Temporal genetic structure and relatedness in the Tufted Duck Aythya fuligula suggests limited kin association in winter

Sample collection and microsatellite genotyping

Our sampling comprised 191 individuals collected in the winters of 2007–2008 (n = 65) and 2008–2009 (n = 126) from Oberkirch at Lake Sempach, Switzerland (47°09′N, 08°07′E), by J. Hofer. Tufted Ducks were caught using a baited fence trap (see Hofer et al. 2005) and feather samples were taken. The age and sex of each individual was determined based on plumage characters or through examination of the cloaca. Capture, ringing and the plucking of a few body feathers were carried out under licence of the Swiss Federal Office for the Environment (issued to J. Hofer).

Due to the uneven presence of ducks over time, ducks captured within the same 3-week period were assigned to the same temporal group for subsequent analyses. Three such temporal groups were defined in the winter of 2007–2008 (2007-1 to 2007-3) and six in 2008–2009 (2008-1 to 2008-6). The exact time windows and the resulting sample sizes are indicated in Table 1. Tests showed that shifting the temporal windows by 1 week in either direction did not have a significant effect on the observed patterns of genetic diversity, population structure or relatedness (results not shown).

Genomic DNA was extracted using a standard phenol-chloroform extraction protocol (Sambrook et al. 1989). Samples were screened at 11 autosomal microsatellite loci that had been initially isolated in other related duck species; Mallard Anas platyrhynchos: Caud13 (Huang et al. 2006), Apl12, Apl36 (Denk et al. 2004) and Aph13 (Maak et al. 2003); Muscovy Duck Cairina moschata: CmAAT28 (Stai & Hughes 2003); Common Eider Somateria mollissima: Smo4 and Smo11 (Paulus & Tiedemann 2003); Spectacled Eider Somateria fischeri: Sfiμ2 and Sfiμ4 (Fields & Scribner 1997); and Common Merganser Mergus merganser: MM03 and MM05 (Hefti-Gautschi & Koller 2005). The PCR protocols and multiplex details are provided in Stovicek et al. (2011) and Liu et al. (2012).

Genetic diversity

The number of alleles (N_A), allele frequencies, observed (H_O) and expected heterozygosities (H_E) and Wright's inbreeding coefficient (F_IS) for all temporal groups were calculated with arlequin v.3.5 (Excoffier & Lischer 2010). The same program was used to test for deviations from Hardy–Weinberg equilibrium (HWE) and linkage equilibrium. Significance was assessed based on 1 000 000 steps in the Markov chain and 10 000 dememorization steps for HWE and 10 000 permutations for linkage equilibrium. In addition, allelic richness (A_R) was estimated using fstat v.2.9.3.2 (Goudet 2002). Significance levels were adjusted for multiple tests using the sequential Bonferroni procedure (Rice 1989). The frequency of null alleles at each locus was estimated with the program freena for each temporal group separately (Chapuis & Estoup 2007).

Genetic differentiation between temporal groups

Analysis of molecular variance (amova; Excoffier et al. 1992) implemented in arlequin was used to investigate the hierarchical genetic structure among the temporal groups. First, we assessed the amount of genetic variation explained by: (1) differences between winters and (2) differences between temporal groups within each winter. The amovas were also used to detect differences between males and females in each winter with temporal groups within winters pooled (i.e. four groups in total). Next, pairwise F_ST values between all pairs of temporal groups were calculated using the estimator of Weir and Cockerham (1984) in arlequin. Significance was assessed based on 10 000 permutations, with significance levels adjusted for multiple testing using the sequential Bonferroni procedure (Rice 1989).

The genetic substructure was further investigated with structure v.2.3.1 (Pritchard et al. 2000). We assumed an admixture model with correlated allele frequencies (Falush et al. 2003) and used 500 000 Monte Carlo Markov Chain (MCMC) repetitions following a burn-in of 100 000 iterations. The number of groups (K) varied between one and nine, as we had nine temporal groups. For each value of K, we carried out 10 independent runs. The most likely number of groups was assessed with the ad hoc statistic Delta K described by Evanno et al. (2005) and visualized in structure harvester v.0.6.92 (Earl & von Holdt 2012).

Analysis of genetic relatedness

For each pair of individuals we calculated the Queller and Goodnight (1989) estimator of relatedness (R_QG) using genalex v.6.3 (Peakall & Smouse 2006). The genetic relatedness coefficient is defined as the proportion of ancestral alleles that are shared between descendant individuals (Lynch & Walsh 1998). The arithmetic mean and the standard deviation of R_QG were calculated across all individuals within a temporal group using the bootstrap approach with genalex. The bootstrap procedures were repeated with 999 iterations based on the observed allele frequencies at each locus. We tested for significant differences of the observed mean relatedness from a random assortment of individuals (expected R = 0) for each temporal group. The 95% confidence intervals were obtained using 999 permutations at the allelic level with genalex. The same analyses were applied to subsets of males and females separately in each winter but with temporal groups within winters pooled (i.e. four groups in total). Unfortunately, the number of individuals of known age was small (seven first-winter individuals in 2007–2008; 28 in 2008–2009). As a consequence, it was not possible statistically to explore potential differences in the patterns of relatedness between age classes. To determine the number of microsatellite loci that provide stable estimates of pairwise relatedness, we conducted a rarefaction analysis based on R_QG using the online program re-rat (Schwacke et al. 2005). This method allows for the calculation of relatedness coefficients when a given number of loci are available.

