Population Genetics Parameters

Sex involves an alternation of syngamy and meiosis. In outcrossing sexual species, random mating allows alleles to spread across populations, while segregation and recombination (here in the sense of crossing-over) associated with meiosis generate new genotypic and haplotypic combinations. This strongly contrasts with the case of selfing and asexual species. In such species, alleles cannot spread beyond the lineage they originated from because mating occurs within the same lineage (selfers) or because syngamy is suppressed (asexuals). Recombination, secondly, is not effective in non-outcrossers. In selfers, while physical recombination does occur (r 0), effective recombination (r e) is reduced because it mainly occurs between homozygous sites, and it completely vanishes under complete selfing: for tight linkage, r e = r 0(1 − F IS), where F IS is the Wright’s fixation index [19], whereas for looser linkage, effective recombination is more reduced than predicted by this simple expression [20,21,22]. In asexuals, physical recombination is suppressed (r 0 = r e = 0). High levels of linkage disequilibrium (nonrandom association of alleles between loci) could therefore be expected in selfers and asexuals. The observed data are mainly consistent with these predictions. In the selfing model species Arabidopsis thaliana, LD extends over a few hundreds of kb, while in maize, an outcrosser, LD quickly vanishes beyond a few kb [23]. In a meta-analysis, Glémin et al. [24] also found higher LD levels in selfers than in outcrossers. Beyond pairwise LD, selfing also generates higher-order associations, such as identity disequilibria (the excess probability of being homozygote at several loci, [25]) that alter population genetics functioning compared to outcrossing populations (e.g., [26]).

Theory also predicts that the effective population size, N e, depends on the breeding system (Fig. 2). First, compared to outcrossers, selfing is expected to directly lower N e by a factor 1 + F IS by reducing the number of independent gametes sampled for reproduction [27]. From a coalescent point of view, selfing reduces coalescent time (again by the same factor 1 + F IS). Under outcrossing, two gene copies gathered in a same individual either directly coalesce or move apart at the preceding generation. Selfing prolongs the time spent within an individual, hence the probability of coalescing [19, 28]. In diploid asexuals, the picture is less obvious. Since genotypes, not alleles, are sampled, Balloux et al. [12] distinguished between the genotypic and allelic effective size. The genotypic effective size equals N, not 2N, i.e., the actual population size, similarly to the expectation under complete selfing. On the contrary, the allelic effective size tends toward infinity under complete clonality because genetic diversity within individuals cannot be lost [12]. This corresponds to preventing coalescence as long as gene copies are transmitted clonally [29, 30]. However, very low level of sex (higher than 1/2N) is sufficient to retrieve standard outcrossing coalescent behavior [29, 30], and as far as natural selection is concerned (see below), the genotypic effective size is what matters [31]. The ecology of selfers and asexuals may also contribute to decreasing N e as they supposedly experience more severe bottlenecks than outcrossers [32, 33]. On the contrary, higher population subdivision in selfers could contribute to increasing N e at the species scale. However, Ingvarsson [34] showed that, under most conditions, the extinction/recolonization dynamics is predicted to decrease N e in selfers, at both the local and metapopulation scale. Finally, because of low or null effective recombination, hitchhiking effects—the indirect effects of selection at a locus on other linked loci—reduce N e further [35]. Under complete selfing or clonality, because of full genetic linkage, selection at a given locus affects the whole genome. Most forms of selection, and especially directional selection, reduce the number of gene copies contributing to the next generation by removing deleterious alleles to the benefit of advantageous ones. Because of linkage, such a reduction spreads over the rest of the genome, globally reducing the effective population size (sensu lato) in non-outcrossing species. Background selection, the reduction in N e due to the removal of deleterious mutations at linked loci, can be particularly severe in highly selfing and clonal population, potentially reducing N e by one order of magnitude or more [22, 36]. And this effect is expected to be stronger in asexuals than in selfers [36]. In the predominantly selfing nematode C. elegans, nucleotide diversity has been shown to be reduced genome wide by both background selection [37] and selective sweeps [38], and in a comparative analysis, the effect of linked selection has shown to be more pronounced in selfing than in outcrossing species [39].

As genetic diversity scales positively with N e μ, where μ is the mutation rate, selfers are expected to be less polymorphic than outcrossers. Asexuals should also exhibit lower genotypic diversity, but the prediction is not clear for allelic diversity (see above). However, because of the lack of recombination, haplotype diversity should be lower for both breeding systems. The effect of selfing on the polymorphism level is well documented, and empirical data mainly agree with the theoretical predictions. Selfing species tend to be more structured, less diverse, and straightforwardly more homozygotes than outcrossers [6, 24, 40, 41]. Much fewer data exist regarding diversity levels in asexuals, but the available datasets confirm that genotypic diversity, at least, is usually low in such species (see discussion in ref. 12). At the population level, a recent comparative analysis of sexual and asexual Aptinothrips rufus grass thrips confirmed the expected lower nuclear genetic diversity of asexual populations while also evidencing that some asexuals with extensive migration can feature very high mitochondrial genetic diversity [42].

These predictions concerning polymorphism patterns implicitly assumed that mutation rates are the same among species with contrasted breeding systems. However, modifications in breeding systems can also affect various aspects of the species life cycle potentially related to the mutation rate. In asexuals, for instance, loss of spermatogenesis can reduce mutation rates, while loss of the dormant sexual phase can increase them (reviewed in [43]). Mutation rates can also be decreased in non-outcrossers due to the loss of recombination, which can be mutagenic [44, 45]. In selfers, meiosis and physical recombination do occur. However, the specific mutagenic process during meiosis depends on the level of heterozygosity, such as indel-associated mutations (IDAM): heterozygote indels could increase the point mutation rate at nearby nucleotides because of errors during meiosis [46, 47]. Consistent with this prediction, the IDAM process more strongly affects the outcrossing wild rice, Oryza rufipogon, than the very recent selfer and weakly heterozygous domesticated rice, O. sativa. A. thaliana, a more ancient and mostly homozygous selfer, is very weakly affected by IDAM [48]. Overall, these processes should globally contribute to lowering mutation rates, and thus polymorphism, in selfing and asexual species.

