LOCAL ADAPTATION MAINTAINS CLINAL VARIATION IN MELANIN-BASED COLORATION OF EUROPEAN BARN OWLS (TYTO ALBA)

Authors


Abstract

Ecological parameters vary in space, and the resulting heterogeneity of selective forces can drive adaptive population divergence. Clinal variation represents a classical model to study the interplay of gene flow and selection in the dynamics of this local adaptation process. Although geographic variation in phenotypic traits in discrete populations could be remainders of past adaptation, maintenance of adaptive clinal variation requires recurrent selection. Clinal variation in genetically determined traits is generally attributed to adaptation of different genotypes to local conditions along an environmental gradient, although it can as well arise from neutral processes. Here, we investigated whether selection accounts for the strong clinal variation observed in a highly heritable pheomelanin-based color trait in the European barn owl by comparing spatial differentiation of color and of neutral genes among populations. Barn owl's coloration varies continuously from white in southwestern Europe to reddish-brown in northeastern Europe. A very low differentiation at neutral genetic markers suggests that substantial gene flow occurs among populations. The persistence of pronounced color differentiation despite this strong gene flow is consistent with the hypothesis that selection is the primary force maintaining color variation among European populations. Therefore, the color cline is most likely the result of local adaptation.

The relative role of adaptive versus neutral processes in generating and maintaining genetic and phenotypic variation among as well as within species is still under debate (e.g., Nei 2005; Lynch 2007). Whereas the neutral forces of mutation, drift, and migration result in stochastic allele frequency changes over time and space, natural selection is a directed process eliminating deleterious alleles from populations and carrying advantageous alleles to fixation. If directional selection is the prevailing force of evolution, a paradox emerges: although in the long run natural populations should lose variation at both the phenotypic and functional genetic levels, they usually exhibit high genetic diversity and often extensive phenotypic variation. In homogeneous environments mutation and balancing selection may account for the maintenance of diversity (e.g., Star et al. 2007). However, environments are rarely constant, neither at the spatial nor the temporal scale, and over the evolutionary timescale populations are expected to adapt to the prevailing environmental conditions, that is, to be locally adapted. This selection linked to spatiotemporal environmental heterogeneity may be a major force promoting and maintaining phenotypic and genetic diversity both within and among populations (Felsenstein 1976; Hedrick et al. 1976; Hedrick 1986). Although a potential role of adaptation to temporally fluctuating selection in the maintenance of phenotypic variation has been rarely acknowledge (but see Grant and Grant 2002), local adaptation to spatially varying selection has been widely studied. Among the best-documented sources of local adaptation figure host–pathogen interactions (for reviews see Kaltz and Shykoff 1998; Sotka 2005; Greischar and Koskella 2007). Other examples of local adaptation include selective pressures induced by climate and temperature (McKay et al. 2001; Savolainen et al. 2007), altitude (Storz and Dubach 2004; Samietz et al. 2005; Byars et al. 2007), water salinity (Gomez-Mestre and Tejedo 2003; McCairns and Bernatchez 2008), interspecies competition and predation (Abjornsson et al. 2004; Grondahl and Ehlers 2008), or soil color-related camouflage against predators (Hoekstra et al. 2005).

Local adaptation may however not be fully achieved, as exchange of individuals adapted to alternative environments may move populations away from the locally optimal phenotype (Postma and van Noordwijk 2005; Räsänen and Hendry 2008). In addition, local adaptation restricts gene flow if the selection gradient between the alternative environments is steep enough to reduce immigrants’ probability to reproduce (Nosil et al. 2005). This implies that gene flow can impede local adaptation, whereas inversely local adaptation can lead to reproductive isolation and ultimately speciation (Nosil 2008; Räsänen and Hendry 2008).

