Role of Introduction History and Landscape in the Range Expansion of Brown Trout ( Salmo trutta L.) in the Kerguelen Islands

Human-mediated biological invasions constitute interesting case studies to understand evolutionary processes, including the role of founder effects. Population expansion of newly introduced species can be highly dependant on barriers caused by landscape features, but identifying these barriers and their impact on genetic structure is a relatively recent concern in population genetics and ecology. Salmonid populations of the Kerguelen Islands archipelago are a favorable model system to address these questions as these populations are characterized by a simple history of introduction, little or no anthropogenic inﬂuence, and demographic monitoring since the ﬁrst introductions. We analyzed genetic variation at 10 microsatellite loci in 19 populations of brown trout ( Salmo trutta L.) in the Courbet Peninsula (Kerguelen Islands), where the species, introduced in 3 rivers only, has colonized the whole water system in 40 years. Despite a limited numbers of introductions, trout populations have maintained a genetic diversity comparable with what is found in hatchery or wild populations in Europe, but they are genetically structured. The main factor explaining the observed patterns of genetic diversity is the history of introductions, with each introduced population acting as a source for colonization of nearby rivers. Correlations between environmental and genetic parameters show that within each ‘‘source population’’ group, landscape characteristics (type of coast, accessibility of river mouth, distances between rivers, river length . . . ) play a role in shaping directions and rates of migration, and thus the genetic structure of the colonizing populations. with variations in nature of the coast and river characteristics (accessibility, length, width . . . ). We analyze the inﬂuence of both introduction history (time, place, and numbers of individuals) and some landscape features on the dispersal and colonization dynamics of an anadromous ﬁsh. We also discuss the results in terms of brown trout potential for local adaptation, enabling efﬁcient colonization and settlement in new territories.

Following the introduction of a species in a new environment, small populations can evolve in isolation from other individuals of the species. Rapid loss of genetic variation in the derived founder population can occur under the combined effect of genetic bottleneck and genetic drift (Nei et al. 1975). Range expansion involving long-distance migration at a slow rate can also contribute to further reduction in genetic variation in colonizing populations compared with the source population (Hewitt 1999;Hansson et al. 2000). On the other hand, rapid population expansion will be likely to retain genetic diversity within the founder population, through low genetic drift and higher gene flow (Hewitt 1999;Zenger et al. 2003). Historical patterns can then be modulated by local environmental constraints that trigger different ecological responses of individuals, shaping the genetic diversity into its contemporary structure (e.g., Castric et al. 2001;Dionne et al. 2008). Human-mediated biological invasions are mostly recent, and sometimes well documented (Sakai et al. 2001; Lee 2002). They constitute good field opportunities to understand evolutionary processes, among which the role of founder effects (Estoup et al. 2001;Clegg et al. 2002;Abdelkrim et al. 2005) and local population adaptation.
Colonization of a new environment is one of the situations in nature that depart from the migration/drift equilibrium hypothesis which is assumed by many population genetics models used to quantify genetic variation within and among population (Kinnison et al. 2002;Castric and Bernatchez 2003;Ramstad et al. 2004;Poissant et al. 2005). Landscape features can lead to barriers to migration and/or population expansion. They thus shape the patterns of structuring of the genetic diversity and the speed with which these patterns will eventually stabilize. Identifying these barriers and their impact on genetic structure is a relatively recent concern in population genetics and ecology (Sork et al.