2.2 Breeding Systems and Selection Efficacy

2.2.1 Drift and Recombination: Parallel Reduction in Selection Efficacy in Selfers and Asexuals?

The effective population size strongly affects the outcome of natural selection. The probability of fixation of a new mutation is a function of the N e s product, where s is the selection coefficient ([49] and see Fig. 3). As N e is reduced, a higher proportion of mutations behave almost neutrally. Weakly deleterious alleles can thus be fixed, while weakly advantageous ones can be lost. Genetic associations among loci generated by selfing and clonality also induce selective interferences [26, 50]. Because of their reduced effective population size and recombination rate, selection is thus expected to be globally less effective in selfers and asexuals than in outcrossers, which should result in various footprints at the molecular level (Table 1). Assuming that most mutations are deleterious (with possible back compensatory mutations), both the ratio of non-synonymous to synonymous polymorphism, πN/ πS, and the ratio of non-synonymous to synonymous substitutions, dN/dS, are predicted to be higher in selfers and asexuals than in outcrossers. Codon usage should also be less optimized in selfers and asexuals than in outcrossers.

Fig. 3
figure 3

Substitution rates relative to the neutral case (dN/dS) in outcrossers (thin lines), selfers (bold line), and asexuals (dotted lines) for different mutation dominance levels. The fitness of the resident, heterozygote, and homozygote mutant genotypes are 1, 1 − hs, and 1 − s, respectively. For asexuals, it is necessary to consider two substitution rates corresponding to the initial fixation of heterozygotes and the ultimate fixation of complete homozygote mutants from an initially heterozygote population [31]. Population size: N = 10,000. To highlight the difference between selfers and asexuals due to segregation, demographic and hitchhiking effects reducing N e in asexuals and selfers are not taken into account

Full size image

Contrary to polymorphism surveys, few studies have tested these predictions empirically (Table 2). In the few available comparative studies, contrasted patterns were observed between selfers and asexuals. Compared to sexual ancestors, recent asexual lineages show a marked increase in the dN/dS ratio in Daphnia ([51] but see below), Timema stick insects [52], gastropods Campeloma [53] and Potamopyrgus [54], and the plant Boechera [55], in agreement with theoretical predictions (Table 2). However, no significant effect of asexuality on dN/dS was found in four aphid species [56] and in the plant Ranunculus auricomus [57]. Bdelloid rotifers, long considered as ancient asexuals (see below), exhibit a higher πN/πS ratio but not a higher dN/dS ratio than comparable sexual groups, suggesting that mildly deleterious mutations can segregate at a higher frequency in asexuals but are eventually removed. A higher πN/πS ratio in asexual lineages than in sexual relatives was reported from transcriptome data in Oenothera primroses [58] and Lineus nemerteans [59]. Note however that in the latter case, the increased πN/πS is primarily explained by the hybrid nature of the asexual Lineus pseudolacteus (Table 2). The recent origin of asexuality through introgression also challenges the interpretation of elevated dN/dS ratio in the mitochondrial genome of asexual lineages of Daphnia pulex [51], as less than 1% of mutations on the branches leading to asexual lineages would have arisen after the transition to asexuality [60]. Here, rather than being the direct cause of genomic degradation, asexuality may have evolved in already-degraded lineages.

Table 2 Summary of studies comparing patterns of molecular evolution between different breeding systems
Full size table

All predictions are not equally supported by data in selfers. Polymorphism-based measures mostly support reduction in selection efficiency in selfers in various plant species, and this was recently confirmed by a meta-analysis of genome-wide polymorphism data ([6] and see Table 2). On the contrary, as far as dN/dS or base composition are compared, most studies, in plants, fungi, and animals, did not find evidence of relaxed selection in selfers (Table 2). A recent origin of selfing is often invoked to explain that effect of selfing is rarely observed in species divergence (e.g., [61, 62,63,64]), whereas a recent transition to selfing can leave a clear signature of relaxed selection at the polymorphism level [65]. In contrast, in the freshwater snail Galba truncatula where selfing is supposed to be old and ancestral to a clade of several species, relaxed selection in the selfing lineage was also observed at the divergence level [66]. The same rationale should apply to asexual species. However, in Campeloma, Potamopyrgus, Timema, and Boechera, clonality is also recent, yet the expected patterns are observed at the divergence level. The reduction in N e could simply be less severe in selfers than in asexuals as predicted by background selection models [36]. Furthermore, complete selfing is hardly ever noted in natural populations; residual outcrossing typically occurs. Among hitchhiking effects, some are very sensitive to the recombination level, such as Muller’s ratchet [67], weak Hill-Robertson interferences [50], or hitchhiking of deleterious mutations during selective sweeps [68, 69]. If such mechanisms are the main cause of reduction of N e in selfers, then even a low recombination rate could be enough to maintain the selection efficacy. This is suggested by genomic patterns across recombination gradients in outcrossing species. In primates, no effect of recombination on the selection efficacy has been detected [70]. In Drosophila, Haddrill et al. [71] found little evidence of reduced selection in low recombining regions, except when recombination was fully suppressed, as in Y chromosomes. Differences between selfers and asexuals could thus simply result from different degrees of residual outcrossing. However, as stated above, selfers and asexuals also fundamentally differ as far as segregation is concerned, as we now discuss in more detail.

2.2.2 Segregation: Dealing with Heterozygotes

Selfing affects the selection efficacy by increasing homozygosity and thus exposing recessive alleles to selection. This effect can counteract the effect of reducing N e. Considering the sole reduction in N e due to non-independent gamete sampling, selection is less efficient under partial selfing for dominant mutations but more efficient for recessive ones (Fig. 3, and see ref. 72). More precisely, Glémin [73] determined the additional reduction in N e (due to hitchhiking and demographic effects) necessary to overcome the increased selection efficacy due to homozygosity. This additional reduction can be high for recessive mutations. On the contrary, the lack of segregation in asexuals reduces selection efficacy and increases the drift load, as heterozygotes can fix [31]. The effects of selfing and clonality on the fixation probability of codominant, recessive, or dominant mutations are summarized in Fig. 3. Note that segregation may also have indirect effects. When recombination is suppressed, Muller’s ratchet is supposed to reduce N e and contribute to the fixation of weakly deleterious alleles [74]. In selfers, the purging of partially recessive deleterious alleles slows down the ratchet [67], which suggests that the fixation of deleterious alleles at linked loci would be lower in selfers than in asexuals. The same mechanism also contributes to weaker background selection in selfers than in asexuals (see above, [36]). In the extreme case of intra-gametophytic selfing, purging could be even more efficient at removing deleterious alleles [11], as it has been suggested for moss species [75]. Segregation at meiosis could thus partly explain the differences between selfers and asexuals, but more data are clearly needed to confirm this hypothesis.

