On some genetic consequences of social structure, mating systems, dispersal, and sampling

Significance

Many species live in socially structured populations, forming cohesive units with kin structure. Yet, sociality has been neglected by population geneticists under the assumption that social groups can be seen as small demes subjected to significant genetic drift. Such demes are usually considered to be susceptible to inbreeding, with inbreeding avoidance becoming a major force explaining dispersal strategies. We find that social structure is highly effective in maintaining high genotypic and genetic diversity levels, without invoking sex-biased dispersal or inbreeding avoidance mechanisms. These findings should change the way we perceive social groups.

Abstract

Many species are spatially and socially organized, with complex social organizations and dispersal patterns that are increasingly documented. Social species typically consist of small age-structured units, where a limited number of individuals monopolize reproduction and exhibit complex mating strategies. Here, we model social groups as age-structured units and investigate the genetic consequences of social structure under distinct mating strategies commonly found in mammals. Our results show that sociality maximizes genotypic diversity, which contradicts the belief that social groups are necessarily subject to strong genetic drift and at high risk of inbreeding depression. Social structure generates an excess of genotypic diversity. This is commonly observed in ecological studies but rarely reported in population genetic studies that ignore social structure. This heterozygosity excess, when detected, is often interpreted as a consequence of inbreeding avoidance mechanisms, but we show that it can occur even in the absence of such mechanisms. Many seemly contradictory results from ecology and population genetics can be reconciled by genetic models that include the complexities of social species. We find that such discrepancies can be explained by the intrinsic properties of social groups and by the sampling strategies of real populations. In particular, the number of social groups and the nature of the individuals that compose samples (e.g., nonreproductive and reproductive individuals) are key factors in generating outbreeding signatures. Sociality is an important component of population structure that needs to be revisited by ecologists and population geneticists alike.

Sociality is one of the most striking features in the Animal Kingdom. A large number of animal species, including humans, are social. Social systems have evolved in several distinct taxa, such as insects, birds, and mammals (1–4). Whereas some animals are highly social and live in groups for their entire life, others form groups only for a short period. The diversity of social organizations ranges from eusociality in insects or communal breeding in vertebrates to solitary life in some mammalian species (5–9). Even among closely related species, this diversity can be wide (2, 10). For example, among the great apes, gorillas live in single-male harem systems, whereas chimpanzees live in large multimale/multifemale aggregations forming hierarchical communities of related males with coalition strategies (11, 12), and orangutans exhibit a semisolitary lifestyle where the territory of one male can include the territory of several related females (13).

For decades, much of the theoretical work on sociality has focused on its origins (14–18). It is generally thought that group-living confers fitness benefits to individuals (19–21). This view has had a profound impact on the interpretation of social behavior. Indeed, it is still hotly debated whether these benefits are at the origin of sociality, or if they are a side-effect of the existence of kin groups with stable bonds that appeared for other reasons (6).

Sociality has been discussed by population geneticists for several decades, but mostly from a theoretical point of view (22–26). A mathematical theory was first formalized by Chesser (23, 24), who showed that the accumulation of genetic variation within and among breeding groups can be predicted from mating and dispersal strategies. This work demonstrated that breeding groups may appear outbred despite high coancestry among the group members. However, this and other seminal studies (22, 25–27) have been neglected and sometimes misinterpreted by population geneticists and ecologists alike. For example, empirical studies documenting outbreeding within social groups (SGs) have interpreted it as a consequence of inbreeding avoidance or as evidence for sex-biased migration. Although this is in agreement with Chesser’s study on female philopatry (23), it overlooked Chesser’s later results (24) showing that social structure can lead to outbreeding even without sex-biased migration.

Ecological data on relatedness and dispersal show that social species exhibit kin recognition and sex-biased dispersal. For example, active kin recognition has been described in many cooperatively breeding passerine species (28), and in mammalian species, where SGs are generally composed of related females, males are usually the dispersing sex (1, 29). The amount of genetic data gathered from social species has also increased significantly over the last decades. Most genetic studies within social units report a higher proportion of heterozygotes than expected under random mating [i.e., evidence for heterozygote excess (outbreeding)]. This finding was described, for example, in yellow-bellied and alpine marmots (30, 31), white-tailed deers (32), and in black-tailed prairie dogs (33).

