Impacts of acidification on brown trout Salmo trutta populations and the contribution of stocking to population recovery and genetic diversity

Abstract Anthropogenic acidification in SW‐Scotland, from the early 19th Century onwards, led to the extinction of several loch (lake) brown trout (Salmo trutta) populations and substantial reductions in numbers in many others. Higher altitude populations with no stocking influence, which are isolated above natural and artificial barriers and subjected to the greatest effect of acidification, exhibited the least intrapopulation genetic diversity (34% of the allelic richness of the populations accessible to anadromous S. trutta). These, however, were characterised by the greatest interpopulation divergence (highest pairwise D EST 0.61 and F ST 0.53 in contemporary samples) based on 16 microsatellite loci and are among the most differentiated S. trutta populations in NW‐Europe. Five lochs above impassable waterfalls, where S. trutta were thought to be extinct, are documented as having been stocked in the late 1980s or 1990s. All five lochs now support self‐sustaining S. trutta populations; three as a direct result of restoration stocking and two adjoining lochs largely arising from a small remnant wild population in one, but with some stocking input. The genetically unique Loch Grannoch S. trutta, which has been shown to have a heritable increased tolerance to acid conditions, was successfully used as a donor stock to restore populations in two acidic lochs. Loch Fleet S. trutta, which were re‐established from four separate donor sources in the late 1980s, showed differential contribution from these ancestors and a higher genetic diversity than all 17 natural loch populations examined in the area. Genetically distinct inlet and outlet spawning S. trutta populations were found in this loch. Three genetically distinct sympatric populations of S. trutta were identified in Loch Grannoch, most likely representing recruitment from the three main spawning rivers. A distinct genetic signature of Loch Leven S. trutta, the progenitor of many Scottish farm strains, facilitated detection of stocking with these strains. One artificially created loch was shown to have a population genetically very similar to Loch Leven S. trutta. In spite of recorded historical supplemental stocking with Loch Leven derived farm strains, much of the indigenous S. trutta genetic diversity in the area remains intact, aside from the effects of acidification induced bottlenecks. Overall genetic diversity and extant populations have been increased by allochthonous stocking.

derived farm strains, much of the indigenous S. trutta genetic diversity in the area remains intact, aside from the effects of acidification induced bottlenecks. Overall genetic diversity and extant populations have been increased by allochthonous stocking.

K E Y W O R D S
acid tolerance, adaptation, introgression, microsatellites, population bottlenecks, sympatric populations 1 | INTRODUCTION Worldwide, many salmonid populations have become extirpated (Hendry et al., 2003), largely as a result of anthropogenic causes.
There is now a particular interest in how best to restore these populations, especially in situations where natural recolonisation cannot occur. Only about a quarter of reintroductions have resulted in self-sustaining populations (Houde et al., 2015). Restoration stocking failures can occur because the original factors that led to the extinction still exist, or due to random demographic fluctuations (Moritz, 1999). The intrinsic potential for local adaptation in salmonids Garcia de Leaniz et al., 2007) makes restoration particularly challenging with attempts potentially failing due to inadequate adaptive matching of introduced fish (Allendorf & Waples, 1996).
Several approaches have been proposed to overcome restoration failure involving "matching or mixing" (Lesica & Allendorf, 1999).
These include using a donor population genetically similar to the extinct one; i.e., genetic or ancestry matching (Houde et al., 2015), which assumes that genetically similar fish are likely to be best adapted to the environmental conditions in which the previous population existed. Use of within-catchment local sources probably gives increased fitness from local adaptation and decreased risks from straying (Garcia de Leaniz et al., 2007). Also, local salmonid sources are likely to share greater genetic similarity with the historic population as a result of common ancestry, although postglacial colonisation by multiple lineages (McKeown et al., 2010) means that this is not necessarily the case for brown trout Salmo trutta L. 1758. Another approach involves the selection of a source population from a similar environment; i.e., environmental matching (Houde et al., 2015). Such populations may possess genes that are adaptive for the environment of the extirpated population, which may be especially appropriate when the environment has changed substantially in the intervening period. A further stocking option is to use fish from a population with a high level of genetic variation, hence increasing the potential for local adaptation to evolve. High levels of genetic variation can also be produced by mixing fish from multiple genetically dissimilar populations (Houde et al., 2015;Huff et al., 2011). Mixing can involve genetically distinct populations with common ancestry, or from similar environments, so matching and mixing approaches are not mutually exclusive. Mixing may also be appropriate where a single source population cannot sustain the removal of sufficient fish for reintroduction.
In situations where a wild S. trutta population is present in reduced numbers, supplemental stocking of fertile farm strain has been used frequently in an attempt to boost the angling catch. The efficacy of stocking farm-reared S. trutta, however, is generally considered to be low (Ferguson, 2007;Pinter et al., 2017). In Britain and Ireland, the farm strains used are often derived solely, or partly, from the first S. trutta farms established in Scotland at Solway (1880; 54 58 0 46 00 N, 03 39 0 27 00 W) and Howietoun (1881; 56 04 0 20 00 N, 03 57 0 10 00 W), which involved broodstock of Loch Leven (56 12 0 N, 03 23 0 W) origin (Armistead, 1895;Maitland, 1887). As stocking with these domesticated strains has been widespread over the past 130 years, the extent to which native gene pools of S. trutta have been lost or modified has been the subject of much debate. There is now strong evidence indicating that such genetic changes can affect the fitness, life-history characteristics and other genetically based aspects of the populations resulting in stocking being counterproductive relative to the aim of increasing S. trutta numbers (Ferguson, 2007). Thus, genetic assessment of S. trutta populations is important in establishing the effectiveness of stocking in different circumstances. It is also required to determine the extent of introgression by hatchery-reared S. trutta and identify pure indigenous populations of high conservation value.