Introduction of Brown Trout in the Kerguelen Archipelago
The Kerguelen Archipelago is located in the sub-Antarctic region (southern hemisphere, lat 49.37°S, long 69.50°E). This Archipelago (area of 7215 km 2 ) is constituted by a main island (Grande Terre) surrounded by 300 smaller islands. Except in the eastern part, the coastline is very indented, with narrow isthmus, deep bays and fjords. The open sea is cold (between 0.5 and 4.5°C), often rough but void of ice. The oceanic origin of the island, its relative distance from any continent, and its position on the polar front explain the absence of a natural freshwater ichthyologic fauna (Davaine and Beall 1997). Since the late 1950s, several species of salmonids have been introduced in a limited number of rivers, mainly brown trout (S. trutta), Atlantic salmon (Salmo salar), and brook charr (S. fontinalis) (Davaine and Beall 1997). After these introductions, brown trout has colonized the Archipelago and, 40 years later, is present in most rivers in the eastern part of the main island. The study is focused on the Courbet Peninsula, situated at the east of the Kerguelen's main island (Figure 1). Mark-recaptures studies conducted for 24 years on both sides of the Prince of Wales Peninsula (south-east of Courbet) have shown that this peninsula acts as a barrier to brown trout mobility (Davaine P, unpublished data) and Courbet can be considered as an independent system as far as brown trout colonization is concerned (Figure 1). The Courbet Peninsula presents contrasted landscape features. Its northwestern part is covered by a high mountain range (basaltic rocks), the coast is lined by high cliffs, and the river mouths are at the end of narrow fjords. The eastern part is a wide zone consisting of quaternary deposit plains after the melting of the ice caps, and the coast is made up of large sand and pebbles beaches.
Available information about brown trout introduction in the Courbet Peninsula was obtained from the review work of Davaine and Beall (1997). According to these authors, brown trout was initially introduced in 3 different rivers of north part of Courbet Peninsula and its east part (westward bounded by the Prince of Wales Peninsula): Studer River, Chateau River, and Nord River. Although several attempts were done with cohorts of the same origin and during the same period, only 3 introductions were successful in the 2 former rivers (Davaine and Beall 1997). In these 3 cases, the introduced fish originated from a French domestic stock raised in a hatchery located on the Nive River (Pyrénées). In 1959, 721 1-year juveniles were successfully introduced in the Studer River (north-west of Courbet). In 1962, 22 4-year adults (from the same cohort) were stocked in the Chateau River (east of Courbet), and 2000 fry, from the same origin, were introduced in the Studer River. In 1981, 15 000 fry from a cross between 2 males and 3 females from the Chateau population were transferred to the Nord River. Field surveys during the following decades have reported the presence of trout populations in almost all the rivers and brooks of the peninsula (Table 1). In all naturally colonized rivers, brown trout is the only salmonid present (and actually the only fish), except in the rivers from the Norwegian Bay, where it is found in sympatry with brook trout (S. fontinalis), also introduced at the same period. Results from the field surveys have led to the hypothesis that rivers S1-S5 have been colonized by individuals from the Studer River, N6 and N7 from the Nord River, and C8-C16 from the Chateau River (Davaine and Beall 1997).

Biological Material
Salmo trutta populations from 19 rivers of the Courbet Peninsula have been sampled between January 2001 and March 2002: the 3 original introduction rivers (Studer, Nord, and Chateau) and 16 rivers that have been colonized by brown trout (S1-C16) ( Figure 1 and Table 1). Fish were sampled by nondestructive electrofishing in several stations along the watershed. Population densities were estimated for each station, and each fish was measured, weighed, and sexed whenever possible. Scales were sampled, and the adipose fin was cut and stored in absolute ethanol. Age was determined from the size histogram of the population for the young stages and confirmed by scale reading of a subsample. For each river, 30 one-year fish (juveniles) from a single station were randomly chosen for genotyping. This age was chosen to make sure the individuals studied were born in the river because Kerguelen trout do not migrate before they reach 2 years of age (Jarry et al. 1998). Thirty adult sea trout from the 3 source rivers (i.e., individuals returning in the river after a growth period in the sea) were genotyped. Hereafter, ''sea trout'' will designate fish that migrate to the sea for part of their life cycle, whereas ''migrant'' will refer to fish that are immigrant from other streams.

