Large-scale population genetic structure in Bonelli's Eagle Aquila fasciata

Study species

Bonelli's Eagle is a long-lived bird of prey (maximum age of reproduction: 25 years; Real & Mañosa 1997), with deferred maturity (age of first successful breeding: 4 years; Soutullo et al. 2008), relatively low fecundity (mean number of fledglings per breeding pair: 0.6–1.4; Del Moral 2006), and high adult survival (annual adult survival rate: 87.1–96.7%; Soutullo et al. 2008). Generation length varies across subpopulations but is between about 9 and 13 years (estimated from demographic parameters in Real & Mañosa 1997, Soutullo et al. 2008). Breeding individuals are territorial, and show strong pair bonding and stay with high tenacity in the breeding territory within and between years (Bosch et al. 2010). However, after fledging there is a transient nomadic phase before territorial recruitment, during which individuals range widely over hundreds or thousands of kilometres (Balbontín & Ferrer 2009, Cadahía et al. 2009, 2010). In a sample of Bonelli's Eagles marked as nestlings, territorial recruitment occurred primarily within or close to the natal subpopulation, with mean distance (± sd) from birthplace to the breeding territory of 136.9 ± 127.8 km (range 23–430 km; n = 22) for females, and 72.6 ± 60.9 km (19–239 km; n = 19) for males (recalculated from Hernández-Matías et al. 2010).

Sampling and microsatellite genotyping

Blood samples from 245 Bonelli's Eagle nestlings were collected in the Iberian Peninsula, Morocco and Cyprus (Table 1, Fig. 1). Samples within the western Mediterranean were considered representative of seven subpopulations (Fig. 1), identified on the basis of regional clusters of breeding territories largely separated by distribution gaps (Del Moral 2006). Three of these (west and southwest Spain and Morocco) contain the largest number of individuals, whereas the remaining subpopulations are much smaller and geographically peripheral to the core subpopulations. All these subpopulations were relatively close to each other (< 1000 km) and within the known range of movements undertaken by immature Bonelli's Eagles during the transient nomadic phase (Balbontín & Ferrer 2009, Cadahía et al. 2009, 2010). In contrast, Cyprus was far from the western Mediterranean populations (> 3000 km), with major discontinuities in the species’ distribution between the two localities (Ferguson-Lees & Christie 2001).

All samples were preserved in Queen's lysis buffer (Seutin et al. 1991) until DNA extraction and further laboratory analysis. DNA was extracted using a blood extraction kit from Amersham-Pharmacia (Uppsala, Sweden). DNA was diluted to a final approximate concentration of 10 ng/μL. All individuals were genotyped for a total of 17 microsatellite loci, of which 14 were specifically developed for Bonelli's Eagle (Mira et al. 2005) and three for Spanish Imperial Eagle Aquila adalberti (Martinez-Cruz et al. 2002). PCR-amplification was carried out as described by Mira et al. (2005).

Genetic variation

Samples were obtained from 140 different breeding territories (Table 1), with some nests yielding up to two nestling samples in a given year, and some breeding territories yielding nestling samples over several years. The overall sample thus included an unknown proportion of siblings and half-siblings, which might bias the allele frequencies and consequently the statistical estimators (Ruzzante 1998). Because of this, instead of the overall sample (n = 245) we used 10 subsets of individuals from different territories, each with 140 individuals selected by randomly taking a single nestling from each breeding territory. Each genetic analysis was carried out independently for each of the 10 subsets and results were then averaged across the 10 subsets to obtain the overall estimates. Basing analysis on individuals from different territories was judged to overcome eventual bias due to kinship, as preliminary analysis using Queller and Goodnight's (1989) genetic estimate of relatedness, R, and the software spagedi (Hardy & Vekemans 2002), indicated that within each population the mean relatedness between nestlings from different territories was negligible (mean R < 0 for all populations; Table S1), and thus we are confident that our sampling scheme could be used for further genetic analyses.

