QTL LINKAGE MAPPING OF ZEBRA FINCH BEAK COLOR SHOWS AN OLIGOGENIC CONTROL OF A SEXUALLY SELECTED TRAIT

Authors


Abstract

Mate choice based on sexual ornaments can impose strong selection, which raises the question of how genetic variation in ornaments is maintained. One mechanism that has been proposed is genic capture. If ornament expression is influenced by general condition and condition is under polygenic control, selection will be inefficient in removing genetic variation. Here we analyze whether the genetic architecture of beak color in a population of zebra finches supports this hypothesis. Zebra finch beak color is commonly assumed to be under strong selection by mate choice, although some of the evidence is ambiguous. We show that beak redness has a heritability of 34% in our population and that it is strongly genetically correlated between the sexes, suggesting that it is largely controlled by the same genes in males and females. We mapped variation in beak redness based on 1404 single-nucleotide polymorphism (SNP) markers genotyped in a large pedigree. We find evidence for linkage on four chromosomes (Tgu1, Tgu5, Tgu13, Tgu21), which together explain a large part of the additive genetic variance. Our finding of genomic regions with major additive effects is not consistent with directional selection and genic capture, but rather suggests a role of antagonistic pleiotropy in maintaining genetic variation.

Mate choice can impose strong directional selection on target traits, which is expected to erode genetic variation in sexually selected traits. A loss of genetic variation will in turn erode the genetic benefits of choosing mates based on ornament expression (“lek paradox,”Borgia 1979). Contrary to this prediction, however, sexually selected traits often show more genetic variation than other traits (Pomiankowski and Møller 1995). Genic capture has been suggested as one solution to this paradox (Rowe and Houle 1996; Tomkins et al. 2004). According to genic capture hypothesis, ornaments could “capture” and display overall genetic quality if the expression of sexually selected ornaments is condition dependent and condition is (at least partly) under genetic control. Genetic variation can then be maintained by mutation–selection balance because overall genetic quality is likely to be influenced by a multitude of genes of small effect and thus offers a large mutational target (Merilä and Sheldon 1999). Hence, a critical prediction of the genic capture hypothesis is that the sexual ornament is under highly polygenic control (Tomkins et al. 2004).

Besides the many genes that might affect ornaments through condition dependence, every specific sexual ornament may also be influenced by genes that are more directly involved in ornament production. Mutations in such genes will tend to have a large impact on the level of ornamentation. Such mutations would undermine the value of the ornament to honestly signal genetic quality (condition) if they affect trait values in a way that is unrelated to overall condition. Mutations that increase ornamentation without imposing a cost will be driven to fixation by sexual selection, whereas mutations that reduce ornamentation will rapidly be eliminated. Under persistent directional selection, polymorphisms at these loci might be retained in the gene pool only if these major genes have antagonistic pleiotropic effects (antagonistic pleiotropy).

Although many quantitative genetic studies have shown that there is a genetic basis for prominent, sexually selected traits (e.g., Price and Burley 1993; Pomiankowski and Møller 1995; Price 1996; Birkhead et al. 2006), there is limited information on the distribution of number and location of loci that are involved in the production of sexual ornaments (Chenoweth and McGuigan 2010). Only recently has it become feasible to map genetic regions that show standing genetic variation for fitness-related traits and traits involved in intrasexual competition (Beraldi et al. 2007; Chan et al. 2010; Johnston et al. 2010). There are only a few studies that have analyzed the genetic basis of sexually selected traits that have evolved by mate choice (Johns et al. 2005; Wright et al. 2008; Kitano et al. 2009; Tripathi et al. 2009; Magalhaes and Seehausen 2010). These studies seem to indicate that sexually selected traits are sometimes under oligogenic control, that is, trait expression is influenced by a few genes with relatively strong effects.

Here we study the genetic basis of zebra finch beak color, a sexually dimorphic ornament that is presumed to be involved in mate choice and for which genic capture has been suggested as the causal mechanism that maintains genetic variation (Birkhead et al. 2006). Some studies have shown that males with dark red beaks were more attractive to females than males with orange-red beaks (Burley and Coopersmith 1987; Price and Burley 1994; Blount et al. 2003), but in other populations (including ours) such a preference has not been found (Collins et al. 1994; Collins and ten Cate 1996; Forstmeier and Birkhead 2004), and is thus at least not as strong or the preference is context dependent (Simons and Verhulst 2011). The pronounced sexual dimorphism in beak color clearly suggests that the trait has been subject to sexual selection, even if the current evidence is somewhat ambiguous. Several studies have shown that beak coloration is expressed in a condition-dependent manner (Blount et al. 2003; Alonso-Alvarez et al. 2004; Birkhead et al. 2006; Gautier et al. 2008; Bolund et al. 2010a,b), which adds to the plausibility of the genic capture scenario.

Zebra finch beak color is largely caused by carotenoids that cannot by synthesized de novo by birds, but have to be incorporated through the diet (McGraw and Ardia 2003; McGraw et al. 2003; Alonso-Alvarez et al. 2004; McGraw and Toomey 2010). Birds that were fed the carotenoids lutein and zeaxanthin had more reddish beaks, showed an enhanced cell-mediated immune response, and were more attractive to females (Blount et al. 2003). This is in line with the idea that the use of carotenoids has to be traded-off between immune function and sexual signaling (Olson and Owens 1998; McGraw and Ardia 2003; Chew and Park 2004; Bertrand et al. 2006), which would be a mechanism that induces condition dependence.

