A Complex Evolutionary History in a Remote Archipelago: Phylogeography and Morphometrics of the Hawaiian Endemic Ligia Isopods

Compared to the striking diversification and levels of endemism observed in many terrestrial groups within the Hawaiian Archipelago, marine invertebrates exhibit remarkably lower rates of endemism and diversification. Supralittoral invertebrates restricted to specific coastal patchy habitats, however, have the potential for high levels of allopatric diversification. This is the case of Ligia isopods endemic to the Hawaiian Archipelago, which most likely arose from a rocky supralittoral ancestor that colonized the archipelago via rafting, and diversified into rocky supralittoral and inland lineages. A previous study on populations of this isopod from Oʻahu and Kauaʻi revealed high levels of allopatric differentiation, and suggested inter-island historical dispersal events have been rare. To gain a better understanding on the diversity and evolution of this group, we expanded prior phylogeographic work by incorporating populations from unsampled main Hawaiian Islands (Maui, Molokaʻi, Lanaʻi, and Hawaiʻi), increasing the number of gene markers (four mitochondrial and two nuclear genes), and conducting Maximum likelihood and Bayesian phylogenetic analyses. Our study revealed new lineages and expanded the distribution range of several lineages. The phylogeographic patterns of Ligia in the study area are complex, with Hawaiʻi, Oʻahu, and the Maui-Nui islands sharing major lineages, implying multiple inter-island historical dispersal events. In contrast, the oldest and most geographically distant of the major islands (Kauaʻi) shares no lineages with the other islands. Our results did not support the monophyly of all the supralittoral lineages (currently grouped into L. hawaiensis), or the monophyly of the terrestrial lineages (currently grouped into L. perkinsi), implying more than one evolutionary transition between coastal and inland forms. Geometric-morphometric analyses of three supralittoral clades revealed significant body shape differences among them. A taxonomic revision of Hawaiian Ligia is warranted. Our results are relevant for the protection of biodiversity found in an environment subject to high pressure from disturbances.


