Farm salmon escape event: levels of farm/wild hybridisation

Materials and Methods

A genetic survey was undertaken in 2020 and 2021. In both years, fin clip tissue samples of fry (0+ fish) were collected around the Scottish west coast in the vicinity of the escape event. In 2021, at the request of the Environment Agency, samples of both fry and parr were collected from English northwest coast rivers. These samples were used to:

• Estimate the prevalence of introgression in juvenile salmon populations in the vicinity of the North Carradale fish farm prior to any potential influence of the 2020 escape event.
• Estimate the prevalence of introgression following the 2020 escape event.
• Compare the prevalence of introgression within the rivers between years.

Survey design

To ensure that the samples collected were representative of salmon populations in the Scottish survey area, and were not affected by sampling biases, a formal statistical survey design was employed. This survey design followed the methodology of the National Electrofishing Programme for Scotland (NEPS) (Malcolm et al., 2020) and used a Generalised Random Tessellation Stratified (GRTS) (Stevens and Olsen, 2004; Kincaid and Olsen, 2017) design which provided a spatially balanced survey across the area of interest.

The survey design employed was an equal probability stratified survey with over-samples. The sample frame (i.e. the rivers that could potentially be sampled) included all rivers greater than Strahler River Order 2 that were below impassable barriers for salmon. The design included sample sites split across four strata; Carradale/Arran (East coast of Kintyre peninsula and West coast of Arran), Argyll, Clyde (including Lomond) and Ayrshire (Figure 1). The sites were surveyed in the autumn/winter of 2020 and early spring of 2021 prior to any effects of the 2020 farm escape, and again in the autumn/winter of 2021 after escaped fish would have had the opportunity to spawn.

Together with the GRTS sites, additional sites were added to the survey (termed here: ad-hoc sites). These were: 1) additional sites within the GRTS strata, but not part of the formal survey design, that were of particular interest to the local fisheries managers in these areas; 2) sites requested by NatureScot on the river Bladnoch, which is designated a Special Area of Conservation (SAC) with regard to Atlantic salmon; and 3) sites requested by the Environment Agency on SAC rivers along the northwest coast of England, which may also have been impacted by the escape event. In the case of the English rivers, sampling was undertaken in 2021 only.

Sample collection

Fish were sampled by the staff from the representative bodies in each area (Argyll Fisheries Trust, Ayrshire Rivers Trust, Clyde River Foundation, and Loch Lomond Fisheries Trust, NatureScot, Environment Agency). In Scotland, caudal fin clip samples were collected from fry (juvenile salmon spawned in the year of collection) in both 2020 and 2021. This approach allowed direct comparison between years using the same age class. The samples collected in 2020 provided a ‘background’ point estimate of prevalence of introgression and numbers of first generation (F1) farm/wild hybrid fish at the sites. The samples collected in 2021 included fry whose genetic composition could potentially have been impacted by any escapees breeding with wild individuals, with the production of F1 hybrids in this cohort. Numbers of F1 hybrid fish were compared across the two sample years to identify any impacts. In order to provide direct comparisons with previous studies of introgression in Scotland (Gilbey et al., 2021), the same sampling protocols were followed. Target samples of up to 30 fry were to be obtained from each site. Where no fry were captured within 10 minutes of arrival at a site, surveyors proceeded to use over-samples (spare sites that maintain the integrity of the overall design) to maintain the targeted site numbers.

There was no opportunity to sample in 2020 in England as the English SAC rivers were not included in the survey until requested by the Environment Agency in 2021 and, so, both parr and fry were collected in 2021. Here, levels of introgression and numbers of F1 hybrid fish were compared between age classes, taking into account the potential influence of differential mortality between age classes of farm/hybrid/wild individuals, which is known to occur (e.g. Skaala et al., 2012; Solberg et al., 2015; Glover et al., 2017).

