Carlos J. De Luna, Centre of Equine and Animal Science, Writtle College, Lordship Road, Writtle, Chelmsford CM1 3RR, UK. Tel.: +44 0 1245 42 42 00; fax: +44 0 1245 42 04 56; e-mail: firstname.lastname@example.org Alan R. Hoelzel, School of Biological and Biomedical Sciences, Durham University, South Road, Durham, DH1 3LE, UK. Tel.: +44 0 191 334 1325; fax: +44 0 191 334 1201; e-mail: email@example.com
Abstract Determining the mechanisms that generate population structure is essential to the understanding of speciation and the evolution of biodiversity. Here, we investigate a geographical range that transects two habitat gradients, the North Sea to North Atlantic transition, and the temperate to subpolar regions. We studied the harbour porpoise (Phocoena phocoena), a small odontocete inhabiting both subpolar and temperate waters. To assess differentiation among putative populations, we measured morphological variation at cranial traits (N = 462 individuals) and variation at eight microsatellite loci for 338 of the same individuals from Norwegian, British and Danish waters. Significant morphological differentiation reflected the size of the buccal cavity. Porpoises forage in relatively shallow waters preying mainly on benthic species in British and Danish waters, and on mesopelagic and pelagic fish off the coast of Norway. We suggest that the observed differentiation may be explained by resource specialization and either adaptation or developmental responses to different local habitats.
Intraspecific differentiation along a contiguous geographical area due to vicariance events or lasting geographical barriers is common (see Wiley, 1988). However, such effects are less common in the marine environment, where movement is typically unrestricted over vast distances for mobile marine species. Nevertheless, some highly mobile marine species show population structuring, sometimes over quite a fine geographical scale, not necessarily associated with isolation by distance (e.g. Goodman, 1998; Bahri-Sfar et al., 2000; Guarniero et al., 2002; Natoli et al., 2005, 2006). In some cases, oceanographic and behavioural characteristics have been proposed to explain strong population differentiation (e.g. Hoelzel et al., 1998, 2007; Naciri et al., 1999; Bekkevold et al., 2005; Pilot et al., 2010). The implication is that prey availability or prey choice sometimes drives dispersal pattern and range, and this may be especially true in social species where cultural transmission could play a role (see Hoelzel et al., 2007).
The harbour porpoise (Phocoena phocoena) is a small odontocete that inhabits subpolar and temperate coastal and continental shelf waters of the Northern Hemisphere. In the eastern North Atlantic, its distribution includes the Barents Sea and west coast of Norway, around the coasts of Iceland, in the North and Celtic Seas, and around Danish waters in the Skagerrak and Kattegat seas. It is considered common in this region where an estimated 340 000 individuals have been reported (Hammond et al., 2002). Possible migration routes include the Kattegat Sea, the English Channel, the Bay of Biscay and the coast of Portugal and north-west Africa (Rosel et al., 1999). They are found in groups of up to 10; though, little is known about any possible social interactions (but see Hoek, 1992).
A number of studies have addressed the question of population structure for this species in the North Atlantic and North Sea using molecular markers. Andersen et al. (2001) compared six regions [inner Danish waters (IDW), the Danish North Sea (DKNS), the British North Sea (BNS), Norway, Ireland and West Greenland] using 12 microsatellite loci and found significant differentiation for all pairwise comparisons for both assignment tests and measures of FST (though the latter were small, ranging from 0.002 to 0.014). Earlier studies found significant differentiation at the mtDNA control region comparing Norway with the BNS (FST = 0.06; Tolley et al., 1999), and the BNS with Ireland (ϕST = 0.002; Walton, 1997). Another study comparing 10 microsatellite DNA loci across a geographical range from the Black Sea to Norway found differentiation between the Black Sea, Spanish coast and the North Sea (Fontaine et al., 2007b). A study focussing on populations in the Danish Sea and the Baltic reported a population genetic boundary to the east of Denmark between the Skagerrak and the Belt Sea (Weimann et al., 2010). Taken together, these studies suggest a pattern of relatively fine-scale population differentiation along the geographical range between the Black Sea and the North Sea similar to that seen for another cetacean species, the bottlenose dolphin (Tursiops truncatus; Natoli et al., 2005; Nichols et al., 2007).