Studying the interplay of selection and gene flow thus represents a central issue to the understanding of the evolution of local adaptation and phenotypic diversity (Bridle and Vines 2007). Clines in continuous populations are of particular interest in this context. Although in discrete populations differentiation in fitness-relevant phenotypic traits may reflect historical selection, the persistence of clinal variation requires recurrent spatially heterogeneous selection along environmental gradients to counterbalance the homogenizing effect of gene flow (Haldane 1948; Slatkin 1973). Many important environmental variables vary continually in space, and may lead to strong clinal variation in fitness-related traits (Huey et al. 2000). However, clines can also be an outcome of neutral evolution, being generated by genetic drift in populations connected through spatially limited gene flow (isolation-by-distance) (Endler 1977), by admixture of previously isolated populations (secondary contact) (Slatkin 1973; Barton and Hewitt 1985), or by spatial population expansions (Klopfstein et al. 2006; Excoffier and Ray 2008). Such clines arising as a result of colonization history have been demonstrated recently for phenotypic traits in two plant species (Keller et al. 2009, see also Vasemägi 2006 for a discussion on clines in gene frequencies in Drosophila). Thus, before invoking selection to explain clinal phenotypic variation, the null hypothesis of neutral evolution should be tested (Gould and Johnston 1972; Storz 2002).

One striking example of clinal phenotypic variation is found in the barn owl (Tyto alba) (Roulin et al. 2009). Across all major areas of its almost worldwide distribution, this species displays geographic variation in predominantly genetically determined pheomelanic coloration (Roulin et al. 1998; Roulin and Dijkstra 2003). The color cline is most pronounced on the European continent, where coloration of the body underside continuously varies from white in the southwest to reddish-brown in the northeast (Roulin 2003). Although less work has been carried out on pheomelanic compared to eumelanic color variation in this species, observations suggest that the degree of pheomelanism could be of functional importance. In Central Europe, it is related to diet, breeding rate, and growth rate (Roulin 2004; Roulin and Altwegg 2007; Roulin et al. 2008). However, neutral models of cline evolution have not been tested so far.

The comparison of the levels of population differentiation at putatively selected traits to differentiation at neutral genetic markers allows disentangling adaptive from neutral phenotypic differentiation (for reviews see Merilä and Crnokrak 2001; McKay and Latta 2002; Leinonen et al. 2008; Pujol et al. 2008; Whitlock 2008). If traits evolve neutrally, the proportion of their variation among populations should on average be identical to the proportion of among-population variation in allele frequencies at neutral loci. QST represents a metric of population differentiation at a quantitative trait, and under neutrality thus should equal FST, its analog of genetic differentiation at neutral genetic markers. When QST for a given trait exceeds FST, this indicates that phenotypic differentiation has been driven by selection for different phenotypes in different populations on the trait under study or on genetically correlated traits (Wright 1951; Spitze 1993). To test whether clinal color variation in the barn owl is maintained by selection on the European continent, we compared the differentiation of coloration and neutral genetic variation among 18 populations.

Methods and Materials

SAMPLING AND COLOR MEASUREMENTS

To measure barn owl coloration across Europe, breast feathers from a total of 373 birds were sampled by collaborators working in survey programs in 18 populations (Fig. 1). Sampling was conducted during the breeding season 2007, except for two populations in The Netherlands and Hungary (2004–2005). To minimize the probability that individuals were immigrants from other populations, only nestlings were considered and only one of them was randomly chosen per brood to reduce the probability that individuals are closely related (Table 1, adults were added in two populations with small sample size). The sampling area covered a maximum distance of 2395 km between Évora, Portugal, and Budapest, Hungary (Fig. 1, Table 1). The minimal and mean distances between two populations were 31 km and 864 km (SD: 531 km), respectively.

Figure 1.

Map showing the population sampling locations. The biggest city in an 80 km radius around the actual sampling area is indicated.

Table 1.  Population samplings’ characteristics.
PopulationCountryDistance from Évora (km)NHoHeComments
F M
ÉvoraP   07140.680.68-
BilbaoE 638650.630.676 juveniles, 5 adults
La RochelleF 986930.630.648 juveniles, 5 adults
NantesF109210130.650.65-
Le HavreF1404870.650.66-
GenevaCH144410170.620.65-
TroyesF15001990.670.65-
StuttgartD17777130.700.64-
HeidelbergD1801980.710.68-
GroningenNL194315140.640.63-
MagdeburgD210615150.590.64-
LeipzigD21266130.640.65-
BerlinD219813140.700.66-
KielD221910100.700.64-
RibeDK22531070.620.65-
BrnoCZ22819110.680.68-
RostockD23421470.640.67-
BudapestH2395970.710.68-

