|Home | About | Journals | Submit | Contact Us | Français|
The Hawaiian silversword alliance (Asteraceae) is one the best examples of a plant adaptive radiation, exhibiting extensive morphological and ecological diversity. No research within this group has addressed the role of geographical isolation, independent of ecological adaptation, in contributing to taxonomic diversity. The aims of this study were to examine genetic differentiation among subspecies of Dubautia laxa (Asteraceae) to determine if allopatric or sympatric populations and subspecies form distinct genetic clusters to understand better the role of geography in diversification within the alliance.
Dubautia laxa is a widespread member of the Hawaiian silversword alliance, occurring on four of the five major islands of the Hawaiian archipelago, with four subspecies recognized on the basis of morphological, ecological and geographical variation. Nuclear microsatellites and plastid DNA sequence data were examined. Data were analysed using maximum-likelihood and Bayesian phylogenetic methodologies to identify unique evolutionary lineages.
Plastid DNA sequence data resolved two highly divergent lineages, recognized as the Laxa and Hirsuta groups, that are more similar to other members of the Hawaiian silversword alliance than they are to each other. The Laxa group is basal to the young island species of Dubautia, whereas the Hirsuta group forms a clade with the old island lineages of Dubautia and with Argyroxiphium. The divergence between the plastid groups is supported by Bayesian microsatellite clustering analyses, but the degree of nuclear differentiation is not as great. Clear genetic differentiation is only observed between allopatric populations, both within and among islands.
These results indicate that geographical separation has aided diversification in D. laxa, whereas ecologically associated morphological differences are not associated with neutral genetic differentiation. This suggests that, despite the stunning ecological adaptation observed, geography has also played an important role in the Hawaiian silversword alliance plant adaptive radiation.
In many actively evolving groups, species and infraspecific taxa (i.e. subspecies and varieties) may represent incipiently diverging lineages (Darwin, 1859; Shaw, 2002; Manier, 2004; Gamble et al., 2008; Mulcahy, 2008). However, as many organisms exhibit phenotypic plasticity and reproductive isolation is not absolute, taxa recognized on the basis of geography, ecology or morphology may not represent genetically distinct evolutionary units (Zink, 2004; Sotuyo et al., 2007). This problem is exacerbated in groups that have undergone recent adaptive radiations, because the degree of genetic divergence among sister taxa may be low, whereas ecological and morphological divergence can be large (Schluter, 2000; Hughes and Eastwood, 2006; Givnish et al., 2009; Meudt et al., 2009).
The Hawaiian silversword alliance (Asteraceae) offers a prime example of a plant adaptive radiation and the difficulties associated with untangling the evolutionary patterns of a rapid diversification. Over the past 5 Myr (Baldwin and Sanderson, 1998), the 33 members of the Hawaiian silversword alliance have undergone a rapid radiation, colonizing every major island in the Hawaiian archipelago, exhibiting a diverse array of morphological characteristics and occupying almost every available habitat in the archipelago (Carr, 1985; Baldwin and Robichaux, 1995). Despite this exceptional diversity, phylogenetic and population genetic studies have had difficulty resolving evolutionary relationships within species (Friar et al., 2001, 2007, 2008; Lawton-Rauh et al., 2007; Remington and Robichaux, 2007) and among species (Witter and Carr, 1988; Baldwin, 1997; Friar et al., 2006). When examining the relationship of taxa within the group, two drivers of diversification are apparent: (1) ecologically based isolation among sister taxa (Robichaux et al., 1990; Baldwin and Robichaux, 1995; Friar et al., 2006), and (2) geographically based isolation with sister taxa being isolated between islands or distinct geographical regions within islands (Carr, 1985; Friar et al., 2001). Previous studies have focused largely on the ecological drivers of diversification within this group, but the role of geography has not previously been addressed.
Genetic data provide a tool for dissecting the evolutionary relationships of recently derived taxa, independent of morphology and geography. Studies that employ multiple data types (Schaal et al., 1998; Schaal and Olsen, 2000; Gamble et al., 2008; Pineiro et al., 2009) are particularly suited to understanding the complexities of population-level evolutionary processes and relationships that are indicative of both historical and contemporary diversification. However, conflict among data types, particularly DNA sequence data from different genomes, has been widely documented (e.g. Rieseberg and Soltis, 1991; Birky, 2001; Funk and Omland, 2003) raising questions about whether any specific gene tree is truly representative of recent evolutionary history (Maddison, 1997; Barraclough and Nee, 2001; Degnan and Rosenberg, 2009).
In plants, plastid (or chloroplast) capture, in which the plastid genome has a unique evolutionary history relative to the nuclear genome, has long been recognized as a potential source of phylogenetic error (Rieseberg and Soltis, 1991), but its impact at the species level remains largely unaddressed (Tsitrone et al., 2003; Okuyama et al., 2005; Sotuyo et al., 2007). Additionally, as plastid DNA is maternally inherited in most flowering plants (Birky, 2001), phylogeographical reconstructions based solely on plastid DNA generally document the pattern of seed dispersal, which may be independent of long-distance gene flow via pollen. Consequently, significant questions exist as to whether evolutionary groups recognized on the basis of plastid DNA represent true evolutionary entities or artefacts of the evolutionary history of a plastid genome with a relatively small complement of genes.