The two opposite effects of drift and segregation in selfers should also affect adaptive evolution. In outcrossers, new beneficial mutations are more likely to be rapidly lost if recessive, as they are initially present in heterozygotes and masked to selection—a process known as Haldane’s sieve [76]. By unmasking these mutations in homozygotes, selfing could help adaptive evolution from recessive mutations [72, 73]. However, this advantage of selfing disappears when adaptation proceeds from pre-existing variation because homozygotes can also be present in outcrossers [77]. Selective interference in selfers also reduces their advantage of not experiencing Haldane’s sieve, especially for weakly beneficial mutations [21], and the effect of background should globally reduce the rate of adaptation [73, 77, 78]. Conversely, the lack of segregation in asexuals delays the complete fixation of an advantageous mutation. Once a new advantageous mutation gets fixed in the heterozygotic state, additional lag time until occurrence and fixation of a second mutation is necessary to ensure fixation [79]. Little is known about the dominance levels of new adaptive mutations, but a survey of QTL fixed during the domestication process in several plant species confirmed the absence of Haldane’s sieve in selfers compared to outcrossers [80]. This mostly corresponds to strong selection on new mutations or mutations in low initial frequencies in the wild populations. More generally, the effect of selfing on adaptive evolution will depend on the distribution of dominance and selective effects of mutations and the magnitude of genetic drift and linkage.

Few studies have tested for difference in positive selection between selfers and outcrossers. In their survey of sequence polymorphism data in flowering plants, Glémin et al. [24] found, on average, more genes with a signature of positive selection in outcrossers than in selfers assessed by the McDonald-Kreitman test [81]. An extension of this method—where non-synonymous vs. synonymous polymorphism data are used to calibrate the distribution of the deleterious effects of mutations and then attribute the excess non-synonymous divergence observed to positive selection [82]—was applied to one plant [83] and one freshwater snail dataset. In both studies, a large fraction of non-synonymous substitutions was estimated to be adaptive in the outcrossing species (~40% in the plant Capsella grandiflora and ~55% in the snail Physa acuta), whereas this proportion was not significantly different from zero in the selfer (Arabidopsis thaliana and Galba truncatula, respectively). Based on methods where the dN/dS ratio is allowed to vary both among branches and sites, a comparative analysis of two outcrossing and two selfing Triticeae species [84] suggested that adaptive substitutions may have specifically occurred in the outcrossing lineages. This would contribute to explaining why selfing lineages did not show a higher dN/dS ratio than outcrossing ones (see above and Table 2). So the data available so far support an increased rate of adaptation in outcrossing species, suggesting that the effects of drift and linkage overwhelm the advantage of avoiding Haldane’s sieve. A similar approach was used in Oenothera species suggesting also reduced adaptive evolution in clonal compared to sexual lineages [85].

Finally, the classical assumption of a lack of segregation in asexuals must be modulated. First, in some form of asexuality, such as automixis, female meiosis is retained, and diploidy restoration occurs by fusion or duplication of female gametes. Depending on how meiosis is altered, automixis generates a mix of highly heterozygous and highly homozygous regions along chromosomes. The genomes of such species could thus exhibit a gradient of signatures of selfing and diploid clonal evolution [86]. Secondly, mitotic recombination and gene conversion in the germline of asexual lineages can also reduce heterozygosity at a local genomic scale. Mitotic recombination has been well documented in yeast (see review in ref. 87) and also occurs in the asexual trypanosome T. b. gambiense [88] and in asexual Daphnia lineages [60, 89, 90]. If its frequency is of the order or higher than mutation rates, as reported in yeast and Daphnia, asexuals would not suffer much from the lack of segregation at meiosis. Especially, during adaptation, the lag time between the appearance of a first beneficial mutation and the final fixation of a mutant homozygote could be strongly reduced [87]. However, such mechanisms of loss of heterozygosity also rapidly expose recessive deleterious alleles in heterozygotes and generate inbreeding-depression-like effects [60].

2.2.3 Selection on Genetic Systems

So far, we have only considered the immediate, mechanistic effects of breeding systems on population genetic parameters. Breeding systems, however, can also affect the evolution of genetic systems themselves, which modulates previous predictions. Theoretical arguments suggested that selfing, even at small rates, greatly increases the parameter range under which recombination is selected for [91,92,93]. These predictions have been confirmed in a meta-analysis in angiosperms in which outcrossers exhibited lower chiasmata counts per bivalent than species with mixed or selfing mating systems [94]. Higher levels of physical recombination (r 0) could thus help break down LD and reduce hitchhiking effects. This could contribute to explaining why little evidence of long-term genomic degradation has been observed in selfers, compared to asexuals.

Breeding systems may also affect selection on mutation rates. Since the vast majority of mutations are deleterious, mutation rates should tend toward zero, up to physiological costs of further reducing mutation rates being too high (e.g., [95, 96]). Under complete linkage, a modifier remains associated with its “own” mutated genome. Selection should thus favor lower mutation rates in asexuals and selfers (e.g., [95, 96]). However, Lynch recently challenged this view and suggested a lower limit to DNA repair may be set by random drift, not physiological cost [97]. Such a limit should thus be higher in asexuals and selfers. Asexuality is often associated with very efficient DNA repair systems (reviewed in [43]), supporting the view that selection for efficient repair may overwhelm drift in asexual lineages. Alternatively, only groups having high-fidelity repair mechanisms could maintain asexuality in the long run. More formal tests of mutation rate differences between breeding systems are still scarce. The phylogenetic approach revealed no difference in dS, as a proxy of the neutral mutation rate, between A. thaliana and A. lyrata [61], nor did a mutation accumulation experiment that compared the deleterious genomic mutation rate between Amsinckia species with contrasted mating systems [98]. A similar experiment in Caenorhabditis showed that the rate of mutational decay was, on average, fourfold greater in gonochoristic outcrossing taxa than in the selfer C. elegans [99]. Recent mutation accumulation experiments on Daphnia pulex suggested a slightly lower mutation rate in obligate than in facultative asexual genotypes, except for one mutator phenotype which evolved in an asexual subline [90]. Overall, these results do not support Lynch’s hypothesis of mutation rates being limited by drift in asexual and selfing species. However, such experiments are still too scarce, and quantifying how mutation rates vary or not with breeding systems is a challenging issue that requires more genomic data.