Pheomelanin-based plumage color of each individual bird was measured from one to five breast feathers (mean: 4.03, SD: 1.10), depending on the number of feathers available. To measure color, reflectance spectra from four points per breast feather were captured with a S2000 spectrophotometer (Ocean Optics, Dunedin, FL) and a dual deuterium and halogen 2000 light source (Mikropackan, Mikropack, Ostfildern, Germany). For each reflectance spectrum, the brown chroma was calculated following Montgomerie (2006). The brown chroma represents the contribution of the red part of the spectrum (600–700 nm) to the complete visible spectrum (300–700 nm). For each individual, the brown chroma was averaged (1) per feather (average among point measurements) and (2) per individual (average among feathers). The repeatability of assessing coloration was very high (97.6% of among-individual variance) as shown by the repeated measurement of coloration of 14 individuals twice one year apart.

MOLECULAR ANALYSES

Genomic DNA from all 373 individuals was extracted from the basal 1 mm of breast feather quills. Extractions were performed either on a BioSprint 96 extraction robot using the BioSprint 96 DNA blood kit or using the DNeasy blood and tissue kit, following the manufacturer's protocols (Qiagen, Hilden, Germany).

To estimate neutral genetic differentiation among barn owl populations, individuals were genotyped at seven polymorphic microsatellite loci (Ta-206, Ta-210, Ta-216, Ta-218, Ta-220, Ta-306, and Ta-414, Burri et al. 2008). Polymerase chain reactions (PCR) were performed in two PCR multiplexes (Table S1). Multiplex PCR reactions were run in a final volume of 8 μL, containing 2.5 μL of Multiplex PCR Kit buffer (Qiagen), 12 ng of DNA, and the multiplex primer mixes with forward primers fluorescently labeled. PCR conditions included an initial denaturation step at 95°C for 15 min, 34 cycles of denaturation at 94°C for 30 sec, primer annealing at 57°C for 1 min 30 sec, and primer extension at 72°C for 1 min. A final step at 60°C for 30 min was used to complete primer extension. Fragment analysis was run on an ABI 3100 automated sequencer using a Gene Scan™ 500 ROX™ size standard and allele lengths were assigned using GENEMAPPER 4.0 software (Applied Biosystems, Foster City, CA).

To account for sex in color analyses, molecular sex determination for all individuals was performed using the method described in Py et al. (2006). This method allows distinguishing sexes based on a length dimorphism between sex chromosomes in a segment of the SPINDLIN gene. In total, 187 males and 186 females were used in the analyses; details per populations are reported in Table 1.

ANALYSES OF POPULATION STRUCTURE

Data from all seven microsatellite markers were used to estimate the levels of differentiation among all populations. After verifying that populations were in Hardy–Weinberg equilibrium (FSTAT 2.9.4, updated from Goudet 1995) and checking for the presence of null-alleles (MICRO-CHECKER 2.2.3, Van Oosterhout et al. 2004), we computed FSTs (Weir and Cockerham 1984), to estimate neutral genetic differentiation between populations. Confidence intervals were estimated by running 1000 bootstrap iterations over loci. In addition, we estimated the differentiation statistic DEST between populations (Jost 2008) and Slatkin's RST (Slatkin 1995). In contrast to FST, DEST partitions total genetic variance into statistically independent within- and between-population components and thereby guards against deflated differentiation measures that can arise in measures such as FST if within- exceeds between-population genetic diversity (Jost 2008). RST (Slatkin 1995) accounts for microsatellite mutation pattern and is better suited than FST when mutation is important relative to migration (Slatkin 1995; Balloux and Goudet 2002). To test for a geographic structure of isolation-by-distance, we plotted neutral genetic differentiation (FST, DEST, and RST) against the geographic distance between populations. Significance of the regression was tested by a nonparametric, permutation-based, Mantel test, running 1000 bootstrap iterations. Pairwise RST were estimated in Arlequin 3.1 (Excoffier et al. 2005). All other analyses were conducted in FSTAT 2.9.4 or the R package HIERFSTAT (Goudet 2005).

