Radiation of tropical island bees and the role of phylogenetic niche conservatism as an important driver of biodiversity

Abstract

Island biogeography explores how biodiversity in island ecosystems arises and is maintained. The topographical complexity of islands can drive speciation by providing a diversity of niches that promote adaptive radiation and speciation. However, recent studies have argued that phylogenetic niche conservatism, combined with topographical complexity and climate change, could also promote speciation if populations are episodically fragmented into climate refugia that enable allopatric speciation. Adaptive radiation and phylogenetic niche conservatism therefore both predict that topographical complexity should encourage speciation, but they differ strongly in their inferred mechanisms. Using genetic (mitochondrial DNA (mtDNA) and single-nucleotide polymorphism (SNP)) and morphological data, we show high species diversity (22 species) in an endemic clade of Fijian Homalictus bees, with most species restricted to highlands and frequently exhibiting narrow geographical ranges. Our results indicate that elevational niches have been conserved across most speciation events, contradicting expectations from an adaptive radiation model but concordant with phylogenetic niche conservatism. Climate cycles, topographical complexity, and niche conservatism could interact to shape island biodiversity. We argue that phylogenetic niche conservatism is an important driver of tropical island bee biodiversity but that this phylogenetic inertia also leads to major extinction risks for tropical ectotherms under future warming climates.

1. Introduction

At its heart, evolutionary biology attempts to explain how new species arise and evolve to occupy a myriad of niches. A predominant paradigm in evolution is that species arise from adaptive radiation into new niche spaces, with gene flow between the new and ancestral populations subsequently inhibited, eventually leading to speciation [1–4]. Phylogenetic niche conservatism provides an alternative, whereby the inability of a lineage to adapt to new or changing environments promotes speciation when populations become isolated [5–7]. This model is particularly relevant to lineages with narrow climatic niches, especially those that are spatially fragmented across existing landscapes or constrained to isolated refugia during climate extremes, promoting allopatric speciation [7].

Distinct shifts in climate since the beginning of the Quaternary enable the relative roles of adaptive radiation versus phylogenetic niche conservatism to be explored as instigators of speciation. A key issue in assessing the role of phylogenetic niche conservatism for climate-driven speciation is that continental species may alter their latitudinal distributions in response to climate [6], rather than only elevation. Depending on continental geography, such latitudinal shifts can lead to population fragmentation or range expansion [8]. However, terrestrial biota of isolated islands and archipelagos have limited opportunities for latitudinal shifts, restricting responses to shifts in elevation [9].

Island biogeography therefore provides an opportunity to compare adaptive radiation and phylogenetic niche conservatism as drivers of speciation, in a system where changes in elevational range are not confounded by other (e.g. latitudinal) shifts. The ‘taxon cycle’ provided an initial paradigm for how speciation within islands could be explained by adaptive radiation. Using Fijian ants as a model system, Wilson [10] argued that colonizing ant species were more likely suited to coastal habitats that reflect their origin, with subsequent speciation driven by niche expansion into inland and highland ecosystems. Early data and some subsequent ant studies from the South Pacific islands support this adaptive radiation model [10,11].

However, no compelling empirical studies have indicated that speciation in island taxa has been driven by niche conservatism and past climate cycles, forcing lineages into fragmented elevational refugia (though sea-level changes have been implicated: [12]). Any such studies would require islands with substantial variation in topography [13]. Topographic complexity has been associated with in situ speciation in island biogeography models. However, this has often been attributed to adaptive responses to an increased number of possible microclimatic niches rather than effects of climate cycles on lineages with narrow thermal tolerances [11,14]. Discriminating between these adaptive and non-adaptive mechanisms is possible empirically if phylogenetic histories are sufficiently resolved.

There are also theoretical reasons to expect that any non-adaptive responses to climate change might be larger in the tropics compared to temperate or boreal regions. Tropical ectotherms are expected to be less tolerant of changing climates because they have evolved in environments that experience much lower thermal variation [15,16]. For example, tropical Drosophila species demonstrate lower genetic variation in cold and desiccation traits than do temperate species [17]. More broadly, ‘Rapoport's Rule’ is the observation that species tend to have more narrow latitudinal ranges when their overall distribution is closer to the equator, interpreted as indicating that lower exposure to climate variation selects for narrower climatic tolerances [17,18].