Dubautia laxa (Asteraceae) is a widely distributed member of the Hawaiian silversword alliance, occurring on five of the six major islands of the Hawaiian archipelago (Carr, 1985). Partially due to its wide distribution, D. laxa harbours extensive infraspecific variation with four currently recognized subspecies and as many as 12 varieties being recognized previously (Carr et al., 2003). Two subspecies have multi-island distributions, D. laxa subsp. hirsuta occuring on Kaua‘i, O'ahu and Lana'i and D. laxa subsp. laxa occurring on O'ahu, Moloka'i and Maui, and two subspecies, D. laxa subspp. bryanii and pseudoplantaginea, are endemic to the Ko'olau Range on O'ahu (Table 1; Fig. 1). The apparent centre of diversity of D. laxa is O'ahu, where all four subspecies occur and populations of subspecies bryanii, laxa and pseudoplantaginea are found in close physical proximity in the Ko'olau Range. D. laxa subsp. hirsuta is relatively isolated on O'ahu, where it is restricted to, and is the sole subspecies occuring in, the Wai'anae Range. Among the subspecies of D. laxa, there is considerable morphological and habitat variation with subsp. hirsuta exhibiting the most distinctive characteristics (Table 1). When considering the biogeographical relationship of populations distributed on Lana'i, Moloka'i and Maui, we will use the term Maui Nui, which refers to the single large island composed of these smaller islands that began to separate about 0·6 Mya (Price and Elliott-Fisk, 2004).
Despite the current taxonomic treatement of D. laxa (Carr, 1985), it is unclear if the recognized subspecies represent unique evolutionary entities or if the ecological and morphological diversity within this group is a product of recurrent local adaptation or phenotypic plasticity. Further complicating our understanding of this species is the widespread occurrence of hybridization among members of the Hawaiian silversword alliance (Carr and Kyhos, 1981, 1986; Carr, 1985; Carr et al., 2003). The available data on hybridization within the alliance indicate that the barriers that do exist to prevent hybridization among species are ecological in nature rather than genetic (Friar et al., 2006; Carr, 1985). This finding can be extended to D. laxa in which subspecies are differentiated according to morphology and habitat, but are frequently not geographically separated, allowing potential gene flow among subspecies. This is particularily apparent in the Ko'olau Range on the island of O'ahu where three of the four subspecies of D. laxa are sympatric and subspecies are frequently physically closer to populations of other subspecies than they are to consubspecific populations.
In the present study, we use nuclear microsatellite and plastid DNA sequence data to investigate genetic differentiation among populations, subspecies and islands to determine if genetic data support the recognition of divergent infraspecific taxa and how distinct lineages relate to geographical distribution. Furthermore, the unique distribution of D. laxa allows us to pose questions about the impact of geography on diversification at multiple spatial scales. This research aimed to address three questions related to the role of geography in the diversification of D. laxa. (1) Are sympatrically distributed subspecies on the island of O'ahu distinct evolutionary units, suggesting that ecological adaptation is the primary force maintaining distinct lineages? (2) Is isolation among islands sufficent to lead to divergence both within and among subspecies with multi-island distributions? (3) What has been the impact of the pattern of island colonization on diversification in this group? Together, these questions allow us to address the role of geography in diversification within the Hawaiian silversword alliance.
The subspecies of Dubautia laxa have been distinguished on the basis of capitulescence number, receptacular bract length, flowering time, leaf morphology, leaf pubescence, whole plant architecture and habitat (Carr, 1985). A matrix of these characters is provided in Table 1.
D. laxa was sampled from 13 populations on four islands (Fig. 1; Table 2). Only a single subspecies occurs on each island except for O'ahu where D. laxa subsp. hirsuta (Dlh) is restricted to the Wai'anae Range and D. laxa subsp. bryanii (Dlb), D. laxa subsp. laxa (Dll) and D. laxa subsp. pseudoplantaginea (Dlp) are found in the Ko'olau Range, frequently in close physical proximity. Thirty individuals were sampled from each population, except for Dlh Ka'ala and Dll Moloka'i, which are represented by 29 and 27 individuals, respectively (Table 2). All leaf tissue samples were shipped on ice to Rancho Santa Ana Botanic Garden, where they were deposited in a –80 °C freezer until extraction. DNA extractions were performed using a modified CTAB protocol (Friar, 2005) with the addition of 1 % (w/v) Caylase (Cayla Inc., Toulouse, France) and 100 µL Phytopure Nucleon Resin (Amersham Biotechnologies, Amersham, UK).
For plastid DNA sequencing, three individuals of Dubautia plantaginea from the island of Lana'i were sampled as an outgroup. D. plantaginea is not thought to be a close relative of D. laxa, but it is grouped in the same section of the genus, it co-occurs with D. laxa on several islands and has been thought to hybridize with this species in nature. The following members of the Hawaiian silversword alliance were also included in the phylogenetic analysis using data collected for other projects in the E. Friar lab: Argyroxiphium caliginis, A. grayanum, A. kauense, A. sandwicense, Dubautia arborea, D. ciliolata, D. latifolia, D. linearis, D. menziesii, D. platyphylla, D. reticulata, D. scabra and D. waianapanapaensis and Madia sativa (a California tarweed).
