Tracing the Drivers of Range‐Wide Bowhead Whale Genomic Structure and Diversity Authors: Michael V. Westbury, Nicholas A. Freymueller, Andrea A. Cabrera, Lutz Bachmann, Steven H. Ferguson, Mads Peter Heide‐Jørgensen, Kit M. Kovacs, Christian Lydersen, Olga Shpak, Øystein Wiig, Damien A. Fordham, Eline D. Lorenzen Published: 2026 ABSTRACT ABSTRACTIn a rapidly warming Arctic, genetic variation might serve to buffer organisms against the effects of environmental change, such as sea ice loss and ocean warming. Yet, this concept remains largely unexplored because comprehensive genome‐wide studies across the full ranges of Arctic marine taxa are rare. Bowhead whales ( Balaena mysticetus ) have a strong association with sea ice and polar water masses, a long history of human exploitation, and a circumpolar distribution, making them a valuable model for evaluating how past environmental and anthropogenic factors have shaped contemporary population variation. We analysed both nuclear and mitochondrial genomes from bowhead whales sampled across the species' range, encompassing all four recognised stocks. Firstly, our results indicate the existence of three genetic groupings instead of four: (1) Okhotsk Sea; (2) East Greenland‐Svalbard‐Barents Sea; and (3) a population containing two recognised stocks—the Bering‐Chukchi‐Beaufort and East Canada‐West Greenland. We utilised high‐resolution ecological niche modelling and bowhead whale movement data to reconstruct inter‐stock habitat connectivity over the last 11,700 years, finding that this explains our identified genetic groupings. Bowhead whale populations exhibit little‐to‐no evidence of recent inbreeding and retain high genetic diversity relative to other mammalian species despite centuries of intensive commercial whaling. The most vulnerable population is that in the Okhotsk Sea, which has the lowest genetic diversity, most inbreeding, and the highest realised genetic load. Collectively, our findings elucidate the recent history and dynamics of bowhead whales, offering valuable baseline data and context on present‐day genetic structure and diversity to support effective conservation and management strategies. Introduction Interactions between genetic and ecological processes are fundamental to understanding contemporary patterns of phylogeographic structure and diversity and how they have been impacted by global change (Westbury, Brown, et al. 2023). This is because contemporary genetic variation is conditioned by past climatic and environmental change along with anthropogenic influences that individually, or in concert, drive population structure and adaptive potential (Theodoridis et al. 2020). Understanding their relative roles is essential for determining current and future threats to the long‐term survival of a species. In marine ecosystems, anthropogenic drivers have modified species' contemporary genetic structure and diversity differently than in terrestrial systems. The latter have been subject to human‐induced pressures across many millennia, including through direct hunting, changes in fire regimes, and landscape modification (Barnosky et al. 2011). In contrast, marine environments have remained largely unaffected by anthropogenic influences until the past ~500 years (Jackson et al. 2001). Consequently, contemporary genetic patterns in marine species are inferred to be more directly linked to natural environmental drivers (Westbury, Cabrera, et al. 2023), although anthropogenic impacts, such as ongoing climate change and industrial exploitation, are becoming more pronounced (Jackson et al. 2001). Bowhead whales ( Balaena mysticetus ) are the only baleen whale endemic to the Arctic seasonal ice zone. They are among the most biologically distinctive mammals, having a slow life history, extreme longevity, and exceptional insulation through their thick blubber layer, which constitutes 40%–50% of their body mass (George and Thewissen 2020). Bowhead whales have been harvested sustainably by Arctic Indigenous peoples for millennia and remain of profound cultural and subsistence importance (Huntington et al. 2021). However, populations of bowhead whales suffered major declines across their circumpolar range during four centuries of commercial whaling (~1530–1914) (Thewissen and George 2021). They were ideal targets due to their slow swimming speed, relatively docile nature, and ease of retrieval after killing, as they float and could be towed. International agreements giving bowhead whales formal protection did not come into effect until 1931, after commercial exploitation had already been abandoned due to whaling no longer being profitable (Freymueller et al. 2026; Tønnessen and Johnsen 1982). Bowhead whales occupy a wide range of Arctic marine environments, from regions of dense, consolidated sea ice to more dynamic marginal ice zones and open water areas dominated by polar water masses. Across their contemporary range, seasonal movements are influenced by a combination of sea ice and oceanographic conditions that structure zooplankton availability, with the relative importance of these drivers varying among populations (Citta et al. 2012, 2015; Dyke and England 2003; Fortune et al. 2020; Heide‐Jørgensen et al. 2021). The fossil record for bowhead whales, which is rich compared to other marine mammals (Dyke et al. 1996; Dyke and England 2003), indicates that distributional shifts during the Late Quaternary resulted from shifting sea ice conditions (Dyke et al. 1996; Foote et al. 2013). However, palaeogenomics and habitat modelling of bowhead whales in the Atlantic sector of the Arctic suggest population stability across the Holocene (past 11,700 years), despite environmental perturbations (Westbury et al. 2026). At a global scale, suitable bowhead whale habitat inferred from ecological niche models is forecast to rapidly decline by the end of the 21st century, indicating an uncertain future for this charismatic species (Chambault et al. 2022; Freymueller, Lorenzen, et al. 2025). The International Whaling Commission, which sets overarching conservation guidelines of bowhead whales across their circumpolar range, currently recognises four geographically segregated management stocks (Figure 1A), based on satellite telemetry and genetic data (Baird and Bickham 2021). These stocks are named: Bering‐Chukchi‐Beaufort (BCB); East Canada‐West Greenland (ECWG), which was previously split into the Hudson Bay and David Strait stocks (Rugh et al. 2003); East Greenland‐Svalbard‐Barents Sea (EGSB); and the Okhotsk Sea (OKH) (Suydam et al. 2021). Bowhead whales undertake seasonal movements within these stock boundaries, and dispersal events between different stocks are rarely detected (Heide‐Jørgensen et al. 2012). Whether connectivity between stocks has occurred regularly in the past and will persist in the future remains unknown. Genetic population structure among stocks of bowhead whales. (A) Map of sample localities (white symbols; triangles indicate the 24 samples new to this study and squares denote existing samples) and the four management stocks that bowhead whales regularly travel within throughout the year (coloured shading); four of the 24 samples new to this study were omitted from downstream nuclear genomic analyses due to either low nuclear coverage (< 0.005×) or high relatedness. (B) Median‐joining haplotype network constructed in PopART, based on the 32 haplotypes present among 48 mitochondrial genomes. Haplotypes are represented by circles, the relative size of which are proportional to the number of individuals. Colours represent stock of origin. Missing haplotypes are indicated by black dots. Hashes indicate the number of substitutions among haplotypes, denoted by a number for > 5. (C) Nuclear genomic Principal Components Analysis (PCA) created from pseudohaploid base calls. Percentages of variation explained by PC1 and PC2 are indicated in parentheses. (D) Nuclear genomic admixture proportions created using genotype likelihoods and number of populations set at K = 2. (E) Nuclear genomic F ST values between individuals estimated from pseudohaploid base calls, grouped into the four bowhead whale management stocks. Patterns of genetic structure and diversity across the circumpolar range of bowhead whales are not fully understood. Previous studies have used genetic markers with limited power to discern structure, or have had a regional focus (Baird and Bickham 2021). Maternally inherited short mtDNA regions have been used to uncover population size changes over time and a lack of local population structure (Borge et al. 2007; Phillips et al. 2012; Rooney et al. 2001). Microsatellite studies found some evidence of population structure within the BCB stock (Jorde et al. 2007), but this was later refuted (Givens et al. 2010) due to inherent challenges with reproducibility and reliability of microsatellite data. More recently, whole‐genome sequencing has led to more robust and reproducible stock‐specific demographic analyses (Cerca et al. 2022; de Greef et al. 2024; Keane et al. 2015; Westbury et al. 2026). However, these efforts have focused on individual