Introduction
The Hawaiian Islands are well known for their rich biodiversity and high rate of endemism [1].Their remoteness, representing the world's most isolated major archipelago, along with the progressive formation of these islands, are considered crucial for the striking diversification observed in several Hawaiian terrestrial organisms, which include: the Hawaiian Drosophila [2][3][4], the silversword alliance [5], succineid land snails [6,7], honeycreeper birds [8,9], and others [10][11][12][13].In contrast, endemism in Hawaiian marine invertebrates is strikingly lower than that in Hawaiian terrestrial organisms [14], and with the exception of intertidal Cellana limpets [15], there are no documented marine radiations within the Hawaiian Archipelago.Such disparity suggests that opportunities for allopatric differentiation within the archipelago have been very limited in the marine realm, while abundant in the terrestrial realm.Populations of non-vagile invertebrates endemic to specific coastal patchy habitats located at the interface between sea and land, however, have the potential to be highly isolated and, thus, show elevated levels of allopatric differentiation [16].
The Hawaiian endemic shrimp Halocaridinia rubra, which inhabits anchialine coastal pools, shows evidence of withinand between-island divergence and is comprised of multiple highly divergent lineages [17,18].An additional interesting case of diversification in a coastal patchy habitat of the archipelago is that of oniscidean isopods in the genus Ligia, which most likely arose from a rocky supralittoral ancestor that arrived to the Hawaiian Archipelago via rafting, and diversified into rocky supralittoral and inland (hereafter, terrestrial) lineages.High levels of genetic differentiation were detected among populations of Hawaiian Ligia from different localities in the islands of Kauaʻi and Oʻahu [19], but populations from other islands have not been studied.Therefore, further phylogeographic analyses of Hawaiian Ligia including populations from previously unsampled islands are needed to better understand the biodiversity and evolution of this group.
The genus Ligia has a worldwide distribution and includes 42 currently recognized species [20][21][22][23][24][25][26]; most of which are restricted to rocky supralittoral areas [19,27].Eight species, however, are strictly terrestrial, inhabiting montane habitats of tropical regions [19,22], and are believed to derive from supralittoral forms [28].Supralittoral Ligia exhibit morphological, physiological, and behavioral characteristics that are intermediate between ancestral marine and fully terrestrial isopods [29].Some of the biological characteristics of supralittoral Ligia confer them extremely low vagility [30], such as: direct development (lack a planktonic larval phase, as all peracarids); active avoidance of the open sea (they remain in the area between the exposed intertidal and the supralittoral); extremely low desiccation tolerance (a reason for which they stay close to the water line and are most active at night); limited motility underwater and on sandy shores (rendering them highly vulnerable to predators in these environments).
The biology of supralittoral Ligia severely constrains the movement of these isopods outside the rocky beaches they occupy, effectively isolating populations.This is reflected in the striking radiations of supralittoral Ligia reported in different regions of the world, with extraordinarily high levels of allopatric genetic differentiation, even between localities separated by few kilometers ( [19,[30][31][32] and unpublished data).These observations, as well as those by Hurtado et al. [16], challenge earlier suggestions that supralittoral isopods are highly dispersive species, based on their common presence in beaches around the world [33].Phylogeographic patterns of Ligia in different regions have been shaped by past tectonic events [30], environmental factors, such as sea surface temperature [31], as well as oceanic dispersal events, probably through rafting (unpublished data).
Taiti et al. [19] investigated whether the origin of terrestriality in L. perkinsi populations from the Hawaiian islands of Kauaʻi and Oʻahu occurred as a single event (i.e., L. perkinsi from both islands constitute a monophyletic group sister to a lineage of L. hawaiensis) or as two independent events, one in each island (e.g. two clades each showing reciprocal monophyly of L. hawaiensis and L. perkinsi from the same island).They conducted Maximum Parsimony phylogenetic analyses with sequences of two mitochondrial genes (COI and 16S rDNA), and included individuals of L. perkinsi and L. hawaiensis from Kauaʻi and Oʻahu.They obtained strong support for the monophyly of the Ligia lineages endemic to the Hawaiian Archipelago (also observed in [30]), which were divided into three main clades: one comprised of the L. perkinsi from Kauaʻi; another clade comprised of L. perkinsi from Oʻahu; and a third clade comprised of L. hawaiensis individuals.The L. hawaiensis clade was divided into three lineages: one comprised of the Kauaʻi individuals and the other two comprised of the Oʻahu individuals.The two main L. perkinsi clades were paraphyletic (L.perkinsi from Oʻahu was sister to a clade of L. perkinsi from Kauaʻi + L. hawaiensis), thus, the results were inconclusive as to whether a single or two origins of terrestriality occurred.In addition, they observed high divergences among populations of L. hawaiensis, implying long-standing isolation among them, and with phylogeographic patterns suggesting inter-island dispersal events have been rare throughout the history of the Hawaiian endemic Ligia lineages [19].
Herein, we expanded on Taiti et al.'s [19] previous work by incorporating populations from previously unsampled main Hawaiian Islands (i.e., Maui, Molokaʻi, Lanaʻi, and Hawaiʻi), increasing the number of gene markers, which include nuclear genes, and applying more current phylogenetic approaches.Sampling across all main Hawaiian Islands enables a better understanding on the diversity and evolution of endemic Hawaiian Ligia.Specifically, we asked: (1) whether the younger islands (i.e., Maui, Lanaʻi, and Molokaʻi, all with an age ~ 1.3 My, and Hawaiʻi with an age of 0.4 My; ages from [37]) harbor highly divergent L. hawaiensis lineages, as observed in the older islands (i.e., Kauaʻi [5.1 My] and Oʻahu [3.7 My]); (2) whether evolution of Ligia in the Hawaiian Islands followed a pattern consistent with the progression rule (i.e., lineages from older islands are basal to those from younger islands), a back dispersal pattern (lineages from younger islands colonized the older islands), or an unresolved and/or highly stochastic pattern (indicative of a complex evolutionary history, probably with frequent inter-island dispersal); and (3) whether the new data shed light on the origin of terrestriality in Kauaʻi and Oʻahu (i.e., a single origin or independent origins in each island).Lastly, we incorporated geometric-morphometric analyses to test for differences among three highly divergent supralittoral lineages and determine whether these methods can be used for their discrimination.

Sampling
Our molecular dataset included twenty-four L. hawaiensis populations (i.e., localities) from across the main Hawaiian Islands and four L. perkinsi populations from Kauaʻi and Oʻahu (Figure 1A, Table S1).The collection of these specimens did not require a permit from the Department of Land and Natural Resources of Hawaii, and these isopods are not considered endangered or protected species.We also included publicly available sequences for L. hawaiensis and L. perkinsi (Table S1).We were unable to collect L. hawaiensis in the islands of Niʻihau and Kahoʻolawe, because they are private property and a state reserve, respectively.Populations were sampled by hand and preserved in 70-100% Ethanol.As outgroups, we included specimens from Ligia vitiensis, Ligia occidentalis, and Ligia exotica, as previous research [19,30] and preliminary analyses suggest they are the closest extant relatives to Ligia from the Hawaiian Archipelago.

Molecular methods
We extracted total genomic DNA of Ligia individuals from pleopods/legs using the DNEasy Blood & Tissue kit (Qiagen Inc.), following standard protocol instructions.We PCRamplified a 710-bp fragment of the Cytochrome Oxidase I (COI) mitochondrial (mt) gene for 1-10 individuals per population (Dataset S1) using the primers and conditions published by Folmer et al. [38].A subset of these individuals (essentially one individual per population; see Figure S1) was then amplified and sequenced for three additional mitochondrial genes using previously published primers and conditions: ~490-bp of the 16S rDNA gene (primers 16Sar/ 16Sbr; [39]); ~495-bp of 12S rDNA (primers crust-12Sf/ crust-12Sr; [40]); and a 361-bp fragment of the Cytochrome-b (Cytb) gene (primers 144F/151F and 270R/272R; [41]).We also amplified two nuclear genes for a subset of individuals (1-5 per population, see Figure S1): a ~1,000-bp region of the 28S rDNA gene (primers 28SA/28SB; [42]) and a ~710-bp region of the alpha-subunit of the Sodium Potassium ATPase (NaK; primers NaK forb/NaK rev 2; [43]).PCR-products were cleaned with a mixture of Exonuclease I (New England Biolabs) and Shrimp Alkaline Phosphatase (USB Scientific) and cycle sequenced at the University of Arizona Genetics Core (UAGC).We assembled sequences and removed primer regions using Sequencher 4.8 (Genecodes).None of the protein-coding sequences exhibited premature stop codons or frame shifts, suggesting they are not pseudogenes.

