Sex-biased dispersal in a northern ungulate population

In most mammals dispersal is male-biased and in many polygynous ungulates female philopatry and matrilineal grouping involve small-scale genetic structure. We have through sex-related differences in microsatellite allele distribution addressed sex-biased dispersal in a spatially expanding northern ungulate population. The Norwegian red deer population (Cervus elaphus atlanticus) has the last hundred years grown substantially and expanded spatially after a major decline from 300 to 100 years ago. Previous Bayesian analyses suggest a present division of genetic variation into five geographically separated subpopulations. Among these subpopulations the overall Fst values were 0.067 (SE=0.014) for males and 0.094 (SE=0.017) for females. Pairwise Fst values were significantly higher for females than males, demonstrating a stronger genetic structure among females, and that dispersal has been lower in females than males. Accordingly, a higher number of male than female first generation dispersers were identified among the five subpopulations using Bayesian assignment with prior population information, but significantly so only with relaxed stringency levels of assignment. The identified male-biased dispersal distances varied from 30 to 300 kilometers suggesting male biased dispersal on a large scale in red deer.


Introduction
Dispersal is sex-biased when the members of one sex are faithful to their natal range and / or matrilineal group (philopatric), while members of the other sex are more likely to disperse (Prugnolle & de Meeus, 2002).Most mammalian species exhibit male-biased dispersal and female philopatry (Greenwood, 1980;Clutton-Brock, 1989;Prugnolle & de Meeus, 2002), but recent applications of genetic tools suggest wide variation in the direction, magnitude and timing of sex-biased dispersal (Handley & Perrin, 2007).Theoretical studies show that philopatry in concert with polygony may involve small scale genetic differentiation among social groups (Chesser, 1991b;Chesser, 1991a), which may have important evolutionary consequences for kin and localized selection (Coltman et al., 2003).In many ungulate species polygony and philopatry thus involves a pronounced fine-scale genetic structure (Mathews & Porter, 1993;Petit et al. 1997;Purdue et al., 2000;Coltman et al., 2003;Nussey et al., 2005).Sex-biased dispersal at the subpopulation and meta-population levels has however been poorly documented (Prugnolle & de Meeus, 2002), but has recently been reported for turtles (Bowen & Karl, 2007) and bats (Chen et al., 2008).
The red deer (Cervus elaphus) is a highly polygynous (Pemberton et al., 1992) and philopatric species (Clutton-Brock et al., 1982b;Clutton-Brock et al., 2002) with many populations genetically structured at the subpopulation level (Hartl et al., 1990;Herzog & Gehle, 2001;Kuehn et al., 2004).On the small Scottish island of Rhum, philopatry and male-biased dispersal may explain the observed extreme finescale genetic structure, which has declined with time as population density has increased (Nussey et al., 2005), but may also account for genetic structure on a larger scale among management blocks (Nussey et al., 2006).In the large northern population in Norway, a genetic structure of five subpopulations has been suggested, which may be explained by previous strong genetic drift and subsequent limitations to gene flow during population growth and spatial expansion (Haanes et al., 2010).These limitations to gene flow include both isolation by distance and geographical barriers like fiords and steep mountains.A drastic decline from 300 to years ago limited the population to a few hundred individuals along the west coast (Collett, 1909;Ingebrigtsen, 1924), but after one century of growth and spatial expansion the population is today distributed throughout southern Norway, counting between 000 and 130 000 individuals (Langvatn, 1988;Forchhammer et al., 1998;Langvatn, 1998).We have investigated whether dispersal is sex-biased on a large scale within the spatially expanding Norwegian red deer population.Specifically we have assessed sex-biased dispersal through sex-related differences in genetic structure between subpopulations and by identifying first generation dispersers through Bayesian assignment analyses.et al., 2010).Solid grey shading indicates the approximate distribution of the population around 1900 (Collett, 1909;Langvatn, 1998), rectangles show combinations of locations to obtain at least 15 individuals per locality and the Bayesian clusters are indicated by transparently shaded ovals.
Bayesian assignment (STRUCTURE 2, Pritchard et al., 2000) without prior information about population structure has previously been used to show that the 15 sampled Norwegian red deer localities can be divided into five subpopulations (Haanes et al., 2010).For each of a different number of genetic clusters (K∈ [1,7]), an admixture model (α=1, α max =50) with uniform priors, correlated allele frequencies (Falush et al., 2003), 100 000 burnins cycles and 500 000 MCMC iterations was run 10 times.The log likelihood of the data (ln Pr(X K)) was highest for K = 5 and the statistic delta K (Evanno et al., 2005) was pronounced higher for K = 2 and K = 5 (Table 1).Since delta K is negatively related to the increasing variance among repeated runs and often increasing posterior probabilities with higher K values, it reflects the main genetic structure of the data set   (Evanno et al., 2005), and the genetic structure of Norwegian red deer was interpreted as a main dichotomy with a lower hierarchical level of five subpopulations (Haanes et al., 2010).The proportionate cluster membership was for most individuals and localities much higher in one of the clusters (Fig. 2, Appendix Table S1), and the data was divided into five well geographically separated clusters separated by tens to hundreds of kilometers (Fig 1).As summary genetics not have been published for the five identified Norwegian subpopulations, each of the five identified subpopulations was assessed through exact tests of Hardy-Weinberg equilibrium (HW) across the 14 loci using GENEPOP 3.4 with the default settings (Raymond & Rousset, 1995).Sequential Bonferroni correction was used to adjust for all repeated tests (Rice, 1989).To assess genetic variation we used FSTAT 2.9.3 (Goudet, 2001) to calculate the allelic richness (El Mousadik & Petit, 1996) and the gene diversity (Nei, 1987) for each subpopulation across loci.Summary statistics from the 15 locations can be found in Appendix Table S2 and Appendix Table S3 shows details on allelic frequencies among the five subpopulations.
To assess sex-biased dispersal we used FSTAT 2.9.3 (Goudet, 2001) to calculate pairwise F st values for males and females separately (Weir & Cockerham, 1984) between these five subpopulations (assuming HW when no significant deviations were detected).The numbers of adult individuals per sex per subpopulation are given in Table 2. To assess whether dispersal between the subpopulations was higher for males than females we used one Student's paired t-test to assess for differences in F st values (JMP 7.0.1., 2007).
Sex-biased dispersal may be detected through individual Bayesian assignment (Prugnolle & de Meeus, 2002;Freeland, 2005;Handley & Perrin, 2007).To identify first generation dispersers (n=419), we used Bayesian assignment with the five subpopulations predefined according to prior information on where each individual was sampled (POPINFO=1, GENSBACK=0, Pritchard et al., 2000).Model settings included admixture (α=1, αmax=50), only moderate migration (v=0.05), 100 000 burnins and 500 000 iterations.Among individuals assigned to other subpopulations than where they were sampled, STRUCTURE identifies significant first-generation dispersers.Markand recapture has shown that STRUCTURE can detect most natal dispersers with moderate genetic structure (e.g.F st ≥0.06) and that almost 100% accuracy may be obtained with less structure using high stringency levels of assignment (Berry et al., 2004).The numbers of significant first-generation dispersers identified by STRUCTURE that were adult and where the sex was known were assessed according to different stringency levels of assignment, the one with the highest q value or values higher than 0.5, 0.7, or 0.9.In addition, an alternative Bayesian assignment algorithm (Rannala & Mountain, 1997), implemented in GENE-CLASS 2 (Piry et al., 2004), was also used to detect first-generation migrants, using the same data set (n=419) and 10 000 simulations to account for Type I errors with two different alpha levels (0.05 and 0.01; Paetkau et al., 2004).Only adult first-generation dispersers where the sex was known were considered.To test if the number of dispersers was higher for males than for females we used Fisher Exact tests (one-tailed) because of low and zero table values (Bhattacharyya & Johnsen, 1977).