Despite much interest in carotenoid-based ornaments in animals (e.g., Olson and Owens 2005), relatively little is known about genetic pathways involved in the uptake, conversion, and incorporation of carotenoids in the tissue of vertebrates. Walsh et al. (2011) have compiled 11 candidate genes based on work in a variety of animals and have shown that most of them were expressed in the carotenoid-rich plumages and beaks of a passerine bird (red-billed quelea, Quelea quelea). The candidate genes include ninaB and ninaD, scavenger receptor class B genes that mediate carotenoid uptake in Drosophila (von Lintig et al. 2001; Kiefer et al. 2002), BCDO2 that is involved in producing carotenoid-rich yellowish bare parts in chicken (Eriksson et al. 2008), BCO2 that is associated with a rare phenotype of carotenoid-enriched tissue in sheep (Våge and Boman 2010), and BCMO1 that shows genetic polymorphisms associated with levels of circulating carotenoids in humans (Ferrucci et al. 2009). It is currently unclear if any of these loci are also involved in producing carotenoid-based ornaments in zebra finches, although they have the potential to constitute major genes for zebra finch beak color.

Our aim in this article is to clarify the genetic architecture of zebra finch beak color. We therefore present a quantitative genetic analysis, followed by a marker-based linkage analysis for identifying genomic regions that contribute to the heritability of zebra finch beak color. Specifically, we tested a critical prediction of the genic capture hypothesis. Under the genic capture model, we expect that variation in beak color is associated with a large number of genes that contribute to condition. Under this hypothesis, correlations between individual polymorphic loci and variation in the trait are expected to be very weak, because the genetic variance is spread across many loci. Hence, we would not expect to find significant associations between a locus and the trait, despite substantial heritability. In contrast, under the oligogenic model we expect to find few genomic regions that are strongly associated with beak color. We would further predict that the identified genomic regions contain genes involved in carotenoid metabolism, although limited knowledge about the causal pathways might limit the opportunity to pinpoint them.

Materials and Methods

STUDY POPULATION AND PEDIGREE

Our study population is derived from a large and genetically diverse stock of zebra finches that is held at Seewiesen, Germany (referred to as Seewiesen-GB in Forstmeier et al. 2007b). The population was derived from the population in Sheffield, United Kingdom, that was used by Birkhead et al. (2006) to study genic capture. Our linkage analysis is based on genotypes and phenotypes of 1019 individuals from four consecutive generations (to which we refer as P, F1, F2, and F3). Parentage was ascertained for all individuals by genotyping all offspring and their potential parents using 10 highly variable microsatellite markers (Forstmeier et al. 2007a). No misassignments were subsequently discovered when genotyping the population for single-nucleotide polymorphisms (SNPs). The genetic parents of the P generation individuals were also known (they were all bred in cages containing a single breeding pair) and the full pedigree information covers five generations (1378 individuals). Most of the individuals of the P generation and all of the F1 and F2 generations (84% of all phenotyped individuals) had been cross-fostered to foster parents at the egg stage to separate genetic and early-rearing effects.

PHENOTYPIC DATA

We measured beak color by photo-spectrometry (for details see Bolund et al. 2007). In short, we used a hand-held spectrometer with a deuterium-halogen light source that was placed perpendicular to the beak surface. Reflectance was recorded every 0.273 nm in the range of 260–820 nm, which includes the human-visible spectrum as well as reflectance in the UV (visible to birds, Bennett et al. 1996). We averaged reflectance spectra across five measurements from five locations on the upper mandible.

Birds were measured on several occasions (sessions) at an average age of 369 ± 346 days (mean ± SD, range: 87–1706 days). All measurements were taken when birds were in nonbreeding condition (without nest boxes, none of the birds was involved in incubation or chick rearing). On average, individuals had 1.9 measuring sessions (401 individuals with one, 375 with two, 230 with three, and 15 with four sessions). On 640 occasions, individuals were measured when they were less than 120 days (about 17 weeks). At this age, beak color still changes slightly and approximately linearly with age (Burley and Coopersmith 1987; also see “Results”). Therefore, we included age as a covariate in our analyses, whereby all individuals older than 120 days were given the age value 120 because beak color did not change systematically beyond this age (Burley and Coopersmith 1987; also see “Results”). Few measurements were taken between days 120 and 135 (15 measurements, 1.7% of total), hence shifting the threshold by two weeks does not influence the results. Of the individuals measured at ages below 120 days, 386 were measured again later. These repeated measurements facilitated the separation of age effects from subject effects.

