The genetic architecture of male colour differences between a sympatric Lake Malawi cichlid species pair

Authors


  • Present address: N. J. Barson, Biodiversity and Ecological Processes Group, Cardiff University, Cardiff CF10 3TL, UK; M. E. Knight, Department of Biological Sciences, University of Plymouth, Plymouth PL4 8AA, UK.

G. F. Turner, Department of Biological Sciences, University of Hull, Hull, HU6 7RX, UK.
Tel.: +44 (0) 1482 466425; fax: +44 (0) 1482 465458; e-mail: g.f.turner@hull.ac.uk

Abstract

The genetic basis of traits involved in reproductive isolation is a key parameter in models of sympatric speciation by sexual selection, a potential mechanism driving the explosive radiation of East African cichlids. Analysis of hybrid crosses between two sympatric Lake Malawi cichlid species, representing the extremes of the extant colour distribution, generated Castle–Wright estimates of four to seven loci controlling colour differences. Segregation patterns deviated from a purely additive model with a significant contribution from dominance, and possibly also epistasis. Evidence was found for a strong influence of autosomal loci. As departures from simple additive variation could effect the operation of models of sympatric speciation, dominance and epistasis should not be neglected.

Introduction

The cichlids of the East African Great Lakes, Malawi, Victoria and Tanganyika have undergone explosive speciation leading to flocks of more than 500 species in Lakes Malawi and Victoria (Genner et al., 2004). One hypothesis for this extensive radiation is that these cichlid fishes have diversified rapidly through sympatric speciation driven by sexual selection either by female choice for male colour (Dominey, 1984; Turner, 1994), or male–male competition (Seehausen & Schluter, 2004). Empirical evidence supports the role of sexual selection on male colour pattern in these radiations (Seehausen et al., 1997; Seehausen & van Alphen, 1998; Seehausen et al., 1999a; Knight & Turner, 2004; Maan et al., 2004). Sympatric speciation in African cichlids has long been debated (e.g. Fryer & Iles, 1972; Van Doorn et al., 1998; Seehausen & van Alphen, 1999) and has some support from genetic studies of cichlid populations (Schliewen et al., 1994; Shaw et al., 2000).

Many closely related pairs of cichlid fish species differ clearly in male breeding colour, indicating that speciation may often involve divergence in male colour and associated female preferences (e.g. Seehausen et al., 1997). Sympatric congeners are more likely to have differing colour patterns than allopatric ones (Seehausen & van Alphen, 1999), prompting the suggestion that the same colour type can arise independently in different parts of the same lake within the same genera, a form of parallel evolution (Smith & Kornfield, 2002; Allender et al., 2003). The recurrence of similar colouration within and between genera in geographically isolated sections of the lake (Konings, 2001) raises the possibility that multiple divergences involve the same genes. Changes in the same gene have been implicated in the development of convergent melanic colour patterns in distantly related bird species (Hoekstra & Price, 2004; Mundy et al., 2004), but not in populations of rock pocket mice (Hoekstra & Nachman, 2003; Nachman et al., 2003).

Theoretical studies suggest that divergent mate choice could drive rapid sympatric speciation, but the conditions under which this can be achieved are restrictive (Arnegard & Kondrashov, 2004). In order to reduce the computational complexity, models of sympatric speciation tend to assume that simple additive genetics and a few loci control traits involved in sexual isolation (Gourbiere, 2004). Whether speciation is modelled as occurring in response to ecological selection and is reinforced by assortative mating (Dieckmann & Doebeli, 1999; Kondrashov & Kondrashov, 1999; Gourbiere, 2004), or in response to sexual selection alone (Turner & Burrows, 1995; Higashi et al., 1999; Arnegard & Kondrashov, 2004; Gourbiere, 2004), the probability of speciation decreases as the number of loci controlling assortative mating (trait or preference) increases.

Here, we investigated the genetic architecture of male colouration in a blue–yellow pair of sympatric, reproductively isolated Lake Malawi cichlid fish species, through the production of hybrid crosses. The number of loci estimated to control the differences in male colour is shown to fall within the range that would allow models of sympatric speciation by either ecological or pure sexual selection to operate. The inheritance of male colour is found not to conform to a simple additive genetic architecture as has been generally assumed in models of sympatric speciation.

