Mating preferences can drive expansion or contraction of major histocompatibility complex gene family

Abstract

Major histocompatibility complex (MHC)-based mating rules can evolve as a way to avoid inbreeding or to increase offspring immune competence. While the role of mating preference in shaping the MHC diversity in vertebrates has been acknowledged, its impact on individual MHC diversity has not been considered. Here, we use computer simulations to investigate how simple mating rules favouring MHC-dissimilar partners affect the evolution of the number of MHC variants in individual genomes, accompanying selection for resistance to parasites. We showed that the effect of such preferences could sometimes be dramatic. If preferences are aimed at avoiding identical alleles, the equilibrium number of MHC alleles is much smaller than under random mating. However, if the mating rule minimizes the ratio of shared to different alleles in partners, MHC number is higher than under random mating. Additionally, our simulations revealed that a negative correlation between the numbers of MHC variants in mated individuals can arise from simple rules of MHC-disassortative mating. Our results reveal unexpected potential of MHC-based mating preferences to drive MHC gene family expansions or contractions and highlight the need to study the mechanistic basis of such preferences, which is currently poorly understood.

1. Introduction

The major histocompatibility complex (MHC) genes have a dual function: in addition to the crucial role in vertebrate immune response, they also serve in social communication [1–3]. MHC genes code for molecules binding fragments of proteins from both self and pathogens (referred to as antigens). An adaptive immune response is initialized when the T-cell receptor identifies an MHC–antigen complex as one bearing a foreign oligopeptide. Pathogen species can evade immediate immune reaction by a rapid evolution of their oligopeptide repertoire. This, in turn, creates a positive selection pressure in hosts and leads to the maintenance of high polymorphism of MHC genes due to heterozygote advantage and/or negative-frequency-dependent selection [4–6]. MHC genotype can also serve as a social cue (e.g. to relatedness or to immunocompetence of a potential mating partner [1,7]). Preferences for MHC-dissimilar partners, reported for some species (reviewed in) [7,8], may, in turn, enhance MHC polymorphism [9–11].

While the evolution of MHC polymorphism has been extensively investigated by theorists [6,12–16], the selection pressures on the number of MHC variants expressed by single individuals have received less attention (but see [17]). A particular MHC molecule is able to bind a limited range of different antigens [18–20], thus individuals expressing many kinds of MHC molecules should be able to bind many different antigens, and therefore raise the immune response to many different pathogens. Indeed, heterozygosity at individual MHC loci and the number of MHC variants expressed across loci can be negatively associated with infection [4,21]. Also, increased pathogen pressure seems to select for high copy number [22,23]. However, the expansion of the MHC gene family is likely to be limited by constraints such as a higher chance of deletion of lymphocytes with receptors reacting with MHC molecules presenting self-oligopeptides during thymic education [24,25]. Even though there is some evidence for optimizing selection by pathogens on individual diversity in a few species [26–28], in general, species vary widely in the number of MHC copies in their genomes ranging from a very few in chicken [29] or humans [30], for example, to dozens in some rodents [31] or passerine birds [21]. It does not seem feasible that such striking differences in the number of MHC variants in different species can be explained only by the differences in parasite pressure alone, as it should be more than an order of magnitude to account for the observed difference in MHC variant numbers [17]. Here, we investigate whether mating preferences for MHC-dissimilar mates impact the evolution of individual MHC diversity.

Preferences for MHC-dissimilar mates may arise from two selective pressures: for the avoidance of mating with kin and for increased resistance to parasites in progeny (reviewed in [1,7]). The first explanation stresses the costs of mating with relatives, which in wild population can lead to an extreme loss of fitness (inbreeding depression) even at the moderate level of inbreeding [32]. MHC alleles are useful markers of kinship [33]—their extreme polymorphism implies that sharing an allele often results from common ancestry. Thus, avoiding mates with the same MHC alleles is an effective mechanism to avoid inbreeding. The second explanation is associated with the primary function of MHC molecules in immune response, which is presenting foreign antigens to T cells: MHC heterozygotes should be able to present a broader spectrum of antigens than homozygotes, thus raising an immune response against a wider range of genotypes of infectious organisms [4]. In consequence, preferences for MHC-dissimilar mates should result in improved immunity of progeny (although this benefit may be outweighed by the immunity cost of expressing too large an MHC diversity in species with highly duplicated MHC—see above).