Five microsatellite loci developed for the Hawaiian silversword alliance were used as described in Friar et al. (2000). The microsatellite loci were Ap3MS1, MKMS2, MKMS3, MKMS4 and 42-2. Each primer population pair was optimized for annealing temperature and MgCl2 concentration. PCR amplifications were carried out in 15-μL volumes, which included 3 µL genomic DNA (at 5 ng μL−1), 0·6 µL of each primer (at 10 p.p.m.), 0·9 µL dNTPs (at 2·5 mm), 0·5 U Taq polymerase (Promega, Madison, WI, USA) and 1·5 µL 10× amplification buffer (0·1 m Tris-HCl, 0·5 m KCl, 5 %, v/v, glycerine, pH 8·3). Optimized amplification temperatures and MgCl2 concentrations ranged from 52 to 60 °C and from 3 to 5 mm, respectively (data not shown). Forty cycles of amplification were carried out in an MJ Research, Inc. (Waltham, MA, USA) PTC-100 thermocycler, with 1 min denaturing at 94 °C, 1 min annealing at primer-specific temperatures and 30 s extension at 72 °C. Amplification products were visualized on 1 % agarose gels to verify fragment size.
Once amplifications were optimized, microsatellite loci were amplified using primers labelled with the fluorescent dyes 6-FAM or HEX (Applied Biosystems, Foster City, CA, USA) for analysis on an Applied Biosystems 3100 genetic analyser. When possible, fluorescently labeled amplified products were multiplexed. All amplification products were co-loaded with a ROX-350 size standard according to the manufacturer's specifications (Applied Biosystems). Electrophoretic results were analyzed using GENESCAN software (version 3.7.1, Applied Biosystems).
PCR was employed to amplify the non-coding plastid psbA-trnH intergenic spacer and rpl16 intron. PCR amplifications were carried out in 20 μL volumes, which included 2 µL genomic DNA (at 10 ng μL−1), 1 µL of each primer (at 10 pmol), 1 µL dNTPs (at 2·5 mm), 1·5 U Taq polymerase (Promega) and 2 µL 10× Promega Taq polymerase buffer with MgCl2; 0·2 µL bovine serum albumin (at 10 mg μL−1) was added to problematic samples. Annealing temperatures were varied from 52 to 56 °C. Amplifications were carried out in an MJ Research, Inc. PTC-100 thermocycler, with initial denaturing at 94 °C for 4 min, followed by 35 cycles of 1 min of denaturing at 94 °C, 1 min annealing at primer-specific temperatures and 3 min extension at 72 °C. The annealing temperature was increased by 0·5 °C per cycle to a maximum of 72 °C. The 35 cycles were followed by a final 10 min extension at 72 °C. Amplification products were visualized on 1 % agarose gels to verify fragment size. Primer pairs used for the PCR were F71 (5′-GCTATGCTTAGTGTGTGACTCGTT-3′, Jordan et al., 1996) and R1516 (5′-CCCTTCATTCTTCCTCTATGTTG-3′, Kelchner and Clark, 1997) for the rpl16 intron and psbAF (5′-GTTATGCATGAACGTAATGCTC-3′) and trnHR (5′-CGCGCATGGTGGATTCACAATC-3′) for the psbA–trnH intergenic spacer (modified from Sang et al., 1997).
Amplified PCR products were purified using PEG precipitation prior to direct sequencing (Johnson and Soltis, 1995). Cycle sequencing was performed on an MJ Research PTC-100 thermocycler, using BigDye Terminator v3.1 (Applied Biosystems). Purified PCR products were subjected to 35 cycles of amplification, with 30 s denaturing at 94 °C, 15 s annealing at 50 °C and 4 min extension at 60 °C. The amplification primers for the two regions were used as forward and reverse primers for sequencing. Due to the size of the rpl16 intron, the internal primers F543 (5′-TCAAGAAGCGATGGGAACGATGG-3'; Butterworth and Wallace, 2004) and rpl16IR (5′-ATTAATGGAGAAGCTATGG-3′; Prince and Kress, 2006) were used to ensure complete coverage of the fragment. Labelled products were purified by Sephadex column cleaning and then sequenced using an Applied Biosystems genetic analyser.
Each population was characterized by the observed number of alleles (AO), the effective number of alleles (AE), observed heterozygosity (HO), expected heterozygosity under Hardy–Weinberg expectations (gene diversity; HE) and the inbreeding coefficient (FIS), calculated as FIS = 1 – HO/HE. One-tailed t-tests and F-tests were performed to determine significant differences in means and variances between pairs of populations and subspecies using StatView 5·0.1 (SAS Institute, Cary, NC, USA). Bonferroni corrections were used to correct for multiple comparisons (Rice, 1989). Tests for fit to Hardy–Weinberg expectations and pairwise linkage disequilibrium were performed using Arlequin ver. 2·000 (Schneider et al., 2000). Genetic differentiation between populations, subspecies and islands was assessed using an analysis of molecular variance (AMOVA; Excoffier et al., 1992).