Effective salmonid conservation and management requires an understanding of the roles of natural and anthropogenic influences on population genetic structure (Small et al., 2007). Salmo trutta exhibits complex genetic structuring, with high levels of genetic differentiation often occurring at small geographic scales, both allopatrically and sympatrically (Andersson et al., 2017a(Andersson et al., , 2017bFerguson, 1989;Verspoor et al., 2019). Genetic differences can arise as a result of spawning in different localities and the accurate natal homing typical of salmonids.
These spawning groups may diverge genetically over generations as a consequence of genetic drift and natural selection. The varying balances between reproductive isolation produced by homing to natal breeding areas and gene flow among populations caused by successful reproduction of straying individuals (effective straying) results in different levels of genetic population structuring, which may or may not be related to geographic distance among populations (Bond et al., 2014). Compared with other salmonids, relatively little is known of the conditions and timescales required for detectable allopatric and sympatric differentiation to evolve in S. trutta (Jorde et al., 2018).
While intra and interpopulation genetic variation is a major component of biodiversity, it has received relatively little attention from organisations responsible for the management and conservation of non-endangered, but ecologically important, species (Mimura et al., 2016).
Many freshwater lochs (lakes) occur in the upland area (200+ m asl) in south-west Scotland. These range in size from <1 ha to Loch Doon at 820 ha ( Figure 1) and angling records indicate that most currently contain S. trutta. The lochs are drained by several river systems ( Figure 1). In addition to many natural waterfalls, some of the rivers have hydroelectric dams, constructed mainly in the 1930s. These barriers are partially or completely impassable, resulting in many lochs being reproductively and genetically isolated from upstream movement of S. trutta. Both river-resident and anadromous (sea trout) S. trutta occur in the lower reaches of these rivers, although artificial barriers have reduced the incidence of the anadromous forms, as elsewhere in Europe (Ferguson et al., 2019).
The area consists largely of granitic rocks, often overlain by peat and poorly-drained, acidic soils. This base-poor topography and associated low buffering capacity, together with high rainfall, geographical position and prevailing winds, resulted in the area being the worst affected in Scotland by industrially driven acidification in the latter part of the 20th century (Harriman et al., 1987). Diatom studies of loch substrates indicate that acidification started in the early part of the 19th century (Battarbee et al., 1985), coincident with the early stages of the industrial revolution. The increase in acidity reached its peak in the years after 1950 with the pH falling in several lakes to below 4.5, which is often regarded as the lower tolerance limit for species such as S. trutta (Gjedrem & Rosseland, 2012;Jellyman & Harding, 2014). However, no simple pH threshold can be set, as many other factors are often involved. These include heritable tolerance of acidic conditions (Gjedrem & Rosseland, 2012), level of calcium, which N F I G U R E 1 Diagrammatic map (not to scale) of south-west Scotland showing the relative positions of rivers (in italics) and lochs (in roman font) from which Salmo trutta were sampled or are referred to in the text. Additional details are given in Table 1. , Natural and artificial barriers that are likely to be passable to upstream migrating S. trutta, at least for certain sizes of fish and under some water flow conditions; , barriers considered impassable to upstream migrants reduces the toxic effects of low pH and labile aluminium, the toxicity of which is reduced by dissolved organic carbon Serrano et al., 2008). Extensive coniferous afforestation in south-west Scotland from 1950s onwards probably exacerbated the acidification due to interception of acid deposition by the forest canopy, this being particularly important in relation to some spawning streams . In the early 1980s, UK and international action to reduce the emissions of sulphur and nitrogen from power stations (Kernan et al., 2010) resulted in a c. 80% reduction in UK SO 2 emissions (Helliwell et al., 2011). Improvements in pH and labile aluminium levels took place in the lochs of south-west Scotland, especially during the second half of the 1980s (Ferrier et al., 2001).
Survival studies in Loch Fleet in 1984 using S. trutta eggs and fry showed that these stages could not survive in the loch water as a result of low pH (mean 4.4), low calcium (1 mg l −1 ) and high labile aluminium concentration (200 μg l −1 ; Turnpenny et al., 1988). In all five lochs, waterfalls, impassable to upstream movement of S. trutta, prevented upstream recolonisation after environmental conditions improved.
In the 1978-79 and 1984 surveys, low numbers of S. trutta were found in many other lochs relative to earlier records. For example, in Loch Grannoch, one of the most acidified lochs, the annual S. trutta catch was c. 1000 fish in 1940 but this declined steadily to <100 fish in the early 1970s, even with greatly increased fishing effort (Harriman et al., 1987). Loch Riecawr, for which angling catch records exist since the early 20th century, showed a tenfold decline in numbers of S. trutta caught per year by anglers from 1925 up to the 1970s, subsequently followed by an increase (Harriman et al., 2001).