Irrespective of the selective advantage, preferences for MHC-dissimilar mates were reported in some species across several vertebrate taxa, such as rodents [34,35], fish [36,37], reptiles [38], birds [39] and humans [40], but nonetheless, they are not universal (e.g. [41–43]). Meta-analyses of human and non-human primates [44] and non-human vertebrates [8] showed that trends for preferences for dissimilar mates were non-significant overall. However, both reported considerable variations of effect sizes associated with phylogeny and species-specificity. Such differences in effect sizes between taxa/species can potentially affect selection for duplication and diversification of MHC genes within genomes, resulting in interspecific differences in individual MHC diversity. For example, in species where MHC-based mating preferences are strong, duplication and diversification of MHC genes may be favoured, as it will increase bearer's dissimilarity to potential partners. Conversely, expressing many MHC variants may increase the risk of sharing a variant with an unrelated individual, which might decrease chances to find a compatible mate. Here, we use computer simulations to demonstrate that simple rules for MHC-disassortative mating could indeed strongly affect the evolution of MHC duplication and divergence. Interestingly, depending on the details of MHC-based mating mechanism, it can select for or against MHC gene family expansion. MHC-based mating preferences thus have a strong potential to explain interspecific variation in individual MHC diversity.

2. Methods

We implemented MHC-based mating rules in a simulated population in which MHC diversity is subject to selection by parasites, to reflect a typical situation encountered in wild populations. To simulate host–parasite coevolution, we used a theoretical framework we originally developed for investigating how pathogen diversity (measured by the number of pathogen species to which a host was exposed) and mutation rates impact the MHC gene copy number [17], but we added a step involving mate choice. The general framework for simulations of MHC-pathogen coevolution is based on an approach successfully used in previous research on the evolution of MHC diversity [6,9,13]. It simulates MHC antigen binding sites, as well as antigens, as strings of 0 and 1 s (bit strings). Pathogen recognition is based on the similarities in bit sequences between the host's MHC alleles and the pathogen's antigen. A detailed description of the model is in electronic supplementary material, S1.

In brief, the first phase of each generation simulated selection by pathogens as implemented by Bentkowski & Radwan [17]. MHC molecules were simulated as 16 bit-long strings (each bit can be thought of as a peptide binding residue of the MHC molecule; there are 12–23 such residues in human MHC molecules [45]). Each host had two chromosomes harbouring MHC genes subject to mutations, deletions and duplications. Mutations changed particular bits, resulting in new MHC variants (we use term variants rather than alleles, because due to duplication and deletion, chromosomes could have different number of loci). Our model did not allow for recombination between the MHC genes.

Pathogen's antigens were 6000 bits long. If the antigen had at least one 16-bit-long substring identical with one of the host's MHCs, the antigen was recognized by the host immune system and the pathogen cleared. Finally, host viability was proportional to the number of pathogens recognized. Similarly to previous work [17], we introduced an intrinsic cost for expressing too high MHC diversity (see electronic supplementary material, S1 for details), which reflected constraining mechanisms such as depletion of T cells due to increased negative selection [24,25].

The second phase of each generation (specific for this work) simulated MHC-dependent mate choice. Individuals were drawn for mating with probability proportional to their viability. For the sake of simplicity, we did not differentiate between sexes (i.e. our hosts are equivalent to out-crossing hermaphrodites). The programme selected one random individual (the chooser) and offered it 10 random mates, from among which one was chosen as a mate according to mating rules described below. The pair produced one offspring. Reproduction continued until the host population of 1000 individuals was recreated.

We modelled several variants of hypothetical mechanistic mating rules that minimized the probability of mating with kin. The first rule selected the partner with the minimal number of MHC variants shared with the chooser (MinShared scenario henceforth). Such a strategy would be very effective in avoiding mating with close relatives (as individuals carrying identical alleles have increased probability of being related). However, such a rule could compromise the efficiency of finding a mate if allele sharing with unrelated individuals was common, e.g. in small populations. Therefore, we modelled an alternative mating rule, assuming that in addition to the suppressive effect of shared alleles, there is a stimulatory effect of different alleles. This rule (henceforth PropShared) minimized the proportion of shared alleles (min(|A∩B|/|A∪B|)), where A and B are sets of variants in two potential mating partners). For completeness, we also simulated a mating rule that gave preference to the partner having the largest number of MHC variants different from those carried by the chooser (MaxDiffer). In a control (Random) scenario, mates were matched randomly.

