Atypical panmixia in a European dolphin species (Delphinus delphis): implications for the evolution of diversity across oceanic boundaries

Authors


Correspondence: A. Rus Hoelzel, School of Biological and Biomedical Sciences, Durham University, South Road, DH1 3LE, UK.

Tel.: +44 191 334 1325; fax: +44 191 334 1201;e-mail: a.r.hoelzel@durham.ac.uk

Abstract

Despite the scarcity of geographical barriers in the ocean environment, delphinid cetaceans often exhibit marked patterns of population structure on a regional scale. The European coastline is a prime example, with species exhibiting population structure across well-defined environmental boundaries. Here we undertake a comprehensive population genetic study on the European common dolphin (Delphinus delphis, based on 492 samples and 15 loci) and establish that this species shows exceptional panmixia across most of the study range. We found differentiation only between the eastern and western Mediterranean, consistent with earlier studies, and here use approximate Bayesian computations to explore different scenarios to explain the observed pattern. Our results suggest that a recent population bottleneck likely contributed significantly to the differentiation of the Eastern Mediterranean population (in Greek waters). This interpretation is consistent with independent census data that suggest a sharp population decline in the recent past. The implication is that an unperturbed population may currently show panmixia across the full study range. This exception to the more typical pattern of population structure seen for other regional dolphin species (and for common dolphin populations elsewhere in the world) suggests particular ecological or life-history traits distinct to this species in European waters.

Introduction

Barriers to gene flow are often physical, thereby affecting a range of different species in the same way. Physical barriers in the oceans may often be porous, but still influence a broad range of species. For example, a number of species show evidence of restricted gene flow across the Almeria-Oran thermal front separating the Mediterranean Sea from the North Atlantic (e.g. sea bass, Dicentrarchus labrax, Naciri et al., 1999; triplefin fish, Tripterygion delaisi, Carreras-Carbonell et al., 2006; Mediterranean mussels, Mytilus galloprovincialis, Quesada et al., 1995; cuttlefish, Sepia officinalis, Perez-Losada et al., 2002; and bottlenose dolphins, Tursiops truncatus, Natoli et al., 2005), even though there is extensive movement across the front as well. One likely mechanism restricting gene flow for this system is the difference in prey resource found in the relatively cold or warm water on either side of the front, and the evolution of local habitat dependence or adaptations (see discussion in Natoli et al., 2005). At the same time, isolating factors that are ecological, temporal or behavioural (e.g. see Via, 2001) have the potential to affect different species in different ways.

Cetaceans have high dispersal potential; however, studies have shown that instead of having large panmictic populations, these species often show considerable genetic and morphological structuring on a regional scale (e.g. Hoelzel, 2002). Given the rarity of hard boundaries to dispersal in the oceans (such as rivers or mountain ranges may represent boundaries to some terrestrial species), differences in local environments have been proposed as an important mechanism promoting divergence in cetaceans (e.g. Natoli et al., 2005, 2006; Fontaine et al., 2007; Amaral et al., 2012).

Along the European coastline, most cetacean species studied exhibit marked population structure, with genetic breaks generally overlapping spatially with well-described environmental transitions. The main transitions are between the southern and northern North Sea (Andersen et al., 2001; Banguera-Hinestroza et al., 2010); between the North Sea and Atlantic (Natoli et al., 2005; Fontaine et al., 2007); between the Mediterranean and the Atlantic (Garcia-Martinez et al., 1999; Natoli et al., 2005); and between Western and Eastern Mediterranean basin (Natoli et al., 2005; Gaspari et al., 2007b). For the common dolphin (Delphinus delphis), there had been evidence for population differentiation only between the Eastern and Western Mediterranean (Natoli et al., 2008; Mirimin et al., 2009).