Sequence alignments and mitochondrial phylogenetic analyses
We aligned the ribosomal DNA gene fragments (i.e., 16S rDNA, 12S rDNA, and 28S rDNA) with the MAFFT algorithm [44] assuming the Q-INS-I strategy as implemented in the GUIDANCE server [45].Because of the high divergence among lineages of Ligia (see Results), several regions of ambiguous alignment were observed for these genes.Therefore, we used the GUIDANCE server [46] to estimate confidence scores for each nucleotide position (100 independent alignments based on different bootstrap guide trees were conducted), and removed all positions with a confidence score below 1.00, as well as several positions for which alignments were considered ambiguous (Datasets S2 and S3).We estimated pairwise genetic distances with the Kimura-2-Parameter (K2P) correction (excluding ambiguous sites) in MEGA v5.05 [47] for the COI and the 16S rDNA gene fragments separately.
We determined the most appropriate model of DNA substitution for each mitochondrial gene fragment and the mitochondrial concatenated dataset from among 1,624 candidate models, by evaluating their corresponding likelihood scores on a fixed BioNJ-JC tree, under the Akaike Information Criterion (AIC), corrected AIC (AICc), and the Bayesian Information Criterion (BIC) (Table S2) using jModeltest v2.1.1 [48].The chosen model was used in phylogenetic searches, with two general exceptions.First, when the selected model was not available in a particular software, we applied the next more complex model available (Table S3).Second, as the joint estimation of Γ and I parameters can be problematic (see RAxML manual; and pages 113-114 of [49]), we used the simpler Γ when the chosen model included both Γ and I parameters.We also implemented several partitioning schemes: (a) all positions within a single partition; (b) partitioned by gene; and (c) the best partitioning scheme according to the BIC implemented in PartitionFinder v1.0.0 [50].We used the following parameters in PartitionFinder searches: branch lengths = linked; models = all; model selection = BIC; search = greedy; and a priori partitioning combining each gene and codon position.
We carried out maximum likelihood (ML) searches in RAxML v7.2.6 [51][52][53] and GARLI v2.0 [54].RAxML consisted of 1,000 bootstrap replicates followed by a thorough ML search under the GTR +Γ model run under the Rapid Bootstrap Algorithm, whereas GARLI analyses consisted of 1,000 bootstrap replicates under the appropriate model of evolution identified by jModeltest.All other settings were as default.We calculated majority-rule consensus trees for each analysis with the SumTrees command of DendroPy v3.10.1 [55].
We carried out Bayesian phylogenetic reconstructions in MrBayes v3.1.2[56,57] and Phycas v1.2.0 [58].We implemented polytomy priors [59] in Phycas to alleviate the potential overestimation of posterior probabilities (i.e., "star-tree paradox") known to affect Bayesian approaches [60].We present the number of independent MCMC runs, chains, and generations in Table S3, with all other parameters as default.We determined if Bayesian analyses had reached stationarity by: (a) stable posterior probability values; (b) high correlation between the split frequencies of independent runs as implemented in AWTY [61]; (c) small and stable average standard deviation of the split frequencies of independent runs; (d) Potential Scale Reduction Factor close to 1; and (e) an Effective Sample Size (ESS) > 200 for the posterior probabilities, as evaluated in Tracer v1.5 [62].Samples prior to stationarity were discarded as "burnin" (Table S3).To estimate the posterior probability for each node, we built majority-rule consensus trees of the stationary stage of each run using the SumTrees command [55].S4).Nodes receiving 100% for all methods are denoted with an *.NS: less than 50% node support.Samples examined by Taiti et al. [19] are marked with †. doi: 10.1371/journal.pone.0085199.g001

Nuclear gene analyses
Given the low variation levels observed in both nuclear genes amplified (see Results), we visualized relationships between nuclear alleles on networks constructed using the cladogram estimation algorithm of Templeton et al. [63] as implemented by TCS v1.21 [64].We calculated the 95% most parsimoniously plausible branch connections between alleles, with all other settings as default.