stocks or specific geographic regions, leaving significant gaps in our understanding of genetic structure from a range‐wide perspective. Comprehensive genomic analyses across the full distribution range of bowhead whales are necessary to provide a holistic view of their genetic diversity, population structure and connectivity, and demographic history. Past environmental factors influencing bowhead whale distribution have been explored through ecological niche models, offering reconstructions of past and future habitat suitability (Chambault et al. 2022; Foote et al. 2013; Freymueller, Lorenzen, et al. 2025). These studies have primarily focused on quantifying changes in suitable habitable area across the Holocene and forecasting potential changes in the coming decades. However, they are yet to be used to explore the spatial structure of suitable habitat patches through time, or to quantify genetic connectivity patterns between different bowhead whale stocks. Doing so could provide important insights into the role of climate and environment in shaping current genetic structure and diversity. Here, we explore determinants of range‐wide patterns of contemporary genomic population structure and diversity in bowhead whales, and of their past demographic history, to establish an ecological and evolutionary baseline for understanding the threats from near‐future climatic change and increased human activity in the Arctic. We analyse nuclear genomes from 39 contemporary bowhead whale individuals and 48 mitochondrial genomes sampled across the four management stocks (Figure 1A). We evaluate whether changing environments shaped bowhead whale phylogeography and range‐wide population structure by interpreting our genomic findings alongside estimates of inter‐stock connectivity across the Holocene, derived from bowhead whale ecological niche modelling (Freymueller, Lorenzen, et al. 2025) and movement data (Chambault et al. 2022; Ferguson, Clarke, et al. 2021; Heide‐Jørgensen and Teilmann 2025; Kovacs et al. 2020). Materials and Methods Genetic Data Samples We used samples from contemporary bowhead whale individuals collected across the distribution range of the species. Our samples represent all four recognised management stocks (Figure 1A, Table S1): East Greenland‐Svalbard‐Barents Sea (EGSB); East Canada‐West Greenland (ECWG); Bering‐Chukchi‐Beaufort (BCB); and the Okhotsk Sea (OKH) (Suydam et al. 2021). In total, 24 samples were new to this study: ECWG = 7; BCB = 7; and OKH = 10 (Table S1). Samples from the ECWG (Central Canada) and BCB stocks were obtained from carcasses through Inuit subsistence harvests or strandings as part of long‐term community‐based monitoring programmes coordinated by Fisheries and Oceans Canada and local Hunters and Trappers Organisations (Harwood et al. 2020). All tissue samples were collected by Inuit beneficiaries exercising subsistence hunting rights, and no additional animal care approval was required (Wenzel et al. 2016). Sample DL7641 from the ECWG (West Greenland) was obtained during a harvest, whereas the remaining ECWG (West Greenland) samples were obtained from skin biopsies of free‐ranging whales using a minimally invasive biopsy approach. The OKH samples were obtained as skin biopsies from free‐ranging whales and imported with CITES Permit №21RU000001 (issued 11.01.2021). All non‐OKH samples were imported under CITES exemption DK014 to the Globe Institute, University of Copenhagen. We sourced published bowhead whale genetic data from various publications (Cerca et al. 2022; Keane et al. 2015; Nyhus et al. 2016; Sasaki et al. 2005; Westbury et al. 2026). The EGSB stock samples were originally collected as skin biopsies from free‐ranging whales under appropriate national permits, as described in the original publications (Cerca et al. 2022; Nyhus et al. 2016). Despite the availability of sequencing data for additional EGSB individuals B, C, D, E, and F, we excluded them because they might be duplicate samples of the individual ‘A’ (Cerca et al. 2022; Nyhus et al. 2016). We analysed nuclear genomes from 43 individuals, including the 24 individuals new to this study (Table S1). We removed two individuals with very low nuclear genomic coverage (< 0.005×) from the new samples, both from BCB, and two individuals that we identified in close pairs of relatives, one from ECWG and one from OKH (see further details in Data Processing and Relatedness sections below). We thus based our population analyses on the remaining 39 nuclear genomes: OKH = 9; BCB = 5; ECWG = 13; EGSB = 12. We also analysed 48 mitochondrial genomes (mitogenomes), of which 43 overlapped with the nuclear samples (Table S1); the four individuals removed from the nuclear analysis were included here: OKH = 11; BCB = 7; ECWG = 15; EGSB = 15. Based on our initial analyses of nuclear population structure, which suggested that the BCB and ECWG individuals represent a single population, we used high coverage (> 20×) representative individuals from each genetic population for individual‐based analyses of diversity, demography, and genetic load (Table S1). Two individuals with newly generated high‐coverage data were from Igloolik, Central Canada (CGG_1_024320) and the Okhotsk Sea (CGG_1_024474), and were used to represent the BCB/ECWG and OKH populations, respectively. We used published data from individual 17–19, sampled in the Greenland Sea between Greenland and Svalbard, to represent the EGSB population (Cerca et al. 2022). Data Generation For the 24 individuals new to this study, we extracted DNA from tissue samples using a DNeasy blood and tissue kit following the manufacturer's protocol. We fragmented the extracted DNA from the new ECWG and BCB samples to ~450 bp using a Covaris M220 ultrasonicator. We built Illumina sequencing libraries from the sonicated DNA using the BEST protocol (Carøe et al. 2018), using Illumina adapter mix concentration of 20uM and 15 cycles during the indexing PCR step. We cleaned the indexed libraries using a SPRI bead DNA purification method. For the OKH samples, the extracted DNA was shipped to Novogene (UK), fragmented, and built into Illumina sequencing libraries. All samples were sequenced on an Illumina Novaseq 6000 at Novogene using 150 bp PE reads. DNA previously extracted for individual CGG_1_24320 (Westbury et al. 2026) was sent to StarSEQ, Germany, to have additional Illumina sequencing libraries constructed and sequencing performed to ~25× coverage using an Illumina Nextseq 2000. Data Processing All sequencing data, whether generated within this study, or previously, was subject to the same data processing steps specified in this section. We trimmed adapter sequences and removed reads shorter than 30 bp from the raw reads using skewer v.0.2.2 (Jiang et al. 2014). We merged overlapping paired‐end reads using FLASHv1.2v11 (Magoč and Salzberg 2011), using the default parameters. We mapped the merged and unmerged reads to the bowhead whale reference genome (Keane et al. 2015) including the mitochondrial genome (Genbank accession: KY026773.1) using Burrows‐wheeler‐aligner (BWA) v0.7.15 (Li and Durbin 2009), the mem algorithm, and default parameters. We parsed the output and removed duplicates and reads of mapping quality < 30 using SAMtools v.1.6 (Li et al. 2009). Mapping to the reference bowhead whale nuclear genome resulted in genome‐wide coverages between 0.001× and 26.55× (median = 3.27×, mean = 6.44×, Table S1). Two individuals, CGG_1_024323 and CGG_1_024325, both from the BCB stock, had very low nuclear genome coverage (< 0.005×) and were therefore excluded from downstream nuclear genomic analyses. Mapping to the reference bowhead whale mitochondrial genome resulted in mitochondria‐wide coverages between 97.3× and 7920.9× (median = 1631.3×, mean = 2859.0×). Sex Scaffold Identification We identified putative sex chromosome scaffolds in the bowhead whale reference genome by aligning it to the Cow X (Genbank accession: CM008168.2) and Human Y (GenBank accession: NC_000024.10) chromosomes. We performed the alignments using Satsuma synteny v2.1 (Grabherr et al. 2010) with default parameters. Relatedness We assessed the nuclear genomic relatedness of all 41 individuals with sufficient coverage (> 0.2×) using NGSrelate v2 (Hanghøj et al. 2019). As input, we calculated genotype likelihoods using ANGSD v0.921 (Korneliussen et al. 2014). We obtained genotype likelihoods using the GATK algorithm (‐GL 2), specified the output as a binary beagle file (‐doGlf 3), and applied the following filters: only include reads with a mapping and base quality greater than 20 (‐minmapQ 20 ‐minQ 20); only include reads that map to one location uniquely (‐uniqueonly 1); a minimum minor allele frequency of 0.05 or greater (‐minmaf 0.05); only call a SNP if the p‐value is less than 1e−6 (‐SNP_pval 1e‐6); infer major and minor alleles from genotype likelihoods (‐doMajorMinor 1); skip triallelic sites (‐skipTriallelic 1); remove sex scaffolds and scaffolds shorter than 100 kb (‐rf); and call allele frequencies based on a fixed major and an unknown minor allele (‐doMaf 2). We determined a relatedness