2.5. Population genetic differentiation and clustering
For differentiation and structural analyses, loci under Linkage
Disequilibrium (LD) were removed using plink v2.00a3.3 (Purcell et al.,
2007) (–indep-pairwise: 50 [’kb’] 5 0.20) (Table 2). Pairwise Fst
values (Weir & Cockerham, 1984) were estimated with the functionstamppFst from ‘StAMPP’ R package (Pembleton et al., 2013), and
significant differences for each population pair were evaluated with
confidence limits at 95% after 1000 bootstrap iterations. Heatmaps
based on Fst were created using pheatmap in R (Raivo Kolde,
2019).
Isolation by distance (IBD) was evaluated by correlating the Fst genetic
distances with linear geographic distances among populations, using a
Mantel test (Mantel, 1967) with function mantel.randtest in the
‘ade4’ package in R (Dray and Dufour, 2007) with 1000 permutations. The
test was performed for Menorca and Mallorca/Cabrera archipelago
separately to control for distance effect between major archipelagos.
To reconstruct evolutionary relationships among island populations, we
first used a Discriminant Analysis of Principal Components (DAPC),
integrated in the ‘Adegenet’ R package (Jombart, 2008). In this
multivariate statistical approach, sample genetic variance is
partitioned into groups to maximize discrimination, without prior
knowledge on genetic structure. The number of meaningful PCs retained
was chosen by a cross-validation with the function xvalDapc(Jombart & Collins, 2015) with default parameters (1000 replicates).
Population structure, including contemporary and historic admixture
among populations and their relationships, were further investigated
using ADMIXTURE v1.3.0 (Alexander et al., 2009), setting the numbers of
co-ancestry clusters (K ) from 2 to 16 (maximum number of
populations). Cross-validation was performed with the -cv option
(10-fold) to retain the best value of K clusters (lowest cv error).
Membership probabilities were calculated according to the retained
discriminant functions and specimens showing above 90% of assignment
probability to an island different from their source were considered as
recently translocated, or potential mislabeled and removed (none
detected).