3.3 Reconstructing scenario of breeding site colonization
A split between ancestral Atlantic and Indian populations (Fig. 2a) occurring 1.76 My ago (95% CI range 0.99-2.60) was inferred based on all 9 gene trees (2.71 My ago (1.17-4.72) using only mitochondrial markers). West and east Atlantic ancestral populations split around 1.38 My ago (0.78-2.04), baroli and boydi split at 0.85 My ago (0.44-1.32), and nicolae and bailloni at 0.70 My ago (0.33-1.13). The *BEAST analysis based on all nine markers showed the same topology though with generally lower divergence times and higher confidence intervals (Supplementary Material 8). ABC analyses also supported a similar scenario of ancestral population divergence: best retained topologies using mtDNA markers and all nine markers suggested, starting from oldest to newest splits, nicolae being ancestral, leading to the appearance of boydi (Fig 2d, Supplementary Material 5). Then lherminieri diverged from boydi , andbaroli diverged from boydi . Finally, baillonidiverged from nicolae (Fig. 2d Supplementary Material 4). Our phylogenetic trees placed the Central Pacific taxon dichrouswithin the Indian clade, thus supporting the putative scenario of Atlantic lineages diversifying from Indian Ocean rather than from Pacific ancestors (Fig. 2b,c). Within lherminieri , ABC analyses suggested a northerly stepping stone colonization process, from Martinique to the Bahamas (Supplementary Material 5). Similarly, the most likely scenario of population divergence within baroli was colonization from the Canaries to the more northerly Azores.
Finally, Bayesian Skyline analyses inferred current effective population sizes to be around 104 individuals (see Supplementary Material 10). Mean Ne seemed to increase slowly over time, but high confidence intervals precluded detection of any sudden change in population sizes. Confidence intervals on current population sizes were also very large, in particular for the individuals breeding on Réunion (see Supplementary Material 10). The demographic parameter estimations were consistent with a constant population size over time for each lineage, as suggested from Fu’s Fs results (which are more suitable than Tajima’s D to estimate population expansion; Ramirez-Soriano et al. 2008; Supplementary Material 7).
Discussion