Methods

Experimental animals

The study species were Pseudotropheus zebra (blue males; PZ) and the undescribed Pseudotropheus‘zebra gold’ (yellow males; PG). Adult individuals of both species were obtained from Nkhata Bay, Malawi. These fishes live in rocky habitats, which typically have high water transparency. Their depth distributions overlap, but significant differences in microsatellite allele frequencies indicate that there is minimal gene flow between them (van Oppen et al., 1998). Laboratory experiments indicate that the species are reproductively isolated by direct mate choice (Knight et al., 1998). F1 hybrids were generated by keeping females of one species in aquaria with males of the other species, but without access to conspecific males. F1 hybrids readily mated to produce F2 and backcross progeny (backcross to P. zebra; BZ and P.‘zebra gold’; BG), which were reared to maturity in the tropical aquarium system at the University of Hull.

Photography

Cichlid colour pattern can be influenced by motivational state, which is, in turn, influenced by social setting (Baerends & Baerends-van Roon, 1950). Thus, to get a good measure of the relative extent of colour elements, males were brought into a standardized dominant territorial motivational state prior to photography. Males were placed into individual compartments of a tank (125 × 40 × 40 cm), which was divided into four compartments (floor dimensions c. 31 × 25 cm) each containing a single male, with a single rear area (125 × 15 cm) into which were placed females of both parental species. All sections were divided by transparent partitions, with a 5-mm gap at the bottom of each partition to allow water to flow through the tank. The tank was filtered by an external filter, which removed water from the female compartment and returned it to the top of all of the male compartments allowing chemical, in addition to visual, communication. This arrangement allowed males to become territorially motivated by the sight and smell of conspecific males and females. Males were approximately matched for size with the other males being photographed at the time to minimize intimidation. Males failing to show territorial colouration were kept in the tank while the other males were changed. This invariably led to them gaining full territorial colouration.

Males were photographed on at least three different days using a digital camera (Olympus Camedia C-2500L; Olympus, London, UK). At least two photographs from each day were selected for colour analysis.

Ninety-five males were photographed: nine P.‘zebra gold’, 10 P. zebra, 11 F1 (from two broods of P. zebra♀ × P.‘zebra gold’♂), 25 F2 (from eight broods), 19 backcross to P. zebra (from two broods P. zebra♀ × F1 ♂) and 21 backcross to P.‘zebra gold’ (from four broods of P.‘zebra gold’♀ × F1 ♂; one brood of F1 ♀ × P.‘zebra gold’♂). The reciprocal cross was not generated in the F1 generation and only one direction of backcross was generated to P. zebra.

Image analysis

Photographs were cropped to remove the fins; only the body was then analysed. The area of yellow and blue colouration and the total area of the fish were measured using scion image 4.0.2b (Scion Corporation, Frederick, MD, USA) and the percentage area of yellow and blue calculated. A macro was written to measure the area of yellow and blue from the red, green, blue (RGB) stack (which splits the image into three layers representing the primary colours of light, red, green and blue). To calculate the area of yellow, the red and green images were added together and normalized and then the inverse of the blue stack was added to the normalized red + green image and then this was normalized. Estimates of the area of yellow were made on a 0–110 density slice of the final image. Thus, for each pixel, Pixel = yellow if x = 0–110, Pixel ≠ yellow if x > 110 where: x = {[(RED + GREEN) 0.5] + [255−BLUE]} 0.5 and ‘RED’, ‘GREEN’ and ‘BLUE’ refer to the RGB stacks. This estimation was felt to be appropriate, as there were no red or green colour elements and the area selected matched what the observer perceived as yellow in the photographs. The area of blue was estimated directly from a 0–127 density slice of the blue stack. The total area of the fish was estimated by density slicing the red stack between 0 and 255 and measuring the area.

Estimating of the number of colour loci

The Castle–Wright method estimates the effective number of loci (nE), or the minimum number of normally segregating loci that would be required to explain the observed phenotypic means and variances of parental, F1, F2 and backcrossed generations (Castle, 1921; Wright, 1952). This method assumes that one line is fixed for alleles which increase a character and the other for alleles which decrease it. Lande (1981) generalized the Castle–Wright method for use with wild populations where the character had been subject to sustained natural or sexual selection.