One way to explore phylogenetic niche conservatism is to infer how traits have evolved over phylogenetic trees. Retention of ancestral climatic niches across speciation events, when alternative niches are presumably available, would support a phylogenetic niche conservatism model. On the other hand, gradual extensions of niche range by daughter species would suggest adaptive radiation. Bayesian phylogenetic methods allow ancestral traits and their rates of evolution to be inferred (e.g. [19,20]) using constant-rate and relaxed Brownian motion (random walk) models of evolutionary change [21]. For example, it is possible to examine whether trait changes are concentrated at the base of a tree, which would suggest early adaptive radiation. When applied to climatic niches, these issues are important for understanding how past climates have influenced biodiversity and when asking how well species will adapt, or not, to changing climates.

Fiji is a tropical archipelago in the Pacific Ocean, consisting of several hundred islands of varying ages and sizes. Three of these islands first emerged during the Oligocene and exhibit substantial ranges in elevation up to 1324 m. Endemic bee species of the genus Homalictus (Halictidae) arose from a single dispersal event into Fiji during the Quaternary [22]. Homalictus species are generally communal ground-nesting species and often generalist pollinators, yet little has been published on their habits [23]. Most Fijian Homalictus species forage on both weedy and native plants, indicating polylectic diets, while H. fijiensis is a super-generalist pollinator [24]. Until recently, only four Fijian species were known, but genetic and morphological studies indicate a much more speciose clade [25], with many species only recorded from high elevations [26]. Here, we combine molecular phylogenetic analyses and elevation distribution data, which reveal at least 22 candidate species (independent lineages) and demonstrate how elevational niches have evolved since the Quaternary colonization of Fiji. We further highlight that such phylogenetic signals could indicate climate-related extinction risks.

2. Results

Data from cytochrome c oxidase subunit I (COI) sequences (630 bp) for 764 specimens, and consilience with morphology and single-nucleotide polymorphism (SNP) data (8381 filtered SNP loci) for 94 specimens indicate the existence of 22 Homalictus species in our samples (figures 1 and 2). The maximum credibility all-sample tree from BEAST [27] for COI indicates that more distal nodes tend to have higher posterior probability support than more basal nodes (electronic supplementary material, Appendix and figure S1). Some poorly resolved clades have slightly different relationships across analyses, due largely to stochastic sampling of tree topologies. Complete concordance between currently described species [26], the highly supported SNP-based phylogenetic tree for five species where SNP data were available, and the COI-based all-sample tree for all 22 species in our study suggests that COI sequences are able to recover species trees (figures 2 and S3, electronic supplementary material Appendix and Results).

Figure 1.

(a) Maximum credibility tree from Bayesian phylogenetic analysis of Fijian Homalictus based on analysis of the COI gene showing inferred elevation along branches (metres above sea level), where blue branches indicate higher elevation than red branches. Species are indicated in boxes, where colour refers to geographical locations in map inset (b), with an ‘I’ indicating the location of Homalictus sp. I. Map inset (c) shows all Homalictus collection sites across the Fijian archipelago.

Figure 2.

(a) A COI phylogenetic tree of the Fijian Homalictus clades coloured by geographical location where terminal triangle depth indicates branch depth for that species and (b) a box plot of elevation in metres above sea level and sample size. Hashed tree branches indicate clades with a median elevation below 800 m.a.s.l. and shadowed branches indicate SNP tree topology. Branch and boxplot colour refer to geographical location in figure 1b. Node values indicate posterior probability where it exceeds 0.85. The approximate maximum elevation for the sampled highland regions are as follows: Mt. Batilamu (1110 m.a.s.l.), eastern Monasavu dam (1087 m.a.s.l.), Mt. Tomanivi (1324 m.a.s.l.), De Voux's peak, Taveuni (1195 m.a.s.l.), and Mt. Nadarivatu (1054 m.a.s.l.). (Online version in colour.)

Eighteen of 22 Fijian bee species had a median elevation of over 800 m.a.s.l. and only four species had median elevations from below 800 m.a.s.l. (figure 2b). Most species have very narrow elevational bands with only four species having an elevational range greater than 500 m (H. fijiensis, H. hadrander, H. groomi, and H. sp. O) and these form a monophyletic group (figure 2a). Fourteen species were only recovered from single mountain peaks (figure 2).

