In several chapters to this point, we have referred to “good genes” and “genetic quality.” These topics play more important roles in chapters to come. This chapter explains the nature of good genes, considers the evolutionary processes that give rise to and maintain variation in genetic quality in natural populations, and discusses literature on potential markers of quality. In chapter 6, we argued that women’s ornaments have evolved, through sexual selection, as signals of superior quality and condition, which, in part, may reflect genetic quality. We elaborate that argument here. We also discuss potential signals of genetic quality that men possess by virtue of evolution via sexual selection.
The processes that maintain genetic variation in fitness compose a broad topic within evolutionary biology. For present purposes, we are interested in specific implications of this genetic variation: implications for mate choice. In sexually reproducing species, individuals receive two kinds of resources from their mates that affect their own fecundity or that of offspring, and thereby their own fitness: DNA (which combines with mate choosers’ own in offspring) and nongenetic material benefits (delivered either as a result of adaptation in mates to offer benefits or as by-products of mates’ adaptations designed for functions other than to deliver benefits). Selection favors choice of mates that, all else equal, provide DNA that promotes offspring (and hence mate choosers’ own) fitness. In his classic paper on sexual selection, Trivers (1972) introduced the term “good genes” to refer to choice of individuals who possess alleles that could benefit choosers’ offspring.
Sexual selection theorists and researchers now distinguish three different types of genetic benefits relevant to mate choice (Jennions & Petrie, 2000):
First, mates can offer intrinsic good genes. Individuals who offer intrinsic good genes possess alleles that are associated, on average, with relatively high fitness (over several generations or more). On a continuum of genetic variation in fitness, these individuals are at the high end of the distribution. Put otherwise, they possess alleles that are currently favored by directional selection. They hence can pass on alleles that provide fitness advantages to any mate chooser’s offspring (and subsequent descendants). Meaningful variation in the extent to which mates offer intrinsic good genes implies meaningful additive genetic variation in fitness (i.e., some alleles are in fact favored by directional selection, and others disfavored by it). In the next section, we discuss how much genetic variation in fitness persists in natural populations and the processes that maintain it.
Second, mates can offer compatible (or complementary) genes. These individuals possess specific alleles that work well with the mate chooser’s own alleles (either at the same locus or at different loci) to promote fitness in offspring but do not work well with all mate choosers’ alleles (e.g., Zeh & Zeh, 2001). Variation in the extent to which mates offer compatible genes implies meaningful nonadditive genetic variation in fitness. That is, it implies that, at some influential genetic loci, heterozygotes possess mean fitness not equal to the mean fitness that homozygotes possess (e.g., heterozygotes possess higher fitness than homozygotes) or that there exist epistatic effects on fitness—nonadditive interactive effects on fitness involving allelic variation at different loci. In the third section of this chapter, we discuss candidate exemplars of these kinds of effects.
Third, mates can offer diverse genes. As a number of scholars have noted (e.g., Ellison, 1994), lineage extinctions are evolutionarily important events. A lineage extinction occurs when all of a focal individual’s descendants (whether at the first generation or the nth generation) fail to reproduce. If the environments to which descendents are exposed are unpredictably variable in nature, individuals may enhance fitness by diversifying offspring, including through genetic diversification. (More to the relevant point, an allele that predisposed an individual to genetically diversify offspring could be favored by selection over multiple generations of descendents.) Diversification is a form of bet hedging. Given environmental uncertainty, one’s descendents could get “lucky” (possess alleles favored in future environments) or “unlucky” (possess alleles disfavored by future environments). Genetic diversification is purportedly a means of reducing variance of the reproductive outcomes of descendents in the face of such uncertainty, which can be favored through avoidance of lineage extinction, even if the mean of those outcomes is not altered or slightly lower (e.g., Gillespie, 1977). In the fourth section of this chapter, we consider these and other arguments.
The Additive Genetic Coefficient of Variation Intrinsic good genes, once again, are alleles favored by directional selection (or, in recent evolutionary history, have been favored). Collectively, their aggregate effects on fitness contribute to and account for additive genetic variation in fitness. One fundamental question concerning intrinsic good genes and their importance in mate choice is how much additive genetic variation in fitness exists in natural populations.
For any quantitative trait measured on a ratio scale, additive genetic variance can be quantified with the additive genetic coefficient of variation, or CVA. (A ratio scale is a measure that has a meaningful zero point—a zero that implies no quantity of the trait—and units of measurement that linearly relate to quantities of the trait. Height measured in centimeters is a ratio-level scale.) The CVA is the square root of the additive genetic variance (in a sense, the “additive genetic standard deviation”) of the trait divided by the trait mean. Typically, this value is multiplied times 100, which converts a proportionate measure into a percentage measure. (The additive genetic variance is the variance in the trait associated with simple additive effects of alleles, aggregated across all loci at which allelic variation affects the trait. It does not reflect all genotypic effects on traits, for it does not reflect dominance [e.g., heterosis] and most epistatic effects. Because dominance and most epistatic effects are not heritable in a strict sense [effects in the parental genotype do not predict effects in the offspring genotype], they are not relevant to measuring genetic variation pertinent to mate choice for intrinsic good genes. See Fisher, 1930, and Falconer & Mackay, 1996.)
A simple example can illustrate the calculation and meaning of the CVA. Suppose men’s height averaged 70 inches in a population, with a standard deviation of 3 inches. Suppose further that the heritability of men’s height in that population is 0.85. The trait variance, then, is 9, and the additive genetic variance is 7.65 (9 × .85). The CVA is the square root of 7.65 divided by 70, times 100: approximately 4. This means that the standard deviation in height due only to differences associated with allelic effects on height is 4% the mean of height. (Though we have chosen hypothetical values, men’s height does appear to have a CVA close to 4, at least in Western populations; see, e.g., Miller & Penke, 2007.)
The implications of the size of the CVA for individual differences can be appreciated further in terms of a ratio of the values on the trait possessed by individuals near the top of the distribution (say, 2 standard deviations above the mean) and the values on the trait possessed by individuals near the bottom of the distribution (say, 2 standard deviations below the mean). In this instance, it would be approximately 1.17 (most readily calculated as [100 + 2(4)]/[100 – 2(4)]). That is, individuals who, by virtue of the independent effects of their alleles, are near the top of the distribution in terms of being predisposed to being tall are 17% taller than individuals who are near the bottom of the distribution in terms of being predisposed by their alleles to being tall. One can also say that those near the top of the distribution in being predisposed to being tall are 8% taller than the average person ([100 + 2(4)]/100 = 1.08).
What Is the Additive Genetic Variance of Fitness? The size of the CVA of fitness has major implications for the importance of selecting a mate for intrinsic good genes. If the CVA of fitness is 4, similar to that of height, then a mate at the top 2% of the distribution in genetic fitness, paired with a randomly chosen individual, would produce offspring that are, on average, 4% more fit than offspring produced by two randomly chosen individuals or two individuals of average genetic fitness. (Here, “genetic fitness” is shorthand for variation in predisposition to high fitness by virtue of the additive effects of individuals’ alleles. Because individuals contribute only half their alleles to an offspring, the mean fitness advantage of one’s offspring due to additive genetic effects, when mates are individuals randomly chosen from the population, is only half that possessed by the individual him- or herself; see Falconer & Mackay, 1996.) That advantage might be enough to drive selection for adaptations for choice of mates who possess good genes—but perhaps not, if the costs of that mate choice (e.g., in currencies of waiting time or costs of resisting other suitors) are considerable.
Until the early 1990s, one major reason to question whether mate choice for intrinsic good genes could be profitable was serious doubt that the amount of genetic variance in fitness in natural populations could render choice for good genes profitable. Fisher’s (1930) fundamental theorem of natural selection states that directional selection on a trait reduces its additive genetic variation. Fitness, by definition under directional selection, should thus have most of its additive genetic variation exhausted by selection, “which creates a serious difficulty for the good genes hypothesis” (Charlesworth, 1987, p. 22; for similar expressions of skepticism, see Maynard Smith, 1978; Partridge, 1983; Taylor & Williams, 1982). Naturally, if there is no additive genetic fitness variation in a pool of potential mates, there is no reason to choose one mate over any other for additive genetic benefits. And even if some small amount of genetic variation in fitness persists, the costs of good-genes mate choice may not exceed the benefits of obtaining a mate with relatively good genes.
Fitness itself is a trait difficult to measure. One way to estimate the genetic variance in fitness is to estimate the genetic variance of fitness-relevant traits (fitness components) purportedly under strong directional selection. Two major fitness components are fecundity and longevity. Fifteen years ago, David Houle (1992) first brought attention to what was at the time a very surprising fact: Fitness components appear to actually have CVAs that are substantially larger, not smaller, than those of many ordinary traits (e.g., morphological traits such as height). To draw this conclusion, Houle (1992) examined scores of studies in the literature, many on Drosophila. Whereas ordinary morphological traits or traits under stabilizing selection have CVAs less than 5, Houle found that fitness traits typically had CVAs of 10+. Sgro and Hoffman (1998) reported similar findings on Drosophila: Whereas fecundity had a mean CVA of 20, wing length had a CVA < 2 (see also Hughes, 1995). Using direct and indirect means of estimation based on observations of a number of different species, Burt (1995) estimated the CVA of fitness itself to typically be 10–30. (For other relevant findings, see Gardner, Fowler, Barton, & Partridge, 2005.)
Overall, more recent longitudinal studies on natural populations of relatively long-lived organisms reveal similar patterns. In males, lifetime reproductive success (RS) has an estimated CVA of 17, 16, and 32 in collared flycatchers (Merilä & Sheldon, 2000), great tits (McCleery et al., 2004), and red deer (Kruuk et al., 2000), respectively. (Merilä and Sheldon, 2000, noted that the value for flycatchers is probably an underestimate, as extra-pair paternity was not assessed in the study. But Brommer, Kirkpatrick, Qvarnström, & Gustafsson, 2007, reported a lower estimate.) Female RS in these same species was estimated to have a CVA of 29, 6, and 0. Sampling variability renders each individual estimate very imprecise; the variation in the estimates may not be highly meaningful. The means across studies of 22 (for males) and 12 (for females)—considerably more stable—are substantial, however, and in the range anticipated by Houle (1992) and Burt (1995). In red deer, male body size is a strong predictor of reproductive success and also had a substantial CVA (32; Kruuk et al., 2000). By contrast, ordinary morphological traits not predictive (in a linear fashion) of RS in these species (e.g., tarsus length and wing length in birds, jaw length in red deer) had low CVA values (always < 4 and, on average, close to 2).
These values mean that there could be very substantial benefits to choice based on intrinsic good genes. If the CVA of fitness is 20, mating with an individual two standard deviations greater than the mean of genetic fitness translates into 20% greater expected fitness of offspring, relative to mating with an individual at the mean of genetic fitness (all else equal). But, of course, for mate choice for good genes to actually work in this way, mate choosers need to be able to rely on a reasonably good indicator or signal of intrinsic genetic fitness (not merely overall fitness, which contains much variation due to nongenetic and nonadditive genetic sources) to identify individuals who possess good genes. Later, we discuss this critical issue in more detail.
What evolutionary processes cause and maintain genetic variation in fitness? One major cause is mutation. Deleterious mutations in germ cells caused by DNA copying errors occur at each genetic locus at some small probability. The mean effects on fitness of mutations vary; whereas some deleterious mutations are lethal, most probably have very minor effects (no more than a few percentage points’ effect on fitness in the heterozygous state; e.g., Eyre-Walker, Woolfit, & Phelps, 2006). Mutations that cause death prior to reproduction (or that prevent reproduction for any other reason), even in the heterozygous state, are eliminated by selection immediately. For all other deleterious mutations, there is a nonzero probability that they will pass through to the following generation, with the probability proportional to the mutation’s mean harmful effect on fitness. Those mutated alleles with weak effects on fitness may persist for many generations, affecting many individuals, before being eliminated by selection. (Indeed, for mutations with very weak effects, there is some small probability that they will become fixed in the population through random drift, particularly when the size of the population is small; see Lynch et al., 1999.) In an idealized population at equilibrium, the rate at which negative effects on fitness due to mutation are eliminated by selection per generation is equal to the rate at which negative effects on fitness are introduced by fresh mutations per generation; the population is said to be in mutation-selection balance (e.g., Fisher, 1930). At equilibrium, there exist a certain number of mutations in the population (aggregated across all loci), which have a distribution of fitness effects. Not all individuals, however, have precisely the same number of mutations or mutations with precisely the same effects on fitness. (Naturally, nearly all mutations are ones several to many generations old, which individuals presently in the population inherited, not fresh ones that originated in the germ cells that produced current individuals; as Keller & Miller, 2006, put it, “Most mutations are a family legacy, not an individual foible”; p. 397.) The variation across individuals in the effects of the mutations they possess at equilibrium produces a characteristic standing genetic variation in fitness due to deleterious mutation (Fisher, 1930; Houle, 1992). Genetic variation in fitness can also be generated by positive selection for rare favorable mutations not yet at fixation, but, because favorable mutations are rare, they probably account for much less variation than do deleterious mutations (see, e.g., Smith & Eyre-Walker, 2002).