From the resulting spectrometric data, we extracted three components that summarize most of the relevant variation (see Bolund et al. 2007). These are (1) the position of the peak in the UV part of the spectrum (<470 nm, “nmUVmax”), (2) the point where the curve reaches half of the difference between the minimum value of the curve and the peak in the UV (nmUVhalf), and (3) the point where the curve reaches half of the difference between the minimum value of the curve and the peak in the red part of the spectrum (>600 nm, nmRedhalf). These three wavelengths characterize the “trough” shape of the curve with males having a wider trough than females (i.e., lower nm-UV-half and higher nm-Red-half). Furthermore, we used the reflectance at three characteristic points of the spectrum (residuals after controlling for overall brightness) that were maximally different between males and females (358 nm “res358,” 570 nm “res570,” and 707 nm “res707,” see Bolund et al. 2007). Spectrometric measurements reflect human-visible beak redness as was shown by a strong correlation with color chart scores of beak color (Bolund et al. 2007).

We performed a linear discriminant analysis (LDA) on the six variables derived from the spectrometric data (see above). We included only fully adult birds (≥120 days of age) in the discriminant analysis to capture sexual dimorphism in adults, but the coefficients were applied to all measurements (Table 1). The resultant main axis reflects sexual dimorphism in beak color and thus presumably the direction of (past and/or current) sexual selection. Alternatively, we could have used a principle component analysis (PCA) to summarize contemporary variation in beak color. A PCA on the pooled dataset of males and females will also be dominated by the differences between the sexes and is thus similar to a LDA. Discriminant scores were highly correlated with each of the six original traits (Table 1) and the correlation was strongest with traits that showed higher heritabilities (correlation between the heritability of a trait and its correlation with discriminant scores: r = 0.95). Furthermore, discriminant scores correlated highly with the first principal component both when the principal component analysis was applied to data from both sexes (r = 0.95), and when applied to the sexes separately (males: r = 0.86, females: r = 0.86). This shows that discriminant scores (mean: 0.35, SD: 1.71) also reflected the main axis of variation within each sex. In the following, we will refer to discriminant scores as “beak redness” for brevity.

Table 1.  Coefficients of a linear discriminant analysis on six traits measured from spectrometric reflection curves of zebra finch beak color, correlations of discriminant scores with the original traits and the heritability of spectral characteristics.
 Discriminant function coefficient (β)Correlation with discriminant scores (r)Heritability (h2)
nmUVmax−0.01860.690.157
nmUVhalf 0.00170.760.162
nmRedhalf−0.12070.980.338
res358−0.03480.770.227
res570 0.08200.870.274
res707−0.07230.760.198

GENETIC CORRELATIONS

We used the MCMCglmm function from the MCMCglmm package in R 2.11.1 (Hadfield 2010) to fit animal models in a Bayesian framework. Fitting male and female beak color scores as two separate traits allowed us to estimate genetic correlations between the sexes. Models are based on the same data as described above and included permanent environment effects (individual identity not linked to the pedigree), maternal effects (mother identity), shared early-rearing effects (foster brood identity), and additive genetic relatedness (inferred from pedigree data) as random effects, and inbreeding coefficient (based on our five-generation pedigree) and age as fixed effects. For the variance–covariance matrices, we used the following priors that are uninformative for the genetic correlation (see documentation for the MCMCglmm package, http://cran.r-project.org/web/packages/MCMCglmm/vignettes/CourseNotes.pdf): 0.002 for all variances, 0 for all covariances and 3 as a degree of believe parameter. We ran two MCMC chains for 160,000 iterations with a burn-in phase of 60,000 iterations and a thinning interval of 100. We check convergence using the Gelman–Rubin criterions (Gelman and Rubin 1992), implemented in the coda package in R (Plummer et al. 2010). The critical genetic correlation reached an R score of 1.07 (97.5% percentile: 1.10).

SNP GENOTYPING

We genotyped all individuals for 1920 SNPs using an Illumina GoldenGate Assay (Fan et al. 2003) at the SNP Technology Platform of Uppsala University, Sweden. Details of SNP identification, call rates and quality control have been described previously (Backström et al. 2010). The analysis was based on 1396 SNPs that were polymorphic in our population and that have been used to construct a linkage map of the zebra finch genome (32 linkage groups, total length of 1479 cM, Backström et al. 2010). The linkage map covered c. 92% of the currently annotated zebra finch genome. Marker spacing varied substantially across the genetic map with an average density of 1.1 ± 3.4 cM between adjacent markers, but markers were denser (in terms of genetic distances) in the central part of the chromosomes, whereas marker spacing was much wider in the telomeric regions. This is an effect of a heterogeneous recombination landscape in the zebra finch genome, with highly elevated recombination rates toward chromosome ends (Backström et al. 2010).

QTL MODELS

We used the linkage-disequilibrium linkage-analysis (LDLA) tool of the GridQTL cluster (Hernández-Sánchez et al. 2009) in our quantitative trait locus (QTL) analysis. The LDLA tool implements variance component-based QTL analysis by fitting mixed-effect models using restricted maximum likelihood (REML) estimation (George et al. 2000). Additive effects were estimated from identity-by-decent (IBD) probabilities from flanking markers. In an initial scan, we estimated QTL effects at 5-cM intervals across the full genetic map. Subsequently, we refined this to 1-cM intervals on chromosomes that showed evidence of linkage in the initial scan.

The LDLA module offers an option to estimate IBD probabilities of alleles in the founder population based on the allele frequencies among founders and population genetic parameters that have to be provided. This option is supposed to increase the power of the analysis and is not supposed to introduce bias, unless the parameters provided are far off their true values (Hernández-Sánchez et al. 2009). However, we observed susceptibility to rare alleles and to details of the population settings (own unpublished data) and therefore prevented the estimation of historical IBD probabilities by setting the number of generations since the establishment of the population to zero. Hence, our analysis is solely based on information from meioses in the pedigree.