2.3 Breeding Systems and Genomic Conflicts

Outcrossing species undergo various sorts of genetic conflict. Sexual reproduction directly leads to conflicts within (e.g., for access to mating) and between sexes (e.g., for resource allocations between male and female functions or between offspring). In selfers and asexuals, such conflicts occur because mates are akin or because mating is absent [100, 101]. Outcrossers are also sensitive to epidemic selfish element proliferation and to meiotic drive, because alleles can easily spread over the population through random mating. In contrast, selfers and asexuals should be immune to such genomic conflicts because selection only occurs between selfing or asexual lineages so that selfish elements should be either lost or evolve into commensalists or mutualists [102].

2.3.1 Relaxation of Sexual Conflicts in Selfers and Asexuals

Some genes involved in sexual reproduction are known to evolve rapidly because of recurrent positive selection [103]. Arm races for mating or for resource allocation to offspring are the most likely causes of this accelerated evolution. In selfers and asexuals, selection should be specifically relaxed on these genes, not only because of low recombination and effective size but mainly because the selection pressure per se should be suppressed. According to this prediction, in the outcrosser C. grandiflora, 6 out of the 20 genes that show the strongest departure from neutrality are reproductive genes and under positive selection. This contrasts with the selfer A. thaliana, for which no reproductive genes are under positive selection [83].

More specifically, two detailed analyses provided direct evidence of relaxed selection associated with sexual conflict reduction. In the predominantly selfer C. elegans, some males deposit a copulatory plug that prevents multiple matings. However, other males do not deposit this plug. A single gene (plg-1), which encodes a major structural component of this plug, is responsible for this dimorphic reproductive trait [104]. Loss of the copulatory plug is caused by the insertion of a retrotransposon into an exon of plg-1. This same allele is present in many populations worldwide, suggesting a single origin. The strong reduction in male-male competition following hermaphroditism and selfing evolution explains that no selective force opposes the spread of this loss-of-function allele [104, 105]. In A. thaliana, similar relaxed selection has been documented in the MEDEA gene, an imprinted gene directly involved in the male vs. female conflict. MEDEA is expressed before fertilization in the embryo sac and after fertilization in the embryo and the endosperm, a tissue involved in nutrient transfer to the embryo. In A. lyrata, an outcrossing relative to A. thaliana, MEDEA could be under positive [106] or balancing selection [107], in agreement with permanent conflicting pressures for resource acquisition into embryos between males and females. Conversely, this gene evolved under purifying selection in A. thaliana, where the level of conflict is reduced.

Male vs. female diverging interests are also reflected by cyto-nuclear conflicts. When cytoplasmic inheritance is uniparental, as in most species, cytoplasmic male sterility (CMS) alleles favoring transmission via females at the expense of males can spread in hermaphroditic outbreeding species, leaving room for coevolution with nuclear restorers. Maintenance of CMS/non-CMS polymorphism leads to stable gynodioecy [108]. In selfers, CMS mutants also reduce female fitness—because ovules cannot be fertilized—and are thus selected against. In the genus Silene, the mitochondrial genome of gynodioecious species exhibits molecular signatures of adaptive and/or balancing selection. This is likely due to cyto-nuclear conflicts as this is not, or is less, observed in hermaphrodites and dioecious [109,110,111]. Although less studied, cyto-nuclear conflicts are also expected in purely hermaphroditic species. In a recent study in A. lyrata, Foxe and Wright [112] found evidence of diversifying selection on members of a nuclear gene family encoding transcriptional regulators of cytoplasmic genes. Some of them show sequence similarity with CMS restorers in rice. Given the putative function of these genes, such selection could be due to ongoing cyto-nuclear coevolution. Interestingly, in A. thaliana, these genes do not seem to evolve under similar diversifying selection, as expected in a selfing species where conflicts are reduced.

2.3.2 Biased Gene Conversion as a Meiotic Drive Process: Consequences for Nucleotide Landscape and Protein Evolution

GC-biased gene conversion (gBGC) is a kind of meiotic drive at the base pair scale that can also be strongly influenced by breeding systems. In many species, gene conversion occurring during double-strand break recombination repair is biased toward G and C alleles (reviewed in [113]). This process mimics selection and can rapidly increase the GC content, especially around recombination hotspots [114, 115], and, more broadly, can affect genome-wide nucleotide landscapes. For instance, it is thought to be the main force that shaped the isochore structure of mammals and birds [116]. gBGC has been mostly studied by comparing genomic regions with different rates of (crossing-over) recombination (reviewed in [116]). However, comparing species with contrasted breeding systems offers a broader and unique opportunity to study gBGC. gBGC cannot occur in asexuals because recombination is lacking. Selfing is also expected to reduce the gBGC efficacy because meiotic drive does not occur in homozygotes [117]. To our knowledge, GC content has never been compared between sexual and asexual taxa, but there have been comparisons between outcrossers and selfers.

As expected, no relationship was found between local recombination rates and GC-content in the highly selfing Arabidopsis thaliana [117], and Wright et al. [118] suggested that the (weak) differences observed with the outcrossing A. lyrata and Brassica oleracea could be due to gBGC. Much stronger evidence has been obtained in grasses. Grasses are known to exhibit unusual genomic base composition compared to other plants, being richer and more heterogeneous in GC-content [119], and direct and indirect evidences of gBGC have been accumulating [119, 120,121,122]. Accordingly, GC-content or equilibrium GC values were found to be higher in outcrossing than in selfing species [24, 84, 120]. Difference in gBGC between outcrossing and selfing lineages has also been found in the plant genus Collinsia [123] and in freshwater snails [66], although difference in selection on codon usage cannot be completely ruled out.