Average allelic richness, average observed heterozygosity (H_O) and unbiased average expected heterozygosity (H_E; Nei 1978) were estimated within each subpopulation using genetix 4.05 (Belkhir et al. 2004). Estimates of allelic richness using rarefaction were computed with hp-rare 1.0 (Kalinowski 2005) for a rarefaction size equal to the minimum sample obtained (n = 6; Cyprus and central Spain). Departure from Hardy–Weinberg equilibrium within each population was assessed with F_IS, estimated with f (Weir & Cockerham 1984), followed by a permutation test (10⁴ permutations) of the null hypothesis of random mating (H_O : F_IS = 0) implemented in genetix (Belkhir et al. 2004). Linkage disequilibrium was also tested with genetix. To control for multiple comparisons, we used Storey and Tibshirani's (2003) q-value method for estimating the false discovery rates (FDR) from the distribution of P-values. To obtain FDR of 5% in multiple testing, a P-value was considered significant at the 5% level only when the q-value was also ≤ 5% (Storey & Tibshirani 2003).

Population genetic structure

The distribution of individuals among a priori subpopulations was first visualized with factorial correspondence analysis (FCA) implemented in genetix. Levels of differentiation between subpopulations were then quantified with the widely used F_ST estimator θ (Weir & Cockerham 1984), and significant departures from H_O of no differentiation were tested with the permutation (10⁴ permutations) procedure implemented in genetix (Belkhir et al. 2004). To limit bias in this estimator due to a possible unbalanced distribution of polymorphism among populations, we also compared results with those obtained using the D_est (Jost 2008) estimator, corrected for the level of polymorphism, with statistical significance tested using the bootstrap resampling procedure (2 × 10⁴) implemented in the package demetics (Gerlach et al. 2010) for r (R Development Core Team 2012).

Isolation by distance (IBD) was assessed by correlating the pairwise estimates of F_ST and D_est with geographical distances between subpopulations using the Mantel (1967) test implemented in genetix (10⁴ randomizations). A similar analysis was carried out at the individual level using Queller and Goodnight's (1989) coefficient of relatedness, and relating pairwise relatedness to geographical distances with the permutation procedure (10⁴) implemented in spagedi (Hardy & Vekemans 2002). Because the exact location of many nests was unavailable, distances were always computed considering the geographical centre of each population. Cyprus was excluded from analyses because it was too far away from the remaining populations, whereas Morocco was excluded due to uncertain provenance of individuals.

To examine the extent to which a priori geographically defined subpopulations actually corresponded to genetic units, we used the Bayesian clustering method implemented in structure 2.3.1 (Pritchard et al. 2000, Falush et al. 2003). Analysis assumed the admixture model with α inferred from the data, and correlated allele frequencies with λ = 1 (Falush et al. 2003, Pritchard & Wen 2004), and were based on Markov chain Monte Carlo (MCMC) sampling with 2 × 10⁵ iterations, following a burn-in period of 10⁵ iterations. Information on the geographical location of individuals was used as recommended by Hubisz et al. (2009), which allows structure to be detected at lower levels of divergence. Ten independent runs of structure were performed for each K-value from one to nine. The optimal number of genetic units (K) was selected by considering for each value of K the log-likelihood given the number of clusters (Ln P(X|K) and the standardized second-order rate ∆K of change of Ln P(X|K) (Evanno et al. 2005). Although the latter approach has been criticized, it may still provide useful estimates when populations deviate from the finite island model, as for instance in the case of hierarchically structured populations (Waples & Gaggiotti 2006).

structure was also run using a hierarchical procedure, to assess whether dominant genetic structure could be hiding fine-scale genetic structure (Evanno et al. 2005). The procedure started by running a first round of structure using the overall sample, and then selecting the optimal number of genetic units using the ∆K approach (Evanno et al. 2005). At this stage, subpopulations clustering as a single genetic unit with membership probability > 0.90 were excluded from further analysis, and a second run of structure was performed on the individuals from each genetic unit including more than one subpopulation. The procedure was repeated until no further subpopulation could be identified as a distinct genetic unit. Outputs of structure were visualized and the Evanno et al. (2005) method implemented using harvester (Earl & vonHoldt 2012).