The Bayesian analysis STRUCTURE ver. 2.3.1 was used to assess population structure and ancestry (Pritchard et al., 2000; Falush et al., 2003). Burn-in and run lengths of 50 000 replicates were used for all analyses. As gene flow is expected among populations, the admixture model using correlated allele frequencies was implemented. The analyses including all sampled individuals used the subspecific designation of each population as a prior incorporated into the STRUCTURE algorithm (Hubisz et al., 2009). Subsequent analyses examining subsets of the sampled individuals did not use any information about subspecific designation as a prior. The number of inferred populations, K, ranged from 1 to three more than the actual number of populations included in the analysis. As the log-likelihood values obtained by STRUCTURE continued to increase, albeit slightly, with increasing values of K, the method presented by Evanno et al. (2005) was used to determine the ‘true’ number of populations. This method examines the rate of change in log-likelihood values between successive K values over 20 replicates to obtain ΔK.
Primer regions were trimmed and sequences were edited in Sequencher v. 4.2.1 (Gene Codes Corp., Ann Arbor, MI, USA). Edited sequences were aligned manually using Se-Al v2·0a11 (Rambaut, 1996). GenBank accession numbers for each dataset are as follows: psbA–trnH IGS: EU341877–EU341889, GU226075–GU226130; rpl16 intron EU341890–EU341902, GU226131–GU226186.
The plastid regions were combined into a single concatenated dataset for all analyses. Nucleotide diversity (π, Nei, 1987), population mutation parameter (qw, Watterson, 1975), number of segregating sites (S) and number of haplotypes (H) were calculated by population and subspecies for the combined plastid DNA dataset using DNASP version 4·0 (Rozas et al., 2003). The number of haplotypes was also calculated by hand after coding gaps larger than 1 bp (see below).
Sequence gaps larger than 1 bp were coded using the simple indel coding method (Simmons and Ochoterena, 2000; Simmons et al., 2001) as implemented by the program GapCoder (Young and Healy, 2003). Optimal maximum-likelihood (ML) settings were determined using likelihood ratio tests of multiple DNA substitution models as implemented by Modeltest v. 3·6 (Posada and Crandall, 1998). ML phylogenetic trees were constructed using PAUP v. 4·0b10 (Swofford, 2003). All phylogenetic trees are rooted with the California tarweed Madia sativa. Additionally, a minimum-spanning haplotype network was computed using TCS v. 1·18 (Clement et al., 2000) with gaps treated as a fifth state and the parsimony criterion set to 95 %.
A total of 386 individuals representing 13 populations of D. laxa were analysed for five microsatellite loci. Locus Ap3MS1 was the most diverse, exhibiting 19 alleles across all populations, with a maximum of 14 alleles within the Dlh Kilohana population and a minimum of six alleles within the Dlh Kaala population (data not shown). Across all populations, loci MKMS2, MKMS3 and MKMS4 had four, five and five observed alleles, respectively. Locus 42-2 was fixed for all individuals except in the Dlb Laie and Dlb Poamoho populations, which each had one heterozygous individual with the same rare allele.
Locus Ap3MS1 showed significant deviation from expected Hardy–Weinberg frequencies for all populations except Dlp Manana. Locus MKMS2 also showed significant deviation from expected Hardy–Weinberg frequencies for all populations except Dll Moloka'i, Dll Kukui and Dll Hanaula. Dlh Lana'i showed significant deviations from expected Hardy–Weinberg frequencies for MKMS3. Both Ap3MS1 and MKMS2 showed a deficiency of heterozygotes and dimorphic allele size distributions.
Tests for pairwise linkage disequilibrium among loci showed significant disequilibrium among loci for most populations. Only Dlh Kilohana, Dlh Wekiu and Dlh Kaala had no significant linkage disequilibrium among loci. Significant linkage was observed between Ap3MS1 and MKMS2 in Dlb Poamoho, Dlb Manana, Dll Konahuanui and Dlh Lana'i, between Ap3MS1 and MKMS3 in Dlp Manana, between Ap3MS1 and MKMS4 in Dlb Laie, Dlb Manana and Dll Moloka'i, between Ap3MS1 and 42-2 in Dlb Poamoho, between MKMS2 and MKMS3 in Dll Konahuanui, between MKMS2 and MKMS4 in Dll Moloka'i, and between MKMS3 and MKMS4 in Dlb Laie, Dll Poamoho, Dll Kukui and Dll Hanaula. Due to the apparent randomness associated with the observed linkage between loci, it is likely that these results are due to deviations from Hardy–Weinberg equilibrium within populations (Schneider et al., 2000).