coefficient (RAB) > 0.25, equivalent to at least first cousins, as closely related and we removed whichever individual had the lowest coverage from downstream nuclear genomic analyses. Individuals DL7642 and CGG_1_024313 (both from ECWG) had a RAB relatedness score of 0.44, and individuals CGG_1_024478 and CGG_1_024482 (both from OKH) had a RAB relatedness score of 0.26. We therefore removed the individuals with the lowest coverage (CGG_1_024313 and CGG_1_024482; 1.3× and 3.3×) from each pair for further analyses. We also obtained inbreeding coefficients (F) for each individual from the NGSrelate output. Population Structure We used both pseudohaploid consensus base calls and genotype likelihoods for the PCA. We generated the pseudohaploid identity‐by‐state base calls and genotype likelihoods in ANGSD (Korneliussen et al. 2014) using the following filters and parameters: ‐uniqueOnly 1, ‐GL 2, ‐remove_bads 1, ‐minMapQ 20, ‐minQ 20, ‐SNP_pval 1e‐6, ‐skipTriallelic 1, ‐doGlf 2, ‐domajorminor 1, ‐minmaf 0.05, minimum number of individuals required to consider the site in the analysis as 20 (‐minind 20), compute pseudohaploid consensus base calls (‐doIBS 2), only consider a site if the minor allele is present in at least two individuals (‐minminor 2), and output a covariance matrix from the identity‐by‐state base calls (‐doCov). Additionally, we constructed a covariance matrix from genotype likelihoods using PCAngsd v0.98 (Meisner and Albrechtsen 2018). We determined the significance of PCA axes using a Tracy‐Widom test on the eigenvalues in R (Patterson et al. 2006; Wang et al. 2020) based on a significance value of p < 0.01. To build a neighbour‐joining tree, we generated a pairwise distance matrix from ANGSD when performing the pseudohaploid base calls (setting the parameter ‐makematrix 1). We converted the distance matrix into a neighbour‐joining phylogenetic tree using fastME v2.1.6.1 (Lefort et al. 2015) and default parameters. We calculated admixture proportions using the same genotype likelihoods calculated for the PCA with NGSadmix (Skotte 2013). We ran NGSadmix specifying K = 2, K = 3, and K = 4. To evaluate the reliability of the NGSadmix results, we independently ran each K up to 50 times. We considered K values reliable if we retrieved consistent log‐likelihoods (within a value of 3) from at least two independent runs. For the fixation index (F ST), we ran ANGSD with the same filtering parameters, but with the addition of calling consensus pseudohaploid bases (‐dohaplocall 2). We calculated F ST from the pseudohaploid base call file in 100‐kb, non‐overlapping sliding windows, with a minimum requirement of 100 sites per window using the available popgenWindows.py (https://github.com/simonhmartin/genomics_general). To investigate the impact of differences in coverage between our sampleset, we repeated the analyses but with the 12 EGSB individuals downsampled to ~2× using SAMtools. We used D‐statistics (ABBA‐BABA) to investigate gene flow between non‐sister populations (Durand et al. 2011). We calculated D‐statistics for all individuals using a random base call approach in ANGSD (‐doabbababa 1), specifying only autosomal scaffolds > 100 kb with the following parameters: ‐minmapQ 20, ‐minQ 20, ‐blocksize 1,000,000, ‐uniqueonly 1. To infer the ancestral sequence (‐anc) we used a right whale ( Eubalaena glacialis ) individual (BioSample: SAMN32746534), which we mapped to the bowhead whale reference genome using the same approach as for bowhead whales. We summarised the results using a block jackknife approach with the R script available in the ANGSD toolsuite (ANGSD_jackknife.R). We subsequently extracted only comparisons which fit the known population structure of our samples, that is, ((ECWG, BCB), EGSB), ((BCB, ECWG), OKH), ((EGSB, BCB), OKH), and ((EGSB, ECWG), OKH). This allowed us to investigate whether there was relatively more gene flow between OKH and any of the remaining management stocks as well as between EGSB and either ECWG or BCB. We considered all Z‐scores > |3| to be significant. Genetic Diversity We calculated autosomal heterozygosity in one high‐coverage (> 20×) individual per population using ANGSD: 17–19 (20.94×, EGSB), CGG_1_024320 (26.55×, ECWG/BCB), CGG_1_024474 (20.30×, OKH). To ensure comparability with previous studies, we followed the filtering parameters of Cerca et al. (2022). These filters included: use the SAMtools algorithm to calculate genotype likelihoods (‐GL 1); the minimum read depth for a site to be 5 (‐setMinDepthInd 5); ‐minmapq 25, ‐minq 25; ‐uniqueonly 1; ‐remove_bads 1; calculate sample allele frequencies (‐doSaf 1); use the reference genome as the ancestral sequence (‐anc); only consider autosomal scaffolds > 100 kb (‐rf); adjust quality scores around indels (‐baq 1); and fold the site frequency spectrum (‐fold 1). We called the folded SFS from the allele counts using realSFS, part of the ANGSD toolsuite, in 100 kb non‐overlapping sliding windows. We generated the 100‐kb sliding‐windows bed file using bedtools (Quinlan 2014). We obtained mean nuclear genome‐wide heterozygosity values from 25 other mammalian species for comparison from Cerca et al. (2022). We calculated population levels of nucleotide diversity from the pseudohaploid base call file generated for the F ST analysis in 100‐kb, non‐overlapping sliding windows, with a minimum requirement of 100 sites per window using the available popgenWindows.py (https://github.com/simonhmartin/genomics_general). Based on the population structure results, we pooled individuals into three populations, EGSB, OKH, and ECWG/BCB. To correct for differences in coverage, we repeated this analysis with the 12 EGSB individuals downsampled to 2×, and calculated mean differences between the original and downsampled versions. We then used that to correct for the nucleotide diversity in the remaining stocks due to lower‐coverage individuals. To investigate inbreeding we used ROHan (Renaud et al. 2019) to calculate runs of homozygosity (ROH). We ran ROHan using default parameters, but specified a window as being a ROH if it had an average heterozygosity of < 5 × 10−5, as opposed to 1 × 10−5, and limited the analysis to only autosomal scaffolds > 1 Mb. Genetic Load We calculated the genetic load of the three high‐coverage individuals using SnpEff v5.1 (Cingolani et al. 2012). As input, we created a VCF file containing only variant positions using BCFtools v1.15 (Narasimhan et al. 2016) for the autosomes > 100 kb and with the minimum mapping and base quality filters set to 20 (‐q 20 ‐Q 20). This VCF also included the Eubalaena glacialis individual mapped to the bowhead whale reference genome, which we also used in the D‐statistics analysis. We built a SnpEff database using the basic config file, but with the addition of the bowhead whale reference genome and its annotations using the build function. We filtered the annotated output VCF based on three groupings: loss of function (LoF), nonsynonymous mutation, and synonymous mutation. We designated SNPs as LoF if the site was annotated as “stop_gained”, “frameshift_variant”, “splice_acceptor_variant”, or “splice_donor_variant”. We designated SNPs as nonsynonymous if a site was annotated as “missense_variant”, and synonymous if a site was annotated as either “synonymous_variant”, “start_retained” or “stop_retained_variant”. We further filtered for sites where the alternative allele was designated as a LoF or nonsynonymous variant and Eubalaena glacialis was homozygous for the reference allele. If a site was homozygous, we only considered this of interest if the bowhead individual had the derived allele—the Eubalaena glacialis allele was considered the ancestral allele. To normalise our values, we divided the number of LoF or nonsynonymous sites by the number of synonymous variants relative to the reference allele. We considered a heterozygous synonymous variant as a single change and a homozygous synonymous variant as two changes. Demographic History We ran demographic analyses on diploid consensus genomes from the three high‐coverage individuals using a pairwise sequential markovian coalescence model PSMC (Li and Durbin 2011). We called diploid genome sequences using SAMtools and BCFtools, specifying a minimum quality score of 20, minimum coverage of 10, and specified only to include autosomes > 100 kb in length. We ran PSMC specifying standard atomic intervals (4 + 25*2 + 4 + 6), a maximum number of iterations of 25 (‐N), maximum 2 N0 coalescent time of 15 (−t), and initial theta/rho ratio of 5 (−r), and performed 50 bootstrap replicates. We checked PSMC results for overfitting so that after 20 iterations, at least 10 recombinations were inferred to have occurred in the intervals each parameter spans. We plotted the output using a bowhead whale mutation rate of 1.2 × 10−8 per generation (Suárez‐Menéndez et al. 2023), assuming a generation time of 35 years (Rooney et al. 2001). Mitochondrial Genomes We built consensus sequence mitochondrial genomes using a majority rules approach (‐doFasta 2) in ANGSD only considering bases with quality scores > 30 (‐minq 30), reads with