Recent dispersal

To estimate the recent exchange of individuals between genetic population units identified by structure, we used assignment tests which allocate each individual to the population from which its multilocus genotypes were most likely to be derived (Waser & Strobeck 1998), using arlequin (Excoffier et al. 2005). As sampling was based on nestlings, the population of birth of each individual was known without error. Therefore, assignment of an individual to a population different from the one in which it was sampled was taken to indicate a foreign origin of at least one of its parents.

Recent migration rates (last two generations) were also estimated using the Bayesian approach implemented in bayesass 3.0 (Wilson & Rannala 2003). The program was run using a large number of MCMC iterations (> 10⁷), a burn-in of 10⁶ and a sampling frequency of 10³. To ensure consistent and accurate estimates, several runs were started with different random number seeds, and MCMC chains were examined for evidence of convergence and mixing using tracer (Rambaut & Drummond 2004). When there were inconsistent estimates or convergence problems, analyses were repeated using longer runs and by using different random seed numbers.

Properties of microsatellite loci

In total, 123 alleles were detected across 17 microsatellite markers and 245 birds genotyped (Table S2). The number of alleles per locus in the overall sample varied from two to 22, with an average of 7.2 alleles per locus. The observed and expected heterozygosities were the lowest for the populations from southwest Portugal and Cyprus (Table 1). Mean allelic richness was also lowest for these two subpopulations after correcting for differences in sample size (Table 1).

Linkage disequilibrium tested on 136 pairs of loci at the level of each subpopulation only indicated statistically significant departures from independent segregation after correcting for multiple comparisons (FDR < 0.05) in one and seven pairs of loci, for southwest Portugal and eastern Spain, respectively. As the linked loci were inconsistent among populations, these findings suggest statistical rather than physical linkage, and thus all loci were used in subsequent analyses. The only inbreeding coefficient (F_IS) significantly different from zero at FDR < 0.05 was found for the population from western Spain (Table 1), suggesting a significant departure from Hardy–Weinberg equilibrium.

Population genetic structure

Factor correspondence analysis highlighted strong differentiation between eastern (Cyprus) and western Mediterranean populations (Fig. 2). In a second analysis that excluded Cyprus, there was evidence of sharp differentiation of Bonelli's Eagles from southwest Portugal, with most individuals occurring in a cluster well separated from the other Iberian and North African populations (Fig. 2). Individuals from the remaining populations clustered close to each other, with some clinal variation apparent from north to south (Fig. 2).

The F_ST and D_est estimators were highly correlated (r = 0.96, P < 0.001), although D_est-values were consistently larger than F_ST-values (Table 2). Both estimators pointed to significant differentiation among all subpopulations after correcting for multiple comparisons (FDR < 0.05), particularly between the population from Cyprus and those sampled in the western Mediterranean (Table 2). The subpopulation of southwest Portugal also showed marked differentiation from the remainder, whereas differentiation among the other subpopulations was relatively low, albeit statistically significant for all pairwise comparisons (Table 2).

Genetic and geographical distances between populations were significantly correlated for both F_ST (Mantel test: r = 0.56, P = 0.005) and D_est (Mantel test: r = 0.43, P = 0.038) estimators. However, differentiation observed in relation to the subpopulation of southwest Portugal showed consistent positive deviations from the isolation-by-distance trend lines, whereas the opposite was found for comparisons involving the other populations (Fig. 3, Fig. S1). When the population of southwest Portugal was removed from the analysis, there was an increase in the correlation between genetic and geographical distances for both F_ST (Mantel test: r = 0.66, P = 0.008) and D_est (Mantel test: r = 0.56, P = 0.055), although P-values were larger due to smaller sample sizes. In contrast to analyses at the population level, pairwise relatedness between individuals of different populations was not significantly related to geographical distance (r = 0.054, P > 0.85; Fig. S2).

