Repeated range expansion and niche shift in a volcanic hotspot archipelago: Radiation of C4 Hawaiian Euphorbia subgenus Chamaesyce (Euphorbiaceae)

Abstract Woody perennial plants on islands have repeatedly evolved from herbaceous mainland ancestors. Although the majority of species in Euphorbia subgenus Chamaesyce section Anisophyllum (Euphorbiaceae) are small and herbaceous, a clade of 16 woody species diversified on the Hawaiian Islands. They are found in a broad range of habitats, including the only known C4 plants adapted to wet forest understories. We investigate the history of island colonization and habitat shift in this group. We sampled 153 individuals in 15 of the 16 native species of Hawaiian Euphorbia on six major Hawaiian Islands, plus 11 New World close relatives, to elucidate the biogeographic movement of this lineage within the Hawaiian island chain. We used a concatenated chloroplast DNA data set of more than eight kilobases in aligned length and applied maximum likelihood and Bayesian inference for phylogenetic reconstruction. Age and phylogeographic patterns were co‐estimated using BEAST. In addition, we used nuclear ribosomal ITS and the low‐copy genes LEAFY and G3pdhC to investigate the reticulate relationships within this radiation. Hawaiian Euphorbia first arrived on Kaua`i or Ni`ihau ca. 5 million years ago and subsequently diverged into 16 named species with extensive reticulation. During this process Hawaiian Euphorbia dispersed from older to younger islands through open vegetation that is disturbance‐prone. Species that occur under closed vegetation evolved in situ from open vegetation of the same island and are only found on the two oldest islands of Kaua`i and O`ahu. The biogeographic history of Hawaiian Euphorbia supports a progression rule with within‐island shifts from open to closed vegetation.

There are 17 Euphorbia species native to the Hawaiian Islands as recognized by the current morphologically based classification. One of them, E. haeleeleana belongs to Euphorbia subgenus Euphorbia (Dorsey et al., 2013). It represents a separate colonization event and is outside the scope of this study. The remaining 16 named species form a clade within Euphorbia subgenus Chamaesyce section Anisophyllum, hereafter referred to as Hawaiian Euphorbia (Yang & Berry, 2011). Euphorbia section Anisophyllum comprises about 400 species and mainly distributed in warm areas in North and South America (Halford & Harris, 2012;Yang et al., 2012). Members of the section are commonly small, weedy herbs, and all but three species exhibit C 4 photosynthesis (Yang & Berry, 2011). Like other typical C 4 plants, distribution of Euphorbia in the continental North and South America is mainly in warm, dry, and exposed habitats. In contrast, however, Hawaiian Euphorbia species occupy a wide variety of habitats, including coastal strand, dry forests, wet forests, and bogs, and they range in habit from subshrubs to trees 10 m tall (Figure 1). Four of the species have two or more recognized varieties. Ten species are endemic to a single major island, whereas the remainder is known from two or more major islands (Table 1). Six species and four varieties of Hawaiian Euphorbia are federally listed as endangered (marked with "*" in Table 1). A prior phylogenetic study with taxon sampling throughout section Anisophyllum suggested that Hawaiian Euphorbia originated following allopolyploidy, with their closest relatives being small herbs occurring in dry, warm, and exposed habitats in southern United States, northern Mexico, and the Caribbean, including E. cinerascens, E. leucantha, E. mendezii, E. stictospora, and E. velleriflora (Figure 1f;Yang & Berry, 2011). Given the overlapping distribution of these putative mainland close relatives, the allopolyploidy event likely happened before dispersal to the Hawaiian Islands. The longdistance dispersal most likely occurred via the tiny seeds (typically 1-2 mm long) that adhere to birds with their mucilaginous seed coat (Carlquist, 1966(Carlquist, , 1980Price & Wagner, 2004).

Following their arrival on the Hawaiian Islands, Hawaiian
Euphorbia became woody, and some species lost the mucilaginous seed coat and developed larger seeds (Carlquist, 1966). Yet all species retained C 4 photosynthesis like their close mainland relatives (Pearcy & Troughton, 1975;Sporck, 2011). C 4 photosynthesis is a specialized adaptation typically providing a competitive advantage under low CO 2 availability and/or in hot, dry environments (Sage & McKown, 2006). By contrast, Hawaiian Euphorbia species such as E. remyi grow Euphorbia is thus an interesting model group for understanding the evolution of photosynthetic systems (Sage & Sultmanis, 2016).
In this study, we sequenced seven chloroplast and three nuclear markers to reconstruct the history of radiation in Hawaiian Euphorbia.  (Lorence & Wagner, 1996). Although Hawaiian Euphorbia occurs on all major islands, our samples focused on six of TA B L E 1 Distribution of the 16 named Hawaiian Euphorbia species on the six major Hawaiian Islands. Habitat types are sorted from wetter habitats generally at higher elevations to lower elevation and drier ones, and ages of islands are ordered left to right from older to younger (Koutnik, 1987;Koutnik & Huft, 1990;Lorence & Wagner, 1996;Morden & Gregoritza, 2005). Taxa with an "*" are federally listed as endangered. See Riina and Berry (2016)

