Rapid global warming is severely impacting Arctic ecosystems and is predicted to transform the abundance, distribution, and genetic diversity of Arctic species, though these linkages are poorly understood. We address this gap in knowledge using palaeogenomics to examine how earlier periods of global warming influenced the genetic diversity of Atlantic walrus (Odobenus rosmarus rosmarus), a species closely associated with sea ice and shallow-water habitats. We analysed 82 ancient and historical Atlantic walrus mitochondrial genomes (mitogenomes), including now-extinct populations in Iceland and the Canadian Maritimes, to reconstruct the Atlantic walrus’ response to Arctic deglaciation. Our results demonstrate that the phylogeography and genetic diversity of Atlantic walrus populations were initially shaped by the Last Glacial Maximum (LGM), surviving in distinct glacial refugia, and subsequently expanding rapidly in multiple migration waves during the late Pleistocene and early Holocene. The timing of diversification and establishment of distinct populations corresponds closely with the chronology of the glacial retreat, pointing to a strong link between walrus phylogeography and sea ice. Our results indicate that accelerated ice loss in the modern Arctic may trigger further dispersal events, likely increasing the connectivity of northern stocks while isolating more southerly stocks putatively caught in small pockets of suitable habitat. 

The 82 mitochondrial genomes were obtained from historical and ancient specimens collected from different museums around Europe and Canada. Of these, 28 were obtained from previous publications (Keighley et al. (2019)) and 10 samples originally from Svalbard obtained from Star et al. (2018). Ancient DNA laboratory work was undertaken at the Globe Institute from the University of Copenhagen, Denmark following stric aDNA laboratory guidelines. Shotgun sequencing was performed on a range of Illumina technologies (MiSeq, HiSeq 2500 and HiSeq 4000) at the Danish National High-throughput Sequencing Centre. Throughout all laboratory work, samples were randomly given a unique sample number, with different groupings for extraction, library build, amplification, and sequencing to ensure no clustering of samples from a particular locality or time period. Samples run on the Illumina HiSeq 4000 were dual-indexed due to the risk of index-hopping (Van der Valk et al. 2020). Reads were trimmed, filtered, and aligned using the PALEOMIX (v1.2.13.4) BAM pipeline (Schubert et al. 2014).