To test if plumage color differed significantly between populations and sexes, we conducted a two-way analysis of variance (ANOVA). An analysis of covariance (ANCOVA) was then used to test for clinal variation in mean coloration, entering the populations’ distance to the southernmost population (Évora, Portugal), sex, and their interaction as independent variables.

COMPARING COLORATION AND NEUTRAL GENETIC DIFFERENTIATION

Testing whether a phenotypic cline evolved by neutral processes or by selection requires that phenotypic variation is compared to the neutral patterns of evolution driven by population history and demography. Although migration and genetic drift have an equal effect all over the genome, selection affects only regions harboring the quantitative trait loci (QTL) underlying the phenotypic trait it acts on. Thus if selection causes divergent evolution of phenotypes among populations, either because selection is exerted on coloration itself or on genetically correlated traits, phenotypic differentiation is expected to exceed neutral differentiation, especially if populations remain interconnected by gene flow such as in cline models. Contrasting the geographic structures of FST and QST, that is, contrasting the respective correlations of FST and QST with geographic distance between pairs of populations constitutes a more robust test of geographically gradually varying selection than comparing overall value of QST and FST. Indeed the latter comparison does not depend on the absolute magnitudes of phenotypic and genetic differentiation. With this approach, selection is inferred from a significant difference between the slopes of the respective regressions: if geographically disruptive selection is stronger than the homogenizing effect of gene flow, phenotypes will diverge more markedly among populations with increasing distance than populations differ at neutral genetic markers.

QST calculation requires experimental estimates of additive genetic variances (in common gardens for example), but for many species such estimates are impossible to obtain for logistical reasons. So in practice, many authors calculated QST with assumptions on the determinism of the trait under study and tested the sensitivity of the results to those assumptions (see e.g., Saether et al. 2007 and references therein). To make the strict distinction between those surrogates of QST and “true”QST, Saether et al. (2007) proposed to call them PST (for phenotypic or pseudo-QST). Melanic color, such as the pheomelanic trait we analyzed here, is often not dependent on nutritional intake, but on mostly genetically determined melanin deposition (Mundy 2006). Although in the barn owl most of the variation in coloration is genetically determined at least in Switzerland, such as shown by cross-fostering experiments (h2= 0.81 ± 0.09, Roulin et al. 1998; Roulin and Dijkstra 2003), we did not experimentally estimate additive genetic variance for barn owl color in this study. We therefore are strict by referring differentiation in plumage coloration to PST rather than QST.

PST is a function of the within- (σ2w) and between-population phenotypic variances (σ2b), heritability (h2), and the proportion of the between-population phenotypic variation due to additive genetic effects (g, and 1 −g corresponds to the environmental effect). Pairwise PST-values for color were calculated as follows (Wright 1951; Spitze 1993):

image

Within- and between-population phenotypic variances were assessed by extracting mean squares (MS) from a two-way ANOVA on color, with factors including population and sex. Within-population MS are an unbiased estimate of the within-population variance 2w). Between-population variance 2b) can be estimated as

image

where MSb and MSw are the within- and between-population MS. n0 is a weighted average of sample size for each comparison and following Sokal and Rohlf (1995, p. 179–217) is calculated as

image

where a is the number of populations to be compared and ni the number of individuals in the ith population. For further details about the procedure we refer to Storz (2002).

To estimate the effect of geographic distance on color differentiation between populations, we plotted pairwise PST against pairwise geographic distances between populations. As for neutral genetic differentiation, significance of the regression was tested using the nonparametric, permutation-based Mantel test. Finally, to investigate whether selection was involved in the evolution of the color cline, we tested whether population history alone explained the spatial structure of phenotypic differentiation, or whether the latter persisted if phylogeographic effects inferred from neutral genetic variation were accounted for. Tests of this kind usually involved partial Mantel tests among PST as a response matrix and FST and geographic distances as first and second explanatory matrices, respectively (Storz 2002; Saether et al. 2007). However, PST and FST are supposed to be identical under the null hypothesis of absence of selection on color or genetically correlated traits. Thus, a simple way to test this hypothesis is to contrast the matrix of pairwise differences between PST and FST, PSTFST, with the matrix of geographic distances. Under our null hypothesis, the two matrices should be uncorrelated, and a positive correlation would indicate a strong signal that selection is acting on the color polymorphism or correlated traits.

