Leveraging Whole Genomes, Mitochondrial DNA and Haploblocks to Decipher Complex Demographic Histories: An Example From a Broadly Admixed Arctic Fish

ABSTRACT

The study of phylogeography has transitioned from mitochondrial haplotypes to genome‐wide analyses, borrowing from population genomics methods along the way. Whole‐genome sequencing allows the study of both mitochondrial and nuclear DNA and provides the density of markers to investigate recombination along the genome. This level of resolution could unravel complex histories of admixture between lineages, which are commonly observed in species evolving in recently deglaciated habitats. In this study, we sequenced 1120 Arctic Char genomes from 33 populations across Canada and Greenland to characterise patterns of genetic variation and diversity, and how they are shaped by hybridisation between the Arctic and Atlantic glacial lineages. Mitochondrial genomes across the study area were predominantly of Arctic origin, except in Greenland, where we observed some Atlantic descent. Through admixture analyses and demographic inferences on nuclear markers, we identified that all Canadian populations under the 66th parallel showed introgression from the Atlantic lineage, leading to higher genetic diversity. By scanning the genome using local principal component analyses, we identified putative large low‐recombining haploblocks as local ancestry tracts from either lineage. Since haplotypes might retain different signatures of postglacial histories by sheltering sequences from recombination, we attempted to infer origins of recolonisation using whole genomes vs. ancestry tracts for the Arctic lineage. Despite limitations, we unveiled clues suggesting a complex postglacial history in Arctic Char. Overall, our study demonstrates that, even at low depth, making the most of whole‐genome sequencing by analysing several genomic compartments provides a versatile and powerful way to address phylogeographic dynamics.

1. Introduction

Understanding a species' evolutionary and demographic history is central to interpreting the processes that produced its contemporary genetic diversity and population structure. Since its inception, the field of phylogeography has aimed to interrogate the microevolutionary processes responsible for the geographical distribution of genetic lineages (Avise et al. 1987). For 2 decades, phylogeography was dominated by the study of gene trees based on mitochondrial DNA (mtDNA) sequences because of their easy sequencing, short length and simple uniparental inheritance without recombination (Avise et al. 1987; Edwards et al. 2015). With the advent of high‐throughput sequencing, phylogeographic studies now commonly use thousands of nuclear DNA (nuDNA) markers such as single nucleotide polymorphisms (SNPs). Frequent cases of discordance between mitochondrial and nuclear genetic variation have highlighted the limitations of mtDNA (Ballard and Whitlock 2004; Toews and Brelsford 2012; Zink and Barrowclough 2008). As an example, mitochondrial markers are matrilineally inherited, leading to potential impacts of sex‐biased migration on genealogies (Osada 2011; Wood et al. 2005). Despite these limitations, mtDNA remains a valuable tool in phylogeographic studies: being haploid, nonrecombining and having a lower effective population size, they provide a rich source of information independent from the nuclear genome. Used in conjunction with nuclear markers, they offer a more comprehensive understanding of species' evolutionary histories and genetic diversity.

The integration of genome‐wide nuclear markers into phylogeography has also blurred the distinctions between this field and population genetics (Edwards et al. 2015). Population genomics continues to rely heavily on averaging the genome‐wide genetic variation of SNPs, for example, as a means to investigate population structure, while ignoring fluctuations in genetic variation at a finer genomic scale (Bhatia et al. 2013; Peter 2022). However, these ‘vertical’ approaches, stacking information from individual genetic markers across individuals, usually ignore or purposely try to reduce the impacts of linkage disequilibrium, neglecting the ‘horizontal’ signal contained in haplotypes (Leitwein et al. 2020). Haplotypes are usually defined as DNA sequences where alleles for multiple markers are inherited as a block. Unlike SNP‐based approaches focusing on mutations and assuming independence between markers, haplotypes carry information about recombination events and have the potential to increase accuracy in the inference of selection, gene flow and population structure (Martin et al. 2017; Shipilina et al. 2023). In particular, long haplotypes, hereafter haploblocks, may be maintained in low recombination regions or within structural rearrangements, providing information about deeper coalescent times than short haplotypes within high recombination regions (Lawson et al. 2012).

Phylogeographic studies are particularly relevant for arctic species, as their habitats and ranges were among the most impacted by the glacial cycles of the Pleistocene. As demographic bottlenecks and extended periods in allopatry lead to the accumulation of divergence between populations that survived in distinct glacial refugia (Hewitt 1996, 2000; Shafer et al. 2010), postglacial recolonisations are commonly the stage for secondary contact between previously isolated lineages (Bernatchez and Wilson 1998; Hewitt 1999). These cases of hybridisation between lineages lead to an admixture of genetic backgrounds, producing a mosaic of local ancestry tracts that tend to shorten through the recombination of chromosomes over generations of backcrossing (Leitwein et al. 2020; Liang and Nielsen 2014). Whole‐genome averaging approaches can struggle to properly identify admixture at a relatively recent glacial timescale, as isolation‐by‐distance can produce similar genomic signatures (Wiens and Colella 2024). Therefore, integrating multiple aspects of the genome (e.g., mtDNA and haplotype information) may help distinguish between true admixture events and patterns that arise due to gradual changes in genetic variation over geographical space.

Arctic char ( Salvelinus alpinus ) is a widespread circumpolar species with the northernmost distribution observed in freshwater or diadromous fishes, most of which coincides with regions covered by ice sheets at the last glacial maximum, around 18,000 years ago (Bernatchez and Wilson 1998; Hocutt and Wiley 1986). Arctic Char has been called ‘the most diverse vertebrate on earth’ (Klemetsen 2010) with many localities harbouring morphs diverging in size, diet, habitat choice, phenology, migratory tactic or life‐history traits (Reist et al. 2013). Some authors recognise dozens of subspecies, and five main glacial lineages have been defined using mtDNA (Brunner et al. 2001; Moore et al. 2015). However, these fail to offer a perfectly discrete classification of wild populations around the globe, as postglacial secondary contact zones have been described between most pairs of neighbouring mtDNA lineages (Dallaire et al. 2021; Gordeeva et al. 2021; Jacobsen et al. 2022; Salisbury et al. 2019). With their unique blend of moderate philopatry, higher dispersal rate than other anadromous salmonids (Moore et al. 2013), and range spanning the most recently deglaciated region around the globe, Arctic Char offers unique opportunities for the study of the distribution of genetic diversity in postrecolonisation populations. Notably, the origin of the postglacial recolonisation of the ‘Arctic’ mtDNA lineage (as first named by Brunner et al. 2001), presently found throughout most of the Canadian Arctic, has remained cryptic, while both Beringia and the Arctic Archipelago have been suggested as potential refugia (Moore et al. 2015).