Multiple regression analyses indicate that sampling effort (measured as the number of DNA sequences obtained for each elevational band) had a relatively small and only marginally significant effect on the number of species detected (β = 0.536, t = 3.213, p = 0.049). Elevation had a much larger and clearly significant impact (β = 0.925, t = 5.540, p = 0.012). We can therefore conclude that species' distributions are strongly tied to elevation, and that this is not solely an artefact of differing sampling efforts across different elevations.

To test for mode of evolution, elevational shifts were optimized on a species-level phylogeny using BayesTraits v. 3.0 [20] under a range of models (e.g. early burst or punctuated changes). We tested if changes in elevation were phylogenetically conserved (λ), associated with speciation events (κ), or relatively constant through time (δ), as indicated by tree-transformation parameters. When λ = 0 it indicates that a trait is evolving among species completely independently of phylogeny, while λ = 1 indicates that trait evolution is entirely explained by phylogeny. Our λ estimate was 0.37 (95% highest posterior density (HPD) = 7.79 × 10⁻⁵, 0.86) for median elevation and 0.59 (95% HPD = 0.14, 0.99) for minimum elevation (table 1; electronic supplementary material, Appendix and figure S2), indicating moderate to strong trait correlations with phylogenetic history. Values of κ stretch or compress individual phylogenetic branch lengths: when κ = 0 trait evolution is independent of branch length (change occurs at nodes suggesting punctuated evolution), κ > 0 indicates trait change is associated with branch length (trait change occurs along branches, rather than concentrated at speciation nodes). Our κ estimate was 1.21 (95% HPD = 0.09, 2.99) for median elevation and 0.98 (95% HPD = 3.37 × 10⁻⁴, 1.91) for minimum elevation (table 1; electronic supplementary material, Appendix and figure S2), indicating that elevational changes are strongly associated with branch lengths, and not concordant with a punctuated model of evolution, where elevational changes should instead be associated with speciation events. Delta scales the length of basal versus terminal branches: δ < 1 stretches basal branches and δ > 1 stretches terminal branches (i.e. the rate of trait evolution varies with distance from the phylogenetic root). Our δ estimate was 2.18 (95% HPD = 1.17, 3.00) for median elevation and 1.95 (95% HPD = 0.489, 2.99) for minimum elevation (table 1; electronic supplementary material, Appendix and figure S2) providing no evidence for an early burst model of change, but is instead consistent with approximately constant rates across time. The inferred median and minimum elevations for the most recent common ancestor of Fijian Homalictus were 828 m.a.s.l. (95% HPD = 636, 1020) and 687 m.a.s.l. (95% HPD = 451, 947), respectively (table 1). The strong phylogenetic signal, strong correlation between change and branch length, and lack of early burst changes all indicate that a Brownian motion model is appropriate.

Table 1.

Results from BayesTraits analyses estimating different parameters for both median and minimum elevation for each clade. The parameters estimated are lambda (λ), kappa (κ), delta (δ), and alpha (α), where λ, κ, and δ represent standard estimates of character change across a phylogeny and α is the inferred ancestral elevational state at the root node. Values chosen to be the best fit by Akaike's information criterion with 100 bootstrap replicates are shown in italics.

elevation	parameter	λ (95% HPD lower, upper)	K (95% HPD lower, upper)	∂ (95% HPD lower, upper)	α (95% HPD lower, upper)
median	λ	0.37 (7.79 × 10-5, 0.86)			852 (546, 1143)
	K		1.15 (0.06, 2.19)		835 (383, 1276)
	∂			2.18 (1.17, 3.00)	832 (709, 1044)
	λ and K	0.36 (4.10 × 10-7, 0.85)	1.21 (0.09, 2.99)		852 (569, 1138)
	λ and ∂	0.40 (3.18 × 10-5, 0.90)		1.84 (0.52, 3.00)	828 (636, 1020)
minimum	λ	0.59 (0.14, 0.99)			710 (326, 1080)
	K		0.89 (1.69 × 10-5, 1.80)		713 (254, 1184)
	∂			1.95 (0.89, 2.99)	675 (491, 883)
	λ and K	0.59 (0.13, 1.00)	0.98 (3.37 × 10-4, 1.91)		707 (350, 1076)
	λ and ∂	0.59 (0.12, 1.00)		1.70 (0.47, 2.97)	687 (451, 947)