mapping quality scores > 30 (‐minmapq 30), and sites with at least 10× coverage (‐minInddepth 10). We aligned the mitochondrial genomes using Mafft v7.310 (Katoh and Standley 2013) and default parameters. This resulted in an alignment of 16,389 bp. We constructed an unrooted haplotype network for the 48 complete mitochondrial genomes of the bowhead whale using the Median‐joining network (Bandelt et al. 2000) as implemented in PopART (Leigh and Bryant 2015). We performed stock diversity estimates with DnaSP v.6.12.03 (Rozas et al. 2017). Diversity statistics included the number of haplotypes (h), the number of segregating sites (S), the haplotype (HD) and nucleotide (π) diversity (Nei 1987), and Tajima's D (Tajima 1989). Gaps and missing data were considered and excluded only in pairwise comparisons. We inferred the demographic history of bowhead whale mitochondrial genomes by employing the Bayesian skyline plot method (Drummond 2005) implemented in BEAST v.2.6.1 (Bouckaert et al. 2014) using all mitochondrial genomes generated in this study and 107 previously published mitochondrial genomes from radiocarbon dated fossil individuals from the Canadian Arctic Archipelago and Svalbard regions (Westbury et al. 2026). We repeated this analysis twice. One run included all individuals (n = 155). We also ran the analysis without the 11 Okhotsk individuals to avoid the effects of population substructure, as the OKH stock was relatively distinct to the other populations (Heller et al. 2013). As input, we aligned the mitochondrial genomes using Mafft v7.392 (Katoh and Standley 2013) and extracted 38 regions, including protein‐coding regions, rRNAs, tRNAs and control region, based on GenBank annotation coordinates. We combined sequences of these 38 regions into six subsets: (i) first codon position of the protein‐coding regions; (ii) second codon position of the protein‐coding regions; (iii) third codon position of the protein‐coding regions; (iv) tRNAs; (v) rRNAs; and (vi) control region. We identified the best‐fit partitioning scheme and substitution model for the six subsets using Partitionfinder v.2.1.1 (Lanfear et al. 2017). The best partitioning scheme and substitution models, based on the corrected Akaike Information Criterion, were used as input for Beast2. We analysed each partition using unlinked substitution models, but linked genealogy and molecular clocks. We assumed six groups of coalescent intervals and a strict molecular clock with a rate of one unit per substitution per site. We sampled from posterior distributions of parameters using MCMC methods. MCMC sampling consisted of 100 million burn‐in steps followed by 1000 million steps, sampled at every 100,000 steps. We assessed chain mixing and convergence to stationarity using Tracer v.1.7.1 (Rambaut et al. 2018), and required minimum effective sample sizes of 5000 for all parameter estimates. Habitat Connectivity We leveraged summer habitat suitability generated using ecological niche models (ENM) of bowhead whales at 35‐year time steps (corresponding to the average generation time of the species (Pacifici et al. 2013)) throughout the Holocene (Freymueller, Lorenzen, et al. 2025). This ENM was built using occurrence data from fossils, historical whaling records, as well as contemporary sightings and satellite tracking (Freymueller, Lorenzen, et al. 2025). This ENM was projected at a 60 × 60 km pan‐Arctic projection, as this spatial scale allowed for accurate representation of the opening of known channels (e.g., Lancaster Sound, Nares Strait) due to sea‐level rise and ice‐sheet deglaciation (see next paragraph). Occurrence data was matched in time and space to modelled palaeoclimate reconstructions from a HadCM3B‐M2.1 coupled atmosphere–ocean general circulation model (Armstrong et al. 2019). Cross‐validation tests show that the ENM did well at predicting occurrence as a response to climatic and environmental conditions, and also identified known contemporary bowhead whale summer congregation hotspots (Freymueller, Lorenzen, et al. 2025). To account for sea‐level dynamics operating at resolutions finer than our ENM projections, we generated a high‐resolution, temporally changing, land‐sea mask for the Holocene. To do this, we statistically downscaled ICE‐7G_NA deglaciation reconstructions (Roy and Peltier 2018) from a 1° × 1° resolution to a 5 × 5 km resolution using GEBCO 2014 bedrock bathymetry data (https://www.gebco.net/) (Weatherall et al. 2015) using the delta‐ method (Figure S1) (Beyer et al. 2020). Reconstructions from the ICE‐7G_NA model are available at 500‐year time steps for the Holocene, which we bilinearly interpolated to the 35‐year time steps used in the ENM. To identify and mask out potential dispersal corridors that would not have been traversable by bowhead whales due to being glaciated, we used fine‐scale (rasterized to 5 × 5 km) ice sheet margin (i.e., edge) reconstructions for: the Laurentide, Cordilleran, & Innuitian (Dalton et al. 2023); Greenland and Iceland (Dalton et al. 2020), Fennoscandian (Stroeven et al. 2016); and Eurasian (minus the Fennoscandian) ice sheets (Hughes et al. 2016). We treated areas above sea level and areas covered by ice sheets as impenetrable. We then upscaled these fine resolution ice sheet‐land‐sea masks to the ENM resolution where a 60 × 60 km grid cell was classified as land/ice sheets if its centroid was covered by the 5 × 5 km mask (Figure S1 and Video S1). This approach retained the finer resolution dispersal channels without enabling dispersal across impenetrable barriers like isthmuses (e.g., northern Kamchatka). We calculated connectivity between the four bowhead whale stocks throughout the Holocene using the landscapemetrics and terra R packages (Hesselbarth et al. 2019; Hijmans et al. 2024). We did this by first identifying “core patches” of summer habitat (grid cells with all eight surrounding cells also being suitable; Figure S2) and calculating pairwise travel distances between all core patches, accounting for temporally evolving land and ice sheet barriers. We used core habitat patches, rather than single cells, to better capture spatial population dynamics and to exclude potential edge effects (Laurance 2000). We then identified minimum travel distances (inter‐stock travel distance) between core patches of all four bowhead whale stocks (Figure S3). We calculated the following inter‐stock travel distances: (i) OKH ↔ BCB; (ii) BCB ↔ ECWG; (iii) ECWG ↔ EGSB via Lancaster Sound/Nares Strait; (iv) ECWG ↔ EGSB via south of Greenland; (v) BCB ↔ EGSB via the northern edge of the Canadian Arctic Archipelago/Greenland; and (vi) BCB ↔ EGSB via the northern edge of Siberia, see Figure 4. We included core patches outside contemporary management stock boundaries to account for potential dispersal corridors that existed during the Holocene and have since been lost due to climate change and effects of commercial whaling (Baird and Bickham 2021). For ECWG ↔ EGSB and OKH ↔ BCB travel distances, we selected the northern/southern edge of Greenland and the southern edge of Kamchatka, respectively, to demarcate if an area was closer to a given population, as marine species would have to pass through these areas to minimise travel distance between stocks. Thus, these land masses reflect natural biogeographic barriers to marine species. For BCB ↔ ECWG and BCB ↔ EGSB travel distances, given there were no obvious biogeographic barriers or choke‐points that minimised distance between these populations, we defined the halfway distance between contemporary management stock boundaries as a demarcation line for stock affinity. We investigated whether more marginal (low‐quality) habitat substantially shortened inter‐population travel distances throughout the Holocene (Figure S4). We did this by setting increasing thresholds for occurrence in our ENM projections (Peterson et al. 2011). In addition to the 10% occurrence threshold used by (Freymueller, Lorenzen, et al. 2025), we included 12.5% and 15% occurrence thresholds. Setting thresholds > 15% resulted in the near absence of nearly all core patches (and suitable habitat in general) in the Okhotsk Sea throughout the Holocene, which is unlikely, given the region's whaling history and present‐day persistence of bowhead whales. We compared our estimates of inter‐stock travel distances to observed bowhead whale movement distances using available tracking data for 83 individuals, totalling 53,955 geolocations over 4092 days (Heide‐Jørgensen and Teilmann 2025; Chambault et al. 2022; Kovacs et al. 2020). We removed position reports that were classified as poor (Freitas et al. 2008) and calculated cumulative bowhead whale daily travel distance as a function of total track days. We estimated bowhead whale cumulative travel distance to be ~7000 km*yr.