Heritability h2 and the between population additive genetic component g are often different from 1. Following Roulin and Dijkstra (2003), heritability was set to 0.81 and, as an assumption, g to 1 in the first place (Fig. 2, bottom left). However, we tested the robustness of our results with respect to different hypotheses on the proportion g of phenotypic variance. The above analyses thus were repeated by varying g between values of 1 (all phenotypic variance due to additive genetic effects) and 0.01 (1% phenotypic variance due to additive genetic effects), which is a broader interval than generally tested in comparable studies (Storz 2002, g= 0.15–1; Saether et al. 2007, g= 0.05–1). In this sensitivity analysis, we fixed h2 to 1 given that Roulin and Dijkstra (2003) found very high heritability in a Swiss barn owl population, and that overestimation of h2 leads to a deflation of PST, which is a conservative bias for the question addressed here.

Figure 2.

Graphs showing the neutral genetic and color differentiation between populations. Shown are the linear regressions of differentiation against geographic distance between pairs of populations. Top panels: neutral genetic differentiation in terms of FST (left) and DEST (right). Bottom left panel: color (PST; h2= 0.81, g= 1) and neutral genetic differentiation (FST). Bottom right panel: sensitivity of PST against the variation of the proportion of the between-population phenotypic variance due to additive genetic effects (g). PSTs are indicated by a solid line for three values of g. Points are not drawn for clarity. Phenotypic differentiation increases significantly more with distance than neutral genetic differentiation, even when only 1% of the phenotypic variance observed between populations is due to additive genetic effects (g= 0.01).

Results

NEUTRAL GENETIC POPULATION STRUCTURE

The neutral genetic structure of barn owl populations across Europe was very low but significant, with an overall FST of 0.011 (99% confidence interval 0.007–0.016). Despite the weak population structure, a slight but significant pattern of isolation-by-distance was found, indicated by the positive correlation between pairwise FST and geographic distances between populations (Mantel test: R2= 0.175; P= 0.001) (Fig. 2, top left). Similar differentiation and isolation-by-distance was found for DEST (Fig. 2, top right) and RST (data not shown), except that sampling variance was high for the latter. Therefore, only FST was used in the following analyses, as it is statistically directly comparable to PST. Furthermore, Balloux and Goudet (2002) showed that FST is better suited than RST when population structure is low and when microsatellites violate the stepwise mutation model such as is the case for some of the markers used in the present study.

GEOGRAPHIC STRUCTURE OF COLORATION

Our study confirmed previously reported patterns of barn owl color variation quantified by museum skin measurements (Roulin 2003; Roulin et al. 2009). Mean plumage color per population, measured in terms of brown chroma, varied between 0.253 and 0.325 and marked differences of mean coloration were found both among populations and among sexes (two-way ANOVA, population: F= 19.331, P < 0.001, sex: F= 14.078, P= 0.002). A strong clinal pattern of color variation was found when plotting the mean color per population by sex against distance from Évora, Portugal (Fig. 3). An analysis of covariance revealed significant effects of distance from Évora and sex, but no differences in geographic variation between sexes (ANCOVA, distance from Évora: t= 9.687, P < 0.001; sex: t=−2.607, P= 0.014; interaction distance from Évora*sex: P= 0.99). Accordingly, the geographic structure of plumage color was very high, with an overall PST of 0.353 (with h2= 0.81, following Roulin and Dijkstra 2003 and g= 1).

Figure 3.

Change of mean coloration across Europe. Shown is the linear regression of mean coloration by sex per population against the distance of each population to the south-western-most population (Évora, Portugal). Male's values are depicted by open circles and a dashed line, and female's values by filled circles and a solid line.

