Study area

The central coast of BC is a remote network of islands and naturally fragmented mainland landmasses with limited (but increasing) industrial anthropogenic disturbance. The area is characterized by a wet and temperate climate, and annual precipitation typically exceeds 350 cm (Darimont and Paquet [17]). A core area (~2000 km²) centered on Bella Bella (52°10’ N, 128° 09’ W) served as the location for this study (Additional file 1). This landscape is surrounded by ocean, which separates a mainland landmass (823 km²) and five main islands ranging in size from 150–250 km². Distances from island to mainland range from 250 m to 1450 m. Observational and genetic data (Darimont et al. [19]; Navid [23]) suggest that wolf packs, defined by the multi-year association of genetically and morphologically distinct individuals, have either island or mainland home ranges. However, one group (Yeo-Coldwell [YC]) primarily uses island habitat but also a portion of the adjacent mainland. Other units are either mainland groups (Upper Roscoe [UR], Lower Roscoe [LR] or island groups (Cunningham-Chatfield [CC], Denny-Campbell [DC]). Moreover, wolves are commonly observed swimming among landmasses, and home ranges of social groups often include multiple islands or mainland landmasses (e.g. peninsulas; Paquet et al. [18]; Darimont [24]; McAllister and Darimont [25]).

Sampling

One thousand and seventy-four (1074) wolf faecal samples were collected between winter 2003 and winter 2004. We collected the following number of samples per season: spring: n = 416 summer: n = 297 fall: n = 292 winter: n = 69. Sampling areas included wildlife trails, logging roads, and electrical power rights-of-way. We preserved each sample in a 50-ml Falcon tube with 95% ethanol. We selected samples for genetic analysis based on characteristics of the samples and collection sites that best predicted amplification success (minimal physical decay, high moisture content, canopy cover; Navid [23]). We extracted DNA from faecal samples with Qiagen QIAamp® DNA Stool Mini Kits and the ‘Protocol for isolation of DNA from larger amounts of stool’ (QIAamp® DNA Stool Mini Kit handbook, http://www.qiagen.com/literature/). We performed DNA extractions in a room physically separated from amplified PCR products and used exclusively for this study to reduce the risk of contamination. Final purified extracts were refrigerated at +4°C until use.

Microsatellite amplification

We amplified a panel of 14 microsatellite markers (13 autosomal and one Y chromosome marker). These were FH2001, FH2010, FH2017, FH2054, FH2088, FH2096, FH2422 (Breen et al. [26]), FH3313, FH3725 (Guyon et al. [27]), PEZ06, PEZ08, PEZ15, PEZ19 (Halverson J. in Neff et al. [28]), and the Y-chromosome marker MS41B (Sundquist et al. [29]). We genotyped n = 477 faecal samples. Polymerase chain reaction (PCR) conditions optimized for the markers, based on the Qiagen multiplexing kit, were: initial denaturation at 95°C for 15 minutes, then 35 cycles of denaturation at 94°C for 30 sec, annealing at 58°C for 90 sec, extension at 72°C for 60 sec, with final extension at 60°C for 30 min. Organisation of markers into multiplexes is shown in Additional file 2. Amplified PCR product was loaded into a 6.5% denaturing polyacrylamide gel, and run on a LICOR4300s DNA analyzer. Genotyping was done with a LICOR’s SAGA GT version 3.3 microsatellite analysis software.

We accepted for further analyses samples that amplified at least 9/14 loci, and used the Excel Microsatellite Toolkit (Park [30]) to test for the presence of matching profiles. We consolidated matches (i.e., profiles with ≥ 75% matching alleles to account for uncertainty in genotyping) into one profile and retained the profile with the highest amplification rate. Matches were tested across all samples, but only observed within wolf groups and 33 profiles were removed.

N = 116 individual profiles were identified. Only 28 wolves (24%) were identified as males, and we observed six alleles for marker MS41B (209, 211, 213, 217, 219, 221). We used MICRO-CHECKER 2.2.3 (van Oosterhout et al. [31]) to assess possibilities of null alleles, large allele dropout, and scoring errors due to stutter peaks. We repeated genotyping of 50 samples collected during the fall season (deemed to represent 50 different individuals from all four groups [CC, DC, LR, YC] based on the abovementioned criteria) to evaluate data quality and estimate genotyping error. Here we estimated per-locus error rates based on the percentage of loci that did not show the same result twice (Additional file 3). Loci for which we obtained the same results twice were accepted as duplicated loci. From these results we obtained duplicate genotypes comprising five or more loci for 18 individuals (i.e. every locus in each of these 18 genotypes provided consistent results when re-tested). Based on amplification and error rates (Additional file 3) we removed MS41B, PEZ08, FH2017, and FH3313 from further analyses.