Each model controlled for sex, age, and inbreeding coefficient as covariates. Furthermore, we included a categorical fixed factor (“session,” 24 levels, all sessions were sex specific) that controls for differences in housing conditions prior to measuring, seasonal, and cohort effects and differences between measuring devices. Conditional on these fixed effects, we estimated variance components for (1) genome-wide additive genetic effects based on the relatedness in the pedigree, (2) maternal effects (identity of the mother), (3) shared early-environment effects (identity of the nest of rearing), and (4) additive IBD effects at a particular locus. When estimating effects of genetic variation on the Z chromosome, we also included a random effect controlling for genetic variation on the W chromosome modeled as the identity of the founder mother in the maternal line of descent within our pedigree. Because the estimated W effects are partly confounded with the maternal effect, we did not include them in the analysis of autosomal QTLs.

After scanning for additive QTL effects, we also searched for dominance QTL effects. We did so by fitting the same additive QTL models augmented with a matrix that contains the probability estimate for sharing both alleles at the locus under consideration. Unlike additive QTL effects, dominance QTL interactions are only shared between two individuals if they share both alleles at a locus. Fitting a dominance-sharing probability matrix together with the additive IBD matrix allows estimating whether sharing both alleles explains phenotypic similarity beyond what is expected from the additive effect of the two alleles.

SIGNIFICANCE TESTING AND CONFIDENCE INTERVALS

We present uncertainty estimates for variance components based on the results from GridQTL that used the underlying ASReml software for estimating variance components and their standard errors (estimate ± SE). All fixed and random effect estimates that we present (with the exception of the genetic correlation, which was estimated by MCMCglmm) are based on REML estimation using GridQTL.

We used likelihood ratio tests (LRT) to test for the significance of QTL effects. We did so by testing the likelihood ratio (LR =−2 × (ln(LH0) − log(LH1)), where LH1 is the likelihood of the full QTL model and LH0 is the likelihood of the (null) model fitted without the QTL effect) against a 50:50 mixture distribution of point mass at zero and a χ21 distribution (Slate 2005). The significance of dominance was also tested using LRT, comparing the likelihood of models including dominance and additive QTL effects with a model including only additive QTL effects (other random and fixed effects were included in all models).

Because we were testing for many potential QTLs across the genome (every 5 cM), we used adjustments of the significance level to account for multiple testing. It was not computationally feasible to obtain empirical P-value distributions. Therefore, we used an approximation to achieve a genome-wide significance threshold (Lander and Kruglyak 1995). Approximate P-value thresholds are 1.28 × 10−4 (LR 13.37) for significant linkage (expected once by chance in 20 genome scans) and 2.32 × 10−3 (LR 6.75) for suggestive linkage (expected once by chance in every genome scan). These thresholds assume an infinitely dense linkage map. Because this is not fulfilled in our study, these thresholds will be too strict. Thresholds are sometimes adjusted if marker density is sparse (e.g., by 20% for marker spacing of 10 cM, Slate et al. 2002; Tarka et al. 2010). We did not use this option because our marker spacing was denser than 10 cM and any adjustment seems arbitrary. However, this means that the thresholds are overly strict and we therefore also considered peaks that came very close to the thresholds to avoid type II errors.

We defined confidence intervals around linkage peaks as all locations where −log(P) values dropped by less than one unit compared to the maximum value on the chromosome (Lander and Botstein 1989; Pavlicev et al. 2008). Furthermore, we defined core peak regions as those where −log(P) values dropped by less than 0.5 units compared to the maximum value on the chromosome. The core peak regions do not constitute 95% confidence intervals for the location and are included here purely for descriptive purposes.

CONVERSION TO THE PHYSICAL MAP AND SEARCH FOR CANDIDATE GENES

Marker order was very similar between the genetic and the physical map (Backström et al. 2010). We delineated confidence regions on the physical map by linearly interpolating between the positions of the closest markers on the genetic map. Linear interpolation will sometimes be inaccurate due to variation in recombination rates across the chromosome. This is particularly problematic when interpolating between distantly located markers on the genetic map. Hence, physical locations should be interpreted as landmarks rather than clear-cut boundaries.

We extracted candidate genes involved in carotenoid metabolism from the literature (see references in the introduction). Specifically, we used the Ensembl (http://www.ensembl.org) pipeline (version 62) to search the zebra finch genome for locations of gene families and genes with the following search terms: “ninaB,”“ninaD,”“scavenger receptor B,” and “carotene” as well as all 11 genes listed in Walsh et al. (2011). ninaA and ninaD did not yield hits for the zebra finch, hence we searched for zebra finch orthologs of ninaA and ninaD in Drosophila using the “Orthologs” function in Ensembl. The search for carotene yielded BCO2 and BCMO1 as the only genes in the zebra finch (both listed in Walsh et al. 2011) and the search term “scavenger receptor B” also yielded results that overlapped with genes listed in Walsh et al. (2011). Furthermore, we downloaded gene ontology terms (biological process) using BioMart (http://www.ensembl.org/biomart) for all genes in QTL core peak regions. We manually investigated the list of terms for candidate terms with special attention given to metabolic, transport processes and immune function.