Geometric-morphometric methods
We captured digital images of the dorsal side of L. hawaiensis specimens using QCapture v3.1.2and an Olympus QColor3 digital camera attached to an Olympus SZ61 stereomicroscope.We removed all pereopods (i.e., legs) to ensure specimens laid flat.Dissected pereopods and pleopods were not used for morphometric comparisons.The male pereopods and pleopods, which are fundamental in distinguishing Ligia species, did not show distinct differences among the L. hawaiensis populations [19].During dissections, we determined and noted the sex of each specimen by visually inspecting the endopod of the 2 nd pleopod.Sex was noted as either: male (M); gravid female (F); or juvenile/non-gravid female (J).We characterized body shape by digitizing 27 landmarks (LMs), using TpsDig v2.16 [65], on the periphery of Ligia bodies (Figure 2A).We included landmarks that capture taxonomically informative regions and can be noted unambiguously.For example, we placed landmarks on the medial and the lateral boundaries of the eyes at the body periphery.These landmarks capture the relative size of the eyes and the inter-ocular distance, characters used to distinguish Ligia species [19].We characterized the relative width of body segments and overall body shape, also important in Ligia taxonomy [19,25,27,66,67], by placing landmarks on the lateral posterior tergite tips.Lastly, we captured the shape of the pleotelson, another trait used in Ligia taxonomy [19,25,68], by placing landmarks at its posterior tip and the lateral posterior points.
As the body plan of Ligia is bilaterally symmetric, all but the pleotelson tip LMs are anatomically homologous and should not be treated as independent in statistical analyses.As suggested by Zelditch et al. [69], we reflected and averaged homologous landmarks across the body midline as defined by the pleotelson tip and the midpoint between the medial eye LMs.Corrected landmarks were centered, scaled and rotated, to best align with the consensus, using the method of generalized least squares, and projected to a flat shape space using tpsRelw v1.49[70].We calculated principal components of aligned coordinates to yield orthogonal shape variables, retaining the minimum number of components needed to represent ≥ 95% of the overall variation.These principal components were taken as shape variables to test for differences in shape between lineages, sexes, and sizes.We used the centroid size (the summed square distances of landmarks from the centroid; [71]) as an estimate of body size.

Statistical analyses
We carried out full factorial MANCOVA analyses of shape variables as a function of lineage, sex, size, and all interactions, to discern the meaningful effects on body shape.When interaction terms were not significant, we removed them from the model, in a hierarchical manner, and repeated analyses.We estimated effect strengths by calculating partial eta squared values (η p 2 ), which is the multivariate analog of R 2 in simple regression models [72].We further explored differences between lineages with quadratic Discriminant Function Analyses (DFAs) with equal probability priors.To focus on these between-group differences, we first accounted for continuous predictors by conducting a preliminary MANCOVA and saving residual variation [73].We used successful classification frequency as an intuitive metric of the power of morphological divergence to correctly assign an individual to its genetic lineage based solely on its morphology.All DFA results were validated using leave-one out cross validation (LOOCV).All statistical tests were carried out in JMP v9.0.1, except for η p 2 values, which were calculated in Microsoft Excel® using the E & H matrix output from JMP. Lastly, we visualized shape differences between all main effects by producing thin-plate-spline transformations of LM positions in tpsRegr v1.37 [74], using canonical scores for the clade effect as the predictor variable block [75].

Mitochondrial phylogenetic results
The final concatenated mitochondrial dataset (mt) included 32 individuals from throughout the Hawaiian Archipelago, and five individuals from three Ligia species (L.exotica, L. occidentalis, and L. vitiensis) as outgroups (Dataset S2).We excluded 219 poorly aligned positions (16S rDNA: 113-bp; 12S rDNA: 106-bp), resulting in a final alignment of 1758 characters, 581 of which were parsimony informative (Table S2).Selection criteria in jModeltest did not agree on a single model for the dataset.The Bayesian Information Criterion (BIC) selected a model with three substitution rates (rate matrix: 012010, see jModeltest manual), variable nucleotide frequencies (+F), and both +I and +Γ parameters.Akaike Information Criteria strategies (AIC, AICc) selected a slightly more complex model consisting of four substitution rates (rate matrix: 012313, see jModeltest manual) and +F, +I, and +Γ parameters.Given the low weights observed for these models under the AIC and AICc (Table S2), and that the 95% confidence interval included the BIC selected model, we applied the latter in GARLI analyses.For all other software (e.g.RAxML, MrBayes), we applied the GTR +Γ model instead, as the chosen models cannot be implemented.
Phylogenetic relationships inferred from the mitochondrial dataset are shown in Figure 1B.Node support values for each analysis are provided in Table S4.The endemic Hawaiian Archipelago Ligia clade (i.e., ingroup) was highly supported: 100 Bootstrap Support (BS) and Posterior Probability (PP).Within this clade, we observed three basal lineages with divergences among them between 11.85 and 16.74% COI K2P (Table 1): (1) Clade A (lavender in Figure 1), which is a wellsupported clade (94-99 BS; 100 PP) that included some L. hawaiensis populations from Maui and Hawaiʻi, and represents a new lineage that was not previously identified by Taiti et al. [19]; (2) Lineage B (black in Figure 1), which includes the L. perkinsi population from Oʻahu (B1), and was previously reported by Taiti et al. [19]; and (3) Clade CDEF (blue, green, orange, and red, respectively, in Figure 1; 60-70 BS and 78-98 PP), which contained all L. perkinsi samples from Kauaʻi (Clade C) and the rest of the L. hawaiensis samples from the archipelago (Clade DEF).
Within Clade A, we detected three divergent lineages (maximum COI K2P divergence = 5.88%; Table 1): (1) one found in a single Maui population (A1); (2) another in a single population from western Hawaiʻi (A2); and (3) the last in eastern Hawaiʻi (A3, A4).The analyses suggest the Maui lineage represents the most basal split within Clade A and that lineages from Hawaiʻi form a monophyletic group; support for this relationship, however, was variable (66-85 BS; < 65 PP).