In the first round of structure, Ln P(X|K) increased up to K = 4 and then levelled off, whereas ∆K peaked for K = 2, although showing a second, smaller peak for K = 4 (Fig. 4). At K = 2, the analysis separated nestlings from southwest Portugal (group membership probability = 0.954) from those of the other seven subpopulations (0.966). At K = 4, the analysis separated the subpopulations of Cyprus and southwest Portugal as clearly distinct genetic units (Fig. 1). The other two clusters were less clearly defined: one cluster included most of the individuals sampled from the coastal Mediterranean regions of Spain and North Africa (Morocco, and southwest and eastern Spain), and the other including most individuals sampled from the central Iberian Peninsula (northern Portugal, and western and central Spain) (Fig. 1). The central Iberian group included about 20% of the genotypes from the coastal Mediterranean group, particularly from southwest and eastern Spain, but the opposite was not observed (Fig. 1). A second round of analyses excluding southwest Portugal returned K = 3 (Fig. S3), confirming the separation of Cyprus, central Iberian and coastal Mediterranean subpopulations. The third round of analysis excluding Cyprus returned K = 2, separating the central Iberian and coastal Mediterranean subpopulations (Fig. S4).

Recent dispersal

In assignment tests, all nestlings from southwest Portugal and Cyprus were assigned to their subpopulation of origin, thus indicating that their parents were also from the same subpopulations. Furthermore, no nestling sampled from a different subpopulation was assigned to either southwest Portugal or Cyprus, thus pointing to few or no recent exchanges of individuals from these subpopulations with other subpopulations. In contrast, there was evidence of recent movements between the coastal Mediterranean and central Iberian populations, with 1.8% of the nestlings sampled in the former assigned to the latter, and 6.1% of those sampled in the latter assigned to the former. Foreign assignments were mostly recorded in the subpopulation from western Spain.

Inferences on recent migration rates using bayesass could not be drawn, as there were inconsistencies among runs and the analysis did not converge. Detailed analysis of the sampling of the MCMC chains using tracer indicated that bayesass was particularly poor in resolving directional migration rates between coastal Mediterranean and central Iberian subpopulations, with posterior probability distributions often showing bimodality.

Spatial population structure: current vs. historical processes

The current patterns of genetic structure observed in western Mediterranean Bonelli's Eagles may be a consequence of a long-standing pattern of distance-dependent dispersal or may be due to recent anthropogenic habitat fragmentation coupled with demographic decline and genetic drift (e.g. Martínez-Cruz et al. 2007). Although there is at present little evidence to support either hypothesis, the information available suggests that historical rather than recent processes may underlie the patterns observed. First, although many Bonelli's Eagle subpopulations suffered a marked demographic decline, those analysed in the present study showed no evidence of genetic bottlenecks (Mira 2006), probably because the population reduction was recent, the population size remained relatively high and the species’ long generation time provided a buffer against the genetic effects of population fluctuations (Hailer et al. 2006, Brown et al. 2007). Secondly, there is no evidence for recent reductions in population connectivity, as no major changes in the range and spatial configuration of Bonelli's Eagle subpopulations have been reported during recent decades (Del Moral 2006). Thirdly, in comparable long-lived species, for instance the Spanish Imperial Eagle, genetic structure resulting from anthropogenic processes was associated with demographic declines and fragmentation levels that were older and far more pronounced than those observed in Bonelli's Eagle (Martínez-Cruz et al. 2004, 2007). Despite this, it cannot be ruled out that recent anthropogenic processes affected contemporary genetic structure, particularly in small and peripheral populations that are prone to genetic drift (Martínez-Cruz et al. 2004, 2007, Alcaide et al. 2009). Clarifying these issues would require a more detailed analysis of the demographic trend and history of Bonelli's Eagle (Martínez-Cruz et al. 2007).

Distance-dependent dispersal was inferred from the pattern of isolation by distance among Bonelli's Eagle subpopulations, and by the observation that recent migrants were few in each subpopulation and originated primarily from neighbouring subpopulations. In contrast, pairwise relatedness between individuals sampled in different subpopulations was not correlated with distance, but this was probably because their true relatedness was inevitably low (they could at most be cousins, i.e. born from two brothers breeding in different subpopulations), and because kinship estimates based solely on genetic data are often imprecise (Van Horn et al. 2008). Taken together, these results are consistent with mark-recapture and satellite tracking studies pointing to a high degree of natal philopatry in the Bonelli's Eagle, with most individuals breeding within or close to their natal subpopulation (Cheylan et al. 1996, Hernández-Matías et al. 2010), despite the occurrence of a few recruitment events hundreds of kilometres from the individual's birthplace (Cheylan et al. 1996, Balbontín & Ferrer 2009, Cadahía et al. 2009, Hernández-Matías et al. 2010). These long-distance recruitments are likely to have precluded strong divergence among subpopulations, although they were probably sufficiently rare to maintain the observed levels of genetic differentiation and isolation by distance (Wright 1943, Kimura & Maruyama 1971) as observed in other philopatric species with high mobility (Alcaide et al. 2009, Barlow et al. 2011).