We used the program ml-relate (Kalinowski et al. 2006) to calculate maximum likelihood estimates of relatedness (R) and the likelihood of four relatedness categories (unrelated: R = 0; close kin (e.g. half-siblings, aunt–niece): R = 0.25; full-siblings: R = 0.5; parent–offspring: R = 0.5) to determine the proportion of a specific relatedness category among all pairwise comparisons (dyads) in each winter. Furthermore, we applied a likelihood ratio test (at the 95% confidence level, 1000 simulations) using the same program to assess the likelihood of a given relatedness category against the remaining three categories (Kalinowski et al. 2006).

Mantel tests were performed to assess the overall correlation between the logarithm of temporal distances (difference between capture dates of two individuals in days) and relatedness coefficients among all pairs of individuals for each winter separately. The significance of the association was assessed based on 999 permutations in genalex.

Genetic diversity

We analysed samples from 191 individuals in two successive winters (Table 1, Supporting Information Table S1), 135 males, 55 females and one unsexed individual. Genotypes of these individuals were obtained based on 11 microsatellite loci, but locus MM03 was not included in further statistical analyses because it showed significant departures from HWE in most temporal groups (Supporting Information Table S2), probably due to a common null allele (estimated null allele frequency: 5.65–39.53%). The null allele frequencies of the remaining loci ranged from 0.25% to 5.05%. Five of the remaining 90 locus- and sample-specific tests showed significant deviations from HWE after Bonferroni correction, affecting four different loci (Table S2). F_IS within temporal groups was not significantly different from zero except in the group 2008-2 (Table 1). We found no significant year effects in the average allelic richness (Mann–Whitney U-test, P = 0.36). The overall observed and expected heterozygosity were very similar among temporal groups (Table 1). There was no evidence of linkage disequilibrium.

Genetic differentiation between temporal groups

The amova based on all nine temporal samples showed no evidence of genetic differentiation between the two winters (P = 0.96) or among samples within winters (P = 0.87): pairwise F_ST values between all temporal groups ranged from −0.02 to 0.009 with all P-values > 0.05 (Supporting Information Table S3). Furthermore, we did not observe genetic differentiation between males and females in the same winter (P-values in both winters were larger than 0.05, results not shown). structure also detected no evidence of temporal genetic substructure with no support for the K > 1 population (Supporting Information Fig. S1).

Patterns of relatedness

Rarefaction analysis suggested that there was little effect on mean relatedness estimates when more than eight loci were used (Supporting Information Fig. S2), suggesting that our microsatellite markers should produce unbiased estimates. The average pairwise relatedness coefficients within temporal groups ranged from −0.006 to 0.03 (Table 1) and none of these values was significantly different from zero (Fig. 1a). Similarly, none of the average relatedness coefficients was significantly larger than zero if males and females were tested separately in both winters (Fig. 1b).

Pairwise relatedness between individuals decreased significantly with the logarithm of the temporal distance between sampling dates if the analysis was performed for each sex separately (Mantel test: winter 2007–2008, males: r = −0.30, P < 0.001; females: r = −0.16, P = 0.01; winter 2008–2009, males: r = −0.18, P < 0.001; females: r = −0.14, P < 0.001) (Fig. 2), but not if males and females were pooled (Mantel test: r = 0.016, P = 0.13 in 2007–2008; and r = −0.01, P = 0.17 in 2008–2009). The sex-specific analysis remained significant if untransformed temporal distances were used (results not shown).

The maximum likelihood estimates suggested that the signal of kinship within short temporal periods was primarily derived from dyads with expected relatedness coefficients of 0.25 (close kin such as half-siblings, uncle-nephew, etc.) (Fig. S3): 244 out of 2080 dyads (12%) and 1034 of 7875 dyads (13%) in winter 2008–2009 were probably close kin. However, only 71 and 322 close kin relationships in each winter remained after the likelihood ratio tests. Among these, 15 and 77 pairwise comparisons in winter 2007–2008 and 2008–2009 were between an adult and a first-winter individual and the remaining comparisons were between adult birds. Less than 1% of the dyads in winter 2007–2008 and 2008–2009 were likely to be full-siblings or parent-offspring, respectively. The dominant proportions of dyads were unrelated individuals (87% in winter 2007–2008, 86% in winter 2008–2009). These results are consistent with the non-significant average relatedness coefficients in each temporal group.