The observed number of alleles (AO), expected number of alleles (AE), observed heterozygosity (HO), expected heterozygosity (HE) and inbreeding coefficient (FIS) are shown for all populations in Table 3. Observed heterozygosity ranged from 0·05 at Dlh Wekiu to 0·37 at Dlb Laie. Grouped by subspecies, Dlh had significantly lower observed heterozygosity and a higher inbreeding coefficient than subspecies Dlb and Dll. No significant differences between heterozygosity values and variances were found among populations within subspecies. The inbreeding coefficient within populations ranged widely, but was >0·50 in all Dlh populations. Due to the significant differences between Dlh and the other three subspecies, the populations were divided into two groups, the Laxa group, containing all populations of Dlb, Dll and Dlp, and the Hirsuta group, containing all populations of Dlh. The division of D. laxa into two groups is strongly supported by plastid sequence data reconstructions presented below.
AMOVA was used to assess the distribution of genetic variation on multiple taxonomic and spatial scales. When the AMOVA was applied to sympatric members of the Laxa group on O'ahu, 4·8 % of the variation was partitioned among subspecies, 8·8 % among populations within subspecies and 86·4 % within populations (Table 4). When AMOVA was calculated for Dlh populations, 46·1 % of the variation was partitioned among islands, 3·2 % among populations within islands and 50·7 % within islands (Table 5). When the AMOVA was applied only to the Laxa group, 14·3 % of the genetic variation was partitioned among islands, 10·5 % among populations within islands and 75·2 % within populations (Table 6).
Distributions of mean L(K) and ΔK values generated from STRUCTURE are shown in Fig. 2. When all populations are included, there is moderate support for dividing D. laxa into the Laxa and Hirsuta groups. The Laxa group exhibited the largest ΔK value at K = 2 (Fig. 2F), whereas the Hirsuta group exhibited the largest ΔK value at K = 3 (Fig. 2E). Cluster assignment is shown for D. laxa in its entirety and the Laxa and Hirsuta groups for the maximum ΔK value (Fig. 3). Within the Laxa group, Dlb from the northern Ko'olau Range and the three Dll populations from Maui Nui are unambiguously assigned to two distinct clusters. The remaining populations of the Laxa group exhibit significant admixture. The admixed populations are located in the central and southern portions of the Ko'olau range on O'ahu, indicating that gene flow is occurring among these populations. Within the Hirsuta group, the three inferred clusters correspond to the island of occurrence. The division of the Hirsuta group by island indicates that there is limited gene flow among islands within this group.
Sequence lengths of the rpl16 intron ranged from 891 to 1029 bp with an aligned length of 1030 bp. A 133-bp deletion was shared between all members of the Laxa group. Intrapopulation sequence variation for this region was observed only in Dll Poamoho. Sequence fragment lengths for the psbA–trnH region ranged from 489 to 538 bp with an aligned length of 560 bp. Population-specific insertion/deletion events were observed for Dlh Lana'i, Dll Konahuanui and D. plantaginea of 6, 15 and 5 bp, respectively. A 20-bp insertion was shared by all members of the Laxa group. The combined data set resolves 11 haplotypes with gaps excluded and 15 haplotypes with gaps coded (Table 7). Only three of the 13 populations sampled, Dlh Kilohana, Dll Konahuanui and Dll Poamoho, exhibited any intrapopulation sequence variation for the combined data set. The Hirsuta group had the highest level of nucleotide diversity (π = 0·00128) followed by subspecies Dlb (π = 0·00078) and the Laxa group (π = 0·00077).
ML analysis was conducted using the General Time Reversible model plus gamma-distributed rate variation across sites, as selected by Modeltest. ML reconstructions resolved two well-supported clades of the D. laxa plastid DNA lineage (Fig. 4). One group, which will be referred to as the Hirsuta group, consists of all populations of Dlh and Dll Konahuanui. The second lineage, which will be referred to as the Laxa group, includes all populations of subspecies Dlb and Dlp and all populations of subspecies Dll excluding Dll Konahuanui. Sequences from D. plantaginea were most similar to the Hirsuta group. With the inclusion of other members of the Hawaiian silversword alliance, D. laxa is resolved as polyphyletic (Fig. 4). The Hirsuta group is resolved as monophyletic and groups with old island members of Dubautia, represented here by D. plantaginea and D. latifolia and all members of Argyroxiphium. Observed haplotypes within the Hirsuta group coalesce within islands. The Laxa group is resolved as paraphyletic, with the young island Dubautia section Railliardia clade nested within it. There is no clear taxonomic or phylogeographical resolution within the Laxa group.
The minimum-spanning haplotype network further illustrates the divergence between the Laxa and Hirsuta groups (Fig. 5). The two groups are separated by a minimum of nine inferred mutational steps, and the Hirsuta group is separated from D. plantaginea by three inferred mutational steps. The haplotype network resolves considerable geographical structure within the Hirsuta group with no haplotypes shared among populations or islands. The Laxa group exhibits limited taxonomic or geographical structure with the single sampled population of Dlp containing a unique haplotype and one Dll population from Maui Nui containing a unique haplotype (see legend to Fig. 5).
Speciation is fundamentally a population-level process that occurs when populations diverge based on morphological, ecological, reproductive and/or genetic characteristics, leading to the emergence of new evolutionary entities (Coyne and Orr, 2004). Considerable variability exists in the application of the taxonomic or theoretical concept of species. When examining genetic data, there is likely to be a continuum below the rank of species with some populations exhibiting complete independence, others being only minimally divergent and many actively exchanging alleles. Investigations that examine the divergence of infraspecific variants within a species can provide information about how the speciation process proceeds and which factors are most important to promote sustained divergence.
