Mitochondrial and nuclear DNA sequence data
Based on variation in 700 bp of the mitochondrial encoded cytochrome b gene (cytb) we identified 34 unique haplotypes among 100 H. rostratus individuals sampled from nine islands. Cytb data was partitioned into codon positions (CP) based on the results of PartitionFinder (76) analysis, and for the Bayesian inference (BI) analyses run through MrBayes (77) using the substitution models HKY+I (CP1), HKY (CP2) and GTR (CP3). Phylogenetic analyses using BI revealed a strongly supported, basal split between specimens from the northern islands + Frégate and specimens from the southern islands (Fig. 1a). The clustering pattern in the cytb phylogeny is congruent with the cytb haplotype network (Fig. 1b). The sharing of some haplotypes between islands could be indicative of either ancient geographic spread followed by vicariance or that some migration has occurred in the recent past between islands.
Northern + Frégate island vs. southern island individuals are separated by p-distances of 3.2–4.8% (between group mean distance = 3.7%; net between group distance = 3.4%: Table 1). The maximum amount of genetic variation within the two main lineages is greater within the southern island group (p-distances up to 1.6%: between individuals from the island of Mahé) than within the northern + Frégate group (p-distances up to 1%: between individuals from La Digue and Praslin).
Four nuclear loci—pro-opiomelanocortin (pomc), brain-derived neurotrophic factor (bdnf), and two anonymous nuclear markers (brev5 and rost5; 78)—were sequenced for a subset of the samples sequenced for cytb (Additional file 2). As expected, mtDNA nucleotide diversity is greater than for nuDNA (see Table 1). No nuclear haplotypes for H. rostratus are restricted to single islands or larger within-island regions except for the samples from Frégate for brev5 (Fig. 2). For brev5 there is a widespread northern island haplotype and a widespread southern island haplotype. Specimens from the eastern island of Frégate and one specimen from the northern island of Curieuse are more similar to southern island individuals. For pomc there are main northern + Frégate and southern-island clusters, though samples from the southern island of Cerf share alleles with two widespread northern island haplotypes; a pattern also observed in the widespread northern island haplotype for rost5. For bdnf and rost5, allele sharing is extensive and there are no clear distinctions between specimens from northern- and southern- island groups. For pomc, specimens from the islands of Mahé, Silhouette and St. Anne all have unique haplotypes.
Inference under the isolation with migration (IM) model (79) supports that no migration is occurring between the northern island + Frégate and southern island groups (Additional File 3). StarBEAST analyses estimate that the northern island + Frégate and southern island groups diverged approximately 1.2 MA [95% HPD: 0.45 – 1.94 MA]. Taken together, these results suggest that the nuclear haplotypes shared between the two mtDNA lineages (northern island + Frégate and southern island groups) seen in the haplotype networks are likely ancestral polymorphisms or incomplete lineage sorting rather than evidence of recent or ongoing migration.
AFLP genetic diversity varies considerably among islands. Using Nei’s (80) unbiased heterozygosity, diversity estimates ranged from ~0.19 (Mahé) to ~0.04 (Praslin) (Table 2). Analyses of AFLP data identified considerable genetic structuring among islands. Estimates of θ were large, and 95% confidence limits of θ did not overlap with zero (θ = 0.3574; upper CI = 0.4548, lower CI = 0.2670).
In all STRUCTURE (81) analyses the optimal clustering of individuals was K = 2. Clusters reflect a northern + Frégate vs. southern island group split (Fig. 3). Intraspecific variation within the northern + Frégate island group is further divided into specimens from the northern islands (Praslin, Curieuse and La Digue) and Frégate (Fig. 3b), with one individual from Curieuse reflecting admixture between the two clusters. Within the southern-island group there is an east-west geographic subdivision of genetic variation (K=2), with individuals from Silhouette and individuals from the closer islands of Mahé and Sainte Anne forming separate clusters (Fig. 3c). Some individuals from each island show small levels of admixture. Within Mahé additional structuring was found (Fig. 4), with a separation between specimens from the most northerly (Bel Ombre, Le Niole, St. Louis, Mt Simpson Estate) and most southerly (Anse Forbans) sampled localities showing very little admixture. The somewhat intermediate (although still northerly) samples (Mt. Coton and Foret Noire) have shared alleles with both the northernmost and southernmost populations, in a clear geographic gradient.
UPGMA and PCA analyses further indicate that the relatively high levels of AFLP structure are explained primarily by the presence of two island groups (Additional file 1), the maximally supported southern islands group (samples from Mahé, Silhouette, Sainte Anne), and a much less strongly (60% bootstrap) supported northern islands (Praslin, Curieuse, La Digue) + Frégate group. Within the northern island + Frégate group there is a strongly supported (92% bootstrap) group comprising samples from Praslin, Curieuse, and La Digue (to the exclusion of Frégate). In the PCA plot (Additional file 1) there is substantial overlap among samples from the southern islands and among samples from the northern islands; the Frégate island population is completely distinguishable. Consistent with our UPGMA dendrogram, estimates of θST obtained from these two main island groups (southern and northern + Frégate) are similar to overall levels of θ obtained when individual islands were treated independently (θST for island groups = 0.3650; upper CI = 0.4666, lower CI = 0.2563). Ordination analyses provide qualitative support for patterns produced in the UPGMA dendrogram. Plots of the first two principal coordinates reiterated the differentiation of northern and southern island groups, and further illustrated the relatively loose alliance of individuals from Frégate with representatives of the northern island group (Additional file 1).
