Inbreeding depression and genetic load of sexually selected traits: how the guppy lost its spots

Authors


 Cock van Oosterhout, Molecular Ecology & Fisheries Genetics Laboratory, University of Hull, Hull HU6 7RX, UK. Tel.: +44(0) 1482 465505; fax: +44(0) 1482 465458; e-mail: c.van-oosterhout@hull.ac.uk

Abstract

Abstract To date, few studies have investigated the effects of inbreeding on sexually selected traits, although inbreeding depression on such traits can play an important role in the evolution and ecology of wild populations. Sexually selected traits such as ornamentation and courtship behaviour may not be primary fitness characters, but selection and dominance coefficients of their mutations will resemble those of traits under natural selection. Strong directional selection, for instance, through female mate-choice, purges all but the most recessive deleterious mutations, and the remaining dominance variation will result in inbreeding depression once populations undergo bottlenecks. We analysed the effects of inbreeding on sexually selected traits (colour pattern and courtship behaviour) in the male guppy, Poecilia reticulata, from Trinidad, and found a significant decline in the frequency of mating behaviour and colour spots. Such effects occurred although the genetic basis of these traits, many of which are Y-linked and hemizygous, would be expected to leave relatively little scope for inbreeding depression. Findings suggest that these sexually selected traits could reflect the genetic condition or health of males, and thus may be informative mate-cue characters for female choice as suggested by the ‘good genes’ model.

Introduction

Mating among relatives generally changes genotype frequencies in populations and can result in a decline in mean phenotype and fitness, a phenomenon known as inbreeding depression (Lynch & Walsh, 1998). Large panmictic populations do not typically suffer from inbreeding depression, because recessive deleterious mutations occur at low frequencies, meaning that they are seldom expressed. The partial dominance hypothesis is empirically perhaps the best supported model of inbreeding depression (Lynch & Walsh, 1998; Roff, 2002), and suggests that the increase in homozygosity of loci during inbreeding exposes the mutation load of deleterious recessives. An alternative explanation for inbreeding depression is given by the overdominance hypothesis, which suggests that inbreeding reduces the frequency of superior heterozygotes, resulting in observed fitness loss (Charlesworth & Charlesworth, 1987). Observations using an RFLP linkage map of rice implicate overdominance, rather than dominance, as the major genetic basis of inbreeding and heterosis (Li et al., 2001). The same study furthermore showed that most QTL associated with inbreeding depression and heterosis in rice appeared to be involved in epistasis (Li et al., 2001). Such observations illustrate that various nonadditive gene actions can result in inbreeding depression (Crow & Kimura, 1970).

Inbreeding depression is often most severe in fitness-related traits that have many loci with nonadditive gene effects. Typically, embryo survival and early development are most affected (Keller & Waller 2002), although inbreeding depression can also affect traits expressed later in life such as longevity (Oosterhout et al., 2000). Studies on Drosophilia show that primary fitness characters such as viability, fertility and egg production generally exhibit the highest level of inbreeding depression (averaging approximately 50%), while morphological characters change by just a few per cent (Roff, 1998; DeRose & Roff, 1999). There is also considerable variation in inbreeding depression across species and populations, which may depend on variation in mutation rate (Lynch & Walsh, 1998), mating system, population structure and population size (Oosterhout et al., 2000). Despite noticeable variation across taxa, almost all species examined show some inbreeding depression. Furthermore, inbreeding depression also affects natural populations and could contribute to an increased extinction risk of small populations (Keller & Waller, 2002).

Hitherto, few studies have examined the effects of inbreeding on sexually selected traits (exceptions include, Meffert & Bryant, 1991; Sheridan & Pomiankowski, 1997; Meffert et al., 1999; Oosterhout, 2000). Inbreeding depression on traits such as ornamentation and courtship behaviour does not compromise the individual's survival, but may impair its attractiveness to the opposite sex and thus probability of mating. As such, sexually selected traits will be closely associated with fitness, and have been suggested to serve as fitness indicators by the ‘good genes’ model (Hamilton & Zuk, 1982; Kokko et al., 2002). Yet another reason to expect that these traits will show inbreeding depression is that (directional) selection through female mate choice will purge all but the most recessive deleterious mutations. The phenotypic variance as a result of these recessive mutations will be released as dominance variation once inbreeding commences.