Catch records (Harriman et al., 2001;McCartney et al., 2003) indicated a rapid natural recovery in S. trutta numbers in the lochs during the 1990s in spite of the fact that many of these, Loch Grannoch for example, still remained chronically acidified (Kernan et al., 2010). Continuous pH recording in early 2017 showed pH values in Loch Grannoch from 4.6 to 4.9 (Galloway Fisheries Trust, 2017).
The diverse landscape ecology, water chemistry and anthropogenic influences, including stocking, of south-west Scotland make it an important and ideal area for studying the effect of these factors on the population genetics of S. trutta, which is a UK Biodiversity Action Plan priority species (JNCC, 2010). The main inter-linked objectives of this study were to determine: (a) the effect of acidification on the contemporary intra and interpopulation genetic diversity and population structure; (b) if restoration stocking has resulted in self-sustaining populations in lochs where S. trutta were considered extinct and the relative success of different strategies for reintroduction; (c) to what extent has stocking with Loch Leven based farm strains resulted in introgression into natural populations; (d) if sympatric sub-structuring occurs within any of the loch S. trutta stocks and how this has evolved; (e) key populations in south-west Scotland of high conservation or scientific value.

| Restoration and supplemental stocking history
With the exception of the fish farms and Loch Leven, the locations referred to below are shown on Figure 1. The land surrounding Loch Fleet, a small (17 ha) oligotrophic upland loch, was limed using calcium carbonate powder in 1986 and 1987 producing an almost immediate improvement in water conditions, with a pH close to 7.0 and elevated calcium and reduced aluminium levels (Turnpenny, 1992). Following successful egg survival trials, restoration stocking of S. trutta was undertaken in May 1987. This involved 300 fish with c. equal numbers of: Little Water of Fleet, the outflowing river below the impassable waterfall (age 1+ years wild S. trutta); Loch Dee, a large loch on the geographically adjacent Ken-Dee catchment (first generation hatchery reared offspring; age 1+, 2+ and 3+ years in the ratio 4:2:1); Solway Fish Farm (age composition similar to Loch Dee stock). The S. trutta were batch marked by fin clipping to allow identification of the three stocks. In July 1988, a second batch of 220 S. trutta, involving the same three stock types and essentially the same age distribution, was introduced. These were marked with individual tags (Turnpenny, 1992). Egg survival experiments were carried out in the years 1988-89 to 1993-94 to check on possible re-acidification. Eggs from S. trutta trapped in the inlet spawning river were used, with the exception of the 1993-94 season when the eggs used were from S. trutta captured in a Loch Grannoch tributary (Turnpenny et al., 1995). This potentially introduced Loch Grannoch stock in addition to the three others above, although <1000 eggs were planted in each of the inlet and the outlet rivers. Improvements in water chemistry in the late 1980s and early 1990s indicated that conditions might be suitable for S. trutta reintroduction in the other four lochs in which S. trutta appeared to be extinct earlier.
In October 1994, 3000 hatchery-reared age 1+ S. trutta produced from Loch Grannoch broodstock were released in Loch Enoch and survived successfully until at least November 1998 (Collen et al., 2000). At the same time, offspring from this source were also stocked into Loch Some stocking with farm strain S. trutta is known to have been carried out in other lochs and rivers in the area. According to Sandison (1983), Loch Mannoch was previously stocked with a Loch Leven strain of S. trutta. Published and anecdotal accounts indicate that farm-strain stocking had been undertaken in Lochs Dee, Harrow and Riecawr (Harriman et al., 1987). The most recent farm-strain S. trutta stocking has been only in the rivers, which are the main target for anglers, with River Girvan having been stocked until recently (R.A. unpublished data: information obtained from local angling clubs).
Stocking with offspring of Riecawr S. trutta also took place in the River Girvan catchment (G. Shaw, Forestry Commission, personal communication).

| Sampling
Most specimens (age 1+ year and older) were obtained by angling, which was carried out with appropriate permissions and according to angling regulations. Juveniles were provided by authorised fishery professionals from rivers being electro-fished as part of their ongoing survey activities. Juvenile S. trutta specimens were taken from several geographically separated sites within each river to provide an overall river profile. In the case of Rivers Annan, Girvan and Loch Doon, specimens were taken from below and above natural barriers. Adult S. trutta were sacrificed by cranial blow as per standard angling practice while juveniles were killed with an overdose of MS-222 anaesthetic. An adipose-fin clip or piece of skeletal muscle was taken, either immediately or at the end of the fishing day and stored in 98% molecular-grade ethanol. Lethal sampling was not considered to be a threat to the populations since most lochs and rivers appeared to have high densities of S. trutta, or angling was carried out within normal bag limits.
In total 2420 S. trutta specimens were obtained, mainly in 2010, 2011 and 2012 (denoted as contemporary samples). Frozen S. trutta specimens from historical loch samples  were available at the Marine Scotland, Freshwater Laboratory Pitlochry. Samples were taken from 23 lochs, the main focus of the study and seven rivers in south-west Scotland. In addition, samples were obtained from Loch Leven and from Howietoun farm, as the local Solway farm, previously used in the area for stocking including Loch Fleet, was no longer in operation.
Sample location details and associated three-letter abbreviations, together with sampling year (e.g., GRA 82 ) where temporal samples were available, are given in Table 1 Table S1). The sector of capture within the loch was recorded for FLE 12 individuals taken in September.