To see if mating rules we simulated can suppress MHC expansion in the absence of intrinsic costs of expressing many MHC variants, we simulated two additional scenarios: MinSharedUnc (same as MinShared, but unconstrained by intrinsic costs of expressing too many MHC variants, i.e. viability directly proportional to the number of presented pathogens) and PropSharedUnc (same as PropShared, but unconstrained). The results are qualitatively similar to scenarios with the intrinsic costs, and are only presented in electronic supplementary material, figure S1. For comparison between all the mating scenarios, see electronic supplementary material, table S1.

As pathogen diversity and mutation rate can affect the evolution of MHC copy number [17], we have evaluated the above scenarios at two pathogen mutation rates (10⁻⁵, 5 × 10⁻⁵ per single bit per reproduction) and three levels of pathogen diversity (4, 8 and 16 species). Each species was initialized as a single string, but during coevolution with hosts, most species would consist of several haplotypes, with large differences maintained between sequences in different species throughout simulations [17]. These parameter combinations lead to the evolution of the number of MHC loci that is representative of a wide range of vertebrate species; increasing parasite diversity above 16 species caused only negligible expansion of MHC gene family, but slowed computation considerably. The parametrization is given in electronic supplementary material, table S2. The source code of the model can be found at https://github.com/pbentkowski/MHC_Evolution.

For evaluation purposes, we considered the last 1250 host generation when the dynamics of the host–parasite coevolution stabilized in terms of the numbers of MHC variants in both populations and individuals (except for some scenarios with no penalty factor for large MHC type number, see above). For that period and for each run, we calculated the mean number of unique MHC variants on both chromosomes (INV, individual number of variants) by first averaging across individuals at a given generation, and then taking the mean over 1250 latest host generations. Simulation results were analysed with linear models, with an average INV as a response variable, and pathogen mutation rate and pathogen diversity as fixed factors. Statistical analyses were done in R 3.4.2 [46].

Furthermore, we explored whether mate choice rules we simulated lead to non-random mating with respect to the number of MHC variants. To standardize this between scenarios leading to different INV values, we calculated the mean INV in the pre-mating but post-parasite host population, and after the mating, we recorded the INV for each reproducing pair. Subsequently, for each choosing individual (with known INV), we calculated the difference between INV in the selected partner and population mean INV. We averaged this number for each number of MHC variants in the choosing partner across the last 1250 generations and all repetitions of simulations to obtain Δ partner NV (for random mating Δ = 0). Results were analysed with linear models, with Δ partner NV as a dependent variable, chooser NV and mutation rate as fixed factors. The model was weighted by the number of cases for a given chooser NV (reflected as datapoint size in figure 2).

Figure 2.

The association between the number of MHC variants in a choosing individual (chooser NV) and the deviation of the number of variants in a selected partner from population average (Δ partner NV) under three MHC-based mating rules (see electronic supplementary material, table S1, for the overview of mating rules) and two pathogen mutation rates (left 10⁻⁵; right 5 × 10⁻⁵ per bit). The dot size corresponds to the $\sqrt{(n)}$, where n is the number of cases for a given chooser NV. Here, runs with 16 pathogen species in the system are shown. For other levels of pathogen diversity (4 and 8 pathogen species), see electronic supplementary material, figures S4 and S5. (Online version in colour.)

3. Results

In our simulations, MHC-based mating preferences had a significant effect on the number of variants per individual (INV) (figure 1), which reflected the number of variants per chromosome (electronic supplementary material, figure S2). The effect was greatly dependent on the mechanistic basis of the MHC-based mating preferences. The mating scenario in which pairs were more likely to mate if they shared fewer alleles (MinShared) strongly selected against MHC expansion (figure 1), even if there was no other cost of possessing additional copy numbers (electronic supplementary material, figure S1). The linear model revealed a three-way interaction between mating preferences, pathogen diversity and pathogen mutation rate, but the interaction was driven by the effect of the two latter variables on INV in the Random scenario (electronic supplementary material, table S3), as previously reported [17]. MinShared algorithm kept INV low across other parameters values (figure 1). Under PropShared scenario, mate choice increased INV, compared to the Random scenario (figure 1), except for a parameter combination with low pathogen diversity and high pathogen mutation rate, as reflected in a significant three-way interaction (electronic supplementary material, table S4). The increase in INV compared to random mating was even more pronounced under the MaxDiffer scenario (figure 1; electronic supplementary material, table S5). Numbers of MHC variants per population reflected average INV in a population (electronic supplementary material, figure S3).