Nuclear gene patterns
We sequenced two nuclear genes for all Study Area lineages with the exception of the L. perkinsi lineage from Oʻahu (B).Multiple attempts to amplify nuclear genes from this population proved unsuccessful.The patterns inferred from the nuclear genes (Figure 3; colors for clades correspond with those in Figure 1) were, in general, consistent with those inferred from the mitochondrial genes.For the NaK gene (Dataset S4), we only observed six alleles, separated by 1-6 steps.For Clade A

Table 1.
Estimates of genetic divergence, as measured by Kimura-2-parameter distances, among the main Ligia lineages in the Hawaii Archipelago and outgroups.

L. exotica L. occidentalis L. vitiensis
Clade A members, we detected two alleles separated by a single step.These alleles were separated from the other alleles by 2-6 steps, which is concordant with the high divergence observed in mitochondrial genes between Clade A and all other lineages.The allele observed for the individual of Clade C (L. perkinsi from Kauaʻi) was divergent from the other alleles by 5-6 steps, also consistent with the mitochondrial results.The NaK results show a closer relationship among members of clades D, E, and F, also congruent with the mitochondrial results.Three alleles were observed for members of these clades, which were separated by only 1-2 steps, with Clade E members from Lanaʻi, Molokaʻi and Maui sharing an allele with a Clade F member from Hawaiʻi (F5); whereas another member of the Clade F from Oʻahu (F1) harbored a unique allele.The Clade D individual from Kauaʻi harbored a unique allele.
For the 28S rDNA gene, we excluded 60 poorly aligned positions, resulting in a final alignment of 967 characters, 31 of which were parsimony informative (Dataset S3).For this gene, we detected thirteen alleles.Seven of these alleles were recovered from Clade E individuals (Figure 3), with the other lineages harboring one or two alleles.Concordant with mitochondrial phylogenetic findings, alleles from different major lineages appeared highly differentiated, with most lineages separated by 10-21 steps.We note, however, that the single allele found in Clade D (Kauaʻi; D1) was separated by only two steps from one of the two alleles observed in Clade F (from the Oʻahu F1 locality).The F1 and D1 alleles were in turn separated by 11 steps from the other Clade F allele (from the Hawaiʻi F5 locality).This pattern could be the result of incomplete lineage sorting or a past hybridization event.Examination of additional informative nuclear markers in individuals from multiple populations per clade is needed to resolve this question.Indeed, multispecies coalescent analyses of numerous unlinked markers are likely needed to resolve relationships with more certainty [76].Nevertheless, as shown by Mateos et al. [77], datasets with three loci, including one with strong phylogenetic signal (i.e., the mitochondrial dataset) and one or two with low phylogenetic signal (e.g.NaK) are not well suited for multispecies coalescent approaches, and different methods lead to different results.Therefore, we consider that at the present stage, the mitochondrial phylogeny represents the most plausible hypothesis.

Geometric morphometrics
We analyzed a total of 84 individuals from six supralittoral localities (D1, E3, E4, E6, F1, and F5), representing three of the major lineages detected for L. hawaiensis and all of the Hawaiian Islands sampled (morphometrics datasets and pictures are available in DataDryad doi:10.5061/dryad.5k56c).We did not include individuals from Clade A due to limited sampling (i.e., < 3 individuals per population).Principal component analysis generated 24 non-zero eigenvectors and the first 11 accounted for 96.2% of the variance, and thus were included in posterior analyses.The full factorial MANCOVA yielded no significance for the three-way interaction term, thus we removed it and repeated the analysis.This simpler MANCOVA model (Table 2) yielded significant results for the main effects of Lineage, Sex, and Size (p < 0.0001; η p 2 > 0.4).One two-way interaction (sex × lineage) was significant (p = 0.02), but was weak (η p 2 = 0.2).Quadratic DFAs of MANCOVA residuals achieved an initial correct assignment of individuals to their lineage in 91.7% of cases.After LOOCV, the successful classification rate dropped to 59.5%, with per-lineage validated correct classification rates as follows: 26.7% for Clade D, 70.6% for Clade E, and 62.9% for Clade F. Individuals from Clade D were more likely to be identified as Clade F (7/15) than to their true lineage (4/15).Most misclassified individuals after LOOCV were assigned to F (17 of 34) and E (14 of 34) (Table 3).