The net position within the loch was available for each individual in the GRA 12 sample. Chi-square analysis was used to test for heterogeneity in position of capture for sub-groups.
Further information about primers, PCR conditions and genotyping is given in Keenan et al. (2013a). LDH-C1* screening of sub-samples consisting of 20 specimens each from non-stocked lochs was carried out following the methodology of McMeel et al. (2001) and presented as the frequency of the *100 allele. Mitochondrial (mt)DNA screening and interpretation were carried out as detailed in McKeown et al. (2010). MtDNA data were primarily included to assist in FLE ancestry determination although several other loch samples were also included to provide baseline data.

| Data analyses
Potential full sibling groups were identified by the maximumlikelihood method implemented in the program COLONY 2.0.5.4 (Jones & Wang, 2010) with the following variables applied: female and male polygamy with no inbreeding; dioecious and diploid; medium run; full likelihood; no updating of allele frequencies; no sibship prior; typing error rate of 0.001. Three replicate runs were carried out in each case and the majority result used where these differed. In accordance with Hansen and Jensen (2005), but taking account of Waples and Anderson (2017), for analyses other than sibship effective population (N e ) estimates, sibling groups were reduced to a maximum of three individuals with the least amount of missing microsatellite data or in numerical sequence otherwise.
All sample pairs were tested for significant genic differentiation using Exact G tests as implemented in GENEPOP 4.7.0 (Raymond & Rousset, 1995), using 10,000 dememorisations, 100 batches and 5000 iterations per batch. Temporal and geographical samples not showing significant genic differentiation in Exact G tests were pooled for subsequent analysis except where there were triangle inconsistencies; i.e., sample A = sample B, sample B = sample C, but sample A 6 ¼ sample C. All subsequent analyses, with the exception of N e estimates, were carried out on the combined samples.
Allelic richness (N AR ) and private allelic richness (N PAR ), the number of alleles or private alleles in a sample were estimated using the rarefaction method in HP-RARE (Kalinowski, 2005). To avoid analytical bias due to a few samples of n < 30, analysis was standardised to a common sample size of 60 genes. Samples from natural populations (i.e., excluding ENO, FLE, LEV, MAN and NAR (Table 1)) were divided into those locations known to be fully accessible to anadromous S. trutta upstream migration as a result of barriers (Table 1 and Figure 1). Note that SHI was excluded from the accessible group as although anadromous S. trutta occur occasionally (R.A. unpublished data) it is above three adjacent barriers, which restrict upstream movement. Samples were also divided into two groups on the basis of the underlying geology of the loch area (Supporting Information Table S1). Statistical significance of difference between groups was assessed using the Mann-Whitney U-test. Spearman's rank correlation was used to determine the degree of correlation between N AR with other physical, chemical and biological data. Both the Mann-Whitney and Spearman's tests were carried out using PAST 3.14 (Hammer et al., 2001).
Differentiation for all sample pairs was measured using Weir & Cockerham's F ST (Weir & Cockerham, 1984) and by Jost's D EST (Jost, 2008), the latter having the advantage of being independent of the level of gene diversity (Jost, 2008), which often leads to an underestimation of the level of genetic differentiation between samples for multi-allelic microsatellite markers. F ST and D EST estimates were calculated using the program diveRsity (Keenan et al., 2013b) and tested for significant deviation from 0 (i.e., no significant genetic differentiation) by randomising multi-locus genotypes between pairs of samples with 1000 bootstrap permutations.
To examine the possible effects of historical stocking on contemporary patterns of population genetic structuring, admixed individuals (identified as described below) were removed from samples of natural populations and corrected N AR , N PAR , H E , F ST and D EST recalculated; i.e., these corrected genetic diversity measures were based on the identified pure clusters rather than the original geographically defined samples.
Three independent approaches, based on different model assumptions and strategies for computation (Jombart et al., 2010;Neophytou, 2014;Neuwald & Templeton, 2013), were employed to describe S. trutta population genetic structuring. In the first instance, the Bayesian clustering method implemented in the program STRUC-TURE 2.3.4 (Pritchard et al., 2000) was used. STRUCTURE analysis followed the hierarchical approach suggested by Rosenberg et al. (2002), which facilitates the identification of major genetic groupings (shared recent ancestry) within the data, eventually refining these down to the population level. Within this hierarchical framework, all major groups identified within a given STRUCTURE run were used separately, as starting points for subsequent runs. In each case, STRUCTURE runs were replicated 20 times for each K value (number of genetic clusters being tested), which ranged from 1 to 10 using the following variables: length of burn-in period = 100,000; number of MCMC reps after burn-in = 100,000; admixture model, allele frequencies correlated models with and without location priors. The ΔK ad hoc approach (Evanno et al., 2005), as implemented in STRUCTURE HARVESTER (Earl & vonHoldt, 2012), was used as a guide to identify the most likely number of clusters. Results of replications were then combined into a single population output using the program CLUMPP 1.1.2 (Jakobsson & Rosenberg, 2007) with the Greedy search method with option 2 for random input orders set to 20,000. CLUMPP output files were used to produce STRUCTURE bar plots illustrating membership of individuals to inferred clusters.