Population substructure of wintering ducks

The Tufted Ducks wintering in Switzerland breed mainly in Scandinavia and Russia (Hofer et al. 2005) and these breeding populations are genetically heterogeneous, as revealed by a range-wide population genetic study (Liu et al. 2012). Under this premise, within-winter population substructure would be expected if different breeding populations of Tufted Ducks use specific wintering sites at different time points during the winter. However, we found no genetic substructure between temporal samples within or between winters. The absence of within-winter population substructure at a local scale may be a consequence of random wintering admixture of Tufted Ducks from different breeding populations. Moreover, the frequent within-winter movements of Tufted Duck (Korner-Nievergelt et al. 2009) resulting in a high turnover rate could also erode temporal patterns at a single wintering site. The genetic composition of populations can also change between years as a consequence of annual variation in demographic factors, such as survival rate, sex ratio and age structure (Robertson & Cooke 1999, Hutchinson et al. 2003). In birds, evidence of significant population substructure between years has been reported, for example in the sedentary passerine Vinous-throated Parrotbill Sinosuthora webbianus, with the genetic changes driven by variable adult survival rates among years (Lee et al. 2009). However, these factors seem less important for long-lived Tufted Ducks with high genetic diversity and large population size.

Temporal pattern of relatedness

We found very low levels of average relatedness within temporal groups and none of these values was significantly different from zero. Our results indicate associations of largely unrelated birds on the wintering grounds but also the presence of small proportions of relatives, mostly pairs of close kin with intermediate relatedness coefficients (half-siblings, aunt–niece, etc.) and a very few probable parent–offspring and full-sibling pairs in wintering Tufted Ducks. Other studies have also detected high proportions of unrelated individuals inferred from estimation of relatedness coefficients in animal aggregations, for example bat colonies (Kerth et al. 2002), fish schools (Fraser et al. 2005, Piyapong et al. 2011) and bird flocks (Liker et al. 2009, Fleskes et al. 2010, Lee et al. 2012). Even when exhibiting low average relatedness coefficients, conspecific groups can still contain several kin-based clusters formed by close relatives, as observed for example in Bechstein's Bat Myotis bechsteinii (Kerth et al. 2002) and wintering roosts of the Eurasian Magpie Pica pica (Lee et al. 2012), but these are too rare to increase the average level of relatedness (Lebigre et al. 2008). This might imply that both communal benefits from living with unrelated conspecific individuals and kinship play a role in the origin of large animal aggregations.

Unexpectedly, we detected a negative association between pairwise relatedness coefficients and the temporal distance between the sampling dates of same-sex individuals, providing evidence of kin in social groups in wintering Tufted Ducks. This association disappeared when both sexes were pooled, suggesting that pairwise relatedness is low among female–male dyads which contribute largely to the pooled dataset. The observed sex-specific patterns might be associated with sex-specific differences in the timing of moult and migration in this species. Males group together and leave the breeding colonies earlier for moult and migration, whereas females and ducklings stay at the breeding grounds until the end of the breeding season (Cramp 1977).

Notably, we found that the occurrence of close kin relationships with an expected relatedness coefficient of 0.25 was higher than that of the closest relationship categories with R = 0.5 (full-siblings and parent–offspring; Fig. S2). Specifically, this category includes likely aunt/uncle–nephew/niece pairs, grandparent–grandchild pairs, offspring from the same mother from multiple years, or half-siblings from the same clutch sired by different fathers. Given that the exact age of each adult was not available, it is not possible to distinguish between these alternatives. Associations of related individuals (i.e. the category with R = 0.25) could arise if males and females from a given breeding colony sometimes stay together during migration and wintering, and if breeding colonies include related individuals. The latter seems likely given the high level of philopatry observed in this species (Blums et al. 2002). The very low frequency of parent–offspring pairs in wintering ducks is also expected because parent–offspring bonds break up at the end of the breeding season and adults and juveniles migrate separately (Cramp 1977).

Collectively, our findings demonstrate the presence of relatively few co-migrating same-sex relatives in long-distance migratory ducks in a wintering area thousands of kilometres from the breeding grounds. Fleskes et al. (2010) found elevated relatedness in captured groups of the Northern Pintail Anas acuta in the non-breeding season but their study focused on a short period in the post-breeding season (September–October). Kin-based structures were also evident in pre-breeding (colony arrival and nest-site selection) and post-breeding (colony departure) Common Eider (McKinnon et al. 2006). We assume that individuals in kin-based groups may benefit more, for example through the facilitation of cooperative anti-predator behaviours (Wright et al. 2001, Griesser et al. 2006) and reduced aggressive interactions in foraging groups (Bednarz 1988, Fraser & Bugnyar 2010). Furthermore, we suggest that kin association could also be beneficial for knowledge-sharing in migratory activities (orientation or timing; reviewed in Helm et al. 2006) as synchrony in migration of ducks from specific breeding colonies should be important for long-distance migrants. Indeed, colony-wise departures for autumn migration were found in Common Eider (McKinnon et al. 2006). More extensive sampling, ideally combined with field observations of marked individuals with known kinship, will be necessary to investigate further the temporal stability of kin associations as well as their potential evolutionary benefits in Tufted Duck and other waterfowl. Nevertheless, our results represent a first step in this direction and demonstrate the use of genetic estimation of relatedness to help better understand flock dynamics during migration and wintering of long-distance migratory birds.