In comparison with the behavioral ecology studies that recognize social units as the basic study unit, most population genetic studies are carried out at the “population” level, where a “population” is frequently a conglomeration of SGs. The latter studies, including those on highly social species, often find no signature of deviation from random-mating, [i.e., the proportion of heterozygotes meets Hardy–Weinberg (HW) proportions, indicating neither outbreeding nor inbreeding (25)]; this is, for example, the case of orangutans (34) and sifakas (35), among others. As a consequence, population geneticists have tended to ignore sociality as an important source of genetic structure or have considered SGs simply as small random-mating units (36, 37). This approach corresponds to the classic Wright–Fisher model (WF) framework in which SGs are isolated demes with small effective sizes. This view has wide-ranging consequences. First, because of the small effective sizes it is expected that SGs should be at high risk of inbreeding depression (36, 38). Second, based on the fact that small demes are subject to high genetic drift, the Bush–Wilson theory (39, 40) postulates that the fixation of chromosomal rearrangements should be frequent and promote an acceleration of the evolution rate in social species. This theory has been criticized by behavioral ecologists who observed higher, rather than lower, levels of heterozygosity in SGs (30).

Researchers working on sociality are thus faced with the conflicting views of population geneticists, who treat SGs as demes, and behavioral ecologists, who emphasize the complexity and details of social systems and have found empirical evidence that SGs are not simple WF units (30, 41–43). This has also led to the formation of a theoretical and communication gap between the two fields, with behavioral ecologists considering themselves more in tune with the complexities of biological realism than population geneticists, who tend to reduce populations to simpler units to make them more amenable to modeling.

Because there are so many social systems and so many peculiarities in wild species, there would probably be no way to find a one-size-fits-all model that would satisfy field biologists and theoreticians alike. Here, we present a simple but general framework to investigate in detail the effects of sex-biased dispersal, philopatry, and breeding tactics on the distribution of genotypes and genetic diversity patterns within and among SGs in a population. By considering key life history parameters, we intend the framework presented here to be general enough to allow the interpretation of empirical observations from a wide array of social species.

We studied different social organizations typically found in a vast range of living species. We simulated socially structured populations under different dispersal patterns and under a variety of mating systems (monogamy, polygynandry, and polygyny), excluding any active mechanism of inbreeding avoidance. Our framework therefore provides a null-model against which theories of inbreeding avoidance, tolerance, or preference can be tested (44). We found that SGs allow species to maximize individual heterozygosity, hence minimizing inbreeding. Our results show that patterns of genetic diversity within and among SGs deviate from expectations under the WF framework. We find that the sampling design (e.g., how many groups and how many reproductive individuals are sampled) is a key and highly underestimated issue that needs to be explicitly accounted for to understand signatures of outbreeding. Altogether, we show that it is possible to reconcile apparently contradictory results by population geneticists and behavioral ecologists and bridge the gap between these two communities.

Results

We modeled populations of social animals as a fully-connected network of SGs (Fig. 1A) among which individuals can disperse. We explicitly considered SGs as age-structured (Fig. 1B) units with overlapping generations and different mating and dispersal strategies without any explicit behavior to minimize inbreeding. This approach is different from demes in classic population genetics models where demes are modeled as WF units. For all of the results presented here we simulated the evolution of populations comprising 50 SGs until equilibrium, and measured genetic diversity and inbreeding using traditional population genetics statistics.

Fig. 1.

Dispersal network and life cycle. A fully connected network of 10 SGs (A), in which each SG is a small unit where individuals undergo a life cycle (B). (A) Nodes represent SGs and edges represent dispersal pathways. The network represents a typical 10-island model according to which individuals can migrate from and to any SG in the population. Note that the figure represents 10 SGs for simplicity, but in all our simulations we considered 50 SGs instead. (B) Within a SG individuals are assumed to go through three stages: young (between birth and weaning), juveniles (between weaning and reproductive age), and adults (above reproductive age). Adults can have a nonreproductive status (non-RS) or a reproductive status (RS), which correspond to individuals that are above reproductive age but do not take part in reproduction events and individuals that actually mate at a breeding season, respectively (SI Materials and Methods). Each box represents an age-class or a RS class. Transitions along age-classes are “natural” transitions, that is, they depend only on the age of individuals (black arrows), whereas movement among reproductive-classes depend on the acquisition/loss of a RS (gray arrows). Note that the transitions from non-RS to RS class (or in the opposite direction) occur following death events and correspond to migration events in the case of the nonphilopatric sex (SI Materials and Methods).

Population Structure, Social Structure, and the Partitioning of Genetic Diversity.

We found that the subdivision of populations into SGs lead to a higher (11–63%) genetic diversity than predicted under the WF framework (Fig. 2, Fig. S1, and Table S1). This finding suggests that classic population genetics models may in many cases underestimate the genetic diversity found in real species and hence lead to biased estimates (e.g., lower effective sizes or mutation rates) when applied to social species. Moreover, individual heterozygosity within SGs (observed heterozygosity) was equal to the expected heterozygosity at the population level (H_T), suggesting that the social structure is maintaining individual diversity at the maximum possible level rather than reducing it (Fig. S2). Therefore, social structure appears to be highly efficient in maintaining diversity within individuals and within SGs.

Fig. 2.