How much of the CVA in fitness in natural populations is due to mutation-selection balance? This has turned out to be a question difficult to answer. Laboratory studies have produced variable answers. Charlesworth (1990) estimated that mutation could account for a CVA in fitness in Drosophila melanogaster of 17. Charlesworth and Hughes’s (2000) estimate for the same species was 8. These estimates rely on assumptions. Furthermore, the amount of fitness variation due to deleterious mutation could vary across species, or even across populations of the same species. It is thought that it is probably larger in vertebrates than nonvertebrates. Though the mean mutation rate per locus in humans appears to be only 60% of what it is in Drosophila (despite many more germline cell divisions per generation in human males; Drost & Lee, 1995; see also Crow, 1997), the number of amino-acid coding genes in the mammalian genome is 4–5 times larger. Hence the number of new deleterious mutations per genome per generation in humans is likely at least double what it is in Drosophila (see Lynch et al., 1999).
The genetic variance in fitness maintained by mutation-selection balance is partly a function of U, the deleterious mutation rate per diploid genome per generation, which is itself a function of the mean rate of mutation per locus and the total number of functional loci in the genome. Eyre-Walker and Keightley (1999) (using indirect methods comparing the genomes of humans with those of close phylogenetic relatives) estimated U to be 1.6 in amino acid coding regions of the human genome (which Eyre-Walker et al., 2006, later revised to be 1.8), a value identical to the minimum value Lynch, Latta, Hicks, and Giorgiana (1998) conjectured based on estimated mutation rates per genome per germline cell division. (Keightley, Lercher, and Eyre-Walker, 2005, note, however, that many mutations of relatively weak effect individually—but with possibly meaningful aggregate effects—have also accumulated in noncoding, regulatory gene regions of the human genome.) Nachman and Crowell (2000) estimated U to be at least 2.0 and possibly greater in humans (see also Kondrashov, 2001).
The genetic variance in fitness maintained by mutation-selection balance also partly depends on the distribution of fitness effects of those mutations and interactions between mutations’ effects. Eyre-Walker et al. (2006) recently estimated that most mildly deleterious mutations in humans have historically had small effects on fitness, with a mean effect no more than a few percentage points. Their estimate of 4.3% negative effect, they note, may overestimate the true value by as much as threefold. (A value of about 2% would be close to estimates for other species; Lynch et al., 1999.)
If U is 2 and the mean effect of mutations on fitness has been 1–3% in humans, the population degrades in fitness by 2–6% per generation by fresh deleterious mutation alone. For a population at mutation-selection equilibrium (as humans may have been historically), with removal of mutations per generation having a net effect on fitness equal (but opposite in direction) to that owing to new mutations, selection against mutations would increase mean fitness 2–6% per generation (Burt, 1995). According to Fisher’s fundamental theorem, the proportional increase in fitness in one generation due to selection (the evolvability or IA of fitness) is equal to the additive genetic variance standardized by mean fitness squared (see Houle, 1992). The square root of this value times 100 gives the CVA (Houle, 1992). In this instance, √.02 × 100 = 15 and √.06 × 100 = 24. That is, empirical estimates of U and the mean effect of mutation, together with idealized assumptions, yield an estimated CVA of fitness due to mutation alone of approximately 20 ± 5—values similar to what Burt (1995) estimated for the CVA of fitness (due to all causes) itself.
This estimate does rely on assumptions. More precise estimates await further research (see, for instance, Gardner et al., 2005; Kondrashov, 2001). Even if the true CVA of fitness due to variation in effects of mutations on fitness alone were only half of this value (10), however, it would be substantial. In all likelihood, a considerable proportion of the genetic variance in fitness in human populations (as well as, most likely, natural populations of other sexual species) has historically been due to variation in mutational effects on fitness across individuals (see, for instance, Lynch et al., 1999).
In light of selection on fitness, how does mutation produce and maintain so much variation in fitness? Houle et al. (1996) argue that the amount of variation generated and maintained in a trait by mutation alone is predicted by its mutational target size, the number of loci at which mutations could affect the trait (see also Houle, 1998). The mutational target size of fitness itself is very large—effectively, the entire functional genome. Though allelic variation due to deleterious mutation at any particular locus has a very small effect on fitness, the cumulative effects, aggregated over many loci at which mutations have effects, add up to a substantial total effect on fitness.
As noted earlier, selection, absent any countervailing process, would eliminate all additive genetic variation in fitness (Fisher, 1930). Deleterious mutation (changes in the genome that cause lack of adaptation of individuals to their environments) is one countervailing process that prevents the complete elimination of genetic fitness variation. Rapid and recurring change in the selective environment (which similarly causes individuals to become less well adapted to their environments) is another. If the direction of selection on allelic variation at an individual locus changes at a sufficiently rapid and recurring rate, at least some meaningful proportion of the time no one allele will be fixed (or nearly fixed) in the population at that locus; allelic variation translating into fitness variation will often be found. If many loci are affected by rapid and recurring changes, the sum total effect on genetic fitness variation could be considerable (e.g., Eshel & Hamilton, 1984). In recent investigations examining the relative fitnesses of wild-type chromosomes in Drosophila, Gardner et al. (2005) found very substantial fitness differences. Perhaps even more notably, however, they found that, through time, patterns of selection for individual chromosomes systematically varied. These patterns imply meaningful gene × (temporally variable) environment interactions. Fluctuating selection, they suggested, may maintain a meaningful proportion of the genetic variation they observed. Some theorists strongly suspect that processes other than simply mutation-selection balance maintain meaningful variation; contrary to expectation of the mutation-selection model that deleterious alleles will be represented in rare frequencies, many polymorphisms found in nature involve multiple alleles of at least intermediate frequency (though their effects on fitness is typically unknown; see Turelli & Barton, 2004, and references therein).
Hamilton and Zuk (1982; Hamilton, 1980, 1982) famously proposed that antagonistic coevolution of hosts and pathogens entails that both hosts and pathogens are sources of relatively rapid changes in the selective environments of the other. That is, evolution of pathogens, in response to host adaptation to them, entails changing selection pressures on hosts. And evolution of hosts, in response to pathogen adaptation to them, entails changing selection pressures on pathogens. (More generally, see also Van Valen, 1973, on Red Queen processes, which he named after the character in Alice in Wonderland who needed to keep running simply to stay in the same place.) Recurrent fluctuating selection on host organisms could maintain genetic variation in fitness. Based on simulations, Eshel and Hamilton (1984) proposed that coevolution with pathogens possessing intermediate intergeneration intervals—such as human macroparasites—can, in theory, best maintain heritable fitness variation in hosts. (Hamilton and Zuk, 1982, also claimed that negative frequency-dependent selection—selection tending to favor rare alleles—operates and plays a critical role in maintaining allelic variation in this particular Red Queen process.)
In theory, any host allele coding for a protein to which a pathogen could adapt could be subject to fluctuating selection (e.g., Tooby, 1982). In practice, selection on host alleles coding for components of immune defense seems the most likely candidate to be “yanked around” by coevolution with pathogens. Major histocompatibility complex (MHC) alleles code for cell-surface markers that components of the immune system use to detect self-and non-self-peptides. At particular points in time, individual alleles may be beneficial in defense against particular pathogens (e.g., Lohm et al., 2002). Temporally varying dynamics of host and pathogen populations could lead to changes in the prevalence of individual MHC alleles, as well as maintenance of MHC genetic variation, particularly in concert with frequency-dependent selection (Hedrick, 2002; though, as we discuss later, heterozygote superiority may well importantly contribute to the maintenance of MHC diversity in humans as well; see Hedrick, 1998; Black & Hedrick, 1997; see also Geise & Hedrick, 2003). Other components of the immune system could similarly be involved in host-pathogen coevolution.
Some of the most important coevolutionary processes maintaining genetic variation in fitness may involve antagonistic coevolution at different loci within a single species—what Rice and Holland (1997) refer to as interlocus contest evolution, or the intraspecific Red Queen.
An example of an intraspecific Red Queen process is maternal-fetal coevolution (see Rice & Holland, 1997, for discussion of other instances). Fetal genes maximally benefit from a greater flow of nutrients from the mother than the level maximizing maternal fitness (Haig, 1993; Trivers, 1974). Hence, a newly arising allele that, when expressed in fetuses, increases the flow of nutrients from mothers to the fetus may be selected and spread. And a newly arising allele at a different locus that, when expressed in mothers, decreases flow of nutrients (e.g., because it undermines a new fetal adaptation) may be selected and spread. At any point in time, most loci involved in the maternal-fetal conflict may be monomorphic (with one allele spread to fixation). But a nonnegligible proportion of them could be in a state of transition in which a recently and currently favored allele has not yet fully spread to fixation. If so, some mothers are expected to be better adapted to the conflict than others. And similarly, some fetuses should be better adapted to the conflict than others. In sum, recurrent genetic variation in fitness could be an outcome of maternal-fetal coevolution.
The intraspecific Red Queen process that has received the most attention to date is sexually antagonistic coevolution (e.g., Arnqvist & Rowe, 2005). In a sexually reproducing species, sexual conflicts of interest typically exist. That is, what gives a male advantage in competition with other males may actually decrease the fitness of his female mates relative to other females, and what gives a female an edge in competition with other females may disadvantage her male mates. Hence males may evolve adaptations at the expense of their female mates. In turn, females may evolve counteradaptations at the expense of male mates. Loci at which alleles code for components of male sexually antagonistic adaptations coevolve with loci at which alleles code for components of female sexually antagonistic coevolution (and vice versa). Just as in the scenario involving maternal-fetal coevolution, even if most loci are monomorphic at any point in time, a meaningful proportion may be in a state of transition in which a recently and currently favored allele has not yet spread to fixation. The resulting variation implies within-sex genetic variation in fitness.
In Drosophila, as in many species, males and females have conflicting interests over the mating rate. Males can benefit from inducing a female to mate when it is not in the interests of the female to do so (e.g., when she has already mated or mated with a male of higher quality). Female adaptations to resist male attempts may fuel selection for counteradaptation in males, leading to selection for counter-counteradaptation in females, and so on. Rice (1996) demonstrated this antagonistic process in a laboratory experiment on Drosophila in which a female line was not permitted to evolve, but a male line could evolve in response to females. After 30 generations, males in this line were better able to induce females to remate and, partly as a result, outreproduced males drawn from the original control stock when both types competed for the same females. The fitness of females, however, was lower when they interacted with these males compared with when they interacted with control males (see also Lew & Rice, 2005). Males had evolved adaptations antagonistic to female interests, to which females were prevented from evolving counteradaptations.
Subsequently, Lew, Morrow, and Rice (2006) and Linder and Rice (2005) estimated the standing genetic variance in female ability to resist male courtship. It was significant and, on average across studies, accounted for a remarkable 40% of the genetic variance in female fitness under normal mating conditions. Other studies provide evidence that male abilities to induce females to mate and effectively supplant stored sperm (offensive tactics), as well as enforcing fidelity of mates and preventing displacement of their own sperm (defensive tactics), are also heritable (e.g., Friberg, Lew, Byrne, & Rice, 2005). Some (perhaps much) of the genetic variation in traits involved in antagonistic interactions between males and females and contributing to fitness is probably due to variants (e.g., mutations) affecting overall vigor (see Kokko et al., 2003). The fact that portions of the genetic variance in males’ offensive and defensive tactics were unique, however, suggests that these traits are affected by some genetic loci evolving independently of overall vigor, possibly through sexually antagonistic coevolution (Friberg et al., 2005).
Genetic variation at a locus that results from persistent sexually antagonistic coevolution may affect the fitness of one sex only: that sex whose sexually antagonistic tactics are affected by variation at the locus. Some heritable fitness variation, then, is expected to be sex-specific—transmitted from a parent to same-sex offspring but not to offspring of the other sex.
In chapters 9–12, we offer hypotheses about ways in which men and women have been involved in sexually antagonistic arms races. Possibly, a meaningful amount of genetic variation in male and female fitness (or fitness in ancestral environments) was produced as an outcome of interlocus sexual conflict. How much variation that might be is unknown at this time.
We have described interlocus sexual conflict: coevolution of alleles at different loci within an organism’s genome fueled by conflict between the sexes. Sexually antagonistic selection can also operate at a single locus. In this instance, selection operates on allelic variants in contrary ways when inhabited by the two sexes. That is, whereas one allele is favored over another when carried in males, another allele at the same locus is favored when carried in females. Selection can lead to stable equilibria maintaining variation. Alternatively, one allele may spread due to stronger selection on one sex but impose fitness costs on the other sex carrying it (e.g., Chippendale, Gibson, & Rice, 2001; Rice & Chippendale, 2001). Both sexes may be pulled away from their phenotypic optimum as a result—but, within the sexes, to variable degrees. On traits affected by a single locus at which sexually antagonistic selection operates, “masculine” males may be more fit than “feminine” males, and “feminine” females may be more fit than “masculine” females. For example, variation in prenatal androgen exposure and resultant variation in digit ratios within both sexes (see chapter 6) may be maintained partly by intralocus sexually antagonistic selection (Manning et al. 2000; see also Saino, Leoni, & Romano, 2006). In chapter 6, we suggested that interlocus and intralocus sexual conflicts may importantly cause the maintenance of genetic variation in human ornamentation.