Results

QUANTITATIVE GENETICS

Beak redness increased linearly with age within the first four months of live (b = 0.26 ± 0.03, t = 7.62, P = 1.1 × 10−9, slope expressed in units change in beak redness per week of age up to day 120), but did not change thereafter (b = 8.7 × 10−5± 1.0 × 10−2, t = 0.084, P = 0.93). Beak redness decreased with inbreeding (inbreeding coefficient, b =−2.70 ± 0.70, t = 3.85, P = 5.8 × 10−4), and significantly differed between sessions (24 levels, F = 17.45, P = 4.2 × 10−11). After controlling for these confounding effects, beak redness showed a narrow-sense heritability of h2= 0.341 ± 0.057 (LRT: P < 10−10), maternal effects were estimated to be 0.035 ± 0.021 (identity of genetic mother, P = 0.0079), shared early-rearing effects were estimated to be 0.028 ± 0.022 (foster brood identity, P = 0.15), and nonshared permanent environment effects were estimated to be 0.119 ± 0.048 (subject identity, P = 0.0060). The W effect was estimated to be 0.077 ± 0.034 (maternal line identity, P = 0.0017). Including W line as a random effect reduced the permanent environment component (0.0685 ± 0.0492), but hardly affect the other components (additive genetic: 0.331 ± 0.056, maternal: 0.029 ± 0.020, foster environment: 0.022 ± 0.021). The genetic correlation in beak redness between the sexes was rA= 0.926 (CI: 0.671–1.062; estimated using MCMCglmm).

ADDITIVE QTL EFFECTS

We found two genomic regions that show suggestive linkage with beak redness (on Tgu1 and Tgu5, respectively) and another two regions that came very close to the suggestive linkage threshold (on Tgu13 and Tgu21, respectively) (Fig. 1). Because our linkage thresholds are too strict (see “Materials and Methods”), we treated the peaks at chromosomes 13 and 21 as suggestive of linkage. Estimates of association were clearly nonsignificant in all other regions of the linkage map.

Figure 1.

Whole-genome scan for additive QTL effects (shown in black). The upper panel shows the significance (in units of −log(P) values and likelihood ratios), whereas the lower plot shows effect sizes (in proportion of phenotypic variance explained). The abscissa shows the genetic positions, with chromosomes separated by vertical lines and named below the axis. Each dot refers to estimates from a 5-cM grid across the chromosomes, except for Tgu1, Tgu2, Tgu5, and Tug13, where each dot refers to estimates from a 1-cM grid. Horizontal dashed lines show the approximate adjusted significance threshold for significant (upper line) and suggestive (lower line) linkage. QTL confidence regions are shown in gray (darker shades show the core peak region). Short vertical lines in the upper part of each plot show marker positions. The gray line in the lower plot shows the variance explained by polygenic effects (excluding the effect of the locus under consideration).

Effect size estimates at the points of maximal significance were 7.0 ± 3.2% (Tgu1), 5.8 ± 2.6% (Tgu5), 5.5 ± 2.7% (Tgu13), and 10.7 ± 5.2% (Tgu21) of the phenotypic variance (after controlling for fixed effects). Effect size plots show that the polygenic effect was reduced in regions with significant additive QTL effects (Fig. 2). This is expected because the (additive) polygenic effect is the sum of all additive effects across the genome. In total, the point estimates for the four QTLs sum up to 29.0% of the phenotypic variance in beak redness. This is only slightly less than the heritability of the trait (34.2%), but is likely to be somewhat overestimated due to the Beavis effect (Beavis 1998).

Figure 2.

Details of the scan (at 1-cM intervals) for additive and dominance QTL effects. See Figure 1 for detailed explanations. Black lines show the significance (upper plot) and effect size (lower plot) of additive QTL effects, whereas gray lines show the respective estimates for dominance QTL effects. The x-axis shows units of cM.

Mapping QTL confidence regions onto the zebra finch genome shows that peak regions cover 3.0–53.2 Mb of sequence in the current assembly (Table 2, Fig. 3). Core confidence regions (see “Materials and Methods”) are slightly narrower (2.2–18.5 Mb). There are many genes in the QTL confidence regions (Tgu1: 53, Tgu5: 776, Tgu13: 344, Tgu21: 114) and even in the core peak regions (Tgu1: 51, Tgu5: 212, Tgu13: 267, Tgu21: 101). The large number of genes in peak regions complicates the search for candidate genes.

Table 2.  Confidence regions of additive and dominance QTL effects and approximate positions on the physical map.
Effect typeChromosomeConfidence regionCore peak region
Genetic map (cM)Physical map (Mb)Width (Mb)Genetic map (cM)Physical map (Mb)Width (Mb)
AdditiveTgu1 8–33113.03–116.83  3.8014–33113.03–115.95  2.91
DominanceTgu260–79  0.11–155.47155.3665–76 22.50–152.66130.16
AdditiveTgu5 9–36  4.83–57.98 53.1619–28 38.29–56.77 18.48
AdditiveTgu1320–53  0.15–16.96 16.8126–47  2.10–16.96 14.86
AdditiveTgu21 0–20  0.64–3.64  3.00 0–15  0.64–2.91  2.26
Figure 3.