In other regions of the world, small-scale genetic differentiation has been reported for the common dolphin, with differentiation in one case being particularly strong and leading to the naming of a new species, the long-beaked form Delphinus capensis (Rosel et al., 1994; Kingston & Rosel, 2004; Chivers et al., 2005; Bilgmann et al., 2008), although the generality of this classification worldwide remains controversial (as long- and short-beaked populations worldwide do not form separate monophyletic lineages; Natoli et al., 2006). Here we focus on the short-beaked common dolphin, Delphinus delphis, the only Delphinus species recognized throughout the study range. However, although there is evidence for significant genetic differentiation among regional populations of Delphinus delphis elsewhere (for example with FST ranging from 0.014 to 0.047 in the Pacific and Indian Oceans, Amaral et al., 2012; and 0.012 to 0.071 among populations separated by 50–600 km along the coast of Australia, Möller et al., 2011), there is little evidence for this level of differentiation in the North Atlantic (Natoli et al., 2006; Mirimin et al., 2009; Amaral et al., 2012). At the same time, the previous North Atlantic studies did not assess all of the same boundaries as investigated for other small cetaceans in European waters, and power was sometimes limited (few loci or small sample sizes) leading these authors to suggest the possibility of undetected population structure. There was also the suggestion of some segregation between different regions in Europe based on morphology (Murphy et al., 2006) and ecophysiology (Amaral et al., 2007).

Here we focus on an intensive sampling of the geographic range of common dolphins in Europe, spanning from Scotland, south along the European coast and into the Mediterranean, providing a high-resolution assessment (based on 15 microsatellite DNA loci) across the several boundaries that have shown evidence for structure for other regional cetacean species. A regular distribution of sampling sites across the European coastal region, especially along the transition zone between the Mediterranean Sea and the Eurosiberian water mass in the north (see Alcaraz et al. 2006), permitted a fine-scale assessment of structure in the context of environmental factors, not previously addressed for this species in this region. We also investigated the previously reported differentiation of the Greek local population to test specific hypotheses about the mechanism for the evolution of differentiation there. Through comparison with earlier studies, we test the hypothesis that apparent marine boundaries to gene flow have differentially affected closely related species, and consider the ecological, environmental or evolutionary factors that may explain the differential degrees of gene flow across shared boundaries for taxonomically similar species.

Materials and methods

Sample collection

A total of 515 samples were collected from stranded, bycaught and biopsied animals (Portugal and Ionian Sea only) representing most of the European range of the species, namely Scotland, Ireland, England, Galicia, six locations along the Portuguese coast, Strait of Gibraltar, Ligurian Sea and two locations in Greece (Kalamos and Korinthiakos Gulf), together with an outgroup sample from Madeira (Fig. 1a). Samples from Scotland, Galicia, Madeira, Gibraltar, Ionian Sea (Natoli et al., 2006, 2008) and Ireland (Mirimin et al., 2009) were analysed previously for a smaller number of loci. All other samples were analysed for the first time in this study. Biopsy samples from Portugal were obtained from six locations along the coast (Fig. 1b) using either: a pole loosely based on the design described in (Bilgmann et al., 2007) or a veterinary air rifle (Pneudart Inc., Williamsport, PA, USA) firing biopsy collection darts with 1.5 cm tip length. Sampling was carried out under a permit from the Portuguese Institute for the Conservation of Nature and Biodiversity (ICNB). Samples from the Ionian Sea were also obtained through biopsy sampling, as described in Natoli et al. (2008).

Figure 1.

Geographical locations of samples used in this study. Number of samples is represented in brackets. (a) Each point represents a mean geographical location of all samples from a given region. Shaded areas represent current known coastal geographic range of common dolphin distribution in Europe based on (Hammond et al., 2002; Bearzi et al., 2003; Evans & Hammond, 2004; Freitas et al., 2004; Brereton et al., 2009). Nonshaded areas represent regions where common dolphin is either rare or absent, or no data is available. (b) Biopsy samples collected along the Portuguese coast.

Laboratory procedures

DNA was extracted using a standard phenol–chloroform protocol (Hoelzel, 1998). All individuals were screened de novo for 15 microsatellite loci (Table S1), with amplification made using two multiplex reactions: set A, amplifying 5 microsatellites; and set B amplifying 10 microsatellites (Table S1). The Qiagen Multiplex PCR kit was used with the conditions: 95 °C for 15 min, 40 cycles at 50 °C (Set A)/57 °C (Set B) for 90 s, 72 °C for 1 min and a final extension at 60 °C for 30 min. All reactions were carried out using 30 ng of the sample DNA. Genotyping was carried out through capillary electrophoresis on an ABI 3130 automated sequencer using the ROX500 size marker and scored using the software provided by the manufacturer. A subset of 73 samples (14.8% of total samples) was replicated for all loci, including samples from both biopsies and strandings.