The minimum number of loci is estimated by using the formula:

image

where, μp1 is the mean of the ith parental line, var[μp1] is the variance of the mean of the ith parental line, and inline image is the segregating genetic variance. The second term in the numerator is a correction for sampling error added by Cockerham (1986). The weighted least squares method of Cockerham (1986) was used to estimate inline image. This method combines information from all line crosses into a single estimate and so achieves greater power. The variance of the estimate is calculated by using the formula:

image

Var (inline image) was estimated through the least-squares analysis.

The Castle–Wright estimator relies on three assumptions about the underlying genetics of the trait under examination: there is additive genetic variation, the loci are unlinked and all loci have equal effects. Zeng (1992) suggested a correction to the estimator that takes into account linkage and loci of unequal effect:

image

where inline image is the recombination index and C is the squared coefficient of variation of effects and ne is estimated by using the formula of Cockerham (1986). The associated variance of this equation is:

image

The recombination index can be estimated from the haploid number of chromosomes (M) as inline image (Lynch & Walsh, 1998). Cα is not known, so a range of corrections were made to represent the effect of different underlying distributions of allelic effects.

Dominance and epistasis

The Castle–Wright method assumes additive allelic effects. The influence of composite effects was tested by using the joint-scaling method (Lynch & Walsh, 1998). The joint scaling test uses a weighted least squares approach to fit the multiple regression, zi = μ + Mi2α + Mi3δ + ɛi, to the line cross means. Where zi is the trait mean in the ith line, μ is the mean of all line means, α is the additive genetic effect, δ is the dominance effect and ɛi is the sampling error associated with the ith line and where epistasis is not included in the model this is implicitly included in the error term. If sufficient line crosses are available, then the effect of epistasis can be added to the model. Unless the effect of dominance can be discounted, this requires more than six lines as there are six parameters to estimate (the six lines P1, P2, F1, F2, B1, B2 and for example F3). Thus, the effect of epistasis was not estimated separately to the model here and is contained in the error term.

Linkage in the expression of yellow colouration

The influence of linkage was assessed by looking at the co-inheritance of yellow colouration in different parts of the body in males from the backcross to P. zebra (BZ). The intensity of yellow colour in the dorsal fin, gular, belly, chest and flank was scored as being between 0 and 3, where 0 represented no yellow and 3 very intense yellow.

Sex linkage of colour loci

The influence of the sex chromosomes relative to autosomal loci was assessed by comparing the phenotypes of the F1 males to males from the backcross to P. zebra. Although the sex determining mechanism of teleost fishes (Orzack et al., 1980; Francis, 1992; Koehler et al., 1995) and of cichlids is known to be labile (Hammerman & Avtalion, 1979; Baroiller et al., 1996), evidence from laboratory breeding experiments (Knight, 1999; Seehausen et al., 1999b) and theoretical expectations (Seehausen et al., 1999b; Lande et al., 2001) have been used to infer the sex determining mechanism in Pseudotropheus (Maylandia) as XX/XY (Seehausen, pers. comm.). As all F1 crosses were in the direction P. zebra female × P.‘zebra gold’ male, all F1 males have a P. zebra X chromosome and a P.‘zebra gold’ Y chromosome. As all backcrosses to P. zebra were to P. zebra females, the origin of the sex chromosomes was the same as in the F1 cross. Thus, the difference in the percentage of the body that is yellow/blue between the F1 and backcross generations must be a result of the increase in the proportion of P. zebra autosomal genes. A one-way anova was carried out to test for a difference in the mean percentage area of yellow and blue in F1 and backcross to P. zebra males. The number of males that exhibited yellow on three body regions, the chest, belly and dorsal fin, was also compared between the two crosses.

Results

Hybrid phenotypes

F1 males all had yellow chests and bellies and predominantly yellow dorsal fins (Fig. 1). The body varied from yellowish to blue. The yellower F1 males were more iridescent than P.‘zebra gold’ males.

Figure 1.

 Colour patterns of parental and hybrid males.

The distribution range between parental phenotypes was well covered by F2 males (Figs 1 and 2). The P.‘zebra gold’ phenotype was fully recovered with three individuals overlapping with the P.‘zebra gold’ distribution. The P. zebra phenotype was never fully recovered. Although two F2 males showed percentages of blue typical of P. zebra, while P. zebra males never had any yellow on the body, the F2 males always did to some extent.