Landscape and Environment Information
Each river was characterized by 2 ecological parameters, integrative of several variables describing coast and river landscape: coast type and river type (Table 1). Coast type is the combination of 3 highly correlated variables: geographical zone, watershed geology, and coastline morphology. This variable aims to best describe the coastline environment in which the sea trout lives. Three classes were defined for this parameter, as follows: (I) low, straight coastline with sand or small pebble beaches, corresponding to the quaternary marsh plains, on a sand/clay ground, (II) straight coastline, low-elevated rocky hills, corresponding to quaternary plains on coarse material, and (III) high cliffs with narrow fjords, corresponding to high basaltic plateaux. Class I was found in the east zone of Courbet (from the Norwegian Bay to the Marville Lake), Class II in the northeast of Courbet, and Class III in the north-west zone. River type is a synthetic variable aiming to describe the morphodynamic characteristics of river, and their accessibility to homing or straying sea trout. It is the combination of 3 variables, also highly correlated: river length, river width, and main water flow in the summer. Three classes were defined for this parameter: (I) long rivers, with easily accessible mouth, total length 13-50 km, mean width 10-30 m, mean water flow 300-1500 l/s, (II) medium rivers, total length 6-12 km, mean width 5-10 m, mean water flow 100-300 l/s, and (III) small rivers with poorly accessible mouth, total length (mean river and tributaries) 3-5 km, mean width (in the zone accessible to sea trout) 1-5 m, and mean summer water flow 30-100 l/s. Note that, due to the history of introduction, each source river is associated with a different coast type, and thus each ''coast type'' class is tightly associated with a group of colonized rivers supposedly derived from the same source population.  1959-1962, 1962, and 1981, respectively. S1-C16 (from north to south) are naturally colonized rivers that have been sampled for the study. The dotted arrows indicate a hypothetical colonization process, based on field observations only (from Davaine and Beall 1997). Details on the sampled rivers are given in Table 1.

Statistical Analysis
For each population, the number of alleles per locus (N a ), the observed heterozygosity (H o ), and the unbiased expected heterozygosity (H e ) (Nei 1978) were calculated using the GENETIX software v. 4.05 (Belkhir et al. 1996). Differences between N a , H o , and H e among groups of samples were tested using a permutation scheme (3000 permutations) according to the FSTAT software v. 2.9.3.2 (Goudet 1995(Goudet , 2001. To test for conformity to Hardy-Weinberg equilibrium (HWE), mean F is per population was computed over loci according to estimators of Weir and Cockerham (1984), using the GENETIX software. Significance levels were assessed from 1000 random permutations of alleles within populations. Linkage disequilibrium (LD) was calculated for each pair of loci, using Fisher's Exact test as implemented in the GENEPOP software package v. 1.2 (Raymond and Rousset 1995).
We used the software BOTTLENECK v. 1.2.02 (Piry et al. 1999) to identify recent reduction in effective population size. The software estimates the expected distribution of heterozygosity from the observed number of alleles, under the mutation-drift hypothesis and a 2-phased model of mutation with 5% multistep mutations (as recommended by the authors for microsatellite markers). A significant heterozygote excess (tested by a Wilcoxon sign-rank test) indicates a possible population bottleneck (Cornuet and Luikart 1996).
Bayesian assignment tests, included in the software STRUCTURE 2.0 (Pritchard et al. 2000), were used to estimate the number of genetic clusters and to evaluate the degree of admixture among them. Structure is based on a Bayesian approach and uses a Markov chain Monte Carlo (MCMC) Coast type: (I) low, straight coastline with sand or small pebble beaches, corresponding to the quaternary marsh plains, on a sand/clay ground, found in the east zone of Courbet (from the Norwegian Bay to the Marville Lake); (II) straight coastline, low-elevated rocky hills, corresponding to quaternary plains on coarse material, found in the north-east of Courbet; (III) high cliffs with narrow fjords, corresponding to high basaltic plateaux in the north-west zone.
River type: (I) long rivers, with easily accessible mouth, total length 13-50 km, mean width 10-30 m, mean water flow 300-1500 l/s, (II) medium rivers, total length 6-12 km, mean width 5-10 m, mean water flow 100-300 l/s, (III) small rivers with poorly accessible mouth, total length (mean river and tributaries) 3-5 km, mean width (in the zone accessible to sea trouts) 1-5 m, and mean summer water flow 30-100 l/s; First observation: first date when fish were seen in the river, either as visitors (V) or as part of a reproductive population (R); First reproduction: estimated date where a reproduction occurred for the first time, based on the date of first observation of fry and/or the age pyramid of the fish at that date.