Data analysis

GenAlex (Peakall & Smouse, 2006) was used to identify matching genotypes based on the 15 microsatellite loci (PID for all 15 microsatellite loci was bellow 5 × 10−12 for all sampling locations). All duplicate samples were removed from the analysis, as were samples that only differed in up to two loci. This resulted in 492 unique individuals being analysed from 15 different locations along the European coast (Fig. 1). The software Powsim (Ryman & Palm, 2006) was used to assess the power of our data set to identify subtle population structure (based on the chi-squared test). Ten thousand simulations were made for a data set composed of 15 populations with sampling numbers based on those obtained in our sampling effort.

Exact tests of Hardy–Weinberg equilibrium and linkage disequilibrium were carried out in the software GenePop (Rousset, 2008) using the MCMC method with 10 000 dememorization steps followed 1000 batches with 10 000 iterations per batch. Pairwise FST between sampling locations was calculated using the software MicroSatelliteAnalyser (MSA) (Dieringer & Schlötterer, 2003). Significance of the pairwise FST values was calculated by 1000 individual permutations with Bonferroni correction. Individual-based clustering without a priori information on sampling location was done using three algorithms. Structure and BAPS employ different Bayesian clustering algorithms that minimize deviations from Hardy–Weinberg equilibrium in each group (Pritchard et al., 2000; Corander & Marttinen, 2006; Corander et al., 2008). Because both algorithms can give biologically implausible results when FST values are low (Latch et al., 2006), they were used as independent confirmation for any patterns obtained. Given that patterns of population structure can also result from location-specific patterns of inbreeding (Gao et al., 2007), we used the Instruct algorithm that is reportedly more accurate in defining population structure in such scenarios (Gao et al., 2007). BAPS was run with four independent chains for maximum number of K between 2 and 15. Such a procedure is suggested by the authors to avoid incorrect identification of local optima (Corander et al., 2008). Structure was run using the admixture model with correlated allele frequencies for 1 × 107 iterations, after 1 × 106 burn-in for K values between 1 and 15, and four independent runs. We ran Structure with and without the locprior option and include the estimation of ΔK (after Evanno et al., 2005). Instruct was run with five simultaneous chains for a total of 2 × 107 iterations after 1 × 107 burnin iterations. The joint inference of population selfing rates and population substructure was used, with K set from 1 to 10. A uniform distribution was assumed as prior for the selfing rates, and convergence among chains was assessed using the Gelman–Rubin statistic.

Genetic diversity was compared between the identified populations using several diversity indices. These included observed, expected and unbiased expected heterozygosity, all calculated with GenAlex (Peakall & Smouse, 2006). Shannon's information index and Allelic richness, which account for differences in sample size were calculated with GenAlex (Peakall & Smouse, 2006) and Fstat (Goudet, 2001), respectively. Permutation tests of individual loci between populations (10 000 permutations) were carried out to assess statistical significance of differences in diversity statistics using the software package RT (Manly, 2007). FIS was calculated using the program GDA (Lewis & Zaykin, 2001) to check for differences in inbreeding. Significance of FIS was assessed by bootstrapping across loci with 10 000 iterations. Isolation by distance was tested using both a Mantel Test and Spatial Autocorrelation using the software Alleles in Space (Miller, 2005). The presence of null alleles was assessed using Micro-Checker (van Oosterhout et al., 2004). Because stranded samples can result in less reliable data due to lower-quality DNA, we analysed stranded and biopsy samples separately. Data sets were further separated according to identified populations.

The software Kingroup (Konovalov et al., 2004) was used to estimate patterns of kinship and assess differences in the pattern of individual relatedness between locations. Pairwise relatedness between all individuals was calculated using the (Lynch & Ritland, 1999) relatedness index. Furthermore, individuals were clustered into hierarchical kinship groups using the Full Sibship Reconstruction (FSR) with decreasing ratios method for the full data set. The proportion of kinship groups relative to the number of samples was calculated for each location as 1−(Kc/N), where Kc = number of groups in each kinship class and N = number of samples.