−1, meaning that possible seasonal one‐way travel distances are ~3500 km (Figure S4). These distances are slightly inflated due to the inclusion of daily foraging patterns not related to directional movements. Calculating maximum travel distance (around land barriers) between known bowhead summering and wintering grounds revealed similar numbers, with distances of ~2500 km and ~3200 km in the ECWG and BCB stocks respectively (Chambault et al. 2022; Ferguson, Higdon, et al. 2021; Finley 2001). This is also consistent with tracking studies for the BCB stock, where bowhead whales move up to ~3000 km seasonally (Ferguson, Clarke, et al. 2021). Therefore, 3500 km represents a reasonable estimate of the maximum likely seasonal movement for bowhead whales. If the bowhead whale stocks were consistently further apart than this upper observed movement distance, dispersal between stocks would be far less likely, and would lead to genetically distinct populations forming over time due to isolation. Results Genomic Population Structure Regardless of the base‐call approach of the nuclear data (genotype likelihoods or pseudohaploid), principal component analysis (PCA) showed that bowhead whales from OKH separated from the other stocks along the PC1 axis (Figures 1D and S5). When using pseudohaploids, EGSB individuals separated from the BCB and ECWG along the PC2 axis, regardless of whether EGSB individuals were downsampled to 2× or not (Figures 1D and S6A). When using genotype likelihoods, the EGSB individuals were widely spread along PC2 (Figure S5); however, when downsampling these EGSB individuals to 2× and using genotype likelihoods, they formed a more distinct cluster (Figure S6B). Regardless of the choice of method, individuals from the BCB and ECWG overlapped considerably, and likely comprise a single cluster. We therefore pooled individuals from these two stocks together for downstream analyses and henceforth refer to them as the ‘BCB/ECWG population’. A Tracy‐Widom test on the eigenvalues indicated that only PC1 explained significantly more variance than expected under a null model of random variation (p < 0.01), whereas subsequent axes likely reflect weaker or non‐significant structure, regardless of the base‐call approach or whether EGSB individuals were downsampled to 2×. The neighbour‐joining tree presented three main clades, with OKH individuals and EGSB individuals each forming their own clades, and a clade with BCB and ECWG mixed throughout (Figure S7). None of the 32 mitochondrial haplotypes were shared among individuals across stocks (Figure 1B). However, the haplotype network did not show any clear phylogeographic structuring, with individuals from each stock distributed across the network. In the admixture proportions analysis, only K = 2 populations converged within 50 replicates. OKH contained mostly ancestry from one population, EGSB contained mostly ancestry from the other population, and BCB/ECWG individuals contained mixed ancestry, overlapping ~20% with OKH individuals and ~80% with EGSB individuals (Figure 1C). Although they did not converge, we also investigated the replicates with the highest likelihoods for K = 3 and K = 4 (Figure S8). For K = 3, OKH individuals contained mostly unique ancestry, BCB/ECWG individuals contained mostly unique ancestry from a second population, and EGSB contained mixed ancestry between the second population (predominant in BCB/ECWG) and its own unique ancestry. For K = 4, OKH individuals contained mostly unique ancestry, BCB/ECWG individuals contained mostly unique ancestry from a second population, and EGSB contained some ancestry predominantly seen among whales from BCB/ECWG, and two more unique ancestry components. The F ST values show a similar pattern to the PCA, with OKH being the most differentiated (Figure 1E). When treating all four management stocks as separate populations, only the comparison between individuals from the BCB and ECWG stocks yielded a median FST < 0 (−0.001609) and a corresponding mean FST of 0.000787, indicating no detectable nuclear genomic differentiation between these two stocks in our analysis. These findings were consistent when downsampling EGSB individuals to 2× (Figure S9). In our mitochondrial analysis, the majority of F ST comparisons were non‐significant (Table S2). Only OKH vs. EGSB was significant at a p‐value threshold of 0.01. Our D‐statistics results were mostly non‐significant (Z < |3|) (Figure S10). The D‐scores from the topology ((EGSB, ECWG), OKH) deviated most from 0. Many were negative, consistent with the assumption of some ancestral gene flow between EGSB and OKH relative to ECWG. However, there were also positive values indicating the opposite pattern. A similar pattern was observed for the topology ((BCB, ECWG), OKH). The topology ((EGSB, BCB), OKH) returned almost exclusively non‐significant values. The topology ((ECWG, BCB), EGSB) returned some significant positive and negative values. Taken together, the absence of consistent directionality across topologies suggests that these deviations are more likely driven by stochastic effects, including differences in effective population size, rather than by sustained introgression (Zheng and Janke 2018). Genomic Diversity Based on the three high‐coverage nuclear genomes chosen as representative of each of the three genetic clusters identified in the structure analyses, we find the lowest mean genome‐wide heterozygosity in the OKH individual (0.000979), followed by the EGSB individual (0.001067) (Figure 2A). The ECWG individual had the highest heterozygosity (0.001226). Visualising the distribution of heterozygosity across the genome from the ANGSD output showed relatively similar distributions across individuals, apart from the OKH individual (CGG_1_024474), with relatively high counts of windows showing low‐to‐no levels of heterozygosity. These genome‐wide heterozygosity values are high relative to other mammalian species (Figure 2B). Genetic diversity of the three inferred bowhead whale populations. (A) Nuclear genome‐wide heterozygosity estimated for our three representative individuals, broken down into 100‐kb windows; an individual from ECWG was used to represent the inferred combined BCB/ECWG population. Dotted lines indicate the mean values. (B) Mean genome‐wide heterozygosity of the three bowhead whale individuals and 25 other mammal species, taken from Cerca et al. (2022). Triangles indicate cetacean species. (C) Population‐level nuclear genome‐wide nucleotide diversity calculated from pseudohaploid base calls in 100‐kb windows. Here, individuals from the BCB and ECWG stocks were pooled and treated as one population (shown in grey) based on the structure analyses (Figure 1C,D). OKH and EGSB were considered as separate populations. Sample size is indicated at the top of the range CV lines/bars. When correcting for differences in sequencing coverage of the population‐level dataset, OKH again had the lowest mean estimate of nuclear genomic nucleotide diversity (0.295). Similar levels of diversity were found in EGSB (0.310) and ECWG/BCB (0.308) (Figure 2C). In the 48 mitochondrial genomes analysed, we found high levels of haplotype and nucleotide diversity across the four stocks (Figure 1B, Table S3). The OKH individuals showed the lowest haplotype and nucleotide diversity. ECWG had relatively low nucleotide diversity, but high haplotype diversity. Estimates of inbreeding using ROHan only found runs of homozygosity > 1 Mb in the OKH individual, constituting 2.47% (1.65%–2.73%) of the genome investigated and a mean length of 1,833,330 bp. No ROH > 1 Mb were found in the other two individuals. The inbreeding coefficients from NGSrelate also did not show signs that any individual was inbred, with all F values being < 0.00012. When considering genetic load due to both missense and/or loss‐of‐function mutations, EGSB had the highest masked genetic load, and ECWG had the lowest (Table S4). The highest realised load based on missense or loss of function mutations was found in OKH. Demographic History To investigate deep‐time demographic trends of the different bowhead whale populations, we implemented PSMC on the nuclear genomes of the three high‐coverage individuals. The three bowhead whale genomes yielded identical demographic trajectories of effective population size (Ne) until ~200 thousand years ago (kya), when the Okhotsk genome diverged from the other two (Figure 3A). At this time, the Okhotsk individual's effective population size continues its trajectory of increasing Ne, whereas the Canadian and Svalbard genomes plateau. At ~120 kya, the two latter individuals decrease in Ne until ~50 kya, after which they increase sharply. Conversely, the Okhotsk Ne decreases sharply at ~50 kya. We note that PSMC‐inferred changes in Ne may partly reflect shifts in population structure and connectivity, such that changes in coalescent rates driven by population divergence or gene flow can generate apparent fluctuations in Ne rather than true changes in population size (Morin et al. 2018). Bowhead whale demographic reconstruction. (A) Deep‐time demographic history calculated from our three representative individuals, using PSMC; an individual from ECWG was used to represent the inferred BCB/ECWG population. Faded lines indicate bootstrap replicates. (B) More recent female demographic history based on a Bayesian Skyline analysis of the mitochondrial genomes of 37 contemporary