at cpDNA loci
Of the 94 individuals analyzed for whole cpDNA genome data, 10 were discarded as they had more than 25% missing data. Of the remaining 84 individuals (with the four M. argentea individuals included), we obtained a total of 146 SNPs (109 SNPs if M. argentea individuals are excluded). We obtained a total of 30 haplotypes in M. fragrans and three additional haplotypes found exclusively in M. argentea . Within M. fragrans , we observed one common haplotype (Hap 3, ~60% of the individuals), that was present in all populations throughout the Moluccas archipelago (Figure 3; Supplementary Figure S4). Two other common haplotypes (Hap 4 and Hap 17) were present only in the North Moluccas. Unique haplotypes were found in all populations (three in Ternate, four in Tidore, seven in Bacan, seven in Ambon and six in Banda). The most common haplotype is placed in the center of the phylogenetic network (Figure 3). We also observe that Hap 3 differs from the M. argentea haplotypes by 48 mutations.
Diversity statistics obtained for the cpDNA dataset (number of segregating sites, nucleotide diversity Π) suggest a higher level of diversity in Bacan and Ternate (Table 3). At the cluster level, the nucleotide diversity is relatively lower in the Southern Moluccas cluster (Π =0.122) than in the Northern Moluccas cluster (Π =0.280). The AMOVA shows that the genetic variance is mainly explained by the cluster (Φ = 41.50% of the total variation, p < 0.001; Supplementary Table S4) and the sub-population (52%) levels. We found no signal of phylogeographic structure within our dataset (N ST= 0.462 < G ST = 0.532; p < 0.05).
Inference of population size changes in M. fragrans
The DIYABCskylineplot analyses show no evidence of significant population change over time in Ternate, Tidore, and Bacan populations, with a Bayes factor (BF , see Jeffreys, 1998) of 0.53, 0.45, and 0.34 respectively (Figure 4). MIGRAINE revealed a population expansion in Ternate, with θ μ >θ ancμ. However, Ambon and Banda exhibit a strong relatively recent bottleneck (BF = 2.17 andBF = 141, respectively). Similar patterns of population size reduction in Ambon and Banda were also shown by our MIGRAINE analyses. The ratio of scaled population sizes θ ratio =θ 0θ A (see Supplementary Table 6) indicates a recent reduction in population size in Ambon and Banda populations.
Reconstructing population history of M. fragrans
The visual inspection of simulated and observed data in the PCA plots indicates that the models considered are able to reproduce the observed genetic diversity patterns across all our datasets (see Supplementary Figure S5, S6 and S7 for the nSSR, the cpDNA and both nSSR and cpDNA data respectively). From these results, we concluded that the proposed scenarios and the prior combinations were adequate to perform ABC analysis.
Model selection according to DIYABC-RF using all three data sets (nSSR, cpDNA, and both nSSR and cpDNA) favored scenario 1 as the best supported scenario, with a posterior probability of 0.73, 0.88 and 0.79 (Figure 5; Supplementary Figure S1. A; Supplementary Table S5). According to Jeffrey’s scale (Jeffreys, 1998)⁠, the Bayes factor of the three posterior probabilities indicates significant support to scenario 1 (Supplementary Table S5). Divergence time (t ) between the two genetic clusters of M. fragrans occurred around 2159 to 8231 generations ago, and the population bottleneck in the South Moluccas cluster occurred around 299 to 402 generations ago (Table 4).