Viaud-Martinez et al. (2007) found cranial differentiation between the harbour porpoise populations in the Black Sea, off the coast of France and in the North Sea, whereas Kompanje & van Leeuwen (2009) described differentiation between skulls from the coast of Africa and those in the North Sea. We tested the hypothesis that phenotypic variation in skull morphology is consistent with differentiation at neutral genetic markers, and with ecological distinctions among habitats within the North Sea and contiguous waters. Here, we focussed on phenotypic variation within the North Sea region, supported by the comparison of the same samples using microsatellite DNA markers. We test the hypothesis that differentiation for phenotype and neutral markers among habitat regions will be correlated, consistent with expectations based on local resource use.
Material and methods
DNA was extracted using a standard phenol/chloroform technique from samples obtained for the majority of the individuals for which skull measurements were also taken (338 of 462 skulls). Samples were screened at eight microsatellite DNA loci: EV104, EV94 (Valsecchi & Amos, 1996), GT011, GT136, GT015 (Bérubéet al., 1998), 417 (Amos et al., 1993), IGF1 (Kirkpatrick, 1992) and TAA031 (Palsbøll et al., 1997) using the reaction conditions described in Andersen et al. (2001). The genotypes for porpoises from the DKNS and IDW are from Andersen et al. (2001). Four of the original 12 loci originally screened by Andersen et al. (2001) did not produce results that could be easily scored for some of the new regional samples and were therefore omitted. Replication among laboratories was carried out for calibration. Microsatellite PCR products were genotyped on a 377 ABI automated sequencer and analysed using ABI Genescan™ and Genotyper™ software (Applied Biosystems, Carlsbad, CA, USA).
A total of 462 skulls were measured from the collection of four European museums: Zoologisk Museum, University of Oslo, Norway; Zoologisk Museum, University of Copenhagen, Denmark; Natural History Museum, London, UK; and National Museums of Scotland, Edinburgh, UK. Fifty were from the Norwegian North Sea–Barents Sea (NOR), 53 from the Danish North Sea–Skagerrak Sea (DKNS), 40 from the IDW, 152 from the BNS and 154 from the Irish Sea (IRL-W; Fig. 1).
Choice of cranial traits
Sixteen bilateral characters (Fig. S1) were chosen for this study following Perrin’s nomenclature (Perrin, 1975). All traits were measured using precision callipers and measured to the nearest 0.001 cm, except to the nearest 0.01 cm for CBL, ML and LOR (see Fig. S1). Three repeated measurements for each trait of every skull were taken, and the callipers were reset to zero after each measurement. The median of the three was used (Zar, 1984). Measurements were taken on the left side of the skull only because of the directional asymmetry present in odontocetes associated with echolocation (Yurick & Gaskin, 1987). No measurements were attempted on missing or worn structures.
Polymorphism was estimated as the number of alleles per locus, number of private alleles per putative population, allelic richness, observed heterozygosity and expected heterozygosity (see Table S1). Deviation from Hardy–Weinberg equilibrium and linkage disequilibrium were tested using an analogue of Fisher’s exact test with a Markov chain method (100 000 iterations, 5000 dememorization steps, sequential Bonferroni correction applied) as described by Guo & Thompson (1992) using arlequin 3.5 (Schneider et al., 2000). FIS and allelic richness were calculated in fstat 2.9.3 (Goudet, 2001). A Kruskal–Wallis test was employed to test for differences in allelic richness among subpopulations, due to the fact that the data did not distribute normally even after transformation.
The software structure (Pritchard et al., 2000) was used to assign individual genotypes to putative populations (K) without prior information on sample origin. The admixture model was assumed and the analysis was performed considering both the independent and the correlated allele frequency models. Burn-in length was set at 500 000 and simulation length at 1 000 000 replications. Each test of K (ranging from 1 to 6) was repeated four times. An assessment of the rate of change in the log probability for successive values of K (ΔK) was also undertaken to detect the uppermost hierarchical level of structure (Evanno et al., 2005). FST (Weir & Cockerham, 1984) and genotypic assignments were calculated using the software arlequin 3.5; statistical significance for FST was calculated by permutation tests with bootstrapping and 1000 iterations.
To test for significant differences in skull morphology between sexes and between age classes within putative populations, all measurements were standardized over the total length of the skull (CBL) to control for the effect of size. This gave a relative ratio for each trait. For each population, a multivariate analysis of variance (manova) was used to find differences in the relative skull characters between age classes and sex. Normality was tested using the Kolomogorov–Smirnov test, and homogeneity of variances assessed using Levene’s test.