To trace the evolution of elevational range across the full phylogeny, and incorporate phylogenetic uncertainty in our inferences of niche evolution, we traced elevation across all post-burnin trees sampled in the COI all-sample analysis (see above), using BEAST 1.10. The actual elevation of each sequenced specimen was used, and modelled using standard and relaxed Brownian motion models [28]; both gave very similar results but the latter fitted the data better (Bayes Factor score > 1000) and is presented in figure 1. Other deviations from Brownian motion—such as punctuated and early burst evolution—were not indicated (see BayesTraits above). From our BEAST analysis, the inferred ancestral elevation of the most recent common ancestor of Fijian Homalictus was 896 m.a.s.l. (figure 1), highly consistent with the BayesTraits results above. Most [17,18] speciation events have involved no major elevational transitions, and there were only three to four speciation events that involved transitions from highland to lowland habitats in one of two daughter species (H. fijiensis, H. taveuni, H. sp. I and O; the clade including H. fijiensis, H. groomi, and H. sp. O could represent one or two elevational transitions; figure 2).

3. Discussion

The COI sequence, SNP, and morphological data considered together indicate that the 22 major clades we have identified here comprise valid biological species (electronic supplementary material, Appendix and figure S3). Our phylogeny of 22 species entails 21 speciation events, with branch transformation parameters λ, κ, and δ providing support for a niche conservation model of speciation over adaptive radiation (table 1; electronic supplementary material, Appendix and figure S2). The low frequency of elevational shifts (three or four elevational transitions from 21 speciation events) suggests that adaptive radiation, at least in terms of climatic niche shifts, has not been a major driver or correlate of speciation (figure 2). The clade containing H. fijiensis, H. hadrander, H. groomi, and H. sp. O likely has a more eurythermic common ancestor compared to its sister clade (H. sp. N) due to its wider elevational range. Eurythermy could be an important trait allowing lowland insular species to persist during cooler glacial periods. Trait reconstruction indicates the common ancestor of Homalictus occupied an elevation that would roughly correspond to between 800 and 900 m.a.s.l. in today's climate (figure 1).

Extensive anthropogenic habitat destruction in many lowland regions since human habitation of Fiji approximately 3500 years ago [29] is unlikely to have caused reductions in lowland Homalictus species diversity for several reasons. Firstly, lowland rainforest (under 600 m.a.s.l.) makes up 78% of all natural forests, with upland (601–800 m.a.s.l.) and montane rainforest (over 801 m.a.s.l.) accounting for only 8% and 4%, respectively [30]. Secondly, a generalist diet in Fijian Homalictus enables the use of introduced and weedy plants that may have been brought to Fiji by the earliest human settlers and up until current times [24]. Finally, nesting preference for bare or sparsely vegetated ground indicates likely resilience of these bees to habitat clearing.

Climate conditions associated with specific elevations in Fiji today are likely to have been different over past climate cycles [13,31]. Groom et al. [22] estimated a crown age for Fijian Homalictus in the Mid-to-Late Pleistocene, but that may well be an underestimation since those analyses did not incorporate many of the recently discovered species. Regardless, Homalictus lineages in Fiji will have experienced multiple glacial–interglacial cycles such that thermal niches associated with highlands in the current climate are likely to have episodically extended to lower elevations in the past [32–34]. However, past climates would also have included warming maxima similar to the current time [32–34]. Narrow thermal tolerances and poor adaptive capabilities of some tropical ectotherms [16,18], the topographical complexity of Fiji, and Quaternary climate cycles could act synergistically. These synergies could produce repeated cycles of population admixture and isolation as species moved into lower elevations during glacial maxima, and retreated to highland refugia during interglacial periods, such as the present [5]. In this sense, phylogenetic niche conservatism combined with climate cycles could have driven repeated cycles of allopatry and speciation [5].

It is possible that speciation by both niche conservatism and by adaptive radiation occur concurrently and could act synergistically. Isolation initiated by climatic niche conservatism could subsequently be promoted by differential adaptations to other local conditions [35] in addition to genetic drift. Despite such potential interactions, our results clearly indicate widespread phylogenetic niche conservatism for elevational niches. This contrasts with studies on Fijian ants where initial colonization by lowland, coastal-adapted species was followed by gradual adaptive expansion into higher inland elevations [10,11,36]. Instead, our results for the Fijian bees support a very different model where ancestral niches are retained and speciation arises by geographical fragmentation of this niche space, promoting allopatry.