Figure 1.

The effect of MHC-based mating rules minimizing the number (MinShared), minimizing the proportion (PropShared) of shared variants and maximizing the number of different variants (MaxDiffer), on the number of unique MHC variants in an individual, compared to random mating (Random). For each mating rule, two pathogen mutation rates were tested (probability of mutation 10⁻⁵ and 5 × 10⁻⁵ per bit during reproduction) and three levels of pathogen diversity (4, 8 and 16 pathogen species that infect the hosts). Boxplots show medians and quartiles from 20 runs of each scenario. See electronic supplementary material, table S1, for the overview of mating rules. Extended results can be viewed in electronic supplementary material, figures S1–S3. (Online version in colour.)

We explored whether mating rules with respect to MHC similarity lead to correlations in NV between mating partners. We found that irrespective of the mating rule, the deviation of the mating partner's NV from the population mean NV was negatively correlated with NV in the choosing partner (figure 2; electronic supplementary material, tables S6–S8). In the MinShared scenario, there was additionally a significant NV * mutation rate interaction, with the slope being significantly steeper for lower pathogen mutation rate in the MinShared scenario (electronic supplementary material, table S7), and in the MaxDiffer scenario, the relationship was significantly nonlinear (electronic supplementary material, table S8).

4. Discussion

Theoretical studies demonstrated that preferences for MHC-dissimilar partners could significantly affect the evolution of MHC polymorphism, increasing it over the level naturally selected by parasites [9,11]. Here, we show that MHC-based mate choice has potentially even more profound consequences for the evolution of the number of MHC genes in the genome.

However, the direction of the effect depends on the detailed mechanism underlying mating preferences. If the mechanism is based on avoidance of sharing identical alleles (our MinShare scenario), then expansion of the MHC gene family is strongly opposed. Remarkably, in MinShare scenario, individual MHC diversity remained very low, compared to random mating, even in the absence of other mechanisms opposing MHC genomic expansion. The outcome is very different, however, if the proportion of shared alleles is minimized (our PropShared scenario), i.e. when the suppression of mating due to shared alleles can be counterbalanced by the stimulation of mating owing to different alleles. Under this scenario, each additional MHC variant in the genome acquired by diversification of a duplicated locus would increase chances of finding a mate. Consequently, our PropShared scenario selected for enhanced MHC expansion compared to random mating scenario (with both scenarios requiring intrinsic costs to MHC genomic expansion to stabilize). Our results thus suggest that differences in mechanisms of MHC-based mating preferences may contribute to striking interspecific variation in individual MHC diversity, comprising from a few [29,47] to several dozens of loci [21,48].

Despite being crucial for predicting the consequences for the evolution of MHC gene number, the discrimination between alternative mechanisms underlying mate choice for dissimilar mates have attracted little attention in experimental work. Researchers usually use just one of the possible measures of MHC similarity, such as proportion of shared alleles [35], the number of alleles which are different [49] or ones that are shared [50] between mates. We advocate that in future studies, these three measures are examined simultaneously. The actual rule guiding partner choice in a given species can be inferred by testing which of these measures best explains the observed data. Empirical studies manipulating MHC-related signals on which mate choice is based can provide the best insight into the mechanism. Mice, fish and humans perceive ligands of MHC molecules, indicating the identity of MHC in the way similar to casts revealing the shape of moulds [51]. As to how this translates into the mechanistic basis of mating preferences, however, is only partly understood. Spiking water with additional MHC molecules changes female stickleback preferences towards males [52]. However, it would be interesting to know if adding ligands identical to the ones bound by self-MHC to water conditioned with MHC ligands from unrelated males could suppress willingness to mate, as would have been expected if kin-avoidance would be a principal mechanism driving the preferences. Finally, it is crucial to bear in mind that MHC-based mate choice can often be cryptic and detectable only after examining genotypes of the progeny in comparison with its parents [39,49].

