Our results provide evidence that selection for dominant individuals in a population can contribute to the genetic evolution of other traits via their genetic correlations with social dominance. We obtained this result by modelling social dominance as a phenotype and estimating both its direct and indirect genetic effects, and for the first time, we detected significant genetic correlations of both components with fitness, health, and morphological traits. As expected, our bivariate models showed that social dominance is a heritable trait, which is consistent with the heritability of other social behaviours [63] that have been reported for domestic [17] and wild populations [64, 65].
We found antagonistic genetic correlations between winning agonistic interactions and traits associated with cows’ health and fitness. Cows that won more fights had on average a lower fertility, decreased milk yield and poorer udder health [34, 35]. The most dominant cows had also ‘male-like’ morphological features, such as a bulkier body in terms of thorax depth and front muscularity. As expected, the genetic correlations of the direct and indirect components of social dominance with the other traits were symmetrical, i.e. of the same magnitude but with opposite signs. These results are important for two reasons: first, they show how social behavioural traits, such as social dominance, are genetically correlated with morphological, health and life-history traits. Second, our results suggest that selection for social dominance is not only associated with a clear evolutionary response in this trait, but that this evolutionary response is also ‘dragging’ along some correlated traits. In fact social dominance showed a strong trend of change in its EBV, with the direct component showing a marked increase and, of course, the indirect component showing a mirroring decrease. This means that, while the phenotype cannot evolve, the genetic merit for social dominance of individuals born more recently is markedly greater than that of their ancestors. In other words, cows in our population have evolved an increased fighting ability in response to selection for social dominance; but it is impossible to observe this at the phenotypic level, because the mean of a 1/0 trait cannot change. However, the traits correlated with social dominance can change: we found that morphological features that are linked with greater frontal muscularity had increased EBV, while the EBV for mammary health decreased. These traits changed according to their correlation with the target of selection, i.e., social dominance. In 2012, selection breeding programs in Aosta cattle introduced weights of 50% for milk yield, 40% for social dominance and 10% for meat yield, but the main unofficial target trait under selection has always been social dominance in this breed, with milk and meat yields being of very little importance in the selection choices [17, 66, 67]. Although we cannot state that the observed genetic trends are only due to the selective pressure for social dominance, we argue that it was the most important factor that drove the evolutionary change of the correlated traits. Higher genetic merit for social dominance, given the antagonistic correlation with milk yield, is not likely to be a consequence of selection for milk yield, and the ‘meat’ component is by far the lowest trait in the index (10%). While selection for milk yield might have played a role in the trend observed for SCS, it is very unlikely that it is responsible for the trends that we observed for morphological features. Indeed, this is because the trend of change observed for muscularity is opposite to what would be expected if only milk yield was included in the selection index, which would lead to a more ‘female-like’ body conformation spreading in the population. Thus, our results provide strong evidence that social dominance as a behavioural trait can not only evolve, but can also impact the evolution of other traits.
Variance constraints in dyadic contest outcomes
A social phenotype such as dominance can only be modelled by considering at the same time both the variance attributed to the genetic makeup of the focal (direct) and the variance attributed to the genetic makeup of the opponent (indirect) individuals. Since the phenotype associated with each fight is the same for both the focal and opponent individuals, variances are expected to be equal, and their covariance to have the same magnitude but with opposite signs. On the one hand, certain software or packages (e.g., MCMCglmm) can add this constraint to the animal model, which ensures that the correlation between direct and indirect components is − 1 and the total heritability for social dominance is 0. On the other hand, an unconstrained model can test the size and completeness of the dataset, since the values that it estimates freely approximate these features. Our unconstrained single-trait model obtained a correlation of − 0.989 between direct and indirect components of social dominance and a total heritability of 0.003. These very close approximations to − 1 and 0 were also obtained in all subsequent bivariate analysis (see Additional file 2: Table S4). The heritability estimates for the direct and indirect components were 0.126 and 0.121, respectively, not so different from that obtained by the constrained model (0.164). The difference in the estimates of heritability between the constrained and the unconstrained models is probably due to the fact that, in the constrained model, we could not insert herd as a random effect, neither for the focal nor for the opponent individual, since it would have impeded chain convergence. In our bivariate analysis, the slight differences in (co)variances between direct and indirect effects do not carry any biological meaning and are likely just an artefact that is caused by the use of a real-life dataset. Taken together, we believe that our unconstrained analysis was more than sufficiently powerful to recover realistic estimates, as it always approximated very closely both the features of the social dominance model and the heritability obtained with the constrained single-trait model. However, it must be noted that our constrained model also produced good estimates despite running only on a small sub-sample of the dataset, which indicates that a constraint might help save power and accuracy when the dataset is small. Finally, differences between the direct and indirect correlations do not have biological meaning, and the most realistic estimate of their magnitude, when they differ, would probably be an average of the two posterior distributions.