Some sexually selected traits are known to show strong Y-linked inheritance (e.g. Houde, 1992; Brooks & Endler, 2001a). Without epistasis such traits are not expected to show inbreeding depression (Crow & Kimura, 1970), because phenotypic effects of deleterious mutations on the hemizygote (haploid) Y-chromosome are always fully expressed and cannot show dominance (or overdominance). Consequently, inbreeding does not expose the deleterious effects of such loci, and without epistatic gene interactions with dominant (autosomal) loci, inbreeding will not change phenotypic value of purely Y-linked traits.

The guppy Poecilia reticulata (Peters) is well known for studies on sexual selection (Houde, 1997). Behavioural studies on the role of male colour pattern elements in sexual selection show that carotenoid colours (orange) are particularly important in female mate choice, possibly because they reflect the nutritional and health status of a potential mate (Kodric-Brown, 1989; Grether, 2000). Females may benefit directly from colour pattern cues by avoiding the risk of parasite infection, or indirectly when carotenoid colours provide cues to ‘good genes’ that can be inherited by the offspring (Houde, 1987; Endler & Houde, 1995).

The ornamentation of male guppies has also received attention in classical genetic studies that mapped many of its colour pattern genes (Winge, 1927). In more recent quantitative genetic studies, the heritability of many male sexual traits (i.e. colour pattern elements and courtship behaviour, Houde, 1992; Brooks & Endler, 2001a) were estimated. These studies show that heritability estimates can exceed unity, indicating that male sexual traits have some Y-linkage (Falconer, 1996). In fact, at least 20 colour pattern genes have been identified on or near the nonrecombining section of the Y chromosome (Winge, 1927; Winge & Ditlevsen, 1947), and these genes are inherited as a Y-linked supergene (Yamamoto, 1975). Guppy colour pattern is furthermore encoded by 17 alleles that recombine between the X and the Y chromosome, and five autosomal genes that control body colour rather than particular colour spots (Yamamoto, 1975). The autosomal and X-linked genes become expressed only in the males, as they are sex-limited and hormone-mediated (Hildemann, 1954). The ornamental genes of the Y-linked supergene are all located within two recombinational map units (centimorgans) of the sex-determining locus (Winge, 1927).

The reduced recombination rate at this ‘hot spot’ for colour genes can result in genetic degeneration by Muller's ratchet, background selection, the Hill–Robertson effect, and the ‘hitch-hiking’ of deleterious alleles by favourable mutations (Rice, 1987, 1994; Charlesworth & Charlesworth, 2000). Mutation accumulation of this Y-linked supergene will not result, however, in inbreeding depression because these mutations are hemizygous. Nevertheless, sexually selected traits may still show inbreeding depression when the expression of these traits is condition-dependent. Inbreeding depression on Y-linked traits can also occur when epistatic interactions between (dominant) autosomal loci and hemizygous Y-linked genes result in impaired expression of sexually selected traits (Crow & Kimura, 1970).

Here, we analysed the effects of three generations of full-sib inbreeding on male colour pattern and mating behaviour in fish from a natural lowland guppy population. Although this inbreeding experiment was not designed to discriminate between condition-effects and epistasis, the observation of inbreeding depression would be consistent with the ‘good genes’ model, as it implies that hemizygous, sexually selected traits could reflect the general condition and/or genetic quality of potential fathers.

Materials and methods

Guppy population

Poecilia reticulata was collected from two lowland localities, subject to high predation regimes, on the Tacarigua river in Trinidad in February 1999. Sixty-one wild-caught females were each mated to a different male, and their ( F1 ) offspring used as founders of the inbred lines. Because we cannot exclude the possibility that females may have used stored sperm from a previous mating, some F1 sibling groups may have been sired by more than one male, and were possibly half sibs rather than full sibs. Therefore, critical analyses on epistasis (see below) have been conducted twice, assuming the F1 to represent half or full sibs. From each inbred line, one male and one female were chosen at random from the surviving offspring as parents of the next generation. Females produced four to 64 young ones per brood and large families were culled to eight young at the day of birth. If there were no surviving offspring of either sex at maturity, the line went extinct. Extinct lines were not replaced by surviving lines, and so the breeding scheme resulted only in selection within lines. Inbreeding was carried out for three generations. The adapted protocol for full-sib inbreeding is similar to that used by Hedrick (1994 ).