Diversity in socially structured populations. Diversity (overall heterozygosity, H_T) in the SG model in comparison with diversity in deme-based models for similar levels of genetic differentiation. The figure shows the results for all mating systems simulated under the non–sex-biased dispersal scenario (but the same qualitative effect was obtained under female philopatry) (Fig. S1 and Table S1). Dot sizes correspond to counts. Diversity values were estimated from a random sample (among all age classes and reproductive status), but similar levels were obtained when only RS individuals were sampled. A population subdivided into SGs accumulates more genetic diversity than a comparable traditionally structured population into demes. This is evidenced by the fact that most points are located above the dashed line.

Interestingly, we found that genetic diversity was always high, whichever sampling strategy we used, whereas genetic differentiation was influenced by the sampling scheme (see below). The influence of sampling on genetic differentiation is in agreement with empirical studies reporting lower differentiation among adult males or females in comparison with offspring (45–47) (Fig. S1 and Table S1). High diversity however seems to be an intrinsic property of social structure.

Quantifying Inbreeding and Outbreeding in Social Systems.

We found that the genotype proportions deviated from HW expectations (i.e., random mating) toward an excess of heterozygotes (i.e., apparent outbreeding rather than inbreeding, F_IS < 0, where F_IS is a measure of departure from HW) (Fig. 3). This pattern was consistent across all mating systems analyzed, with or without sex-biased dispersal (Fig. 3 and Fig. S3), suggesting that social structure minimizes inbreeding without the need of any inbreeding avoidance mechanism. Inbreeding avoidance mechanisms have been shown to exist and may contribute to increase even more heterozygosity levels. Although there were quantitative differences among the distinct mating systems examined, some rather different systems produced similar ranges of F_IS values (e.g., polygynandry and extreme polygyny in Fig. 3: red and blue curves, respectively), whereas other apparently similar social systems would generate noticeably different (although overlapping) ranges of values. Overall, the consistent excess of heterozygotes within SGs was in agreement with the range of values observed in many natural species (Fig. 3, Upper) and suggests that social structure is unlikely to lead to inbreeding even in the absence of any active inbreeding avoidance mechanism.

Fig. 3.

Quantifying inbreeding in socially structured populations. Inbreeding coefficient (F_IS) under the four different types of social structure simulated. (Upper) Real data extracted from the literature on yellow-bellied marmot (30), white-toothed shrew (36), soay-sheep (98), red-howler monkey (45), fuit bat (46), and white sifaka (47) (see Table S3 for details). (Lower) The simulated datasets under four types of mating system. Black, red, blue, and green correspond to monogamy, polygynandry, extreme polygyny, and moderate polygyny, respectively. The solid and dashed lines represent female philopatry and non–sex-biased dispersal, respectively. In real species SGs typically show a deviation from HW equilibrium expectations toward an excess of heterozygotes (F_IS < 0). Such excess is found in several mammalian species that form groups ranging from small pair-bonded units to relatively large aggregations. Negative F_IS values are expected when populations are subdivided into SGs, whatever mating schemes, and dispersal strategies. The F_IS values obtained in the simulation study were within the range of values measured for many real species.

Sampling Design Can Explain Contradictory Observations.

The results above were obtained by sampling randomly within SGs (i.e., combining individuals from different ages and reproductive classes). This is the typical sampling design of empirical studies and is therefore the most appropriate to assess and confront with real data. Most empirical studies are based on noncontrolled sampling schemes. The results obtained from this type of sampling are thus what any modeling should explain. The fact that the F_IS values we found in our simulations overlap with those reported for many social species shows that our model is realistic enough to reproduce the results of some field studies (Fig. 3).

Our framework also allows us to select which individuals are sampled and, thus, to quantify the effect of different sampling schemes on genetic diversity and differentiation patterns. For example, it allowed us to analyze samples composed of reproductive status (RS) individuals only. Because RS individuals are the only type of individuals in classic deme-based models, we also examined the effects of such sampling scheme, even though such biased sampling is rarely—if ever—done in field studies. Our results suggest that the way SGs are sampled affects the expected proportion of heterozygotes (expected heterozygosity, H_ES) as much, if not more, than differences in mating strategies and dispersal (Figs. 4 and 5). As described in detail below, the issue of sampling design is a neglected but important one that can explain contradictory results reported by behavioral ecologists and population geneticists.

Fig. 4.

The effect of pooling SGs on deviations from random-mating. Inbreeding coefficient (F_IS) distributions obtained from data simulated under the extreme polygyny non–sex-biased dispersal scenario. The dashed line is the F_IS distribution obtained within SGs (same as represented in blue in Fig. 3), whereas solid lines are F_IS values obtained when samples from different SGs are pooled and analyzed as one single unit. From left to right, blue lines correspond to grouping of 2, 4, and 10 SGs. This figure shows that the heterozygote excess signal (F_IS < 0) is gradually lost when samples from more than one SG are pooled (solid lines). Indeed, the pooling of four SGs provided results that do not differ significantly from theoretical expectations under an ideal population with random-mating. This also shows that the sampling strategy can lead to spurious signals of deficit in heterozygotes (F_IS > 0, solid lines), which is an important result as in genetic studies of social species, samples from more than one SG are usually analyzed together. This is misleading because pooling individuals from different SGs can lead to the conclusion that populations are structured into random-mating units when they actually have a different mating dynamics. Note that all of the other mating systems analyzed showed the same pattern (see Fig. S4 and Table S2).