In the current study, two distinct lineages were resolved within D. laxa using both plastid DNA and nuclear microsatellites. When the data were analysed independently for the lineages, significant structure among populations and islands was observed. Based on these results we feel that D. laxa should be divided into two groups. The Laxa group is composed of subspecies Dlb, Dll and Dlp. These subspecies were not significantly genetically distinct, exhibit overlapping morphological and ecological traits, and have largely sympatric distributions. The Hirsuta group is composed solely of subspecies Dlh and occurs in a unique habitat, and is genetically distinct from and has an allopatric distribution relative to the other members of D. laxa. One population of Dll, Konahuanui, appears to be an infraspecific hybrid, with the nuclear genome of the Laxa group and the plastid genome of the Hirsuta group.
The level of plastid DNA divergence between the Laxa and Hirsuta groups is striking. Based on inferred mutational steps among plastid DNA haplotypes (Fig. 5), the two groups are separated by a minimum of nine mutational steps with no observed intermediate haplotypes. Furthermore, phylogenetic analyses including species throughout the Hawaiian silversword alliance showed that the plastid genome of both groups are more similar to other members of the Hawaiian silversword alliance than they are to each other (Fig. 4). The most comprehensive phylogenetic studies of the alliance have been conducted by Baldwin and collaborators using both nuclear ribosomal DNA (Baldwin and Robichaux, 1995; Baldwin and Sanderson, 1998) and plastid DNA (Baldwin et al., 1990, 1991), but they did not sample widely within species. All phylogenetic studies using nuclear DNA have found D. laxa to represent a cohesive evolutionary unit, sharing a common ancestor (Baldwin and Robichaux, 1995; Baldwin and Sanderson, 1998; M. E. McGlaughlin, unpubl. res.), which contrasts sharply with the current plastid DNA reconstructions in which D. laxa is polyphyletic. There are clear similarities to Baldwin's plastid DNA phylogeny (Baldwin et al., 1991) in both of the major clades resolved in the current study, but the substantial divergence between the Hirsuta and Laxa groups is an unexpected result.
Although not as distinct as the plastid DNA data, microsatellite analyses show consistent divergence between the Laxa and Hirsuta groups. When all populations of D. laxa were used in a single analysis, using the Bayesian assignment procedure implemented in STRUCTURE, there was a clear clustering of populations belonging to the Laxa or the Hirsuta groups, with very limited admixture between those groups (Fig. 3). The unity of members of the Hirsuta group is striking because populations span >300 km on three islands and Dlh populations on O'ahu and Lana'i are geographically isolated from other Dlh populations and are physically much closer to populations of the other three subspecies. When the Hirsuta group was analysed singly, three clusters that conformed to the island of occurrence were resolved. The two Kaua'i Dlh populations, Kilohana and Wekiu, were not substantially differentiated from each other, but the Kilohana population was moderately admixed with Dlh Kaala (on O'ahu) and slightly admixed with Dlh Lana'i. Within the Laxa group, two population clusters were resolved that were not admixed representing the geographical extremes of its distribution, Dll populations from Maui and Moloka'i and Dlb populations from the northern Ko'olau Range. The remaining Laxa group populations exhibited significant admixture between the two pure cluster types.
Despite significant morphological and ecological diversity, sympatrically distributed populations of the Laxa group on the island of O'ahu do not show any clear patterns of genetic differentiation. This is supported by AMOVA, which found <5 % of the microsatellite variation was distributed among subspecies (Table 4), the extensive admixture observed in the STRUCTURE analysis (Fig. 3) and the sharing of plastid DNA haplotypes among subspecies (Fig. 6). Together, these results indicate that sympatrically distributed subspecies have not diverged genetically, at least not detectably. The lack of resolution among subspecies could be due to the maintenance of ancestral polymorphisms, continuing gene flow or recent divergence. Sympatric speciation remains an actively debated topic in speciation research (Sites and Marshall, 2003; Fitzpatrick et al., 2008; Hendry et al., 2009), with most examples being constrained to uncommon geographical circumstances (Coyne and Orr, 2004; Savolainen et al., 2006; Nosil, 2008). Our data do not support the suggestion that infraspecific taxa in D. laxa represent incipient lineages diverging based on localized ecological adaptation. Rather, these data support the generally held idea that the disruption of gene flow among sympatric populations is difficult to achieve.
Considerable genetic divergence was observed between populations occurring on different islands within both the Laxa and Hirsuta groups. Although Dll is distributed on both O'ahu and Maui Nui, a greater level of divergence is observed between the two than among subspecies within the Laxa group. However, the dominant plastid DNA haplotype found on O'ahu (haplotype A) is found in every Dll population on Maui Nui, indicating that the dispersal between O'ahu and Maui Nui occurred relatively recently and there has not been substantial divergence between the two. Within the Hirsuta group, almost 50 % of the microsatellite variation is distributed among islands and no plastid DNA haplotypes are shared among islands. This high level of divergence is probably due to the age of the Hirsuta group and the limited amount of inter-island gene flow. Furthermore, these data indicate that geographical isolation among islands is a significant impediment to gene flow, particularly in the form of seed dispersal.