Results
With the original data set divided according to the five identified subpopulations, only one locus (OarF-CB193) was found to be deviating from HW in one subpopulation (nr.5, P<0.002).The level of genetic variation estimated through allelic richness and gene diversity was equal among the five subpopulations, but relatively low (Table 3).
For adults with a known sex, across the 14 genotyped loci the proportion of missing alleles was two percent for both females and males.Only one locus (BMC1009) was for females in one subpopulation (nr.4) found to be deviating from HW (P<0.004).HW was therefore assumed for the remainder of analyses.Among the five Norwegian red deer subpopulations overall F st values were 0.067 (SE=0.014)for the 167 males and 0.094 (SE=0.017)for the 112 females.Pairwise F st values among subpopulations separate for each sex varied from 0.01 to 0.14 and all were significantly different from zero except one, which involved a very low F st value (Table 4).In a pairwise t-test the F st values among subpopulations were significantly higher for females than for males (t=2.24,P=0.03, df =9), with an average difference of 0.023 (SE=0.010).
Individual-based Bayesian assignment identified more males than females as first-generation dispersers among the five subpopulations.For the STRUCTURE analysis the difference was significant only with relaxed levels of assignment while for both alpha levels significantly more males were detected through the GENECLASS analysis (Table 5).In the STRUCTURE analysis, some individuals were assigned with coefficients varying between 0.4 (highest q) to 0.6 because of partial assignment to more than one subpopulation, and statistical tests were not significant when these individuals were omitted at the higher stringency levels.Among males identified as dispersers between subpopulations (lowest stringency level in STRUCTURE but lowest alpha and highest stringency in GENECLASS), the distance in kilometers between sampling location and the outer edge of the assigned cluster (Fig. 1) varied from 30 (n=2), to 100-150 (n=5) and 200-300 (n=6).These were adult stags (n = 8) weighing from 107 to 143 kg (n=5), and subadults (n=5) weighing from 61 to 88 kg (n=4), respectively.