A manova was used to test for significant differences in skull characters between putative populations. The manova was extended and a discriminant function analysis (DFA) was performed. DFA was used to classify the porpoises into populations (see Tabachnick & Fidell, 1996). The adequacy of the DFA classification was determined by the percentage of correct classifications, assuming that there was an equal probability by chance (20%) of each skull to being classified into any of the a priori five subsamples (NOR, BNS, IRL-W, DKNS and IDW). Classification percentages substantially > 20% for any given subsample would indicate that the discriminant functions were satisfactory for predicting group membership. The Mahalanobis distance was used to allocate individuals into a population by estimating the distance of the mean vector of each individual from the mean vector of each population. DFA classified each porpoise to a population based on the assumption that the shorter the distance of the mean of a case with respect to the population mean, the higher the probability that the sample belongs to that particular population. Wilks’λ was used to test whether the multivariate means of the groups were significant on each discriminant function (Field, 2009). Pairwise t-tests were performed to compare the dimensions (in centimetres) of specific skull traits between putative populations.
Heterozygote deficiency was observed at just one locus in one population (see Table S1), and its omission did not alter the results (data not shown), so all eight loci were retained (raw data provided through DRYAD doi: 10.5061/dryad.cd168nj1). Each pair of loci was tested for linkage disequilibrium and genotypic independence confirmed. Allelic richness was not significantly different among populations ( = 1.64, P = 0.54).
Figure 2 illustrates the results from the analysis undertaken in structure (correlated allele frequency results shown). There was consistency among different runs for the estimation of P(X|K) and α was stable across the run. K =4 had the highest likelihood for both the independent and the correlated allele frequency models. However, the plot suggests three distinct subpopulations: Norwegian (NOR), British (BRIT – incorporating BNS and IRL-W) and Danish (DK – incorporating DKNS and IDW), with the putative fourth population assigning mostly in the BRIT population. We ran structure again including only the BRIT population samples, and this time the highest likelihood K was K = 1. For the result where K = 4, there was no simple correspondence between the differential assignments in the BRIT sample and the BNS and the IRL-W subpopulations, and no subdivision indicated for the DKNS and the IDW samples. This was reflected in the fixation index analyses where the BNS vs. IRL-W (FST = 0.001) and DKNS vs. IDW (FST = 0.003) samples were not significantly differentiated. All FST values for pairwise comparisons among the remaining three putative populations were significant after Bonferroni correction at P < 0.001 (NOR vs. BRIT, FST = 0.050; NOR vs. DK, FST = 0.046; BRIT vs. DK, FST = 0.040). Assignment to source population (Arlequin method) was 100% for NOR, 99.1% for BRIT and 97.8% for DK. A method that detects the highest level of structure (Evanno et al., 2005) showed strongest support for K = 2 separating the BRIT population from the rest (Fig. S2) when all samples were included in the analysis.
Tests for normality and homogeneity of variance all showed no significant deviation (all P-values > 0.07), indicating that parametric comparisons by analysis of variance were appropriate. The manova did not find significant differences between sexes or age classes, so the data set was pooled for analysis within each population. The manova comparing the three putative populations defined by genetics (NOR, DK and BRIT) showed significant differences for all traits (all P <0.001, Bonferroni correction applied). Due to the fact that the results of the genetic analyses indicated the presence of three putative populations, two DFAs were performed. On the first DFA, we tested the five a priori subsamples (NOR, BNS, IRL-W, DKNS and IDW), and the percentages of successful classifications for each of the five sampling regions are presented in Table 1. The second DFA was performed based on the results of the first DFA where IDW and DKNS samples were classified into one single group, as were the BNS and IRL-W samples (Table 2). Therefore, for our data set, both the genetic and morphometric data supported the division of the sample set into the three population samples (NOR, BRIT and DK). The results below refer to the second DFA.
Table 1. Adequacy of classification results for the first discriminant analysis. Left column indicates the original group, whereas the top row indicates the predicted group. Values are as percentage. Correct classifications are italicized and in bold (sample size in parenthesis).
Table 2. Adequacy of classification results for the second discriminant analysis after reclassification into three main populations. Left column indicates the original group, whereas the top row indicates the predicted group. Values are as percentage. Correct classifications are italicized and in bold (sample size in parenthesis).