Sexually antagonistic genes cause genetic variation in fitness within the sexes. Because sexually antagonistic alleles are favored in one sex but not the other, however, these genes may not strongly covary with overall heritable fitness: Parents whose sons benefit from the alleles produce daughters that are harmed by them, and vice versa. The presence of sexually antagonistic genes can favor alleles that affect the sex ratio of offspring, such that masculine parents produce relatively more males, and feminine parents produce relatively more females (e.g., Rice & Chippendale, 2001). Evidence indicates that people relatively masculine in certain ways do have more sons and relatively feminine people have more daughters (see Kanazawa & Vandermassen, 2006). An alternative adaptive explanation for sex ratio of offspring effects is that the sex ratio is biased in favor of males when the father has indicators of good genes. In species in which reproductive skew is greater for males than for females (e.g., humans), males are benefited by alleles promoting general fitness more than are females. Hence, men who, in the Kinsey study on sexual behavior, reported that they had a large number of premarital sex partners (perhaps reflecting their attractiveness to females) tended to have more sons than those reporting that they had few premarital partners (Gangestad & Simpson, 1990). (For evidence of similar adaptive sex-ratio alterations in zebra finches, see Burley, 1986; in collared flycatchers, see Ellegren, Gustafsson, & Sheldon, 1996; in reindeer, see Røed et al., 2007. But a meta-analysis on birds yielded no mean effect; Ewen, Cassey, & Møller, 2004.) Future studies may be able to tease apart the alternative explanations.
Because intralocus sexually antagonistic selection can result in genetic differences affecting fitness in opposite directions for these sexes, these differences are thought to possibly obscure mate choosers’ abilities to pick out mates who possess intrinsic good genes for both sexes. (A male who appears to have good genes offers genes benefiting sons but not daughters.) Large amounts of variation in fitness due to sexually antagonistic genes, then, may interfere with selection for mate choice for good genes (Pischedda & Chippendale, 2006).
Sexually antagonistic loci drive the genetic correlation of fitness for the two sexes (the correlation between the genetic differences in each sex affecting reproductive success) to be negative. Several recent studies have estimated these correlations. Chippendale et al. (2001) estimated a genetic correlation of –0.16 for Drosophila melanogaster, Foerster et al. (2007) estimated it to be –0.48 for red deer, and Brommer et al. (2007) found a correlation of –0.85 for collared flycatchers. In red deer, male variance in fitness is much greater than female variance in reproductive success, and, hence, it pays females to choose fit males despite the cost of doing so to female offspring. In collared flycatchers, however, Brommer et al.’s (2007) estimates provide little evidence for good-genes sexual selection.
The degree to which the genetic fitnesses of the sexes are correlated—positively or negatively—in traditional humans remains unknown.
Adaptive mate choice for intrinsic good genes requires additive genetic variance in fitness. Houle’s (1992) and Burt’s (1995) broad generalization that fitness components and, presumably, fitness itself tend to have large additive genetic coefficients of variation in animal populations is almost certainly true, even if occasional exceptions exist. Less well understood are the relative contributions of the evolutionary processes that maintain heritable fitness variation. Mutation-selection balance probably accounts for a substantial amount of this variation, but how much (30%, 60%, 90%?) is unknown and probably varies across species.
Some comparative analyses indicate a role for mutations driving mate choice for good genes. As it happens, male gametes typically have higher rates of fresh mutations due to the greater number of cell divisions in sperm than in eggs (see Miyata, Hayashida, Kuma, Mitsuyasu, & Yasunaga, 1987), which yields stronger sexual selection for good genes in males than in females. In fact, sexual selection on males through female choice appears to be stronger in species with strong male mutational biases (Bartosch-Härlid, Berlin, Smith, Møller, & Ellegren, 2003). And in pair-bonding birds, the extra-pair paternity rate is predicted by the mutation rate (Møller & Cuervo, 2004). As we discuss in chapter 10, extra-pair copulation may be one means by which females in pair-bonded species exert sire choice favoring good genes. (See Petrie and Roberts, 2007, however, for a model that suggests that strong sexual selection can increase the mutation rate, rather than vice versa.)
Antagonistic coevolutionary processes probably fuel meaningful variation as well, at least in some species. How much they have contributed to fitness variation during hominin evolution is not known. Nor are the relative contributions of interspecific (e.g., host-pathogen) and intraspecific (e.g., mother-fetus, male-female) conflicts of interest that fuel antagonistic coevolution. Other processes we have not discussed (e.g., other forms of antagonistic coevolution, spatially variable selection, epistatic variance that may be released and expressed as additive genetic variance during periods of rapid change [see Pomiankowski & Møller, 1995]) may play some role as well.
The big picture we deal with in this chapter is mate choice for good genes. The most relevant question in that regard is how much genetic variation in fitness exists, not what causes it. And in that regard, the received view today is very different from what it was 20 years ago. Large reservoirs of genetic variance in fitness typically persist in natural populations. That is not to say that mate choice for good genes will evolve in every species—not even every species in which large amounts of genetic variance in fitness exist. Byrne and Rice (2006) found evidence that males of a laboratory population of Drosophila melanogaster prefer mating with larger, more fecund females. In the same population, however, genetic benefits of remating with a high-quality male could not pay for the direct costs of remating (Orteiza, Linder, & Rice, 2005). And, hence, an allele that increased female resistance to remating at a cost of missing out on genetic benefits garnered through remating was favored (Stewart, Morrow, & Rice, 2005). Females in this population may still favor high-quality males as first mates. But there is no evidence that female remating tactics are effectively designed to obtain good genes; remating behaviors appear to be driven by male and female antagonistic adaptations (see also Byrne & Rice, 2005). In light of the genetic variation in fitness, however, it should not be surprising that good-genes sexual selection occurs in many species.
The Viability-Indicator Model Versus the Fisherian Model The goodness of good genes, once again, refers to their positive influence on offspring reproductive capacity. Intrinsic good genes have this effect independent of the mate chooser’s own genes; their effects are additive.
For most of the past several decades, two different good-genes models were thought to be separate. The viability-indicator model (see Andersson, 1994) held that mate choosers prefer mates who possess indicators of viability. These individuals live longer than other potential mates (e.g., they specifically have greater juvenile and/or adult survival, as well as correlates such as better phenotypic condition or resistance to disease). Because viability is partially heritable, these mates’ offspring also live longer. All else being equal (e.g., equal age-specific fecundities), long lives promote reproductive success.
The second good-genes model was the Fisherian model. This model proposes that individuals prefer mates who are attractive but live no longer than others (or may die even earlier than others). Because their offspring are also attractive, however, their offspring have greater reproductive success. Particularly when males are the highly ornamented sex and have more variable reproductive success, the effect of a male mate’s attractiveness on his son’s attractiveness and reproductive success is particularly strong (e.g., Kirkpatrick, 1982; Lande, 1981). Hence this model is sometimes referred to as the “sexy son” hypothesis (Fisher, 1930). The Fisherian model, like the viability-indicator model, is a good-genes model. Because this model does not assume that preferred mates or their offspring have any greater ability to survive than other potential mates and their offspring but, rather, have greater reproductive success merely because they are “attractive,” this model has sometimes been posed as one that presumes that preferred mates are not “intrinsically” better than other mates. This has particularly been true when the preferred trait is presumed to be an “arbitrary trait” that is first preferred solely because of a sensory bias (through a process we described in chapter 5). Individuals preferred by a sensory bias are no better than any others in terms of fitness initially and only become “better” because of spread of genes for the arbitrary preference, which occurs due to linkage with genes for the preferred trait (see, e.g., Andersson, 1994).
As we described in chapter 5, a trait initially preferred because of a sensory bias can also ultimately become a viability indicator. As the trait preferred due to sensory bias becomes exaggerated over time, it becomes increasingly dependent on overall condition. Individuals in the best condition can afford to “pull off” growing ornamentation that is most exaggerated. In Rowe and Houle’s (1996) terms, as the trait becomes increasingly dependent on condition, it “captures” the genetic variance in condition and, thereby, genetic variance in fitness. What was initially a trait preferred solely because of a sensory bias has, over time, become a viability indicator.
The Viability-Indicator Model and the Fisherian Model Merely Define Two Ends of a Continuum of the Same Sexual Selection Model Kokko, Brooks, McNamara, & Houston (2002) recently offered an important insight about the viability-indicator and the Fisherian models. They are not two distinct models. Rather, they are variations on the same fundamental model. Both are intrinsic good-genes models, and equally so. They differ only in their assumptions about what an individual who possesses good genes does with the advantage: being particularly good only at attracting mates or being particularly good at surviving, as well.
Selection will shape individuals’ allocation of resources (e.g., energy, time) to fitness-promoting activities (e.g., growth, somatic repair, immune function, developing ornaments) in ways that maximize their fitness. Two individuals who have different quality presumably differ in the resources available to them for fitness-promoting activities (or the efficiency with which they can allocate those resources to fitness-promoting activities). And, as we discussed in chapter 5, an individual of good quality may optimally allocate those resources in ways different from the way an individual of poor quality optimally allocates its resources.
The viability-indicator model presumes that individuals allocate resources in a way such that an individual of good quality survives, on average, better than does an individual of poor quality. (For instance, the former may dedicate greater resources to immune function or somatic repair.) Because high-quality individuals are more resourceful, they presumably can simultaneously allocate more resources to developing traits that attract mates (e.g., ornaments), as well.
The Fisherian model presumes that individuals allocate resources in a way such that an individual of good quality survives, on average, no better than does an individual of poor quality (or, in fact, may actually survive less well). That is, the former may dedicate fewer resources to immune function or somatic repair than does the latter. (A high-quality individual may, then, be more disease-prone and less healthy than a poor-quality individual; see Getty, 2002). Obviously, for high-quality individuals to truly be high quality, in this scenario they must allocate more resources to developing traits useful for attracting mates and, to offset the fact that they survive less well, be much better at attracting mates than low-quality individuals.
What circumstances would lead individuals of high quality to put so much of their resources into competing for mates, such that they actually die earlier, on average, than individuals of poor quality? In theory, these are circumstances in which mate choosers particularly value good genes and pay minimal costs (e.g., in search time) for maintaining very high standards for mate choice (e.g., leks). In short, species for which the Fisherian pattern fits are ones in which sexual selection (typically on one sex’s attractiveness only) is extreme (see Kokko et al., 2002, 2003).
The viability-indicator model and the Fisherian model anchor two ends of a continuum. At one end of the continuum, individuals of high quality (and good genes—hence also the offspring of high-quality individuals) are much better at surviving, and fairly minimally better at attracting mates, than low-quality individuals. At the other end, individuals of high quality (and good genes—hence also their offspring) are much better at attracting mates, and even worse at surviving, than low-quality individuals. But these are merely two ends of a continuum; the actual pattern of how good genes translate into fitness (and the fitness of offspring) within a particular species can fall anywhere along the continuum (Kokko et al., 2002).
Where species actually tend to fall on the sexual selection continuum is an empirical question. Jennions, Møller, and Petrie (2001) conducted a meta-analysis of all animal studies, reporting a correlation between a sexually selected male trait and male survival. On average, more highly ornamented males survive better than less ornamented males. The mean across the species studied, then, falls toward the viability-indicator end of the continuum. That result, of course, does not rule out some species clearly falling at the sexy-son end of the continuum. Relatedly, Møller and Alatalo (1999) performed a meta-analysis of the literature of animal studies examining an association between a sexually selected male trait and offspring survival. Consistent with Jennions et al.’s (2001) analysis, they found a mean positive relationship between male ornamentation and offspring survival: In general, more ornamented males sire offspring with greater survival than do less ornamented males (and, in fact, this association is similar in size to that between male ornamentation and degree of male care; Møller & Jennions, 2001). (For other work on ornamentation of sires and health or overall phenotypic condition, see, e.g., Johnson, Thornhill, Ligon, & Zuk, 1993; Zuk, Thornhill, Ligon, & Johnson, 1990; Zuk, Thornhill, Ligon, Johnson, Austad, et al., 1990.)
These meta-analyses examined only male ornaments. In theory, because females are typically under less intense sexual selection than are males, attractive females should be better not only at attracting mates but also at producing high-viability offspring and, perhaps, at surviving (i.e., female ornaments of quality should typically be explained by a model at the viability-indicator end of the sexual selection continuum; see chapters 5 and 6).
Once again, adaptive mate choice for intrinsic good genes requires additive genetic variance in fitness. By contrast, adaptive mate choice for compatible genes requires nonadditive genetic variance in fitness (see Neff & Pitcher, 2005). Choice for compatible genes involves choosing a mate to create a combination of maternally and paternally derived alleles in offspring that promote fitness independent of the individual fitness effects of the alleles (their additive effects). Nonadditive genetic variance due to fitness effects of favored combinations of alleles can come in two forms. Dominance deviations are unique effects associated with combinations of two alleles at single loci (e.g., heterozygote superiority). Epistatic effects are nonadditive effects associated with combinations of two or more alleles at different loci (e.g., Falconer & Mackay, 1996).