Fig. 5.

Sampling design and inbreeding signals. Inbreeding coefficient (F_IS) estimated from different sampling schemes within SGs obtained under the extreme polygyny non–sex-biased dispersal scenario. The dashed line represents F_IS estimated when using the typical sampling design in ecological studies (random sample within SGs), whereas the black solid line represents F_IS values estimated when considering samples composed of reproductive individuals. The gray line represents the F_IS distribution obtained from equivalent population genetics classic models assuming deme sizes and sex ratios as in the SGs’ simulations (see SI Materials and Methods for details). The figure shows that datasets similar to those obtained by field biologists and conservation geneticists produce signatures of outbreeding (excess of heterozygotes; F_IS < 0), whereas a sampling design based on the general assumptions of classic population genetics' models produces a spurious signal of random mating (neither excess of heterozygotes, nor a deficit; F_IS = 0) (see also Fig. S5 and Table S2).

In most population genetic studies the SG is rarely the unit analyzed; instead, sampled individuals are usually combined using geographical criteria, and thus, samples include individuals from several SGs. We found that F_IS values consistently increased when samples from different SGs were pooled and analyzed as if they corresponded to a single unit. The greatest increment was caused by combining samples from two SGs, with diminishing effects of additional SGs. When many SGs were pooled no major deviations from random-mating were detected (Discussion, Fig. 4, Fig. S4, and Table S2).

Population genetics theory predicts a deficit in heterozygotes (i.e., a positive F_IS) when samples from different random-mating demes are combined, which is referred to as the Wahlund effect (48, 49). The expected positive F_IS values are based on the implicit assumption that demes are at HW (F_IS = 0). Our results showed that even though single SGs are characterized by an excess of heterozygotes (F_IS < 0), this departure from random mating is more difficult to detect when samples from two or more SGs are pooled together (Fig. 4). This finding is consistent with a Wahlund effect where the pooled units are not at HW equilibrium but rather exhibit an excess of heterozygotes. Thus, F_IS values reported in population genetic studies are often overestimates that misrepresent the departures from random mating caused by social structure. However, these results also show that population genetics theory, in this case the Wahlund effect, still applies even when the unit of study is more complex than a simple WF deme.

Our results also suggest that in addition to the effect of combining samples from different SGs, knowing the social status of sampled individuals is more important than population geneticists would typically recognize (but see refs. 22, 50, and 51 for important but undercited exceptions). When samples comprised only RS individuals, we did not find major deviations from random-mating, neither within SGs nor when SGs were pooled (Fig. 5, Fig. S4, and Table S2).

Taken together, our results suggest that signatures of outbreeding reflect an organizational level below that of the population/deme that should be recognized and reconciles the two apparent contradictory observations: (i) no departures from random-mating even in highly social-species (described in population genetic studies) (Fig. 5 and Fig. S5); and (ii) genetic evidence for outbreeding (described in ecological studies) (Fig. 3, Fig. S3, and Table S2).

Discussion

Species are spatially subdivided at several hierarchical levels. How these subunits should be represented or modeled is likely to depend on the questions asked. For several decades population geneticists have used simple but very powerful and useful hierarchical representations to describe the overall relationships between such subunits and to document the patterns of genetic variation. Population genetics models have played a major role in clarifying the role of space in maintaining genetic variation and in generating patterns that could not be easily reproduced without structure (52, 53). However, in most studies this classic approach to population structure has tended to ignore the complex relationships that can exist among individuals living within SGs and under the various mating systems that social species exhibit. This deme-based viewpoint has had a long-standing impact on the interpretation of genetic data and has even produced an influential theory of molecular evolution. Here, we showed that: (i) it is possible to build a simple and flexible framework with an underlying model explicitly incorporating social structure and mating systems; (ii) this model predicts patterns of genotypic and genetic diversity that are different from those predicted by deme-based models; and (iii) the sampling strategies are an important issue that can explain real data and the lack of convergence between what ecologist report and what populations genetics’ studies find. The parameters used in our model are also more easily translated in real-world data than population genetics parameters (such as effective size or migration rates).

The Evolution of Social Behavior.