Wilks’λ test was significant for DF1 (λ = 0.004, = 116.2, P <0.001), but not for DF2 among the population centroids (λ = 0.368, = 21.0, P >0.05). Table 3 shows the structure matrix of the DFA. DF1 explained 98% of the variance. The population centroid of the Norwegian population showed a value of −15.69 for DF1, which represented the largest degree of separation among the groups (Fig. 3). Table S2 shows the t-test pairwise comparisons for each trait among populations, with WON, OL, ML and maximum width of the palatine (MWP) (see Fig. S1) showing the strongest effects. Although the Norwegian population was most differentiated (at DF1), the British and Danish populations also show separate clusters (at DF2; Fig. 3).
Table 3. Pooled within-subpopulation correlations between discriminating variables and standardized canonical discriminant functions. Variables ordered hierarchically by absolute size of correlation within function. The largest absolute correlations with either discriminant function are shown in bold and italicized.
Assignment methods and FST supported the designation of three genetically differentiated populations; though, the strongest differentiation was between BRIT and the rest (Fig. S2). As described previously, an earlier study based on 12 microsatellite DNA loci (Andersen et al., 2001) had found small but significant differentiation between BNS and IRL-W (FST = 0.005) and between DKNS and IDW (FST = 0.004). However, these comparisons were not significant for our data (possibly due to reduced power), and the result in structure was K = 1 when the BRIT sample was considered on its own. A division into three populations for our data set was also reflected in the clusters based on cranial morphology (Fig. 3); though, in this case the Norwegian sample showed the greatest differentiation.
As shown in the results, the morphological trait that showed the highest correlation with the discriminant function 1 was the MWP. Pairwise comparisons of MWP between the three proposed subpopulations showed that this trait was significantly different among them (Table S2). The palatine bones are situated behind the maxillae and they form the roof of the hard palate. The trait with the next highest correlation to DF1, the length of the maxilla (ML), was also significantly different among subpopulations. The maxillae form the roof of the mouth and they hold the upper teeth. Another trait contributing to DF1 was the length of the mandible (LOR), and it was significantly different in pairwise comparisons between the Norwegian subpopulation and the other two subpopulations.
These three traits are all anatomically involved in forming the oral cavity, and therefore involved in the shape of the snout or beak, and thus physically associated with feeding. The fact that the Norwegian population showed relatively shorter maxillae and mandibles along with narrower palatines means that they possessed proportionally smaller beaks than their British and Danish counterparts. The implication is that there may be local adaptation associated with differences in feeding strategy or prey choice. Consistent with this, Fontaine et al. (2007a) found a clear distinction between the carbon and nitrogen stable isotope ratios for samples from Norway (close to the location of our samples) and Denmark. They suggested that porpoise are taking pelagic prey in the deeper Norwegian waters, compared to more coastal or demersal prey in the relatively shallow waters off Denmark and in the North Sea. They further proposed that shifts over the seasons reflecting local oceanographic conditions suggest a lack of extensive migration (c.f. Tolley & Heldal, 2002).
Analyses of stomach contents (Aarefjord et al., 1995; Börjesson et al., 2003; Lockyer & Kinze, 2003) found that harbour porpoises from the Atlantic mid-coast of Norway preyed mainly on mesopelagic and pelagic fish like capelin (Mallotus villosus), herring (Clupea harengus), saithe (Pollachius virens), haddock (Melanogrammus aeglefinus), blue whiting (Micromesistius poutassou) and greater argentine (Argentina silus), whereas those from the relatively shallow North Sea, Skagerrak Sea and IDW preyed mainly on benthic species like gobids, ammotydids, sprat (Sprattus sprattus), whiting (Merlangius merlangus), cod (Gadus morhua) and Atlantic hagfish (Myxine glutinosa).
Aarefjord et al. (1995) suggested that the difference may be related to access to benthic species, with access possible only in relatively shallow waters. If foraging depth does distinguish these populations, our data on differentiation in orbit length (dominant factor in DF2, and significantly different among regions; Tables 2 and S2) may be relevant if greater orbit size relates to larger eyes and improved vision at depth (Land & Nilsson, 2002). Although DF2 was not strongly supported, the Norwegian population showed significantly larger average OL measurements for pairwise comparisons (see Table S2). The relative importance of vision and echolocation during foraging is not known for this species, but it is likely that vision plays a significant role (see Land & Nilsson, 2002).
The British and Danish populations were also differentiated at traits associated with the overall width of the skull (e.g. mandibular width, MW), together with the size of the orbit (OL; see Table S2). The British population had wider mandibles and smaller orbits on average. The width of the skull may also affect the buccal cavity and feeding strategy; however, the diet of harbour porpoise in UK waters is dominated by benthic species as in Danish waters, especially whiting and sand eels (Ammodytidae; Pierce et al., 2007).