Experimental design

The fish were reared in a semiclosed recirculation system with a total water volume of approximately 1800 L comprising 440 transparent Dispo™ Safe jars of 3.7 L and a biofiltration unit with UV sterilizer. The flow rate in jars was approximately 3 L h−1, and fish were kept in full-sib families until sexual maturation. Fish were fed ad libitum with flake food and Brine shrimp (Artemia salina) at least twice a day, and maintained in a 12 : 12 light : dark cycle at 25 (±0.5) °C.

Male colour pattern and courtship behaviour was compared across three generations. Common environment effects might therefore confound inbreeding depression. However, by using partially overlapping generations, and analysing a mixture of males from different generations simultaneously, temporal effects and common environment effects were minimized. Furthermore, our design allows for a direct comparison of grandfathers, fathers and sons (i.e. patrilines) which has an important advantage of analysing the effects of inbreeding on individuals with very similar (or almost identical) Y chromosomes.

Colour pattern analysis

All colour pattern and behavioural analyses were conducted on males at least 1 month after they had reached full maturity, as indicated by complete development of the gonopodium (see Houde, 1997). Males were sedated using 120 mg L−1 MS222 for 60 s and placed at 40 mg L−1 MS222 during photographing. The left lateral side of 190 males was photographed using a 50-mm macro-lens-fitted camera with an indirect flash and shutter speed t=125−1 s and aperture F=5.6. Colour slides were projected onto a 70 × 50 cm poster frame fitted with graph paper, and the contours of all black and orange spots and stripes were measured. Repeatability of colour pattern measurements was analysed on two independent measurements on two photographs of 40 randomly picked males. We calculated the relative area of colour pattern elements to account for the effects of inbreeding depression on the size and total body area of fish.

Male mating behaviour

Observations on male mating behaviour of 202 males were carried out in a 46 × 30 × 26 cm tank with a 30-W light bulb fitted 10 cm above the water surface. The same 16 adult (nonvirgin) females were used throughout the experiment. Additionally, two control males were observed at each recording event to account for temporal variation. Eight experimental males were placed in the tank the night before observations commenced; a heterogeneous mix of age-classes of males from all three generations was used. The males were observed in random order in two repetitions of 5 min each using an event-recording program (Etholog, version 2.2.4).

The recorded components of male mating behaviour were divided into state events (that have a measurable time) and frequency events. State events include the time spent following females (‘Following’), charging females (‘Charging’), luring a female away from other males (‘Luring’) and displaying (‘Sigmoiding’). Displays were recorded when the focal male moved in front of the female, arched his body in an S-shaped posture and quivered, and we distinguished between the intensity of sigmoiding behaviour (Baerends et al., 1955). We recorded high-intensity sigmoiding when males completely spread their caudal and dorsal fins and moved up and down distinctly. A low-intensity display is generally shorter, and the fins are kept at least partially folded (Baerends et al., 1955). Charging or chasing females was recorded when two (or more) males were in pursuit of the same female and jockeying for position (Houde, 1997). The recorded frequency events include the number of times a male performed sneaky mating attempt by ‘Gonopodial Thrusts’, which was recorded when the male made physical contact with the female's genital pore without her co-operation or prior display (Baerends et al., 1955).

Repeatability of behavioural traits among days was analysed by recording 30 random fish twice over a period of 7–29 days. Repeatability within observation days was calculated using observations recorded during two repetitions of 5 min each for all 202 males.

Statistical analyses

Logarithmic [log10(x + 0.01)] transformations were used to account for scale effects and nonadditivity (Lynch & Walsh, 1998). Normality of residuals, and homogeneity of variances for both colour pattern elements and behavioural traits were tested using a W-test (normality) and Bartlett test, respectively.

An analysis of covariance was applied in a General Linear Model (GLM) to quantify the effects of inbreeding on colour pattern and mating behaviour. The factor ‘F-coefficient’ has three factor levels corresponding to the three levels of inbreeding, indicating whether the trait shows significant inbreeding depression. Analysis with the F-coefficient F=0.25, F=0.375 and F=0.5 for the first, second and third generation are reported only, but qualitatively similar results were obtained when assuming half-sib families in the F1 generation. Additionally, male age (in days) was used as a covariate to test for the effects of age on male sexual traits. As this covariate did not explain significant variation in male colour pattern or courtship behaviour in the GLM with the factor ‘F-coefficient’ in the model, we omitted male age from the analysis.