Shape variation in L. hawaiensis
We present thin-plate-spline transformations, which represent the average shape deformation, for the three supralittoral lineages examined (D, E, and F; Figure 2B).Along with the visualizations we present photographs of individual specimens having the greatest canonical axis scores.The three lineages examined differ most prominently in two aspects of shape: the width of the body, relative to the total body length, especially in the mid-body, and head size and conformation, particularly with respect to the distance between the eyes, as measured by the distance between their medial boundaries.We also detected minor differences in the distal-most point of the pleotelson.On average, individuals from Clade D have an oblong-ovate body with a mid-body narrower than individuals from clades E and F, and the distance between the eyes appears to be ~2/3 of the total eye width.Clade E individuals exhibit, on average, an ovoid shape and eyes that appear to be separated by a distance similar or greater than the total eye width, and the distal-most point of the pleotelson protrudes more prominently than in either Clade D or F. Individuals from Clade F, on average, exhibit an overall body shape anteriorly broadened relative to those from Clade D or E; a small head size with less crowding of the eyes than in Clade D; and distance between the eyes appears to be equal to the width of the eyes.Finally Clade F had a less prominently protruding pleotelson.It should be kept in mind that these morphological differences are those due to clade effects estimated in the overall context of the MANCOVA, while controlling for allometry, sex/maturity status, and clade-specific allometry.We also present thin-plate-spline transformations for sex/ maturity categories (Figure S2 A).Differences between the sexes appear to be mostly localized in the anterior end of the body, with females having wider segments than both males and juveniles.On average, juveniles and males appear to have very similar body shapes, but the latter appear to have less convexity of the body around the midline.Shape differences between sizes were also evident, with larger individuals having a more elongated body and smaller head, relative to smaller individuals (Figure S2 B).