Genetic analysis

Genotyping of wild samples

All fish were genotyped using a set of 74 Single Nucleotide Polymorphic (SNP) genetic markers, developed to maximise the power to detect and quantify levels of introgression between farmed fish of Norwegian origin (which the escapees in this case were) and wild Scottish fish (Gilbey et al., 2021). DNA was extracted from fin tissue using a Chelex extraction protocol (Walsh et al., 1991). Genotyping was carried out on a Fluidigm EP1 platform (Fluidigm, San Francisco, CA, USA) following the manufacturer’s protocols. Samples with < 95% successful genotype calls were removed from the analysis.

Family structure

The family structure of the samples was examined at each site. This step was undertaken as samples containing many full-sibs (sharing both parents) may influence interpretation of results. This is of particular importance, as in most cases the samples consisted of fry alone and, as such, dispersal from the redd may have been limited compared to older fish (Eisenhauer et al., 2020 and references therein). For each site, the presence of full-sibs was examined using maximum likelihood estimations as implemented in the COLONY 2.0.6.6 software package (Jones and Wang, 2010). Numbers of full-sibs in each year were compared using paired-t tests carried out in R (R Core Team, 2015).

Reference samples

Hybrid fish were identified by comparing individual genotypes of the wild sampled fish to two sets of reference samples using the approaches outlined in Gilbey et al. (2021) following the procedures described below.

This analysis was performed using the same wild reference panel as used in Gilbey et al. (2021) based on wild fish captured from around Scotland as these have previously been shown to be representative of Scottish wild fish. However, within this study, a new set of farm reference samples were used. These new farm reference samples were obtained from Mowi and comprised tissue samples from 100 fish of the same breeding line as the escaped fish (Mowi pers. comm.). Using these reference fish allowed maximum power for the analytical quantification of genetic interbreeding from the specific escape event. The farm fish were genotyped for the 74 SNP markers as described above, and reference centroids produced consisting of 100 in silico generated reference fish produced in HYBRIDLAB 1.0 (Nielsen et al., 2006), as described in Gilbey et al. (2021).

Simulating farm x wild crosses

To identify first generation (F1) crosses, a set of simulated F1 crosses were produced. Their hybrid status was then determined, as described below for wild fish. Genotypes of 500 simulated F1 hybrid fish were produced in HYBRIDLAB by random mating between Mowi farmed fish of the same breeding line as the escapees and wild fish captured in 2020 only. Such crossings provided the most accurate match to any F1 fish that may have been captured in 2021 and which might be a result of mating between the actual escapees and wild fish in 2020. The distribution in individual levels of introgression measured in the simulated F1 fish provided confidence bounds with which to classify any 2021 F1 hybrid fish captured in the wild (mean ± 2 x standard deviation).

Estimating levels of introgression and identifying first generation hybrids

The proportion of the genome of an individual fish that is of farm strain origin was estimated by comparing the genetic signature of that fish against the reference samples. The probability of belonging to the wild reference sample, P(wild), was then determined using a systematic Bayesian clustering approach applying Markov Chain Monte Carlo (MCMC) estimation, as implemented in the STRUCTURE 2.3.4 software package (Pritchard et al., 2000). This was performed using 50,000 repetitions as burn-in, followed by a further 100,000 repetitions with no a priori information of sampling locality/origin, and assumed two populations (wild and farm). Each fish was analysed separately, together with the farm and wild reference populations, to prevent biases that may be introduced if all samples were included in a single analysis (Kalinowski, 2011; Karlsson et al., 2014). For each fish, the probability of belonging to the wild centre point P(wild) was individually calculated and recorded. Expected P(wild) probability distributions range from around 1 for pure wild fish to around 0 for pure farmed fish, with F1 hybrids (first generation) having P(wild) values distributed around 0.5 (Gilbey et al., 2021). To compare levels of introgression to that observed previously (Gilbey et al., 2021), the same hybrid identification cut-off P(wild) of 0.747 was used (i.e. the value below which the fish is classified as a hybrid). However, some care must be taken in interpreting these observations, as the cut-off used in the previous analysis was developed based on different reference samples than those used here. However, it is utilised here to allow the observations in the current analysis to be put into a qualitative comparative context with the previously published work.