Launey et al. Landscape Genetics of Salmo trutta in Kerguelen
algorithm to find the posterior probability that individuals belong to each of K clusters assuming linkage equilibrium and HWE across multiple, unlinked loci. The analysis was performed with a ''burn-in'' period setting of 50.000 and 100.000 MCMC repetitions in order to estimate the number (K) of genetically distinct clusters. The analysis was run for different a priori values of K (from 1 to 10). For each K, the posterior probability of the number of populations was computed, using an admixture model, with allelic frequencies correlated among populations. The statistic DK was used to identify the greatest rate of change between each subsequent K (Evanno et al. 2005) and thus determines the uppermost level of structure. Ten replications were performed for each value of K.
Cavalli-Sforza's genetic distance (Cavalli-Sforza and Edwards 1967) was chosen to assess genetic divergence between populations. This measure of distance is purely based on drift and should perform better than measures based on mutational process (such as standard Nei's D) in populations with a small divergence time (Goldstein et al. 1995). Distances were calculated using the PHYLIP 3.57 software package (Felsenstein 1989). The distance matrix was visualized as a neighbor joining (NJ) tree (Saitou and Nei 1987, using PHYLIP). Robustness of the nodes of the unrooted tree obtained was assessed by bootstrapping over loci (500 pseudoreplicates).
F st (Weir and Cockerham 1984) were computed for the whole data set and within each ''source group.'' F st values were compared between groups using a permutation scheme as implemented in FSTAT 2.9.3.2 (Goudet 1995(Goudet , 2001 using 3000 permutations. Pairwise F st were computed for each pair of population to estimate gene flow, using GENETIX. Migration rates between populations were estimated by identifying migrants based on population exclusion methods in GENECLASS 2.0 (Piry et al. 2004) using the Bayesian method (Rannala and Mountain 1997) and the likelihood ratio of L home to L max ). Monte Carlo resampling was performed with 10 000 simulated individuals and an assignment threshold of 0.05. Mean self-assignment rates in groups of populations from the same coast type or river type were compared using a Student's t-test. A higher self-assignment rate indicates a lower gene flow.
The relative importance of gene flow in shaping population genetic structure was assessed by examining the evidence for isolation by distance (IBD). The linear arrangement of rivers along shorelines allows for the application of a stepping-stone model of migration to our system, resulting in decreased genetic similarity with genetic distance. Geographical distances (GeoD) were measured along the coastlines. F st /(1 À F st ) was used as a measure of genetic distance (GenD) between each pair of population, as suggested by Rousset (1997). Correlations between geographical and genetic distances were estimated by a Mantel test (Mantel 1967, using GENETIX). Two series of tests were performed: one with all the populations included and one within each source group. If our hypothesis of colonization from one source population to the nearby rivers holds, and if the stepping-stone model applies, we expect to see significant IBD within each source group. Slopes of the regression lines were compared between coast types (Student's t-test).

Genetic Admixture Analysis
Bayesian assignment analyses for S. trutta revealed that the uppermost level of structure was identified for K 5 2. To parallel the number of introductions, assignment tests were also performed for K 5 3. For K 5 2, one cluster contained all the ''Studer group'' populations (Studer and S1-S5), whereas the other cluster contained all other populations ( Figure 2). Little admixture was observed between these 2 clusters, with strong admixture barriers at the geographic borders (between S5 and C6). For K 5 3, the Studer cluster was maintained. The third putative cluster contained populations from the Lake Marville and easternmost rivers, but with extensive admixture with the Chateau cluster.