Assessing bottleneck signatures was done using the software Bottleneck (Cornuet & Luikart, 1996), with significance tests done for all mutation models with 1000 iterations, and with 70% proportion of the infinite allele model (IAM) for the two-phased model (TPM). A test for mode shift in allele frequencies was also carried out. The modified Garza-Williamson M-ratio test (Garza & Williamson, 2001) was calculated using the software Arlequin (Excoffier et al., 2005). Approximate Bayesian computation (ABC), as implemented in the software package DIYABC (Cornuet et al., 2008), was used to investigate different demographic scenarios that might explain the patterns of population differentiation identified in previous analyses. In ABC, several competing population scenarios are simulated, and statistical tests are then used to assess which scenario better fits the observed data. Four different scenarios were tested (Fig. 2), representing: (1) simple divergence with the ancestral population effective size (Ne) being the sum of Ne for the daughter populations; (2) divergence at time t2 in the past with one of the populations experiencing a change in Ne later at t1 (representing a bottleneck event in that lineage); (3) divergence occurring simultaneously with a change in Ne in one of the populations (representing a founder event); and (4) divergence occurring simultaneously with a reduction in Ne in one of the populations that then experiences another change in Ne. These parameters were based on population units identified in the previous analyses and on results obtained from demographic history analyses (see 'Results' for more details).

Figure 2.

Scenarios used to simulated data sets in DIYABC. Scenario 1 represents simple divergence; Scenario 2 represents divergence followed by a change in the size of one population; Scenario 3 represents divergence with simultaneous change in population size; Scenario 4 represents a scenario where divergence occurs due to a founder event. See text for more details.

Two sets of simulations were made with 1 × 106 data sets for each scenario. In one set, uniform priors were used for all summary statistics, whereas in the other Ne and different event timings were constrained according to Table 1. Two levels of constraints were applied differing only in the Ne priors used. The fit of each simulation to the observed data was assessed through a principal component analysis (PCA) and best fit assessed using the logistic regression method. Estimates for different Ne and event timings were done by averaging the parameter values obtained in the 6000 simulated data sets closest to the observed data (determined by logistical regression) for the best fitting scenario, following guidelines distributed with the software package.

Table 1. Prior distribution of the several parameters modelled in DIYABC for the simulation where the priors were constrained to fit particular scenarios. Distributions were uniform
LevelParameterMinimumMaximumAdditional condition
1Ne110100< Ne2
1Ne21010 000> Ne1
1Ne31010 000 
1t0110t1
1t11100t0; < t2
1t2110 000t1
2Ne1101000< Ne2
2Ne210100 000> Ne1
2Ne310100 000 
2t0110t1
2t11100t0; < t2
2t2110 000t1

Results

Tests for Hardy–Weinberg equilibrium showed no consistent pattern among sampling locations or significant deviation after Bonferroni correction (Table S2), and the overall mean (±SE) was very similar for Ho and He (0.706 ± 0.037, 0.709 ± 0.038, respectively). There was no significant linkage disequilibrium among loci. Microchecker showed no evidence of null alleles for biopsy samples from both identified populations (Greece vs. Europe; Table S3). Stranded samples from Europe (all Greek samples were biopsies) showed evidence for null alleles for five loci. However, the estimated frequency of null alleles was low for each locus and given that biopsy samples showed no evidence of null alleles, we interpret this result as being the result of allelic dropout due to lower-quality DNA obtained from stranding samples. As allelic dropout would be expected to distort allele frequencies and increase the chance of finding a significant difference and as there were no similar problems detected for the one population that showed differentiation (Greece), we retained analysis of all loci.

Power analysis indicated that we would be able to detect an FST of 0.01 with 100% confidence, 0.001 with 60% confidence and 0.0001 with 13% confidence. Pairwise FST comparisons showed no significant differentiation after Bonferroni correction apart from comparisons with Greece (Table S4). Locations with n < 10 were not included in the analysis, and any inference from the three locations with n < 20 may be affected by sampling effects. Pooling all samples from the Portuguese coastal region into a single population also showed no significant differentiation for pairwise comparisons with the remaining populations apart from Greece (compared to Greece, FST = 0.039, P = 0.0045; compared to all other locations, FST ranged from −0.00011 to 0.0056, no significance). FST values involving Greece (0.032–0.048) are comparable to that found for other dolphin species across this same boundary: 0.032–0.048 in this study; 0.045 for the bottlenose dolphin (Natoli et al., 2005); and 0.0047–0.0161 for the striped dolphin (Gaspari et al., 2007b). The potential for comparative inference is limited by some differences in methodology; however, the work was done in the same laboratory and these studies used a number of the same loci. Neither the Mantel test nor spatial autocorrelation showed any relationship between genetic and geographic distance (Mantel test: r = 0.029, P = 0.998; Spatial autocorrelation: v = 0.014, P = 0.994).