| Laboratory procedures
Genomic DNA extraction, plus PCR amplification and sequencing of both ITS and cpDNA followed the protocols in Yang and Berry (2011). A total of seven chloroplast (cpDNA) noncoding regions were sequenced: atpI-atpH spacer, psbB-psbH spacer, psbD-trnT spacer, rpl14-rpl36 spacer, rpl16 intron, trnH-psbA spacer, and the trnL-F region. For the ITS region, sequences with continuous superimposed peaks were excluded. Two of these excluded PCR products, E. celastroides var. kaenana 5840 and E. kuwaleana 5700, were cloned following the protocol of Yang and Berry (2011) to evaluate allelic variation. The second intron of the nuclear lowcopy gene LEAFY and intron of glyceraldehyde 3-phosphate dehydrogenase subunit C (G3pdhC) were PCR amplified and cloned following the protocol in Yang and Berry (2011), except that at least 24 clones from each PCR product were sequenced to recover all copies. Copy-specific primer pairs were designed for both LEAFY and G3pdhC, and at least eight clones were sequenced from each copy-specific PCR reaction (Supporting Information Methods in Appendix S2).

| Phylogenetic analysis
Each of the seven cpDNA and three nuclear data sets were analyzed separately using maximum parsimony (MP) in PAUP* (Swofford, 2003 (Müller, 2006) and were treated as the second character set of the cpDNA matrix.
Two independent runs of four chains each (three heated, one cold), starting from random trees, using a temperature of 0.2, were run for 10 million generations, using the model GTR + I + γ selected by AIC in MrModeltest v2.3 (Nylander, 2004). Trees were sampled every 1,000 generations. Parameters were unlinked between the two partitions except tree topologies. The binary indels were subject to "rates=gamma." A branch length prior "brlenspr=unconstrained:exponential(100.0)" was applied to the nucleotide partition to prevent unrealistically long branches (Marshall, 2010). Diagnostic parameters were visually examined in the program Tracer v1.5 (Rambaut & Drummond, 2007) to verify stationary status. Trees sampled from the first 1 million generations were discarded as burn-in, and the remaining 18,002 trees were used to compute the majority rule consensus (MCC) tree and posterior probability (PP) for each branch of the MCC tree.
Maximum likelihood (ML) analysis was carried out using RAxML v7.4.2 (Stamatakis, 2006), partitioning nucleotides versus indels. The nucleotide substitution model was set to GTR + γ, and 500 rapid bootstrap (BS) replicates were performed, followed by a thorough search for the best tree.

| Cross validation of date constraints and molecular dating using cpDNA
The Hawaiian island chain was formed by the Pacific plate moving northwestward over a fixed hot spot (Carson & Clague, 1995). We assumed that a new island was colonized soon after it emerged (Fleischer, Mcintosh, & Tarr, 1998), and that given the extremely small colonizing population, deep divergence from ancestral polymorphisms in the colonizing population was highly unlikely.
We cross-validated these two assumptions with a preliminary analysis estimating the stem ages of Maui Nui and Hawai`i clades

| Phylogeographic reconstruction
Discrete phylogeographic analysis (Lemey, Rambaut, Drummond, & Suchard, 2009) was used to reconstruct the pattern of dispersal along the island chain from the cpDNA data set. Phylogeographic analysis was carried out in BEAST using two independent continuous-time Markov chains by manually editing the xml file generated by BEAUti from the previous molecular dating analysis following Lemey et al. (2009). Most recent common ancestor of all Hawaiian accessions was set to Kaua`i according to molecular dating results. Convergence diagnostic parameters were visualized in Tracer, and the first 6 million generations were discarded as burn-in.

| Assignment of vegetation types
We categorized coastal strand, scrub, and dry forest habitats as "open vegetation." Open vegetation is either fully exposed or has relatively open canopy coverage. It is generally low in elevation, though the upper elevation limit of lowland dry forest varies from 150 to 1,500 m depending on the island and the aspect of the slope, and the montane dry forests species E. olowaluana occurs in elevation as high as 2,800 m on Hawai`i (Gagné & Cuddihy, 1990;Koutnik & Huft, 1990). Both mesic and wet forests, which generally occur at relatively high elevation, have a closed forest canopy and were categorized as "closed vegetation." Montane bogs, although not protected by a closed forest canopy, are specialized forest openings surrounded by wet or mesic forests and were categorized as closed vegetation.

