Cooperation is predicted to evolve more easily when social partners are genetically related [1–5]. In many species, individuals are able to assess the genetic similarity of conspecifics using heritable phenotypic cues, for example by comparing their own cues to those of their social partners (self-referent phenotype matching; ), or by directly identifying individuals with the same genotype as them at a particular locus ("green beard" recognition; ). Such cue-dependent cooperation is thought to be common in all domains of life, including plants , fungi , bacteria , vertebrates , insects , slime moulds  and sessile marine invertebrates .
Cooperation based on genetic cues is a conundrum because it requires polymorphic recognition loci, yet cooperation is predicted to erode this genetic variation (e.g. [15–17]). This problem, sometimes termed Crozier’s paradox, applies whenever individuals with common recognition cues receive greater average fitness returns from social interactions. For example, individuals with common cues might be aggressively rejected less often [15, 18, 19], or might receive altruism from a greater proportion of the population [17, 20]. Disproportionate fitness benefits for individuals with common recognition alleles should produce positive frequency-dependent selection at recognition loci, depleting the genetic variance necessary for kin recognition.
The origin and maintenance of polymorphic genetic recognition cues remains incompletely understood despite substantial theoretical and empirical research (e.g. [15, 17, 21–24]). Previous models of cue-dependent cooperation have treated the cooperative behaviour and the recognition cue phenotype either as products of a single locus (effectively a green beard locus; [15, 25–29]) or separate loci [17, 19, 20, 30]. In one such two-locus model, individual recognition alleles increased in frequency when in linkage disequilibrium with the cooperative allele, but were prevented from fixing by non-cooperating "freeloaders" with the same recognition allele . Intermediate recombination rates produced cycles in which cooperation and multiple recognition alleles could coexist, suggesting that genetic recognition systems can remain somewhat stable under some conditions. By contrast, subsequent analytical models and simulations suggested that a very restrictive combination of high population structure, low recombination and frequent mutation is required to preserve recognition cue diversity . Hence, the paradox of highly variable genetic recognition cues remains largely unresolved, at least under the assumption that recognition loci function only in the selection of social partners. The prevailing consensus is therefore that recognition loci likely have more than one function (i.e. they are pleiotropic), and that the pleiotropic function introduces negative frequency-dependent selection that preserves recognition cue diversity [9, 15, 17, 23].
Such negative frequency-dependent selection could be provided by pathogens and parasites, for example when parasites evolve to preferentially infect hosts with common recognition cues (e.g. [17, 31, 32]). Some loci used in kin recognition also affect parasite resistance, notably the major histocompatibility complex (MHC) of vertebrates (e.g. ). Additionally, social insects use heritable chemical cues to identify colony members  as well as allospecific social parasites; these parasites may evolve to chemically mimic the commonest host genotypes [34–36]. Host-parasite interactions may similarly maintain cue diversity in parasitic species. Copidosoma floridanum parasitoid larvae identify and attack unrelated larvae present inside the same host using cues present on a membrane surrounding the larvae . Common cues might therefore confer protection against unrelated larvae, but the membrane also defends larvae from the host’s immune system, which may select for rare phenotypes . Rare cues have also been proposed to improve the precision of intra-specific recognition and thereby reduce the frequency of costly errors [9, 14, 29].
Recognition cues that inform social behaviour might also be used in the context of mate or gamete choice (e.g. [17, 39]). A pleiotropic function in mate choice might preserve genetic polymorphism at recognition loci by at least four non-exclusive mechanisms.
Firstly, individuals with rare recognition cues might be attractive to or compatible with a greater proportion of potential mates, for example if individuals avoid inbreeding by discriminating against potential mates with similar recognition cues, or if matings between partners with the same recognition locus genotype are infertile . Individuals with rare cues would therefore have a sexually-selected advantage that might counterbalance their disadvantage in social interactions. This hypothesis reflects the well-known population genetic result that disassortative mating can increase genetic diversity provided that it creates inequalities in mating success .
Secondly, disassortative mating with respect to recognition loci should increase their heterozygosity. Assuming that cooperation increases the fitness of common genotypes, this would increase the fitness of heterozygotes, promoting genetic polymorphism. This hypothesis [proposed in 15] assumes that cooperation occurs primarily or solely between individuals that share both alleles at the recognition locus (which we call "2-allele matching").
Thirdly, for species in which cooperation occurs between individuals sharing at least one recognition allele ("1-allele matching"), we suggest that the increase in heterozygosity caused by disassortative mating could provide "hiding places" for rare recognition alleles. Under 1-allele matching, rare alleles may receive substantial amounts of cooperation when sharing a body with a common allele. Disassortative mating causes rare alleles to exist as heterozygotes even more often than predicted under random mating. This could prevent rare alleles from being lost from the population, increasing the ability of mutation and migration to maintain recognition locus diversity .
Fourthly, we propose that disassortative mating for recognition cues may indirectly lead to disassortative mating for condition. This is because matings between individuals with common and rare cues should be more frequent under disassortative mating, and individuals with common cues will tend to be in better condition since they receive greater average payoffs in social interactions. Under disassortative mating, individuals with rare cues might tend to have a mating partner in better condition than themselves and vice versa, potentially improving the relative fitness of individuals with rare recognition alleles.
Previous models of genetic kin recognition have mostly assumed random mating and haploidy (precluding heterozygosity), so the effect of mating systems on Crozier’s paradox is currently unclear. Here, we explore in detail the hypothesis that mate choice affects the evolution of cooperation based on genetic recognition cues using individual-based simulation. We also review the available literature in order to evaluate the relative importance of mate choice in stabilising cue-dependent cooperation in diverse taxa.