In this study, we used low‐coverage whole‐genome sequencing to investigate the recolonisation of the recently deglaciated Canadian Arctic by two lineages of Arctic Char and their putative hybridization. We sequenced 1120 Arctic Char genomes (average coverage = 2×, with 100 samples in 5 populations further sequenced at a depth of 8×) from 33 river systems. We aimed to cover most of the 4000 km wide range of the previously defined Arctic mitochondrial lineage in northern Canada and western Greenland to characterise patterns of genetic variation, diversity and admixture with the neighbouring Atlantic lineage. We combined information from both mitochondrial and nuclear genomes including mtDNA haplotype (hereafter mitotype) networks, genome‐wide admixture and ancestry, low‐recombining haploblocks and demographic inference to resolve the history of the secondary contact between the Arctic and Atlantic glacial lineages. Finally, to highlight the complexity of this recent evolutionary history, we aimed to infer the origin of postglacial recolonisation of the Arctic lineage, that is, the position of its refugia during the last glaciation, by comparing derived allele distributions across the genome and inside haploblocks descending from the Arctic lineage potentially carrying different nonrecombining ancestries.

2. Material and Methods

2.1. Library Preparation, Sequencing and Data Preprocessing

Anadromous Arctic Char adipose fins were collected at the mouth of rivers across Arctic Canada and Western Greenland between 2007 and 2023 and stored in ethanol 95% (4°C) or RNAlater (−20°C). Whole‐genome libraries were prepared following Mérot et al. (2021) (which modified protocols by Baym et al. 2015; Therkildsen and Palumbi 2017). Briefly, genomic DNA was extracted using NucleoMag Tissue kits (Macherey‐Nagel) and RNAase (Qiagen) following the manufacturer's instructions. Fragments shorter than 1 kb were removed using Axygen magnetic beads in a 0.4:1 ratio, then each extract was quantified with AccuClear Ultra High Sensitivity dsDNA Quantification kits and diluted to 1.5 ng/μL. Sample extracts (2.5 μL) were then used for Nextera library preparation: (1) tagmentation, that is, the simultaneous fragmentation of the DNA and addition of partial adapters; (2) a two‐step PCR procedure (8 + 4 cycles) with the KAPA Library Amplification Kit to attach Illumina dual index barcodes (custom primers derived from Nextera XT sets A, B, C, and D) and amplify the libraries; and (3) size selection for fragments 400–700 bp in length using Axygen magnetic beads in a two‐step cleaning protocol. Longer fragments were removed by keeping the supernatant in a 0.35:1 bead solution, then shorter fragments were removed by keeping the beads in a 0.7:1 solution. The final distribution of fragment length in libraries was assessed by Agilent BioAnalyzer for a subset of 10 samples per plate. For the low‐coverage dataset, we performed sequencing on lanes of Illumina NovaSeq 6000 (150 bp paired reads), with around 96 pooled individual libraries per lane (range of 75 to 106), yielding on average 31.6 million reads per sample for 1178 samples in 33 populations. For the medium‐coverage dataset, we selected 100 random samples in five populations (20 per population) and repeated sequencing on five lanes to obtain an average of 135.9 million additional reads per sample.

All data were prepared following the pipeline described at https://github.com/enormandeau/wgs_sample_preparation. In brief, raw sequences were trimmed using fastp (Chen et al. 2018) and aligned on the ASM291031v2 reference genome with bwa mem (minimum alignment quality of 10) (Li 2013). Note that the reference genome used was assembled using sequencing data from a Dolly Varden ( Salvelinus malma ) or an S. alpinus × S. malma hybrid (Christensen et al. 2021) but remains the closest high‐quality reference for Arctic Char at the time of our analysis. Duplicate reads were then removed using picard (Picard Toolkit 2019), indels were realigned using GATK (Van der Auwera and O'Connor 2020), and overlapping ends of paired reads were clipped using GATK. The average per‐base depth of coverage was then estimated using mosdepth (Pedersen and Quinlan 2018), and 61 samples with genome‐wide coverage under 1× were excluded from analyses on nuclear DNA. The remaining samples had an average sequencing depth of 1.80× (sd = 0.48×, Figure S1), hereafter named the ‘2× dataset’. For the five populations targeted by additional sequencing, aligned reads from both sequencing batches were merged (average depth of coverage = 7.88×, sd = 1.13×) to create the ‘8× dataset’.

2.2. mtDNA and nuDNA Genotyping and Filtering

The depth of coverage was much higher on the mitochondrial genome (mean = 196×), allowing us to call haplotypes for the whole mitogenome (16,659 bp) for every individual in the 2× data set using angsd ‐doFasta 1 (Korneliussen et al. 2014). Sequences were inspected in Geneious 2022.0.1 (https://www.geneious.com) to remove individuals with more than 5% missing data and mask invariant positions. We used popart (Leigh and Bryant 2015) to construct a median‐joining network of mitotypes extracted for the D‐loop control region (998 bp), COI (1551 bp) and cytb (1141 bp), using an ε parameter of zero. An individual sampled in population 30 matched the D‐loop sequence for Salvelinus fontinalis (AF154850, Doiron et al. 2002) and was kept as an outgroup for the mitotype networks but was removed from every subsequent analysis.

Nuclear SNP positions were first detected using ANGSD v0.937 (Korneliussen et al. 2014) (−GL 2 ‐remove_bads 1 ‐minMapQ 30 ‐minQ 20 ‐skipTriallelic 1 ‐uniqueOnly 1 ‐only_proper_pairs 1 ‐min_Maf = 0.01) on both the 2× and 8× data sets independently with the following coverage filters (for 2× data: at least 1× in 75% of samples, maximum average depth of 4×; for 8× data: at least 3× in 75% of samples, maximum average depth of 20×). The sex of sampled fish was unknown. As such, SNPs on sex‐linked regions identified by Beemelmanns et al. (2025) were not considered to avoid potentially unbalanced sex ratios being confounded with population differentiation.

The 15.6 (2×) and 21.5 (8×) million retained SNP positions were tested for their deviation from expected patterns of heterozygosity and read ratio using the calcLR function in ngsParalog (Linderoth 2018). In brief, ngsParalog analyses the individual distribution of reads to detect signs of mismapping or collapsed paralog regions. ngsParalog likelihood ratios were compared to χ ² distribution and SNPs with a p value (adjusted by applying the Bonferroni procedure) under the threshold of 0.001 were considered deviant (2×: 9.63 million, 62% of all SNPs; 8×: 11.20 million, 52%). These numerous deviant SNPs were removed from the site list for each dataset, as they are expected to have their inferred genotype biased by mismapped reads related to paralogy or repeated elements, two features that are expected to be especially prominent in salmonid genomes following a recent whole‐genome duplication (Dallaire et al. 2023). For analyses based on site‐frequency spectra (SFS), we masked the reference genome in 150 bp regions centred on every deviant SNP, as well as all coding genes and a 1‐kb region around them.