Sociality has traditionally been explained using ecological arguments over time scales beyond those of interest to most population geneticists. With few exceptions, population genetics theory has rarely been invoked to explain sociality or the evolution of social species [but see the Bush–Wilson hypothesis below (39, 40)]. Our results show that social structure has a fundamental impact on the genetics of populations. In comparison with expectations from deme-based models, socially structured populations exhibit an increase in genetic and genotypic diversity, the latter being often associated with fitness traits (54, 55). Several studies have found that higher individual heterozygosity is associated with fitness in an important range of traits. For example, heterozygosity was found to increase neonatal survival and attractiveness in harbor seals (56, 57) and territory size in a group-living bird (58). In addition, studies on the MHC show that heterozygosity increases longevity in baboons (59) and the immune response in HIV-infected Mauritian macaques (60). Sociality is usually explained by the fitness advantages it supposedly provides (21, 61), as for example, increased mating opportunities and offspring survival (62–65). Because it also incurs costs, such as higher risks of disease infection or parasite transmission, and inbreeding in small groups, the maximization of observed heterozygosity would therefore appear to act against some of the main costs typically associated with group living. Although costs and benefits of group living have been discussed for more than three decades (14, 21), interpretations of social behavior and explanations for sociality are still controversial today (6, 20). One possible reason for some of these debates is the difficulty in quantifying the relative importance of different evolutionary forces. For example, our results suggest that inbreeding avoidance might be easier to achieve through selection simply because social structure naturally generates outbreeding signatures. Our results therefore add a new dimension to this debate, suggesting that maintaining the social structure is in itself an advantage.

Inbreeding in Socially Structured Populations.

Inbreeding depression occurs in many natural populations. Because inbred individuals typically have decreased survival and fertility (66), inbreeding depression is expected to be a powerful selective force favoring any behavior that increases individual heterozygosity. However, this is still a hotly debated issue and it is not clear how heterozygosity contributes to fitness increase. Balloux et al. (55) have suggested that inbreeding is not necessarily correlated with individual heterozygosity. These authors noted that apparent heterosis could result from linkage between markers experiencing balancing selection and other genomic markers. Under such a scenario regions under balancing selection would drive the positive correlation between heterozygosity and fitness in other genomic regions. Our results show that the situation may be even simpler because individuals are represented in our model by independent markers without any linkage with other loci and we do not assume any type of selection. Thus, the excess of heterozygotes detected in our data cannot be an outcome of selective mechanisms. In addition to heterosis, another common explanation is that the outbreeding signatures found in many species are a consequence of sex-biased dispersal strategies. In particular, this heterozygosity excess is seen as a consequence of inbreeding avoidance mechanisms (44, 67–69). There is indeed significant experimental evidence of inbreeding avoidance and kin recognition (70, 71).

We show that in a network of SGs where individuals are replaced at random among all of the potential reproductive individuals available in the network, an excess of heterozygotes is observed. Therefore, there is no need to invoke inbreeding avoidance mechanisms to explain the outbreeding signature observed in so many natural systems (see also refs. 23 and 24). Furthermore, there may also be no need to invoke inbreeding avoidance as a necessary cause for the evolution of sex-biased dispersal. In fact, an increasing number of studies (reviewed in ref. 44) suggest that inbreeding avoidance may not be as widespread as it was previously believed. In most species individuals with a high inbreeding coefficient are likely to be rare and, within a population, inbreeding may vary widely among individuals. In addition, the costs of outbreeding may have been neglected and many species may be either tolerant to inbreeding or may even prefer to mate with relatives [“inbreeding preference” (44)]. Szulkin et al. (44) also discuss the difficulty in identifying inbreeding avoidance strategies in animals, especially because there are no clearly identifiable null-models to which real populations can be compared. As these authors rightly note, a precise measure of individual inbreeding requires access to the exact pedigree or an exhaustive sampling of the population under study. This is rarely possible and may therefore lead to biases in the estimates of inbreeding obtained in most samples. Our results confirm this point and our framework thus provides a null-model against which real case studies can be properly compared.

Although we do claim that our framework provides new results beyond those of classic population genetics models, we also wish to stress that deviations from HW expectations have already been anticipated by theoretical deme-based models. Such models accounted for the effects of variance in reproductive success, unbalanced sex-ratios, and sex-biased dispersal (22, 72–75). In small populations the absence of selfing in dioecious organisms [i.e., separate sexes (51)], differences in allelic frequencies between males and females (73, 74), and sex-biased dispersal (22) can result in an excess of heterozygotes. Unfortunately, these results have been either ignored or partly misinterpreted. For example, behavioral ecologists have interpreted Chesser’s theoretical work as if it suggested that outbreeding was a consequence of sex-biased dispersal. This interpretation may come from the fact that Chesser (23) did indeed use a model with female philopatry. However, in another study Chesser (24) actually showed that sex-biased dispersal was not necessary to generate outbreeding within SGs. Our study confirms that sex-biased dispersal is not required to explain the observed excess of heterozygotes but it also shows that modeling SGs as small WF units is not sufficient. Moreover, we identified nonreproductive individuals as central in generating signals in real datasets beyond those predicted by classical models.