gBGC can also affect functional sequence evolution, leaving a spurious signature of positive selection and increasing the mutation load through the fixation of weakly deleterious AT→GC mutations: gBGC would represent a genomic Achilles’ heel [124]. Once again, comparing outcrossing and selfing species is useful for detecting interference between gBGC and selection. gBGC is expected to counteract selection in outcrossing species only. The Achilles’ heel hypothesis could explain why relaxed selection was not detected in four grass species belonging to the Triticeae tribe [84]. In outcrossing species, but not in selfing ones, dN/dS was found to be significantly higher for genes exhibiting high than low equilibrium GC-content, suggesting that selection efficacy could be reduced because of high substitution rates in favor of GC alleles in these outcrossing grasses. In outcrossing species, gBGC can maintain recessive deleterious mutations for a long time at intermediate frequency, in a similar way to overdominance [125]. This could generate high inbreeding depression in outcrossing species, preventing the transition to selfing. In reverse, recurrent selfing would reduce the load through both purging and the avoidance of gBGC, thus reducing the deleterious effects of inbreeding. Under this scenario, gBGC would reinforce disruptive selection on mating systems. In the long term, gBGC could be a new cost of outcrossing: because of gBGC, not drift, outcrossing species could also accumulate weakly deleterious mutations, to an extent which could be substantial given current estimates of gBGC and deleterious mutation parameters [125]. Whether this gBGC-induced load could be higher than the drift load experienced by selfing species remains highly speculative. Both theoretical works, to refine predictions, and empirical data, to quantify the strength of gBGC and its impact on functional genomic regions, are needed in the future. Grasses are clearly an ideal model for investigating these issues, but comparisons with groups having lower levels of gBGC would also be helpful.

2.3.3 Transposable Elements in Selfers and Asexuals: Purging or Accumulation?

Considering the role of sex in the spread of selfish elements, TEs should be less frequent in selfers and asexuals than in outcrossers because they cannot spread from one genomic background to another through syngamy. However, highly selfing and asexual species derive from sexual outcrossing ancestors, from which they inherit their load of TEs. TE distribution eventually depends on the balance between additional transposition within selfing/clonal lineages on one hand and selection or excision on the other. Following the abandonment of sex, large asexual populations are expected to purge their load of TEs, provided excision occurs, even at very low rates. However, purging can take a very long time, and, without excision, TEs should slowly accumulate, not decline [126]. In small populations, even with excision, a Muller’s ratchet-like process drives TE accumulation throughout the genome [126]. Transition from outcrossing to selfing should also rapidly purge TEs, but as for asexuals, in small fully selfing populations, TEs can be retained [127]. Using yeast populations, it was experimentally confirmed that sex increases the spread of TEs [128, 129]. TE numbers were also found to be higher in cyclically sexual than in fully asexual populations of Daphnia pulex [130,131,132] (Table 3), contrary to what was described in the parasitoid wasp Leptopilina clavipes and in root knot nematode species (Table 3). It should be noted that several comparative studies on asexual arthropods, nematodes, primroses, and green algae did not evidence any significant effect of breeding system on TE content or evolution (Table 3). At larger evolutionary scales, the putatively ancient asexual bdelloid rotifers strikingly exemplify the fact that asexuals can purge their load of TEs. Unlike all sexual eukaryotes, they appear to be free of vertically transmitted retrotransposon, while their genome contains DNA transposons, probably acquired via horizontal transfers [133, 134]. Examples of TE accumulation in asexuals are less common, maybe because species are doomed to extinction under this evolutionary scenario [135]. However, the increase in genome size in some apomictic lineages of Hypericum species may result from this process [136].

Table 3 Summary of studies comparing transposable element distribution and dynamics between different breeding systems
Full size table

In selfers, the distribution of TEs depends not only on the population size but also on the mode of selection against TEs [127, 137]. Under the “deleterious” model, TE insertions are selected against because they disrupt gene functions. According to the “ectopic exchange” model, TEs are selected against because they generate chromosomal rearrangements through unequal crossing-over between TE at nonhomologous insertion sites. Under the first of these two models, homozygosity resulting from selfing increases the selection efficacy against TEs, while under the second one, under-dominant chromosomal rearrangements are less selected against in selfing than in outcrossing populations [127, 137]. A survey of Ty1-copia-like elements in plants suggests that they are less abundant in self-fertilizing than in outcrossing plants, thus supporting the “deleterious” rather than the “ectopic” exchange model [127]. The distribution of retrotransposons in self-incompatible and self-compatible Solanum species also supports the “deleterious” model, even though most insertions are probably neutral [138] (Table 3). In the selfer Arabidopsis thaliana, selection efficacy against TEs seems to be reduced compared to its outcrossing sister species A. lyrata [139, 140], but comparison of the two complete genomes revealed a higher load of TE in A. lyrata and a recent decrease in TE in number in A. thaliana, in agreement with the date of transition to selfing [141]. In the Capsella genus, while the very recent selfer C. rubella possesses a slightly higher number of TEs than the outcrossing C. grandiflora, the oldest selfer C. orientalis exhibits a significantly reduced load of TE [142] (Table 3). Other selfish elements, such as B chromosomes, are also less frequent in selfers, in support of the view that inbreeding generally prevents selfish element transmission [102].

2.4 Breeding Systems, Ploidy, and Hybridization

Atypical breeding systems are often associated with polyploidy [143], and the reasons for this association are not entirely clear. Polyploid mutants might be more likely to establish as new lineages in selfers and asexuals than in obligate outcrossers if crosses between polyploids and diploids are unfertile or counterselected. This is because at low population frequency a polyploid mutant will experience the disadvantage of mostly mating with diploids—the minority cytotype exclusion principle [144, 145]—unless it reproduces asexually or via selfing. In addition, by doubling gene copy number, polyploidy might alleviate the fitness cost of recessive deleterious mutations being exposed at homozygous state in selfers [146]. Kreiner et al. [147] reported that in Brassicaceae the rate of production of unreduced gametes is higher in asexuals than in outcrossers, suggesting that mating systems can influence not only the establishment rate but also the mutation rate to polyploidy.

Recent genome-wide data analyses have revealed that a number of polyploid selfers or asexuals actually correspond to allopolyploids (e.g., [59, 148,149,150,151]), highlighting the possibility that hybridization plays a role in breeding system and ploidy evolution. Hybridization between facultative asexuals might cause immediate transition to obligate asexuality if the two progenitor genomes are so divergent that meiosis is impaired—e.g., due to chromosomal rearrangements, or in case of genetic incompatibilities affecting genes involved in sexual reproduction [16]. Numerous selfing or asexual lineages, either diploid or polyploid, are known to be of hybrid origin (e.g., [13, 152,153,154,155,156,157]). Hybridization would therefore appear as a potential cause, and polyploidy a potential consequence, of atypical breeding systems [16], but more genome-wide data are obviously needed to draw firm conclusions on these complex relationships.