The Bayesian analysis of population structure program (BAPS 5.3; Corander et al., 2003Corander et al., , 2008 was used as the second approach to identify clusters of genetically similar individuals and to assign individuals to clusters based on their multi-locus genotypes, using BAPS's "clustering of individuals" option. Unlike STRUCTURE, which relies on an ad hoc statistic to identify the best number of clusters explaining the data, BAPS infers the optimal number of clusters directly (Corander et al., 2004). The program was initially run with all samples for a maximum K = 40 to identify the optimal K-value explaining the data. Subsequent BAPS runs were then sequentially carried out for all samples in fixed K + 1 steps from K = 2 to K = best K value (as identified in the previous step), to recover hierarchical relationships among population samples comparable to the STRUCTURE hierarchical analysis.
The discriminant analysis of principal-components method of Jombart et al. (2010), which is implemented in the function dapc of the R adegenet package (Jombart, 2008), was used as the third independent analytical approach. The identification of the best number of clusters (or populations) explaining the data was done using find.cluster (with the Bayesian information criterion; BIC) and with a maximum number of clusters set to 50. To minimise potential analytical biases in the dapc analysis, the number of retained PCA were chosen to optimise the α-score, as recommended in the dapc manual, using the function optim.a.score.
BAPS was used to identify significantly admixed individuals within inferred populations by identifying the original samples or BAPS clusters from which each individual's alleles originate (Corander et al., , 2008 using the output file from the initial mixture clustering. For this admixture analysis, a minimum cluster size of 20 was used in order to remove small groups of outlier individuals. Following guidelines provided in the BAPS manual, runs involved 100 iterations to   (Nei et al., 1983) were constructed using POPTREE2 (Takezaki et al., 2010). One tree was constructed using all of the original samples while a second tree was constructed using contemporary south-west Scotland samples from natural populations only, with BAPS determined admixed individuals removed. Confidence for the tree nodes was assessed by bootstrapping (10,000).
Effective population size (N e ) was estimated using: (a) the biascorrected version of the linkage disequilibrium (LD) method (Waples & Do, 2008); (b) by the sibship frequency (SF) method (Wang, 2009); (c) where data were available, by the temporal method of Jorde and Ryman (2007). The NeEstimator 2.01 software (Do et al., 2014) was used for both LD and temporal methods. Allele frequency criteria of ≤0.05, 0.02 and 0.01 were used. Jackknifing over loci was used to obtain 95% confidence intervals for the estimates. For the temporal method, a generation time of 3 years was used based on observations of maturity in the samples obtained (authors' unpublished data). The SF method was carried out using Colony 2.0.5.4 (Jones & Wang, 2010). Correlation between N e values obtained using the LD and SF methods was tested using Spearman's correlation coefficient as above. It should be emphasised that the primary aim of the N e analyses was not to accurately determine N e but rather to identify populations where values are or were low, such that increased genetic drift would be expected.

| RESULTS
While full sibs were observed in 69% of the samples examined, the actual number of full sib families in each case was small, with 73% of these consisting of two sibs only ( Note. GRA1, 2, 3, and FLE 1 and 2 refer to the separate populations identified in those lochs in the GRA 10-12 , the FLE 11 and the FLE 12 samples respectively. LD, N e based on linkage disequilibrium method with minimum frequency 0.01 as recommended by Waples and Do (2008) for sample sizes of this magnitude; Sib,N e based on the sibship method (Wang, 2009), assuming non-random mating; Temporal, N e based on temporal method of Jorde and Ryman (2007), with a minimum allele frequency of 0.01, where the estimate is based on that sample and the preceding temporal one. and total catchment area (Spearman's ρ = 0.74, P < 0.01). No significant correlation was found between N AR and the minimum-recorded pH for a loch or with the 2010-12 angling catch (Supporting Information Table S1). For all samples, N AR was positively correlated with H E (Spearman's ρ = 0.81, P < 0.001).
The lowest D EST value overall (Supporting Information Table S3) was NEL 11 vs. VAL (0.003), which was not significant (95% C.I.  Figure 4). In the remaining cases, where other admixed individuals were noted, these involved putative sources from the same catchment. The LDH-C1*100 allele frequency ranged from 0 to 0.9 (Table 1). GRA 10-12 had a LDH-C1*100 allele frequency of 0.34 while in NEL 11 and VAL these were 0.9 and 0.82 respectively (Table 1). Assuming a native frequency of 1.0 in NEL and VAL, based on allele frequency proportionality (Taggart & Ferguson, 1986) the maximum overall genetic contribution of GRA would be 27% to VAL and 15% to NEL 11 , similar to the BAPS admixture results. The mtDNA haplotype 4.7 was present in GRA 10-12 at a frequency of 0.643 and haplotype 3.8 at a frequency of 0.262 (Table 3). However, both NEL 11 and VAL were fixed for haplotype 3.8, indicating, at most, a limited maternal contribution from GRA.

In independent analyses of samples from individual lochs only FLE
and GRA indicated further sub-structuring. BAPS analyses of FLE 11 and FLE 12 samples analysed separately indicated an optimal K = 4.
With a fixed K of 2, as seen in the overall STRUCTURE analysis   Table S3) and also by both the STRUCTURE placement ( Figure 2) and NJ tree (Supporting Information Figure S2).
