Small coastal streams—Critical reservoirs of genetic diversity for trout (Salmo trutta L.) in the face of increasing anthropogenic stressors

Abstract We used microsatellite markers to investigate levels and structuring of genetic diversity in trout (Salmo trutta L.) sampled from 16 rivers along the south coast of Cornwall in southwest England. This region is characterized by many small coastal streams with a few larger catchments. At a regional level, genetic structuring of contemporary populations has been influenced by a combination of events, including the last Ice Age and also more recent human activities over the last millennium. All populations are shown to have gone through strong genetic bottlenecks, coinciding with increased exploitation of mineral resources within catchments, beginning during the Medieval period. At more local levels, contemporary human‐induced habitat fragmentation, such as weir and culvert construction, has disproportionally affected trout populations in the smaller catchments within the study area. However, where small catchments are relatively unaffected by such activities, they can host trout populations with diversity levels comparable to those found in larger rivers in the region. We also predict significant future loses of diversity and heterozygosity in the trout populations inhabiting small, isolated catchments. Our study highlights how multiple factors, especially the activity of humans, have and continue to affect the levels and structuring of genetic diversity in trout over long timescales.

through fixation due to drift or selection. In a study of particular relevance to the topic, Spielman, Brook, and Frankham (2004) explored the impact of genetic factors on extinction risk for threatened populations and species and articulated the idea that reduced population genetic diversity correlates with reduced reproductive fitness and an elevated risk of future extinction linked to genetic factors. More generally, Spielman et al. (2004) linked the degree to which a population is threatened with population size, with small populations being more likely to be classified as threatened than large populations.
Freshwater habitats are among some of the world's most threatened, suffering from five major human-mediated threats, namely over-exploitation, pollution, modification of flows, habitat degradation, and the spread of invasive, non-native species (Dudgeon et al., 2006).
These anthropogenic impacts directly affect the physio-chemical conditions within freshwater habitats and, consequently, have strong influences on the aquatic biota within these habitats (Schinegger, Trautwein, Melcher, & Schmutz, 2012) affecting biodiversity at multiple levels. However, while in freshwater systems-as in their terrestrial counterparts-the strength of the effect of these human impacts on intraspecific diversity and differentiation will be dependent on population size and magnitude of selection (Einum, Fleming, Cote, & Reynolds, 2003;Frankham et al., 2017), the size of a river catchment will also play a potentially critical role, with populations in smaller catchments being particularly at risk (Consuegra, Verspoor, Knox, & García de Leániz, 2005;Whelan, 2014). Moreover, there are a number of specific threats from which fish populations in small streams are likely to suffer more than larger rivers. Of particular interest are issues with access for anadromous fish species, connectivity between small streams and larger catchments or the sea, the impact of barriers to fish movement and water flow fluctuations (Griffiths, Koizumi, Bright, & Stevens, 2009;Palm, Laikre, Jorde, & Ryman, 2003;Thorstad, Økland, Aarestrup, & Heggberget, 2008). Barriers, both natural and man-made, can impact rivers by dividing continuous habitat into smaller patches (Jones et al., 2019). From a genetic perspective, this subdivision can have multiple adverse effects on fish populations. Barriers that prevent movement of fish between habitat patches can result in reductions in both census and effective population sizes and increased inbreeding, which, in turn, can lead to reduced levels of genetic diversity, and an increase in genetic structuring (Frankham et al., 2017;Griffiths et al., 2009;Montgomery et al., 2000;Palm et al., 2003).