2.5 Breeding Systems and Genome Size Evolution

As argued above, breeding systems can affect many aspects of genome content and organization. They should also affect the whole genome size. Following Lynch’s theory [1], genome size should be higher in selfers and asexuals because of their reduced effective population size, hence reduced ability to get rid of useless, slightly costly sequences. However, the picture is probably more complex. First, because of the recent origin of many selfing and (at least some) asexual lineages, relaxed selection may not have operated longly enough to impact genome size. Second, because of their immunity to selfish element transmission, selfers and asexuals should exhibit lower genome size, especially in groups where TEs are major determinants of genome size. Hence, it is not clear whether genetic drift or resistance to selfish elements (or other processes) is the most important in governing genome size evolution in various breeding systems.

Meta-analyses performed in plants provided equivocal answers. Analysis of the distribution of B chromosomes showed a strong and significant positive association between outcrossing, the occurrence of B chromosomes, and genome size [102, 158]. However, after phylogenetic control, only the association between breeding systems and B chromosomes remains. Whitney et al. [159] simultaneously tested the effect of breeding systems (using outcrossing rate estimates) and genetic drift (using polymorphism data) on genome size in seed plants. Raw data showed a significant effect of both breeding systems and genetic drift, according to theoretical predictions. However, no effect was observed after phylogenetic control, leading the authors to reconsider the hypothesis of a role of nonadaptive processes in genome size evolution. Similarly, phylogenetic comparative analysis of 30 primrose species (Oenothera) covering several transitions to asexuality showed no significant relationship between reproductive mode and genome size [160].

Because breeding systems can evolve quickly, more detailed analyses at a short phylogenetic scale are needed to get a clearer picture of their effects on genome size evolution. Moreover, breeding systems are often correlated with other life history traits, such as lifespan, which can make it hard to clarify the causes and consequences of the observed correlations. A detailed analysis of genome size in the Veronica genus suggests that selfing, not annuality, is associated with genome size reduction [161]. A comparison of 14 pairs of plant congeneric species with contrasted mating systems also suggested a genome size reduction in selfers [162]. However, this could partly have been due to the four polyploid selfing species of the dataset—polyploidy can lead to haploid genome size reduction because of the loss of redundant DNA following polyploidization. A better understanding can be gained from the comparative analysis of genome composition and organization, not only genome size. In Caenorhabditis nematodes, the observed reduction in genome size is not driven by reduction in TEs but by a global loss of all genomic compartments [163]. This pattern contradicts the hypothesis of relaxed selection in selfers against the accumulation of deleterious genomic elements. Alternatively, it could be explained by deletion bias and high genetic drift in selfers. However, in mutation accumulation lines, insertions predominate over deletion in the selfing C. elegans, and deletions occurred at the whole gene level instead of being at random among genomic compartments, as predicted under a general deletion bias (see discussion in ref. 163). In this genus, Lynch’s hypothesis that evolution of genome size should be driven by changes in N e does not apply. Alternatively, the authors suggested that it is a more direct consequence or even an adaptation to the selfing lifestyle, although the underlying mechanisms still remain unclear.

3 A Genomic View of Breeding System Evolution

Because breeding systems can strongly affect genome structure and evolution, conversely, genomic approaches offer new powerful tools to reconstruct breeding system evolution and to test evolutionary hypotheses, especially concerning long-term evolution.

3.1 Genomic Approaches to Infer Breeding System Evolution

3.1.1 Genomic Characterization of Breeding Systems

Genetic markers have long been used to determine breeding systems and quantify selfing rates or degrees of asexuality. For instance, current selfing rates can be inferred using molecular markers through F IS estimates or preferably—although more time consuming—through progeny analyses [164,165,166]. Multilocus-based estimates that take identity disequilibrium into account greatly improve the simple F IS-based method that is sensitive to several artifacts such as null alleles ([167], see also refs. 168, 169). This method, implemented in the RMES software [167], has proven to give results very similar to progeny-based methods [170]. To take advantage of the information potentially available in sequence data, coalescence-based estimators have also been proposed to infer long-term selfing rates, and they have been implemented more recently in a Bayesian clustering approach in the INSTRUCT software package [171]. However, this approach mostly captures information from recent coalescence events so that such approaches still estimate recent selfing rates [28]. Much more information about long-term selfing rates can be derived from LD patterns [19], but this has not been fully exploited for selfing rate estimators (for instance, LD is not taken into account in INSTRUCT). Similarly, recombination can be inferred using genetic markers or sequence data, and more generally, various methods have been proposed to characterize the degree of clonality in natural populations (for review see ref. 172) and recently implemented in the R package RClone [173].

Initially, such methods were applied with few markers, from which only global descriptions of breeding systems were deducible. Thanks to the considerable increase in sequencing facilities, it has become possible to finely characterize temporal and spatial variations in breeding systems. In A. thaliana, an analysis of more than 1000 individuals in 77 local stands using more than 400 SNP markers revealed spatial heterogeneity in outcrossing rates. Local “hotspots” of recent outcrossing (up to 15%) were identified, while other stands exhibited complete homozygosity with no detectable outcrossing [174]. Interestingly, at this local scale (from 30 m to 40 km), outcrossing rates have been found to be twofold higher on average in rural than in urban stands; hence, selfing could be associated with higher disturbance in urban stands.