COMPARING COLORATION AND NEUTRAL GENETIC DIFFERENTIATION

The overall differentiation between populations was more than 30-fold larger for color (PST= 0.353) than for microsatellites (FST= 0.011). The linear regression of pairwise PSTs against pairwise geographic distances between populations revealed a very strong and highly significant pattern of isolation-by-distance on color differentiation (R2= 0.551; P= 0.001) (Fig. 2, bottom left). More importantly, the positive correlation between pairwise PSTFST differences and pairwise geographic distances demonstrates that the isolation-by-distance observed on color differentiation holds when eliminating the baseline level of differentiation resulting from historical and demographic factors (Mantel test R2= 0.528; P= 0.001).

To test whether these results were robust against changing assumptions on h2 and g, we tested the robustness of our results to those assumptions in a sensitivity analysis (Fig. 2, bottom right). As aforementioned, a heritability of one is the most conservative assumption to demonstrate that PST exceeds FST. By testing values of g between 1 and 0.01, we showed that even when only 1% of the variance between populations is due to additive genetic effect, there is still a significantly stronger isolation-by-distance on color than on the neutral genetic markers (R2= 0.043; P= 0.013). This analysis thus showed that our conclusion on the involvement of selection in color differences among population is extremely robust to any realistic assumptions on the determinism of the trait.

Discussion

Based on the comparison of the geographic differentiation of coloration to the neutral genetic population differentiation throughout Europe, we show that the barn owl's clinal pheomelanic coloration is not the result of genetic drift, but the most likely result of local adaptation. In accordance with observations from many other birds species (Crochet 2000 and references therein), the neutral genetic structure among European barn owl populations is low (overall FST= 0.011). The weak but significant pattern of isolation-by-distance suggests that regular effective migration leads to extensive and spatially weakly restricted admixture of neutral genetic variation among populations. This conclusion is supported by barn owl ring recovery data (Cramp 1985; Paradis et al. 1998; Marti 1999). A recent analysis (following Paradis et al. 1998) of data provided by the Swiss Ornithological Institute (Sempach, Switzerland) revealed a mean dispersal distance of 65 km (median 23 km) with a standard deviation of 121 km (n= 321) and dispersal movements as far as several hundreds of kilometers. Although in theory more detailed insights into the genetic population structure and quantitative measures of gene flow could be obtained from genetic data using clustering approaches or coalescent-based methods, we refrained from performing these analyses, as neither of them is expected to provide satisfying inference of population structure or migration rates with data from almost continuous populations with weak differentiation (FST= 0.011) and isolation-by-distance as observed in our data (Falush et al. 2007; Faubet et al. 2007).

Given the high rates of gene flow uncovered by genetic analyses, coloration is expected to be homogenized among European barn owl populations if such coloration evolved by purely neutral processes. However, color differentiation remains strong after accounting for neutral genetic population differentiation, suggesting that strong recurrent selection on coloration or genetically correlated traits is involved in the maintenance of the clinal coloration polymorphism. This conclusion remains consistent even when coloration is assumed to be completely heritable (h2= 1) and to its largest extent dependent on variation in environment between populations (g= 1%, i.e., 99% of the variation between population is due to environmental effects). Experimental evidence from Swiss barn owls shows very limited environmental dependence of melanin-based coloration (Roulin and Dijkstra 2003). The parameter space used in our sensitivity analysis is thus far beyond realistic estimates of the environmental component (1 –g) and conservative for heritability (h2). Pujol et al. (2008) recently criticized the use of in situ phenotypic measures from natural populations to infer selection (the PST approach). The conducted sensitivity analyses permit to settle this problem, and the congruence of the results obtained by FST and DEST confirms the weak population structure indicated by genetic data not being an artifact of high within-population genetic diversity.