Discussion
In this study, we conducted phylogeographic analyses of the Ligia lineages endemic to the Hawaiian Archipelago including a comprehensive sampling of these isopods across the main Hawaiian Islands, which greatly enhances our understanding on the diversity and evolution of this group.Previous genetic work by Taiti et al. [19] on Ligia from the Hawaiian islands of Kauaʻi and Oʻahu reported highly divergent lineages of the terrestrial L. perkinsi and the coastal L. hawaiensis in these islands.By including populations from previously unsampled main Hawaiian Islands, we detected additional lineages of L. hawaiensis and expanded the known range of others.We report the discovery of a new lineage of L. hawaiensis (i.e., Clade A), distributed in Maui and Hawaiʻi, which is highly divergent (> 10% COI K2P) from previously reported Ligia lineages in the Hawaiian Archipelago.We also discovered that the distribution of a lineage (Clade E) previously sampled in a locality in Oʻahu (E1) extends to Molokaʻi, Lanaʻi, and Maui.Similarly, we discovered that Clade F, which was previously identified in Taiti et al. [19] from Oʻahu, contains two additional lineages; one distributed in Maui (F3) and one in Hawaiʻi (F4-F8).Divergence among the three Clade F lineages is > 3% COI K2P.Considering that intra-specific divergences in marine invertebrates are typically < 3% for the same COI fragment used in this study [78], it is likely that some of the divergent Hawaiian Ligia lineages represent cryptic species, thus, the taxonomy of the group needs revision.High levels of genetic differentiation among populations of the Hawaiian Ligia are congruent with studies of Ligia in other parts of the world ( [30][31][32] and unpublished data).Biological characteristics of these isopods severely restrict their dispersal potential, contributing effectively to the isolation of populations, and, in the long-term, to allopatric genetic differentiation [30].Nonetheless, phylogeographic patterns indicate that past dispersal events have been important in shaping the evolutionary history of these isopods in the Hawaiian Archipelago.
The monophyly of the endemic Hawaiian Ligia lineages is well supported, suggesting that evolution of this group likely occurred within the archipelago.Lineages identified as the closest relatives of the endemic Hawaiian Ligia are highly divergent and found in other Pacific localities (unpublished data).Large divergences observed among Hawaiian Ligia lineages also suggest a long evolutionary history for this group.In addition, the phylogeographic patterns observed among Hawaiian Ligia, although not fully resolved, do not support simpler patterns of evolution, such as colonization from older to younger islands (i.e., progressive rule), or vice versa (i.e., from younger to older islands), but rather a complex evolutionary history.Illustrating the complexity of the evolutionary history of the Hawaiian Ligia, the three most basal lineages include one found only in the younger islands (Clade A), a terrestrial lineage restricted to Oʻahu (Clade B), and one that is the most diverse.This latter clade (CDEF) includes supralittoral and terrestrial lineages, and divergent lineages restricted either to the older island of Kauaʻi or to the other islands.Dispersal and local extinctions likely contributed to shape this complex evolutionary history.
Despite the complex evolutionary history of Hawaiian Ligia, some phylogeographic patterns emerge that appear congruent with the geological history of the main Hawaiian Islands.First, Kauaʻi, the oldest of the main Hawaiian Islands, harbors only one endemic L. hawaiensis lineage (Clade D), which is highly divergent from the other lineages.This is consistent with the older geological history of this island and its high degree of isolation, as no overland connections are thought to have existed between Kauaʻi and other Hawaiian islands [37].Second, in contrast to the pattern observed in Kauaʻi, sharing of highly divergent lineages is observed among the other main Hawaiian Islands, suggesting inter-island dispersal among these islands.The geological history of these islands may have provided opportunities for the exchange of colonizers from divergent lineages.Molokaʻi, Maui, Lanaʻi, and Kahoʻolawe, are thought to have existed as a single landmass, the Maui Nui Complex, throughout most of their geological history, first splitting up some ~0.6 million years ago (Mya) and retaining land connections during glacial low sea level stands [79].The Maui Nui complex is thought to have been connected via a land bridge to Oʻahu between 2.2-1.9Mya forming the short lived Oʻahu Nui complex [37,79].These past connections may have facilitated the dispersal of Clade E, restricted to the Maui Nui islands and Oʻahu.The low genetic divergences observed among Clade E populations (< 2.5% COI K2P) may indicate a recent history of isolation among populations of this clade.Dispersal of Clade F across these islands may also have been facilitated by these connections.Third, historically, the oceanic channel separating Maui and Hawaiʻi does not appear to constitute a very effective barrier for inter-island dispersal of coastal organisms that disperse through rafting, as two divergent lineages are shared between Hawaiʻi and Maui (i.e., clades A and F).Remarkably, however, members from Clade E were not found in Hawaiʻi, despite their wide distribution in the Maui Nui islands.
Evolution of terrestriality appears to have occurred early during the diversification of the Hawaiian Ligia, as clades B and C are highly divergent and occupy a relatively basal position in the phylogeny of Hawaiian Ligia.Our results confirm the paraphyly of L. perkinsi, previously observed by Taiti et al. [19].Therefore, from the phylogeographic patterns it is uncertain (1) whether evolution of terrestriality occurred independently in each island (i.e., Kauaʻi and Oʻahu) or (2) whether terrestriality evolved once.Both of the above hypotheses are equally parsimonious assuming a coastal ancestor for the endemic Hawaiian Ligia clade, which we consider most likely, as the closest lineages are coastal and because of the predominantly coastal nature of the genus.In addition, the remoteness of the Hawaiian archipelago makes colonization by a coastal lineage more likely than by a terrestrial one.Under the observed phylogenetic patterns, an independent origin of terrestriality would have required two steps: the evolution of terrestriality along the branches of clades B and C. Conversely, a single origin of terrestriality may only be explained by invoking the adaptation to terrestrial habitats followed by a reversal to the marine habitat (i.e., 2 steps).Taiti et al. [19] considers the independent adaptation to terrestrial habitats in Hawaiian Ligia as the most plausible explanation for the origin of L. perkinsi populations in Oʻahu and Kauaʻi.They consider a shift from terrestrial to seashore conditions unlikely, given that most of the species in the genus occupy supralittoral habitats, and that terrestrial forms appear to have derived from supralittoral forms [28].Similar conditions could have acted in Oʻahu and Kauaʻi that facilitated the independent colonization of freshwater habitats.Nonetheless, not enough is known about the biology of L. perkinsi to discard the possibility of a reversal.This species is often found in the rocky shores of streams [34], with populations in Oʻahu occurring less than 3 Km from shore and at low altitude [80].Furthermore, L. perkinsi exhibits the highest osmoregulatory capabilities observed among Ligia species [81].Post-hoc analyses from Approximately Unbiased (AU) tests implemented in CONSEL [82] rejected the monophyly of L. perkinsi (p = 1 × 10 -6 ), which would have supported a single origin of terrestriality.These tests also rejected the monophyly of L. hawaiensis (p = 2 × 10 -5 ), as well as the monophyly of L. perkinsi from Kauaʻi + L. hawaiensis from Kauaʻi (p = 1 × 10 -5 ).
Similar to the Hawaiian Ligia, the shrimp Halocaridinia rubra, which inhabits anchialine coastal pools, also shows multiple highly divergent lineages, with allopatric distributions [17,18].In contrast with Ligia, however, presence of a highly divergent lineage in multiple islands is rare in the shrimp, with only one case reported among the three islands studied (i.e., Oʻahu, Maui, and Hawaiʻi).A shrimp lineage that was found mainly on the west coast of Hawaiʻi was also found in one locality of Oʻahu, represented by a single haplotype that was very close (only one substitution) to some haplotypes present in Hawaiʻi.Therefore, inter-island dispersal appears to have been more limited in this shrimp.
Morphological differences have been previously reported between L. perkinsi lineages (B, C), but not between L. hawaiensis lineages (D, F) [19].Previous comparisons, however, relied on classic taxonomic characters.Geometricmorphometric approaches have proven useful in identifying differences between otherwise cryptic species in other invertebrate taxa [83][84][85][86], including crustaceans [87,88].They also have been used to test body shape differences at the intraspecific level in isopods [89].Our geometric-morphometric analyses revealed significant differences in body shape among the three highly genetically divergent supralittoral lineages examined (D, E, and F).Rates of correct assignment of quadratic DFAs after LOOCV were low, however, indicating body shape alone is of limited use as a taxonomic diagnostic tool.On average, lineages differ in traits widely used in the taxonomy of Ligia, including relative body width, distance between the eyes, and the protrusion of the distal-most point of the pleotelson [19,25,27,66,67].Interestingly, individuals from Clade D exhibit average differences with respect to the other two clades in traits that were previously used to describe Ligia kauaiensis, such as a narrow oblong-ovoid body shape and shorter distances between the eyes [90].This species, now considered a synonym of L. hawaiensis [20,91], was first described from individuals from the shores of Kalihiwai Bay, Kauaʻi [90], the same location (D1) included in our molecular and morphometric analyses.Therefore, the deep genetic divergence of Clade D (at least 13.57% COI K2P), its discrete geographic distribution, and apparently distinct morphology suggest that L. kauaiensis may be a valid species.
We also detected significant differences in the average overall body shape between sexes, with females exhibiting wider anterior segments than males and juveniles.These differences may be caused by the development of the ventral marsupium (i.e., brood pouch) in females.This structure forms from thoracic sterna and overlapping oostegites prior to copulation in mature females, and was present in all samples identified as females.As we did not observe obvious differences between males and juveniles, and females were classified based on the presence of the marsupium, a temporary structure [92], differences between the sexes in L. hawaiensis may be temporary, and thus, not relevant to the taxonomy of Ligia in the Hawaiian Archipelago.Lastly, we detected an effect of body size on body shape for L. hawaiensis, with larger individuals exhibiting a more elongated body with relatively smaller head than smaller individuals.Allometry has been reported also for Ligia pallasii, where in contrast to our results, larger individuals (usually males) tend to be disproportionality broader than smaller ones [93,94], a pattern that we have also observed in other Ligia from the northern East Pacific (unpublished data).