The hypothesis of philopatry contributing to maintain small to moderate levels of genetic differentiation between spatially disjunct subpopulations is not incompatible with the apparent homogeneity of the mitochondrial control region in western Mediterranean Bonelli's Eagles (Cadahía et al. 2007), as differences between the patterns derived from mtDNA and microsatellites are widespread in birds (e.g. Brito 2007, Hull et al. 2008b). Female-biased dispersal in Bonelli's Eagle (Hernández-Matías et al. 2010) might have contributed to homogenized allele frequencies of maternally inherited mitochondrial markers over large geographical regions, while retaining the signature of distance-dependent dispersal in autosomal markers. It is more likely, however, that failure to detect spatial structure was due to the low diversity (four haplotypes) of this marker, reducing the statistical power to detect genetic differentiation. As gene flow among populations appeared to be moderate, genetic structure was probably only detectable at the fine resolution allowed by a large number of microsatellites.

Geographical population configuration: effect on genetic diversity and gene flow

Geographical patterns of population configuration also seemed to play a role in the genetic patterns observed, probably because they resulted in some clusters of populations exchanging individuals more often among themselves than with other clusters. This was the case of subpopulations located along the Mediterranean coastal regions of the Iberian Peninsula (southwest and northeastern Spain) and northwestern Africa (Morocco), which are close to each other and emerged as a consistent genetic unit in the structure analyses. This clustering suggests that there may be some exchange of individuals across the Mediterranean, although larger sample sizes from northwest Africa would be needed to confirm this hypothesis. Although this exchange has not been directly observed (Cheylan et al. 1996, Real & Mañosa 2001, Balbontín & Ferrer 2009, Cadahía et al. 2009, 2010, Hernández-Matías et al. 2010), it should be expected given the close proximity of subpopulations on either side of the Strait of Gibraltar. Another genetic cluster emerging from the structure analyses included most individuals from northeast Portugal, and western and central Spain. In this central Iberian cluster there was a significant representation of genotypes from the Mediterranean cluster, most pronounced in the populations of central and, to a far lesser extent, western Spain. This pattern is suggestive of an asymmetrical exchange of individuals between the two clusters. Asymmetrical migration was supported by assignment tests and is in line with recent metapopulation models suggesting that the Bonelli's Eagle population in southwest Spain is a source of colonists for western and central Spain (A. Hernández-Matías pers. comm.). Unfortunately, this pattern could not be explored further due to the unreliable estimates of recent migration rates provided by bayesass, possibly because the estimation method fails when genetic differentiation is too low and migration rates are too high (Faubet et al. 2007).

The geographical position of the small subpopulation of southwest Portugal (< 20 breeding females in 1990; Beja & Palma 2008), in the periphery of the species’ core range, probably contributed to its low genetic diversity and high divergence in relation to other subpopulations. These patterns are consistent with the expectations of increased genetic drift and reduced gene flow in small and peripheral subpopulations in a metapopulation system (Eckert et al. 2008), as already observed for other birds of prey with high movement capabilities (Martínez-Cruz et al. 2004, 2007, Alcaide et al. 2009). It is noteworthy that similar patterns of high genetic divergence from the remaining populations of the Iberian Peninsula have been reported for a range of vertebrates inhabiting southwest Portugal, some of which have been recognized as highly circumscribed endemic taxa (e.g. Mesquita et al. 2005, Godinho et al. 2008, Gonçalves et al. 2009). Divergence of these populations is generally associated with the emergence of the Caldeirão Mountains in the Early Pliocene (3.4–5.3 Ma) (Gonçalves et al. 2009), which is probably much older than the process promoting genetic divergence of the Bonelli's Eagle subpopulations inhabiting the same region, as suggested by its lack of mtDNA differentiation (Cadahía et al. 2007). In both instances it may be argued that the remote geographical position of these mountains, located at the southwesternmost tip of the Iberian Peninsula and thus partially surrounded by the sea, and the relatively large gap between it and other mountain ranges with comparable environmental conditions, may be factors limiting gene flow and thus promoting genetic differentiation even in the case of a highly mobile species. Clearly, it would be interesting to investigate whether similar patterns of genetic differentiation occur in other bird species that also show rather isolated populations in the same region.