As demonstrated, our data rule out that the color cline established as a result of spatially restricted gene flow. However, patterns closely resembling the ones expected from selection acting along an environmental gradient can be generated by the neutral process of surfing that has found little attention in the study of phenotypic clines so far (Klopfstein et al. 2006; Excoffier and Ray 2008). Alleles responsible for whitish or reddish-brown colorations could have gradually increased in frequency by genetic drift acting at the front of a past spatial population expansion out of either end of today's cline. Although surfing might be a frequent phenomenon leading to phenotypic clines, especially when starting from standing genetic variation (Excoffier and Ray 2008), several findings argue against this neutral scenario in the case of the barn owl color cline. (1) Experimental evidence suggests that coloration indeed is of functional relevance (Roulin 2004; Roulin and Altwegg 2007; Roulin et al. 2008). (2) Long-range dispersal as observed in the barn owl renders surfing unlikely compared to species that disperse over short distances in a stepping-stone-like manner. (3) Apart from clinal variation on the European continent, color clines in the barn owl independently evolved in North and South America and in Africa (Roulin et al. 2009). As surfing is a stochastic process affecting random regions in the genome, it appears unlikely that coloration would have been involved in surfing events four times independently. (4) Last and most importantly, surfing would be most likely if the colonization of Europe after the last glaciation occurred out of a single refugium. However, as already suggested by Voous (1950) half a century ago and confirmed by preliminary mitochondrial data obtained in our laboratory (S. Antoniazza, R. Burri, L. Fumagalli, J. Goudet, and A. Roulin, unpubl. data), Europe seems to have been colonized from at least two regions. Even though postglacial colonization might have brought into secondary contact two distinct color morphs that evolved in allopatry, and the color cline in first place could have established neutrally by admixture, the maintenance of the cline despite extensive gene flow requires the recurrent action of selection.

SPATIALLY HETEROGENEOUS SELECTION AND CLINE EVOLUTION

The most eminent question for the evolution of phenotypic diversity in species with continuous repartition across whole continents is how selection can restrict homogenization of phenotypic traits in the presence of high rates of gene flow (Nosil 2008; Räsänen and Hendry 2008). In barn owls, several lines of evidence suggest that both indirect selection and direct selection are involved in the maintenance of the color polymorphism. Reddish-brown individuals invest more into parental care (Roulin et al. 2001; Roulin 2006) and they grow faster in body mass than white ones when rearing conditions are relaxed (Roulin et al. 2008). The linkage of color to life-history components in this case seems to result from physical linkage or pleiotropy of the respective genes (see also Roulin 2006; Ducrest et al. 2008), whereas the correlation of the color polymorphism with diet (Roulin 2004) may establish by direct selection on color as a consequence of different foraging success upon alternative prey. Altogether, this indicates that barn owl color phenotypes occupy different ecological niches (Roulin 2004; Roulin and Altwegg 2007; Roulin et al. 2008), suggesting that divergent selection among niches maintains the color cline observed at the continental scale.

Depending on the niche distribution across the continent, different scenarios may explain the evolution of the color cline. In the simplest model, selection might be exerted by an environmental gradient creating locally homogeneous niches. In each region individuals of nonadapted phenotype are counter-selected, but strong recurrent immigration maintains phenotypic variation. Close inspection of the patterns of color variation in barn owls reveals that even though mean coloration tightly fits a linear geographic cline, color variation within populations is usually extensive. It seems thus more likely that habitat is also heterogeneous at the local scale, with most niches present all over the continent, but at gradually changing frequencies. Immigrants of any phenotype may thus settle and reproduce almost throughout the continent. The maintenance of the color cline in such heterogeneous landscapes involves different processes: (1) ecological selection prevents invasion of niches by nonadapted phenotypes, and (2) niche frequency and competition for niches determine the local frequencies of phenotypes, and thereby the local mean coloration. The strong phenotypic differentiation at the continental scale would be the result of local ecological selection acting in conjunction with a cline in niche frequency and selective pressure. Populations may then even almost freely exchange neutral genetic diversity, because selection intensity on progeny of subsequent generations will rapidly decrease with increasing level of back-crossing when mating is most likely with locally adapted individuals. Additional sampling effort, spatially explicit simulations of selection and population-level radio-telemetry observations in conjunction with the use of high-resolution environmental maps will help elucidate the spatial scale at which the selection pressures involved in barn owl color evolution are acting.