The mtDNA haplotype 4.7, present in GRA 10-12 at a frequency of 0.643, is the only haplotype that was private to any of the four putative FLE ancestors (  Table S3) between all pairs of these three groups (GRA1-2, 1-3, 2-3) were sig- Overall, these three groups formed 28%, 37% and 35% of the sample. There was significant heterogeneity in the distribution of the three groups within the loch with GRA 12 group 1 being present at a greater frequency than either GRA 12 2 or GRA 12 3 in the northern half of the loch (χ 2 P < 0.02). The latter two groups were not significantly different in their distribution within the loch.
Effective population size (N e ) estimates for the linkage disequilibrium and temporal methods are only given for a minimum allele frequency of 0.01 (Table 5), as this showed the lowest number of ∞ estimates and is also appropriate for sample sizes of the magnitude used ( Waples & Do, 2010 (Harriman et al., 1987;Maitland et al., 1987;Turnpenny et al., 1988). However, the current study shows that a small remnant population of S. trutta As well as this N e based estimate of a substantially reduced population in Grannoch, angling records also indicate substantially reduced numbers. Thus, the annual Grannoch S. trutta catch was approximately 1000 fish in 1940 but this declined steadily to <100 in the early 1970s, even with greatly increased fishing effort (Harriman et al., 1987). In the lochs where clearly remnant populations survived, netting of Loch Neldricken and Round Glenhead in 1978-79 (Harriman et al., 1987)  The successful re-establishment of Loch Enoch S. trutta, after an absence of at least 70 years, was in spite of it having borderline water chemistry conditions (i.e., mean pH 4.8) (Collen et al., 2000). This successful re-establishment is therefore likely to be due to a genetically increased tolerance of Loch Grannoch S. trutta to acid conditions.
Loch Grannoch translocated S. trutta fry, together with introduced eggs and subsequent alevins, survived much better than the equivalents from Loch Dee (mean pH 5.2 in 1981; Burns et al., 1984) in common-garden experiments undertaken in the Loch Enoch outflowing river in 1991 and 1993 (Collen et al., 2000). Loch Dee S. trutta have previously been shown to have increased tolerance of low pH compared with S. trutta from other waters and a farm strain from higher pH conditions (Battram, 1990;McWilliams, 1982). Acid tolerance is a quantitative trait with large genetic variation among natural populations and with a higher heritability than usually found for fitness traits in fishes (Gjedrem & Rosseland, 2012). Salmo trutta survived in Loch Grannoch in spite of a minimum pH of 4.2 and aluminium >300 μg l −1 being recorded (Harriman et al., 1987), albeit numbers being much reduced as discussed in Section 4.1.
By 1989, S. trutta were well-established in Loch Fleet and age 0+ and 1+ years fish were found in the inlet and in the outlet downstream of the loch for the entire 7 km above the waterfall, although density was very low after 1 km (Howells et al., 1992). By 1993, the loch S. trutta density was some five times that of the stocking density (Turnpenny et al., 1995). Salmo trutta from all three deliberately As noted above, Loch Dee S. trutta have also been shown to have increased tolerance for low pH conditions. In keeping with their lower tolerance of acid conditions, even though a similar number and age range of Solway farm and Loch Dee S. trutta were stocked and the majority of the inlet spawning run as captured in the trap in 1987 comprised farm S. trutta (Turnpenny, 1992), Solway's overall contribution was low. MtDNA analysis would suggest that female contribution was lower than male contribution for Solway, as has also been reported for farm Atlantic salmon Salmo salar L. 1758 breeding in the wild (Fleming et al., 1996). Farm strain S. trutta generally show poor survival and reproduction in the wild in both rivers and lakes (Ferguson, 2007;Pinter et al., 2017). In 1988, S. trutta of Loch Dee origin were found to be the predominant spawners (Turnpenny, 1992). The contribution of Water of Fleet was also low even though these were from the same river some 10 km downstream below the impassable waterfall and were translocated wild S. trutta rather than hatchery reared offspring.
Garcia de Leaniz et al. (2007) have argued that, as a result of local adaptation, salmonids from within the same catchment would be genetically similar and thus more likely to be successful for population re-establishment. However, this region of Water of Fleet is a known anadromous S. trutta spawning area (J. Graham, Galloway Fisheries Trust, personal communication) and thus their genetic propensity for anadromy (Ferguson et al., 2019) may have meant that they migrated out of Loch Fleet, the impassable waterfall preventing subsequent return. Burger et al. (2000) have shown that life-history adaptations were critically important for the establishment of river and shorespawning populations of sockeye salmon Oncorhynchus nerka (Walbaum 1792) in an Alaskan lake. In addition, the pH below the waterfall in the Little Water of Fleet was considerably higher than in Loch Fleet (Harriman et al., 1987) and thus S. trutta from this locality may not have been suitable for the lower pH environment.
Loch Grannoch S. trutta offspring, in spite of being used only for monitoring hatching success in 1993 and not as part of the original deliberate stocking, contributed to a similar extent overall as Solway farm and Water of Fleet and had a higher contribution than these stocks to the outlet group. Indeed, the Loch Grannoch contribution is surprisingly high given that <1000 eyed eggs were used in each of the inflowing and outflowing rivers in 1993-94 only (Turnpenny et al., 1995). Loch Grannoch S. trutta could have been first to mature as age 1+ years in 1995-96 and primarily not until 1996-97 as age 2+ years, at which time the other S. trutta would have been well-established.
