Studying patterns of population structure across the landscape sheds light on dispersal and demographic processes, which helps to inform conservation decisions. Here, we study how social organisation and landscape factors affect spatial patterns of genetic differentiation in an ant species living in mountainous regions. Using genome-wide SNP markers, we assess population structure in the Alpine silver ant, Formica selysi. This species has two social forms controlled by a supergene. The monogyne form has one queen per colony, while the polygyne form has multiple queens per colony. The two social forms co-occur in the same populations. For both social forms, we found a strong pattern of isolation-by-distance across the Alps. Within regions, genetic differentiation between populations was weaker for the monogyne form than for the polygyne form. We suggest that this pattern is due to higher dispersal and effective population sizes in the monogyne form. In addition, we found stronger isolation-by-distance and lower genetic diversity in high elevation populations, compared to lowland populations, suggesting that gene flow between F. selysi populations in the Alps occurs mostly through riparian corridors along lowland valleys. Overall, this survey highlights the need to consider intraspecific polymorphisms when assessing population connectivity and calls for special attention to the conservation of lowland habitats in mountain regions.

Sampling and genotyping

Formica selysi lives in riverine ecosystems throughout the European Alps and the Pyrenees mountains (Seifert, 2002). J. Purcell and A. Brelsford sampled workers in 152 colonies from 13 localities ranging from 180 m to 1450 m in elevation (1-32 colonies per locality, Table S1). In each locality monogyne and/or polygyne colonies were sampled within a 1 km2 area (Table S1). The sampling localities were situated along the Rhine River or tributaries (3 localities, east Switzerland and west Austria), along the Upper Rhône River or tributaries (6 localities, west Switzerland) and along tributaries of the Lower Rhône River (4 localities, France; Figure 1, Table S1). Each locality represents a separate population.

We genotyped one worker per colony. We extracted DNA from the head and thorax of each worker using the Qiagen DNeasy Blood and Tissue kit, following the protocol for insect tissue. We obtained double-digest RAD sequence data by following the ddRAD-seq protocol described in Brelsford et al. (2016). In brief, we digested genomic DNA using restriction enzymes EcoRI and MseI, ligated inline barcoded adapters, removed DNA fragments shorter than 250 bp using AMPure magnetic beads, carried out PCR amplification of each individual in triplicate, during which we added a second unique adapter for each independent plate, and carried out a final size selection on the pooled libraries, to retain sequences in the 400-500 bp range. The resulting libraries were sequenced on the Illumina 2500 Hi Seq platform of the Lausanne Genomic Technologies Facility.

Bioinformatics

Demultiplexing and quality control of raw sequences were done with the process_radtags pipeline in STACKS v. 2.2 (Catchen et al., 2013). Clean reads were aligned to an upgraded version of the reference genome of Formica selysi (Brelsford et al. 2020, NCBI, GenBank accession number: GCA_009859135.1), using BWA v. 0.7.17 (Li and Durbin, 2009). Single Nucleotide Polymorphisms (SNPs) and genotypes were called with the ref_map pipeline in STACKS, using default parameters. The initial consensus output catalogue from the populations program contained 628,232 RAD loci, with average length of 84.7 bp and average sample coverage of 28.9x. In total, 323,797 SNPs were retained, distributed across 99,299 polymorphic RAD loci.

Further SNP filtering was done using the VCFtools (Danecek et al., 2011) and the “VcfR” R package (Knaus BJ, 2017). Genotypes with quality score lower than 20 and sequencing depth lower than three-fold were considered missing data. We retained one random polymorphic site per RAD locus, to avoid bias due to linkage disequilibrium. We removed sites with heterozygosity higher than 0.70, to exclude merging paralogous loci (Paris, Stevens and Catchen, 2017). We only retained SNPs with minor allele frequency higher than 0.01 and mapping to one of the 27 chromosome-length scaffolds of the reference genome. We further removed individuals with more than 30% of missing data and selected SNPs present in 95% of the individuals retained. The resulting dataset had 13,421 SNPs, of which 923 were on chromosome 3, which contains the non-recombining social supergene (Purcell et al., 2014), and 12,498 were in the remaining 26 chromosomes.