The effect of inbreeding on male colour pattern was analysed further by estimating the number of detrimental equivalents (DE) for black and orange spots. We adopted a regression technique developed by Morton et al. (1956) that summarizes the deleterious consequences of inbreeding for traits that can be classified by incidence (e.g. the presence and absence of colour spots). To quantify the effects of inbreeding, we simply counted the number of colour pattern elements per guppy. An estimate for the genetic load expressed as the number of DE was then calculated as b=–Ln (Pf/Pf′)/ΔF, with variance Var(b) ≈ F2 ((1 − Pf)/(PfNf) + [(1 − P0)/P0N0]), where b is the effective number of deleterious equivalents, Pf and Pf′ are the proportion of individuals with given number of colour pattern spots and inbreeding coefficients F and F′, and ΔF is the increase in the individual's inbreeding coefficient (Lynch & Walsh, 1998).

Inbreeding depression can be potentially reinforced by epistatic effects between loci (Crow & Kimura, 1970), resulting in a nonlinear decline of phenotypic value with inbreeding coefficient. To overcome problems of nonindependence of commonly applied regression techniques (Lynch & Walsh, 1998), we also analysed the change in colour pattern within patrilines. The change in pigment area per increment in F between grandfathers to fathers (F1 to F2), and fathers to sons (F2 to F3) were calculated. A Wilcoxon's signed-rank test (Sokal & Rohlf, 1995) was employed to analyse the relative change in colour pattern over generations, and to test the null hypothesis of a linear decline in colour pattern over F-coefficients. Because we could not exclude the possibility that some F1 siblings were sired by different males, we have assumed that the F1 generation was produced by the mating of half sibs, resulting in a more conservative analysis and downward-biased estimate of number of DE and regression slope b.

Results

Repeatability analysis

The repeatability values for colour pattern traits were high and ranged from 0.7851 (±0.0783) for area of the black spots to 0.9922 (±0.0059) for area of orange spots. The between-day repeatability of behavioural traits for 30 males ranged between 0.42 (±0.09) for gonopodial thrusts to 0.64 (±0.07) for high-intensity sigmoid display. Luring showed a low repeatability (R = 0.17 (±0.08)) and was omitted from further analyses. Within observation days repeatability was higher [ranging from 0.55 (±0.05) for following females to R=0.72 (±0.03) for charging]. There is no evidence for temporal changes in the courtship activity of both control males over the 29-day observation period (‘Following’F1,55=3.69, ‘Charging’F1,55=1.39, ‘Sigmoid with high intensity’F1,55=0.08; ‘Sigmoid with low intensity’F1,55=2.04, all test results n.s). We interpreted this as no evidence for a directional shift in female response over the experimental period.

Inbreeding depression: colour pattern

All colour pattern elements and standard length showed significant inbreeding depression over three generations, in that all traits exhibited a significant decline in phenotypic value over generations (Table 1; Fig. 1).

Table 1.  Simple linear regression analysis on male guppy traits with generation as independent variable.
Traitd.f.r2 -adj Slope b (SE)tP
  1. The table shows the degrees of freedom (d.f.) of the test, adjusted r2 value, i.e. the proportion of variation explained by generation, the regression coefficient b with its SD within parentheses, the t-value testing the null hypothesis, the coefficient is equal to zero, and the P-value for this test.

Log (length)18913.1%−0.2866 (0.0549)−5.220.000
Log (% black)17410.1%−3.629 (0.8014)−4.530.000
Log (% orange)1859.2%−1.620 (0.3761)−4.310.000
Log (% stripe)1893.1%−0.1254 (0.0488)−2.570.011
Figure 1.

Inbreeding depression for the natural log-transformed relative area of black and orange colour spots (mean and SEM). The analysis shows a significant quadratic components for both the area of ‘black’ ( F1,188 =6.52, P =0.012) and ‘orange’ pigments ( F1,188 =4.32, P =0.039).

Figure 1 shows that there appeared to be no decline in the area of colour pattern elements orange and black in the F2 , but that there was inbreeding depression in the F3 . In order to test for nonlinearity between the total surface area of colour pattern elements and the inbreeding coefficients, least-square regression analyses were calculated, showing significant quadratic components for both the area of black and orange pigments (see Fig. 1 ).