While stochastic factors may have played a part, the acid tolerance of Loch Grannoch S. trutta, as discussed above, is likely to have contributed to their success, especially as by 1994 the pH and calcium concentrations were declining again (Howells & Dalziel, 1995). This was particularly so in the outlet, as only part of the catchment, which included the inflow, was limed (Howells et al., 1992). A minimum pH of 4.6 was recorded in the outlet in 1993 (Turnpenny et al., 1995).
The greater success of S. trutta of Loch Grannoch origin in the outlet rather than the inlet may also have been the result of the outlet spawning group being much slower to establish and therefore there was less competition for the Grannoch juveniles. Although fry were detected in the outlet in 1990 and 1991, albeit at much lower densities than in the inlet, none were detected in 1992 and 1993 possibly due to high spring flows washing fry downstream (Turnpenny et al., 1995). However, declining pH is more likely to have been responsible The Loch Fleet S. trutta stock showed higher genetic diversity than the 17 other natural loch populations sampled in the area, undoubtedly as a result of four genetically distinct ancestors. Thus the mixing strategy has resulted in higher genetic diversity, which is potentially important in maximising the capacity of a population to adapt to its new environment and to future environmental change (Fraser, 2008). Several authors have argued against a mixing strategy on the grounds that hybridisation between genetically distinct stocks can result in offspring of lowered fitness due to outbreeding depression through loss of local adaptation or the disruption of co-adapted gene pools in the F 2 and later generations (Huff et al., 2010). However, it is likely that concerns about outbreeding depression have been overstated as several studies have indicated that, although outbreeding depression can occur in early generations, selection can quickly overcome this and result in hybrid superiority in later generations (Houde et al., 2011;Whiteley et al., 2015;Wells et al., 2019). As demonstrated in Loch Fleet, a mixing approach is likely to be the best option for population re-establishment, except where there is clear evidence of a donor population with adaptive qualities appropriate to the environmental conditions as is the case with Loch Grannoch.
However, a combination of the two strategies can be effectively employed as occurred fortuitously in Fleet.
Self-sustaining stocks arising solely from restoration stocking are shown to be present in Lochs Enoch, Narroch and Fleet some 12-24 years after the initial re-establishment; such restoration was the primary objective of the Loch Fleet project (Howells et al., 1992).
These successes contrast with the generally reported findings in the literature, which indicate that reintroductions have often failed to yield self-sustaining naturalized populations (Anderson et al., 2014).
However, although stated in general terms, these reports are contrary to the S. trutta findings here and the fact that many new selfsustaining S. trutta populations have been established world-wide (Newton, 2013), suggesting that this species may differ from other salmonids in its ability to establish new populations, possibly as a result of its high genetic diversity and life history plasticity (Ferguson, 1989;Ferguson et al., 2019).

| Stocking with farm-strain S. trutta
A population that owes its origin to stocking with a Loch Leven derived farm strain of S. trutta is that in Loch Mannoch. This loch was artificially created by construction of a dam in 1919 and is first mentioned for its fishing by Maxwell (1922), the author noting that it is "heavily stocked with S. trutta." In this situation of a newly created loch there would have been few, if any, native S. trutta to compete with since an impassable waterfall downstream of the dam would have prevented natural colonisation. Also, at the time of stocking, 100 years ago, the farm strain would have been considerably less domesticated than today (Ferguson, 2007).  (Harriman et al., 1987). It seems likely that these were farm fish, possibly as a result of being unable to cope with the acidic conditions.
Several studies have shown a decrease in admixture over time in S. trutta native populations once stocking has ceased (Harbicht et al., 2014;Valiquette et al., 2014). Stocking with farm strains was more prevalent in lowland regions of rivers in south-west Scotland, where the pH is higher (Harriman et al., 1987) and which are generally of more interest to anglers in the area than the lochs. Loch Leven S. trutta influence was found in all of the rivers examined, in most cases at a low level. The highest influence was found in the River Girvan, which is not surprising as this river is known to have had the most recent stocking, with this continuing up until the early years of this century (local angling clubs, personal communication).

| Sympatric populations within lochs
Sub-structuring (i.e., genetically distinct and thus reproductively isolated sympatric S. trutta populations) was found only in Lochs Fleet and Grannoch. Two main genetically distinct populations were found in Loch Fleet with an inlet spawning population and an outlet spawning one, the latter comprising three sub-populations. Reproductive isolation due to inlet (lacustrineadfluvial) and outlet (allacustrine) spawning occurs in other lakes, e.g., Lough Melvin, Ireland (Ferguson, 2004). As only a short stretch of the outlet is available for spawning it is likely that spawning also occurs in the adjacent shores of the loch where suitable gravel is present and where diffuse groundwater flow from surrounding land or wind action can provide sufficient oxygenation for the developing embryos (Whitlock et al., 2014). Thus, the outlet population may comprise individuals from several discrete spawning areas in and around the outlet. Lake shore spawning of S. trutta has been demonstrated in several upland Norwegian lakes, especially where groundwater influx occurs and is potentially an important strategy where harsh weather conditions occur; e.g., periodic bottom freezing of rivers (Heggenes et al., 2009;Thaulow et al., 2014). These authors found genetic differentiation among juvenile S. trutta from separate sites within a lake and between adjacent lake and river juveniles. The heterogeneity of the two main Loch Fleet populations in September with respect to spatial position in the loch relative to the inlet and outlet rivers further emphasises their distinctness. This distribution possibly reflects the movement of S. trutta to the areas of the loch adjacent to the spawning streams ready for the spawning migration, which for Loch Fleet has been shown to be in October or early November (Turnpenny, 1992 (Bernatchez et al., 2016). Due to its underlying geology, Loch Doon was much less affected by acidification than other lochs and although there is some evidence from anecdotal angler accounts of partial reduction in numbers in the 1980s, the fact that angling continued throughout suggests that the reduction was much smaller than for Loch Grannoch. Continuous S. trutta catch records from 1908 onwards are available for the adjacent Lochs Macaterick and Riecawr and while these again show a reduction in catches in the 1970s, moderate catches persisted throughout (Harriman et al., 2001). It is clear from both angling records (see Section 1) and N e estimates here from the 1980s and 1990s that S. trutta numbers were reduced considerably in Loch Grannoch.