The level of genetic variation contained within populations can also offer insight into evolutionary history and how populations are interacting genetically. D. laxa, like most members of the Hawaiian silversword alliance, is believed to be self-incompatible (Carr et al., 1986), thereby increasing the importance of the level of genetic variability contained within populations. The genetic diversity data further support the distinction among the Laxa and Hirsuta groups. Microsatellite variability in the Laxa group was high in all sampled populations. Populations of the Hirsuta group had significantly lower observed heterozygosity and exhibited significantly higher levels of inbreeding than those of the other three subspecies. In relation to the other subspecies, Dlh has the smallest mean population size and the greatest distance among populations and it is found in a rare habitat, bog margins. At the other end of the spectrum is subsp. bryanii, which exhibited the highest observed levels of genetic diversity, is found in large populations, in close physical proximity to consubspecific populations and populations of other subspecies (<5 km apart) and on exposed ridges affording the greatest potential gene flow by generalist pollinators.
The observed microsatellite heterozygosities for Dlb, Dll and Dlp are within the range of values reported from natural populations of the closely related genus Argyroxiphium (mean HO = 0·24; Friar et al., 2000, 2001) and generally higher than values reported for species in Dubautia section Railliardia (mean HO = 0·17; Friar et al., 2006, 2007; E. A. Friar, unpubl. res.). The differences in genetic variation between D. laxa and members of section Railliardia are most likely due to the younger age of the Railliardia taxa, which occur on islands with a maximum age of 1·76 Myr, although some populations do exhibit similar levels of variation. For all subspecies, the expected heterozygosity is higher than values reported from allozyme data for D. laxa (HE = 0·07) and other members of the silversword alliance (mean HE = 0·07; Witter and Carr, 1988).
The plastid DNA sequence data provide information about genetic variability on a different time scale relative to the nuclear microsatellites due to the slower mutation rate of the plastid genome. The difference between the Laxa and Hirsuta groups was particularly pronounced for the plastid DNA data. The number of haplotypes resolved for the two groups are comparable despite a larger sampling of individuals and populations belonging to the Laxa group (24 vs. 15; Table 7). The Laxa group is characterized by low levels of nucleotide diversity and a dominant haplotype (haplotype A) occurring in 62 % of the populations. The low levels of plastid DNA diversity suggest that the Laxa group has undergone a historical population bottleneck, possibly when the group originated, and has a more recent origin. By contrast, the Hirsuta group has a much higher level of nucleotide diversity and no haplotypes are shared among populations. Furthermore, two of the four populations of the Hirsuta group exhibit multiple plastid DNA haplotypes when gaps are coded.
The combination of nuclear microsatellite and plastid DNA sequence data provides extensive data to untangle the pattern of dispersal of D. laxa through the Hawaiian Archipelago. Based on the plastid DNA diversities and the phylogenetic placement of the Hirsuta group, it appears that D. laxa originated on the island of Kaua'i. At a later date members of the Hirsuta group dispersed from Kaua'i to O'ahu, establishing the populations currently found in the Waianae Range. The Hirsuta group then dispersed again to Lana'i. The subdivision of the Hirsuta group by islands is supported by both data sets and AMOVA analyses. The lack of shared plastid DNA haplotypes among islands within the Hirsuta group suggests that there was a single seed dispersal event to colonize each island. Furthermore, the fact that the Dll Konahuinui population contains a Hirsuta group plastid DNA haplotype indicates that there has been seed dispersal between the Waianea and Ko'olau mountain ranges on O'ahu. Denser sampling in the vicinity of the Konahuinui population would further clarify the frequency of seed dispersal between the two mountain ranges.
The plastid genome of the Laxa group appears to have a complex history with two possible evolutionary scenarios. First, the plastid DNA of the Laxa group could have originated from the Hirsuta group on O'ahu. This scenario would require substantial divergence of the Hirsuta group plastid DNA leading to the origin of the Laxa group. The shared ancestry and recurrent gene flow via pollen between the two groups could have maintained the relative cohesion of the nuclear genome. A second scenario is that the plastid DNA of the Laxa group represents a plastid capture event from an unsampled Dubautia species. A plastid capture scenario has also been suggested to explain evolutionary patterns observed in old island members of Dubautia (Baldwin, 1997) and in D. scabra (Baldwin, 1997; Baldwin et al., 1990; Friar et al., 2008), a member of the Railliardia clade. Determining the likely scenario for the origin of the Laxa group could be aided by additional sampling of Dubautia species from Kaua'i and the inclusion of other regions from the plastid genome. However, given the recently documented case of plastid capture in D. scabra (Friar et al., 2008) and the degree of divergence among the D. laxa plastid lineages, we feel that a plastid capture event involving an unsampled Dubautia species is the most likely scenario.
