2.3 Population genetic analyses of 96 SNP array
Allele frequencies and deviations from Hardy–Weinberg proportions measured as F IS and their associated significance levels for the 96 SNP fluidigm array were obtained from GENEPOP (version 4.3; Raymond, 1995; Rousset, 2008). Holm’s (1979) sequential Bonferroni approach was applied to adjust significance levels when evaluating the results from multiple testing. F ST(Weir & Cockerham, 1984) was estimated using FSTAT v2.9.4 (Goudet, 2003). CHIFISH v5.0 (Ryman & Palm, 2006; available at http://www.zoologi.su.se/~ryman/) was used for F ST significance testing.
We also computed Nei´s (1973) parametricF ST(F ST=(H T-H S)/H T) using GenAlEx v6.5 (Peakall & Smouse, 2012) to allow direct comparison with F STcalculated from Pool-seq data where only Nei´sF ST is possible to obtain. Confidence intervals for Nei’s F ST were calculated using the following equation: F ST ± tdf√s2/n (s2 is the variance ofF ST among loci), and for Weir & Cockerham’sF ST using FSTAT (Goudet, 2003). We note that Nei´s pairwise F ST is typically around half that of Weir & Cockerham´s. To avoid confusion, we consistently try to indicate the type of F ST we refer to.
We assessed the most likely number of populations (K ) using STRUCTURE v2.3.4 (Pritchard, Stephens, & Donnelly, 2000) using the default model allowing population admixture and correlated allele frequencies. We used a burn-in of 250,000 steps and 500,000 Markov chain (MCMC) replicates to estimate Q(assignment probability for each individual to each cluster) and likelihoods for different K (= 1–15). Estimation of the most likely K was repeated over ten runs and the output was analyzed using STRUCTURE HARVESTER v0.6.94 (Earl & vonHoldt, 2012). Mean individual Q values to each deme over runs were derived from CLUMPP (Jakobsson & Rosenberg, 2007). We based our estimation of the most likely value ofK on the mean likelihood value from STRUCTURE, ΔK(Evanno, Regnaut, & Goudet, 2005) from STRUCTURE HARVESTER, and on results from KFinder v1.0 (Wang, 2019).
We also explored population structure using BAPS v6.0 (Corander, Marttinen, & Mantyniemi, 2006) and the details from this analysis are provided in Appendix S2. We constructed an individual-based neighbor-joining phylogenetic tree based on Nei’s D A distance estimates (Nei, Tajima, & Tateno, 1983) from the 96 SNP array using POPTREE2 (Takezaki, Nei, & Tamura, 2009), and MEGAX 10.0.5 (Kumar, Stecher, Li, Knyaz, & Tamura, 2018). We used the default number of bootstrap replications (1,000); the tree was condensed to only include branches with bootstrap support of at least 70%.

2.4 Pool-seq data processing and variant calling

We assessed the quality of the raw sequence reads of each pool using FastQC v0.11.5 (Leggett, Ramirez-Gonzalez, Clavijo, Waite, & Davey, 2013), and the results from different pools were jointly evaluated using MultiQC v1.5 (Ewels, Magnusson, Lundin, & Käller, 2016). Low quality bases (phred score <20) and Illumina adapters were trimmed off the reads using BBDuk as implemented in BBTools v38.08 (http://sourceforge.net/projects/bbmap/). The trimmed reads were mapped against the brown trout reference assembly (comprising 2,371,863,509 bp; https://vgp.github.io/genomeark/Salmo_trutta/) using the Burrows–Wheeler Aligner v0.7.17 (BWA, using bwa mem algorithm; Li & Durbin, 2009). Resulting bam files were sorted, merged per pool and filtered for paired reads using SAMtools v1.8 (Li et al., 2009). The quality of the obtained bam files per pool were evaluated with Qualimap v2.2.1 (García-Alcalde et al., 2012) and summarized in MultiQC v1.5. Read depth histograms obtained from Qualimap were assessed to define minimum and maximum depth thresholds for subsequent population genomic analyses. SAMtools was applied for variant calling using minimum mapping quality and base quality scores of 20 and the parameter “base alignment quality” (BAQ; “-B”) to reduce false SNPs caused by misalignments, resulting in one mpileup file for the two pools. We used the ‘identify-genomic-indel-regions.pl’ script of PoPoolation2 v1201 (Kofler, Pandey, & Schlötterer, 2011) to remove any indels from the mpileup file. No SNPs were kept from the error-prone 5 bp upstream and downstream of indels. A synchronized file was created for downstream analyses using the ‘mpileup2sync.jar’ script of PoPoolation2.