The MHC has been investigated in multiple species, including humans, as a possible target of mate choice for compatible genes. The MHC exists in all vertebrates. Loci within the MHC are often polymorphic, sometimes extremely so. In humans, for instance, three loci (the A, B, and DRβ loci) are highly polymorphic, with up to hundreds of different possible alleles at an individual locus and, typically, no allele being more than 20% of the alleles at a locus in a population. MHC alleles are codominantly expressed, and hence heterozygotes can potentially present a wider range of foreign peptides to T-cells and thereby effectively defend against a broader array of pathogen strains. Studies on chinook salmon (Arkush et al., 2002) and mice (McClelland, Penn, & Potts, 2003; Penn, Damjanovich, & Potts, 2002) found evidence for MHC heterozygote superiority (but see Lohm et al., 2002). In humans, MHC heterozygote superiority has been found in resistance to hepatitis B infection (Thurz, Thomas, Greenwood, & Hill, 1997). Furthermore, human couples that possess a common MHC allele produce an underrepresentation of homozygotic offspring, indicating in utero selection against homozygotes (see, e.g., Hedrick & Black, 1997b). (One-quarter of the offspring of couples who share one allele at a MHC locus should be homozygotic at that locus. In fact, less than one-quarter are homozygotes.) Their conceptions more frequently end in spontaneous abortions, as well (Ober, Hyslop, Elias, Weitkamp, & Hauck, 1998). Heterozygote superiority is probably one reason why MHC loci are highly polymorphic (Hedrick, 1998). (For one study showing that pathogen diversity leads to MHC diversity in humans, see Wegner, Reusch, & Kalbe, 2003.)
To choose a mate with whom an individual will produce offspring that are MHC heterozygotes—a mate that possesses compatible MHC alleles—the individual should choose one who shares no (or few) MHC alleles. Mice can detect MHC identities in scents of other mice (Yamazaki, Beauchamp, Curran, Baird, & Boyse, 2000). Based on scent, they prefer mates who possess dissimilar MHC genotypes (Penn & Potts, 1999). Preferences for MHC dissimilarity (or other forms of self-referenced preferences; see, e.g., Milinski et al., 2005) also exist in species of birds (Freeman-Gallant, Meguerdichian, Wheelwright, & Sollecito, 2003), fish (Aeschliman, Haberli, Reusch, Boehm, & Milinski, 2003; Milinski et al., 2005), and lizards (Olsson et al., 2003), though not all species possess them (e.g., Sommer, 2005).
Studies examining MHC preferences in humans have yielded generally supportive but mixed evidence. Preferences for the scent of opposite-sex individuals with dissimilar MHC genotypes have been detected in three of four studies of normally ovulating women (Santos, Schinemann, Gabardo, & Bicalho, 2005; Wedekind & Füri, 1997; Wedekind, Seebeck, Bettens, & Paepke, 1995; cf. Thornhill et al., 2003) and two of three studies of men (Thornhill et al., 2003; Wedekind & Füri, 1997; cf. Santos et al., 2005). (In another study, women preferred the scent of MHC-similar men, but its preference measure may not tap sexual attraction; Jacob, McClintock, Zelano, & Ober, 2002).
A study of Hutterites found married couples to be more MHC-dissimilar than expected by chance (Ober et al., 1997); studies of South American Indian and Japanese couples did not (Hedrick & Black, 1997a; Ihara, Aoki, Tokumaga, Takahashi, & Juji, 2000). A study of romantically involved couples in the United States detected disassortative mating at MHC class-I loci but not at a class-II locus (Garver-Apgar, Gangestad, Thornhill, Miller, & Olp, 2006). (Class I alleles are expressed on nearly all cells and function in resistance to intracellular infection. Class II alleles are expressed only on leukocytes and function in resistance to extracellular infection. Because they are expressed on skin cells, Class I alleles may be more readily detected through scent; see Leinders-Zufall et al., 2004.)
Despite the substantial amount of research done on MHC, it is not the most thoroughly documented case of choice for compatible genes in humans; incest avoidance is. Avoidance of inbreeding is beneficial not because siblings (and other close relatives) possess poor genes. On average, of course, individuals who are opposite-sex siblings to at least one other person have fitness close to the mean in the population. (Because their parents had at least two offspring of different sexes, they may even have slightly higher than mean fitness.) Siblings are poor mate choices because of nonadditive genetic effects expressed in offspring produced by siblings who mate with one another. A sibling may have about the same number of mutations, on average, as a randomly chosen individual but, relative to two randomly chosen individuals, two siblings are much more likely to have mutations at the same locus. In the heterozygous state, mutations at coding sites typically have nonzero deleterious effects on fitness (i.e., mutations are not completely recessive; as noted earlier, the deleterious effect on fitness averages a few percentage points). But two mutations at the same locus typically have a joint effect on fitness much larger than double their effect in a heterozygous state (perhaps, on average, five times the effect; Lynch et al., 1999). Put otherwise, there are large nonadditive components to the effects of mutations. These nonadditive genetic effects drive selection for adaptations that lead individuals to avoid mating with closely related individuals. Adaptations that function to avoid incestuous mating, then, are classic instances of adaptations for mate choice to acquire compatible genes for offspring. (For research on the cues that people use to discriminate siblings from nonsiblings and thereby avoid sex with individuals likely to be siblings, as well as sex differences in use of these cues, see Lieberman et al., 2003, 2007.) Whether humans have adaptation to avoid inbreeding by refraining from mating with individuals more distantly related than first-degree relatives is unknown. Preference for MHC dissimilarity may function as a mechanism of inbreeding avoidance rather than to increase heterozygosity per se (e.g., Penn & Potts, 1999).
Aside from preferences for dissimilar MHC alleles and incest avoidance, we know of no other well-documented adaptations for mate choice for compatible genes in humans. As Zeh and Zeh (1996, 2001) note, both interspecific and intraspecific antagonistic coevolutionary processes create genetic incompatibilities that may be targets of adaptive mate choice (or, as it may be, mate avoidance). For instance, a mother who can effectively suppress a fetus’s attempts to extract maternal resources may actually be best off selecting a mate who provides a fetus with capabilities of keeping its own in a tug-of-war with its mother. (See Zeh & Zeh, 1996, 2001, and Jennions & Petrie, 2000, for discussions of other possible nonadditive genetic effects fueling compatible gene choice.)
As Zeh and Zeh (1997) also argue, however, adaptations that function for compatible gene choice may often operate postcopulation in sperm or zygote selection. Individuals typically are not selected to advertise traits related to compatible genes; as Jennions and Petrie (2000) note, males have little reason to advertise features that females might find compatible, as males may be selected to mate as widely as possible. (Whether that should be true in a system involving biparental care and mutual mate choice is unclear.) Indeed, MHC recognition is probably achieved through adaptation that detects incidental by-products of MHC identities (e.g., self-peptides shed by skin cells; e.g., Leinders-Zufall et al., 2004), not a system involving targets’ adaptation to signal MHC identities. Many compatible gene effects may pertain to single loci that, unlike MHC identities, do not produce by-products that perceivers can detect. By contrast, postcopulatory mechanisms may be particularly effective at executing female choice for compatible genes; females may evolve ways to assess DNA’s “identities” (and hence compatibilities) once sperm is in the female reproductive tract, which is simply not possible before copulation. As well, a conceptus that is not highly fit due to incompatibilities may be selectively aborted (Haig, 1991). The underrepresentation of MHC homozygotes in the offspring of human couples who share MHC alleles (see previous discussion; Hedrick & Black, 1997b) may be the result of postcopulatory adaptation to select for compatible genes. (For reviews of postcopulatory mechanisms of choice for compatible genes, see Jennions & Petrie, 2000; Simmons, 2005.)
Zeh and Zeh (1997, 2001) suggest that females in some species possess adaptation to mate polyandrously, which functions to allow postcopulatory mechanisms to select compatible genes among different males’ sperm. In chapter 10, we discuss the possibility that women mate polyandrously to achieve these or other effects garnered by “running” sperm competitions. Though we are skeptical that, in humans, the benefits of polyandrous mating to achieve sperm selection have historically outweighed the costs, the possibility that women possess adaptation to do so under some conditions cannot be ruled out.
One possible additional form of mate choice for compatible genes in humans could be driven by intralocus sexually antagonistic selection. As noted earlier, this form of selection leads each sex to be compromised by what is adaptive and selected for in the other sex. Individuals differ, however, in the extent to which they are moved away from the optimum phenotype for their sex by this form of selection. If male reproductive success is skewed, male “masculinity” may relate to reproductive success in a concave-upward curvilinear fashion. A female who is highly feminized and one who is relatively masculine may produce sons who differentially benefit from receiving genes from a masculine sire. The fitness of the son of the feminized female will be compromised by his mother’s feminizing genes. The son of the masculine female will not be. Because masculinity relates to fitness in a rising, curvilinear fashion, the benefit of receiving masculine genes from a father is greater for the son of the masculine female. Hence, perhaps, females masculinized by sexually antagonistic genes adaptively prefer more masculine men, as a function of adaptive choice for compatible genes.
The ratio of the lengths of the 2nd and 4th digits, once again, may be affected by sexually antagonistic genes (Manning et al., 2000). Relatively low digit ratios are masculine, whereas relatively high digit ratios are feminine. Scarbrough and Johnston (2005) asked women to rate the attractiveness of male faces, which were manipulated to vary in their masculinity. In general, women’s choice of the man most attractive for a short-term relationship (a sex partner) was more masculine than the man rated most attractive for a long-term relationship. The man most preferred as a long-term partner by women with masculine digit ratios, however, was more masculine than the man most preferred as a long-term partner by women with feminine digit ratios—consistent with the compatible genes hypothesis we outlined. The question of whether these differences in women’s preferences truly do reflect adaptation for choice of a mate with compatible genes requires further study.
As noted previously, genetic diversification of offspring has traditionally been thought of as a form of bet hedging in the face of future uncertainties: reducing variance in outcomes, even at a cost of mean outcomes (e.g., Gillespie, 1977). This model of the selective advantages of genetic diversification has been questioned (Yasui, 1998). In a nutshell, the argument against this model of selection is that a gene that promotes diversification within families at a cost (e.g., costs of polyandry) will not spread through selection if an alternative allele can achieve the same level of diversification across families, but without a cost (for a discussion, see Jennions & Petrie, 2000).
Nonetheless, there are other avenues through which genetic diversification can be advantageous—not only by decreasing variance in fitness outcomes but also by increasing mean fitness outcomes (e.g., Yasui, 1998). In some interaction contexts, full siblings may negatively affect each other’s outcomes more than half siblings do. Thus, for instance, two siblings are likely to be exposed to pathogens that each carries. If a pathogen adapts to and thrives within one sibling (e.g., because it has evolved counteradaptation to the individual’s immune defenses), the likelihood that the pathogen will transmit to and thrive within the other sibling is a function of the similarity between the siblings. Full sibs are more similar to one another than are half sibs. Hence, a female may benefit from producing half sibs (with different fathers) rather than full sibs (with the same father). (We note that sibling-sibling conflicts of interest about parental investment, which are greater in half sibs than in full sibs, may more than offset these benefits in some species; see Trivers, 1974.)
A mate chooser can diversify offspring, however, even when producing full sibs. Specifically, a mate who is heterozygous at a particular locus will produce offspring more diverse at that locus than are the offspring produced by a mate who is homozygous. If females particularly benefit from diversifying offspring at loci involved in pathogen recognition and immune defense, it may pay them to prefer a mate who is heterozygous at MHC loci. In a study performed with colleagues, we found precisely this effect. Women strongly preferred the scent of men who were heterozygous at all three loci we assayed (A, B, DRβ) over the scent of men who were homozygous at one or more loci (Thornhill et al., 2003). Possibly, women possess adaptation that functions to diversify their families’ MHC alleles by leading them to prefer, as long-term mates (with whom they have, historically, typically had multiple offspring) men heterozygous at MHC. We discuss this finding and interpretation further in chapter 9.
Evidence that sexual selection has operated on women to shape preferences for mates who possess intrinsic good genes has been indirect. Researchers have asked whether women prefer male traits that in ancestral conditions may have been associated with intrinsic good genes. These traits may not be associated with fitness in many modern environments. The extraordinary health care and lifestyle changes that have occurred in the past century, for instance, may have substantially altered correlations between phenotypic features and fitness.
One candidate trait that may have been related to fitness in ancestral environments is developmental instability. Developmental instability is an individual’s proneness to the imprecise expression of developmental design due to genetic and environmental perturbations. These perturbations, importantly, include mutations and pathogens and, hence, major factors that contribute to genetic variation in fitness. Because of its conceptual link to maladaptation (see Leung & Forbes, 1996; Møller, 1997, 1999; Parsons, 1992), developmental instability became a focal point of research on sexual selection in the 1990s (Møller & Swaddle, 1997; Møller & Thornhill, 1998b).