We then estimated inbreeding depression within patrilines, and compared the decline in area of colour spots using reference phenotypes (grandfathers) with identical Y-linked colour pattern genes. The general pattern of decline in phenotypic values between generations was remarkably consistent across individual inbred patrilines (see Fig. 2). Wilcoxon's signed rank tests with 14 grandfather–father–son comparisons shows that the loss in colour pattern was more severe between generation F2 to F3 (i.e. fathers to sons), than from F1 to F2 (grandfathers to fathers) (see Fig. 2).

Figure 2.

The effects of inbreeding on the relative surface area of black and orange within patrilines during three generations of full-sib mating. Within patrilines there is a small decline in the relative pigment area of colour spots between the first and second generation, and a significantly greater loss between the second and third generation for both black ( Ts =16, d.f.=13, P  < 0.025) and orange ( Ts =9, d.f.=13, P =0.01).

In addition to the decline in total area of colour elements, the number of colour spots also declined significantly during three generations of inbreeding, with the number of black spots reducing by 25.1% and orange spots by 19.2%. The number of DE estimated over generations one to three revealed a high mutation load for the number of orange (DE=2.31 ± 0.80) and black spots (DE=4.89 ± 0.75).

Inbreeding depression: male mating behaviour

Courtship activity showed significant inbreeding depression, with the average time (±SE) males courted females declining from 67.0 (±2.5)% in the F1 and 48.1 (±2.6)% in the F3. Various aspects of male mating behaviour declined over generations, but it was especially the intensity with which males were courting and pursuing females that changed most markedly. For instance, although F3 males did not show a significant reduction in frequency of sneaky matings by gonopodial thrusts [b=−0.395 (±0.259), t161=−1.52, n.s] and low-intensity sigmoid displays (b=−0.085 (±0.395), t161=−0.21, n.s), significant inbreeding depression was observed for high-intensity displays [b=−2.64 (±0.68), t129= −3.87, P < 0.001]. Inbred males also charged and jockeyed females less frequently [b=−0.245 (±0.049), t161=−5.03, P < 0.001], which does not involve overt aggression, but rather appears to be an attempt to remain with the female and exclude other males from copulation (Houde, 1988). Finally, average sigmoid duration significantly declined over generations (Fig. 3), suggesting that inbred guppies were less motivated to display. Although the data presented in Fig. 3 appears to show curve-linearity, the quadratic component is not significant at 5% level.

Figure 3.

Mean sigmoid duration (±SE) in seconds exhibited by males over three generations of inbreeding, showing significant inbreeding depression for low-intensity sigmoid [ r2 =8.6%, b = −0.131 (±0.034), T =−3.89, d.f.=161, P  < 0.001] and high-intensity sigmoid [ r2 =4.3%, b =−0.213 (±0.089), T =−2.39, d.f.=129, P =0.018]. The quadratic components are not significant for low-intensity sigmoid ( F1,160 =3.20; P is n.s) and high-intensity sigmoid ( F1,128 =0.941; P is n.s).

Discussion

Guppies from a wild population showed significant inbreeding depression for both male colour pattern and mating behaviour during three generations with inbreeding. Up to 25% of the variation in colour pattern over generations was explained by inbreeding, indicating a considerable load of deleterious mutations for secondary sexual traits. A previous study on guppy colour pattern demonstrated that male colour pattern showed inbreeding depression, although this analysis was based on a small sample with only a few lines per population (Sheridan & Pomiankowski, 1997). Our analysis on presence and absence of orange and black spots revealed a large number of DE for black [DE=4.89 (±0.75)] and orange [DE=2.31 (±0.80)]. These estimates are equivalent to (or even higher than) the estimates made for many primary fitness characters (e.g. growth rate, survival), which typically range between 0.1 and four lethal equivalents per gamete per trait (Lynch & Walsh, 1998; Oosterhout et al., 2000). Furthermore, the observed inbreeding depression for male courtship was also greater than what generally has been found for behavioural traits of other organisms, such as Drosophila melanogaster (Miller et al., 1993) and the butterfly Bicyclus anynana (Oosterhout, 2000; Oosterhout et al., 2000). Only male song frequency in Drosophila montana, a trait presumably closely related to fitness, showed comparative levels of inbreeding depression (see Aspi 2000). The severe loss in progeny fitness might explain why guppies have evolved outbreeding mechanisms, such as the preference of both males and females to court and mate with unfamiliar (possibly unrelated) individuals (see Hughes et al., 1999; Kelley et al., 1999).