Estimates of additive and dominance QTL effects mapped to chromosomes (Tgu1, Tgu2, Tgu5, Tgu13, and Tgu21). The upper plot shows the significance of the effect, the lower plot shows effect sizes. Additive effects are shown in black, dominance effects in dark gray. Estimates from the genetic map were mapped onto the physical map by interpolating genetic positions between markers. The x-axis shows units of Mb on chromosomes. Note that the physical map is inverted relative to the genetic map for Tgu1, Tgu5, and Tgu13. Furthermore, some markers do not have an unambiguous location in the genome assembly, hence same parts of the genetic maps are missing from Figure 3 (Tgu1 markers > 120 cM, Tgu5 markers > 60 cM, Tgu13 markers < 30 cM).

DOMINANCE QTL EFFECTS

After accounting for additive QTL effects, we found close-to-suggestive dominance effects on Tgu2 (Fig. 4). At the point of maximal significance, the dominance effect on Tgu2 was estimated to 8.3 ± 3.4% of the phenotypic variance. Notably, this peak (as well as other clearly nonsignificant dominance peaks) hardly affected estimates of the polygenic effect, but reduced the estimated permanent “environment” effect. This is expected if the polygenic effect captures additive genetic variation, although, intralocus dominance interactions are constant within individuals and will hence be attributed to the “permanent environment” effect, unless modeled explicitly.

Figure 4.

Whole-genome scan for dominance QTL effects while controlling for additive QTL effects. See Figure 1 for detailed explanations. Deviating from Figure 1, the upper panel shows the significance of the dominance effect in black and the lower plot shows the variance explained by the dominance effect in black, the additive genetic variance (polygenic plus local additive effects) in dark gray and the variance explained by the permanent environment effect in pale gray.

CANDIDATE GENES

We searched for candidate genes that are known to be involved in carotenoid metabolism in animals. Most of the candidate genes were on chromosomes with no evidence of linkage to beak color in our population of zebra finches (Table 3). Some candidates were not covered by our genetic map (two orthologs of Drosophila ninaB and gene CD36 on chromosome “unknown” and StARD1 [equivalent to StAR1 listed in Walsh et al. 2011] on Tgu22). BCO2 (an ortholog of BCDO2 in chicken) is located on Tgu24 at position 1.5 Mb, which is involved in carotenoid metabolism in sheep and chicken (see introduction). We only have three markers on Tgu24 and the closest marker is at 4.9 Mb, which means that the power to detect associations is low even if BCO2 is causally involved. Overall, our study provides no evidence that any of the candidate genes is associated with beak color in our population of zebra finches. We also scanned gene ontology annotations of genes represented in QTL core regions for candidate biological processes. Although we found some genes with the GO term “metabolic process” (GO: 0008152), such genes are widespread and occur on most chromosomes. One of our peaks (on Tgu13) includes several HLA class II genes that are functional in immune response (GO: 0006955). However, besides two genes on Tgu21 (GO 0006955: ENSTGUG00000002459 and ENSTGUG00000002461), we do not find indications of immune function genes in other peak regions.

Table 3.  Candidate genes for carotenoid metabolism and transport in animals. Physical and approximate genetic positions in the zebra finch genome are listed.
Candidate geneEnsembl IDChromosomePhysical position (Mb)Genetic position (cM)Remarks
BCMO1ENSTGUG00000004509Tgu11 2.529Ortholog of ninaD in Drosophila
BCO2ENSTGUG00000000294Tgu24 1.5< 0Ortholog of ninaD in Drosophila, ortholog of BCDO2 in chicken
RPE65ENSTGUG00000010240Tgu827.027–46Ortholog of ninaD in Drosophila
ninaBENSTGUG00000014582Tgu_unknown  Ortholog of ninaD in Drosophila
ninaBENSTGUG00000016002Tgu_unknownOrtholog of ninaD in Drosophila
CD36ENSTGUG00000002619Tgu1A10.164–65 
CD36ENSTGUG00000016390Tgu_unknown 
SCARB1ENSTGUG00000004988Tgu15 1.8 4–6Ortholog of ninaD in Drosophila, equivalent to SR–BI
SCARB2ENSTGUG00000001690Tgu4 1.93Scavenger receptor class B gene family
DMBT1–1ENSTGUG00000004547Tgu19 5.419–20Scavenger receptor class B gene family
APODENSTGUG00000008991Tgu914.326–27 
GSTA2ENSTGUG00000012882Tgu390.859–60 
PLINENSTGUG00000008109Tgu1012.928–32 
StARD1ENSTGUG00000004778Tgu22 2.8No markersEquivalent to StAR1
StARD3ENSTGUG00000002220Tgu27 1.7 0–22Equivalent to MLN64
StARD4ENSTGUG00000001003TguZ20.826–28Equivalent to StAR4
StARD5ENSTGUG00000007905Tgu1012.027–28Equivalent to StAR5

