Introduction

Octocorals, sessile cnidarians, are found in marine habitats ranging from intertidal to abyssal waters and are distributed from the Arctic to the Antarctic (Bayer 1961). Gorgonians, octocorals containing a sclero-proteinaceous axis, are ecologically important and the dominant macrofauna on many Caribbean reefs. There can be up to 40 octocoral species on a single Caribbean coral reef (Sánchez et al. 1997, 1998; Sánchez 1999), and octocoral communities produce dense and colorful gardens, which provide three-dimensional structures for many reef dwellers and scenic value for underwater activities (Fig. 1). Despite their importance, evolutionary relationships among Caribbean octocorals have received little attention. In the present study, we present the first phylogenetic analyses of Caribbean octocorals using mitochondrial DNA sequences.

Fig. 1.
figure 1

Plexaura spp., Pseudopterogorgia spp. Caribbean gorgonian octocorals in their natural habitat. Fore-reef terrace (8 m) at Carrie Bow Cay, Belize

Full size image

Historically, shallow-water Caribbean octocorals were placed in the order Gorgonacea (Bayer 1961). The order has been superceded and most of the shallow-water species are now classified in the Holaxonia, which is a nominal suborder of the order Alcyonacea (Bayer 1981). However, the term Gorgonacea is still in use in several databases (e.g. GenBank-taxonomy, NCBI, and UNEP-WCMC database for animals). The term "gorgonian" or "gorgonian coral" is colloquially used throughout the literature to refer to all of the branching octocorals that contain a sclero-proteinaceous axis (gorgonin) with varying degrees of calcification (Bayer 1961; Grasshoff 1999; Sánchez et al., unpublished data). These are primarily members of Holaxonia, but also include taxa from the suborders Calcaxonia and Scleraxonia.

Most holaxonians in the Caribbean have been assigned to the families Plexauridae and Gorgoniidae. Plexaurid colonies are characterized by a thick coenenchyme (the connecting tissue between polyps), canals surrounding the axis, and large, irregularly ornate sclerites sometimes up to several millimeters long (Bayer 1961). Gorgoniidae, on the other hand, have slim branches and small sclerites, usually <0.3 mm long, which are regularly ornate (Bayer 1961; Grasshoff and Alderslade 1997). Both family descriptions include characters that are not clearly defined at the generic level and based on those familial traits some species could be classified in either family. For instance, some descriptions of species in the Plexauridae say "presence or not of a strong polyp sclerites armature" and "loculated axis or not" (Bayer 1961). However, the absence of those characters is associated with gorgoniids. In Gorgoniidae, on the other hand, descriptions include traits such as "tubercles arranged in whorls, but irregular in some species" (Grasshoff and Alderslade 1997), which can correspond to some plexaurids as well. The size of the sclerites and the thickness of the coenenchyme are also quite variable, and there are many intermediate and overlapping subgroups. Species of Muriceopsis (Plexauridae), for instance, have thin coenenchyme and sclerites of intermediate size (Sánchez 2001); Plexaurella spp. (also Plexauridae) have very thick coenenchyme with small (<0.5 mm) sclerites that are regularly ornate. Both Muriceopsis spp. and Plexaurella spp. have surface sclerites that are very different from those of other "typical" gorgoniid or plexaurid species. There is no clear division between Plexauridae and Gorgoniidae, and characters that can discern ancestry are lacking.

The first records of Caribbean reef octocorals go back to Linnaeus (e.g. Gorgonia ventalina Linnaeus, 1758). Numerous species have been described from the fauna, but not until Bayer (1961) was the extent of gorgonian octocoral diversity, nor its taxonomic, geographic, and ecological status, clarified and revised. There have been further additions to the fauna (e.g. Bayer 1991; Lasker et al. 1996; García-Parrado and Alcolado 1997; Sánchez 2001), but no dedicated attempts to review the fauna. Nor is there a complete and robust phylogenetic hypothesis for these octocorals. Kükenthal (1919) presented a phylogeny of some octocorals (Gorgonacea), but it relies upon untenable support (Bayer 1956). Using information from surface sclerites, Bayer (1953) proposed convergent evolution for gorgoniid colony architectures such as sea fans (Gorgonia spp. and Pacifigorgia spp.) and sea leaves (Phyllogorgia spp. and Phycogorgia spp.). Gerhart (1983) presented the first cladistic approach using presence and absence of terpenoids and their biosynthetic pathways in an analysis of 18 gorgonian species. Gerhardt identified classification issues that have not been resolved, such as the grouping of Plexaurella spp. with gorgoniids (Gorgonia spp. and Pseudopterogorgia spp.). Morphological characters have provided consistent phylogenetic information for within-genus cladistic relationships (Eunicea spp.: Sánchez 1998; Muriceopsis spp.: Sanchez 2001). Bayer's and Gerhart's differing hypotheses were based on mostly independent evidence, i.e. sclerites (Bayer) and secondary compounds (Gerhart). The two phylogenies present fundamentally different hypotheses that can be tested. We used coding mitochondrial DNA sequences to examine the relationships among a group of 28 species to specifically test whether Plexaurella spp. group with gorgoniids, and if sea fans (Gorgonia spp. and Pacifigorgia spp.) evolved independently from hypothesized non–sea fan ancestors.