There has been considerable progress in analytical methodology allowing inferences of individual‐ or population‐scale genetic indices using genotype likelihoods (e.g., allele frequencies, population structure, level of admixture, genetic diversity) with high accuracy when the sample size per population is high enough, like in our study (Alex Buerkle and Gompert 2013; Lou et al. 2021; Nielsen et al. 2012). As such, we estimated genotype likelihoods for 5.98 million nondeviant SNPs in the 2× data set using ‐doGlf ‐GL2 in ANGSD. We used ngsLD (Fox et al. 2019) to estimate linkage between SNPs within 500 kb using a random subset of half the samples and pruned the dataset with the graph method until no SNP pairs within 200 kb had an r ² above 0.1, leaving 375,236 SNPs in the LD‐pruned data set.

2.3. Population Structure, Differentiation and Diversity

The general distribution of genetic variation and population structure was first assessed by performing a principal components analysis on the LD‐pruned 2× data set in PCAngsd 1.10 (Meisner and Albrechtsen 2018). We then explored the hierarchical population structure using NGSadmix (Skotte et al. 2013) on the LD‐pruned 2× data to estimate individual admixture while varying the number of ancestral populations (K) from 2 to 33.

Pairwise population differentiation was estimated using allele frequency difference (AFD) and F_ST index. For AFD, we estimated allele frequency in every population for all 5.98 million SNPs (2×) using angsd ‐domaf 1, then calculated the average absolute MAF difference between each pair of populations. For F_ST, we used ANGSD and the reference genome masked for deviant SNPs and genes to estimate sample allele frequency (SAF) for each population, then we used the winSFS streaming option to compute two‐population SFS after shuffling genomic positions to break linkage disequilibrium patterns (Rasmussen et al. 2022). The SAF and 2D‐SFS served as input for the realsfs fst index, print, stats and stats2 functions in ANGSD to calculate genome‐wide weighted F_ST (Reynolds et al. 1983). We tested for the presence of isolation‐by‐distance through linear regression: we compared the marine distance between population pairs, estimated using the R package marmap (Pante and Simon‐Bouhet 2013), with either their AFD or linearised F_ST (Slatkin 1993).

We estimated nuclear genetic diversity in populations with Watterson's estimator (ϴ_W) and nucleotide diversity (ϴ_π), as well as individual heterozygosity. ϴ statistics were computed with ANGSD's realsfs saf2theta and thetaStat do_stat functions using the SAF by population described above and a 1D‐SFS produced by winsfs. To account for the variable coverage along the genome and across populations, we compared per‐site ϴ in windows of 100 kb (step of 10 kb) and excluded windows with a number of sites (denominator) under the 5th percentile (2600 sites) in any population. Individual heterozygosity was estimated by computing a per‐sample SAF and dividing the number of heterozygous sites (SAF = 1) by the number of sites.

We also estimated diversity metrics on whole‐mitogenome sequences, after removing mitotypes from the Atlantic lineage (i.e., those with D‐loop sequences closely matching Atlantic mitotypes identified in Jacobsen et al. 2022 and Salisbury et al. 2019), those with more than 5% of missing data, and masking positions with remaining missing values. The 1074 Arctic mitotypes were 16,483 bp long and were analysed by population with the pegas (Paradis 2010) package in R for (1) ϴ_W with theta.s function, (2) ϴ_π with the nuc.div function, (3) Tajima's D with the Tajima.test function, (4) mitotype diversity using the haploFreq function and the formula corrected for sample size from (Nei and Roychoudhury 1974), and (5) the population frequency of the mitotype most frequently observed across the whole data set. We tested for geographical patterns in populational mtDNA diversity using regression models for those five variables with the marine distance from the westernmost population (pop.1) as the explicative variable. We used the Akaike information criterion (AIC) (Akaike 1974) to compare the null, simple and second‐degree polynomial models.

2.4. Secondary Contact, Introgression and Demographic Inference

To test hypotheses regarding the admixture between the Arctic and Atlantic lineages of Arctic Char in Canada, we downloaded S. alpinus raw sequences (PRJEB62707) from a landlocked population in an inland lake of Sweden (Lilla Stensjön) (Saha et al. 2024, on NovaSeq 6000, 150 bp reads), which we assumed descends purely from the Atlantic lineage. We also acquired unpublished data from a Dolly Varden ( Salvelinus malma ) population in Babage River, Yukon, Canada (Labrecque et al. in prep), sequenced in our lab following a protocol identical to that used in the present study. Those sequences were trimmed, aligned and cleaned as described in the ‘data preprocessing’ section, then we estimated genotype likelihoods and population MAF for the LD‐pruned SNP list. We obtained an average depth of coverage of 3.61× (sd = 0.29×) for the Swedish population and 2.41× (sd = 0.80×) for S. malma .

We repeated the NGSadmix analysis for K = 2 with the Swedish population. We then converted the estimated MAF of all 35 populations of S. alpinus and S. malma to approximate allele count by multiplying the MAF by the number of sampled alleles (twice the sample size) to fit the Treemix file format. We computed f₄ statistics in Treemix 1.13 (Pickrell and Pritchard 2012) with S. malma as the outgroup (H₄), the Swedish population as the potential introgressor (H₃), and all pairs of Canadian and Greenlandic populations as H₁ and H₂.

Based on the previous results, we build alternative demographic models to test the presence of ancient or recent gene flow between the Atlantic lineage, represented by the Swedish population and Southern Arctic Canada (see Figure 4B). We estimated the unfolded 3D‐SFS between COP (North, pop. 4), TJL (South, pop. 25) and LLS (Sweden, pop. 34) using winsfs. Alleles were polarised based on an ancestral reference genome constructed by aligning WGS data from four closely related species (Atlantic Salmon, Salmo salar ; Lake Trout, Salvelinus namaycush ; Rainbow Trout, Oncorhynchus mykiss ; and Chinook Salmon, Oncorhynchus tshawytscha ) on the Salvelinus sp. reference genome (ASM291031v2), as in Dallaire et al. (2023). We used the most common allele to create a consensus for each species using angsd ‐doFasta, then merged the consensus in Geneious 2022.0.1 to keep only positions covered in at least 2 consensuses. The fit between the observed SFS and models with 0 no, (1) ancient, and (2) recent gene flow was compared by running 100 runs of each in fastsimcoal 2.7 with the following parameters: ‐d ‐C 10 ‐n 500,000 ‐L 50 ‐s 0 ‐M. To estimate parameters in the best‐supported model (lowest AIC and Δ_Likelihood), we complete 50 runs of fastsimcoal on SFS estimated on independent subsamples of linkage groups (35 out of 40) with the following parameters: ‐d ‐C 10 ‐n 10,000 ‐L 40 ‐s 0 ‐M ‐c 5.