Anadromous salmonid species, known for their strong homing fidelity to their natal rivers (Keefer & Caudill, 2014), are often found in spatially structured metapopulations (Schtickzelle & Quinn, 2007). This philopatry is highly adaptive, increasing the likelihood that fish will find suitable spawning and juvenile habitats (Keefer & Caudill, 2014) and giving rise to the aforementioned spatially structured metapopulations. However, straying of fish to non-natal rivers also occurs (King, Hillman, Elsmere, Stockley, & Stevens, 2016;Valiente, Beall, & Garcia-Vazquez, 2010) and is recognized as an important evolutionary feature of salmonids, especially in range expansion and colonization of newly open habitats, and can facilitate gene flow between rivers (Horreo et al., 2011;Keefer & Caudill, 2014).
The role of small streams (for which there is no universally accepted definition; Biggs, Nicolet, Mlinaric, & Lalanne, 2014) in the ecology and population genetics of brown trout (Salmo trutta L.) is not well understood (Thomson & Lyndon, 2018;Whelan, 2014). There is a disproportionately large number of these streams in southwest England, where the Devon/Cornwall peninsula precludes the development of large dendritic catchments and it has been suggested that the relatively small populations of trout residing in such small streams may collectively make a significant contribution to the genetic diversity of the species, particularly in relation to those fish exhibiting an anadromous (sea trout) life cycle (Consuegra et al., 2005;Whelan, 2014), albeit with the caveat that small populations also increase the effects of genetic drift, often leading to distinct but genetically depauperate populations (e.g., Paris, King, & Stevens, 2015;Perrier, Ferchaud, Sirois, Thibault, & Bernatchez, 2017). Indeed, while these coastal streams may not be good holding habitats for resident trout, they may offer significant areas of spawning habitat for anadromous fish that are able to access them.
In general, small populations are more likely to suffer from the detrimental effects of genetic bottlenecks, inbreeding, and genetic drift. These processes can lead to the loss of genetic diversity and inbreeding depression (Vandewoestijne, Schtickzelle, & Baguette, 2008). Ultimately, this loss of genetic diversity can lead to a reduction in individual fitness and an increased risk of local population extinction (Vandewoestijne et al., 2008). The effects of genetic drift and inbreeding can be counteracted by both gene flow from other populations and mutation. Over recent timescales (i.e., since the end of the Quaternary glaciations), however, insufficient time has elapsed and mutation rates are generally too slow for mutation to have contributed significantly to increasing levels of genetic diversity (Ho & Larson, 2006). Gene flow, therefore, is the only process that can be relied upon to maintain diversity within small populations. For gene flow to be effective, there has to be a high degree of connectivity between populations, which can be a major problem for populations inhabiting rivers with significant barriers to upstream migration.
In this paper, we explore the genetic structure of-and connectivity between-trout populations in small streams and larger catchments along the south coast of southwest England. This coast is characterized by a few larger catchments and numerous small streams, and resident trout were sampled from both for genetic analysis. This approach allows the genetic structure of small populations of coastal stream trout along this section of coast to be set in a wider context and enables us to address the relative importance of population size and demographic factors in shaping contemporary patterns of genetic variation in trout in small streams.  Figure 1, Table A1). Fish were caught during routine Environment Agency electrofishing surveys. The sampling scheme employed was designed to reduce the collection of potentially related individuals by targeting 1+ or older fish. However, in some sample locations where fish densities were low, fry were also sampled to increase sample sizes (and were later excluded from the analysis if found to be part of a full-sib group-see below).