Together with the effects of small population size and peripheral geographical position, ecological divergence may also have played a role in the genetic pattern observed in Bonelli's Eagles from southwest Portugal, as suggested for other wide-ranging species (Stenseth et al. 2004, Musiani et al. 2007, Sacks et al. 2008), including birds of prey (Hull et al. 2008a). The subpopulation of southwest Portugal is ecologically unique in that individuals nest largely in trees (Palma et al. 2006), whereas individuals in other subpopulations nest predominantly on cliffs (Carrascal & Seoane 2009b). In these circumstances, an eventual behavioural imprinting to the natal nest type might contribute to non-random dispersal (Davis & Stamps 2004) and thus to some reproductive isolation between ecotypes (Beltman & Metz 2005). Restricted dispersal between populations with contrasting nest-site selection patterns has long been hypothesized in birds of prey (Fox 1995), particularly in the case of cliff- and tree-nesting Peregrine Falcons Falco peregrinus (e.g. Wegner et al. 2005). This hypothesis requires further testing because in birds of prey, the extent of imprinting to the fledging site may vary widely (e.g. Tordoff et al. 1998, Mannan et al. 2006, Rutz 2008), and the information available to support a link between nesting habitat divergence and reproductive isolation remains scarce and ambiguous (Nesje et al. 2000, Hull et al. 2008a, Riegert et al. 2010).

Conservation implications

Despite recent positive trends in some subpopulations, the conservation of Bonelli's Eagle remains a high priority (Del Moral 2006). Findings from this study confirm the view that management of Bonelli's Eagles should assume a metapopulation structure (Real & Mañosa 1997, Soutullo et al. 2008), with distance-dependent dispersal between spatially discrete subpopulations, and occasional recruitment hundreds of kilometres from an individual's natal site (Cheylan et al. 1996, Cadahía et al. 2009, Hernández-Matías et al. 2010). Furthermore, the study suggests that the geographical scale of this metapopulation may be larger than usually assumed (Real & Mañosa 1997, Soutullo et al. 2008), given the possible exchange of individuals between the Iberian Peninsula and northwestern Africa. These results have fundamental consequences for the development of demographic models, and thus to the design of conservation management strategies. For instance, the assumption of an equal probability of individuals dispersing to any subpopulation, including the natal one, was incorporated into a recent metapopulation model for the Bonelli's Eagle, which suggested that pre-adult mortality plays a key role in determining overall population trends (Soutullo et al. 2008). This result is in marked contrast to earlier demographic models that assumed local subpopulations to be isolated, concluding that adult mortality had the largest impact on population growth rate (Real & Mañosa 1997). The two results have explicit conservation implications, because adult and pre-adult Bonelli's Eagles often have disjunct spatial distributions and require distinct management actions (Real et al. 2001, Cadahía et al. 2010). New modelling studies focusing on the entire Bonelli's Eagle metapopulation and accounting for distance-dependent dispersal are thus needed to strengthen conservation prescriptions.

The association of high genetic and ecological divergence in Bonelli's Eagles from southwest Portugal merits particular attention due to the conservation implications. This is one of the few subpopulations in western Europe that is increasing and expanding into neighbouring regions (Beja & Palma 2008), with preliminary mark-recapture and genetic data suggesting that individuals from this population are occupying forested and agricultural landscapes that would be unavailable to cliff-nesting individuals (L. Palma, P. Beja, R. Godinho & N. Ferrand unpubl. data). This population may therefore need to be regarded as a distinct management unit, although further research is required to assess the extent to which ecological divergence is actually contributing to reduce gene flow with the other, cliff-nesting, populations (Crandall et al. 2000, Fraser & Bernatchez 2001).