FIGURE 4.

Evidence for gene flow between the Atlantic and South Arctic. (a) Simplified demographic model showing ancient divergence between the Arctic and Atlantic, followed by isolation with migration of the North (COP, pop. 4) and South (TJL, pop. 24) Arctic. (b) Support (AIC and Δ _Lik, i.e., difference between observed and predicted likelihood of the SFS) for alternative fastsimcoal models (100 runs per model) for 0 absence of gene flow, (1) ancient or (2) recent gene flow between the South Arctic and Atlantic lineage (represented by the Swedish population), as highlighted in red in panel a. (c) f ₄ test for introgression from the Atlantic lineage (represented by the Swedish population), where a negative f ₄ statistic supports introgression in population H₁ (only tests with populations from northern regions are shown) and a positive f ₄ supports introgression in population H₂ (detailed on the y‐axis). Similar lcWGS data from Dolly Varden ( Salvelinus malma ) was used as an outgroup.

2.5. Introgression and Recombination Landscapes

To investigate the fine‐scale genomic landscape of admixture between the Arctic and Atlantic lineages, we scanned the genome for patterns of genetic variation distinct from the genome‐wide signal, using local PCAs as described in Huang et al. (2020) and Mérot et al. (2021). Briefly, we scanned each chromosome independently by computing the covariance matrix between individuals in nonoverlapping windows of 100 SNPs in PCAngsd and detecting clusters of consecutive windows presenting similar PCA patterns using the pcdist function and multidimensional scaling (MDS) in the R package lostruct. Windows with MDS values beyond three standard deviations from the chromosomal mean on one of the first five MDS axes were kept, and clusters within 20 windows of each other were merged. We only considered clusters longer than 10 windows.

In each outlier cluster, we tested for the presence of patterns on PC1 indicative of the three genotypes of a chromosomal inversion or other types of nonrecombinant haploblock. The R package dbscan allowed for the grouping of individuals based on their density distribution on PC1 while allowing for uncategorised individuals between clusters. We iteratively ran dbscan while increasing the ε parameter by increments of 0.005, starting from 0.01, until three groups were detected or ε = 0.15 was reached. We further cleaned the dbscan results by setting individuals with PC1 values outside 3 standard deviations from the mean of any group as intermediate genotype. The three remaining groups were labelled haplogroup AA, AB and BB, starting from the most frequent one, and the intermediate genotypes were labelled A0 and 0B.

To test if the putative haploblocks were associated with lower recombination rates, we estimated recombination using LDhat for each chromosome and population in the 8× data set. To produce the bcf file in input, we genotyped previously identified nondeviant SNPs in ANGSD using ‐doBcf 1 ‐doGeno 4 ‐doPost 1 ‐postCutoff 0.95 with more stringent filters (population MAF > 0.05, minimum of 4× in 90% of individuals). Population‐scaled recombination rates (ρ/kb) were estimated in LDhat 2.2 (Auton and McVean 2007) while controlling for demography with LDpop (Kamm et al. 2016) using a θ estimate of 0.001, a block penalty of 5 and other parameters set following the pipeline in Brazier and Glémin (2024). The LDhat output was cleaned by removing plateaus of invariable recombining rates longer than 22 kb (99th percentile of the length of all plateaus). The weighted mean of recombination rates inside each putative haploblocks and other MDS outliers was compared to the chromosome‐wide average through mixed‐effect models with the chromosome and population set as random effects. Haploblocks were only considered in recombination maps that covered at least 60% of their length (see e.g., chromosome NC_036857.1 for TJL in Figure S10, displaying invariable plateaus over the long regions overlapping haploblocks).

2.6. Inference of the Origin of Recolonisation

To infer the origin of the recolonisation after the last glacial maximum, and thus a putative glacial refugium, we computed the directionality index (ψ) for each population pair from their genome‐wide unfolded 2D‐SFS, following Peter and Slatkin (2013). Ancestral alleles private to either population were excluded. We then used the Time Difference of Arrival algorithm (Gustafsson and Gunnarsson 2003) on the ψ matrix following Prior et al. (2020). This approach uses the expected spatial distribution of derived alleles in recent range expansions to triangulate the origin of recolonisation. Briefly, a pairwise comparison of ψ and geographical distance between focal populations and the assumed origin was performed for every possible origin, using a 0.5 decimal degrees step across the study area. For origins with a positive slope between ψ and geographic distance, we rescaled the coefficient of regression (R ²) from 0 to 1, and origins with a higher R ² were considered more likely. For this analysis, we excluded populations out of Canada, as well as population 30, which had a high level of Atlantic lineage ancestry.

To test how introgression from the Atlantic lineage impacted the inference of the origin of recolonisation, we repeated the ψ and TDoA analyses while focusing on the diversity inside the Arctic haplogroup for the putative haploblocks identified using local PCAs. For each population pair and haploblock, we randomly selected 10 homozygotes for the Arctic haplogroup (AA) by population and computed their SAFs and 2D‐SFS. We then summed the per‐haploblock SFS for pairs of populations with at least 10 AA individuals each and computed a directionality index from the summed spectrum. TDoA was performed on the resulting ψ matrix as described above.

To assess the robustness of this analysis, we repeated this process over 100 replicates on resampled AA individuals (n = 10) and a random subset of 90% of the putative haploblocks. To control for the small number of nonrecombining haploblocks and the difficulty in precisely delineating these putative local ancestry tracts, we added 100 replicates where each haploblock was replaced by a random genomic region of equal length on the same chromosome. TDoA R ² maps from the putative haploblocks and random regions were each rescaled as above, then averaged and compared with the genome‐wide TDoA signal.

3. Results

3.1. Distribution of Nuclear Genetic Variation Across the North American Arctic

Nuclear genome‐wide variation at 5.98 million SNPs (average coverage = 2×, Figure S1) in 1120 anadromous Arctic Char from 33 Canadian and Greenlandic rivers (Figure 1a and Table S1) displayed a strong structure associated with rivers, which grouped into two major genetic clusters corresponding to Northern and Southern areas. Genome‐wide pairwise F_ST ranged from 0.012 to 0.381 among Canadian rivers, while comparisons including Greenlandic sites reached up to F_ST = 0.632 (pop. 1 and 33) (Figure 2a). The general structure associated with longitude was visible on the first axis of a principal components analysis (Figure 1b), explaining 63.9% and 5.5% of the genome‐wide variance, respectively, before and after stringent LD‐pruning (375,236 SNPs). As expected for homing salmonids, each river system appeared genetically differentiated in an NGSadmix analysis (Figure S2).