The primary measure of developmental instability used by biologists is fluctuating asymmetry (FA). FA is deviation from perfect symmetry on a bilateral trait that is, on average, symmetrical in the population. For instance, the breadth of people’s right and left wrists is, on average, very similar. Some people, however, have slightly larger right wrists. Others have slightly larger left wrists. FA of wrist breadth is calculated as the absolute deviation of the right and left wrist breadths (in some applications, standardized as a proportion of wrist size by dividing by the individuals’ own or the population average wrist size). In research conducted at the University of New Mexico, colleagues and we have measured a number of asymmetries in humans—of the ears, elbows, wrists, ankles, feet, and fingers (see, for instance, Furlow, Armijo-Pruett, Gangestad, & Thornhill, 1997; Furlow, Gangestad, & Armijo-Prewitt, 1998; Gangestad, Bennett, & Thornhill, 2001; Gangestad & Thornhill, 1997a, 1997b, 1998, 2003 a,b; Gangestad, Thornhill, & Yeo, 1994;Thoma, Yeo, Gangestad, Lewine, & Davis, 2002; Thoma et al., 2005; Thornhill & Gangestad, 1994, 1999a, 2006; Thornhill, Gangestad, & Comer, 1995; Thornhill et al., 2003). Other researchers have measured similar sets of traits (e.g., Livshits & Kobylianski, 1989; Manning, 1995; Manning, Scutt, & Lewis-Jones, 1998; Manning & Wood, 1998; Rikowski & Grammer, 1999; Waynforth, 1998). The asymmetry that exists in these traits is very small, the mean being 1–2 mm, so small that you cannot detect it reliably through normal social interaction. Hence, the asymmetries we measure cannot serve as cues by which individuals assess others’ developmental imprecision (though it is possible that small asymmetries can influence observable performance). The reason we measure them, then, is because they purportedly are markers of underlying developmental imprecision, which may substantially affect the overall phenotypic fitness of individuals. We aggregate across many traits (typically 10) because, as we discuss later, each individual trait’s asymmetry very poorly reflects overall developmental instability (Gangestad & Thornhill, 1999; Leung, Forbes, & Houle, 2000).
Many studies have examined associations between FA and fitness components in a wide variety of species. Møller (1999) reviewed studies available at the time and found that, indeed, on average low FA is associated with relatively high fitness (e.g., survival, fecundity). This broad conclusion must be tempered with a caveat: Not all results have supported a link between FA and fitness. Though many null results may very well be due to low power and sampling variability (Type II error; see Gangestad & Thornhill, 2003b), in some species there may be no link. As we noted earlier in this chapter, as well, in some species individuals of high quality may survive no longer than individuals of low quality (e.g., Kokko et al., 2002; Getty, 2002).
One specific component related to fitness that has received much attention in work on FA is disease infection and parasite loads. Møller (2006) recently reviewed this literature. Once again, a robust, mean positive association between asymmetry and disease or parasite load exists (see also Thornhill & Møller, 1997). This literature includes studies on humans. In one study, we found that more symmetrical individuals reported lower frequencies of infections in the past three years (Thornhill & Gangestad, 2006). In a Mayan population, Waynforth (1998) found an association between FA and incidence of major disease. Overall, however, results on humans have been mixed (see Rhodes, 2006). Whether the mixed results are due to modern health care obscuring associations between developmental instability and health is unknown.
A number of studies have specifically examined associations between FA and immune activation (e.g., immunoglobulin levels) during development. Once again, links do, on average, exist: Activation of the immune system during development is associated with higher levels of asymmetry (Møller, 2006). Future research may examine these links in humans.
In humans, associations between body FA and features of the brain have specifically been examined. Human brains are characteristically asymmetrical in particular ways (e.g., the planum temporale is larger in the left hemisphere than in the right hemisphere), partly due to specialization of function in the two hemispheres (e.g., in the case of the planum, specialization of language function in the left hemisphere). One can calculate an individual’s atypical brain asymmetry (deviation from mean left vs. right differences). In two studies, Thoma and colleagues (Thoma et al. 2002; Thoma et al., 2005) found that body asymmetry predicted atypical brain asymmetry; individuals with asymmetrical bodies also tended to have brains that were atypically asymmetrical. Atypical anatomical asymmetries may reflect atypical brain organization. Neuropsychological tasks can detect the extent to which some specific functions (e.g., perception of phonemes) are processed primarily in the right or left hemisphere. Individuals with high FA tend to perform simple cognitive tasks in the two hemispheres in an atypical fashion (Yeo, Gangestad, Thoma, Shaw, & Repa, 1997). A number of studies have demonstrated a negative association between FA and intelligence (as assessed by standard psychometric procedures; Bates, 2007; Furlow et al., 1997; Luxen & Buunk, 2006; Prokosch, Yeo, & Miller, 2005; Rahman, Wilson, & Abrahams, 2004 [though not in women]; Thoma et al., 2005; but for a study that failed to find an association, see Johnson, Segal, & Bouchard, 2007). Thoma et al. (2005) found that FA predicts intelligence independently of brain size. Furthermore, FA predicts slow processing in a simple reaction time task (Thoma et al., 2006). More generally, developmental instability appears to be associated with a variety of neurodevelopmental disorders (e.g., schizophrenia, schizotypy, ADHD, dyslexia; see Yeo, Gangestad, Edgar, & Thoma, 1999). Possibly, developmental instability compromises adaptive organization of functions in the brain and/or, at a more molecular level, neural or metabolic function (e.g., Yeo, Hill, Campbell, Vigil, & Brooks, 2000).
Associations between fluctuating asymmetry and functional operation of other organ systems in humans have not been examined. One might expect, however, that, just as developmental instability compromises the integrity of brain development, it also affects the precision with which other important systems are developed. As the human brain is composed of a particularly complex and energy-demanding set of features, with a large proportion of all genes expressed in brain tissue, it may particularly reveal deleterious effects of developmental instability (see, e.g., Gangestad & Yeo, 1997; Yeo et al., 1999).
The precise developmental processes by which fluctuating asymmetry arises are not well understood (e.g., Van Dongen, 2006). In theory, FA is the outcome of developmental noise—random expression caused by mutations, infection, toxins, and the like—and a developmental system’s sensitivity to those perturbations (e.g., lack of buffering or repair mechanisms). Asymmetry, then, can result from high levels of perturbations (e.g., high levels of mutations), high levels of sensitivity to perturbations, or both. Part of the difficulty of interpreting precisely what FA means about an individual is that we do not know the extent to which it reflects the many possible variations that could cause it. And, indeed, the relative impact of causes need not be the same across different species (or, within a species, across different environments).
In general, asymmetry results when populations of cells on one side of the body stop growing while corresponding populations of cells on the other side continue to grow. Babbitt (2006) recently proposed, “variation in fluctuating asymmetry is in large part due to the random exponential growth of cell populations that are terminated randomly around a genetically programmed development time” (p. 258). One major issue concerns whether systematic processes affect termination of development and, if so, what those might be. One possibility is that genetic mutations disrupt adaptively programmed timing of development. Kjaer et al. (2005), for instance, compared individuals from four families with long polyalanine expansions in the HOXD13 gene (known to affect ontogeny of the digits) to individuals from the same families without mutated HOXD13 alleles. Those with long expansions had greater FA of the hands and feet than those without long expansions.
A more pervasive systematic process leading to asymmetry might be oxidative stress. Respiration naturally produces, within cells, unstable reactive oxygen species (ROS) as by-products. ROS can react with proteins or lipids in a cell or, indeed, DNA itself, thereby damaging cell structures. Organisms possess adaptive systems to quickly stabilize ROS (e.g., through production of antioxidants such as superoxide dismutase and catalase). When production of ROS outpaces antioxidant activity, a condition referred to as oxidative stress occurs and cell structures are often damaged. A wide variety of conditions can cause oxidative stress. In fact, one adaptive inflammatory response to pathogens in vertebrates is to produce ROS directed against the pathogens. A cost of this response is potential cell damage. (For instance, destruction of the gastric epithelium as a result of Helicobacter pylori infection in humans—the creation of ulcers—is at least partly due to extracellular ROS released by phagocytes directed against the bacteria; e.g., Ramarao, Gray-Owen, & Meyer, 2000.) Cellular damage may affect regulation of cell replication or lead to cell death. As already noted, infection during development appears to be associated with high FA (Møller, 2006). Oxidative stress may partly or fully mediate this effect. (A possibly interesting side note is that, as some pathogens have adaptations that combat oxidants directed toward them—e.g., Ramarao et al., 2000—host-pathogen coevolution may revolve partly around production of antioxidants. This coevolution may maintain genetic variation in pathogen resistance, as well as developmental instability, in hosts.) More generally, any stressor (e.g., mutation, biotoxin) that causes oxidative stress or interferes with antioxidant adaptations may cause cell damage and replication. Possibly, then, cell damage caused by ROS is a common mediator of the effects of perturbing events and FA (see also von Schantz, Bensch, Grahn, Hasselquist, & Wittzell, 1999).
These specific ideas are speculative; obviously, they demand empirical tests. But a broader point of emphasis here is that, in general, more research is needed before we will understand well the processes that lead to FA. And only after we have a better understanding of these processes will we have a good idea of why (and when) FA taps fitness-related traits (e.g, Van Dongen, 2006).
A pertinent issue in this regard, still unresolved, is the extent to which the asymmetries of different characters are caused by systematic factors shared across characters or ones specific to characters. For instance, are the systematic individual differences that underlie variation in asymmetry of human ear dimensions the same individual differences that underlie variation in asymmetry of human finger lengths? Or are these individual differences largely unshared? The correlation between any two traits’ FA tends to be very small—about .05 on average in a large survey of over 1,000 correlations (Gangestad & Thornhill, 1999; see also Gangestad et al., 2001). This finding has led some observers to claim that FA of any two characters largely reflects character-specific causes of developmental imprecision, not organism-wide causes (e.g., Van Valen, 1962).
In fact, however, this conclusion is almost certainly wrong, at least for many species. One must take into account the extent to which systematic individual differences contribute to a single trait’s FA. The effects of developmental error on asymmetry have large, unsystematic, random components. Hence, even if the same individual were to grow the same trait twice under precisely the same conditions and experience precisely the same amount of developmental error, the individual would not grow precisely the same amount of asymmetry. (Random developmental errors on the right and left side on one occasion might happen to be in opposite directions and accentuate asymmetry, whereas on another occasion they might happen to cancel each other out, yielding little asymmetry.) The amount of variance in a single trait’s asymmetry due to systematic individual differences places a limit on the correlation between the asymmetry of two different characters (or, in fact, the correlation that would be observed if individuals could grow the same trait twice). A number of methods can be used to estimate this value, typically referred to as the repeatability of FA (see, for instance, Van Dongen, 2006; Gangestad & Thornhill, 2003b). The median estimated repeatability across a large number of data sets is less than 0.10 (Gangestad & Thornhill, 1999, 2003b). In our most recent data on human traits, we found it to be 0.076 (Gangestad et al., 2001). In 12–14 measurements of skeletal bones of eight species of nonhuman primates, Hallgrímsson (1998) found a remarkably similar average value: 0.072. Though higher values are occasionally observed (e.g., Lens & Van Dongen, 1999), in mammalian skeletal or skull measurements values larger than 0.10 are hardly ever seen (Gangestad & Thornhill, 2003b). (Recently, Graham, Shimizu, Emlen, Freeman, & Merkel, 2003, have questioned the assumptions behind estimations of repeatability and propose alternative assumptions that, if true, imply that true values are even smaller than our estimates. Van Dongen, Talloen, & Lens, 2005, however, found that estimates from real data are very similar based on analyses that assume a range of different models, including alternatives posed by Graham et al., 2003.)
The implication is that, even if two traits’ asymmetries shared 100% of their systematic individual differences in propensity to develop asymmetrically, they would correlate, on average, only about 0.07–0.08. The fact that they correlate, on average, around .05 (or, in the human data we collected, about 0.045; Gangestad et al., 2001) means that over 50% of the systematic variance in two traits’ asymmetry is shared. That is, most of the systematic individual differences underlying a trait’s asymmetry are, in fact, organism-wide variations in developmental instability (see also Lens & Van Dongen, 1999). (If our estimated repeatability of .07 is itself slightly overestimated due to improper assumptions [Graham et al., 2003; Van Dongen et al., 2005], the amount of shared effects could near 100%.)
In fact, this result is not terribly surprising, given that fluctuating asymmetry does relate to fitness, on average. If each individual trait’s asymmetry merely reflected components of developmental integration and stability specific to that trait, it would not seem likely that the trait’s asymmetry would covary with fitness. (An exception might be made if the asymmetry itself directly affected performance, but that is not the case with the very small asymmetries in finger lengths, ear size, etc., that we measure.) For trait asymmetry (or even a composite of different traits’ asymmetries) to plausibly predict fitness, those asymmetries should typically reflect systematic variation across individuals in their propensity to develop precisely in an organism-wide fashion. (Indeed, the correlations between body asymmetry and atypical brain asymmetry and organization require shared variance in developmental processes across traits of the body and brain; cf. Polak & Stillabower, 2004.)