Figure 2.

 Distribution of percentages of blue and yellow on bodies of Pseudotropheus zebra (n = 10, filled triangles), P.‘zebra gold’ (n = 9, open diamonds), and F2 males (n = 25, grey squares).

Backcrosses to P. zebra sometimes fully recovered the P. zebra phenotype but many individuals retained some yellow in the chest (five out of 19) and belly (13/19) and some yellow in the dorsal fin (12/19) (Fig. 1, Table 1). While the flanks were predominantly blue, some yellow was present in nine out of 19 individuals. The amount of yellow in all of these regions of the body seemed to vary continuously.

Table 1.   Distribution of yellow body colouration in backcross to Pseudotropheus zebra males. Extent of yellow is graded by eye from 3, entirely yellow, to 0, no yellow visible.
Male numberYellow chestYellow bellyYellow gularYellow flankYellow dorsal fin
  1. Total number of individuals with any yellow in that region of the body. Yellow in the trailing edge was ignored as this occurs in P. zebra.

7933312
8300000
8400000
850–1200–10–1
8622201
8700000
8800000
8900000
9000000
11901001
1200200–11
1210200–11
1220200–10–1
12302010–1
1240200–11
12901001
1410–110–10–12
14201000
1430–11011
Total5133912

Most fish from the backcross to P.‘zebra gold’ had a short submarginal blue to black band extending from the anterior of the dorsal fin and pale blue/white lappets (Fig. 1). These features were also present in F1 and some F2 males, but not in wild-type P.‘zebra gold’.

Estimated minimum number of colour loci

The data for percentage of blue was approximately normally distributed, as assumed by the Castle–Wright estimator, but the percentage yellow data had to be arcsine transformed before most lines conformed to normality. Plots of the mean percent blue (raw data) and yellow (arcsine transformed) against the variance (Fig. 3b,c) conform reasonably well to the assumptions of the Castle–Wright–Lande estimate (see Fig. 3a) and were used in the subsequent calculations. The F1 variances of both percent of the body blue and percent yellow were higher than expected. These high F1 variances may indicate environmental effects on male colour (Maan et al., 2006). Both the F1 and F2 means were displaced towards P.‘zebra gold’.

Figure 3.

 (a) Theoretical distribution of the means and variances of a quantitative trait for different lines on a transformed scale of measurement. This distribution is assumed by the Castle–Wright–Lande estimate. (b and c) Plots of mean percent of body blue (b) and yellow (c) against its variance (blue data are untransformed, yellow data have been arcsine transformed). BG is the backcross to Pseudotropheus ‘zebra gold’ line, BZ is the backcross to P. zebra line, PZ is the wild parental P. zebra population, and PG is the wild parental P. ‘zebra gold’ population.

The Castle–Wright estimates indicate that a minimum of three loci control the area of blue (ne = 2.56 SE 1.6) and four the area of yellow (ne = 4.3 SE 1.6). Correcting for linkage using the recombination index and leaving allelic effects equal shows the estimate to be almost unbiased (Cα = 0 ne blue = 2.77 SE 1.42; ne yellow = 5.05 SE 2.37). The estimate is not considerably elevated if a normal distribution of allelic effects is assumed (Cα = 0.25 ne blue = 3.22 SE 1.77; ne yellow = 6.07 SE 2.96). The number of loci increases approximately two-fold if a negative exponential distribution is assumed (Cα = 1 ne blue = 4.55 SE 2.84; ne yellow = 9.10 SE 4.73), increasing to four- to five-fold if a more leptokurtic, L-shaped distribution is assumed (Cα = 4 ne blue = 9.87 SE 7.1; ne yellow = 21.26 SE 11.83). However, as the distribution becomes more leptokurtic, the effect of the additional loci is very small compared with the major loci and this situation may therefore reduce to a major gene system in behaviour.

