3.3. Phylogeographic scenario
Prior to phylogenetic reconstruction, we assessed past migration events among populations using allele frequencies on Treemix and OptM analysis to support the best number of edges. OptM analysis indicated a single migration edge explaining 98.9% of the variance. The event with the greatest weight occurred from Colomer into Cabrera-Escatlasang (mean migration weight: 0.454032, mean p-value: 3.22478e-11), followed by a migration event from Dragonera into Escatlasang-Cabrera (mean migration weight: 0.380029, mean p-value: 6.71e-05) (Figure S1).
A midpoint rooted ML tree was then built using consensus sequences for each population on the concatenated SNP dataset (Figure 6). The phylogenetic tree supports the major grouping obtained by the ADMIXTURE, Fst and DAPC analyses, largely in line with the population geographic distribution (Figure 1). Menorcan and Cabrera/Mallorca populations formed clearly distinct phylogenetic clades (bootstrap, bs =100%). Within the Mallorca/Cabrera clade, further substructuring indicated three well differentiated clades: i) the southern Mallorca islets of en Curt, Moltona and Guardia (bs= 99%)), ii) Cabrera islets of Foradada and Esclatasang (bs=76%), with Cabrera as a basal lineage, and iii) the two northern Mallorcan islands of Colomer and Dragonera (bs=80%) (Figure 1 and Figure 6). Within the Menorca clade, substantial support is given only to the grouping of Porros de Cavalleria (PRCV) and Revells (RAVL) (bs=73%), while the small islet of Porros de Fornells (PORROS) represents an independent lineage, highly differentiated from to the rest of Menorcan populations (bs=89%). The remaining Menorcan populations were poorly resolved, consistently with their lower Fst values (Figure 4) and higher admixture levels (Figure 5B).
Tree rooting using non-reversible models on IQ-tree consistently identified the population of Colomer (Mallorca) as the most likely ancestral lineage. The island showed the highest rootstrap value (86,13%, after 10K bootstrapping on consensus sequences, all sites; values <2% for all other branches) (Figure S2). This result was robust to the use of consensus sequences with variant sites only and to subsampling to one random specimen per population, all consistently recovering the highest rootstrap value for Colomer (71.35 and 69.54, respectively). Root testing provided the maximum p-value of the approximately-unbiased (AU) test for the branch leading to Colomer (p=0.85). All branches within the Menorca clade showed a p<0.05, significantly excluding Menorca as a potential source of the ancestral population. Rootstrap values and root testing considering only the archipelago of Menorca did not provide any clear root support within Menorca.