Body width and head trait dimensions are strongly sexually dimorphic in H. rostratus (Fig. 5). Females have wider bodies than males regardless of island of origin. In general, females have smaller adjusted mean values for head length and width, and shorter IO (see Methods for description of morphological measurement abbreviations), IN, EN, ET, and TN distances. The notable exception is females from Curieuse, which have larger adjusted mean values for head length and width, and longer IO, EN, ET, and TN distances. This difference is reflected in the significant interaction between sex and island of origin for all but one trait (IN distance) for head morphology.
Numbers of folds, scales and vertebrae (PF, VF, SF, VERT, SR, and PFS) vary significantly only with island of origin (Fig. 5; Additional file 1), although a significant interaction for vertebrae suggests that variation in this character is also somewhat dependent upon the sex of the individual. There are no distinct patterns in the differences for number of folds, folds with scales, scale rows, or vertebrae among islands.
The PCoA plots (Fig. 6) for both males and females show little overlap between southern and northern island samples, and much more overlap and scatter for samples from individual islands within each of these two main groups. The Frégate samples now overlap much more extensively with the southern island group samples than the northern island group. The PCA plots (Additional file 1) generally agree with the PCoA plots in terms of the similarity of samples from the different islands. For males, the first two principal components account for 92% of the variation in the data, with the first principal component (PC1) explaining 86% of the total variance. Factor loadings for PC1 were high and positive for head length and width (0.64 and 0.47, respectively) and moderately positive for other head traits (range: 0.11-0.35). Factor loadings for PC2 were high and positive for HW (0.74), moderately positive for IO (0.26) and BW (0.11), high and negative for HL (-0.61) and low and negative (-0.02 to -0.08) for other head measures. For females, the first two principal components account for 89% of the variability in the data, with PC1 explaining 82% of the total. Similar to the pattern in males, factor loadings for PC1 were high and positive for head length (0.6) and width (0.53), and other head trait dimensions (range: 0.1-0.36). Factor loadings for PC2 were high and positive for HW (0.72), moderately positive for BW (0.2) and IO (0.24), and moderately negative for HL, EN, and ET (-0.26–-0.46).
UPGMA dendrograms based on Mahalanobis distances among islands for males and females have identical branching patterns and differ only in branch length (Fig. 6). This analysis recovers two groups of islands separating the northern and southern islands, supporting the findings from the mtDNA and AFLP analyses. However, a notable difference to the genetic data is the placement of the Frégate samples, which are joined by a long branch to the group of southern islands (rather than clustering with the northern islands).
The mean level of island group (northern island + Frégate, and southern island groups) subdivision for all morphological characters was 0.18 (PST) in both males and females. Levels of phenotypic subdivision were high for VERT and PF in both males and females (males: 0.46 and 0.47, respectively; females: 0.52 and 0.48, respectively). Values for all remaining traits for males and females were between 0.09 and 0.24.
Although there is substantial evidence of sexual dimorphism in H. rostratus, as demonstrated by the ANCOVA analysis, there is no evidence that morphological variation among islands differs by sex. Mahalanobis distance between islands is highly correlated in males and females (Mantel test; r = 0.9, p = 0.001) (Additional file 1). A linear regression of PST for traits in females versus males shows a high correlation between the amount of island subdivision in the two sexes (r = 0.98, p < 0.001) (Additional file 1). The slope of the least-squared regression line is not significantly different from one (lower CI = 0.80; upper CI = 1.02) and the intercept not significantly different from zero (lower CI = -0.01; upper CI = 0.03), indicating that male-female traits have evolved in a similar manner.
Combined data set analyses
Simple Mantel tests of distance matrix associations revealed significant correlations between all of the datasets apart from between mtDNA vs. female morphometric data and between male morphometric data vs. geographic distance (Table 3). These relationships are indicative of isolation-by-distance.
The partial Mantel test results indicate a significant association between genetic (mtDNA and AFLP) and geographic distance while controlling for morphology (male and female). There is a marginally significant association between AFLP and morphological (male and female) data while controlling for geography (Table 3).
The correlations between Nei’s d (AFLP) and Da (mtDNA) between islands was r = 0.81 (p = 0.24). The relationships between molecular genetic divergence at AFLP markers and Mahalanobis distance were significant for males (r = 0.49, p = 0.024) and marginally significant for females (r = 0.43, p = 0.052). The correlations between the mtDNA based Da and Mahalanobis distances for males and females were positive but not significant (males: r = 0.25, p = 0.134; females: r = 0.18, p = 0.147).