The same climatic factors that could drive tropical ectotherm speciation [5,6,35] could also determine extinction risks with globally changing climates. For example, a narrow climatic tolerance means that tropical ectotherms are expected to be particularly vulnerable to changing climates [16,37,38]. Many lineages have altered their distributions in the direction locally expected with climate change and 41% of species that have been examined have responded to recent mild (0.6°C) global warming [39]. Mountaintop species are predicted to be particularly vulnerable to climate change because of their limited ability to disperse in response to warming climates [40]. There is a global trend of declining distributions among montane species as their lower elevational extents shift towards mountain peaks [41]. Several studies indicate that some species are already nearing their elevational limits [26,42–44].

Tropical ectotherm taxa that have demonstrated strong elevational tracking with past climate cycles are at risk from globally warming climates as elevational shifts in distribution are associated with a reduction or loss of viable habitat [40,45,46]. Local extinctions of some highland taxa due to elevational tracking of climate have already been recorded [40,43,45], and one Fijian Homalictus species (H. achrostus) is suspected to be verging on extinction or be functionally extinct [26]. The generality of niche conservatism-driven speciation across various taxonomic groups in the tropics is an important and pressing area of future research. While islands provide a simplified system to examine these patterns, similar patterns might be found in montane continental systems where niche conservatism might result in speciation as lineages track climate latitude and altitude simultaneously. Such investigations will have important implications for our understanding of how biodiversity arises and will inform us about broad-scale climate extinction risks.

Our data and the arguments above, combined with narrow climatic envelopes of many tropical ectotherms [16,46,47], shows the potential importance of the niche conservatism model of speciation, as a contrast to the adaptive radiation model. Indeed, if Darwin had studied these Fijian bees instead of Galapagos finches, he might have come to rather different conclusions about the origin of species. Our results advance fundamental questions of island biogeography [48], and have three important implications for understanding the role of climate cycles in the island and tropical biodiversity. First, they support the notion that speciation events resulting in the rich biodiversity of tropical ecosystems might be driven, at least in part, by niche conservatism as well as adaptive radiation. Second, they suggest that topographical complexity and climate cycles might strongly interact to shape island biodiversity. Finally, our inferred elevational niche conservatism suggests widespread yet clade-specific extinction risks for tropical invertebrates for warmer and more variable future climates. This indicates a need to explore evolutionary limits to thermal adaptation when assessing the susceptibility of tropical insular ecosystems to future climates.

4. Material and methods

(a). Sample locations and collection methods

Collections throughout Fiji were made between 2010 and 2017 from multiple localities including the main islands of Viti Levu, Vanua Levu, Kadavu, and Taveuni, as well as multiple small islands in the Lau group (electronic supplementary material, Appendix and figure S5). Sampling of specimens at each location was not biased towards particular species because, for these very small bees, only H. achrostus could be easily identified in the field due to its distinctive colouration; all other species required microscopy or DNA sequencing for species identification.

Samples were collected from 3 m to 1324 m.a.s.l. (highest elevation of Fiji) by sweep netting both native and introduced plants, and from nesting aggregations along roadsides. For each collection site, latitude, longitude, and elevation were recorded using a Garmin 550 (Garmin Ltd., USA); latitude and longitude were then checked against satellite images (Google Earth) to confirm accuracy. Once collected, bees were immediately transferred into vials containing 98% ethanol. Vials were kept cool at approximately 5°C and ethanol was replaced within a week of collection to reduce DNA degradation.

Maps of Fiji were produced in ArcMap [49] and a digital elevation model of the archipelago was provided by Fiji Lands Information System.

(b). Sampling bias and elevational species richness

It was not possible to evenly sample bees across all geographical and elevational regions of Fiji because physical access to many regions was restricted by terrain and lack of roads. Access constraints could therefore affect sampling effort and this, in turn, could influence the ability to recover true species richness in different elevational bands. Here, we quantize sampling effort as the number of DNA sequences obtained for different elevations, categorized into 200 m.a.s.l. bands. Because specimens were only identified to species levels after DNA sequencing, the number of obtained sequences represents sampling effort. We examined whether this sampling effort may have influenced our estimates of species richness using multiple regression, where the number of detected species was the dependent variable and the number of sequences (sampling effort) and elevational bands were the independent variables. The relative importance of sampling effort and elevation band for detected species richness can then be explored by regression β-values and their statistical significance.