| cpDNA phylogeny and molecular dating suggested a Kaua`i/Ni`ihau origin of Hawaiian Euphorbia
We obtained sequences of all seven chloroplast noncoding regions from each of the 164 DNA accessions included in this study.
The aligned matrix was 8,278 bp in length (alignment statistics in Supporting Information Table S2 Nui-based clade 2, which is a coastal strand species that occurs on all main islands. We used cpDNA only for dating and phylogeographic analyses to track dispersal via seeds or vegetative fragments. Using island age for molecular dating can potentially be biased by delayed arrival long after island formation, multiple dispersal events, local extinction, and ancestral polymorphism. Another consideration is that at the time Kaua`i formed ca. 5 million years ago (Ma), the adjacent island of Ni`ihau was of similar size and prominence (Price & Clague, 2002 Lerner et al., 2011;Sherrod et al., 2007). Both taxa in the Hawai`i clade, E. multiformis var.
microphylla and E. olowaluana, also occur on Maui Nui (Koutnik, 1987), and it is likely that the "Hawai`i clade" diverged on Maui Nui before dispersing to Hawai`i.

| Distribution of species richness across islands and habitats
The number of overall species per island is highest in O`ahu (10 species), and decreases towards both older (eight on Kaua`i) and younger islands (six on Maui Nui and four on Hawai`i), showing a humped trend. Species that occupy two or more major islands ("widespread" species) were most numerous on Maui Nui, and single-island endemic species were only found on Kaua`i and O`ahu, the two oldest islands, and absent from the two younger island groups (Figure 4a). The species-habitat plot (Figure 4b) showed that "widespread" species tend to occur in open vegetation, while single-island endemics tend to occur under closed vegetation.

| All three nuclear markers had increased copy numbers compared to mainland relatives and low resolution within Hawaiian Euphorbia
Similar to the cpDNA phylogeny, the nuclear ITS tree highly Both the low-copy nuclear genes LEAFY and G3pdhC showed increased copy numbers among Hawaiian taxa compared to outgroup taxa, but the resolution within each copy was low. Four copies of LEAFY were recovered, but only one copy was detected from the known outgroup species (Supporting Information Figure S2.3 in Appendix S2).
Similarly, six copies of G3pdhC were detected in Hawaiian Euphorbia, among which three had a clear association with known outgroup species (Supporting Information Figure S2.4 in Appendix S2).

| Kaua`i origin and dispersal following progression rule from older to younger islands
Our analyses suggest that Hawaiian Euphorbia first colonized Kaua`i or Ni`ihau, then O`ahu, Maui Nui, and finally Hawai`i, generally following the "progression rule" from older to younger islands (Funk & Wagner, 1995;Hennig, 1966), but with at least one dispersal event in the reverse direction through a widespread coastal species.  (Horn et al., 2014). In order to constrain the root, Horn

| Dispersal through open vegetation with in situ origin of species specialized in closed vegetation
Given that all closely related mainland species are from dry and disturbed habitats (Yang & Berry, 2011), the initial colonization A similar pattern of "upslope migration" is also evident in Hawaiian Artemisia (Hobbs & Baldwin, 2013) and in flightless alpine moths in Hawai`i and Maui (Medeiros & Gillespie, 2011). By contrast, in Hawaiian violets a nuclear ITS phylogeny recovered a "dry clade" and a "wet clade," each having species from multiple islands (Havran, Sytsma, & Ballard, 2009). Given that Havran et al. (2009) relied solely on the ITS marker in a group with a complex polyploidy history, it may not have accurately resolved the evolutionary history of the group (Marcussen et al., 2012). Analyses of the Hawaiian endemic plant genus Schiedea using ITS + ETS + morphology (Wagner, Weller, & Sakai, 2005) and a more detailed assessment using eight plastid and three nuclear loci (Willyard et al., 2011) F I G U R E 3 Maximum clade credibility (MCC) tree recovered from BEAST phylogeographic analysis in Hawaiian Euphorbia. Node labels are mean ages, and node bars are 95% highest posterior density (HPD) intervals. Outgroups are not shown.  Habitat type: 1 = coastal strand 2 = scrub 3 = dry forest 4 = mesic forest 5 = wet forest 6 = bog showed a pattern of multiple shifts to both dry and wet habitats from a presumed mesic ancestor.

| Dynamic history of dispersal and habitat shift with island building and erosion
In addition to our findings of progressive dispersal along the island chain and movements toward closed habitats on individual islands, the timing of the volcanic island formation and erosion adds another dimension to the dynamics of dispersal and habitat shift (Lim & Marshall, 2017;Whittaker, Triantis, & Ladle, 2008). This is evident from the hump-shaped curve of the total species number versus island age relationship typical in volcanic island systems (Lim & Marshall, 2017;Whittaker et al., 2008), As islands become older and eroded, the number of overall species decreases.