Discussion
The differences between the sexes in genetic structure estimated from microsatellite markers suggest that dispersal is more limited in females than males and thus that dispersal is male-biased between the subpopulations of the Norwegian red deer population.The high and significant F st values indicate that the sample sizes were adequate for each sex, as the only non-significant F st value was relatively low.The subpopulations are separated by tens to hundreds of kilometers (Fig. 1) and the differences in genetic structure therefore reflect limitations to dispersal on an intermediate to large scale in this expanding ungulate population.
Sex-specific F st values between the 15 sampled localities indicate a similar bias also for shortdistance dispersal but sample sizes were too low to achieve significant F st values (data not presented).Bayesian analyses offer a powerful alternative for quantitative estimates of sexbiased dispersal (Handley & Perrin, 2007), and the higher number of identified male than female first generation dispersers between subpopulations provide strong support for a malebiased dispersal on a large scale.Most of these originated in subpopulations four and five in the south-east (Fig S1 in supplementary), which lies in relatively flat and low-lying areas compared to the western shore where steeper topography and fiords like the Sognefjorden lying to the south of subpopulation 3 act as major barriers against dispersal (Haanes et al., 2010).Moreover, the majority of detected dispersers dispersed to more distant subpopulations rather than to the closest possible subpopulation (Fig S1).However, the efficiency of the STRUCTURE algorithm is reduced with low levels of genetic differentiation (Berry et al., 2004;Latch et al., 2006) and since our F st values indicate mostly moderate to weak genetic structure (Wright, 1978;Hartl & Clark, 1997), the number of dispersers may have been underestimated.This is reflected by the higher number of first-generation dispersers identified through the GENECLASS analysis, which provides good support for the results and conclusions.
In red deer, male-biased dispersal has been well documented through field studies (Clutton-Brock et al., 1982b).However, genetic methods may give additional insights into how sex-biased dispersal translates into gene flow (Handley & Perrin, 2007).As dispersed individuals successfully reproduce, their genetic contribution will translate into gene flow.The higher male than female dispersal suggested by differences in genetic structure and numbers of dispersers within the Norwegian population will therefore, depending on the mating success of the dispersers, in the next generation translate to gene flow between the subpopulations.Genetic differentiation from polygony and limitations to gene flow by philopatry among social groups (Chesser, 1991b;Chesser, 1991a) have been reported as fine-scale genetic structure in several ungulates like Soay sheep (Ovis aries; Coltman et al., 2003), Mediterranean muflon (Ovis gmelini;Petit et al., 1997) and White-tailed deer (Odocoileus virginianus; Mathews & Porter, 1993;Purdue et al., 2000), and red deer (Nussey et al., 2005;Frantz et al., 2008).On a scale of a few kilometers, malebiased gene flow has been suggested from a much weaker genetic structure in nuclear microsatellite markers than in maternally inherited mitochondrial DNA among red deer management blocks on the Scottish island of Rhum (Nussey et al., 2006).Within the relatively small study area of Rhum, a maximum distance of 22 kilometers for male dispersal has been recorded (Clutton-Brock et al., 1982b).By comparison, our Bayesian analyses identified mainly long-distance male dispersal on a scale of 10 to 100 kilometers, of which most were large adults that probably contribute to reproduction and thus gene flow.Such longdistance dispersal distances have also previously been reported both from the Norwegian population (Collett, 1912;Ahlèn, 1965) and other spatially expanding ungulate populations with sex-related differences in genetic structure such as elk (Petersburg et al., 2000) and whitetailed deer (Long et al., 2005).
Many ungulate species with fine-scale genetic structure from philopatry and polygony have a recent history of demographic growth (Mathews & Porter, 1993;Purdue et al., 2000;Coltman et al., 2003), even though increasing dispersal with population growth and increased density would be expected to break down such structure.On the northern management block of Rhum, red deer were released from the an-nual cull from 1972 and population density was allowed to increase.Here the fine-scale genetic structure actually declined as population density increased until the year 2001, but observed dispersal between population subdivisions did not increase in either sex and the decline was instead explained by an increased female breeding population size and a reduced level of polygony (Nussey et al., 2005).However, during this period overall male emigration increased (Clutton-Brock et al., 1997;Clutton-Brock et al., 2002) and the spatial association between female relatives increased (Albon et al., 1992).By comparison, the growing Norwegian population has during the last century expanded spatially (Langvatn, 1988;Forchhammer et al., 1998;Langvatn, 1998), involving dispersal of both males and females into new areas.In red deer hinds, increases in the density of matrilineal groups may involve increased competition and reduced reproductive success (Clutton-Brock et al., 1982a), and experiments have shown that aggression increases with density (Blanc & Thériez, 1998).We suggest that as the Norwegian red deer population grew (Langvatn, 1988;Forchhammer et al., 1998;Langvatn, 1998), in addition to increased male emigration, hinds started to disperse from core areas as density increased and competition intensified.
Several hypotheses proposed to explain sexbiased dispersal have been classified into three main hypotheses (Prugnolle & de Meeus, 2002;Freeland, 2005); the resource-competition hypothesis (Greenwood, 1980), the local matecompetition hypothesis (Dobson, 1982;Perrin & Mazalov, 1999), and the inbreeding avoidance hypothesis (Pusey, 1987).More recently, cooperative behaviour of kin and enhanced use of local resources have been added as a fourth hypothesis (Perrin & Lehmann, 2001;Le Galliard et al., 2006;Handley & Perrin, 2007).From its limited distribution one century ago (Collett, 1909;Ingebrigtsen, 1924), the Norwegian population has expanded considerably (Langvatn, 1988;Forchhammer et al., 1998;Langvatn, 1998) and competition for local resources in newly established areas can probably be excluded.Due to a strongly male-biased harvest of Norwegian red deer the sex-ratio of the population is skewed towards females (Langvatn & Loison, 1999).Increased bias of the sex-ratio towards females has been shown to decrease male harem holding periods and increase the proportion of males that hold harems (Clutton-Brock et al., 1997), and any pronounced local male competition for female mates seems therefore unlikely in the Norwegian population.Finally, with the social grouping of philopatric females in red deer (Clutton-Brock et al., 1982b;Clutton-Brock et al., 2002), the hypothesis of cooperation among female relatives can not be excluded, but may rather be complementary to inbreeding avoidance.In social mammals, inbreeding avoidance probably has played an important role in the evolution of dispersal, as shown among polygynous sciurid species where the degree of male-biased dispersal increases with sociality (Devillard et al., 2004).Natal faithfulness towards mobile social units may involve similar genetic differentiation as philopatry (Prugnolle & de Meeus, 2002), especially with polygynous mating systems (Chesser, 1991a), and we suggest that hinds during spatial population expansion dispersed in matrilineal groups to maintain the benefits of cooperation and that the long male dispersal distances are related to inbreeding avoidance.A possible management implication of this may apply to particularly small and genetically structured populations where the male proportion in the population should be maintained to avoid inbreeding.This suggests that rather than just considering the five identified genetically differentiated subpopulations as separate management units one should probably rather manage the whole population as one meta-population.