We present one final note on the association between developmental instability and fitness. Though, all else being equal, zero developmental imprecision is optimal, in most instances, processes that maintain developmental precision (e.g., cellular repair mechanisms, antioxidation) do demand energetic resources. Hence allocation to these processes demands trading off allocation to other adaptive processes. In some circumstances, individuals that are most fit in the population may actually be best off allocating fewer resources to maintenance of developmental precision (or compromising it, e.g., through rapid growth) than to the allocation that maximizes the fitness of less fit individuals. (Related to our earlier discussion of the sexual selection continuum, this may be true particularly if sexual selection is very strong, in which case extreme sexual displays by highly fit individuals may compromise developmental precision and other viability-enhancing traits; e.g., Kokko et al., 2002, 2003.) Some of the inconsistency in associations between FA and fitness components reported in the literature probably does reflect true differences across species in the extent to which fit individuals do well by maintaining developmental precision. As we have already noted, in some species there may simply be very little association between developmental instability and fitness.
Perhaps the most intensively studied correlate of FA to date is mating success. Do symmetric individuals achieve greater mating success than less symmetric individuals? And if so, is their greater success at least partly due to the other sex’s preference to mate with relatively symmetric individuals? A decade ago, Møller and Thornhill (1998b) performed an exhaustive meta-analysis of the studies that had been done at that time. Their conclusions were clear: Symmetric individuals do indeed have relatively high mating success. And in many instances, it is because they are attractive to members of the other sex. These conclusions were challenged by Palmer (1999, 2000), who questioned whether the literature might be misleading due to publication bias (in favor of supportive studies). But subsequent analyses that assess bias support the original conclusions (Møller, Thornhill, & Gangestad, 2005; Thornhill, Møller, & Gangestad, 1999). Once again, however, mean positive effects do not imply that positive effects exist in all species or, within species, all environments; in some species no association may exist (at times because there are, in these species, no associations between FA and fitness; see preceding discussion and Møller & Cuervo, 2003).
Over a decade ago, we began asking whether FA predicts number of sex partners in college students, similar to associations examined in nonhuman species. We have now studied over 500 college students of each sex. In sum, we find that, in this population, men’s FA does, whereas women’s does not, reliably do so (see Gangestad & Thornhill, 1999, for an overview; see also Gangestad & Thornhill, 1997b; Thornhill & Gangestad, 1994). We (with Kevin Bennett; Gangestad et al., 2001) estimated the correlation between men’s developmental instability and number of sex partners using latent structural equation modeling, which uses each individual trait’s FA as an independent marker of underlying developmental instability. In a sample of over 200 men, the estimated correlation was –0.4 to –0.5 (with body size and age controlled), a sizable effect.
We have also collaborated with two anthropologists, Mark Flinn and Rob Quinlan, in work in a rural village on the Caribbean island of Dominica. There, too, we find that FA predicts number of romantic partners in men (Gangestad, Thornhill, Quinlan, & Flinn, 2007). In that study we measured number of romantic partners by asking other villagers, not target individuals themselves. The correlation between FA and men’s number of partners was about –.4.
We suspect that female preferences for symmetry per se have little if anything to do with the causal process that drives these associations. Again, FA is our measure of developmental health. It correlates with a variety of physical and behavioral features that may mediate these associations between FA and number of sex partners because women prefer these features, not FA directly. Consider the following examples.
1. In Dominica, FA negatively predicts men’s peer status, as assessed through interviews with men. More symmetric men are seen to be better coalition partners than less symmetric men. Female preferences for men with favorable peer status could lead more symmetric men to have more romantic partners (Gangestad, Thornhill, Quinlan, & Flinn, 2007).
2. U.S. college men who have greater body symmetry may have more masculine faces, as assessed by a variety of sexually dimorphic facial dimensions (Gangestad & Thornhill, 2003a; but see Koehler, Simmons, Rhodes, & Peters, 2004). Even the association between facial symmetry and attractiveness may be partly mediated by other facial features (Scheib, Gangestad, & Thornhill, 1999). (See figures 7.1 and 7.2.)
3. Simpson, Gangestad, Christensen, & Leck (1999) had U.S. college men interviewed for a potential lunch date with an attractive female. As part of the interview, each man was asked to tell the woman, as well as a male competitor—someone else she was purportedly interviewing—why he should be chosen for the lunch date over the other. More symmetric men were more likely to engage in direct intrasexual competitive tactics—directly compare themselves with the other and state that they were the better choice—than asymmetric men were. Other research shows that, compared to asymmetric men, symmetric men are less willing to back down from threats to their status (Gangestad & Thornhill, 1997a) and, perhaps as a result, more likely to get into physical fights, particularly those that they escalated into a fight (Furlow et al., 1998; Manning & Wood, 1998).
Figure 7.1 Scatterplot of the association between facial masculinity and bodily FA in men, showing a significant, linear association (r = –0.27, p = 0.001, N = 141). Variables on both axes have been standardized (converted to z-scores). The line is the least squares linear regression line. The facial masculinity measure is a composite of sexually dimorphic effects (e.g., testosteronization; see Gangestad & Thornhill, 2003a). From figure 1 of Gangestad and Thornhill (2003a).
Figure 7.2 Scatterplot of the associations between facial FA and facial masculinity in men (filled diamonds) and women (open squares). For men, there is a significant linear association as well as a significant curvilinear association (N = 139, linear r = –0.20, p < 0.05; quadratic r [with linear effect partialled out] = 0.39, p < 0.001). For women, there is a significant curvilinear association (N = 151, linear r = 0.03, ns; quadratic r [with linear effect partialled out] = 0.23, p < 0.01). The lines are the least-squares regression lines. Note that, as facial masculinity increases, men tend to have greater facial symmetry than women; as facial masculinity decreases, women tend to have greater facial symmetry than men. From figure 2 of Gangestad and Thornhill (2003a).
4. In college samples, symmetric men appear to be more muscular and vigorous (Gangestad & Thornhill, 1997a) and have lower basal metabolic rates (Manning, Koukourakis, & Brodie, 1997) than asymmetric men.
5. In a study conducted in Jamaica, Brown et al. (2005) found that men’s symmetry predicted the attractiveness of their dance movements. In that study, target individuals shown in video clips were reduced to stick figures through computer technology and, hence, all information aside from that reflected in dance movement was removed.
6. Symmetric men appear to have more attractive voice qualities than asymmetric men (Hughes et al., 2002).
7. As already noted, more symmetric individuals may be more intelligent than less symmetric counterparts.
Again, we suspect that these features (or some set of them) mediate FA’s association with sexual history, partly because of female preferences for them. Intrasexual competitive advantages, independent of female choice, may also play some role. And, as we describe in chapter 8, female preference for the scent of symmetrical men may play a role as well. As we also discuss in chapter 8, women particularly prefer this scent during estrus—when they are fertile in their cycles. Indeed, women particularly prefer a number of these correlates of men’s symmetry—intrasexual competitiveness, facial masculinity, muscularity, vocal masculinity—during estrus. Indeed, we argue that women’s preferences for symmetric men, which lead symmetric men to have a greater number of sex partners than asymmetrical men, is a noteworthy component of women’s estrous sexuality but not a dominant component of women’s extended sexuality.
When Møller (1990) and Thornhill (1992a, 1992b) first demonstrated female preferences for symmetrical males in barn swallows and scorpionflies, respectively, their favored interpretation is that male symmetry reflects individual variation in intrinsic good genes (see also Watson & Thornhill, 1994). Symmetrical males purportedly possess, on average, greater heritable fitness than do asymmetrical males. Female preferences for symmetrical males, then, function to obtain intrinsic genetic benefits for offspring. The heritable variation in developmental instability could be maintained by mutation-selection and/or coevolutionary processes (e.g., host-parasite coevolution). But, for females to prefer symmetrical males to obtain good genes, this heritable variation should relate substantially to general heritable variation in fitness.
Many studies have examined the heritability of FA, dating to work in the 1960s. Amazingly, however, we still know very little for sure about the heritability of developmental instability (for a recent review, see Leamy & Klingenberg, 2005). We perhaps do have a pretty good sense of the average heritability of FA for single traits. In the subsample of studies reviewed by Møller and Thornhill (1997) examining single trait FA, the mean was .041 (see Gangestad & Thornhill, 1999). In a subset of those studies that fully controlled for maternal effects, the mean was .025 (Whitlock & Fowler, 1997). Van Dongen (2000) applied Bayesian hierarchical analysis to 66 heritability estimates from 12 studies of single traits’ FA and found a mean value of 0.046. A more recent review reported a mean of approximately 0.03 (Fuller & Houle, 2003). And very recent studies are consistent with these estimates. For instance, Kruuk, Slate, Pemberton, and Clutton-Brock (2003) reported a mean h2 of .041 across four antler traits’ FA in red deer; Stige, Stagsvold, and Vøllestad (2005) found a mean h2 of 0.065 for two plumage traits in pied flycatchers; Roff and Reale (2004) estimated the FA of four traits in a cricket to have a mean h2 of 0.031. Though some variation in estimates across characters and species are observed, the estimates are remarkably consistent on the whole. Sampling error may in fact account for most of their variation.
A mean h2 of 0.03 is in fact very small. (Indeed, in most studies, a value this small is not statistically significant.) Single trait FA, then, has very low heritability. Just as correlations between single trait FA measures must be interpreted in light of the total variance accounted for by meaningful individual differences, however, so too must heritability. As we noted earlier, the median estimated proportion of variance in a single trait’s FA (at least in mammals) due to individual differences in developmental instability (the repeatability of FA) is about 0.07. Hence, if 100% of the individual differences in developmental instability were due to additive genetic variance, the heritability of the single trait FA would be just 0.07. To estimate h2 of underlying differences in developmental instability, one should simply divide the h2 of single trait FA by its reliable variance. On average, we expect a h2 of approximately 0.03/0.07 (≈ 0.4). If the mean correlation between two traits’ FA (about 0.05) is a better estimate of the proportion of variance in single traits due to systematic differences in FA, then the mean h2 of organism-wide developmental instability could be closer to 0.030/.05 (= 0.6).
These estimates can be illustrated with data from single studies. Again, Stige et al. (2005) estimated the h2 of asymmetry of two plumage features in pied flycatchers. The mean h2 estimate (not statistically significant) was 0.065. Stige et al. (2005) also estimated the repeatability of the traits’ FA, which averaged 0.14. The estimated h2 of developmental instability in this study, then, is 0.065/0.14 = 0.46. The authors nonetheless stated in the title of the article that fluctuating asymmetry in these features is “not heritable” (despite acknowledging that the 95% confidence intervals around their estimates of the h2 of developmental instability contain a value of 1.00!). (Van Dongen, 2000, estimated lower h2 for developmental instability, but the repeatabilities in the studies he analyzed were unusually large. We suspect that his results are not representative, though further empirical work must ultimately decide this issue. See Gangestad and Thornhill, 2003b.)
In our view, the available data are consistent with developmental instability typically being moderately heritable, contrary to the conclusions offered by some observers (e.g., Leamy & Klingenberg, 2005; Van Dongen, 2006). Admittedly, few single studies provide clear, strong conclusions. As Fuller and Houle (2003) note, most studies simply have very little power to detect meaningful heritable variation in developmental instability, particularly when they investigate a handful of traits or fewer. Few reports estimate h2 of underlying developmental instability. Nonetheless, in our view the values typically found for h2 of FA of individual traits and the reliable variance in FA of individual traits points to a fairly clear expectation: an h2 of developmental instability of about 0.4.
Naturally, this value may vary across species or, within species, across environments. Only two studies have estimated the h2 of FA in human skeletal and morphological traits (as opposed to dermatoglyphic FA, which appears to have different causes due to these traits being set by the second trimester of gestation). Livshits and Kobylianski (1989) estimated a composite of eight traits to have an h2of 0.31 in an Israeli sample. More recently, Johnson, Gangestad, Segal, and Bouchard (in press) examined heritability of FA based on samples of monozygotic and dizygotic twins reared apart. They estimated a heritability of a 10-trait composite to be about 0.3. As even aggregates of the FA of 8 and 10 traits do not measure organism-wide developmental instability extremely well, the heritability of developmental instability in these populations is almost surely substantially greater than 0.3. (As we observed elsewhere [Gangestad & Thornhill, 1999], Livshits & Kobylianski’s estimate assumed that all parent-offspring correlation was due to heritable effects, but Johnson et al.’s did not. See also Sangupta & Karmakar, 2007.)
Leamy and Klingenberg (2005) suggest that FA and developmental instability may have large amounts of nonadditive genetic variance, particularly in the form of epistatic variance. In some systems (e.g., mice) that might be the case, though more research is needed before generalizations can be made. Epistatic variation can reflect a history of selection on additive genetic effects (Hansen, Alvarez-Castro, Carter, Hermisson, & Wagner, 2006). As developmental instability probably has been under directional selection in many systems (Møller, 1999), perhaps it should not be surprising if it does indeed have meaningful amounts of epistatic variance. Nonetheless, for reasons already discussed, we suspect that it also typically has additive genetic variance.