The Effect of the Sampling Scheme.

Sampling is increasingly recognized as an important issue in population genetics inference (51, 52, 76–78). Beyond the obvious effect of the sample size, several recent studies (77, 78) have shown that the sampling schemes typically used by population geneticists can be highly misleading in population genetics inference.

Basset et al. (50), and Prout (22) before them, discussed the importance of the timing of sampling (in relation to dispersal events) in generating departures from HW proportions, but these results have been generally neglected even though there are real-case studies where such effects were observed (33, 46, 47). Although none of these studies are directly comparable to ours, they do point out that the interpretation of genetic data is strongly affected by the composition of samples. We found that different samples’ composition (in terms of the status of individuals) can lead to different perceptions of the distribution of genotypic diversity, ranging from signals of outbreeding to no deviations from random mating within SGs. Basset et al. (50) proposed that reproductive events drive a heterozygote excess that is counteracted by dispersal, and hence F_IS values estimated from juveniles are lower (more outbred) than when estimated from adults. Similarly, we found that when samples are only composed by RS individuals, the expected heterozygosity within SGs is larger than when individuals from any age and reproductive class are included.

Unfortunately, in the wild it may be difficult to determine the reproductive status of individuals. This is a crucial point. Because it may not always be possible to identify RS individuals, it is important to be aware that the composition of samples has a great influence on genetic measures when drawing conclusions about social species. Samples often represent a mixture of individuals from different SGs and social status. This finding is similar to the mixing of demes, which is known to generate a Wahlund-effect (48, 49). The distinction is that the SGs are not at HW equilibrium but rather exhibit negative F_IS values and the Wahlund effect brings them closer to, rather than further from, HW proportions.

Our results show that apparent random-mating populations found in traditional population genetic studies are difficult to interpret. On one hand, deviations from random-mating that are usually attributed to ecological processes (inbreeding avoidance) may be a simple outcome of social structure. On the other hand, a real nonrandom mating population may exhibit HW proportions as a result of noncontrolled sampling strategies. As far as population genetic studies are concerned, it may in many cases be appropriate to consider a socially structured population as a WF population (in the sense that it has the expected HW proportions of genotypic diversity at the population level), but it is important to recognize that important deviations are ongoing at the lower genotypic level. We provide a framework that reconciles the results obtained by classic models and the results obtained in empirical genetic and ecological studies, whether they ignore social structure or explicitly account for it. Some theoretical studies have produced important results that are similar to ours (50). However, these results are typically ignored, perhaps because of the fact that they are not reproducible by most existing simulation software. In addition, the last 20 y have led to a major development of the coalescent theory, which makes assumptions that are incompatible with SG modeling. For example, multiple coalescent events are usually ignored (but see ref. 79 for generation by generation coalescent algorithms, or refs. 80 and 81). Similarly, the sample size is assumed to be negligible compared with the effective size (N_e) of the deme.

Moreover, the comparison between models is not always as straightforward as it may seem. Even if one focuses only on deme-based models, if one compares n-island models and nonstructured WF models, N_e expressions are only valid under some specific assumptions. As discussed by Sjödin et al. (82) and by Wakeley and Sargsyan (83), for some models of structure the concept of effective size is not necessarily meaningful. Similarly, Chikhi et al. (77) have shown with others (52, 76, 78) that structured populations exhibit signals of population decrease even though the population size remained constant. That is, patterns of genetic diversity in an n-island model require a “dynamic N_e”, simply because no constant N_e could explain the data. One could find an N_e that would explain the H_T, but not other statistics, such as Tajima’s D. The results presented here go further and suggest that in social species, the concept of N_e may thus not always be the most appropriate concept to understand the gene dynamics.

We believe that there are several differences between our SG model and deme-based models that could contribute to explaining why we found a higher level of genetic diversity (Fig. 2). The first difference is that our model allows for overlapping generations. Because this is expected to reduce the effective size, it is unlikely to explain the differences found here. The second difference is that our model incorporates individuals that do not reproduce but may be recruited when RS individuals die. These individuals can be seen as “reservoirs” of genetic diversity, and may be one reason for the increase in diversity found here. Taken together, these two factors have opposite effects, and we should therefore be cautious in extending our conclusions to all social species and environments. The crucial role played by the non-RS individuals stresses again one of the main points of our study, namely the role of sampling. Non-RS individuals can represent a large proportion of the individuals that make up real populations, and are actually responsible for the genetic signatures found in empirical studies. Non-RS individuals are not explicitly represented in WF models but can store genotypes or alleles that were lost among the RS individuals.

Altogether, we stress that by forcing real data into deme-based models, some signals, such as outbreeding described in many social species, are ignored or seen as noise when these signals actually are what we need to interpret and explain. We need to identify models that can incorporate the complexities of real populations, such as social structure and mating systems (84, 85), as argued by several ecologists in a recent critical review on ecological models (86).