Genomic data may also help characterize breeding systems in species with unknown or ill-characterized life cycles. In yeasts Saccharomyces cerevisiae and S. paradoxus, the analyses of linkage disequilibrium patterns allowed to quantify the frequency of (rare) sexual reproduction events and the proportion of inbreeding and outcrossing during these events [175, 176]. For instance, in the pico-algae Ostreococcus, no sexual form or process has been detected in the lab. However, the occurrence of infrequent recombination (about 1 meiosis for 10 mitoses) inferred from a population genomics approach and the presence of meiosis genes in the genome support the existence of a sexual life cycle [177]. Moreover, a strong negative correlation between chromosome size and GC-content has been observed [178]. In mammals and birds (among others), such a pattern has been interpreted as a long-term effect of gBGC acting on chromosomes with different average recombination rates [116]—small chromosomes having higher recombination rates because of the constraint of at least one chiasmata per chromosome arm. A similar interpretation for Ostreococcus is thus appealing. Genomic data also allow to test whether the theoretical signatures of long-term asexuality are observed in putative asexuals. As an example, whole-genome analyses of the trypanosome T. b. gambiense demonstrated an independent evolution and divergence of alleles on each homologous chromosome (the “Meselson effect” [179, 180]), which is indicative of strict asexual evolution [88]. In contrast, genomic studies of the putatively ancient asexual bdelloids recently uncovered the occurrence of inter-individual genetic exchanges ([181, 182] see below Subheading 3.2.2).

3.1.2 Inferring and Dating Breeding System Transitions

Genomic approaches are also useful for analyzing the dynamics of breeding system evolution. A simple way is to map breeding system evolution on phylogenies, which could provide a raw picture of the frequency and relative timing of breeding system transitions (e.g., [183]). However, these approaches, based on ancestral character reconstruction, are hampered by numerous uncertainties. For instance, in the case of two sister species with contrasting breeding systems, such as A. thaliana and A. lyrata, it is impossible to know whether A. thaliana evolved toward selfing just after divergence (about five million years ago) or only very recently. At a larger phylogenetic scale, inferring rates of transition between characters and ancestral states can be biased if diversification rates differ between characters—this is typically expected with breeding systems for which asexuals and selfers should exhibit higher extinction rates than outcrossers [184].

Thanks to the genomic signatures left by contrasted breeding systems, it is possible to trace back transitions in the past and to date them more precisely. In diploid asexual species, because of the arrest of recombination, the two copies of each gene have diverged independently since the origin of asexuality. After having calibrated the molecular clock, it is thus possible to date this origin from the level of sequence divergence between the two copies. This so-called Meselson effect was observed and quantified in the trypanosome T.b. gambiense, suggesting that this species evolved asexually about 10,000 years ago [88]. However, no Meselson effect has been observed in other presumably ancient asexual species such as oribatid mites [185] or darwinulid ostracods [186], while data refute the possibility of cryptic sex. In such cases, it is thus not possible to infer when recombination actually stopped, presumably because of homogenizing processes such as very efficient DNA repair or automixis. Mitotic recombination could also obscure the pattern predicted under this Meselson effect. Of note, when asexuality originates by hybridization (see above Subheading 2.4), the last common ancestor of the two copies of a gene dates back to the ancestor of the two parental lineages, which can be much older than the hybridization date, faulting the above-described rationale.

Past transitions from outcrossing to selfing have also been investigated, through either population genomics approaches or the evolutionary analysis of self-incompatibility (SI) genes, which are directly involved in the transition to selfing. Since the evolution of selfing requires the breakdown of SI systems, initially constrained S-locus genes are expected to evolve neutrally after a shift to selfing. In A. thaliana, Bechsgaard et al. [187] reasoned that the dN/dS ratio in the selfing lineage should be the average of the neutral dN/dS (i.e., 1) and the outcrossing dN/dS—inferred from sister lineages—weighted by the time spent in the selfing vs. the outcrossing state. They deduced that SRK, one of the major SI genes, became a pseudogene less than 400,000 years ago. SRK, however, is not the only gene involved in SI. Mutations in other genes may have previously disrupted the SI system, thus confusing SRK-based dating. Indeed, coalescence simulations showed that the observed genome-wide pattern of linkage disequilibrium is compatible with the transition to selfing one million years ago [188], suggesting a possible but debated two-step scenario in the evolution of selfing [189, 190]. The persistence of three distinct divergent SRK haplotypes among extant A. thaliana individuals also suggests multiple loss of SI [191], but the recent discovery of the co-occurrence of the three haplotypes in Moroccan populations makes possible the evolution of selfing in a single geographic region [192]. In another Brassicaceae, i.e., Capsella rubella, analyses of both S-locus and genome-wide genes coupled with coalescence simulations suggested that selfing evolved very recently from the outcrosser C. grandiflora, around 50,000 years ago [193, 194] from a potentially large number of founding individuals followed by a strong reduction in N e [195]. In the tetraploid selfer Arabidopsis suecica, which originated as a hybrid between A. thaliana and the outcrossing A. arenosa, the genomic analysis of the S-locus also revealed the origin of selfing, suggesting an instantaneous loss of SI due to the fixation of nonfunctional alleles from both parents around 16,000 years ago [150].

3.2 Matching Breeding System Evolution Theories with Genomic Data

3.2.1 Testing the Dead-End Hypothesis: Comparison Between Selfing and Asexuality

The expected reduction in N e in selfers and asexuals may increase the drift load (accumulation of slightly deleterious mutations) and preclude adaptation. Selfing and clonality are thus supposed to be evolutionary dead ends [17, 18]. The twiggy phylogenetic distributions of asexuals [196] and selfers [183] or self-compatible species [197] suggest they are mostly derived recently from outcrossing ancestors (but see ref. 198). However, this observation may not be sufficient to support the dead-end hypothesis, and neutral models can also explain this pattern [199,200,201]. In a comprehensive and epochal phylogenetic study of several Solanaceae genera, Goldberg et al. [202] went further by testing the irreversibility of transitions. Using a phylogenetic method developed for estimating the character effect on speciation and extinction [203, 204], they showed that self-compatible species have both higher speciation and extinction rates—with the resulting net diversification rates being lower—than self-incompatible species. This was the first direct demonstration of the dead-end hypothesis, and additional results have been obtained in Primula species [205]. On the contrary, in the Oenothera genus, asexuality has been found associated with increased diversification but frequent reversion toward the sexual system, suggesting that the form of asexuality in this group is not an evolutionary dead end [206].