Statistical analyses

We calculated allelic diversity and observed and expected heterozygosity (with correction for sample size bias; Nei [32]) per locus in GENEPOP 3.4 (Raymond and Rousset [33]) and Genetix 4.05.2 (Belkhir et al. [34]), and F_IS according to Weir and Cockerham [35], for mainland and island wolves. We tested for departures from Hardy-Weinberg equilibrium per locus in GENEPOP 3.4 with the Markov chain method (Guo and Thompson [36]). The results were adjusted to account for multiple comparisons with the false discovery rate (FDR, Verhoeven et al. [37]). Subsequently, we performed centered and scaled principal component analyses (PCA) with the adegenet- package (Jombart [38]) in R 2.14.2 (R development Core Team [39]). The PCA approach does not assume genetic equilibrium conditions and is well-suited for identifying spatial patterns such as genetic clines (i.e., gradients rather than separate clusters or complete admixture) that can be difficult to detect (Jombart et al. [40]). We repeated the PCA with the 18 individuals for which we had duplicated genotypes (and thus higher confidence), and all wolves identified as males to confirm the presence of the observed cline. On average, male wolves may disperse longer distances than females and are more likely to join new packs. As our data set comprised relatively few confirmed males, we tested these results separately to check if the island-mainland gradient remained consistent. A high proportion of females in our sample might otherwise have contributed to the observed gradient if females disperse less frequently and/or shorter distances than males. We then performed a spatial PCA (henceforth sPCA; Jombart et al. [40]), which also takes spatial sampling information into account. As multiple samples were at times collected from the same location, we added 100 m of jitter (small amount of noise) to the UTM coordinates. We performed a spatial autocorrelation in GenAlEx (Peakall and Smouse [41] and references therein) to examine the possible existence of isolation-by-distance in our data set. We used distance classes of 5 km to obtain fine-scale results for our study area. Finally, we performed a partial Mantel-test in R with the Vegan package (Oksanen et al. [42]) to examine the relationship between genetic distance and island-mainland habitat type while controlling for geographic distance. This allowed us to test whether there was an effect of habitat type on fine-scale genetic structure after accounting for the effect of geographic distance. Geographic distance and habitat may be co-linear and their effects could be difficult to separate. Consequently, we also examined the relationship between geographic distance and habitat type. For these tests we incorporated co-dominant genotypic and Euclidean geographic distance matrices exported from GenAlEx and a third matrix with island-mainland habitat designations. We used Pearson’s correlation coefficient with n = 999 permutations.

Genetic variation

Allelic diversity and expected heterozygosity for island wolves (6.8, 0.690) and mainland wolves (5.8, 0.63) were relatively high and comparable to values reported for island populations of wolves on the Pacific Coast in southeast Alaska (5, 0.52; Weckworth et al. [10]) and coastal island populations in Arctic Canada (4.2, 0.61; Carmichael et al. [43]). Allelic diversity and expected heterozygosity were somewhat lower for wolves on the mainland portion of our study area, although this might, at least in part, reflect chance effects of our relatively small sample sizes. Comparison with F_IS values from southeast Alaska islands (0.05) and coastal islands in Arctic Canada (0.181) suggest a higher degree of mating among relatives in mainland (0.264) and island (0.211) wolves from coastal BC. However, Carmichael et al. [43] also observed high F_IS values on Victoria Island (0.427, n = 52) and on islands in the High Arctic (0.629, n = 11). Based on the findings from Alaska wolves, continental wolves appear to have higher genetic diversity. We would also expect a similar situation for our study area, as mainland wolves have a wider surrounding area from which to receive immigrants. However, there are known wolf groups on neighbouring islands not included in this study and we cannot exclude the possibility that immigration from these areas may have augmented the diversity in our sample of island wolves.

Non-invasive sampling and genotyping

Allelic dropout in non-invasive sampling (Santini et al. [44]) could, at least in part, explain the lower values for observed heterozygosity, the high number of loci not in Hardy-Weinberg equilibrium, and the positive F_IS values. Our results could also have been influenced by the presence of null alleles. When most of the loci indicate null alleles, however, the MICRO-CHECKER program warns there may not be random mating in the population (panmixia). The PCA and sPCA findings of island-mainland differentiation suggest absence of panmixia in our study area. We therefore believe that island-mainland structure contributed to the frequent reports of null alleles. The study area is difficult to access, and many samples may have been several weeks old and thus affected by exposure to the humid climate (Santini et al. [44]; Navid [23]). Our results are based on analyses of faecal material, where duplicated genotypes were obtained for 15% (18 of 116) individuals. Error rates were high, but we do not expect any consistent bias between areas. Results from the duplicated genotypes accord with the larger dataset, although further sampling and multiple-tube analyses (e.g. Santini et al. [44]) would be necessary for accurate identification of individual wolves and to confirm dispersal events in our study area.