FIGURE 1.

Highly structured distribution of genetic variation across the Arctic. (a) Sampling sites for 1140 Arctic Chars ( Salvelinus alpinus ) sequenced at a coverage of 2× (circle), 5× (triangle) or 8× (square). Colours indicate sampling regions, and the same colour coding is used in all figure panels. Dotted blue lines delimit the ice margin at the last glacial maximum (18 ka, Dalton et al. 2020). (b) Principal components analysis based on 375,236 independent SNPs. 95% confidence interval ellipses are drawn around each sampling site. (c) Median‐joining mitotype network for the mitochondrial D‐loop control region (993 bp), showing divergence between the Arctic (top) and Atlantic (bottom) mitochondrial lineages. For each mitotype shared by n > 1 individual, coloured sections of the pie chart indicate the proportional representation of individuals from each region.

FIGURE 2.

North–South divide in genetic variation. (a) Pairwise genetic differentiation among populations estimated using allele frequency difference (AFD, above diagonal) and genome‐wide weighted FST (below diagonal). Greenlandic and Canadian populations above and below the 66th parallel are divided by solid white lines, and dotted lines divide regions defined in Figure 1a. (b, c) Isolation‐by‐distance as estimated by the linear regression between genetic (either b) AFD or (c) linearised FST and marine distance between populations 1–29. Population pairs were categorised based on their inclusion in the North (blue, pop. 1–16) or South (red, pop. 17–29) group, or including a Northern and a Southern population (green).

Pairwise differentiation among populations, both based on the F_ST index and allele frequency differences (AFD), highlighted a genetic divide between the northern (pop. 1–16) and southern (pop. 17–30) Canadian Arctic around the 67th parallel (Figure 2a). While populations from both regions exhibited patterns of isolation‐by‐distance (IBD), southern populations were more differentiated at equal marine distances than northern populations (Figure 2b,c). Greenlandic populations and pop. 30 (southeastern Nunavik) were comparatively more differentiated than other populations and did not fit the Canadian IBD patterns (Figure S3). Intrapopulation genetic diversity, as measured by nucleotide diversity (ϴ_π), was also higher in southern (median widowed per‐site ϴ_π = 1.13–2.77 × 10⁻³) than in northern (ϴ_π = 0.72–2.09 × 10⁻³) Canadian populations (Figure 3d and Figure S4b,c).

FIGURE 3.

Higher nuclear diversity in admixed populations. (a) Genome‐wide Arctic ancestry, estimated by admixture proportion in NGSadmix (K = 2), of individuals (smaller points) and sampling sites (mean ancestry, circled points) along a spatial gradient from Inuvialuit (left) to Sweden (right). (b) Individual autosomal heterozygosity (heterozygote sites per kb) is higher in the secondary contact zone (intermediate ancestry), as visualised by a fitted quadratic equation. (c) Per‐site Watterson's estimator (ϴ_W) in whole‐mitogenome haplotypes of the Arctic lineage by population. (d) Per‐site nucleotide diversity (ϴ_π) and Watterson's estimator by population, estimated in 100 kb windows along the nuclear genome. The median (point), 25th, and 75th percentiles (error bars) are shown. The Northern (pop. 1–16) and Southern (pop. 17–30) Canadian populations are highlighted by 95% confidence interval ellipses and a dotted line follows ϴ_π = ϴ_W (Tajima's D = 0).

3.2. Admixture Between Lineages Is a Major Determinant of Variation in Southern Populations

Several lines of evidence support the hypothesis that the large intra‐ and interpopulation diversity described earlier in the Southern Canadian Arctic and Greenland is due to admixture between the genetic material descending from two distinct glacial refugia, that is, the Arctic and Atlantic lineages. First, when conducting an admixture analysis (K = 2, NGSadmix) with the North American sampling sites and the Swedish outgroup, southern Canadian and northwestern Greenlandic populations had mixed genome‐wide ancestry for the two groups. Ancestry was distributed along a cline‐like northwest‐to‐southeast gradient, suggesting admixture between the Arctic (i.e., northern Canadian populations) and Atlantic (i.e., Swedish outgroup) lineages (Figure 3a). Within‐population variance in proportions of ancestry was low (Figure 3a), and putatively admixed populations had higher observed heterozygosity (Figure 3b and Figure S4a) and nucleotide diversity (Figure 3d and Figure S4c), supporting the hypothesis of secondary contact between the Arctic and Atlantic lineages.

Second, to further test the hypothesis of admixture between lineages in the southern part of the study area, we computed f ₄ statistics (Patterson et al. 2012; Reich et al. 2009) using a Dolly Varden ( Salvelinus malma ) population from Yukon, Canada, as an outgroup. This test provided support for introgression from the Atlantic lineage (represented by alleles from the Swedish outgroup) towards the southern Arctic populations (Figure 4c). Finally, using demographic modelling performed with fastsimcoal, the model including ancient gene flow between the Atlantic lineage (i.e., Sweden) and the southern Canadian Arctic population (COP) was better supported than models with no or more recent gene flow (Figure 4a,b; see demographic parameter estimation in Figure S5).

3.3. Mito‐Nuclear Discordance in Admixture and Diversity Patterns

To leverage an additional source of historical demographic information and to compare our results with previous studies, we produced mitotype networks for the D‐loop control regions of the mitochondria (Figure 1c). Contrary to our results using the nuclear genome, the Arctic mitochondrial lineage (as first described in Brunner et al. 2001) dominated in all Canadian populations, with only one Canadian sample from the present study carrying an Atlantic mitotype (in pop. 28). In contrast, the Atlantic and Arctic mitochondrial lineages co‐occurred in western Greenland. A single mitotype dominated the Arctic lineage, with 77.5% of samples sharing the most common D‐loop mitotype. In Greenlandic populations, the individuals carrying mitotypes from either lineage did not differ in their nuclear genome ancestry level inferred with NGSadmix (t‐test per population: p > 0.05). The COI and cytb genes produced very similar mitotype networks, separating Canadian samples from Swedish and some Greenlandic samples, but the divergence between haplogroups was lower than for the D‐loop control region (Figure S6).

Whole‐mitogenome genetic diversity inside the Arctic mtDNA lineage also displayed different trends compared with the nuclear genome. While nuclear diversity was the highest in the admixed populations of the South Arctic and Greenland, both mitochondrial ϴ_W and mitotype diversity were higher near the westernmost population than in the southeastern part of the study area (Figure 3c and Figure S7a,d). Mitochondrial nucleotide diversity (ϴ_π) and Tajima's D were lower at intermediate distances from the westernmost population, that is, the centre of the Arctic mtDNA lineage range. Finally, the frequency of the most frequent whole‐mitogenome haplotype, present in 34% of all Arctic lineage samples, was lower in populations at both extremities of the range of the lineage, that is, in Inuvialuit and Greenland (Figure S7e).