(c). DNA extraction and sequencing

Tissue samples for DNA extraction were obtained by removing a single hind leg from each of the 764 specimens. For all samples obtained after 2014, DNA extraction and polymerase chain reaction (PCR) amplification were completed at the South Australian Regional Facility for Molecular Ecology and Evolution (SARFMEE). DNA extraction and PCR amplification of COI prior to the 2014 samples was completed at the Canadian Centre for DNA Barcoding at the Biodiversity Institute of Ontario [22] and amplification used the universal primer pair LepF1 and LepR2 [22,50]. Extractions at SARFMEE followed protocols described by Ivanova et al. [51] with the subsequent DNA eluted into 75 µl of (10 mM Tris-HCl pH 8.0; 0.1 mM EDTA) (TLE) buffer. PCR amplification of the 658 bp fragment of the DNA (COI) was completed using the primers LCO1490 (forward) and HCO2198 (reverse). The 25 µl PCR reactions comprised the following reagents: sterile H₂O (15.9 µl), (1 × Immolase buffer/1.5 mM MgCl₂/0.8 mM dNTP mix/0.05 mg/ml BSA) (MRT) buffer (5 µl), 1 µl (5 µM) of LCO1490, 1 µl (5 µM) of HCO2198, Immolase Taq (0.1 µl), and DNA from the specimen (2 µl). The thermocycling regime comprised one cycle at 94°C for 10 min, then five cycles at 94°C for 60 s, 46°C for 90 s, 72°C for 75 s, followed by 35 cycles at 94°C for 60 s, 51°C for 90 s, 72°C for 75 s, followed by 72°C for 10 min and then 25°C for 2 min.

Sequences were checked against the National Center for Biotechnology Information (NCBI) BLAST database to screen for non-target DNA. Forward and reverse sequences were aligned and chromatograms visually checked before creating final consensus sequences in Geneious v. 10.2.2 [52]. Initial alignments were trimmed to 630 bp to avoid any problems associated with missing data.

(d). Phylogenetic, elevational, and species analyses

The full COI alignment consisted of 630 bp for 764 specimens. Partition finder v. 2 was employed using Bayesian information criterion (BIC) and a greedy algorithm to find the best partition schemes and DNA substitution models from widely used (i.e. MrBayes) models [53–55]. The first and second codon positions were combined into a single partition with an Hasegawa, Kishino, and Yano (HKY) + I substitution model. A generalized time reversible (GTR) substitution model was applied to the third codon position. The BEAST file and parameters for phylogenetic analyses were set using BEAUti v. 1.8.4 [56]. Because of the small numbers of substitutions on each branch, a strict clock was used to avoid overparametrization. To infer changes in elevation across the tree, we included elevation as a continuous trait using a strict or relaxed Brownian motion model (confirmed as adequate given our λ, κ, and δ estimates; table 1). Phylogenetic analyses were implemented in BEAST v. 1.10 [27] with 200 million iterations sampled every 50 000th iteration. Resulting log files were analysed in Tracer v. 1.6 [57] and a burnin of 2.5 × 10⁷ iterations was employed, which was always after stationarity had been achieved. Maximum clade credibility trees and posterior probability support values were obtained using TreeAnnotator v. 1.8.4 [56]. Each run was performed four times for each analysis to ensure consistent results and stationarity. Post-burnin log and tree files for each run were then combined using LogCombiner v. 2.5.2 [58] to generate summary statistics and consensus trees.