bowhead whale individuals, combined with 105 Holocene and 2 Late Pleistocene mitochondrial genomes from Westbury et al. (2026). The 11 Okhotsk individuals were omitted from this analysis, due to their unique demographic history inferred from the PSMC; the Skyline analysis with the full dataset is provided in Figure S11. The bold line indicates the median value and the shaded area shows the 95% credibility intervals. Investigations into the more recent maternal demographic history using mitochondrial Bayesian skyline plots indicate an increase in female Ne between 20 and 15 kya, followed by relative stability until ~2.5 kya, when it begins to decrease until the present (Figure 3B). This recent decrease is sharper when including OKH individuals in the skyline analysis (Figure S11), but may be due to population structure in the dataset (Heller et al. 2013). Habitat Connectivity Travel distances required to maintain connectivity between the BCB ↔ OKH populations regularly exceeded 4500 km for the past 11,700 years (Figure 4B). Given that these distances far exceed the upper observed seasonal movement distances for bowhead whales (~3500 km (Chambault et al. 2022; Ferguson, Higdon, et al. 2021; Finley 2001)), our habitat connectivity modelling suggests that the Okhotsk Sea bowhead whale population has been isolated throughout the Holocene. Spatial connectivity of bowhead whale populations across the Holocene. (A) Map of the approximate distributions of the four bowhead whale stocks. Arrows reflect potential dispersal pathways among stocks. (B) Minimum inter‐stock travel distances during the Holocene. Thick lines denote 75% loess curves of the time series. The different colours show inter‐stock travel distances under sensitivity thresholds that exclude marginal bowhead whale habitat (blue = excluding lowest 10%; black = excluding lowest 12.5%; red = excluding lowest 15%). Dashed horizontal black lines represent the 3500 km upper likely seasonal movement distances for bowhead whales. In contrast, the closest core summer habitat patches between the BCB ↔ ECWG populations were rarely greater than ~2000 km once Lancaster Sound in the Canadian Arctic Archipelago opened ~10.7 kya (Roy and Peltier 2018) (Figure 4). Thus, it can be expected that movement between these two populations was and likely is relatively frequent. From ~9 kya, when the Nares Strait that connects the Atlantic (northern Baffin Bay) and the Arctic (Lincoln Sea) Oceans opened (Georgiadis et al. 2020; Jennings et al. 2011), the nearest distance between the ECWG ↔ EGSB populations shortened to 3000–4000 km in length. This distance approximated the upper 3500 km maximum likely seasonal movement for bowhead whales, starting in the Holocene Thermal Maximum (10–5.5 kya (Axford et al. 2021)) (Figure 4). Distances between ECWG ↔ EGSB populations via a southern Greenland route mostly exceeded 3500 km, only dipping below 3500 km around ~11 kya and again ~5 kya. Distances between BCB ↔ EGSB habitat patches via Siberia were mostly < 1000 km throughout the Holocene. In contrast, BCB ↔ EGSB travel distances north of the Canadian Arctic Archipelago exceeded the 3500 km maximum likely seasonal movement for bowhead whales. Sensitivity analyses show that our estimates of inter‐population connectivity are insensitive to thresholds used in the ecological niche modelling. We show that the exclusion of increasingly marginal habitats did not markedly increase travel distances between stocks (Figure 4B). Furthermore, analysing spatial habitat suitability patterns through time showed that BCB ↔ ECWG and BCB ↔ EGSB travel distances were not sensitive to the exact location of population demarcation edges (Videos S2–, S4). Discussion Understanding how past environmental and anthropogenic factors shaped contemporary genetic patterns is essential for assessing the vulnerability of species in the face of ongoing climate change (Razgour et al. 2019). By integrating range‐wide genomics with ecological modelling, we assessed how environmental change and habitat connectivity during the Holocene have shaped the genomic population structure and diversity of bowhead whales today. We identified three genetically differentiated bowhead whale populations, each showing little‐to‐no signs of recent inbreeding and high heterozygosity values relative to other mammalian species. Population Structure and Holocene Habitat Connectivity Our nuclear genomic data indicate three genetically distinct populations, in contrast to the four management stocks currently recognised by the International Whaling Commission (Baird and Bickham 2021). We find the Okhotsk Sea (OKH) bowhead whales are the most genetically distinct on a nuclear genomic level. Based on our demographic analysis, the OKH population began a unique trajectory, diverging from the others ~200 kya. The relative isolation of the OKH bowhead whales is consistent with required long travel distances to the Bering‐Chukchi‐Beaufort (BCB) population (and all other populations via BCB), exceeding bowhead whale movement capacity of ~3500 km, at least for the past 11,700 years. Thus, our findings indicate long‐term geographic isolation has been a major driver of the pronounced genetic distinctiveness of the OKH population. However, the absence of pronounced phylogeographic structuring among the mitochondrial genomes suggests that some level of connectivity via gene flow with the other populations cannot be excluded. In contrast, our nuclear genomic analyses reveal no strong genetic differentiation between individuals from the BCB and East Canada–West Greenland (ECWG) stocks. We note that the number (n = 5) and spatial sampling distribution (West Canada only) of BCB individuals analysed is limited, which may reduce power to detect subtle structure both across the geographic range of the BCB stock and between BCB and ECWG. We therefore interpret the BCB–ECWG overlap as evidence for no strong differentiation in our dataset. Despite this, the observed pattern is consistent with previous genetic studies reporting limited or context‐dependent differentiation between these stocks (Baird et al. 2018). Genomic similarity between BCB and ECWG is consistent with high habitat connectivity and short (≤ 2000 km) inter‐population travel distances throughout the Holocene. It is also supported by satellite telemetry, which has demonstrated spatial overlap of two individuals from the BCB and ECWG stocks (Heide‐Jørgensen et al. 2012). Although this exchange was attributed to the recent and dramatic loss of sea ice in the Canadian Arctic, our results suggest connectivity between these regions may actually be a longer‐term phenomenon, occurring at least since the opening of Lancaster Sound in the Canadian Arctic Archipelago ~10.7 kya (Roy and Peltier 2018). Individuals from the East Greenland–Svalbard–Barents Sea (EGSB) are differentiated from the BCB‐ECWG group, but only to a small extent. EGSB forms a distinct cluster along PC2 of the PCA and in the neighbour‐joining phylogenomic analysis, with slightly elevated F ST values relative to the comparisons between BCB and ECWG. However, a close affinity of EGSB with BCB and ECWG whales is supported by the admixture analysis, which shows all BCB‐ECWG individuals share a large ancestry proportion (~80%) with EGSB, and that of three major mitochondrial haplogroups, one is shared only between EGSB and ECWG (Figure 1). Time series palaeogenomic analyses of bowhead whale fossils indicate that EGSB and ECWG populations were panmictic across the past 12,000 years—their differentiation only happened during the past 500 years and is likely a direct consequence of commercial whaling (Westbury et al. 2026). Long‐term connectivity between ECWG and EGSB is supported by our deep‐time demographic reconstruction that shows nearly identical demographic trajectories until approximately 30 kya, which corresponds to the temporal limit of the PSMC method (Li and Durbin 2011). Genomic connectivity could have been maintained by very low but continuous levels of gene flow—fewer than five migrants per generation (Westbury et al. 2026). Rare long‐distance dispersal is consistent with our habitat connectivity analysis, which indicates that the distance between ECWG and EGSB during the Holocene was close to the maximum likely seasonal movement for bowhead whales (3500 km). In summary, the genetic differentiation of EGSB is a recent phenomenon, driven by commercial whaling and not by long‐term climatic or environmental drivers. Our analyses could not identify whether north or south of Greenland has been a more likely dispersal route between bowhead whales from ECWG and EGSB in the past. While bowhead whale fossils dating to the Middle and Late Holocene have been recovered from the southernmost part of the Nares Strait and northeast of the Lincoln Sea further north, they are otherwise rare in northern Greenland (Figure S12; Freymueller, Dyke, et al. 2025). This is despite the discovery of fossils from other Arctic marine mammals having been found in the Nares Strait and Lincoln Sea, indicating that poor preservation conditions are unlikely to