LOCAL ADAPTATION AT LARGE SPATIAL SCALES

The present study represents a striking illustration of the levels of phenotypic differentiation that can be achieved in nature despite substantial gene flow. Although the patterns observed in the barn owl might seem particular, we believe that they are far from being restricted to this species. Rather, we expect them to be ubiquitous in nature, especially in species with large distributions and high dispersal propensity. However, so far only a handful of studies combined measures of phenotypic and neutral genetic variation in natural populations displaying phenotypic clines at continental scales (examples include Long and Singh 1995; Merilä 1997; Gockel et al. 2001; Storz 2002; Palo et al. 2003; Ingvarsson et al. 2006; Savolainen et al. 2007; Demont et al. 2008). Compared to the linear gradients and large geographic distances in these cases (>1000 km), other studies that identified phenotypic clines rather reported sharp transition zones of limited width relative to the species’ distribution, with phenotypes changing quickly across hybrid zones (reviewed in Barton and Hewitt 1985) or ecotones (e.g., Mullen and Hoekstra 2008). We put forward that the extent and pronounced linearity of the phenotypic clines and phenotypic isolation-by-distance despite almost absent neutral genetic population differentiation observed in the former studies and in the barn owl are likely a matter of spatial scale. At large spatial scales, the environment varies in numerous ecological dimensions that constitute likely selective agents, and local adaptation at these scales may seem inevitable. Working at such scale permits to minimize the influence of local variation in environmental conditions relative to their variation over the entire study area. As moreover many important ecological parameters are expected to vary linearly across continents, clear-cut patterns of isolation-by-distance on phenotypic traits may easily emerge at large spatial scales, because distance is a good proxy for environmental variation. The confirmation of the action of selection acting at large spatial scales may thus be straight forward. However, the identification of the selective agents such as climatic and ecological variables, in turn, is tremendously flawed with problems of spatial autocorrelation. Finally, a combination of approaches integrating various spatial scales and methods such as landscape genetics will be essential to get track of the detailed selective agents involved in the process of local adaptation.

Conclusion

We believe that the barn owl is a well-suited model for the study of the interplay of gene flow and selection, which in turn is of central importance to an increased understanding of the processes of local adaptation and speciation. This species represents one of only six worldwide distributed bird species, and there may be only few more vertebrate species with comparable autochthonous areas. This implies that the species encounters various environmental conditions, and we expect local adaptation to be common. Might the species show a special propensity for adaptation that explains its global success? The identification of selection, such as conducted here for color, provides only the first essential step toward the understanding of how local adaptation evolves. The world-wide distribution provides exceptional opportunities to confirm hypotheses derived from single populations and compare patterns and processes of phenotypic and genomic evolution among allopatric populations. In-depth studies of phenotype frequencies within populations and detailed description of the color cline in conjunction with the establishment of the species’ phylogeography will provide important information on the evolutionary history of the color cline in Europe and the spatial scale at which selection is acting. Finally, tracking down color evolution to the molecular level will allow us to study whether in allopatric populations the same underlying genes and processes are involved in barn owl color evolution.


Associate Editor: H. Hoekstra

ACKNOWLEDGMENTS

R. Alonso, Athenas Franche-Compté, A. Beaufils, C. Blaize, G. Burneleau, J. De Jong, K. Dichmann, J. Frank, J. Gonin, K.-H. Graef, A. Häller, B. Hartung, M. Hug, P. Iseli, H. Keil, G. Klammer, A. Klein, E. Kniprath, F. Krause, O. Lambert, G. Linde, J. Luge, A. Marques, H. D. Martens, G. Moyne, K. Poprach, R. Riep, I. Roque, O. Schmidt, H. Seeler, J. Soufflot, F. Steiner, F. Valera, F. Ziemann, and I. Zuberogoïta kindly provided samples, without which this study would not have been possible. We thank R. Piault, P. Bize, and A. Dreiss for technical support with color measurements and K. Ghali for support with the organization of the sampling. P. Nosil, H. Hoekstra, and two anonymous reviewers provided helpful comments on an earlier version of the manuscript. The Swiss Ornithological Institute kindly provided Swiss barn owl's ring recovery data. This study was supported by Swiss National Science Foundation grants 3100A0-109852/1 to LF and PPOOA-102913 to AR.