| Sample collection
Fish were anaesthetized using MS-222 prior to removal of adipose their fin using sharp scissors. Fin clips were transferred immediately into tubes containing absolute ethanol. Genomic DNA was extracted from fin tissue following the method of Truett et al. (2000).

| Analysis of genetic structure
Two different analyses were used to test for population struc- Because the STRUCTURE analysis showed there was no evidence of within-river genetic structuring (see Section 3), the DAPC analysis was conducted on the data for each river rather than individual sampling locations. As with STRUCTURE, we performed hierarchical analyses where distinct "outlier" populations were successively removed from the data set and the DAPC repeated. An outlier population was one that exhibited a somewhat distinct cluster of individuals based on plots of DAPC1 v DAPC2 and DAPC1 v DAPC3.
To test for isolation by distance (IBD), a Mantel test (Mantel, 1967) was used to evaluate the relationship between linear

| Bottleneck analyses
Evidence for the presence of genetic bottlenecks in each of the sampled rivers and groups identified in the STRUCTURE analyses was assessed using both the heterozygote excess (Cornuet & Luikart, 1996) and M-ratio (Garza & Williamson, 2001) methods. The program BOTTLENECK (Piry, Luikart, & Cornuet, 1999) was used to test for an excess of heterozygotes against expectations for a population at mutation-drift equilibrium under the two-phase mutation (TPM) model of microsatellite evolution. We set the proportion of multistep mutations in the TPM model to 20% and the variance of TPM to 30%. Significance was tested using the Wilcoxon's sign-rank test for a one-tailed heterozygosity excess. Allele frequency distributions were also examined to determine whether mode shifts had occurred. We also calculated M, the ratio of the number of alleles at a given microsatellite locus to the allelic size range for that locus (Garza & Williamson, 2001). We also investigated long-term changes in effective population size (N e ) using VarEff v1.2 (Nikolic & Chevalet, 2014). The program estimates temporal changes in N e using an MCMC approach.
Estimates of effective size were generated from sampling time to 300 generations in the past (1,200 years assuming a generation time of 4 years; Jensen et al., 2008). We set the effective size prior to 10,000 and used the two-phase model of microsatellite evolution with the proportion of multistep mutations set to 0.2 and assuming a mutation rate of 5 × 10 -4 (Paris et al., 2015). The length of burn-in was 10,000 steps with data generated from 10,000 batches of length ten with a sampling interval of ten steps giving a total of 10 6 data points.
Data were analyzed for each individual river, and the three multiriver groups identified by the STRUCTURE analysis. VarEff requires microsatellite loci to have three or more alleles. Therefore, the data for locus One102b were removed. Additionally, trout populations in some rivers possessed only two alleles for Ssa85. For consistency, data for this locus were also removed for all rivers and groups.

| Gene flow analyses
Two methods were used to estimate historical and contem- The mixing parameters ΔA, ΔF, and ΔM were each set to 0.15. Three runs were performed using 10 7 iterations (with a burn-in of 10 6 iterations) and a sampling interval of 1,000 iterations. Migration rates were calculated as the average of the three runs.
To predict potential future reductions in heterozygosity for each of the 16 sampled rivers, we used the method proposed by Crow and Kimura (1970). Predicted levels of heterozygosity (H t ) at 10, 50, and 100 generations in the future were calculated as: where H O is the current observed heterozygosity, N e is the current ef-

| Data quality assurance
The 18  were found for 12 tests comprising eight loci and ten populations. As none of these significant results were consistent across loci or populations, all loci and populations were retained for further analyses.

| Measures of genetic diversity
Measures of genetic diversity were generally high across all trout populations. A total of 452 alleles were found at the 19 loci. The number of alleles per locus ranged from two (One102a) to 50 (SsaD58).

| Analysis of genetic structure
The STRUCTURE and DAPC analyses were in general agreement, both identifying a group of small "outlier" catchments (Helford River, Kennall, Percuil, Portmellon, Par, and Polperro) and a set of generally larger rivers. For the STRUCTURE analysis, the Evanno ΔK method identified K = 2 as the most likely partition of the data (Figure 2) splitting the rivers into western (Helford River to Portmellon) and eastern groups (Par to Lynher). Hierarchical analysis of these two groups showed further structure within the data set. For the western group, the optimum K was six with the rivers of the Carrick Roads (Allen, Tresillian, and Fal) As the STRUCTURE analyses showed a lack of genetic structuring of trout within rivers, the DAPC analysis was performed at the river level, combining multiple collections from the same river into a single sample. Successive hierarchical DAPC analyses also showed that six of the small streams were distinct. Analysis of the whole data set showed trout in the Helford, Portmellon, Percuil, and Kennall rivers to be distinct ( Figure A1a,b), with the Par and Polperro rivers being distinct in the first hierarchical analysis ( Figure A1c,d). The second hierarchical analysis ( Figure A1e,f) showed a split between the remaining western and eastern rivers.

| Bottleneck analyses
The heterozygosity excess method suggested that fish in only a single river, the Polperro, had gone through a recent bottleneck (Table A3). For all rivers, the allele frequency distribution had a normal L-shaped distribution. However, the M-ratio test indicated that trout in the majority of rivers and STRUCTURE groupings had experienced a more ancient bottleneck. The exceptions were the FLL and SL groups and the River Fowey, with the SL group and the Fowey failing to show a bottleneck only at Θ values greater than 1 (Table A3).
The VarEff results indicated that all rivers and STRUCTURE group- also between the western and eastern groups (Figure 4b).
Using the approach of Crow and Kimura (1970), the larger rivers within the data set are predicted to lose an average of 8.6% (range 2.8%-14.9%) and 33.56% (range 15.70%-42.01%) of their heterozygosity by 100 generations in to the future, based on the LD and sibship methods of calculating N e , respectively ( Figure 5, Table A4).
Conversely, the outlier rivers are predicted to lose an average of 34.5% (range 20.1%-56.9%) and 60.42% (range 33.57%-82.43%) of their heterozygosity, based on the LD and sibship methods of calculating N e , respectively, over the same time period.