Features and limitations of our dataset
It is very difficult to estimate the IGE of behaviour [65, 68] and collecting a dataset of sufficient size and with enough details to estimate the genetic correlations of IGE with other traits is even more complicated [69]. The unique opportunity given by the meticulously recorded ‘Batailles des Reines’ in the Aosta Chestnut-Black Pied cattle breed created the conditions for a decade-long study with tens of thousands of individuals. However, the dataset did have some limitations: for example, we could not model dominance in male individuals to check for sexually antagonistic pleiotropy in the investigated traits, since the males do not participate in the contests. Nonetheless, the artificial nature of the agonistic interactions allowed us to avoid several of the issues often encountered when modelling quantitative genetic data in a natural population [70, 71]. Studies on the genetics of social behaviours in natural populations struggle to partition the variance that is linked to the social environment [2, 65]. Nepotism, brood effects, and the presence of other related individuals tend to inflate the variance that is attributed to genetic effects, especially in small populations [4, 18, 72] and (Susan Alberts, personal communication). Modelling these factors through IGE is often much too complex, given that the nature of the interactions is often communal, highly sparse, and discontinuous [65]. Given how agonistic interactions are structured in the ‘Batailles des Reines’ (see Methods), we could discard issues that are linked with the immediate social environment, such as possible interference of other conspecifics with the two individuals engaged in an ongoing duel (e.g., [73]). In addition, the repeated nature of these battles (i.e., individuals may compete more than once within an annual tournament and over the years) permitted a randomized and standardized evaluation of social dominance. This also allowed us to partition the trait variance relative to the (non-genetic) individual permanent environment, an achievement that is rarely possible in most studies under natural conditions [72]. Two additional specific features of our system are worth mentioning, as they might have partially influenced our results. The first is that cows are chosen to fight within weight classes. In other words, two cows that greatly differ in weight might never face each other in tournaments. We do not have direct evidence that size directly influences social dominance in Aosta cows, but it does in many wild [74] and livestock [75] species, with larger animals being usually more dominant [24]. Thus, we could expect that if some cows were to fight adversaries outside their weight class, it could influence genetic variances and correlations. Since size usually has a large genetic component, a positive correlation between this component and social dominance would entail that our reported genetic variance for the latter might be underestimated. In other words, if there had been no division into weight classes, a greater part of the fights would have been decided by a factor that is strongly controlled by the genetic background of the cows. Moreover, our results show that winning cows have greater thorax depth and front muscularity, which in turn are possibly larger in cows of the ‘heavy’ fight category. This suggests that the correlation that we found between social dominance and this male-like morphological trait is also an underestimation, as across weight categories the relation between male-like appearance and winning might be more pronounced. However, with the current dataset we cannot speculate any further, since morphological traits are measured only once over a cow’s lifetime, while cows usually fight over several years, with their weight (and thus weight category) changing with age. A second factor which may have an influence on our results and is difficult to disentangle is that dominant cows have access to more fights. This raises two issues, the first one being that winning individuals might be repeated more times in the dataset. However, given the tournament-like arrangement of the ‘Batailles’, only a handful of cows out of hundreds get to fight more than 1 to 3 times in one day, thus any effect due to this issue is likely very small. The second issue is a possible winner effect: since cows that win more fights proceed further into the tournament, there might be a positive effect of previous wins—but, as with most repeated measures of social dominance, it would be difficult to disentangle an eventual winner effect from actual genetic quality. Moreover, in the set-up of our study, this is controlled by the fact that only cows that won an identical number of fights can face each other during the tournament, thus ensuring that an eventual winner effect is symmetrical in all fights.
Generalization of our results
In general, our set-up was thus more akin to an experiment under controlled conditions than to observations of a natural situation, and, accordingly, it permitted an unprecedented degree of focus not only on the genetic basis of behavioural traits, but also on their correlation with life-history traits. In natural populations, dominant individuals usually obtain more resources and mating opportunities: differences in individual quality [76] and resource acquisition [21, 23] combine with social feedback effects [24] to create the conditions for an increased overall fitness and health of dominant individuals [77]. In other words, possible antagonistic genetic correlations between social dominance and fitness or health traits in wild populations are more than compensated by the increased gains that social dominance allows. However, in a managed population, access to resources is not limited, which eliminates the confounding factor of differences in acquisition. Thus, it is possibly for this reason that, in our study, we found antagonistic genetic correlations between social dominance and fitness traits that are more rarely found in natural populations [78].