explain the absence of recovered bowhead whale fossils in this region. In contrast, southeast Greenland shows a general scarcity of fossils from all Arctic marine mammal species (Freymueller, Dyke, et al. 2025) which may reflect the acidic nature of regional substrates that likely hinders the preservation of organic material (Westbury et al. 2026). Our habitat connectivity analyses identified that direct dispersal between the BCB ↔ EGSB populations via Siberian waters during the Holocene was plausible, in agreement with the relatively weak genomic differentiation between these stocks. Although the continental shelf along Siberia contained suitable habitat for bowhead whales throughout the Holocene, travel distance between the current designated stock boundaries would be ~3500 km one‐way. With the exception of a single Late Holocene bowhead whale fossil from the East Siberian Sea (Boeskorov and Davydov 2015), no confirmed bowhead whale fossils have (to our knowledge) been recovered between Franz Josef Land Archipelago and the Bering Strait (Dinesman 1999; Forman et al. 1997; Lubinski et al. 1999; Mason 1998). However, Arctic marine mammal fossils are rare in general from the Russian Arctic (Figure S12; Freymueller, Dyke, et al. 2025). Conversely, contemporary bowhead whales have been detected north of Siberia during aerial surveys (Belikov and Boltunov 2002), although such sightings are limited (Boeskorov and Davydov 2015). These observations support the potential for historical connectivity between the BCB and EGSB populations, consistent with our habitat analysis and their relatively high genomic affinity. Population Diversity and Genomic Health Bowhead whales were nearly driven to extinction during four centuries of commercial whaling (Thewissen and George 2021). Despite a long history of over exploitation, the BCB and ECWG management stocks are now relatively abundant, with present‐day population sizes estimated at ~16,000 (Givens and Heide‐Jørgensen 2021) and ~6000 individuals (Givens and Heide‐Jørgensen 2021), respectively. Less is known of the current census size of stocks in the EGSB region and the Okhotsk Sea, but estimates number in the low hundreds (Cooke et al. 2017; Givens and Heide‐Jørgensen 2021; Vacquié‐Garcia et al. 2017). Each of the three bowhead whale populations displayed high heterozygosity values relative to other mammalian species. The investigated individuals also had little‐to‐no signs of recent inbreeding (runs of homozygosity > 1 Mb). No recent decline during commercial whaling was apparent in our mitochondrial demographic reconstruction. Large population declines associated with commercial whaling would be expected to lead to relatively low genetic diversity, high inbreeding, and low nuclear effective population size, yet this pattern is not clearly observed. This absence may reflect time lags between demographic and genetic decline (Gargiulo et al. 2024), as has been reported for ECWG and EGSB (Westbury et al. 2026). The OKH population, shaped by long‐term isolation, exhibits both relatively low genetic diversity and high realised genetic load, consistent with the prolonged effects of genetic drift and limited gene flow. The effects of climate change on the population will likely be all the more severe (Freymueller, Lorenzen, et al. 2025), given that the Okhotsk Sea is already ice‐free during the summer today (Walsh et al. 2017). Bowhead whales in this region do not appear to leave the Okhotsk Sea (Ivashchenko and Clapham 2010) and may be at high risk of predation by killer whales ( Orcinus orca ) (Shpak and Paramonov 2018). Future predictions of suitable bowhead whale habitat based on the ecological models used here indicate the complete loss of suitable habitat in the Okhotsk Sea by year 2060 (Freymueller, Lorenzen, et al. 2025). In combination with our findings of lower genetic diversity and higher inbreeding levels, the OKH population likely has reduced capacity to buffer additional environmental deterioration compared with the other populations. Despite having undergone one of the most dramatic reductions in population abundance due to historical whaling (> 90% decline) (Westbury et al. 2026), the EGSB population retains high genetic diversity relative to the other bowhead populations, perhaps due to high pre‐whaling abundances and/or diversity, but also the time lag between a population bottleneck and genomic response (Gargiulo et al. 2024). EGSB is estimated to have numbered up to 52,500 individuals prior to commercial whaling (Allen and Keay 2006), and palaeogenomes have shown pre‐whaling genetic diversity was slightly higher (~2%) than today (Westbury et al. 2026). However, the apparent high present‐day diversity of EGSB hides underlying genetic vulnerabilities. Our results show that EGSB bowhead whales harbour the highest masked genetic load relative to the other two bowhead populations, with a substantial proportion of deleterious variants present in the heterozygous state. Bowhead whales in the EGSB stock may become more vulnerable as the genomic signatures of their past demographic collapse become evident. For example, if the population decline leads to increased inbreeding in the future, many of the currently masked deleterious mutations may become homozygous, increasing the realised genetic load and posing a risk to population viability and fitness. Taken together, our various lines of evidence illustrate that while the natural environment may have structured genetic variation in bowhead whales across longer time scales, anthropogenic impacts associated with commercial whaling during the past 500 years shifted populations from historically stable evolutionary trajectories toward increased genetic risk—underscoring the need for timely, population‐specific conservation strategies. Future Outlook Looking to the future, continued climatic change and other anthropogenic threats may also reshape population connectivity. Bowhead whales are now spending winters further north—such as in the southern Chukchi Sea—in response to sea ice decline (Citta et al. 2023; Szesciorka et al. 2024). They are more ice‐associated during the winter than in the summer (Ferguson et al. 2010; Kovacs et al. 2020), and seasonal habitat use varies among populations in response to both sea ice conditions and oceanographic features (Heide‐Jørgensen et al. 2021; Ngo et al. 2024; Nowak et al. 2026). Our connectivity analysis focused on summer connectivity because it is the main foraging period of bowhead whales (Matthews and Ferguson 2015). Continued loss of Arctic sea ice could facilitate new dispersal routes between currently separated populations—for instance, increased movement off the Siberian coast, around northern Greenland, or across the polar basin. While such changes may enable gene flow and buffer small populations against isolation, they may also introduce new threats, including: increased disease risks; increased exposure to shipping traffic (Nisi et al. 2024); increased summer overlap with predatory killer whales (Matthews et al. 2020); and competition with sub‐Arctic cetacean species moving their distributions further north (Heide‐Jørgensen et al. 2023; Kovacs et al. 2011). As Arctic sea ice continues to decline and Atlantification and Pacification of Arctic areas increases, the future dispersal potential and adaptive resilience of bowhead whale populations will depend not only on their environmental responses (and those of their prey), but also on their demographic histories and genomic health. In this context, environmental stewardship must account for both historical and future connectivity, the complex interplay between demography and genetic load, and the risks posed by climate‐driven shifts in Arctic marine ecosystems. By integrating range‐wide population genomics with spatially explicit habitat connectivity modelling, this study not only improves our understanding of bowhead whale population dynamics but also provides a scalable framework for predicting the evolutionary and conservation trajectories of other Arctic marine species facing the dual pressures of historical exploitation and accelerating environmental change. Author Contributions Conceptualisation: Michael V. Westbury, Nicholas A. Freymueller, Damien A. Fordham, Eline D. Lorenzen. Formal analysis: Michael V. Westbury, Nicholas A. Freymueller, Andrea A. Cabrera. Resources: Lutz Bachmann, Steven H. Ferguson, Mads Peter Heide‐Jørgensen, Kit M. Kovacs, Christian Lydersen, Olga Shpak, Øystein Wiig. Writing – original draft: Michael V. Westbury, Nicholas A. Freymueller. Writing – review and editing: All authors. Visualisation: Michael V. Westbury, Nicholas A. Freymueller, Eline D. Lorenzen. Supervision: Damien A. Fordham, Eline D. Lorenzen. Funding Acquisition: Eline D. Lorenzen. Funding This work was supported by Villum Fonden (Grant 37352); Danmarks Frie Forskningsfond (Grant 9064‐00025B); Novo Nordisk Foundation Emerging Investigator (Grant NNF24SA0093839) and The Norwegian Research Council (Grant 244488/E10). Disclosure Benefit‐sharing statement: This research was conducted through an international