3.4. Heterogeneity in the Introgression Along the Genome

By scanning the genome for exceptions to the chromosome‐wide population structure, we retained 117 independent regions that contained at least 10 MDS outlier windows of local PCA (e.g., Figure 5a, see all outlier regions in Figures S8 and S9). These regions ranged from 98 kb to 5.26 Mb in length (median = 825 kb, total = 133 Mb). Of these regions, we visually identified 46 that showed three clear clusters on a first PC explaining most of the variance (e.g., Figure 5b). We interpret these regions as putative haplotype blocks, or haploblocks, that is, regions of low recombination where polymorphic sites are inherited in high linkage, creating groups of similar haplotypes, or haplogroups. As such, the three observed clusters correspond to the two homozygotes and the heterozygote for these haplogroups. Using populational‐scale recombination maps based on 8× data (Figure S10), we observed that the putative haploblocks tended to have lower recombination rates than the chromosome‐wide average (Figure S11), which could support this interpretation.

FIGURE 5.

Local ancestry inferred through local PCA. Representative case of a low‐recombining diverged putative haploblock corresponding to a local PCA outlier region (win8) on LG9 (NC_036849.1). (a) Position of the MDS outlier region (red) on the linkage group (above) and standardised PC1 coordinates in PCAs for successive nonoverlapping 100‐SNP windows inside and around the outlier region (below). Each sample is represented by a line coloured according to their haplogroup inferred from panel b. (b) PCA for SNPs in the outlier region where the three major clusters found on PC1 are attributed to the 2 homozygotes (AA, blue, left; BB, yellow, right) and their heterozygote (AB, green, middle). Intermediate genotypes (A0, light blue; 0B, light green) were assigned to individuals outside of the cluster distributions. The histogram (above) represents the density of individuals along the first axis of the PCA. (c) Proportion of inferred haplogroups for the win8 haploblock across sampling sites.

Most of these putative haploblocks corresponded to islands of differentiation between the Arctic and Atlantic lineages (win8 on LG9 provides a representative case, Figure 5). 23 haploblocks (median length = 783 kb, total = 20 Mb) were differentially fixed between the northwestern Canadian Arctic (Inuvialuit, Kitikmeot; henceforth identified as haplogroup ‘A’) and Sweden (haplogroup ‘B’) while showing heterozygotes in southern populations of the Canadian Arctic (e.g., Figure 5c), suggesting that they may be tracts of different ancestry that are yet unbroken by recombination. The individual distribution of heterozygotes differed between haploblocks, supporting their independent genealogy, but haplogroup frequencies by population were highly correlated with the Arctic–Atlantic whole‐genome ancestry inferred earlier (NGSadmix, K = 2). Interestingly, two other MDS outliers (e.g., win21 and win58 in Figure S9) also featured three distinct genotypes on PC1, but their spatial distribution did not match the Arctic–Atlantic admixture gradient.

3.5. Inference of Origin of Recolonisation

Directionality analysis conducted on either the whole genome or restricted to variation inside putative haploblocks of Arctic descent supported conflicting recolonisation histories, respectively from the east and west of the lineage range. For the TDoA analysis on the whole‐genome variation, we observed a positive correlation between population‐pairwise ψ (i.e., the difference in derived allele frequencies) and geographical distance from the origin (maximum R ² = 0.638 before rescaling) when testing for origins in the southeast of the sampling area in Canada (Figure 6a). In contrast, the TDoA on genetic variation in a putatively nonintrogressed fraction of the genome, that is, the Arctic haplogroup for the previously identified 23 local ancestry tracts (e.g., AA in Figure 5, total length 20 Mb, 1.3% of the genome), weakly supported an origin of recolonisation in the western part of the sampling area over 100 replicate runs (maximum R ² ranging from 0.007 to 0.107, Figure 6b, Figure S12a). When repeating the TDoA analysis on 23 random genomic regions (total length 20 Mb), we inferred a southeastern origin of recolonisation (maximum R ²: 0.199–0.483, Figure 6b and Figure S12b), which was highly congruent with the whole‐genome TDoA.

FIGURE 6.

Genome‐wide and Arctic lineage‐specific inference of the origin of recolonisation. (a) Time Difference of Arrival (TDoA) analysis for inference of the origin of recolonisation, where red colour indicates a higher correlation (R ²) between distance from the putative origin and whole‐genome directionality index (ψ). (b) TDoA for ψ calculated in homozygotes for the Arctic haplogroup (10 individuals per population) in 23 putative local ancestry tracks (total length = 20 Mb). The mean rescaled R ² for 100 replicate runs is shown in blue, and the mean rescaled R ² for 100 runs on 23 random genomic regions (total length = 20 Mb) is shown in red.

4. Discussion

It has long been recognised that making phylogeographic inferences based on mitochondria alone can lead to biased interpretations (Galtier et al. 2009). However, the nonrecombining nature of mtDNA can offer advantages in specific cases since it can ‘record’ past events that are not blurred by generations of hybridisation and admixture. At the same time, vast amounts of nuclear data are sometimes insufficient in revealing historical patterns if averaged over the genome, and evolutionary inferences could benefit from analysing heterogeneity along the genome, such as identifying haploblocks that retain historical signals (Leitwein et al. 2020; Shipilina et al. 2023). In this study, we sought to maximise the use of whole‐genome sequencing data by integrating mitochondrial and nuclear analyses, both at the genome‐wide and haplotypic levels, to assess the extent of admixture between two glacial lineages of an arctic fish. Intraspecific hybridisation between glacial lineages and the resulting increased nuclear diversity observed in southern populations was a dominant feature of the study system. Such introgression across lineages obscured much of the genomic signal for postglacial recolonisation from either single lineage. We scanned the genome for regions of distinct genetic variation and identified putative local tracts of conserved Arctic or Atlantic ancestry. By comparing genetic variation within tracts descending from the Arctic lineage to the whole genome, we hoped to uncover the position of this lineage's glacial refugium, despite the heavily admixed genetic background observed in our data. While our inference power remained relatively low, we propose an approach that aims to isolate lineage‐specific genetic variation to resolve complex histories of postglacial secondary contact.