Phenotypic differences between relatively deep compared to shallower water (pelagic vs. coastal) cetacean populations have been investigated previously. A common pattern is for the pelagic phenotype to show a less pronounced beak; though, this is typically seen in delphinid species with much more pronounced beaks in either form than seen in the harbour porpoise (Heyning & Perrin, 1994; Hoelzel et al., 1998; Natoli et al., 2004, 2005, 2006). It is not clear whether there are any functional parallels between beak length in porpoises and dolphins, nor why beak length may confer an advantage in different habitats.
A correlation between intraspecific morphological and genetic differentiation in highly mobile species has been described previously for various taxa, including raptors (e.g. Hull et al., 2008), pinnipeds (e.g. Hoelzel et al., 2001) and dolphins (Hoelzel et al., 1998). However, the correlation is not always strong. Common dolphins (Delphinus delphis) show genetic differentiation between long- and short-beak morphotypes in the eastern tropical Pacific (Rosel et al., 1994), but not for other populations elsewhere in the world (Natoli et al., 2006). Other studies have shown a lack of genetic differentiation between morphologically distinct populations and suggested an important role for phenotypic plasticity in generating differentiation for both plants (Fleischer-Dogley et al., 2010) and animals (Fritz et al., 2007; Zieritz et al., 2010), or suggested a role for both genetics and phenotypic plasticity (e.g. explaining the trophic polymorphisms in pumpkinseed sunfish; Lepomis gibbosus; Robinson & Wilson, 1996).
Differentiation at neutral markers suggests divergence by genetic drift and sufficient time for local adaptation; however, this does not rule out the possibility of phenotypic plasticity. There is some evidence that the population structure of harbour porpoise in the North Atlantic and adjacent waters has evolved relatively recently. Fontaine et al. (2010) suggested that differentiation between the Black Sea and Atlantic populations evolved within the Holocene (approximately 5000 YBP) based on microsatellite DNA markers, whereas Viaud-Martinez et al. (2007) suggested an earlier division, but still since the last glacial maximum (approximately 20 000 YBP) based on mtDNA control region sequence data. Fontaine et al. (2010) further suggested the isolation of porpoises in Iberian waters from those further north only approximately 300 years ago with a predominant northward migration, contemporaneous with the warming trend underway since the ‘Little Ice Age’ period and with the ongoing retreat of cold-water fishes from the Bay of Biscay. Sommer et al. (2008) suggested that the harbour porpoise has been in the Baltic since early in the Holocene based on subfossil records, but Christensen & Richardson (2008) suggest shifts in prey choice (implying a lower trophic position especially after 1960) in the North Sea over the last 150 years based on stable isotope data from samples collected between 1848 and 2002 (though they cannot rule out the possibility that this reflects a change in the isotope signature of nitrogen entering the food web over this period). If a true reflection of changing porpoise diet, it may suggest a relatively recent regional specialization, and potentially a role for a plastic response in cranial morphology.
As seen for some delphinid species in the North Atlantic and adjacent waters, there is a correlation between apparent habitat boundaries and population genetic structure (Natoli et al., 2005; Gaspari et al., 2007). In our harbour porpoise study, the distinction is associated with deeper water habitat off Norway, and a differential choice of prey resource there, and oceanic differences between the IDW and Skagerrak, proximate to the Baltic, and the Scottish and Irish coastal regions more influenced by the North Atlantic. Although these studies do not prove causation, they suggest that the correlation between evident differences in foraging specializations and the evolution of population genetic structure may be promoted by isolation and subsequent drift, differential selection in habitat suited to different specializations, or both (see Hoelzel et al., 2007 for further discussion). Our study strengthens this hypothesis, providing data on associated morphological differentiation relevant to known differences in prey choice.
This study was partially funded by CONACYT (National Council for Science and Technology, Mexico). We thank the authorities of the museums involved in this study for allowing us access to their collections. Specially, we appreciate the help of Richard C. Sabin and Paula Jenkins from the Natural History Museum in London, Jerry Herman and Andrew Kitchener from the National Museums of Scotland in Edinburgh, Carl Christian Kinze from the Zoologisk Museum from the University of Copenhagen and Øystein Wiig from the Zoologisk Museum of the University of Oslo.
Data deposited at Dryad: doi: 10.5061/dryad.cd168nj1