Social Structuring, Karyotypic Evolution, and Speciation Rates.

Rates of speciation and chromosomal evolution have been described to occur faster in genera where social organization is typically found. This observation led Bush et al. (40) and Wilson et al. (39) to hypothesize that the social structure is a key factor in promoting conditions that accelerate the rates of evolution (Bush–Wilson hypothesis). This hypothesis had its roots in the population genetics theory for subdivided populations, and postulates that stronger genetic drift and inbreeding within demes are conditions under which new gene arrangements have a better chance of becoming fixed and spread. The fixation of new mutations in small isolated demes might ultimately boost postmating isolation mechanisms and result in reproductive isolation (39).

However, this apparently reasonable and attractive hypothesis was criticized by field ecologists who found no evidence for inbreeding. Indeed, social structure appears to minimize the loss of genetic diversity in marmots, prairie dogs, sifakas, and deer, among others (26, 30–32, 35, 45). This finding is in agreement with what we find in our model.

Concluding Remarks and Perspectives

Social structure is ubiquitous among mammals and birds. The fact that these two groups exhibit sex-biased dispersal patterns that are widely conserved (male-biased in mammals and female-biased in birds) (1) is striking and calls for an evolutionary explanation that could be related to social structure, kin recognition, and inbreeding patterns. Inbreeding avoidance may be a factor driving sex-biased dispersal evolution, and there is evidence for it, but its quantification requires proper testing with the identification of appropriate null-models and data. Our results suggest that outbreeding rather than inbreeding is an emergent property of social structure. The fact that inbreeding avoidance has been questioned by numerous field studies suggests that our framework can provide the null-models required to test hypotheses regarding the origin, maintenance, and consequences of social structure.

A number of ecological and conservation studies have argued that changes in habitat composition affect group size, mate availability, and mating behavior (9, 87, 88). Moreover, the increasingly reduced and fragmented landscapes (89) may limit dispersal, increase isolation, and promote high relatedness within groups. Predicting the consequences of such changes with appropriate models, such as our framework provides, may thus be critical for conservation strategies.

Historically, SGs have been modeled as small demes (39, 40, 75) and this has introduced a gap between behavioral ecology observations and population genetic theory. Our work shows that social structure leads to genetic properties that cannot be modeled by considering a population as a network of small demes. In particular, we identified non-RS individuals as crucial in generating signals in real datasets that could not be reproduced by deme-based models, and as potential reservoirs of genetic diversity. In a few words, we found differences in terms of (i) genotypic diversity, (ii) genetic diversity, and (iii) sampling effect (role of RS and non-RS individuals, grouping of SGs into arbitrary random mating populations, and so forth). Accounting for fine structure should allow better understanding and quantifying the relative importance of factors, such as selection and sex-biased dispersal.

However, more work is still needed and it is important to recognize that the island model considered here ignores spatial structure. Classic population genetics models have shown that spatial structure leads to important genetic consequences, such as isolation by distance (90). Because most species are spatially structured and dispersal is space-dependent, real populations may be more easily recognized in a classic stepping-stone model where each deme is only connected to its neighbors (91). We carried out additional simulations of SGs and WF demes under a stepping-stone type of structure. The results we obtained confirm that SGs exhibit genetic patterns that are significantly different from those predicted by deme-based stepping-stone models (Fig. S6). Finally, a combination of ecological data and modeling is needed to increase our understanding of the genetic properties of social species. Our framework provides a bridge between population geneticists and behavioral ecologists and allows explicit realistic modeling of the evolution of SGs.

Materials and Methods

General Framework.

We consider a population of dioecious individuals consisting of SGs between which individuals can move according to different prespecified connectivity topologies (Fig. 1A) with or without sex-biased migration. Individuals, males or females, go through a simplified life cycle that involves several stages or age classes (Fig. 1B). As a consequence, an SG consists of a small age-structured unit, and the model can thus be seen as a model with overlapping generations. Newly born individuals go through the following age classes: young (between birth and weaning), juveniles (between weaning and reproductive age), and finally adults (i.e., above reproductive age). We assume that not all adult individuals are effectively reproductive, that is, we consider that only a limited and fixed (i.e., species-specific) number of adults have the possibility to reproduce within an SG at a given reproductive event. These individuals are the effectively reproductive individuals and are identified as having “reproductive status” (RS). At each reproductive event, RS males are sampled with replacement and females without replacement, which allows for several offspring to be sired to a different mother while sharing the same father. For each independent locus, the genotypes of the offspring are sampled from the parents who contribute with one allele each. We only considered microsatellite markers and we assumed that mutations occur according to a stepwise mutation model at an equal rate across all loci.