Genetic correlation between social dominance and fertility
The genetic correlation between social dominance and fertility is especially interesting from an evolutionary point of view. Fertility is a key fitness trait and, as with most life-history traits, its deviations from the population optimum are thought to be under strong negative selection [79]. In our study, fertility had a heritability estimate of ~ 0.025, in line with previous studies [34, 49] and thus, it was surprising to find that it presented strong antagonistic genetic correlations with social dominance. The genetic constraints and ontogeny of these traits appear to be robustly linked, possibly through hormonal pleiotropy [80], as masculinizing hormones, such as testosterone, are often linked to better competitive performances [81]. However, these hormones can also directly affect female fertility through the modification of primary sexual traits [32]. In our study, we found that the breeding value of fertility does not decrease over time, despite the selection of dominant individuals in this population. Given the low heritability of fertility, the slope associated with the change in the EBV of the cohort was higher than that of evolutionary change of the null model only in 52.6% of the MCMC replicates, which means that in spite of the unfavourable correlation, the rate of change is minimal; thus, it is unlikely that the trait will show a phenotypic shift in the near future [82]. However, given that fertility is a key trait for the livestock sector, this result highlights the importance of accounting for the risk that selection for social dominance (targeted or accidental) might have an impact on economically important traits linked with production and health.
Genetic correlation between social dominance and SCS, milk yield
Social dominance also showed an antagonistic genetic correlation with SCS. Dominant individuals had a larger number of somatic cells in milk, which means poorer udder health. SCS has widely been used in animal breeding as a proxy for mastitis risk and overall udder health [51]: a higher SCS has additionally been linked with a deteriorated immune system [83]. In particular, immune responses may be expected to have antagonistic relationships with social dominance [84], as the testosterone that often mediates fighting behaviour can profoundly impact them [85, 86]. Interestingly, we found that the EBV of this trait showed significant signs of degradation, which indicates a potential undesired consequence of the selection for fighting ability. This might be a cause of concern for the future, as SCS is an indicator of great economic value [50]. However, it must also be noted that degradation of SCS might not be caused only by selection for social dominance, as SCS has also an antagonistic correlation with milk yield [50]. Although milk yield is historically less important in the selection programs of Aosta cattle [66], it is a selection target and thus could indeed play a role in the degradation of SCS.
Milk yield showed a lower, but still significant, antagonistic genetic correlation with social dominance: dominant individuals have a decreased ability to provide large quantities of milk. The production of milk is a demanding task, requiring a lot of energy and nutrients. It involves a vast mobilization of resources, which can be greatly increased by direct artificial selection for milk production. For all these reasons, it is not unexpected that it shows an antagonistic correlation with social dominance. However, since milk production is also a target of selection, the effects of the unfavourable correlation with social dominance may be diluted by the weight that is attributed to milk production in the selection index, and for this reason, we do not see a deterioration in milk production EBV. On the other hand, usually cattle that are selected only for milk production show a change in their morphology, as udder size increases, frontal muscularity decreases, and cows assume a more ‘female-like’ body conformation. This is the opposite of what is observed in the Aosta breed, since there seems to be no change in udder morphology, and on the contrary, selection for winners has led to a masculinization of the cows’ body shape.
Genetic correlations between social dominance and morphology traits
We did not find a significant genetic correlation between social dominance and either the morphology of the udder or thinness. Although udder conformation traits are indicators of a ‘female-like’ conformation and would have been expected to trade-off with dominance related traits [32], in our population, udder morphology does not appear to show a clear trend of change, neither negative nor positive, which could be a consequence of the dual selection for both social dominance and milk production. However, front muscularity and thorax depth, two traits that evaluate key aspects of the bulky, wiry and ‘male-like’ appearance of cows, were not only genetically correlated with social dominance, indicating that more ‘male-like’ individuals won more agonistic interactions, but also showed changes in this population, as illustrated by our results. The change in the cows’ appearance over the years is the kind of fast evolutionary response that is typical of artificial selection. Indeed, while research has made a lot of progress on the neural pathways and physiology of aggressiveness and social dominance, it has been more difficult to quantify how selection for social dominance would change the shape of an animal body. Our results show that, as dominant individuals are selected for, the morphology of the population evolves. This evidence of a link between skeletal and muscular development with social dominance is significant because, while the very specific morphological traits used in our study did not have a high heritability, morphology is often associated with greater heritable variation and response to selection than behavioural and life-history traits [87]. Thus, a genetic correlation between morphological traits and social dominance becomes especially important in terms of genetic variability, selection, and potential evolution within social groups; indeed, we show that a population under strong selection for dominance could in fact greatly and rapidly change its morphology over time. Finally, our results have exciting implications for the genetic architecture of social dominance. A lot of work has been done on the genetic architecture of muscularity in cattle [88, 89] and on the genetic and molecular basis of social dominance behaviour in various species [90, 91]. The Aosta breed is a very good candidate to investigate what are the functional elements, genes and pathways that might be common to both traits, and that are thus at the basis not only of the genetic correlation, but also of the common evolutionary trajectory of the two traits.