Genomic data also provide an opportunity to investigate the genetic causes of such long-term evolutionary failures. The increased dN/dS ratios reported in asexuals (see above) suggest that deleterious point mutations contribute to the load. However, in Daphnia rapid exposure of recessive deleterious alleles through mitotic recombination or gene conversion likely has a much stronger effect on clone persistence than their long-term accumulation under Muller’s ratchet [60]. TE could also contribute to the load and to the extinction of asexuals [135], though more data are still needed to unambiguously support this hypothesis (but see ref. 136). The pattern in selfers is less clear. While theory globally predicts a reduction in selection efficacy in selfers, models also highlight conditions under which selection can be little affected or even enhanced in selfers [72, 73, 207], especially regarding TE accumulation [127, 137]. Empirical data on both protein and TE evolution have not revealed any strong evidence of long-term accumulation of deleterious mutation in selfers, as compared to outcrossers, whereas polymorphism data mainly support relaxation of selection in selfers (Table 2). This is in agreement with the recent origin of selfing but makes difficult further inference of the underlying causes of higher extinction in selfers as trait-dependent diversification processes alter the relationship between life history traits and rate of molecular evolution [208]. A reduced ability to respond to environmental changes through adaptive evolution could also contribute to long-term extinction in asexuals (but see ref. 209) and selfers, especially if standing variation is needed to rescue populations experiencing environmental challenges [77, 210]. Few studies, however, have compared the rate of adaptation in selfers and outcrossers (see Table 2). Theoretical predictions regarding this effect, moreover, critically depend on the dominance level of new favorable mutations [72, 73, 77, 210], which are poorly known (but see ref. 80).

While several issues remain open, current knowledge suggests that selfers are less prone to extinction than asexuals. The wider distribution of selfing than clonality in plants supports this view [211, 212]. Selfers could go toward extinction more slowly than asexuals, and the causes of their extinction could differ. Since deleterious mutations should accumulate at a slower rate in selfers than in asexuals, as suggested by theory and current data, this process would likely not be sufficient to drive them to extinction. The reduced adaptive potential could be the very cause of their ultimate extinction as initially proposed by Stebbins [18], which could generally occur before sufficient deleterious mutations have accumulated to be detected via molecular measures of divergence. On the contrary, in asexuals, the accumulation of deleterious mutations could be fast enough to leave a molecular signature and contribute to extinction. Alternatively, demographic characteristics associated with uniparental reproduction, such as recurrent bottlenecks, fragmented populations, and extinction/recolonization dynamics, could be sufficient to drive population extension simply because of higher sensitivity to demographic stochasticity (see also ref. 213). Genomic degradation would only be the witness of the evolution toward selfing and clonality without being the ultimate cause of their extinctions. These hypotheses need to be further investigated by building more realistic demo-genetic model and by better integrating genomic and ecological approaches.

The literature reviewed above focuses on intrinsic factors that may affect the extinction rate of selfing and asexual species, taken as established lineages, compared to their sexual relatives. Alternatively, Janko et al. [199] suggested that if asexual mutants are produced at a relatively high rate and compete with each other, this would imply a rapid turnover between clonal lineages and a young expected age for extant asexuals, without the need to invoke any fitness effect (see also refs. 200, 201). Of note, this model invokes competitive exclusion among clonal lineages, but not between clonal and sexual ones—the ancestral sexual gene pool is assumed to be immune from extinction.

3.2.2 Evading the “Dead End”

The few putatively ancient asexuals known so far seem to escape the mutational load predicted by the dead-end hypothesis and avoid extinction over long evolutionary time scales. For example, fossil evidence and decades of microscopic observations indicate that bdelloid rotifers have apparently persisted for over 40 million years without meiosis, males, or conventional sexual reproduction [15, 214]. As a matter of fact, the first genome assembly published for these organisms confirmed that their genome structure is incompatible with conventional meiosis [215]. However, two independent studies recently demonstrated that bdelloids could experience genetic exchanges between individuals.

A first article by Debortoli et al. [182] evidenced frequent horizontal exchanges of genetic fragments between individuals of the species Adineta vaga (Adinetidae). Such horizontal transfers could be promoted by the peculiar ecology of these rotifers, which experience frequent desiccations damaging their cell and nucleus membranes and thus allowing for the entry of foreign DNA in the cells. In addition, desiccation induces multiple DNA double-strand breaks, facilitating the integration of foreign DNA during repair processes.

Another study by Signorovitch et al. [181] identified a pattern of allele sharing between individuals of the species Macrotrachela quadricornifera (Philodinidae) that was incompatible with strict asexual evolution. The authors suggested that bdelloids had evolved an atypical meiotic mechanism similar to what has been described in some species of primroses (Oenothera), in which chromosomes organize into a ring during meiosis without requiring homologous chromosome pairing [216]. They advocated that even rare events of such unconventional sex could be enough to generate the observed pattern of allele sharing.

In the absence of conventional meiosis and syngamy, bdelloid rotifers might thus have escaped extinction by maintaining some level of genetic exchanges between individuals, either through horizontal gene transfers or unconventional Oenothera-like meiosis. Regardless of the underlying molecular mechanisms, bdelloids should not be considered as “ancient asexual scandals” anymore. These recent results call for a reassessment of the reproductive mode of all supposedly ancient asexuals (see Subheading 3.1.1 above). The rise of genomic studies in recent years will greatly contribute to decipher whether putative asexuals evolve as strict asexuals or have developed new alternatives to sex.

4 Conclusion and Prospects

There is a large body of theory on the effects of breeding systems on molecular evolution. However, some of them have not been clearly verified by empirical data, and numerous questions remain. Genomic data have also partly unveiled the complexity of breeding systems, especially in asexual or presumably asexual species. Promising prospects include (1) analysis of the rate and pattern of transition to selfing/asexuality using densely sampled phylogenies with appropriate breeding system distributions combined with genome-wide molecular data, (2) distinguishing between the different forms of selection with a better characterization of the fitness effect of mutations, (3) explicitly accounting for the possible association between breeding system shifts and non-equilibrium demographic dynamics (e.g., bottlenecks in selfers, clone turnover in asexuals). A large theoretical corpus has already been developed, and thanks to the increasing availability of genomic data, qualitative patterns are now rather well described and partly understood. Another challenge in the future is also to make our predictions and tests more quantitative.

5 Questions

  1. 1.

    What population genetic parameters are affected, and how, by selfing and asexuality?

  2. 2.

    What are the potential problems when comparing the dN/dS ratio between selfers and outcrossers or sexuals and asexuals?

  3. 3.

    What is the evolutionary “dead-end hypothesis,” and how can we test it using phylogenetic and evolutionary genomic tools?