Investigating detailed mechanisms of MHC-based mate choice is not only crucial for testing predictions of our model but could also shed light on selective forces acting on the evolution of such preferences. For example, avoiding individuals sharing MHC alleles might result in a number of alleles in progeny lower than optimum in terms of resistance to pathogens, whereas a mating rule based on minimal proportion, and even more so the rule based on maximal number of different variants, selects for higher number of loci than optimal in the absence of MHC-based mate choice (figure 1). However, costs of expressing higher or lower number of MHC variants compared to naturally selected optimum may be outweighed by benefits of avoiding sib-mating when inbreeding risk is substantial, and the costs of inbreeding are high, as often observed in natural populations [32]. When the risk of incurring a significant inbreeding depression on progeny is low, however, the selection from parasites might become an important factor constraining evolution of MHC-based mating rules. Evolution of optimal mating rules under different combinations of inbreeding risk, inbreeding depression and selection from pathogens is a promising target of future theoretical and empirical investigation.

Meta-analyses [8,53] highlighted that mating preferences towards individuals with high MHC diversity (or heterozygotes in a single-locus case) are more taxonomically widespread than those for MHC dissimilarity (note, however, that under our PropShared rule, individuals with high MHC diversity would also be preferred as mates on average, so distinction between preferences for MHC dissimilarity versus diversity is not always obvious in empirical data). We have not considered such preferences in this paper, as it is obvious that such preferences should select for increased number of MHC variants in the genome. Yet, in the future work, it would be interesting to compare the effectiveness of such rule in driving expansion of MHC gene family, compared to scenarios considered here. More sophisticated scenarios, such as a mate choice for the optimal number of MHC loci [52,54], could also be explored. However, such scenarios assume that mate choice aims at maximizing offspring immunocompetence so that the optimal number of MHC variants in progeny coincides with the number naturally selected by parasites. Thus, mate choice for MHC optimality should not affect INV evolution. Nevertheless, it would be interesting to explore under which circumstances such a mating rule could evolve. In this context, it is noteworthy that our simple mate choice rules resulted in patterns of pairings which are reminiscent of patterns interpreted as evidence for mate choice for optimal MHC number in offspring. In particular, to achieve optimality via an ‘allele counting’ strategy, individuals with low allele number should choose those with high allele number and vice-versa [55,56]. However, our simulations revealed that such patterns could result from simple mate choice rules for MHC dissimilarity. Therefore, caution should be exercised when inferences of ‘allele counting rules' are based on MHC-based pairing patterns in natural populations (or in simulations modelling such populations). Discrimination between true allele counting and simpler rules we simulated can, however, be done with experimental approaches, such as spiking water with additional MHC ligands [52]. Using such an experimental approach, particularly using ligands not occurring in a given population (such as xenogenic ligands used by Milinski et al. [52]) implies that allele counting cannot be a by-product of increased probability of sharing an allele with potential partner by a chooser with many alleles.

Concluding, our simulations showed that mating preferences for MHC dissimilarity might have significant consequences for the evolution of MHC gene family expansion in vertebrate genomes. Differences in specific mechanisms of mating preferences can lead to strikingly different outcomes and have the potential to explain interspecific differences in the number of MHC genes. Disentangling mating rules is thus an essential aim for future research that will provide data for comparative studies testing predictions of our model. A better understanding of the rules will also help to reveal selective pressures that are driving the evolution of MHC-based mating preferences.

Data accessibility

The source code of the model can be found at https://github.com/pbentkowski/MHC_Evolution. The data are deposited at the Dryad Digital Repository: https://doi.org/10.5061/dryad.vmcvdncpc [57].

Authors' contributions

J.R. conceived of the study, designed the general algorithm for simulations, carried out the statistical analyses and drafted the manuscript; P.B. designed, coded and run the simulations, analysed the simulations' output, reviewed and edited the manuscript. Both authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Competing interests

We declare we have no competing interests.

Funding

This research was supported by Polish National Science Centre (grant no. UMO-341 2013/08/A/NZ8/00153) awarded to J.R. Computations were done at the Poznań Supercomputing and Networking Centre (PCSS).