Interpopulation Differentiation and Correlation with Coast Type
The overall F st was 0.068 (P , 0.001) and indicated the existence of a genetic structuring within the populations of the Figure 2. Bayesian assignment probabilities inferred in the program STRUCTURE (Pritchard et al. 2000) for K 5 2 (above) and K 5 3 (below). Each vertical line corresponds to a single individual and colors represent the proportional membership coefficient of that individual to each genetic cluster.  Courbet Peninsula. Pairwise F st between populations are shown in Table 2. As expected, the genetic divergence between Nord and Chateau populations (juveniles samples, F st 5 0.012, P , 0.005) was smaller than that between Nord and Studer (juveniles samples, F st 5 0.037, P , 0.001). Juveniles and sea trout samples from the same introduction rivers were weakly (Studer and Chateau) or not significantly (Chateau) divergent. S1 and C8 populations were highly differentiated from all others. The lowest F st values were found within the rivers of the Lake Marville system (C9-C12; F st from 0 to 0.017, sometimes not significantly different from 0). The rivers in the east of the peninsula (Coast type I and II) were weakly differentiated. By contrast, rivers associated with Coast type III appeared to be more differentiated, both within the area and also from eastern rivers. Overall, F st were 0.066, 0.041, and 0.038 in the Studer, Nord, and Chateau group, respectively. These values were all significantly different from 0 but not significantly different from one another. Estimates of migration rates are given in Table 3. Selfassignment rates were significantly higher for Coast type III rivers than for Coast type I (0.827 and 0.635, respectively, P 5 0.03). They were also significantly higher for River type III than River type I (0.83 and 0.64, respectively, P 5 0.03), and marginally significantly higher for River type II than River type I (0.79 and 0.64, P 5 0.054). This indicates a lower gene flow within Coast type III rivers, and a higher gene flow to and from River type I watersheds. Little gene flow was observed between the different coast types.
The overall pattern of genetic differentiation was illustrated by the NJ tree based on Cavalli-Sforza genetic distances ( Figure 3). This tree showed 2 highly divergent clades supported by bootstrap values of 100 and 98, respectively. The first clade stemmed from the introduction river Studer and grouped all the populations of the ''Studer source group,'' with long branch lengths indicating important differentiation levels between populations. The other clade included the populations of the ''Chateau source group.'' Apart from C8, branch lengths were shorter in this cluster, indicating a low divergence between populations. The Lake Marville system (C9-C12) formed a very homogeneous and coherent group supported by a bootstrap value of 83). The rivers of the ''Nord source group'' were in an intermediate position (with a bootstrap value of 81 for the N6/N7 branch).
The overall Mantel test was significant (r 5 0.319, P 5 0.007), indicating there was a correlation between geographical and genetic distance across populations (Figure 4). This remained valid within each source group: r 5 0.500 (P 5 0.006) and r 5 0.719 (P 5 0.008) for the Chateau and Studer source group, respectively. The slope of the IBD regression line was marginally significantly higher (P 5 0.06) in the Coast type III group than Coast type I group. There were not enough populations associated with Coast type II for IBD testing.

Intrapopulation Variability and Correlation with River Type
Genetic variability statistics are listed in Table 4. Mean number of alleles/locus/pop (across all loci and popula-tions) was 7.03 (±1.1), and mean heterozygosity was 0.748 (±0.04). The introduction rivers showed a significantly higher diversity than the colonized ones, both for the mean number of alleles per population (8.11 vs. 6.63) and for mean H e (0.780 vs. 0.736). However, the colonized populations did not all have the same genetic variability: 3 populations (S2, S4, and S5) had a greater diversity and 2 (S1 and C8) a lower diversity compared with the other colonized rivers (statistically significant for N a and H e in all cases). There were significant differences in allelic richness and heterozygosity between river types, and mainly populations in type III rivers showed significantly reduced genetic variability. After Bonferroni correction for multiple tests, 7 populations exhibited a F is value significantly different from 0, indicating departure from Hardy-Weinberg proportions ( Table 2). Nord and Chateau juveniles samples, as well as S3, showed heterozygote excess. Conversely, 4 populations (S2, C8, C10, and C14) showed heterozygote deficiency. No differences in F is were observed between river types. Of the 990 possible pairwise LD tests, 89 were significantly different from 0 (9%), but there were variations between rivers (Table 4). Five populations (S1, S3, N7, C13, and C15) had 7 or more significant tests, whereas the others had 5 or less; LD was significantly higher in River type III populations (mean LD 3.2, 2.5, and 9 for River type I, II, and III, respectively; P 5 0.03).
Results of the BOTTLENECK analysis revealed a possible recent reduction in population size in 4 samples: S4, N7, C9, and C10.

