Atlantic salmon from large Scottish east coast rivers - genetic stock identification: report

Discussion

Analysis of this dataset has indicated the presence of four genetic clusters, consisting of both individual rivers and larger regions spanning multiple rivers. However, despite significant phylogenetic structuring being present, the levels of differentiation between the units identified were not large enough to translate into robust river-specific assignments using GSI. Indeed, the same three assignment units as had previously been observed were identified, two comprising single river systems (Oykel/Cassley/Shin and Tweed) and the other consisting of all the other sampled North East coast rivers (Gilbey et al., 2016a).

The samples screened were collected as part of NEPS programme, whereby sites were selected using a Generalised Random Tessellation Stratified (GRTS) design rather than targeted sampling based on knowledge of areas of high juvenile density. Such a design allows unbiased sampling to be undertaken across the area of interest. In some cases, however, this resulted in low numbers of fish caught in a number of sites. Notwithstanding this, the aim of the analysis was to characterise populations at the river level, for which the GRTS design delivered an unbiased sampling of sites across the different rivers. It was important that this was achieved, as it is well known that there is population genetic variance within rivers. The design allowed for this variation to be captured utilising the fixed number of samples to be screened within the project.

Microsatellites can be highly variable, with large numbers or variants (alleles). As a result, sufficient sample numbers are required to ensure that all variation has been captured and to be able to accurately estimate allele frequencies. To address this issue, we combined geographically close sites to represent the wider river area. Combining sites in this way may have resulted in disruption of the unique genetic signature of a site (as reflected in the number of sites out of Hardy-Weinberg equilibrium). This may, in turn, have compromised, to some degree, the power to differentiate between sites. However, this study focused on the assignment potential at the river-level, not at the site-level.

For technical reasons during the panel development, the 101 microsatellites were initially selected to be limited in numbers of alleles in comparison to most microsatellites, with an average of eight alleles per marker as reported by Bradbury et al. (2018). Having a lower number of alleles would reduce the impact of small sample sizes, though may not have completely alleviated them. However, considering the close match with previous analysis of the regions with differing numbers of sites/fish and different marker types, it is unlikely that such influences have had a significant impact on the results of the analysis.

Out of the 101 microsatellites, 69 produced reliable and high quality genotypes. Though the omission of nearly a third of the original markers could possibly have influenced the results, we consider that it is unlikely that the addition of the remaining ones would have substantially altered the findings. The markers are randomly distributed throughout the salmon genome (Bradbury et al., 2018) and are generally considered adaptively neutral. A larger number of markers would have resulted in an increased number of alleles, which would, in turn, have increased the power of the assignment analysis. However, power of assignment would not necessarily have related to higher accuracy at the levels of differentiation observed here. That said, the effect of microsatellite number on assignment success could be investigated in the future.

The most distinct sites, with the highest differentiation values from the other sites were Dee_345, Don_1934 and Don_lower, whilst the lowest differentiation was found between and among sites within the rivers Dee, Spey and Deveron. However, despite the low levels of differentiation observed within and between those latter three rivers, many pairwise comparisons were still significantly different, suggesting that these sites harboured genetically distinct populations. Furthermore, based on this panel of neutral markers, an absence of statistical significance in differentiation between sites does not equate to a lack of natural and adaptive differences between these populations. Indeed, previous work has shown both differences between upper and lower parts within each river (Cauwelier et al., 2018a), as well as differences in a set of adaptive markers related to adult return timing (Cauwelier et al., 2018b).

The structuring analysis revealed four genetic groups, two of which included sites from a single river. Previous work has shown the Oykel/Cassley/Shin river system to be genetically distinct from the other East coast rivers further south (Gilbey et al., 2016a). Similar, though less pronounced, differentiation has previously been reported between the River Tweed and the more northern East coast rivers (Gilbey et al., 2016a,b). The third cluster, consisting of sites from both the rivers Don and Dee, including those three most distinct sites, has not been observed before. It is possible that the new panel of microsatellites has resulted in higher resolution amongst this Northeast coast region at these two rivers. However, successful genotyping proved difficult for samples from these Dee and Don sites compared to the remainder of the dataset. All samples were processed using our laboratory standard operating procedures and samples from the various rivers were spread over multiple plates/runs, so as to minimise processing effects related to reagents, staff member etc. Internal QC checks indicated that the quantity of the DNA was insufficient and/or the quality poor from samples originating at these sites, with average scoring percentage for those sites (69.5% ± 6.9%) significantly lower (t-test: t = 19.9, df = 38, p < 0.0001) than for the others (96.1% ± 1.8%). The reasons for the lower DNA quality/quantity from these samples remain unclear. Further analysis would be required to disentangle any issues with sample quality from the potential for the microsatellite panel to have identified a true distinct genetic cluster not seen in previous work.