Dominance and epistasis

Joint-scaling showed that neither the yellow nor blue areas were adequately described by a simple additive model (Table 2). The predicted means of the additive model were significantly different to the observed means (yellow area: inline image = 15.12, P < 0.01; blue area: inline image = 21.4, P < 0.001). The fit was improved by adding dominance into the model, as expected (Lynch & Walsh, 1998). However, the fit was still poor with significant differences between the observed and predicted means (yellow area: inline image = 12.79, P < 0.01; blue area: inline image = 16.75, P < 0.001). As adding dominance always improves the fit (Lynch & Walsh, 1998), it is necessary to test the significance of the improvement by using a likelihood ratio test. For the area of blue, the improvement of the model including dominance is not significant (Λ1 = 2.32; P = 0.128). However, adding dominance significantly improved the fit for the area of yellow (Λ1 = 4.66; P < 0.05).

Table 2.   Joint-scaling test for the percentage of body area (a) blue, raw data and (b) yellow, arcsine transformed data.
LinePGPZF1F2BGBZ
  1. Sample means and means predicted by an additive and additive + dominance model are shown with their SE. The estimated parameters for the modelled mean (μ0), modelled additive component (αc) and modelled dominance component (δc) are also shown.

(a)
Sample mean (%blue)5.0160.0225.2326.8225.6141.75
SE1.532.173.452.892.782.43
Mean: additive model5.2357.2331.2331.2318.2344.23
SE1.371.70.960.961.021.24
Parameter estimates: μ0 = 31.23 SE 0.958; αc = −26 SE 1.212
Mean: additive + dominance model6.0858.927.8530.1716.9743.38
SE1.482.022.411.811.311.36
Parameter estimates: μ0 = 30.173 SE 1.183; αc = −26.408 SE 1.241; δc = −2.318 SE1.521.
(b)
Sample mean (%yellow)68.525.0747.5543.7751.7716.49
SE1.460.483.232.481.892.57
Mean: additive model69.545.1537.3537.3553.4421.25
SE1.20.470.630.630.90.45
Parameter estimates: μ0 = 37.35 SE 0.626; αc = 32.35 SE 0.662
Mean: additive + dominance model68.014.9941.6639.0854.8323.33
SE1.390.482.091.021.11.06
Parameter estimates: μ0 = 39.08 SE 1.018; αc = 31.51 SE 0.734; δc = 2.58 SE 1.193.

Linkage and epistasis in the expression of yellow colouration

The strength of expression of yellow on the throat and chest seemed closely related but the belly could be quite yellow without either throat or chest being yellow in BZ males (Table 1). The most common regions of the body to develop yellow pigmentation were the belly and dorsal fin: it appears that that there is some linkage in the development of yellow pigmentation in these two body regions. There were no occasions, however, when any other region of the body was yellow if the belly was not also yellow.

Sex linkage of colour loci

The difference in the mean percentage area of yellow and blue between the F1 and backcross to P. zebra crosses was highly significant (one-way anova yellow, F1 = 54.23%, Bz = 10.17%F1,28 = 76.69 P < 0.001; blue, F1 = 25.23%; Bz = 41.75%F1,28 = 15.94 P < 0.001). All F1 males had a yellow chest and belly and predominantly yellow dorsal fin (n = 11). In the backcross to P. zebra, 26.3% had some yellow on the chest, 68.4% had yellow bellies and 63.2% had yellow in the dorsal (n = 19), although the intensity and extent was much reduced compared with the F1 hybrids (Fig. 1). As the origin of the sex chromosomes are assumed to be the same in these two crosses, this significant difference suggests autosomal location of the colour loci.

Discussion

Our results indicate that the difference in body colour between P. zebra and P. ‘zebra gold’ is determined by a minimum of four to seven loci. Only the body was analysed so there may be additional loci controlling fin colour. It was not possible from these data to ascertain whether the genes that determine the area of yellow (nE = 4) and blue (nE = 3) are the same.

Although it is possible that the actual number of loci is higher than the Castle–Wright estimate, equally there may have been genetic changes in the time since speciation occurred and this may have increased the number of loci involved in coding for the colour differences. An estimate for the divergence time between the study species is not currently available owing to problems with shared genetic variation within the whole Lake Malawi species flock (Moran & Kornfield, 1993; Kornfield & Parker, 1997; Parker & Kornfield, 1997). This shared variation has been attributed to very recent divergence (Kornfield & Smith, 2000). This is supported by recent divergence estimates (1000–17 700 years) made using sequence data combined with an adjacent short tandem repeat to create HapSTRs (Hey et al., 2004) for three members of the closely related genus Tropheops (Won et al., 2005).