The most likely number of clusters identified by Structure was K = 1, without the locprior option. Using locprior Structure identified two populations, Greece and the rest of Europe (Fig. 3; S1). BAPS gave a posterior probability of K9 = 0.99 for the individual-based algorithm and K2 = 1 for the group-based algorithm. BAPS assesses the changes in marginal likelihood when individuals or groups are moved among clusters. These changes were small for the individual-based run, but substantial for the group-based run, where Greece is clearly separated from the rest. Instruct gave a most likely K = 10, with an Ln/K plot corresponding to a convex curve of increasing Ln with increasing K (Fig. S1). The only consistent pattern for those analyses showing structure (BAPS, Instruct and Structure Locprior) is a division between Greece and all other European locations (Fig. 3). The patterns obtained for the rest of Europe from BAPS and Instruct were inconsistent, not well supported and likely an artefact, as is common when FST values are lower than 0.02 (Corander & Marttinen, 2006; Latch et al., 2006) as for our data set. Given that evidence for structure only exists for Greece in comparison with other sampled populations, genetic diversity statistics were calculated for Greece and for all other European locations pooled. Although means of all diversity measures are apparently lower in the Greek population relative to Europe (Table S5), the differences were not significant. FIS in Greece (−0.0045) is not significantly different from zero, although the value for Europe is (0.0349; 95% C.I.: 00147–0.0564). However, the two values were not significantly different from each other at the 95% confidence level (Mean diff = 0.0448; P-value = 0.0623).

Figure 3.

Ancestry plots for individual-based analysis. (a) Ancestry plot obtained using Structure with the locprior for K = 2; (b) Clustering of individuals plot for K = 9 obtained using BAPS; (c) Clustering of groups for K = 2 obtained using BAPS; (d) Ancestry plot for K = 10 obtained using Instruct. K.G. indicates the three individual from Korinthiakos Gulf in Greece. All plots drawn using the software Distruct (Rosenberg, 2004).

Mean pairwise relatedness is higher among Greek samples (rmean = 0.1071, SD = 0.1583) than among samples from elsewhere in Europe (rmean = −0.0016, SD = 0.0862). The consistency of this pattern among pairwise comparisons is illustrated in a heatmap for samples ranging from southern Portugal to the Adriatic (Fig. 4a). The same pattern was seen for the rest of the sample range in Europe when all samples were included (data not shown). From the Kingroup analysis, there is a positive correlation between sample size and kinship class, but Greece is a clear outlier (Fig. 4b). The small sample (n = 3) of individuals sampled in the Korinthiakos Gulf did not group with other Greek samples in the heatmap (Fig. 4) and also showed an anomalous signature from the Instruct profile (see Fig. 3); however, the latter may be an artefact as K = 10 was not a credible outcome. Previous studies on common dolphin-mating strategy and social structure all indicate promiscuous mating, with kin associations beyond the age of maturity being rare (Murphy et al., 2005; Westgate & Read, 2007; Viricel et al., 2008). Samples in Greece were collected by biopsy sampling over an extended period of time and without including multiple samples from animals seen together at the time of sampling. Therefore, even if the Greek dolphins exceptionally associated as kin, our sampling protocol did not sample associating individuals, making biased sampling a very unlikely explanation for the observed pattern.

Figure 4.

(a) Heatmap of pairwise relatedeness between samples collected from Portimao (Southern Portugal) through to the Strait of Gibraltar and Ligurian Sea (‘West Meditteranean’) and off Greece. The last three samples are from the Korinthiakos Gulf in Greece. (b) Graph showing the proportion of individuals classified by the Full Sibship Reconstruction Analysis at the second-order level of relatedness (‘cousins’; y-axis). The regression shown is highly significant (P = 3.5 × 10−6).