Therefore, the Laxa group most likely originated on O'ahu, diverging from existing Dlh populations. This finding is supported by the high level of diversity for both plastid DNA and microsatellites on O'ahu, plastid DNA haplotypes that are shared between O'ahu and Maui Nui populations, and nuclear phylogenetic reconstructions that find members of the Laxa and Hirsuta groups as sister taxa (Baldwin and Robichaux, 1995; M. E. McGlaughlin, unpubl. res.). However, as the exact origin of the Laxa group is unclear, it is possible that it originated on Kaua'i and then dispersed to O'ahu. Once completely diverged, the Laxa group dispersed from O'ahu to Maui Nui. The current distribution could have resulted from a single seed dispersal event from O'ahu to Maui Nui followed by short-range dispersal between Moloka'i and West Maui, or two or more dispersals from O'ahu. All populations on Maui Nui share the dominant plastid DNA haplotype found on O'ahu, but the Dll Kukui population also contains a rare plastid DNA haplotype that is not observed in any other population. The tight clustering of the Maui Nui Dll in the Bayesian analyses could be interpreted to indicate a single colonization from O'ahu, or high levels of gene flow that have homogenized the nuclear genome of the Maui Nui populations. The dispersal scenarios presented here follow the basic progression rule that has been suggested to occur with other Hawaiian taxa (Funk and Wagner, 1995), under which species gradually move from older to younger islands. Furthermore, a single collection of Dll has been made from the youngest mountain range on Maui, Haleakala (Carr, 1985), but this population could not be located when the region was surveyed for this project.
The phylogenetic reconstructions presented here also provide valuable information about how the plastid genome has been transferred among Hawaiian silversword species colonizing younger islands. It is clear that the Laxa group plastid DNA is the progenitor of the Railliardia clade. Within the combined lineage of the Laxa group and the Railliardia clade, there is a clear pattern of plastid DNA dispersal from older to younger islands, with the most derived members of the Railliardia clade occurring on Hawaii, the youngest island, as concluded from previous analyses of plastid DNA restriction site variation (Baldwin et al., 1990). The rarity of inter-island seed dispersal inferred for both the Hirsuta and the Laxa groups is therefore consistent with interpretations for the Railliardia clade and for the silversword alliance in general (Baldwin and Robichaux, 1995). But what do these data tell us about species formation in the Hawaiian silversword alliance? It is clear from nuclear DNA data for D. laxa and members of the Railliardia clade (Friar et al., 2006, 2007, 2008; E. A. Friar, unpubl. res.) that most members of Dubautia lack barriers to introgression. This potential for introgression among species leads to the conclusion that plastid DNA may not be as valuable for indicating boundaries between species as for providing a source of biogeographical evidence. It follows that extending the plastid DNA data set to include all members of the Hawaiian silversword alliance could lead to a detailed understanding of the plastid coalescent processes in endemic island taxa.
The data presented here clearly illustrate the need to incorporate multiple data types to best understand the evolutionary history of species. Although the nuclear and plastid data sets are largely congruent, the degree of divergence among the Laxa and Hirsuta groups within the two data sets is considerable. Members of the Laxa group, subspecies Dlb, Dll and Dlp, were not clearly distinguishable based on the plastid DNA data. However, the microsatellite data suggest that there are pure populations of subspecies Dlb in the western portion of the Ko'olau range (Laie and Poamoho) and of subspecies Dll on Maui Nui (Kukui, Hanaula and Moloka'i), but that all other populations are admixed between these genetic types. This leads us to suggest that Dlh should be elevated to the rank of species, and D. laxa should be recognized as containing two subspecies, Dlb and Dll, that intergrade throughout most of the Ko'olau Range on O'ahu.
Our data shed light on the importance of geographical separation to reinforce local adaptation and population divergence within the Hawaiian silversword alliance. Consistent genetic differentiation was only observed among allopatric populations, whether it was among Hirsuta or Laxa group populations on different islands. This indicates that although there is considerable ecological and morphological variation among D. laxa populations, the available genetic data do not support the suggestion that sympatric populations are undergoing incipient genetic divergence. It should be noted that divergent selection for particular traits could be occurring, but this was not resolved here due to our usage of neutral genetic markers.
We thank L. M. Prince, J. T. Columbus, R. Robichaux, V. Soza, K. Wood and T. Anderson for laboratory and field assistance. Collection permits, access to preserves and logistical support was generously provided by the Hawaii Department of Forestry and Wildlife, Hawaii Natural Area Reserve System, Hawaii State Parks, Maui Land and Pineapple Company, The Nature Conservancy of Hawaii, Moloka'i and Maui Programs, Schofield Barracks Army Environmental Office, O'ahu National Wildlife Preserve, Alexander Baldwin Company, East Maui Irrigation and the Bishop Museum. This research was funded by the following awards to M.E.M.: the Cynthia Lee Smith Botany Award, the Andrew W. Mellon Foundation, the Howard and Phoebe Graduate Program Award, the Goldhamer Scholarship Award and the Rancho Santa Ana Botanic Garden Alumni Award. Sampling of outgroup taxa for the phylogenetic analyses was supported by a grant to E.A.F. from the W. M. Keck Foundation and Rancho Santa Ana Botanic Garden. Any opinion, finding, and conclusions or recommendations expressed in this material, are those of the authors and do not necessarily reflect the views of the National Science Foundation.