Genetic Diversity and Colonization Dynamics of Brown Trout in the Courbet Peninsula
The influence of historical patterns on the molecular variation was demonstrated in numerous cases, including freshwater fish, although the duration of this historical event is sometimes much longer than in the Courbet situation: from several generations (Kinnison et al. 2002;Poissant et al. 2005) to postglacial times (Costello et al. 2003). In this part of the Courbet Peninsula, the main factor explaining differentiation between brown trout populations, based on assignment tests and phylogenetic reconstruction, is the history of introduction; the divergence between the 3 source populations reflects the introduction chronology. Both STRUCTURE analysis and genetic distance analysis supported the hypothesis that each population introduced in the 1960s (Studer and Chateau) had played the role of source for colonization of the geographically close rivers. This result fitted the hypothesis based on field surveys (Davaine and Beall 1997, Figure 1). The absence of clear structuring between Nord and Chateau groups reflected the Chateau origin of the founding individuals of the Nord population. However, secondary contact through migrants from the east of the island (originating from the Chateau source population) to the Nord river area cannot be ruled out. Stocks introduced in Studer and Chateau originated from the same hatchery population and belonged to the same cohorts. However, more than 2700 young fish (0þ and 1þ) were introduced in Studer and only 22 reproducing adults (4þ) in Chateau, which could have led to slight genetic divergence through founding effect, and subsequent genetic drift over 5-10 generations since introduction. The divergence was observed as soon as 1984 using allozyme markers (Guyomard et al. 1984). Within-population genetic diversity values of Kerguelen S. trutta populations were very close to those of European wild and hatchery populations. Genetic variability data are not available for the French source hatchery population. However, a comparison can be made with other French domesticated brown trout stocks because they usually exhibit very little genetic differentiation (Krieg and Guyomard 1985;Estoup et al. 1998;Charles et al. 2005). Using the same loci as in this study, Charles et al. (2005) found a mean number of alleles of 9.1 (±1.3) and a mean heterozygosity of 0.781 (±0.01) for 4 French hatchery populations. This suggests that the loss of genetic variation was rather limited within the source populations and mostly confined to a reduction in the number of alleles. Comparing reduction in the number of alleles and reduction in heterozygosity can help to assess the length of a colonization process because during a colonization process, it is expected that the effects of bottlenecks initially result in a reduction of the mean number of alleles and only later in a loss of heterozygosity (Luikart et al. 1998). Indeed, in Courbet, populations of the east zone (N7-C16) had significantly fewer alleles than the source populations although the rivers have been colonized the soonest. Populations from the north zone, on the other hand, exhibited a high genetic diversity, similar to that of the introduced populations, although the rivers have been colonized later in time. The higher within-population genetic diversity found in the Studer group rivers could also result from a higher initial genetic diversity in the Studer source population, still existing in contemporary samples of both juveniles and sea trout. Indeed, more diversity is expected between pools of migrants derived from the 2700 fish introduced in Studer than between pools derived from the 22 reproducing adults introduced in Chateau. Thus, if different pools colonize different rivers, the founder effect is expected to lead to more interpopulation variability in the Studer area than the Chateau area.
The individuals sampled here were 1þ juveniles. Sampling at this life stage can lead to catching families rather than samples representative of populations (Hansen et al. 1997). However, this bias is unlikely to have occurred in the Studer area. In this area, rivers are short and trout density low and most of the river parts inhabited by trout were sampled. In contrast, a sampling bias cannot be excluded in the Chateau area where sampling zones were often short with regard to river lengths. This bias seems rather unlikely in the present case because the juvenile and adult samples were very similar in the 3 rivers where both of them were genotyped. However, even if such a bias would occur, it would mostly lead to an overestimate of the genetic differentiation between populations and would not invalidate our main findings, that is, 1) the congruence between the history of introduction and population divergence and 2) the differences in genetic structuring between the north-west and east areas.

Impact of Life History and Environmental Parameters on Colonization
Colonization of the Courbet Peninsula by the brown trout represents an invasion process of an empty territory by an alien species. Regardless of the nature of the invader, it has been observed that such processes follow roughly the same sequence of phases: 1) an initial establishment phase with low spread, 2) an expansion phase marked by increasing spread rates, and 3) a saturation phase when spread rate reaches a plateau (review in Arim et al. 2006). Invasion rate can, however, be restricted by biological constraints (for instance, environmental parameters). New sites will be further colonized only once the newly arisen population has grown sufficiently to produce efficient colonizing groups (Fort and Méndez 1999).