The program Bottleneck showed statistical support for a bottleneck in Greece for all mutation models (Table 2), as well as a mode shift in allele frequencies (Fig. S2-Top). For all the European samples pooled, all tests using the infinite alleles (IAM) and two-phased (TPM) mutation models showed a statistically significant increase in heterozygosity indicative of a bottleneck; however, none of the tests using the stepwise mutation model (SMM) were significant, and allele frequencies showed no mode shift (Fig. S2-Bottom). The M-ratio showed a very low value for the Greek population (M = 0.25); however, it was also low for the rest of Europe (M = 0.41; Garza & Williamson, 2001 suggest a threshold of 0.68 or lower to indicate a likely bottleneck). In the DIYABC analysis, the PCA revealed that simulations with the unconstrained priors had a worse fit than the constrained priors at the lower levels (level 1; see Table 1), but similar to constrained priors at the higher level (Table 1, Figs S3, S4 and S5). Most of the remaining analyses therefore focus on scenarios based on the constrained simulations at level 1. The logistic regression revealed that scenario 2 fit the observed data best (Fig. S6), thus supporting the idea of a bottleneck in the Greek population rather than a founder event (scenario 3). The time of the bottleneck (t1) was estimated as occurring recently at a median of 11.6 generations ago, although initial divergence (t2) was estimated to have occurred earlier at a median of 317 generations in the past. Given an estimated generation time of 14 years (Taylor et al., 2007) and the 95% confidence limits (see Table 3), this would suggest a bottleneck approximately 58–317 years ago following an initial divergence 1190–15 400 years ago.

Table 2. Significance test results from the software Bottleneck
PopulationMutation modelExpected Het. Exc.Observed Het. Exc. P Stand. diff.Wilcoxon H. Def.Wilcoxon H. Exc.Wilcoxon two-tail
GreeceI.A.M.8.05150.000080.000001.00.000020.00003
T.P.M.8.51140.002460.000190.999980.000030.00006
S.M.M.8.57130.015690.005340.999340.000840.00168
EuropeI.A.M.7.49150.000020.000001.00.000020.00003
T.P.M.8.26150.000110.000011.00.000020.00003
S.M.M.8.4890.499920.072640.873810.138430.27686
Table 3. DIYABC parameter estimate using the 6000 data sets simulated under scenario 2 at different levels of constraint (see Table 1). t1 and t2 are represented in generations
Constraint levelParameterMeanMedianMode0.5 percentile0.95 percentile
1Ne167.170.593.423.797.6
1Ne281008370958054809850
1t112.611.611.14.1622.7
1t2440317172851100
2Ne1621645697201960
2Ne2928078806580237020 400
2t148.34735.412.888.4
2t239521967461200

Although the PCA analyses favour the runs with constrained priors at level 1, the posterior parameter estimates sometimes exceeded the upper bound of the prior range (Fig. S7). The posteriors derived using priors constrained at level 2 were always within the prior range and suggested larger values for Ne, but still within a credible range (Table 3, Fig. S7).

Discussion

Common dolphin genetic structure in Europe

Apart from differentiation between the eastern and western Mediterranean Sea described previously for the common dolphin (Natoli et al., 2006, 2008), we found no evidence of structure in European waters, despite having the power to detect small values of FST and sample sets from approximately every 400 km between the Hebrides and the Alboran Sea. These data address earlier concerns that there may be cryptic population structure that remained masked by low power or incomplete sampling (e.g. Amaral et al., 2007; Natoli et al., 2008; Mirimin et al., 2009). Furthermore, although differentiation between Greece and other Mediterranean sites was found, the data suggest that this was driven especially by a recent bottleneck. Both the M-ratio and Bottleneck tests found evidence for a bottleneck signal in both the broader European population and in Greece; however, the signal for Greece was stronger (e.g. Fig. S2). Although standard measures of diversity were not significantly different, all were lower for Greece and kinship analyses showed that Greek individuals are more closely related to each other than individuals in other locations (see Fig. 4).

Evidence for a strong, recent demographic event in Greece stems from our ABC simulations, where data sets simulated under the assumption of a recent strong reduction in one of the populations (Figs S4 and S5) fit the data better than scenarios where no such assumptions are made (Fig. S3). The best supported scenario (Scenario 2 in Fig. 2) suggests a division beginning early in the Holocene, followed by a more recent bottleneck affecting the Greek population, likely associated with further differentiation. Accurate estimation of t in ABC can be problematic, particularly for recent events (< 100 generations) and when using only microsatellite loci (Cornuet et al., 2010). However, the recent dates obtained for a bottleneck by this method are consistent with the other data on kinship and diversity. Based on ABC, the timing for this event was estimated to have been within 50 generations for either set of constrained priors (see t1 in Table 3), suggesting possible anthropogenic influences. Episodes of population depletion have been associated with the development of population structure in previous studies of cetaceans (Pichler et al., 2001; Nichols et al., 2007). The Greek sample size is small, but the assignment and coalescent methods used are relatively robust to problems associated with sampling effects and low power in small samples.