One can also estimate the CVA of developmental instability. If developmental instability is a fitness trait (and taps heritable variation related to fitness), it should, of course, have a high CVA. And, indeed, if h2 of single trait FA is 0.03, the CVA of developmental instability is indeed high—at least 14 and possibly higher. (These estimates do not vary much with the proportion of variance in single trait FA accounted for by developmental instability; see Gangestad & Thornhill, 2003b, for details). Developmental instability, then, appears to have the signature of a fitness trait.
Our working hypothesis is that developmental instability in human populations has indeed historically been a fitness trait (even if, given modern medicine and reliable birth control, it no longer is). We furthermore work with the hypothesis that it has historically tapped meaningful additive genetic variance in fitness. Though we think that much data are already consistent with this view, we realize that additional work on FA, developmental instability, and fitness is needed before firm conclusions can be drawn.
In chapter 6, we suggested that women’s ornaments have been under sexual selection to signal heritable quality in condition, as well as age- and condition-based reproductive value. Ancestrally, females in better condition, for reasons deriving from both their genetic makeup and environmental circumstances, were better able to store gynoid fat, specialized for reproduction, compared with females in worse condition. Differential readiness for reproduction as a function of condition did not require sexual selection. Male preferences were shaped by sexual selection as a result of covariance between female readiness to reproduce and condition. Males who preferred young females who stored gynoid fat and otherwise showed evidence of reproductive readiness (e.g., had features indicative of estrogenization), all else being equal, had greater fitness than those who did not, and potentially for a variety of reasons: Their mates had a longer reproductive lifespan ahead of them; their mates could produce healthier offspring because of their resources; their offspring received genetic benefits from their mates. These preferences then exerted sexual selection on females to display favored traits. Those females in best condition and highest genetic quality paid fewer costs for their marginal increases in investment in these traits due to sexual selection, and the displays stabilized as honest signals of condition and quality.
In chapter 6, we discussed evidence that patterns of gynoid fat deposition and other features affected by estrogen are associated with women’s reproductive health and general condition. If they are indicative of overall quality, we might also expect them to be associated with developmental instability. In fact, some evidence is consistent with this expectation. The most direct evidence comes from a study of about 200 women in Poland, in which reproductive hormones were measured throughout women’s cycles. FA of finger lengths was assessed. Symmetric women had higher estradiol levels over the menstrual cycle than did asymmetric women (Jasiénska, Lipson, Ellison, Thune, & Ziomkiewicz, 2006). In this population, then, symmetric women are more fertile. They should also possess more estrogenized body and facial features, though this study did not examine those associations.
Other studies specifically examining correlations between attractive features in women and their symmetry have yielded mixed results. Grammer et al. (2002) and Schaefer et al. (2006) reported that facial and nude bodily attractiveness in women is positively associated with body and facial symmetry. Koehler et al. (2004) found that women with symmetric bodily traits have more feminine facial features than do asymmetric women. By contrast, we found no significant association between women’s facial femininity and symmetry (Gangestad & Thornhill, 2003a). Women’s facial symmetry predicts their attractiveness (see review in Rhodes, 2006). But reported correlations between women’s facial attractiveness and body symmetry have been fairly small (e.g., Gangestad et al., 1994; Thornhill & Gangestad, 1994). Hughes, Harrison, and Gallup (2002) reported that women’s symmetry predicts their vocal attractiveness. We have not, however, found women’s waist-to-hip ratio (WHR) to be predicted by their symmetry (unpublished data).
One factor that complicates interpretation of this mixed pattern of results is the apparent fact that women’s symmetry changes across the cycle (Scutt & Manning, 1996). These changes may be due to variations in water retention, which can affect symmetry of soft tissues and joint spaces. As result, symmetry measures in women may reflect within-individual variation that obscures meaningful between-individual variation.
Women’s ornaments appear to clearly relate to reproductive health and reproductive value (see chapter 6). Evidence that they reflect variation in developmental instability and genetic quality is less conclusive.
Earlier, we discussed correlates of male symmetry. In short, symmetric men are more “masculinized” than their asymmetric counterparts. Symmetric men appear to have more masculine facial features (Gangestad & Thornhill, 2003a; cf. Koehler et al., 2004). They behave in more intrasexually competitive and confrontative ways (Simpson et al., 1999) and, in U.S. and U.K. samples, report getting in more fights (Furlow et al., 1998; Manning & Wood, 1998). Some evidence suggests that, on average, they are more muscular than asymmetric men (Gangestad & Thornhill, 1997a) and have more masculine body shapes (e.g., broader shoulders and masculine upper-body to lower-body proportions; Brown; Price, Kang, Zhau, & Yu, 2007). They have more attractive voices (Hughes et al., 2002), which may reflect their having deeper, more masculine voices (e.g., Feinberg, Jones, Little, Burt, & Perrett, 2005).
Just as researchers have hypothesized that women’s estrogen-dependent features have been exaggerated as signals of condition and quality through sexual selection, researchers, including us, have argued that men’s masculinized features have been under sexual selection and hence partly function as signals of condition and quality (e.g., Thornhill & Gangestad, 1993, 1999a; Penton-Voak et al., 1999). The scenario we envision is analogous to the scenario we described in chapter 6 for female estrogen-dependent features. Compared with less fit males, males who were in better condition could adaptively afford to allocate more energy and other somatic resources into mating effort and intrasexual competitiveness. A variety of “masculine” physical and behavioral traits promote mating effort. These traits covaried with male quality, then, and could serve as cues of male quality. Female mate preferences for these traits were selected. In turn, men were sexually selected to allocate greater effort into these traits and, as a result, these traits came to function partly as signals of quality.
Just as many female signals of quality are facilitated by a primary reproductive hormone in women—estrogen—many male signals of quality appear to be facilitated by the primary reproductive hormone in men—testosterone (T). T is phylogenetically old—not as old as estrogen but still quite old, originating in early vertebrates (see chapter 8). In many species, a threshold of T must be achieved for many basic reproductive traits in males (such as sex drive) to be engaged (though the T level of most males of reproductive age exceeds that threshold, and T does not appear to have strong dose-dependent effects above the threshold; for work on humans, see Bancroft, 2002). In broad strokes, it also functions to facilitate pursuit of mates, male–male competition and, thereby, access to mates. Just as the precise manifestations of estrogen have been modified within particular species or taxa, so too the precise mechanisms that regulate T have been modified in specific species since its evolutionary debut.
Following Bribiescas (2001), we conceptualize the function of testosterone in a life history framework (see also chapter 4). Specifically, we see T as a modulator of resource allocation—when resources to be allocated include energy but also time and utilization of functional structures, including neural ones. Again, from a life history perspective, organisms have finite time budgets and hence can harvest energy at a finite rate. In allocating time and energy to fitness-enhancing activities, then, they face trade-offs. Human puberty marks a time when allocation of the energy budget is shifted from growth to reproduction. In males, testosterone plays key roles in that shift.
Though production of male sperm cells is very cheap (such that, even when extremely malnourished, men still produce sperm at rates similar to those of well-nourished men; see Ellison, 2001, 2003), reproduction is typically not cheap for males, even in species in which males do not care for offspring. Males must find and compete with other males for mates, activities that require much energy to engage in effectively. For instance, male muscle mass contributes to mating effort in many mammalian species and demands much energy to build, maintain, and use. Sexual dimorphism in musculature, particularly of the upper body, emerges in humans during adolescence. Male increases in upper body musculature are facilitated by testosterone (e.g., Basaria et al., 2002; Bhasin, 2003; Schroeder et al., 2003). Males must trade off allocation of effort to mating and somatic maintenance (e.g., immune function) and, of course, can never afford to shift all energy away from somatic maintenance; how much they can afford to allocate to mating effort depends on a variety of factors, including their condition. In species in which males do engage in biparental care, males also face a trade-off between two forms of reproductive effort, mating effort and parental effort. As we described in chapter 4, men typically experience a reduction in testosterone when mated or when fathers.
At a general conceptual level, T might be thought of as a hormone that facilitates male mating effort (Bribiescas, 2001). More T results in greater mating effort. Less T is associated with less mating effort and more somatic maintenance and/or parental investment. A variety of features facilitate mating, others facilitate parenting, yet others facilitate somatic maintenance and survival. Each feature may be modular in the sense that, for instance, muscle growth involves mechanisms separate from, say, focus on male-male competition. Endocrine hormones, such as T, are messengers in distributed communication systems that can coordinate adaptive changes in whole suites of such modular features. Selection has presumably shaped the T system—the mechanisms that regulate its release and metabolism, as well as the precise distribution of T receptors in structures—such that, based on inputs to the system, it leads to optimal allocations of effort to mating, parenting, and so on, in environments in which the mechanisms were shaped by selection. This view, though undoubtedly overly simplistic, is a reasonable working model (see Bribiescas, 2001; Ellison, 2001, 2003).
Again, though T is not necessarily expensive, its effects—those that facilitate mating effort—are potentially very expensive. Different males under different circumstances and of different quality may be able to afford different levels of these costs (when allocating effort optimally; e.g., Grafen, 1990; Getty, 2006). As we noted in chapter 4, fathers may do best by allocating proportionately fewer of their energetic resources and other resources to mating. And, all else equal, males in good condition optimally allocate more energy to mating effort than individuals in worse condition. Hence, mechanisms regulating T modulate it partly as a function of condition.
In addition to facilitating muscle growth, testosterone facilitates masculinization of the male face (see Swaddle & Reierson, 2002). In one study, women judged men’s masculinity based on facial photographs. Women’s judgments correlated positively with men’s measured T levels (Roney, Hansen, Durante, & Maestripieri, 2006; see also Penton-Voak & Chen, 2004). Testosterone also affects men’s interactions with and patterns of attention to other men (see, for instance, Ellison, 2001; Mazur & Booth, 1998). Testosterone-facilitated male dominance-seeking may be expressed, among other behaviors, in greater selective attention to angry faces (van Honk et al., 1999; van Honk et al., 2000), in less pronounced smiling (Dabbs, 1997), and in more visual attention toward interaction partners (Dabbs, Bernieri, Strong, Campo, & Milun, 2001). In nonhuman animals, T suppresses fear (perhaps particularly of a social nature); two double-blind studies on humans (though women, not men) showed that an administration of T reduced fear in response to pictures of fearful faces (van Honk, Peper, & Schutter, 2005) and a fear-potentiated startle response (Hermans, Putnam, Baas, Koppeschaar, & van Honk, 2006). Reduced potentiation of fear may lead individuals to be more likely to engage in potentially injurious conflict. Men who score higher on a test of power motivation have higher T (Schultheiss, Dargel, & Rohde, 2003). Moreover, large increases in T following a real or imagined victory in a competition are associated with power motivation in men (Schultheiss, Campbell, & McClelland, 1999; Schultheiss & Rohde, 2002). In one study, the opportunity to interact with an attractive woman led to increases in male T, particularly for men who evidenced the greatest interest in her (Roney, Mahler, & Maestripieri, 2003).
Traits such as muscularity and willingness to engage in male–male contests have real costs. Muscles require energy, and contests could result in injury. One question that arises concerns what keeps traits such as facial masculinity and vocal masculinity honest signals of underlying quality. In and of themselves, they do not appear to be expensive to achieve. One plausible answer is that these traits impose socially mediated costs. In some species of birds, badges and patches regulate male-male competition. Males with bigger patches typically win bigger or better territories. Large patches are honest signals of male ability to engage effectively in male-male competition because males who have large badges will be tested. Thus, if small-badged male Harris sparrows are artificially given large badges, they are aggressed against, typically lose competitions, and end up worse off than they would if they had simply been left with their small badges (Rohwer & Rohwer, 1978). Male facial and vocal masculinity may similarly function to regulate male-male competition (see, e.g., Mueller & Mazur, 1997). Consistent with this interpretation, Puts, Gaulin, and Verdolini (2006) found that people perceive men with deeper voice pitch to be more physically and socially dominant than men with higher voice pitch. Moreover, when a man addresses another man whom he believes is less physically dominant than he is, he speaks at a lower pitch. By contrast, when a man addresses another man whom he believes is more physically dominant than he is, he speaks at a higher pitch (Puts et al., 2006).
Masculine male traits facilitate effective performance in contests between men. Effective performance in these contests, as well as the traits that function or relate to effective performance, function to signal quality, including genetic quality. Women, then, prefer these traits in their mates. Naturally, however, men who can win dominance competitions with other men might be expected to gain access to material benefits. Women could profit from mating with masculine, dominant men because such men better deliver material benefits (e.g., food, shelter, physical protection), relative to less masculine, less dominant men. How can we know, then, that female preferences for masculine traits evolved to (at least partly) function to obtain genetic benefits for offspring? Perhaps the function of these preferences is merely to obtain nongenetic material benefits.
