Study populations and sampling
The location of the four study populations (COU, SEL, SEE, and SIE) is presented in Fig. 1. These populations declined in the 1950s probably due to important habitat degradations. COU population was considered extinct (no more natural reproduction) in the 1970s. As a result, supplementation operations have started in 1989 using non-native young-of-the-year (0+) produced in the Favot Hatchery, Brittany. Spawners caught every year in the Aulne River (AUL, Fig. 1) were used to produce juveniles that were stocked in autumn and winter (i.e., at 9–12 months). The four BMS rivers were stocked with variable number of individuals, as indicated in Table 1. From 1989 to 1994 and from 1996 to 2003, approximately 931 000 fish produced by AUL progenitors were stocked (Table 1). In 1995, no fish originating from AUL were stocked but instead about 80 000 juveniles originating from progenitors caught in the Gave d'Oloron River (GAV) were released. COU has been stocked from 1989 to 2003 with a total of 623 518 fish. SEL has been stocked from 1989 to 1996 with a total of 363 500 fish. SEE and SIE were stocked once in 1990 with 14 000 and 10 000 individuals, respectively.
Figure 1. Map of study rivers with sample sizes in each site. Aulne and Gave d'Oloron populations have been used to stock Couesnon, Sélune, Sée, and Sienne rivers.
Download figure to PowerPoint
Table 1. Demographic and stocking data for the Couesnon, Sélune, Sée, and Sienne populations from 1989 to 2009. Stocked fish originated from Aulne except in 1995 where individuals from Gave d'Oloron were used (Given in bold). Estimates of adult population sizes and average 0+ autumn parr productions are indicated
| ||Average population size||Average 0+ parr production||Stocking|
|COU||200||4000||22 000||53 500||25 000||64 500||25 000||16 600|| 50 267 ||43 197||89 020||59 665||48 200||30 782||35 049||33 172||27 566||29 381||23 585||22 988||25 519||20 090||27 094|
|SEL||300||6000||4000||36 000||25 900||30 000||101 000||66 000|| 29 800 ||61 000||9800||–||–||–||–||–||–||–||–||–||–||–||–|
|SEE||800||16 000||–||14 000||–||–||–||–||–||–||–||–||–||–||–||–||–||–||–||–||–||–||–|
All samples were scales collected on anadromous adults caught by anglers and stored by INRA (Institut National de Recherche Agronomique) and ONEMA (Office National de l'Eau et des Milieux Aquatiques). The age of each individual was determined from its scales. As post-stocking samples, we focused on cohorts 2002–2003 (Fig. 1, Table S1). For prestocking samples, we chose cohorts 1977–1978 (SEE and SEL) and 1985 to 1987 (SIE, SEE and SEL). For the two populations where progenitors were collected for the hatchery, we chose samples from cohorts 1969 and 2003 for AUL, and 1984 and 2003 for GAV. A total of 545 samples (from 25 to 79 individuals per temporal sample) were genotyped.
Molecular data analyses
We used Micro-Checker 2.2.3 (Van Oosterhout et al. 2004) to assess the frequency of null alleles and scoring errors attributed to stuttering or large allelic dropout. Allele number and allelic richness were obtained using Fstat 184.108.40.206 (Goudet 1995). Tests for linkage and Hardy–Weinberg Equilibrium (via F IS) were also conducted with Fstat. Expected heterozygosity, He, (Nei 1978) and observed heterozygosity, Ho, were calculated with Genetix 4.05.2 (Belkhir et al. 1996). Fdist 2 (Beaumont and Nichols 1996) was used to verify the neutrality of the markers used. Pairwise F ST and tests of differentiation were conducted in Fstat. Pairwise Nei (Da) genetic distances (Nei et al. 1983) were estimated using Populations 1.2.30 (http://bioinformatics.org/~tryphon/populations/).
As per the studies of Hansen and Mensberg (2009), Marie et al. (2010), and Winkler et al. (2011), admixture between wild and hatchery stocks was estimated at the individual level using the Bayesian method implemented in the Structure software (Pritchard et al. 2000). Structure analyses were performed assuming an admixture model with default settings (i.e., no informative prior was used). We ran Structure from 1 to 6 genetic clusters (k) with 15 replicates for each k. Each run started with a burn-in period of 50 000 steps followed by 300 000 Markov Chain Monte Carlo replicates and estimating 90% credible intervals. We selected the k with the highest likelihood (Pritchard et al. 2000) and according to the ∆k method (Evanno et al. 2005). We then calculated average population admixtures at the best clustering solution. We also estimated an overall admixture in the four BMS populations calculated as the sum of the admixture of each population weighted by its size (number of returning adults).
We used Nemo 2.1.0, a stochastic individual-based genetically explicit framework (Guillaume and Rougemont 2006) to simulate the evolution of a metapopulation made of wild and hatchery individuals. Our aim was to produce expected patterns of admixture based on various demographic scenarios, which could be compared with the observed data. An individual-based approach was used to produce data as similar as possible to molecular data.