Independent data on Mediterranean common dolphin abundance

The genetic evidence suggesting a bottleneck for the Greek population is consistent with independent data obtained from the Mediterranean common dolphin, where a population reduction has been reported. The common dolphin is thought to have been abundant throughout the Mediterranean Sea (Forcada & Hammond, 1998; Bearzi et al., 2003, 2004). However, recent evidence suggests a continuing population decline. Both strandings and sighting records have been declining since the 1970s in various Mediterranean locations (Notarbartolo-di-Sciara et al., 1993; Bearzi et al., 2003, 2004). During extensive population surveys of common dolphin in the Ionian Sea, the mean encounter rate has steadily dropped to extremely low levels in only 7 years (Bearzi et al., 2005, 2006). Our results show that the perceived decline is clearly reflected in the genetic data. Although the cause for the decline is not known, various anthropogenic causes for have been suggested (Bearzi et al., 2003), with one study in the Ionian sea noting that it coincided with a decline in fisheries landings of sardine and anchovy during the past 40 years (in spite of increasing catch effort). The fact that tuna and swordfish also showed similar declines suggest that overfishing of prey resources might be one of the main factors promoting the common dolphin bottleneck (Bearzi et al., 2006). Modelling studies have showed that common dolphins appear to be strongly associated with high concentrations of pelagic schooling fish (Cañadas & Hammond, 2008; Moura et al., 2012), suggesting that they may be particularly sensitive to overfishing on this particular type of fish prey.

Comparison with other cetacean species in Europe

Apart from the apparent boundary near Greece, the European common dolphin shows a remarkable lack of population structure when compared to other European cetaceans (Garcia-Martinez et al., 1999; Natoli et al., 2005; Fontaine et al., 2007; Gaspari et al., 2007a,b; Banguera-Hinestroza et al., 2010), or common dolphins and other delphinid species elsewhere (see Hoelzel, 2009; Möller et al., 2011; Amaral et al., 2012). The geographic range extending from the south of Portugal to the north of Scotland has shown significant population structure for a number of cetacean species, and this is especially well established for the bottlenose dolphin (Natoli et al., 2005; Nichols et al., 2007) and harbour porpoise (Fontaine et al., 2007). Our sampling was robust over this region, including 433 samples (see Fig. 1), but no structure was found. The one boundary found for the common dolphin between the eastern and western Mediterranean basins is also an apparent boundary for various other species including dolphins (see review in Natoli et al., 2004). However, for common dolphins, this may be primarily driven by recent demographic changes, possibly due to anthropogenic interference or other recent regional environmental change.

The two most plausible explanations for the atypical lack of structure for D. delphis in European waters have to do with either demographics or dispersion. A recent invasion of the broader region by this species, or recent population expansion and re-colonization of different regions could explain the lack of structure. However, this would need to be an event that was unique to common dolphins and not affecting the various other regional cetacean species that do show evidence of population structure. Furthermore, although there is a signal for a bottleneck event, our estimates of diversity and Ne suggest a relatively large population, and earlier studies looking at mtDNA found only mixed evidence for a population expansion affecting the broader region (e.g. no significant values for Tajima's D; Natoli et al., 2006, 2008). Therefore, the more parsimonious explanation may be that there is a difference in dispersion behaviour among European populations of these species.

The implication is that patterns of differentiation among European cetacean populations are dependent on species-specific factors related to their biology and ecology, as opposed to a physical boundary with the same effect on all species with a similar life history. Although several studies have found a correlation between environmental differences and genetic breaks in cetaceans (e.g. Natoli et al., 2006; Fontaine et al., 2007; Mendez et al., 2010; Amaral et al., 2012), the exact causative mechanisms have remained elusive. For the common dolphin, the diversity of prey species consumed (Amaral et al., 2012) or habitat features (Natoli et al., 2008; Amaral et al., 2012) have been suggested to promote resource specialization, which may in turn promote population structure (see Hoelzel, 2009). However, for our study populations, no population structure can be found throughout most of the European range, in spite of observed prey differences along the Atlantic coast (Silva, 1999; de Pierrepont et al., 2005; Pusineri et al., 2007; Meynier et al., 2008). Furthermore, the differentiation found in the Eastern Mediterranean may instead be explained by an extreme demographic event.