Both dispersal ability and habitat specialization in Hawaiian
Euphorbia appear to be associated with seed characters. Hawaiian Euphorbia most likely arrived from North America via tiny seeds that adhered to birds through a mucilaginous seed coat (Carlquist, 1966(Carlquist, , 1980Price & Wagner, 2004). A survey of mucilaginous seed coats across Euphorbia sect. Anisophyllum (Jordan & Hayden, 1992) showed that it is present in most mainland species as well as in Hawaiian Islands, also lacks a mucilaginous seed coat. Instead, it is able to float in sea water (Carlquist, 1980), which likely explains its coastal distribution and offers an alternative dispersal mechanism besides sticking to birds. In contrast, neither E. celastroides (widespread) nor E. clusiifolia (Kaua`i endemic) appear to have floating seeds (Carlquist, 1966). In addition to the difference in dispersal ability between species of different vegetation types, endemic species, such as E. clusiifolia and E. rockii have seeds 2-3 times larger in diameter compared to typical widespread species (Koutnik, 1987). Such larger, nonsticky, nonbuoyant seeds may have enhanced seedling survival in forest understory with reduced dispersal ability.

| Radiation of Hawaiian Euphorbia with gene tree nonmonophyly and extensive discordance between cpDNA and nuclear ITS markers
Our results from three nuclear markers supported the results from a previous analysis (Yang & Berry, 2011) that Hawaiian Euphorbia originated from a single colonization following allopolyploidy. Previous results from cloning another nuclear low-copy gene, EMB2765, found three copies in Hawaiian Euphorbia. Two of the copies were associated with different mainland lineages, while a third copy had close relatives unresolved. With increased taxon sampling in this study, both nuclear low-copy genes cloned, LEAFY and G3pdhC, also had increased copy numbers in Hawaiian Euphorbia compared to mainland species. Two of the four copies detected in LEAFY and three of the six copies detected in G3pdhC were not associated with outgroup taxa previously identified using ITS and chloroplast markers. In addition to the increased copy numbers in nuclear low-copy genes, the elevated number of superimposed peaks recovered in the nuclear ribosomal ITS region compared to mainland relatives is also consistent with an allopolyploid ancestor for the Hawaiian Euphorbia.  (Figures 2 and 3). For example, Euphorbia degeneri is restricted to coastal beach habitats and is characterized by distinctive round and upwardly folded sessile leaves (Figure 1c; Koutnik, 1987).   Figure 5).
Despite being highly variable, E. celastroides is still morphologically distinctive, with entire, distichous leaves that are oblong to obovate in shape (Figure 1e). It can be distinguished from the vegetatively similar E. multiformis, also a widespread species, by its erect fruits and appressed cyathial glands, as opposed to recurved fruits and protruding glands in E. multiformis (Koutnik, 1985).
In addition to highly nonmonophyletic gene trees with deeply divergent placements, we also found evidence for more recent hybridization events. Euphorbia multiformis var. microphylla 5622 and 5624 were both collected at the Pohakuloa Training Area of Hawai`i, and they share an almost identical cpDNA haplotype with E. olowaluana accessions from the same area (Figures 2 and   3). In the ITS phylogeny, however, neither E. multiformis var. Certain infraspecific taxa in Hawaiian Euphorbia are geographically and morphologically distinctive enough that it is sometimes unclear whether separate species should be recognized (Koutnik, 1985(Koutnik, , 1987Koutnik & Huft, 1990). We decided not to recircumscribe species based on our results. First, some of the most morphologically homogenous taxa, such as E. degeneri and E. celastroides var. kaenana, are also some of the most polyphyletic.
Second, with the highly dynamic allelic variation and low resolution in ITS, we do not have sufficient information to reconcile the discordance between ITS and cpDNA. Given that with seven cpDNA markers we had only moderate support for the overall relationships in Hawaiian Euphorbia, it will require high-throughput sequencing with a larger number of additional markers to tease apart incomplete lineage sorting and ancient and/or recent hybridization as factors contributing to the tangled relationships among the Hawaiian Euphorbia species.

| CON CLUS IONS
Our analyses of chloroplast regions suggest that after initial colo-

CO N FLI C T O F I NTE R E S T
None declared.

DATA ACCE SS I B I LIT Y
DNA sequence data were deposited in GenBank (Supporting Information Appendix S1). Alignment files in NEXUS format are provided in Supporting Information Appendix S3.