collaboration involving institutions and researchers from across the circumpolar Arctic and beyond, including Denmark, Greenland, Norway, Canada, Australia, and Ukraine. All collaborators contributing samples, data, and expertise are included as co‐authors on this manuscript. Biological samples were obtained through long‐standing collaborations with Arctic research institutions and, where applicable, through Indigenous subsistence harvests; Indigenous hunters contributed samples essential to this study and are acknowledged in the Methods and Acknowledgements. The findings address a priority conservation concern—understanding population structure, connectivity, and genomic vulnerability of bowhead whales in a rapidly changing Arctic—and are directly relevant to management and conservation frameworks, including those of the International Whaling Commission. Study results have been shared among collaborating institutions and contribute to ongoing Arctic biodiversity research, conservation planning, and institutional capacity building. Conflicts of Interest The authors declare no conflicts of interest. Supporting information Figure S1: Generating a fine‐scale spatiotemporally varying land‐sea mask via statistical downscaling. (a) The contemporary bedrock elevation data of the ICE‐7G_NA deglaciation model (which has been bilinearly interpolated to 5 × 5 km resolution from its native 1° × 1° resolution) is subtracted from the fine‐scale (5 × 5 km) contemporary bedrock elevation GEBCO data to produce a delta layer containing fine‐scale anomalies between the two layers (i.e., an additive delta method). (b) This delta layer is added to the ICE‐7G_NA bedrock layers (which have also been bilinearly interpolated to 5 × 5 km from their native 1° × 1° resolution) for all the Holocene time bins to produce fine‐scale bedrock elevation reconstructions. This assumes that the geometry of fine‐scale structures (e.g., sub‐1° features such as subterranean canyons, small bays/fjords, etc.) are preserved over the course of the Holocene. (c) Areas above sea level and areas covered by ice sheets are then masked out to produce fine‐scale reconstructions of the land‐sea mask. These are then upscaled to the 60 × 60 km resolution used in analyses. Figure S2: Identifying core habitat patches and quantifying inter‐stock travel distances. (A) Thresholded habitat suitability projections from the bowhead whale niche model. Areas on land or covered by ice sheets have been removed according to the spatio‐temporally varying land‐sea mask (e.g., 5845 years BP). (B) Core habitat patches (suitable grid cells surrounded on all sides by other suitable grid cells) are identified. (C) The minimum inter‐stock travel distances are calculated accounting for the spatio‐temporally varying land‐sea mask. This process is then repeated for all other time bins and all other niche model‐threshold values. Figure S3:. Rationale for evaluating the role of marginal habitats in shortening inter‐stock travel distances. (A) Theoretical 2‐dimensional representation of a sea‐scape with marginal‐ (blue) and high quality‐habitat core patches (red). (B) Boxplots of pairwise distances between habitat patches when including vs. omitting marginal habitat. The dashed line represents a theoretical maximum bowhead whale travel distance. In this hypothetical example, marginal habitat patches may not sustain many individuals, but they may serve as valuable stopover locations, markedly lowering travel distances between habitat patches, ultimately maintaining meta‐population connectivity over long time periods. Figure S4:. Seasonal movement patterns of bowhead whales. (A) Cumulative travel distances as a function of the number of days a satellite tag was attached to an individual whale. Only one‐third of all tags (n = 83) were attached longer than 60 days. The black curve represents a loess regression of all tracks, with the 95% confidence interval shown in grey. Colours represent different bowhead whale populations: East Canada‐West Greenland (ECWG) = light blue, n = 68; East Greenland‐Svalbard‐Barents Sea (EGSB) = dark blue, n = 13. We were unable to access raw tracking data for bowhead whales in the Bering‐Chukchi‐Beaufort (BCB) stock. However, their annual movements are known to be on the order of 6000 km, which is indicated by the dashed light green line. No satellite tagging data exists for bowhead whales from the Okhotsk Sea (OKH). (B) Map of daily locations (n = 4092) of the 83 whales tracked. Figure S5: PCA‐based population structure. Nuclear genomic PCA created from using genotype likelihoods. Percentages of variation explained by the PC 1 and PC 2 are indicated in parentheses. OKH: dark green; BCB: light green; ECWG: light blue; EGSB: dark blue. Figure S6: PCA‐based population structure. Nuclear genomic PCA created using (A) pseudohaploid base calls or (B) genotype likelihoods. In both analyses, the EGSB individuals were downsampled to 2× to be more comparable to the other individuals. Percentages of variation explained by the PC 1 and PC 2 are indicated in parentheses. OKH: dark green; BCB: light green; ECWG: light blue; EGSB: dark blue. Figure S7: Tree‐based population structure. The neighbour joining tree was built from a distance matrix computed using pseudohaploid base calls. Colours indicate management stocks. OKH: dark green; BCB: light green; ECWG: light blue; EGSB: dark blue. Scale bar indicates pairwise distance. Figure S8: NGSadmix‐based population structure. Nuclear genomic admixture proportions created using genotype likelihoods and number of populations set at K = 3 and K = 4. These analyses did not converge. Figure S9: Evaluating the influence of sequencing coverage on F ST results. (A) F ST values computed from pseudohaploid base calls with no downsampling of EGSB, also shown in Figure 1E. (B) F ST values computed from pseudohaploid base calls with EGSB downsampled to ~2×. ECWG and BCB were treated as separate populations in these analyses. Figure S10: Patterns of gene flow. D‐statistics results calculated from a random base call approach. Filled circles show significant results (Z < |3|). Negative values show a closer relationship between H3 and H1 relative to the relationship between H3 and H2. Positive results show a closer relationship between H3 and H2 than H1. Figure S11: Demographic reconstruction. Skyline plot of female effective population size built from the full suite of 155 mitochondrial genomes and including the 11 OHK individuals. Also included are the 37 other contemporary bowhead whale individuals, combined with 105 Holocene and 2 Late Pleistocene mitochondrial genomes from Westbury et al. (2026). Figure S12: Fossil record for bowhead whales and other Arctic marine mammals. Location of bowhead whale (n = 1077) and other Arctic marine mammal (n = 2315) fossil remains in the Fossil Arctic Marine Mammal database. Bowhead whale remains are shown in circles with colours denoting sample age. Other Arctic marine mammal species (11 species, including the extinct Stellar's sea cow) are shown as diamonds. Blue areas denote current IUCN stock boundary polygons for bowhead whales. Table S2: Population pairwise mitochondrial F ST analysis. The population F ST values are above the grey area and F ST p‐values below. Asterisked F ST value indicates significance (p < 0.01). Sample sizes are indicated. Table S3: Mitochondrial genome diversity statistics—n = number of samples, bp = base pair length, GM = gaps/missing data, S = number of segregating sites, h = number of haplotypes, HD = haplotype diversity; π = nucleotide diversity. Table S4: Genetic load values of the three inferred bowhead whale populations. Genetic load values, calculated using SNPeff, were normalised by dividing by the number of homozygous synonymous mutations relative to the right whale outgroup. Realised load indicates the variant is in a homozygous state and masked load indicates a heterozygous state. LoF = Loss of function. Table S1: Bowhead whale sample information and genetic mapping statistics. For five individuals, only mitogenomes were available. Individual CGG_1_024320 was first sequenced in Westbury et al. 2026 and more deeply sequenced in this study. ‐ attached as spreadsheet. Video S1: Spatiotemporal land‐sea mask of the Arctic over the Holocene with reconstructed ice sheet extent and thickness. Video S2: Projected summer habitat suitability for bowhead whales using a low ecological niche model threshold of 10%. For details, see Freymueller, Lorenzen et al. (2025) Ecology and Evolution. Video S3: Projected summer habitat suitability for bowhead whales using a medium ecological niche model threshold of 12.5%. For details, see Freymueller, Lorenzen et al. (2025) Ecology and Evolution. Video S4: Projected summer habitat suitability for bowhead whales using a high ecological niche model threshold of 15%. Setting suitability values higher than this resulted in the Okhotsk Sea having little to no suitable habitat, which is unlikely given the persistence of bowhead whales there throughout the Holocene. For details, see Freymueller, Lorenzen et al. (2025) Ecology and Evolution.