Habitat modelling studies have suggested that common dolphin distribution and abundance is strongly dependent on small pelagic schooling fish (Cañadas & Hammond, 2008; Moura et al., 2012), with the exact species depending on the most energetic option available locally (Spitz et al., 2010). This suggests that the common dolphin in European waters is an ecological specialist on high-energy pelagic schooling prey (with flexibility with respect to the species chosen), following the annual variation in the distribution of this type of prey (Moura et al., 2012). This foraging strategy has been observed for common dolphins in the North Atlantic (e.g. Silva et al., 2002) and may be expected to promote dispersion and fluid social structures if prey choice is opportunistic and the resource is widely distributed (Murphy et al., 2005; Westgate & Read, 2007; Viricel et al., 2008).

In contrast, the bottlenose dolphin (Tursiops truncatus), for example, exhibits a much wider range of ecologies, as indicated by the different foraging strategies observed (see discussion in Moura et al., 2012), and exhibits a marked pattern of population structure in Europe (Natoli et al., 2005) and elsewhere (e.g. Hoelzel et al., 1998; Urian et al., 2009). In Europe, information on bottlenose dolphin diet is limited, but the available studies suggest a reliance on demersal prey, with local variation requiring distinct foraging strategies (Blanco et al., 2001; Santos et al., 2001; Hastie et al., 2006; Spitz et al., 2006a; dos Santos et al., 2007) and a strong association between distribution and foraging areas (Hastie et al., 2004). This may suggest a foraging strategy requiring tighter social structure for a more effective transmission of different complex strategies and perhaps a greater dependence on learning. Some characteristics of the striped dolphin (Stenella coeruleoalba) are similar to common dolphins in European waters with respect to group size and habitat use; however, they also appear to have different diets in this region (see Spitz et al., 2006b; Pusineri et al., 2007), affiliations among adult female kin (Gaspari et al., 2007b) and population structure across this geographic range (Gaspari et al., 2007b).

Foraging strategies in marine mammals show some dependence on habitat characteristics (e.g. Sargeant et al., 2007; Tyne et al., 2012) and have been shown in some cases to be transmitted from mother to calf (Estes et al., 2003; Krützen et al., 2005; Sargeant & Mann, 2009). Once adopted they are unlikely to be abandoned when there are high energy costs associated with switching (Estes et al., 2003). As a general case, ecological niche specialization coupled with cultural transmission of energetically costly foraging strategies may be what is commonly promoting population structure in cetaceans. Perhaps, the best studied example is that of the killer whale (Orcinus orca) where population structure is found to be strongly associated with specialization on prey resources requiring different hunting strategies and strong matrilineal social structure (Hoelzel et al., 2007; Pilot et al., 2010). If common dolphins in European waters instead adopt a different strategy, more prone to prey switching, this may help explain the lack of population structure. However, further comparative data providing greater details on prey choice among populations across the worldwide distribution of this species would be required to test these ideas further.

Conclusions

Population differentiation caused by biological rather than physical barriers is well described in the literature. However, the scope and range of the relevant mechanisms are still largely unknown. Adaptation to different environments is known to cause assortative mating, particularly in cases where the individual first chooses and then mates in a given habitat (Rice & Hostert, 1993; Via, 2001). For highly social top predators like dolphins, philopatry may be driven by the investment required to exploit local resources efficiently, perhaps reinforced by social traditions (see Hoelzel, 2009 for further discussion). This suggests that the fundamental difference for European common dolphins may be in a highly specialized ecological niche choice that promotes social promiscuity and dispersal over long distances, although further studies will be required to test this hypothesis.

Acknowledgments

We thank Bob Reid, Nick Tregenza, Alex Aguilar, Luís Freitas, Willy Dabin, Georgios Gkafas, the ‘Coordinadora para o Estudo de Mamiferos Mariños’ (CEMMA) and the Tethys Research Institute for providing us with samples. Samples in Ireland were collected with the help of the Irish Whale and Dolphin Group and through dedicated bycatch observer programmes (BY-CARE). Malgorzata Pilot has provided invaluable support with the kinship analysis and comments to earlier versions of the manuscript. This study was funded by the Portuguese Fundação para a Ciência e Tecnologia through the PhD grant SFRH/BD/28012/2006 awarded to AEM. AEM would also like to acknowledge the companies Marina de Portimão, Nautiradar and AngelPilot for providing excellent logistical support during sampling collection efforts in Portugal, and all the tremendous effort put in by all the volunteer crew on board Clavadel, the names of which are too many to include here. This work would have not been possible without their generous contribution and tireless dedication.

Ancillary