4.1. Extensive Admixture Between Lineages and Mitonuclear Discordance

In Arctic Char, we show evidence for the nuclear genetic admixture of the Arctic and Atlantic lineages in most of Arctic Canada under the 66th parallel. This is in stark contrast with the previously described distribution of mitotypes, for which the observed co‐occurrence zones of both lineages are restricted to Labrador (Salisbury et al. 2019) and western Greenland (Jacobsen et al. 2022). While previous phylogeographic work on the species hinted at the introgression from the Atlantic lineage reaching further west at the nuclear level (Dallaire et al. 2021; Moore et al. 2015), here we inferred the presence of Atlantic nuclear alleles thousands of kilometres away from the range limit of the Atlantic mitochondrial lineage, both using f ₄ tests and indirect detection of lineage‐divergent haploblocks.

This kind of mitonuclear discordance has been observed in numerous natural systems (Toews and Brelsford 2012) and could be explained by multiple mechanisms. First, as nonrecombinant haploid sequences, mitogenomes have a population effective size (Ne) that is ¼ of their diploid nuclear counterparts (Hudson and Turelli 2003) and are thus more impacted by genetic drift. Alternatively, sex‐biased dispersion has been suggested in Arctic Char, with males appearing more mobile than females (Dempson and Kristofferson 1987; Moore et al. 2016). As mitochondrial genomes are matrilineally inherited, this could contribute to a discordance in the distribution of nuclear DNA and mitochondrial DNA variation.

We found 13 amino acid changes in coding regions between the most frequent whole‐mitogenome haplotypes of the Arctic and Atlantic mtDNA lineages of Arctic Char. If thermal selection can shape the distribution of mitochondrial sequences (Balloux et al. 2009; Lajbner et al. 2018), selection against the Atlantic mtDNA lineage in the climatic conditions of the Canadian Arctic could explain its near absence outside of Labrador and Greenland, despite signs of introgression from the Atlantic lineage on the nuclear genome. However, Jacobsen et al. (2022) did not find evidence for positive selection at the mitogenomic level. Finally, hybrid incompatibilities are frequent between mitochondrial genes and their nuclearly encoded components (Burton 2022), which could result in stronger selection against introgressing mitotypes.

4.2. Local PCA Reveals More Than Just Inversions

Short‐read whole‐genome sequencing has become increasingly accessible in nonmodel organisms. New analytical frameworks allow for the study of population genetics questions even with low individual depth of coverage, which has opened the door for whole‐genome studies of unprecedented sampling sizes (Lou et al. 2021). However, phasing (i.e., haplotype estimation) might be impossible in these designs without a reference panel (Rubinacci et al. 2021), as haplotypes are most reliably defined and delimited using long‐ or linked‐read data (Meier et al. 2021). It has been suggested that, when they are large and divergent enough (e.g., in regions of low recombination), haploblocks might also be indirectly detectable through genome scans for distinct patterns of genetic variation (Ishigohoka et al. 2024). One such method is the use of local PCAs (H. Li and Ralph 2019), an approach that has risen as a popular approach to detect large recombination‐suppressing structural variants (e.g., inversions) (Huang et al. 2020). The typical signal expected around nonrecombining haploblocks is a PCA forming three distinct clusters on the first axes (generally PC1), with the middle cluster carrying more heterozygous sites than the two others, and higher linkage disequilibrium among clusters than inside each one (Huang et al. 2020).

In this study, we identified local PCA outliers that allowed for the detection of relatively short (250 kb–4 Mb) candidate regions on every chromosome. We analysed the five first MDS axes, but Figure S8 suggests that many additional outlier regions could have been identified on subsequent axes. PC1 shows the expected three‐cluster pattern indicative of haploblocks, that is, the two homozygotes and the heterozygote for a region of low recombination, and inversions may underlie some of those haploblocks. For example, one of our MDS outlier regions indicative of a haploblock (win16) was located on the distal end of LG12, which was recently identified as a putative inversion in Nunavut Arctic Char (Hale et al. 2021). However, we observed that the clustering in three groups was far from perfect in most of the haploblocks, with many intermediate individuals (e.g., Figure 5B). This suggests a suppression of recombination less severe than what has been observed in recent publications investigating inversions (Huang et al. 2020; Mérot et al. 2021). Intriguingly, some candidate haploblocks (including win16) showed additional structure inside an inferred homozygote group on the second axis of the PCA, resulting in a triangular pattern when considering PC1 and 2 (e.g., win50, win89, win91 and win109 in Figure S9), which may result from nested inversions or regions of low recombination maintaining more than two distinct haplogroups. Therefore, overall, we proposed (and confirmed for some) that those regions are characterised by low recombination but do not further speculate with the present data about the precise mechanism protecting the observed haploblocks from recombination.

More interesting is the coherent spatial pattern observed in multiple independent genomic loci, with the two alternative haplogroups being differentially fixed in the Western Canadian Arctic and Sweden, while the haplogroup frequencies by population were highly correlated with the Arctic–Atlantic whole‐genome ancestry. Such correlations further support a scenario of secondary contact and introgression at the nuclear level by revealing blocks of linked polymorphic markers that appear to be inherited from ancestral lineages, after their divergence in allopatry during past glacial periods. Other putative haploblocks, whose spatial patterns did not match the Arctic–Atlantic admixture gradient (e.g., win21 and win58 in Figure S7), could be associated with recent and polymorphic structural variants suppressing recombination such as chromosomal inversions, allowing the accumulation of genetic divergence in sympatry (Michel et al. 2010).

Our local PCA approach constitutes a rather indirect method compared to phasing algorithms based on the Li and Stephens model (Li and Stephens 2003) which require a reference panel (e.g., HAPMIX (Price et al. 2009) and FLARE (Browning et al. 2023)). Alternatively, the Beagle software has made good progress on haplotype phasing using low‐coverage unlinked short‐read sequencing data (Browning et al. 2021) and could be an interesting avenue for further exploration of admixture in lineages of Arctic Char. Here, by exploring how local PCAs highlight local genetic patterns coherent with haplotype structure, we hope to spur interest in improving how lcWGS can be used to their fullest but also advise caution when systematically interpreting local PCA outliers as structural variants like chromosomal inversions.

Here, we only scratched the surface of the haplotypic structure and likely had the power to detect only the longest and most geographically structured haploblocks. However, long haplotypes of foreign ancestry would be expected to have a lower probability of introgressing in the face of purifying selection (Moran et al. 2021). This could suggest that regions that do introgress are under positive selection and might not accurately represent the neutral portion of the genome. A quick look at the structure outside MDS outlier regions (Figure 5a and Figure S8) shows how most of the genome segregates similarly at a finer scale, that is, in smaller blocks. The delimitation of haploblocks could likely be refined by tuning the window size and filtering parameters of the MDS outlier detection. However, some of the imprecision might be linked to reasons both technical (e.g., the uncertainty inherent to the analytical framework of low‐coverage data) and biological (the heterogeneity of recombination along the genome and among populations, breaking ancestry tracks into regions of varying lengths).