Modeled organisms were diploid with separate genders and lived in a structured metapopulation of six demes with the following local carrying capacities: 2000, 7000, 200, 300, 800, and 400 adults, corresponding to the GAV, AUL, COU, SEL, SEE, and SIE populations, respectively (see also Table 1 and Fig. S1). These carrying capacities were assumed to correspond to the population sizes of COU, SEL, SEE, and SIE estimated from average 0+ parr abundance data (Anonymous 2008). These data were collected by local freshwater fishery organizations through annual electrofishing campaigns between 2001 and 2008. To estimate adult population sizes, we first applied a survival rate of 50.3% between 0+ parr and smolt stages estimated by Baglinière et al. (2005) over the 1995–2003 period in the Oir River, a tributary of the SEL River. Second, we applied a 9.67% survival rate between smolt and adult stages, which was estimated by Etienne Prévost (unpublished data) in the Scorff River (French index river for International Council for the Exploration of the Sea) over the 1995–2008 period. The final estimates were rounded to the nearest ten (Table 1). For GAV and AUL, we chose artificially high population sizes to allow high dispersal rates simulating stocking operations. To assess the sensitivity of our approach to varying population sizes, we also implemented eight additional values in BMS populations for one of the most probable scenarios (C7, see Table 2). The four initial population sizes were multiplied by 0.5, 0.7, 0.8, 0.9, 1.1, 1.2, 1.3, and 1.5 (Table 2).
Table 2. Description of combinations of parameters used for the different simulated scenarios. Survival rates are indicated as ratios between stocked and wild fish (the coefficient by which the survival of stocked fish was divided relative to wild individuals is given in brackets). Pairs of dispersal rates are given for wild and stocked individuals. Different mating systems and population sizes were further investigated for one of the most probable scenarios (C7)
|Survival rate of stocked fish relative to wild individuals||Dispersal of wild; stocked fish|
|0.06; 0.06||0.06; 0.15||0.15; 0.15||0.15; 0.24|
|0.05 (20)||A7||B7||C7a , b ||D7|
We implemented the following semelparous life cycle: (i) breed; (ii) dispersal; (iii) random regulation of local populations size, which reduced the pool of competing individuals to the local carrying capacity (with equal sex ratios); (iv) reproduction during which females were assigned a fecundity value drawn from a Poisson distribution with a mean value of 40 offspring and mated with one randomly chosen male (monogamy). We also investigated the potential effect of multiple mating by testing 20% and 50% of polyandry for one of the best scenarios. Adults died after reproduction and the cycle started again. This assumption of semelparity is realistic as the estimated proportion of multispawners in BMS salmon populations is extremely low: from 0.9% to 2.1% for the 1992–2002 and 1972–1982 periods, respectively (Baglinière et al. 2004). We simulated 17 unlinked neutral loci with initially 15 alleles per locus, a mutation rate u = 0.0001, and a recombination rate U = 0.5. Alleles were inherited randomly (i.e., no linkage or epistasis) and sex was set randomly (equal sex ratio). F-statistics and multilocus genotypes were recorded every generation. There was no dispersal between AUL, GAV, and the four other BMS populations before stocking events.
Using dispersal values reported in the literature for Atlantic salmon (Jonsson et al. 2003a; Pedersen et al. 2007), we implemented four pairs of dispersal rates for wild and stocked fish among BMS populations: 0.06/0.06, 0.06/0.15, 0.15/0.15, and 0.15/0.24, respectively (Table 2). These values correspond to the total dispersal from a given population with, for example, a rate of 0.06 meaning a 0.02 fraction of migrants from a river move into each of the three others.
We investigated the effects of a lower survival of hatchery fish compared with the wild individuals by testing 11 hatchery/wild ratios of survival, from 1 to 0.01 (Table 2). Dispersal rates from GAV and AUL to BMS populations were modified at each generation to simulate the evolution of stocking practices. We simulated 44 scenarios with varying combinations of survival and dispersal rates, and for one of the most probable scenarios, we tested the sensitivity of our approach to variations in population size and mating system (Table 2). Each configuration was run three times for 290 generations to reach equilibrium and obtain a genetic structure similar to the observed prestocking situation. Then, we implemented stocking events from the generation 291 on according to the stocking data and we ran the program until generation 296. In BMS salmon populations, most juveniles spend 1 year in freshwater and most adults spend 1 year at sea and spawn at the end of their third year (Baglinière et al. 2005). Given this predominant generation time of 3 years, we modified the dispersal matrix at each generation according to the stocking data averaged over three consecutive years (generation 291: 1989–1991; g292: 1992–1994; g293: 1995–1997; g294: 1998–2000; and g295: 2001–2003). We used adults born at the generation 295 for admixture analyses.
Three Structure runs were conducted for each Nemo run at the best clustering partition (k=3) with the same settings used for molecular data. To estimate population admixtures, we averaged admixture values over the three Structure runs and then averaged values of the three Nemo runs. To identify the simulated scenario(s) that best fitted the observed data, we used two approaches. First, we used chi-square tests to test whether mean-simulated admixture values for each scenario significantly differed from observed values. Second, we used a least-squares approach: we summed the squared differences between observed and simulated individual admixture values for each scenario in each population and defined the best scenario as the one with the smallest sum of squared differences. The best scenario for the situation where admixture rates were averaged over the four populations was defined as the smallest of the four sums of squares, which were divided by the corresponding number of individuals in each population. We also used factorial correspondence analyses computed with Genetix 4.05.2 to display the genetic structure among samples and compare observed and simulated data.