Discussion

We here present a study of segregating genetic variation for beak color in a population of zebra finches. First, we established that there is a very strong genetic correlation between beak redness in males and females. This corroborates previous studies (Price and Burley 1993; Price 1996) and suggests that it is largely the same genes that influence beak color in both sexes. It also justifies our approach to map QTLs for zebra finch beak color jointly for males and females, despite the considerable sex-difference in average beak redness. Second, we identified four genomic regions (on Tgu1, Tgu5, Tgu13, and Tgu21) that show additive associations with variation in beak redness and one region that is suggestive of dominance QTL effects (on Tgu2). Notably, the threshold of suggestive linkage is expected to be found on average only once per genome scan (Lander and Kruglyak 1995) and therefore constitutes a Poisson process with an expectation of one false-positive hit. Hence, the probability of finding four or more suggestive linkage peaks is 0.0189 (calculated from a Poisson distribution) and hence is unlikely to arise by chance alone. Effect size estimates were similar for all loci (5.5–10.7%) and together summed to almost equal the heritability of beak redness (29% compared to 34% heritability). This is likely to be an overestimate (Beavis effect, an overestimation of effect sizes in QTL studies, Beavis 1998). However, the amount of variation explained by our four candidate loci seems high enough to substantially limit the extent to which phenotypic beak redness can reflect additive genetic variation in condition across multiple loci of small effect (genic capture).

The genic capture hypothesis rests on the assumption that zebra finch beak redness is condition dependent. We can distinguish three types of condition dependence: (1) Environmental condition dependence, (2) nonadditive genetic condition dependence, and (3) additive genetic condition dependence. Only the latter is heritable and therefore relevant for the genic capture hypothesis. Although there is clear support for environmental condition dependence (Blount et al. 2003; Alonso-Alvarez et al. 2004; Gautier et al. 2008; Bolund et al. 2010b) and for inbreeding depression in zebra finch beak redness (a nonadditive genetic effect, Bolund et al. 2010a), there is only mixed evidence for the critical additive genetic condition dependence (Birkhead et al. 2006; Bolund et al. 2010b). In support of the genic capture model, Birkhead et al. (2006) found a strong positive genetic correlation between beak redness and one out of several immune assays (response to tetanus) as well as a weak positive genetic correlation with body condition (residual mass). No support for genic capture was found when looking at early growth, because beak redness tended to show a negative rather than positive genetic correlation with early growth (Bolund et al. 2010b). This indicates that “good genes” that would facilitate early growth (e.g., through metabolic efficiency) do not lead to redder beaks. We have here tested another critical prediction of the genic capture scenario and find no evidence for the genic capture scenario. In summary, although beak redness is clearly condition dependent in zebra finches, the extent to which this reflects “good genes” for condition seems to be limited. Further below we discuss the evidence for contemporary sexual selection acting on zebra finch beak color, which is a third critical assumption of the genic capture hypothesis that might not be fulfilled.

The question arises how the polymorphism in major genes is maintained. It seems rather unlikely that each of the associations represents de novo mutations that arose and spread in captivity, so these genetic polymorphisms likely also exist in the wild. Alleles with major effects must be present in high enough frequencies to be detectable in our linkage scan. Without data on other populations (in particular from the wild), it is unclear if major alleles have risen to higher frequencies in our population. There are two potential explanations of how polymorphisms at major genes are maintained (even at low frequencies) in the wild. First, directional selection on beak color might be weaker than commonly assumed, so that polymorphism in major genes is near neutral. Despite repeated evidence for female preferences for redder beaked males in captivity (Burley and Coopersmith 1987; Price and Burley 1994; Blount et al. 2003), we actually do not know whether the additive genetic variation in beak redness is under positive selection in males in the wild. Furthermore, the repeated failure of experimental studies to demonstrate a female preference for (artificial) beak redness itself (Collins et al. 1994; Sullivan 1994; Vos 1995, but see Burley and Coopersmith 1987) might also suggest that the female preference for natural beak redness is only an apparent one. Females might instead have preferred males in good condition (due to environmental and/or nonadditive genetic effects), but may not have paid attention to beak coloration itself. Both effects are difficult to separate in correlational studies. A recent meta-analysis has concluded that the preference for red beaks depends on the on the length of the imprinting phase (Simons and Verhulst 2011), but that study does not explain satisfactorily why experimental studies mostly failed to support this conclusion.

Second, major genes might be subject to antagonistic pleiotropic effects (Chenoweth and McGuigan 2010) and allelic variation with major positive effects on beak redness may be hindered from rising to fixation due to costly side effects. It is, for example, possible that genes involved in beak color have pleiotropic effects on nonsexual fitness (Hine et al. 2011). Interestingly, Price and Burley (1994) reported sexually antagonistic selection on beak coloration in zebra finches. They found sexual selection for more reddish beaks in males, and survival selection favoring less reddish beaks in females. The latter finding, however, requires further confirmation, because higher mortality of phenotypically redder beaked females is hard to reconcile with redder beaked females being less inbred (Bolund et al. 2010a) and better nourished as chicks (Bolund et al. 2010b). Other pleiotropic effects are possible. For example, beak color might be mediated by hormones like testosterone (Ardia et al. 2010) and additive genetic variation in testosterone levels might mediate the costly side effects of high testosterone levels.