We model different mating systems by allowing a variable number of male and female RS individuals per SG. This approach allowed us to simulate different mating systems, such as monogamy and polygyny (Table 1). Other more specific details of the model and implementation can be found in SI Materials and Methods.

Table 1.

The different scenarios of social structure considered in the simulations

Mating system	No. males	No. females	Structure	Dispersing sex
Monogamy	1	1	Island	Male
				Male/Female
Polygynandry	2	2	Island	Male
				Male/Female
Extreme polygyny	1	10	Island	Male
				Male/Female
Moderate polygyny	4	10	Island	Male
				Male/Female

The four types of mating system analyzed differ in the size and sex-ratios of SGs. These are: monogamy, 1 breeding (RS) pair per SG; polygynandry, 2 breeding females per SG that each mate with 1 of 2 males at each breeding season; extreme polygyny, 1 male and 10 females per SG, which intends to represent harem-polygyny; and moderate polygyny, 4 male and 10 females per SG, which intends to represent a dominance-polygyny mating system (see SI Materials and Methods for details).

Simulating Mating Systems.

Four different mating systems (monogamy, polygynandry, extreme polygyny, and moderate polygyny) (Table 1 and SI Materials and Methods) were simulated. We assumed that the maximum number of SGs was fixed to g and that SGs could become extinct (see below for details). In all simulations we assumed that dispersal events were taking place symmetrically among all SGs, that is, individuals could disperse to any other group in the population. In other words, we assumed an implicit “island model” network for dispersal and colonization events (Fig. 1A). Under each mating system we simulated dispersal events assuming either a sex-biased (female philopatry) or non–sex-biased dispersal regime. Female philopatry was chosen because it is most common among mammals (1). Altogether, eight (4 × 2) different combinations of mating system and dispersal were simulated. We aimed to represent some of the most diverse and representative systems known in mammals (2).

Genetic data on social species were taken from the literature. Datasets were distinguished according to the level at which analyses were performed (Table S3). Quantitative results from the SG simulations were compared with the ones obtained in ecological and population genetic studies. See Table S4 for parameter values of the SGs.

Because of the computational costs of our simulations and the required postprocessing analyses, most simulations performed in this study assumed a g of 50. For example, in primates this would correspond to several hundred individuals and habitats of several square kilometers for many species. However, validation tests and comparisons were also performed with larger and smaller numbers of SGs (g = 20, 100, 200). We were interested in both equilibrium and nonequilibrium conditions but, because very little work has been done on the properties of SGs from a population genetic perspective [but see the works of Prout (22) and Chesser (23, 24)], the results presented here focus mainly on equilibrium conditions. SGs were initialized with RS individuals whose genotypes were obtained by sampling from a WF population (i.e., of stationary size). Our simulations were then run for 6,000 generations, even though population genetics theory predicted that equilibrium would be reached in much fewer generations. Allelic frequencies of the founding populations were simulated with scaled mutation rate θ = 20, using a user-friendly software (92) that uses the program ms (93) as an engine to simulate genetic data.

Genetic and Genotypic Diversity in Socially Structured Populations.

To quantify departures from HW proportions (excess or deficit in heterozygotes), we calculated Wright’s F_IS statistic, using Weir and Cockerham’s estimator (94). This was done by using several sampling strategies either at the SG level (as is done in behavioral ecology studies) or by sequentially pooling samples from 2, 4, and 10 SGs (as is usually done in population genetics studies that typically ignore SGs). For each mating strategy, we sampled four individuals from 20 SGs (in a total of 80 samples). Within each SG we randomly sampled the four individuals (i) among all age classes or (ii) among the RS individuals, when possible (except for monogamy, where only the two RS individuals were analyzed).

To compare the effect of social structure with the effect of population structure we compared the diversity obtained (under models of social organization) with the diversity expected from typical population genetics (i.e., deme-based) models of structured populations, while maintaining the same levels of differentiation. We simulated a 50-island model with same number RS individuals per deme as the SG model. The same sex ratios were used for both the deme-based and SG simulations. Because of this, the variance in reproductive success should be similar or the same in both models. Simulations were also performed with two migration schemes: female philopatry (sex-biased migration) and non–sex-biased migration. The deme-based simulated data sets were obtained with the EASYPOP program (95), which allows for the simulation of structured populations with variable sex ratios. Diversity was compared using the overall unbiased heterozygosity according to Nei (96) and the within SG expected and observed heterozygosity according to Nei and Chesser (97).

Genetic differentiation between SGs/demes was measured by the Weir and Cockerham (94) F_ST. To have the same amount of genetic differentiation between demes as observed under the SG scenarios, we simulated datasets varying the migration rate from 0.3 up to 1 in EASYPOP (95). For comparison purposes we used EASYPOP datasets that generated F_ST values within the range observed in the SG simulations (SI Materials and Methods). The EASYPOP values used for the comparison with SGs were sampled randomly among all of the values obtained within this range.