In accordance with the above process, hypotheses can be made about the dynamics of trout populations in Kerguelen. Adult sea trout are the most efficient colonizing individuals because of a very high fecundity and capacity to migrate (Davaine and Beall 1997). However, they are the least abundant class of fish in a population (Thomas et al. 1981;Jarry et al. 1998) and, in addition, the proportion of strayers in brown trout is known to be low within a population (Le Cren 1985). Therefore the colonization of a nearby new river will probably not become effective until the number of sea trout and, consequently, the whole population size have become quite large. Size of the source stock is thus one of the key factors to explain the colonization process. Indeed, asymmetrical gene flow patterns, from large to small populations only, was observed in brook trout S. fontinalis (Fraser et al. 2004) and brown trout (Hansen et al. 2007). Growth and expansion of the source stock can be modulated by local environment constraints. In the 2 cases where introduction rivers were located in a closed bay, other rivers in the bay were rapidly colonized (Norwegian Bay: 6 years after introduction in the Chateau River; Accessible Bay: 8 years after introduction in the Nord River), but it took about 3 times longer to colonize rivers out of the Norwegian Bay. In the absence of such a closed bay harboring several populations in the Studer zone, the growth of a colonizing source stock was slower and could explain in part the delay in colonization of this area. Following this model of population growth and dispersion, the Lake Marville system could have become a new colonization center when it had reached a sufficient population size and could have been the source population for the late colonization of C8, as shown by the STRUCTURE analysis.
We used 2 environmental parameters, ''Coast type'' and ''River type'' to describe the landscape surrounding brown trout populations. Most of the sea trout populations migrate only a few kilometers away from their home river (Bertmar

Total
Coast type III Coast type I Figure 4. Correlation between geographical distances (measured along the coast line between river mouths) and genetic distances in brown trout populations in the Courbet Peninsula.  1979;Le Cren 1985;Berg OK and Berg M 1987), and salmonids are less attracted to a river empty of fish than to a river already harboring a population of the same species (Solomon 1973). Thus, coastal patterns, describing the landscape experienced by trout during its dispersal at sea, should drive the observed patterns of colonization dynamics. High cliffs, indented coasts exposed to the dominant rough oceanic conditions, narrow river mouths at the end of steep fjords which characterize Coast type III are associated with strong interpopulation genetic differentiation, low gene flow between rivers, and strong IBD patterns. The stronger slope of the IBD regression (compared with Coast type I) indicates that physical distances are more difficult to cover by trout migrating along this coast type. This is a setting where landscape characteristics make up a multiple stage barrier to migration and to settlement in a new river. Colonization in this area would be a slow process, under the influence of high ecological constraints, with the discontinuous arrival of small pools of dispersing individuals from nearby rivers and could lead to divergence through founder effects and subsequent genetic drift. A contrasted situation was found for rivers associated to Coast type I. This coast is low and sandy, with soft coastlines, the river mouths are wider and open freely in the sea, and the rivers often offer a longer zone accessible to migrants. The colonization also fitted a model of IBD, but with a lower correlation. Coast type I landscape characteristics appear more favorable to migration and settlement of new populations. Indeed colonized populations appeared rather homogeneous genetically, with low F st values indicating that exchanges of migrants between rivers were more frequent. Both demographic data and interpretation of genetic results indicated that the expansion of populations in the east zone was a smooth and continuous process and occurred faster after introduction of fish in the Chateau River. However, the Coast type parameter is tightly associated with geographical region (and river of origin), and the reality of the correlation between coast type and genetic parameters remains to be tested. The River type parameter, describing ecological characteristics of the colonized river, was associated with successful settlement of brown trout in newly colonized rivers and within-population genetic variability. For instance, most of the observed genetic disequilibria (F is , LD, BOTTLENECK) as well as the lower genetic variability were found in River type III. These rivers are small, brook-like, sometimes geographically isolated rivers (S1 and C8), and generally late-colonized. This indicates that a long accessible river length was necessary for the settlement of a panmictic functional population. Indeed, most populations associated with River type I were in genetic equilibrium. It can be estimated that the populations of the Chateau area have larger effective size than in the Studer zone (i.e., higher trout density on a longer colonized zone, Davaine P, unpublished data), and thus time to migration/drift equilibrium is longer for the Chateau area, promoting low genetic diversity between populations due to important gene flow. River type II is associated with medium genetic variability and little genetic disequilibrium. However, the strong assignment probability of all colonized rivers (even from the Chateau area) to the Studer Sea trout sample (and not the Juveniles sample) indicated that the existing pool of migrants in the Studer area is highly variable, possibly even carrying individuals coming from the Nord or Chateau areas. Despite this variability, landscape features in the Studer area enhanced the genetic divergence caused by initial founder effect, either by limiting the access of migrants in the rivers (Coast type) or because rivers associated with River type II and III have already reached their carrying capacity. The low efficiency of colonization in the Studer area by individuals born in a different environment could also indicate local adaptation.