Conclusions
Our study revealed new lineages and expanded the distribution range of several lineages of supralittoral Ligia in the Hawaiian Archipelago.A previously unknown highly divergent lineage (Clade A) of supralittoral Ligia was found in the islands of Hawaiʻi and Maui.We found that Clade E, a lineage previously reported only from Oʻahu, is widely distributed in the Maui-Nui islands.Clade F, a lineage also previously reported only from Oʻahu, contains two additional moderately divergent lineages with geographically restricted distributions; one in Maui and one in Hawaiʻi.Our results supported the monophyly of all Ligia lineages endemic to the Hawaiian Archipelago, consistent with a single colonization event.Our results, however, did not support the monophyly of all the supralittoral lineages (currently grouped into L. hawaiensis), and/or the monophyly of the terrestrial lineages (currently grouped into L. perkinsi), implying more than one evolutionary transition between coastal and inland forms.The lack of monophyly of currently recognized species and the relatively high genetic divergences of several lineages argues for a taxonomic revision.The phylogeographic patterns of Ligia in the study area are complex, inconsistent with simple evolutionary patterns proposed for the Hawaiian Islands, such as the progression rule.Evidence for multiple historical dispersal events between islands is observed among Hawaiʻi, Oʻahu, and the Maui-Nui islands.In contrast, the oldest and most geographically distant of the major islands (Kauaʻi) shares no lineages with the other islands.Although multiple lineages inhabit several islands, no instances of sympatry at the locality level were detected.Significant body shape differences were observed among three L. hawaiensis clades examined (D, E, and F), one of which was consistent with previous recognition of a separate coastal species in Kauaʻi (i.e., L. kauaiensis).
Finally, our results are relevant for the protection of unique and highly localized biodiversity found in an environment subject to high pressure from human disturbances in the Hawaiian Islands.Destruction and modification of rocky intertidal beaches through activities such as construction and conversion to sandy beaches (e.g. by sand nourishing) can destroy the natural habitat of Ligia and, thus, jeopardize the survival of unique lineages endemic to the Hawaiian Islands.

Figure 2 .
Figure 2. Landmarks (LMs) placement (A) and thin-plate-spline transformations (B) for Ligia hawaiensis (clades D, E, F). A. LMs 1 and 25 represent the posterior margin of the eyes at the body perimeter; LMs 2-11 and 15-24 are the posterior-most point of each segment; LMs 12 and 14 are the lateral posterior points of the pleotelson, whereas LM 13 is the posterior-most point of the pleotelson; LMs 26 and 27 correspond to the inner-most margin of the eyes.B. Thin-plate-spline transformations of LM positions are shown magnified (10 ×) to reach the natural extremes observed in the clades, which are also indicated with images for specimens with the highest probability of being correctly assigned to each clade.

Figure 3 .
Figure 3. Haplotype networks for two nuclear gene fragments for Ligia from the Hawaiian Archipelago.The nuclear markers examined are the alpha subunit of the Sodium-Potassium ATPase (NaK) and the 28S rDNA gene.Colors correspond with those in Figure 1.Locality ID is indicated next to each allele (locality information is provided in Figure 1 and Table 1).Empty circles represent inferred unsampled (i.e., missing) alleles.Numbers within branches indicate number of mutational steps separating haplotypes.The size of circles is proportional to the frequency at which an allele was recovered.Numbers in parenthesis, indicate the number of individuals observed with the corresponding allele.doi: 10.1371/journal.pone.0085199.g003

Table 2 .
Results of multivariate analyses of covariance examining overall body shape in Ligia hawaiensis.