Previous studies of sexual ornaments that are subject to mate choice in other species have also suggested an oligogenic basis of ornaments (eye span in stalk-eyed flies: Johns et al. 2005; comb size in chicken: Wright et al. 2008; display behavior and spine length in stickleback: Kitano et al. 2009; color patterns in guppy and cichlid fish: Tripathi et al. 2009; Magalhaes and Seehausen 2010). Although this supports our conclusion that sexual ornamentation is (at least sometimes) controlled by relatively few genes, there are two aspects that need consideration. First, there might be an issue of publication bias because polygenic control will not leave clear signals in association studies and hence it is less likely that such findings are reported. Second, and very importantly, most of these studies are based on crosses of either sister species (Magalhaes and Seehausen 2010) or populations of the same species (Wright et al. 2008; Kitano et al. 2009; Tripathi et al. 2009) and therefore do not allow conclusions about the genetic architecture of within-population variation. However, at least one study has demonstrated major genes for a sexual ornament based on selection line crosses (eye span in stalk-eyed flies: Johns et al. 2005) and hence for standing genetic variation within populations. Furthermore, a study on a wild population of sheep showed that traits involved in intrasexual competition (hence also sexually selected) are controlled by few loci (horn morphology, Johnston et al. 2010).

Because our data suggested an oligogenic basis of zebra finch beak color, we then checked for candidate genes that are known to be causally involved in carotenoid metabolism. A prediction of major-locus effects is that genes vary in their contribution to trait expression and that major genes are functionally more close to the trait of interest rather than acting indirectly as can be expect for polymorphisms with minor effect. Scans for candidate genes did not result in support for any of the a priori candidates (although our data are inconclusive about BCO2, a strong candidate from studies on Drosophila, von Lintig et al. 2001, chicken, Eriksson et al. 2008, and sheep, Våge and Boman 2010). Furthermore, a scan for candidate gene ontology terms in core peak regions did not yield any strong candidate terms. Hence, despite our prediction that major genes should be mechanistically close to carotenoid metabolism or transport, we cannot pinpoint them to a single gene (or group of genes). Further functional insight into carotenoid processing in animals could remedy this in the future. Alternatively, some regulatory element upstream of carotenoid metabolism and transport could be responsible for the association that we observe.

Fitness-related traits are predicted to have high dominance-related variance in relation to additive genetic variance (Frankham 1990; Crnokrak and Roff 1995; Merilä and Sheldon 1999). This is because directional selection will tend to erode additive genetic variation (even though under polygenic control genetic variation will be maintained at equilibrium), while most of the dominance variance will persist for longer. We found evidence for nonadditive dominance effects on Tgu2. The estimated effect size was almost as large as the permanent environment effect and we assume that it is somewhat overestimated. This is because the power to detect intralocus dominance beyond additive effects is rather low and significant effects are thus likely to be overestimated in their effect size (Ioannidis 2008). The upper limit to the sum of dominance effects should be set by the permanentt “environment” variance component, but the permanent “environment” variance also contains interlocus interactions (epistatic effects), epigenetic effects and nonshared environmental effects. Price and Burley (1993) did not find dominance variation in their population, which might potentially also be due to their coarser color chart scoring or lower statistical power. Our results show that there is some dominance variation for zebra finch beak color, at least in our population.

Our study (like any study that uses pedigree-based QTL associations) focuses on standing variation. This is relevant from a microevolutionary perspective because selection is operating on genetic variation within populations. However, it also implies that we cannot find associations with genes that are essential for producing red beaks if these genes are monomorphic in our population (e.g., because beneficial alleles relatively recently became fixed by directional selection). This could potentially explain why none of the candidate genes are contained within a QTL peak region. Possibly, transcription factors or other control regions that are functionally, but not necessarily physically, associated with the candidate genes are polymorphic and responsible for the observed QTL peaks. Better knowledge of functional pathways might facilitate the discovery of such loci.

Our results show that a few regions with moderate effects explain a large part of the heritability in zebra finch beak redness. This argues against genic capture as the main mechanism for maintaining genetic variation in beak color. Zebra finch beak color seems to be controlled by few genes and genetic polymorphism in such genes (or associated transcription factors) could be maintained by pleiotropic effects and/or antagonistic selection in females and males. Furthermore, we argue that apparent female preferences for red-beaked males might arise from a general female preference for males in good condition, but that polymorphisms in major genes could disrupt the correlation between beak color and condition. Intriguingly, this could also explain the discrepancies in studies on female preferences for male beak color.


Associate Editor: M. Wayne

ACKNOWLEDGMENTS

We are grateful to E. Bolund, C. Burger, and K. Martin, who helped with collecting spectrometric measurements, and to M. Schneider, who helped with laboratory work. Furthermore, we thank N. Backström and H. Mellenius for providing the linkage map. J. Müller and A. Gioti kindly provided advice on the candidate gene search. A. Husby helped with advice on the variance component analysis. A. Corl, S. Chenoweth, and two anonymous referees provided valuable comments on an earlier version of the manuscript. The work was funded by German Research Foundation grants FO340/1–1, FO340/1–2, and FO340/1–3 (W. F.), the European Union grant HPMF-CT-2002–01871 (W. F.), the Max Planck Society (B. K.), the Swedish Research Council (H. E.), and the Knut and Alice Wallenberg Foundation (H. E.).

Ancillary