3.2 Population structure and sex-biased dispersal
Mitochondrial and nuclear results from Fu’s Fs indicated no deviation from neutrality (Supplementary Material 7). However, for Tajima’s D tests, three localities displayed significant negative Tajima’s D for all mitochondrial and nuclear markers: South Reunion (Indian Ocean, see map in Figure 1), Saint-Barthélemy (W Atlantic) and Raso (E Atlantic). In addition, three localities presented significant negative Tajima’s D at mitochondrial loci only (Selvagem and Funchal, E Atlantic; South Reunion) and two at nuclear loci only (Vila, E Atlantic; North Reunion). Patterns of population structure at seven out of 13 localities might therefore be influenced by selection and/or recent demographic changes, in addition to neutral processes.
Gene trees and phylogenetic inference with *BEAST and MrBayes revealed a hierarchical structure composed of two well-supported (posterior probabilities PP ≥ 0.95) reciprocally monophyletic clades corresponding to the two oceans, within which individuals from the five lineages further clustered into monophyletic sub-clades (Fig. 2b,c and S7). All except one of these sub-clades (East Atlantic boydi ) were supported in *BEAST (PP ≥ 0.95) using all concatenated markers. For both mtDNA markers, and all markers concatenated, the central-Pacificdichrous lineage was nested within thebailloni /nicolae clade, although node supports fordichrous position within Indian Ocean clade were weak (between 0.27 and 0.73). Assignment to an ocean basin based on nuclear haplotype networks was however discordant from the mitochondrial data for 33 individuals, for at least one nuclear locus (Fig. 3b and S8): 15 Atlantic individuals fell closely to the Indian phylogroup, and 18 Indian Ocean individuals clustered within the Atlantic group. All of these 33 individuals showed the mitochondrial signature expected based on their geographical sampling location. Interestingly, a baroliindividual showed one haplotypic phase clustering with the baroliphylogroup (mother) while the other haplotypic phase (father) clustered with the nicolae phylogroup for four nuclear markers (the two remaining could not be assigned to any particular lineages). This individual might be the result of hybridization, although further analyses based on additional markers would be necessary to detect more robustly hybridization among lineages. The ambiguous assignment of the other individuals might be due to introgression or incomplete lineage sorting (see below).
In parallel, we used an AMOVA framework with the five nominal lineages now defined a priori, to examine how genetic variants partitioned among and within these taxonomic units. Most of the genetic variance was due to inter-lineage differentiation (88.5% and 58.4% for mitochondrial and nuclear markers, respectively). The variance among sampling localities within lineages accounted for 0.5 to 4.1%, while variance within sampling localities represented 11.0 to 37.5%. PairwiseF ST showed consistently higher values among, than within lineages for both marker types, with mostly non-significant values within each lineage (Table 2a). Moreover, 24 nuclear FST values were found non-significant versus 10 mitochondrial ΦST values (Table 2a). Population average pairwise differences led to similar results, with high structuration for the five nominal lineages (Supplementary Material 7).
Genetic distance increased clearly with geographic distance (Fig. 4a), but Mantel tests were performed only between pairs within the same Ocean (given that each Ocean taxon is likely different species). Tests confirmed that genetic and geographic distances were strongly correlated to each other, both for mtDNA and nuclear markers when analyzing pairs of populations within an Ocean basin (r=0.88 and 0.70, n=45, p <0.005 for mtDNA and nuDNA, respectively, Fig. 4a,b). Between breeding sites and within lineages, isolation by distance could not be reliably tested as the number of populations was too low, but visually it seemed that there was no relation between geographic and genetic distances (Fig. 4).
Indian Ocean populations (nicolae and bailloni ) showed stronger female dispersion as indicated by significantly stronger deficit of heterozygotes and a significantly lower average relatedness in females (Table 3). Conversely, in baroli ,F IS was significantly higher for males and they were less related to each other than females, suggestive of male-biased dispersal. Finally, an ambiguous pattern was found for bothlherminieri (males had stronger deficit of heterozygotes and were significantly more related to each other than females) and boydi(female F IS was significantly higher than maleF IS, though females were more related to each other than for males at one sampling locality). In addition, population structure within lineages, as measured with F ST, was similar between sexes, but between oceans a larger range ofF ST values was observed for males with higher maximum values, suggesting that males were more structured at least for some pairs of populations (e.g. lherminieri vs. Indian Ocean lineages; Table 2b). Overall males seemed more structured than females between oceans, suggesting that females disperse farther, but genetic signal for sex-biased dispersal varied geographically: female-biased in the Indian Ocean, male-biased or inconclusive in the Atlantic Ocean.