Short-and Long-Term Consequences for the Evolution of Brown Trout Populations
Small-range colonization patterns in Courbet remain unclear, particularly the extent of ''step-by-step'' exchange between rivers in both zones. The analysis of historical samples (scale collection) could help to better understand the patterns of drift/migration processes over time and make hypotheses about the future evolution of these rivers (Nielsen and Hansen 2008). The colonization process in Courbet Peninsula is still at a very early stage. Rivers have mostly been colonized in the last 20 years, which correspond to 5-6 generations at the most, and the process has probably not reached migration/drift equilibrium. Persistence of functional populations in colonized rivers and evolution of population structure patterns will depend on population connectivity and subsistence of large core populations, particularly for small, isolated populations with reduced habitats (Hildebrand 2003).
As many salmonids, brown trout exhibits considerable variation in life history over its native range: stream-resident, lake-run, (nonanadromous), and sea-run (anadromous). Salmonid domestic stocks have also been shown to diverge from wild populations, both genetically (Laikre 1999) and phenotypically (Fleming et al. 1994). Several studies of interaction between domestic escapees and wild congenerics in salmonids have shown a lowered fitness of domestic individuals in the wild (McGinnity et al. 2003;Araki et al. 2007). In the case of the Courbet Peninsula, in the absence of wild competitors, brown trout has managed to quickly develop functional populations. Moreover, introduced and subsequent colonizing populations have managed to develop a whole range of ecological behavior in a limited numbers of generation, which allowed for an efficient colonization of the area. Similar behavior plasticity in the adaptation to new environments was observed in sockeye salmon (Oncorhynchus nerka) after its introduction to New Zealand (Quinn et al. 1998). In chinook salmon (Oncorhynchus tshawytscha) that has colonized New Zealand since its introduction at the beginning of the 20th century, Quinn et al. (2001) have shown that New Zealand chinook populations diverge today both in genetic variability and in phenotypic traits. Although this divergence probably initially resulted from phenotypic plasticity, the authors suggest that divergence may also occur in the very early stage of the colonization process, through a ''favored founders effect'' mediated by local environmental conditions. Brown trout populations in Courbet could be at this early divergence stage. There are evidences for differences in some phenotypic traits between Studer and Chateau populations (for instance delayed migration, delayed sexual maturation, and slower growth in Studer; Davaine P, unpublished data). Given the limited generation time since colonization started, it is likely that the observed life-history changes (migratory behavior) are triggered by environmental or demographic cues, rather than under genetic control. However, evidences start to appear that, in salmonids, lifehistory tactics have the potential to evolve in response to selection acting on the tactic itself or indirectly via selection on body size (Thériault et al. 2007).
The ability of brown trout to colonize new habitats, as demonstrated in the case of the Kerguelen Islands, is also interesting on a wider temporal and geographical scale. Global changes, and associated ice melting phenomenon, could lead to the opening of new freshwater ecosystems in temperate or polar areas (MacDonald et al. 2005). The Kerguelen situation is a unique opportunity to study the vulnerability of these new environments to colonization by freshwater fishes, among which salmonids are the first representant (Babaluk et al. 2000). Ongoing survey of brown trout populations in Kerguelen will help to understand the various strategies (ecology, biology, and behavior) that the species can develop to colonize these newly accessible rivers, depending also on habitat and landscape characteristics.