We note that, although masculine, dominant men possibly could provide greater levels of material benefits to female mates, they also might actually deliver fewer material benefits. In chapter 4, we described research on trade-offs between male-delivered genetic benefits and male provisioning that face females in many socially monogamous bird species. For instance, in collared flycatchers, males with large forehead patches establish territories earlier than small-patched males (Pärt & Qvarnström, 1997). They also provide genetic benefits relative to males with small patches (Sheldon et al., 1997). But small-patched males are better providers; they feed more at the nest (Qvarnström, 1999). Large-patched males reserve effort for seeking extra-pair copulations (EPCs) in the same season or, possibly, future seasons. More generally, in species in which females engage in EPC, relatively attractive males provide fewer material benefits (through foraging) than unattractive males provide (Møller & Thornhill, 1998a).
The same could be true of masculine and symmetric men, relative to more feminine and asymmetric men. Indeed, symmetric men tend to have more EPCs (Gangestad & Thornhill, 1997b). Men with masculine bodies and faces report greater success than less masculine men at short-term mating but not in forming long-term, stable pair bonds (Rhodes et al., 2005). Muscular men, relative to less muscular men, similarly succeed at short-term, but not necessarily long-term, mating (Frederick & Haselton, 2007). As Rhodes et al. (2005) conclude, their “findings suggest that individuals of high phenotypic quality have higher mating success (and, we note, for males, particularly short-term mating success) than their lower quality counterparts” (p. 186).
In a study in which we administered questionnaires to both men and women, which they completed privately in separate rooms, we found that symmetric men, relative to their less symmetric counterparts, invested less in their romantic relationships, based on responses by both men and women to a validated measure of self- and partner investment in their relationship (Ellis, 1998). Symmetric men, compared with asymmetric men, were relatively unwilling to give their time to their partners, were dishonest with their partners, and sexualized other women more often. Though they were seen as more able to provide physical protection than asymmetric men, we found no evidence that they were willing to dedicate time to do so (see Gangestad & Thornhill, 1997b). Women, similar to females of other species, may face trade-offs between various forms of benefits—genetic benefits and nongenetic material benefits—that mates can provide (e.g., Gangestad, 1993; Penton-Voak et al., 1999).
Women’s and men’s perceptions of masculine and feminine men are consistent with women’s facing this trade-off. Men with feminine faces are perceived to be warmer, more agreeable, and more honest than men with masculine faces (Fink & Penton-Voak, 2002). Men with masculine faces are seen to be more likely to engage in male-male competition (e.g., get into physical fights) and pursue short-term matings (e.g., sleep with a lot of women, cheat on partners), whereas men with relatively feminine faces are seen to be more likely to be good, stable, long-term mates (e.g., be caring and emotionally supportive, be great with children; Kruger, 2006).
As we have emphasized throughout this book, the primary evidence for function and the selective pressures that gave rise to adaptations is to be found in features’ design. According to the view that women trade off material benefits for genetic benefits to be mated with masculine men, attraction to male masculinity should have been shaped by selection to be conditional—to depend on conditions that affect (or ancestrally would have affected) the relative value of heritable condition and paternal investment. A number of lines of evidence suggest that it is.
1. Preference varies as a function of relationship context. The face women find most attractive in short-term mates is more masculine than the face they find most attractive in long-term mates (Penton-Voak et al., 2003). In one recent study, women were asked to choose which male they’d prefer as a mate, a male shown to have a masculine face or a male shown to have a feminine face. As it happened, mating context drove female preferences. When choosing a sex or affair partner, most women (57% and 66%, respectively) choose the masculine male. When it came to choosing a marriage partner, however, most women (63%) preferred the feminine male. (As we previously described in chapter 4, women even more strongly preferred the feminine male as a son-in-law [73%] and similarly thought that their parents would prefer them to date the feminine male [71%].)
2. Attractive women have a stronger preference for masculine faces. Little, Burt, Penton-Voak, and Perrett (2001) reasoned that attractive women need not trade off male condition and investment as markedly as must unattractive women; masculine men should be more likely to invest in relationships with attractive women. In fact, attractive women do more strongly prefer facial masculinity (Little et al., 2001; Penton-Voak et al., 2003).
3. Preference varies with culture. Penton-Voak, Jacobson, and Trivers (2004) proposed that women’s preference for masculinity should have been selected to be sensitive to cues of the relative value of condition (and genetic benefits) and investment of male mates in their local ecologies. In Jamaica, infectious disease is more prevalent and male parental investment less pronounced than in the United Kingdom. They predicted and found that Jamaican women show greater preference for facial masculinity than do British women.
As we discuss in detail in chapter 9, women also particularly prefer masculine male traits when in estrus, and particularly so as sex partners. Changes in preferences across the cycle, we argue, reflect female design to weight signals of heritable condition more heavily when they are fertile, particularly when selecting a sex partner.
In sum, the design of female preferences, in concert with other evidence, is consistent with the view that male masculine features function partly as signals of heritable quality and, hence, intrinsic genetic benefits to offspring. We know of no alternative hypothesis that can explain the variety of findings consistent with this view.
We noted earlier that FA is negatively associated with psychometric intelligence. Symmetric men score higher on standardized tests of intelligence than do asymmetric men. (The same may well hold for women [e.g., Furlow et al., 1998], though fewer data are available on women. See also Rahman et al., 2004.) Might intelligence be a cue of heritable fitness? Indeed, might intelligence have been sexually selected to signal heritable fitness?
Miller (2000) devoted most of a full-length book to developing the thesis that, indeed, human intelligence has been sexually selected to signal heritable fitness. His claim is that intelligence functions much like the peacock’s tail: Because it is difficult for individuals in poor condition to develop a brain that demonstrates the complexity we recognize as human intelligence, well-developed brains (and their cognitive manifestations) advertise intrinsic good genes.
If, indeed, human intelligence advertises good genes, we suspect that its origins as a signal are more similar to the origins of female ornaments and male masculine traits than to the origin of the peacock tail. In chapter 5, we described two routes through which a trait that ultimately becomes a signal of quality first acquires an association with fitness. The first is the preferred-signal-through-sensory-bias route. In that scenario, the trait is first preferred due to the by-product of a sensory adaptation that does not function in mate choice. The trait then evolves through sexual selection to a point at which it becomes associated with quality because high-quality individuals can best afford to develop large signals. The peacock tail purportedly evolved to be a signal of quality through this route.
In the second route, a functional trait is first correlated with heritable fitness simply because individuals of higher quality can better afford it. Individuals of the other sex prefer the trait because it is a cue of quality. The value of the trait added by its being a cue of fitness leads individuals to dedicate more effort to the trait, exaggerating it as a signal. Female gynoid fat depots and male muscularity do not function purely as signals; they do have sexually selected signaling properties, but they have functions that were not sexually selected, as well. Similarly, human intelligence may have sexually selected signaling properties, though, in our view, it clearly has other functions as well (see Gangestad & Simpson, 2007, for a range of views about the evolution of human intelligence).
If intelligence has evolved as a signal of genetic quality, however, we might expect it to have many of the same correlates that male masculine features do. Intelligent men, similar to masculine men, should be expected to be advantaged in short-term mating and hence be more likely to engage in short-term sexual relations than less intelligent men. They should invest less in their romantic relationships. They should be particularly preferred by women as sex partners, not long-term mates. And they should be more strongly preferred where genetic quality has large effects on fitness.
Most of these expectations are not borne out. Male intelligence does not correlate positively with number of sex partners in college samples (I. Tal, unpublished data; E. White, unpublished data). We know of no evidence that intelligent men are less kind, caring, and investing in their relationships than less intelligent men. And women tend to seek intelligence in long-term mates more than they seek it in short-term mates (Gangestad et al., 2007). Gangestad et al. (2006) did report that, cross-culturally, women’s preference for intelligence in a mate covaries positively with parasite prevalence, just as preferences for physical attractiveness and health do.
One possibility is that intelligence did in fact evolve as a signal of genetic quality, but one that operates in circumstances in which high quality males do not do best by exerting mating effort. Rather, just as attractive males in bird species in which few extra-pair mating opportunities are available do best by provisioning rather than seeking extra-pair mates (Møller and Thornhill, 1998a), perhaps intelligent men do best by investing in their relationships and offspring. Answers to the question of why intelligence would operate in this fashion, but masculinity does not, will require additional theoretical and empirical work.
Another possibility is that some manifestations that display intelligence do in fact operate in much the same fashion that male masculinity does. Humor and creative displays may function as mating effort, which women find attractive in sex partners (e.g., Miller, 2000). The question of why these displays would work in this way although the underlying trait they purportedly display—intelligence—does not will, once again, require additional work to answer.
Earlier in this chapter, we argued that there is much heritable variation in fitness in natural populations. This heritable variation is the basis for the evolution of mate choice for intrinsic good genes. For individuals to be able to choose mates for intrinsic good genes, however, there must be available to them phenotypic features that reflect that heritable variation. Moreover, these features should capture the heritable basis of fitness but not strongly reflect the nonheritable basis of fitness. Hence, good indicators of heritable fitness should not only have much genetic variation (i.e., large CVA values) but should also have high heritabilities (e.g., Pomiankowski & Møller, 1995; Rowe & Houle, 1996).
We have argued that a variety of masculine features, in concert with some other indicators (e.g., coordination, as might be reflected in dance; e.g., Brown et al., 2005), function as signals of genetic quality. To say so does not imply that they covary with fitness in modern environments. It does imply that they did so in ancestral environments and that female preferences for these features evolved because they covaried with fitness. At least in ancestral environments, these traits (or some linear or nonlinear combination of them, reflecting how they are utilized as cues) should have high CVA values. They should also possess high heritabilities. Future work should address their genetic variation and heritabilities.
As we have discussed, these features and preferences for them do possess signatures of having been shaped partly in the context of a system of honest signaling of heritable fitness. In light of these signatures, we once again proceed with the working assumption that, indeed, sexual selection for displaying and choosing mates with intrinsic good genes played a role in the evolution of these features and women’s preferences for them.
Potentially, individuals can choose mates for one or more of three types of genetic benefits for offspring. Intrinsic good genes are alleles that benefit offspring independent of the genotype of the mate chooser. Compatible genes are alleles that work well with the alleles of an individual mate chooser, though not with all the alleles of all mate choosers. Choice for diverse genes leads to diversification of offspring.
Adaptive mate choice for intrinsic good genes requires that there exist additive genetic variance in fitness in the population. Until the past two decades, evolutionary geneticists have typically assumed that fitness has little genetic variance because selection persistently removes it. It now appears that fitness components (e.g., longevity, fecundity) have a lot of additive genetic variance in natural populations relative to traits under stabilizing selection. A variety of processes contribute to the maintenance of additive genetic variance in fitness, despite selection. One important one is mutation-selection balance. Others are Red Queen or antagonistic coevolutionary processes, which lead to relatively rapid changes over time in which alleles are favored. Host-pathogen coevolution is one example (a process that may operate in concert with negative frequency-dependent selection). Other examples include forms of intraspecific antagonistic coevolution, such as that due to maternal-fetal conflicts of interest. One potentially important form of intraspecific antagonistic coevolution is sexually antagonistic coevolution. Though it appears that slightly deleterious mutations can maintain a substantial amount of additive genetic variance, the precise relative contributions of processes responsible for this variation in humans remains unknown.
Although the viability-indicator and Fisherian models of mate choice have traditionally been thought to represent competing views about the nature of good-genes mate choice, in fact they appear to anchor the ends of a sexual selection continuum. These ends differ not in terms of whether individuals who possess good genes (as well as their offspring) are truly high-quality individuals, as once thought. Instead, they differ in terms of how individuals who are of high quality have been selected to allocate their effort, relative to individuals of low quality. At the viability-indicator end of the continuum, high-quality individuals maintain a viability advantage over low-quality individuals. At the Fisherian end of the continuum, high-quality individuals have been shaped to strongly invest in traits that lead to mating benefits and, accordingly, may die, on average, at younger ages than low-quality individuals do.
Adaptive mate choice for compatible genes requires nonadditive genetic effects on fitness, which can be due to dominance (e.g., heterosis) or epistatic effects. Two examples in humans and many other organisms appear to be mate choice for MHC dissimilar individuals and incest avoidance.
Adaptive choice for diverse genes may be selected when individuals in the same family benefit by being different from one another. Women’s preference for the scent of men heterozygous at MHC loci may be one example.
Developmental instability, as reflected in fluctuating asymmetry, is associated with fitness in many natural populations. In modern human populations, developmental instability appears to be at least moderately heritable.
We suggest that, just as a primary female reproductive hormone, estrogen, facilitates the development of sexually selected indicators of genetic quality in women, a primary male reproductive hormone, testosterone, does the same in men. A variety of “masculine” traits, known or likely to be testosterone-facilitated, are associated with developmental instability in men. Because women prefer men who possess these traits as mates, owing to the traits being indicators of intrinsic genetic quality (at least ancestrally), these men typically invest less in long-term relationships with women. As expected, then, women particularly prefer these traits in sex partners rather than in stable longterm mates; attractive women prefer these traits in long-term partners more than do unattractive women (as these men are more likely to invest heavily in relationships with attractive women); and preferences for these traits appear to vary across cultures. In cultures and ecologies in which paternal care is particularly important, women should prefer masculine men as long-term partners less than in cultures and ecologies in which paternal care is less important.