| D ISCUSS I ON
Analysis of trout from 16 rivers and streams in south Cornwall genotyped at 19 microsatellite loci identified highly contrasting patterns of diversity, relatedness, and genetic differentiation, with six small stream trout populations being distinct from those inhabiting geographically proximal larger catchments.
The initial STRUCTURE and hierarchical DAPC analyses showed a strong regional organization of the genetic diversity in the trout populations analyzed, splitting populations into eastern and western groups. During the last glacial period, sea levels were up to 130 m lower than at present and the English Channel was largely dry land. At smaller geographical scales, the genetic structure is often locally complex. Trout populations in the larger catchments and some of the smaller streams display significant isolation by distance, with fish in geographically proximal rivers tending to be genetically similar. However, it is clear that trout in six of the rivers studied here do not fit this pattern and that a mixture of evolutionary, anthropogenic, and environmental processes have acted to alter the levels of genetic diversity and differentiation between them and the generally larger, more "characteristic" rivers of the region. Similar patterns have been found in other fish species. For example, clear regional structure was found in Baltic populations of pike (Esox lucius); however, at local levels this structure was more complex (Bekkevold, Jacobsen, Hemmer- Hansen, Berg, & Skov, 2015). The lack of IBD for the full data set analyzed in the current study suggests that genetic drift is an important process in shaping patterns of genetic variability of trout populations along the south Cornish coast (Hutchison & Templeton, 1999) and this pattern appears to be driven by the marked divergence of fish in six of the smallest catchments. Similarly, a within-river study of genetic structure in brown trout only showed significant IBD after samples from above impassable barriers were removed from analyses (Griffiths et al., 2009), while Pearse, Martinez, and Garza (2011) found that historic patterns of IBD had been erased in contemporary populations of steelhead (Oncorhynchus mykiss) and that this was partly due to fragmentation of rivers by dams.