Fig. 1 .
Fig. 1.Sampling localities of Norwegian red deer divided according to five clusters identified by Bayesian assignment (from Haaneset al., 2010).Solid grey shading indicates the approximate distribution of the population around 1900(Collett, 1909;Langvatn, 1998), rectangles show combinations of locations to obtain at least 15 individuals per locality and the Bayesian clusters are indicated by transparently shaded ovals.

Fig. 2 .
Fig. 2. Individual posterior probabilities of Bayesian assignment to each of two to five clusters (STRUCTURE with K∈[2,5], different colours) among 419 red deer in each of 15 sampled localities (from Haanes et al., 2010).

Fig. S1 .
Fig. S1.Norwegian red deer long-distance dispersers identified through STRUC-TURE with priors on which subpopulation each individual was sampled, their sex (rectangle=male, circle=female), sampling location (arrow points to, indicating the direction of dispersal, and the likelihood of assignment (coefficient q).

Table 2 .
The numbers of adult Norwegian red deer with known sex which were sampled in each of five identified subpopulations.

Table 3 .
Genetic variation in the five identified Norwegian red deer subpopulations (n = 419, the involved locations also given, see Fig1), each represented by estimates of allelic richness (A R ), unbiased gene diversity (H) and inbreeding (F is ) averaged across loci.Standard errors in brackets (SE).

Table 4 .
Pairwise F st values for females and males separately between five geographic subpopulations of Norwegian red deer identified by Bayesian assignment and the difference between pairs (∆-F st ).Probabilities that F st values differ from zero (P) and significance after sequential Bonferroni correction in italic.

Table 5 .
Number of male and female first-generation dispersers among the five identified subpopulations according to STRUC-TURE with different stringency levels as criterion for assignment (CA) and according to GENECLASS with different alpha (α) levels.Whether the number of dispersers is higher for males than females is tested through Fishers exact tests, for which probabilities (P) are given (significant differences in bold**).