The estimated number of loci controlling the colour difference between P. zebra and P. ‘zebra gold’ is under half the haploid number of chromosomes (n = 22) and so it is unlikely that our estimate is limited by the recombination index, which is normally a few times this (Lande, 1981), as some previous studies of sexual isolation have been (Templeton, 1977Drosophila; Shaw, 1996 Hawaiian crickets). This may also suggest that not all chromosomes are involved in the control of these colour differences.

The co-inheritance of yellow on the belly, chest and dorsal fin, suggests that their loci are linked. Linkage has been found between loci controlling a predominantly female melanic colour pattern in cichlids from Lake Malawi (Streelman et al., 2003).

Geographic races of P. zebra recently shown to be at an incipient stage of speciation (Knight & Turner, 2004) differ in whether the dorsal fin and chest/throat are blue or yellow. The pattern of expression in the BZ line suggests that introgression of these colour elements may occur more readily than differences in background body colour.

Neither the area of blue nor the area of yellow observed in the hybrid individuals conformed to the simple additive model assumed by the Castle–Wright method. Joint-scaling suggests that dominance influences the expression of the yellow colouration and that epistasis may also be involved.

Arnegard & Kondrashov (2004) suggest that male display traits often have additive architectures, but the result of this study and those of other previous studies on closely related Hawaiian crickets (Shaw & Parson, 2002) and birds (Mundy et al., 2004) suggest that dominance is frequently involved in the determination of traits involved in sexual isolation. Models of sympatric speciation typically assume that sexually selected traits are controlled by additive loci, and the influence of dominance on the probability of speciation remains little explored.

There were highly significant differences in the mean percentage area of both yellow and blue between F1 and backcross males suggesting that autosomal genes have a large influence on the percentage of yellow pigmentation. An autosomal location of male display trait loci is assumed in some models of sympatric speciation by sexual selection (e.g. Arnegard & Kondrashov, 2004).

The estimate of four to seven loci determining the colour differences between this sympatric pair is at the higher end of values that would permit sympatric speciation under the conditions of several models (Dieckmann & Doebeli, 1999; Kondrashov & Kondrashov, 1999; Arnegard & Kondrashov, 2004; Gourbiere, 2004). With this number of courtship trait loci, some models indicate that speciation is possible only if there is a large ecological fitness differential between extremes (Kondrashov & Kondrashov, 1999), or if ecological differentiation between species is caused by a small number of loci (Dieckmann & Doebeli, 1999). The latter may not be unrealistic, as Albertson et al. (2003) estimated that species differences in some elements of cichlid jaw structure likely to have a major role in dietary differentiation, such as the shape of the teeth and dentary, may be controlled by two or fewer loci. Other putatively adaptive head shape differences were controlled by 4 to 11 loci. The models discussed above dealt with cases of simultaneous divergence in sexual and ecological traits. Several models have analysed the probability of sympatric speciation driven by sexual selection alone. Turner & Burrows (1995) demonstrated this was possible with four loci controlling the male colour, Arnegard & Kondrashov (2004) found it was only likely with two to four loci and Gourbiere (2004) found the probability of speciation dropped exponentially above five loci. Higashi et al.’s model (1999) was insensitive to the number of loci controlling the male display trait, but sympatric speciation in their model may have been facilitated by the assumption that the initial trait variation followed a symmetrical distribution (Arnegard & Kondrashov, 2004).

Haesler & Seehausen (2005) estimated that one to four unlinked loci controlled the differences in female preference in a sympatric sibling species pair of cichlid fish from Lake Victoria. From these estimates, it appears that differences in male trait and female preference can indeed be explained by relatively few loci.

Conclusion

The minimum number of loci controlling the colour differences between P. zebra and P. ‘zebra gold’ is estimated to be from four to seven. This estimate is towards the limit of the number of genes that would make several models of sympatric speciation realistic. There is evidence that dominance and probably also epistasis are involved in the genetic determination of these colour differences.

Acknowledgments

This work was funded by the Natural Environment Research Council (UK) NER/A/S/1998/00011. We thank Andy Gould, Rosanna Robinson, Alan Smith and Rachel Veale for help in the aquarium, Ole Seehausen for numerous helpful suggestions throughout the work and for the loan of his digital camera.

Ancillary