The severe inbreeding depression observed in the sexual traits of male guppies is at first sight slightly surprising. Although male colour pattern and courtship behaviour are correlated with health and viability (Houde & Torio, 1992; Nicoletto, 1993) and important for female mate choice (Kodric-Brown, 1989; Grether, 2000), their genetic basis seems to leave little scope for severe effects of inbreeding. Guppy colour pattern is encoded by only a small number of genes compared with the hundreds to thousands of genes for a typical fitness trait (Lynch & Walsh, 1998), and most colour genes are located on the Y chromosome (Yamamoto, 1975). Subsequently, we will discuss the genetics of inbreeding depression and its consequences on the sexual traits of the guppy. We will then hypothesize why the guppy population of the lowlands in Trinidad may exhibit a considerable load of deleterious mutations.

The genetic basis of inbreeding depression

According to the partial dominance model of inbreeding depression (Charlesworth & Charlesworth, 1987), the loss in fitness associated with inbreeding is caused by the expression of deleterious recessive alleles in the homozygous state at loci. Many colour coding genes in guppies are located on or near the nonrecombining section of the Y chromosome (Winge, 1927, see Introduction), although X-linked colour pattern genes also contribute to the variation in colour pattern through occasional recombination with the Y chromosome (Winge, 1927; Haskins et al., 1961). Similarly, genes on the Y chromosome apparently mediate the inheritance of display behaviour and thrusting, although autosomal factors may also influence courtship activities (Farr, 1983). With a low recombination rate and Y-linked inheritance, the loss in colour pattern elements and inbreeding depression observed for courtship behaviour cannot merely be due to the expression of deleterious recessive homozygotes. The observed inbreeding depression suggests that either the genome of guppies consists of an additional number of (undescribed) colour pattern genes outside the nonrecombining area of the sexual chromosomes, or that epistatic gene interactions have resulted in inbreeding depression. Alternatively, inbreeding depression could have impaired the health or fitness of males, which may be reflected in the condition-dependent expression of colour spots and courtship behaviour.

Epistasis involving dominance of two or more autosomal genes interacting with additive effects of Y-linked colour or behavioural genes may have contributed to inbreeding depression observed in the current study. When recessive alleles are present at a low frequency on two or more autosomal loci, a trait is expected to show initially very little inbreeding effects (Crow & Kimura, 1970). Only when inbreeding continues will loci simultaneously become homozygous for rare recessive alleles such that the detrimental effects become noticeable, which may be augmented at higher levels of inbreeding. Chippindale & Rice (2001) found substantial epistatic fitness variation on the Y chromosome of Drosophila melanogaster, suggesting that such gene–gene interactions might not be uncommon and an important determinant for male fitness.

The extent to which epistasis may be involved in inbreeding depression is an unresolved issue (Wills, 1993; Lynch & Walsh, 1998). If epistasis augments inbreeding depression, a nonlinear relationship is expected between mean phenotype and inbreeding coefficient. To test for nonlinearity in our data set, commonly applied least-squares quadratic regressions were calculated, revealing that the relative surface area of black and orange spots showed a significant quadratic component. Logarithmic transformations can remove nonadditivity (Lynch & Walsh, 1998), but did not eliminate epistatic effects, suggesting that the observed decline in phenotypic values over generations was not an artefact of scale. Lynch & Walsh (1998) point out that nonindependence of data can render the interpretation of least-squares quadratic regressions difficult. For instance, as the analysed individuals are related, the data are not independent, violating one of the basic assumptions in regression theory (Sokal & Rohlf, 1995). We have tried to overcome this problem by employing a Wilcoxon's signed rank test (Sokal & Rohlf, 1995), and showed that within patrilines, the relative loss in colour pattern per increment in F was more severe from generation F2 to F3 than from F1 to F2. Whether this is evidence for epistatic effects remains uncertain. The F2 individuals may have experienced a particularly favourable environment that improved coloration and thus reduced the phenotypic effects of inbreeding. However, the inbred lines were set up such that the generations were largely overlapping, reducing common environment effects. Furthermore, spurious nonlinearity in change in phenotype over inbred generations can be the result of differential extinction of lines. However, in the current analysis this cannot account for the observed large reduction in colour pattern in the F3 as we have corrected for line loss by using only patrilines with males from all three generations.