Evolutionary ecology and genetic differentiation between mainland and island wolves

The partial Mantel test showed no significant relationship between genetic distance and island-mainland habitat type when accounting for geographic distance. However, there was a significant correlation between geographic distance and island-mainland habitat type, suggesting that the two matrices are collinear and their effects cannot be differentiated. The spatial autocorrelation indicated negative autocorrelation from approximately 17–45 km. These results appear to contrast with those of Muñoz-Fuentes et al. [13] who reported that geographic distance was unlikely to explain the spatial structure of wolf mtDNA haplotypes in a broader study of coastal and central BC. Wolves are highly capable dispersers able to travel > 70 km/day (Mech and Boitani [45]), and it seems unlikely that geographic distance alone can explain the island-mainland structure suggested by the sPCA. In such a situation, we would expect the spatial autocorrelation results to show consistent (and increasing) negative kinship-values with geographic distance. In contrast, the 45–50 distance class that represents wolves in the northern- and southernmost parts of our study, which are farthest apart in geographic distance, showed positive values (or, more likely, no autocorrelation). Multiple interacting factors, including distance, water, terrain ruggedness etc., may affect genetic structure in our study area. Although it is essential to evaluate the possible influence of physiography on the differences observed between island and mainland wolves, the observed correlation between geographic distance and habitat type combined with the physical complexity of the landscape make it problematic and potentially misleading to use linear distances for estimating wolf movement.

Water barriers between the mainland and islands might restrict dispersal and gene flow. For example, captive wolves released on Coronation Island in Alaska did not swim 900 m to nearby habitat with abundant food (Klein [46]). In our study, we reject this hypothesis because we commonly observe wolves swimming among landmasses and distances among islands (including the multiple landmasses used by some groups) are often larger than the distances between islands and the mainland (Darimont et al., unpublished data). Immigrants from outside the study area could also influence the observed east–west gradient in genetic profiles. The differentiation seen in YC profiles, for example, may be explained by gene flow from unsampled wolves on the outer islands father west. Similarly, profiles from the UR group, which showed considerable overlap with island wolves, might result from immigration by one or more island wolves with high reproductive success. Furthermore, the presence of intermediate profiles in the north and south of our study area implies an island-mainland gradient. A strict island-mainland dichotomy may thus be simplistic and should be evaluated on a broader geographic scale. Without genetic data from a larger spatial extent, however, we cannot evaluate these hypotheses.

Family group structure might also have influenced our results, especially for long-lived animals for which the genetic influence of one successful breeder can be detected for many generations. Difficulties with amplification of MS41B likely reduced our ability to identify male wolves. A possible higher prevalence of females in the sample might nevertheless exacerbate genetic structuring in species where males are more likely to disperse. However, male wolf profiles we assessed showed a similar island-mainland gradient. Observational and tracking data suggest that wolf group size in the study area was ≤ 10 individuals (Darimont [24]), and it appears unlikely that the observed gradient in genetic profiles could be explained by social structure (i.e., wolf pack membership) alone. The identification of 116 individuals in the study area appears reasonable for a sampling period that included 2 litters, winter pup mortality that may exceed 50%, and the likelihood that 20% of individuals would be solitary or extra-territorial dispersers (Mech and Boitani [47]).

Despite the above-mentioned uncertainties, we offer the working hypothesis that the sharp ecological gradient between island and mainland locations, as revealed by the landscape characteristics and the dietary and parasitic data from wolves in our study area, can influence population genetic structure. Although our study must be interpreted with caution, and should be repeated with genetic profiles of higher quality, the results appear consistent with an increasing body of literature reporting genetic differentiation in wolves and other highly mobile species (see Introduction) influenced by ecological and environmental factors. Dispersal rates and gene flow might differ substantially between island and mainland sub-populations, and the extent to which populations are demographically independent could help define management units (Palsbøll et al. [48]) along the Pacific coast.

Associated morphological or other characteristics observed over time might have allowed TEK knowledge holders to recognize these dissimilar wolf forms. Such intra-specific nomenclature is common among indigenous knowledge holders (Turner et al. [49]). Indeed, in adjacent southeast Alaska, the frequency of the black colour phase among wolves killed by trappers is ~50% on the mainland and only ~20% on the islands (Person et al. [50]). Additional morphological differences among wolves of coastal BC might have led to mainland-island classification by local people.

The evolutionary influence of marine resources, which are pronounced on islands in our study area, can be dramatic for terrestrial wildlife. For example, polar bears (U. maritimus Phipps, 1774) are thought to have evolved from grizzly bears in peripheral areas where marine resources were abundant (Shields et al. [51]). Moreover, wolves of coastal BC (mainland and island populations) were thought to have diverged from interior populations in part because of marine resource availability in coastal zones (Muñoz-Fuentes et al. [13]). Individuals born in this distinct environment are likely better able to survive and reproduce within, compared to beyond, these conditions.