To infer the evolutionary mode and phylogenetic signal in the elevation data, we used BayesTraits v. 3.0 [20]. The tree-transformation models employed in BayesTraits assume that each terminal taxon is a species, hence we repeated the BEAST analysis using only one DNA sample from each species, and elevation data as either the median or minimum for all samples of that species. The (reduced) BEAST analysis used 100 million iterations, sampling every 50 000th iteration; stationarity and burnin was checked as above. The resulting consensus tree was run in BayesTraits using the median and the minimum elevational value for each terminal taxon to estimate λ (degree of phylogenetic signal), κ (degree of punctuated evolution), and δ (degree of early burst, adaptive radiation). The model of best fit for each estimate was chosen using Akaike's information criterion with 100 bootstrap replicates in Tracer [59]. Analyses in BayesTraits used 500 million iterations sampled every 50 000th iteration. Each run was performed four times for each model at each elevation to ensure consistent results. BayesTraits log files for each run were then combined using LogCombiner v. 2.5.2 [58] to generate summary statistics and consensus trees.

We attempted to co-estimate phylogeny and elevational niche evolution, but these analyses repeatedly failed to converge. Thus, to infer elevational changes across the full phylogeny, we mapped elevation across all post-burnin trees sampled in the full COI analysis. This was done using BEAST, under a standard rate-constant Brownian motion model, as well as a rate-variable Brownian motion model, which assumes rates vary across branches according to an uncorrelated relaxed clock [28]. Stationarity and burnin were confirmed as above. Both models gave very similar ancestral state reconstructions, but the latter model fitted better and is shown in figure 1.

Genetic analyses of bee clades were explored using Arlequin v. 3.11 [60]. For each species with multiple haplotypes and a sample size of more than 10 specimens, we calculated haploytype diversity (h) and pairwise F_ST values.

(e). SNP quality filtering and analyses

The thorax and front legs were taken from 19 individuals from H. fijiensis, H. tuiwawae, H. ostridorsum, H. groomi, and H. sp. S, respectively. To perform Restriction-site Associated DNA sequencing (RAD seq), the solid-state method Diversity Arrays Technology was used [61]. The restriction enzymes used were a combination of PstI and HpaII enzymes. Only female specimens were used to avoid the impact of male haploidy on SNP diversity. Post filtering, missing data were capped at 1.16%.

A total of 62 426 SNP loci were called across all species. Using the R package ‘DArTR’ v. 1.0.5 low-quality loci were removed at a threshold of 0.85% removing loci with 15% or more missing values [62], leading to the retention of 8381 SNP loci. The neighbour-joining tree (electronic supplementary material, Appendix and figure S4) was made using the R package ‘ape’ with the ‘nj’ function [63].

Once SNP data were filtered they were subjected to a discriminant analysis of principal components (DAPC) using the DAPC procedure [64] in the Adegenet package in R [65]. The DAPC was used to identify the number of genetic clusters within the SNP data and the relationship between these clusters. DAPC uses synthetic variables constructed as linear combinations from the original alleles, showing the largest between-group variations and lowest within-group variation. Discriminant analysis also provides membership probabilities of each individual to the different clusters. Our DAPC followed protocols outlined by Jombart [64].

(f). Morphological data

To determine if major mitochondrial clades corresponded to biological species, we examined multiple morphological traits. Internal male genitalic traits have been used as major species diagnostic characters for Homalictus species in the southwest Pacific [25,26], and these were examined for 12 species in this study where male specimens were available, along with external female morphology [26]. For the remaining species, only female external morphology was compared with COI and SNP phylogenies.

Data accessibility

Collection and GenBank accession data for Homalictus specimens are provided in electronic supplementary, additional data electronic supplementary material, table S2. The alignments, SNP data, BEAST, and BayesTraits executables are deposited in Dryad Digital Repository: https://dx.doi.org/10.5061/dryad.80gb5mknf [66].

Authors' contributions

J.B.D., M.I.S., and M.P.S. conceived the study and developed the experimental design; J.B.D., E.F., C.M., S.V.C.G., E.D., C.R., O.K.D., M.I.S., and M.P.S. performed the fieldwork; J.B.D., C.R., and O.K.D. edited sequence data and elevation and geographical records; J.B.D., M.S.Y.L., E.F., and C.M. carried out the analyses with advice from M.I.S. and M.P.S.; J.B.D. wrote the manuscript and prepared the figures, with editorial advice from S.V.C.G., O.K.D., M.I.S., M.S.Y.L., and M.P.S.; all authors gave final approval for publication.

Competing interests

The authors declare no conflict of interest.

Funding

Project and personnel funding were provided by the Australia and Pacific Science Foundation, New Colombo Plan program, Flinders University, The South Australian Museum, and Playford Trust Scholarships to J.B.D. and O.K.D.