4.3. Leveraging Local Ancestry to Disentangle Multiple Origins of Recolonisation

In cases of postglacial secondary contact, pinpointing the origin of recolonisation—that is, the position of the glacial refugium—of each lineage can help provide additional context, as the process of recolonisation itself and its direction can have predictable impacts on within‐lineage genetic variation (Hewitt 2000; Peischl et al. 2015). One such impact is increased genetic drift through sequential founder effects along the recolonisation front (Braga et al. 2019; Hallatschek et al. 2007; Slatkin and Excoffier 2012), leading to a decrease in genetic diversity and an accumulation of derived and sometimes deleterious alleles via increased genetic drift (Koski et al. 2019; Rougemont et al. 2020, 2023).

In our data, we investigated the spatial distribution of derived allele frequencies by applying the TDoA approach to the genome‐wide variation and inferred an origin of recolonisation for the Arctic lineage in the southeastern Canadian Arctic. However, the study system shows clear signs of admixture between the Arctic and Atlantic lineages, suggesting ancestry originating from at least two distinct glacial refugia. Postglacial secondary contact is common in similar studies of phylogeography (Nugent et al. 2024; Sommer and Zachos 2009), but it violates a major assumption of the model (Peter and Slatkin 2013), as the high divergence between glacial lineages most likely overshadows any within‐lineage signal of directionality. The Arctic lineage is expected to have recolonised from a smaller source population than other lineages (Moore et al. 2015), suggesting a demographic bottleneck that could explain the higher frequency of derived alleles in the northern, nonadmixed populations.

With the goal of revealing patterns specific to the Arctic lineage, we focused on genetic variation inside putative local ancestry tracts identified earlier. By considering only homozygotes for the haplogroup that we attributed to the Arctic lineage, we hypothesised that we could limit the influence of introgression from the Atlantic lineage and investigate the origin of recolonisation of the Arctic lineage. While these inferences had lower explanatory power and replicability, they globally pointed in a direction opposite to the genome‐wide analysis. This suggests that genetic diversity inside local ancestry tracts might carry distinct lineage‐specific information that can be leveraged to elucidate more complex evolutionary histories. The remaining challenge involves accurately inferring homozygous tracts of a chosen lineage at the genome‐wide level for all individuals, to avoid biased conclusions drawn from a few genomic regions. Overall, we highlight the interest in dissecting horizontal information present in genomes and further refinement of local ancestry tract identification.

Inferring range expansions from the distribution of derived alleles has several limitations, as other evolutionary mechanisms can create asymmetry in two‐dimensional site‐frequency spectra, such as demographic changes unrelated to the expansion (Mestre et al. 2022). Boundary effects can also act as confounding factors, as populations on the outer part of the distribution will experience increased drift compared to central ones, which can create clines in ψ independently from range expansions (Kemppainen et al. 2024). In light of this and in the absence of rigorous hypothesis testing, our key takeaway focuses less on our confidence in the inference for this specific system and more on the potential of genetic information preserved in large blocks of known local ancestry to unravel complex, multilayered evolutionary histories.

Despite previously discussed limitations, other sources of data were consistent with a postglacial recolonisation from the northwestern Canadian Arctic, such as mitochondrial diversity being higher in this region than elsewhere in the range of the Arctic lineage. While species are intuitively expected to have survived glacial periods in refugia south of the ice extent, fossil and genetic evidence suggest that higher‐latitude refugia might have existed (Mee and Moore 2014; Premoli et al. 2010; Shafer et al. 2010; Sommer and Zachos 2009; Wójcik et al. 2010). Although recent reconstructions suggest a complete glaciation of the Arctic archipelago at the last maximum (Figure 1) (Dalton et al. 2020), data from rodents (Fedorov and Stenseth 2002), birds (Holder et al. 1999) and plants (Abbott et al. 2000; Tremblay and Schoen 1999) point to cryptic refugia in the High Arctic, and Banks Island (Inuvialuit, North West Territories) has been proposed as a refugium for caribou (Klütsch et al. 2017). Ultimately, the use of local ancestry as a lens through which to study recolonisation could offer a promising avenue for disentangling complex postglacial evolutionary histories of the High Arctic and elsewhere.

5. Conclusion

Whole‐genome sequencing (WGS) has been increasingly applied over the past decade, with low‐coverage WGS (lcWGS) emerging as a cost‐efficient strategy for generating large genomic datasets across broad geographic scales. Our study highlights the versatility of lcWGS in resolving intraspecific evolutionary processes in widely distributed organisms by uncovering an extensive postglacial secondary contact zone in Arctic Char. The inclusion of cytoplasmic DNA (from mitochondria or chloroplasts in plants) in WGS also promises a more systematic comparison of markers from different independently evolving segments of the genome, providing new phylogeographic insight while integrating earlier research in long‐studied systems. Finally, WGS data offers far more than an increased SNP count compared to reduced representations of the genome. Indeed, it allowed us to investigate admixture between divergent lineages at the haplotypic level and promises interesting avenues to untangle multiple layers of evolutionary histories. We are confident that the scientific community will benefit from exploring such innovative approaches for leveraging horizontal genetic signals, like linkage and recombination, to deepen our understanding of evolutionary dynamics in a diversity of less‐studied systems.

Author Contributions

X.D. designed the research, performed the research and led the analysis of data and writing of the manuscript. E.N. and T.B. contributed analytic tools and analysed part of the data. L.H. and M.M.H. led and coordinated important parts of the sampling effort and provided feedback on the manuscript. C.M. and J.‐S.M. contributed to the design of the research and the writing of the manuscript.

Disclosure

Benefit‐Sharing Statement: We consulted with the indigenous communities providing the biodiversity resources and hired members of local Inuit Hunters and Trappers Associations to help with biodiversity assessments, including the collection of fin clips from Arctic Char harvested for subsistence. The new data generated here is part of our continued engagement to provide genomic resources to these communities to address priority concerns as part of the FISHES project (Fostering Indigenous Small‐scale fisheries for Health, Economy and food Security). The findings of this study will be part of outreach reports that will be shared with all partners involved. Lastly, as described above, all data have been shared with the broader public via appropriate biological databases.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Appendix S1.

Data Availability Statement

All Salvelinus alpinus raw sequencing data analysed in this manuscript is available on Short Read Archive as part of projects PRJNA1031558 (Canada and Greenland) and PRJEB62707 (Sweden). Whole‐mitogenome haplotypes from Canadian and Greenlandic populations are available on Dryad at: https://doi.org/10.5061/dryad.b2rbnzss2. Sequencing data on S. malma is also planned to be deposited as part of a separate publication, presently in preparation. A bioinformatical pipeline presenting the main lcWGS analyses is available at: https://github.com/xav9536/angsd_pipeline.