Social dominance and evolution
Indeed, a key conclusion that can be derived from our study is that selection for social dominance has the evolutionary potential to affect other traits with very diverse genetic architectures [92]. Several of the traits that we considered here are, in fact, linked by antagonistic relationships of their own: for example, mastitis risk is often positively correlated with milk production [53], which in turn is negatively correlated with fertility [93]. Social dominance is, thus, linked to entirely different aspects of the cows’ reproductive biology, which implies the presence of several interconnected mechanisms of pleiotropy [94]. Indeed, in spite of the categorical nature of its phenotypic record, the expression of social dominance involves neural, hormonal, and physiological responses that are regulated by a myriad of factors, and thus, it is very difficult to piece together what are exactly the genomic targets of selection in this trait. The layered genetic architecture of this trait has far-reaching consequences: since social dominance is the most important factor in the selection of the Aosta Chestnut-Black Pied breed, artificial selection of dominant individuals is shown to affect not only the evolution of their behaviour, but also all the physiological and morphological traits that are pleiotropically linked with it. Through several interconnected pathways, the antagonistic genetic correlations with female-like and health-related traits might lead in the long term to a decrease of the average health of the population. The EBV of SCS in our dataset showed a significant degradation of the trait, present in 95.3% of all MCMC replicates. While there is a lag dividing this observation from actual phenotypic change [82], especially given that Aosta cows generally show good phenotypic values for fertility and SCS in comparison with cosmopolitan and intensively selected breeds, if the current selection trend continues, it is conceivable that key life history and health traits could worsen in the population. This is consistent with what has been reported for the Hérens breed, which was historically selected for fighting ability, and tends to show low values for fertility and milk yield [95]. In natural populations, where competition for resources is strong, covariance of life-history traits with social dominance could therefore lead them away from the optimum, causing local maladaptation [96]. Moreover, since life history, fitness, and health traits correlate not only with the direct but also with the indirect component of social traits, their rate of evolutionary change might be even faster [5]. For example, in populations of highly sociable species, better opportunities for dominant individuals might be directly linked with worse opportunities for the subordinates, particularly where resources are scarcer [97], depending on the context [98]. Finally, another major consequence of the selection of covarying traits might be the evolution of the correlations themselves. For example, varying degrees of the stringent endocrine constraint between fertility and social dominance might be differently advantageous and thus, in time, change their frequencies in the population [99]. Future studies should address whether genetic correlations between different traits might themselves be under selection and thus evolving in this and other populations.
Finally, social dominance represents a specific case of IGE, i.e. a competitive model where the effect of the opponent is key to obtaining realistic estimates of heritability for the trait but perfectly mirrors the effect of the focal individual. However, this might not be the case with other traits and situations. For example, our study did not consider any possible IGE that acts directly on fertility, milk production, morphology, etc. In fact, it is possible that these traits are themselves influenced by the diverse genotypes of their groupmates, especially if housed in small groups during their development. Indeed, extensive work has shown that, for example, aggressive behaviour in pigs is linked to decreased groupmates’ weights [100] and that several health traits are linked to significant social effects [101]. In Aosta cows, IGE that act directly on these traits are expected to be more diluted, as they are never housed in small or barren pens and are free to roam during summer. Thus, it was not possible to add this herd-mates effect to our study: in fact, within the matrices shown [Matrices (2) to (4)], we did not include the potential IGE acting on trait 1 (fertility, SCS, morphology, etc.). However, in future studies, provided an adequate dataset is available, it should be possible to perform bivariate analyses of two traits each with their own IGE. Moreover, besides for social dominance, IGE could be essential for the study of several other traits with values that are in part, or completely, dependent on the values of other individuals, such as, for example, leader and follower dynamics [102], or group success [1] where the value of a trait is the same for the entire group.
Add Comment