F I G U R E 3 Correlation between geographic distance (km)
and genetic distance (linear F ST ) for (a) the full data set (r 2 = .003, p = .327), (b) the larger rivers data set (removing the six outlier small stream populations: r 2 = .193, p = .001), and (c) data for trout from the River Fowey only (r 2 = .146, p = .119)   Paris et al., 2015). However, it is now apparent that human activities have been affecting fish populations in Western Europe over much longer time scales (Hoffman, 2005;Lenders et al., 2016).
Southwest Britain has a long history of mining, spanning from the prehistoric, through the Roman and Medieval periods to the present day (Gerrard, 1996). Early mining techniques such as tin streaming required huge volumes of water to wash away soil overlaying the metal ore deposits. This resulted in the construction of weirs, leats (artificial water courses), and dams (Gerrard, 1999(Gerrard, , 2000 to channel water from rivers and streams to where it was needed for streaming. Up until the Industrial Revolution, subsequent processing of the metal ore relied on mills and smelters powered by waterwheels (Gerrard, 1999) that also required diversion and damming of rivers.
However, while it is clear that historic mining has affected the majority of catchments in the area (Bryan & Hummerstone, 1977;Pirrie, Power, Rollinson, Cundy, & Watkins, 2002;Pirrie, Power, Wheeler, et al., 2002), the effects on resident trout populations have been felt most keenly in the smallest catchments. Larger catchments appear able to buffer against localized reductions in fish population size caused by mining activities, perhaps due to the often-patchy distribution in time and space of mining-related activities. Knaepkens, Bervoets, Verheyen, and Eens (2004)  The presence of impassable barriers also threatens the longterm persistence of populations (Crook et al., 2015;Morita & Yamamoto, 2002) making it unlikely that genetic diversity can be increased naturally through the straying of sea trout into these streams to spawn. Reductions in genetic diversity have been found F I G U R E 4 Contemporary and historical gene flow diagrams based on the results of BayesAss and migrate-n, respectively. Rivers abbreviations are as given in Table 1. Rivers are colored in shades of red and green representing the rivers belonging to the western and eastern groups, respectively, as identified in the STRUCTURE analysis ( Figure 2). Arrow direction represents direction of gene flow between rivers. Bumps in the contemporary plot represent selfassignment of fish to their own river Medieval times (Dines, 1956). The Par has been adversely affected by waste from the china clay mining industry. Mining for the clay started in the mid-18th century, releasing large quantities of silt into rivers which would have been particularly detrimental to trout and salmon. Fine sediments can impact fish both directly and indirectly.
Suspended sediments can cause damage to gills, increase stress levels, and affect feeding and growth rates of fish (Kemp, Sear, Collins, Naden, & Jones, 2011). Indirectly, sediments can clog up spawning gravels and reduce the amount of dissolved oxygen available to developing eggs (Kemp et al., 2011). Additionally, significant reductions in abundance and diversity of invertebrates on which trout feed have been reported in clay-affected rivers (Nuttal & Bielby, 1973). Miramichi River in Canada was affected by high levels of copper and zinc in river water leaching from a mine, with very high levels of metals appearing to completely stop fish movement altogether (Saunders & Sprague, 1967). Similarly, Paris et al. (2015) suggested that reduced movement of fish through a region of high toxic metal contamination is responsible for genetic subdivision of brown trout populations in the River Hayle in west Cornwall.
Together, these multiple processes have resulted in genetic erosion of trout in small streams across the region. Exposure to industrial pollution can elicit multiple possible responses including migration, local extinction, or adaptation (Bijlsma & Loeschcke, 2012).
Experimental populations of Chironomus riparius rapidly lost genetic diversity when exposed to environmentally relevant concentrations of tributyltin (Nowak et al., 2009), while brown trout populations inhabiting rivers with high levels of heavy metal pollution had significantly lower levels of neutral genetic diversity than fish in corresponding relatively clean rivers (Paris et al., 2015). However, responses to pollution appear to be species-specific. No differences in microsatellite heterozygosity or allele number were found between a control site and two sites suffering from cadmium/zinc pollution in bullhead (Cottus gobio; Knapen et al., 2009). Similarly, McMillan, Bagley, Jackson, and Nacci (2006)  The significant predicted loses of heterozygosity and the high levels of relatedness, along with fact that genetic drift is a dominant evolutionary force, show that trout in small, isolated rivers are in danger of further severe reductions in diversity and heterozygosity.
Such decreases have serious consequences for the future survival of these populations. For endangered species, reductions in heterozygosity can lower their evolutionary potential (e.g., their ability to cope with future climate change), compromise their reproductive fitness and elevate the risk of extinctions (Spielman et al., 2004). These generalizations apply equally to small, isolated populations of nonendangered species, such as those highlighted here. Five fitness-related traits were significantly correlated to levels of heterozygosity in the three-spined stickleback (Gasterosteus aculeatus), while there was an increased risk of extinction in populations of the Glanville fritillary (Melitaea cinxia) with reduced heterozygosity (Lieutenant-Gosselin & Bernatchez, 2006;Saccheri et al., 1998). Additionally, large outbreeding populations of lake trout (Salvelinus namaycush) have lower numbers of deleterious mutations than smaller, more inbred populations, highlighting that purifying selection may be less effective in such populations (Perrier et al., 2017). Together, all these factors suggest that reductions in levels of diversity and heterozygosity may put small, isolated trout populations at risk of extinction before demographic changes become apparent (Spielman et al., 2004).
We have highlighted that human activities over long timescales

CO N FLI C T O F I NTE R E S T
None declared.  Calculations are based on current H O and effective population size (N E ) estimated using the linkage disequilibrium (LD) method as implemented in NEESTIMATOR v.2 (Do et al., 2014), and the sibship method as implemented in COLONY v 2.0.5.9 (Jones & Wang, 2010). Predicted future H O and predicted percentage loss in H O are given as LD method/sibship method.