As discussed in Section 4.1, the N e of the Loch Grannoch stock was <30 in the 1980s and early 1990s, with the N e in each Loch Grannoch river clearly being considerably lower still. Where there are two or more spatially distinct spawning areas for S. trutta and the N e in each area is small then the spawning groups will diverge as a result of genetic drift exceeding gene flow due to straying (Ferguson, 1989).
Natal homing would serve to maintain this differentiation and the populations may further diverge as they adapt to local conditions and acquire distinct life histories thereby reducing competition (Hendry et al., 2007). Although most of the larger lochs examined in this study probably have two or more spawning areas, it is only where N e is small that sufficient genetic differentiation occurs, resulting in detection with the analyses and number and type of markers used in this study.
Although several studies have reported S. trutta genetic structuring in large lake systems (Ferguson, 2004;McKeown et al., 2010;Swatdipong et al., 2010;Verspoor et al., 2019), these appear to be the result of colonisation by multiple allopatrically differentiated lineages or occur in large lakes where gene flow is limited by distance between rivers. In the case of Loch Grannoch S. trutta, initial analyses do not indicate any differences in morphology or feeding among the three populations (A.F., unpublished data). Two genetically distinct populations in Lakes Trollsvattnet, Sweden (Palmé et al., 2013), which do not differ in feeding ecology and differ only marginally in morphology (Andersson et al., 2017b), are thought to be reproductively isolated due to respective inlet and outlet spawning, although the results were not fully conclusive and other factors may be involved (Andersson et al., 2017a). Sympatric S. trutta populations are likely to be more widespread than hitherto reported as most suitable lakes have not been examined in sufficient detail. Indeed, sympatric populations would be expected in all lakes with multiple spawning locations and in such situations trophic and morphological differentiation would not necessarily be present. However, in the absence of phenotypic differences that allow a priori grouping and where the N e of each population is large and some gene flow exists, it would require detailed sampling, appropriate molecular markers and rigorous statistical analyses to detect the low-level genetic differentiation that is likely to be present (Jorde et al., 2018;Verspoor et al., 2019). In such situations, examining differentiation among samples from actual spawning rivers or locations may be more appropriate than a pooled sample from a lake, although the former may be logistically difficult to obtain in some cases, especially where lake spawning is involved.

| Conservation and wider scientific importance
From a S. trutta conservation standpoint, the most important loch in south-west Scotland is Loch Grannoch due to its genetic distinctness coupled with its increased tolerance of acidic conditions. Although acidification has been reduced in upland freshwaters in Great Britain (Battarbee et al., 2014), predicted climate change poses a threat to this recovery through an increase in rainfall and the intensity and number of storm events resulting in acidifying sea-salt deposition, as well as increased nitrate leaching from soils (Kernan et al., 2010).
These weather-related changes could result in the remobilisation of toxic aluminium and other substances present in catchment peats (Battarbee et al., 2014). Increased CO 2 levels can also result in acidification, an aspect as yet poorly studied in freshwater systems compared with marine ones (Ou et al., 2015). Thus, the Loch Grannoch stock may be an important donor for the restoration, or genetic rescue (sensu Frankham, 2015), of further S. trutta populations in the future. Loch Grannoch S. trutta are of considerable interest for the scientific study of local adaptation and population structuring, given the existence of three genetically distinct populations. A detailed conservation and management plan for Loch Grannoch is urgently required as it was evident during this study that a considerable S. trutta harvest occurs through permitted and especially non-permitted angling due to the loch having a good number of S. trutta of larger size than most other lochs in the area (Supporting Information Table S1). Loch Grannoch should be accorded legal protection status (e.g., as a again merit specific S. trutta management plans. Their genetic isolation means that they provide independent replicates, ideal for the study of parallel and convergent aspects of local adaptation (Merilä, 2014).
Unlike lakes in many other areas of western Europe, the isolated south-west Scotland populations also represent native populations with negligible influence from domesticated farm S. trutta and are thus of considerably increased conservation importance. They are among the most genetically divergent populations so far described for S. trutta in northern Europe with much of the native genetic diversity still intact despite the effects of acidification. Such genetically divergent populations are very important when it comes to conserving overall S. trutta diversity (Kelson et al., 2015;Vøllestad, 2018 (Winfield et al., 2011).

ACKNOWLEDGEMENTS
We are grateful to many organisations and private land-owners for permitting and facilitating our study including Ayrshire Rivers Trust,