The fourth cluster encompassed all the other sites in the dataset, from the River Spey down to the River Tay. Despite the absence of clear structuring between these rivers and an associated inability to robustly assign fish to them, as reported here and in previous work, within-river differences have been observed (Cauwelier et al., 2018a, b). Though this picture is not obvious here, there is an indication that the proportion of “Cluster 2” increases downstream, whilst the proportion of “Cluster 4” tends to decrease (Figure 2B). Again, further, more detailed work would be required to assess this in more detail.

Despite the clustering analysis showing a number of distinct groups, assignment success at the river level did not meet the accuracy threshold, with the exception of the Oykel/Cassley/Shin river system and the River Tweed. Rivers, therefore, had to be combined to represent larger geographic areas and the final set of assignment units was similar to those found previously (Gilbey et al., 2016b). This would suggest that, despite there being genetic differences among sites and rivers, the differences were not sufficient to translate into high assignment success.

The salmon populations from the river Tweed form an interesting complex. Indeed, the Tweed sites have previously both been combined with either the River Tay populations (Gilbey et al., 2016b) or within the larger NE assignment unit (Gilbey et al. 2016a), which also included the River Tay. Assignment success for fish originating from the River Tweed has, in all these cases, been around 80%, this also being our threshold value for combining (or not) rivers into larger units. This would suggest that the genetic differences between salmon populations from the River Tweed and the other NE coast rivers are borderline to being large or not large enough to result in robust assignment.

A number of genetic marker panels have now been employed in rivers covering the East coast of Scotland and all have failed to provide high resolution genetic stock identification, despite there being indications of population structuring within and between these rivers. These panels have consisted of neutral markers that are not directly associated with specific traits and their variants are thus not considered to confer any fitness differences.

An alternative approach to the use of such markers is to examine variation in DNA sequences that are part of genome regions coding for proteins and are linked to trait variations and, as such, are subject to natural selection (i.e. adaptive markers). These adaptive markers could be identified using a whole-genome sequencing approach. In this method, an organism’s whole genome is sequenced and, when applied to multiple individuals and populations, this approach could result in both the detection of neutral and adaptive population divergence (Larson et al., 2014; Narum et al., 2018). This technique has been shown to have the potential to enhance GSI resolution (Fuentes-Pardo et al., 2020).

Recently, other non-genetic based methods have been used for stock identification. These methods examine environmental differences between locations that get incorporated into a number of structures within growing fish, such as their scales, otoliths and eye lenses in the form of trace elements (Veinott & Porter, 2005; Marklevitz et al., 2011; Perrier et al., 2011). Similar principles are applied here as in genetic-based approaches, whereby a trace elements baseline is created to which unknown individuals are assigned. This approach has been employed to successfully distinguish and assign Chinook (Oncorhynchus tshawytscha) (Marklevitz et al., 2011), brown/sea trout (Salmo trutta) (Ramsay et al., 2011; Veinott et al., 2012) and Atlantic salmon (Veinott & Porter, 2005; Perrier et al., 2011).

It may be that enhanced stock identification could be achieved by the combination of both genetic and trace element analysis. Perrier et al. (2011) combined a panel of six microsatellites markers with trace element analysis gathered from otoliths and found high levels of accurate assignment (83% - 100%) to river of origin using this combined approach, even though assignment success based solely on the genetic markers varied between 25% and 47%. This approach of combining both genetic markers and environmental differences reflected and traced in scales/otoliths could be explored in future to assess if enhanced stock identification can be achieved within this large and biologically and commercially important region of Scotland.