Can we conclude that epistasis has contributed to inbreeding depression of these sexually selected traits? The quadratic decline in phenotypic values of traits with strong Y-linked inheritance suggests that interactions exist between their genes and/or the health or condition of the fish. It is possible that epistatic effects involving dominance of autosomal and/or X-linked loci, potentially interacting with additive effects of Y-linked genes, have contributed to the reduction in ornamentation with continued inbreeding. Unfortunately, the current experiment cannot distinguish between condition-dependence and epistasis, and further investigations are required to discriminate between these alternative hypotheses. Whatever the specific underlying cause of inbreeding depression, guppy colour pattern traits and courtship behaviour appear to reflect the genetic or health condition of males, and could thus be informative mate-cue characters for female choice as suggested by the ‘good genes’ model.

Mutation load for secondary sexual characters

Why do we find such a high mutation load for sexual characters in this lowland guppy population? Charlesworth (1998) suggested that synergistic epistasis might account for the mutation load observed in Drosophila melanogaster, which is much larger than predicted by mutation–selection balance. Although synergistic epistasis might have contributed to the inbreeding depression and mutation load observed here for a wild guppy population, there are several alternative explanations.

Evidence suggests that sexually selected traits in guppy populations can be under strong directional selection by female mate choice (Houde, 1997; Brooks & Endler, 2001a, b). Although these traits are not primary fitness characters, the selection and dominance coefficients of residing mutations will be very similar to those of fitness traits, because in large panmictic populations strong directional selection will purge all but the most recessive deleterious mutations (Crow & Kimura, 1970). In a selection-mutation balance, harmful alleles can only obtain an appreciable equilibrium frequency (say q > 0.01) when they are (almost) completely recessive (Crow & Kimura, 1970). These recessive mutations will contribute to dominance variation, and their phenotypic effects will be expressed during inbreeding.

The high mutation load could also be augmented by the reduced recombination rate of the sex chromosomes, which may enhance the accumulation of deleterious mutations through genetic hitch-hiking along with positively selected colour pattern and behavioural genes (Rice, 1987, 1994; Charlesworth & Charlesworth, 2000). Haskins et al. (1970) indeed found that some deleterious alleles on the Y chromosome are tightly linked with colour pattern genes, which could also explain the negative correlation between male ornamentation and offspring survival observed by Brooks (2000).

Furthermore, the Y chromosome is subjected to similar stochastic disadvantages as the genome of asexual organisms with respect to the accumulation of harmful mutations, a process known as Muller's ratchet (Muller, 1964; Charlesworth & Charlesworth, 2000). In the absence of crossing-over and back-mutations, no genotype (i.e. Y chromosome) can produce offspring with fewer mutations than its own load. In finite populations, chance events may cause the class of individuals with the lowest mutation load not to leave any offspring. When this happens, the ratchet ‘clicks’ and another deleterious mutation has been fixed in the population (Muller, 1964).

The mutation load for secondary sexual traits could also be elevated because such traits are phenotypically expressed in one sex only. Although females do not express the gene-controlled colour patterns as a result of a lack of the necessary male hormones (Hildemann, 1954), they do transmit a small number of colour genes to their sons. Because selection can only act on the expressed phenotype, deleterious mutations for male sexual traits may find a refuge in the genomes of females.

There are several valid hypotheses that could explain the high mutation load observed in this lowland guppy population. More importantly, however, this study shows that the mutation load present in natural populations could play an important role in sexual as well as natural selection and warrants further studies in mating-system evolution. The current findings indicate that colour pattern elements and courtship behaviour may provide informative mate-choice cues for females by reflecting the general condition or the genetic quality of males, an observation consistent with the ‘good genes’ model.

Acknowledgments

We thank David Weetman and two anonymous referees for critical comments on the paper. This work was supported by NERC grant (GST/02/2032), and NERC Fellowship of CvO (NER/I/S/2000/00885).

Received 27 July 2002;revised 12 September 2002;accepted 18 October 2002

Ancillary