Although a few species have been included in broader phylogenetic studies, molecular phylogenetic analyses of Caribbean gorgonian corals have not been undertaken. France et al. (1996: 16S lsu-rRNA) and Berntson et al. (2001: 18S ssu-rRNA; see also Berntson 1999), working on anthozoans and octocorals, respectively, report poorly resolved phylogenies for groups containing holaxonian (gorgonian) octocorals. A reanalysis of their data, including insertions and secondary structure, has provided some evidence for two distinct clades of branching octocorals (Sánchez et al. in press). However, the resolution using nuclear and mitochondrial ribosomal gene sequences is limited due to sequence conservation. Other multicopy nuclear genes such as the internal transcribed spacers (ITS) between the 28S and 18S rRNA genes are too variable and cannot be aligned among different holaxonian families (authors' unpublished data). Therefore, analyses using different genes are necessary to discern relationships within and between octocoral families. A complete octocoral mitochondrial genome for Sarcophyton glaucum has been sequenced (Beaton et al. 1998; Pont-Kingdon et al. 1998), providing a variety of potential regions for phylogenetic analyses. For instance, octocorals contain a unique mitochondrial gene, msh1, which codes for a homologue of a DNA mismatch repair protein (MutS) found in eubacteria (Pont-Kingdon et al. 1995; Culligan et al. 2000). France and Hoover (2001) identified more variation in msh1 among octocoral species than in commonly sequenced mitochondrial genes such as cytochrome oxidase I (France and Hoover 2002) and 16S rDNA (France et al. 1996).

In the present study we develop a molecular phylogenetic hypothesis for the major groups of Caribbean gorgonian octocorals. The particular goals of the study were: (1) to determine the phylogenetic signal in partial sequences from the mitochondrial coding genes NADH-dehydrogenase [subunits 2 (ND2) and 6 (ND6)] and mutS homologue (msh1); (2) to reconstruct phylogenetic relationships among 28 species of gorgonian octocorals from the Plexauridae and Gorgoniidae, including species from 12 genera (including Muriceopsis and Plexaurella); and (3) to compare and map major morphological characters onto the molecular hypotheses in order to identify differences and/or congruence between molecular hypotheses and commonly used taxonomic classifications.

Materials and methods

Octocoral genomic DNA

Most of the material was collected during 2000 in the Bahamas at depths of 1–30 m using SCUBA (Table 1). In the Caribbean, we sampled at least two species per genus (except Pacifigorgia and Pseudoplexaura). Since Pacifigorgia elegans was the only species of this genus found in the Caribbean, P. stenobrochis from the eastern Pacific was included in order to verify sequences at the generic level. Additional material was obtained from colleagues (see Table 1 and "Acknowledgements"). The examined material was preliminarily identified using the taxonomic work and key of Bayer (1961) based on dissection of sclerite layers, digestion of the organic matter in sodium hypochlorite, and observations of sclerites under a compound microscope (100×). Voucher specimen (Table 1) identifications were verified by direct comparison with material deposited in the National Museum of Natural History (Smithsonian Institution), which included type material and/or material identified by Bayer (1961), Lasker et al. (1996), and Sánchez (1998, 2001).

Table 1. Studied species, collection sites and dates, museum voucher specimen number (USNM numbers unless otherwise indicated, National Museum of Natural History, Smithsonian Institution, Washington, D.C., USA), and GenBank accession numbers for the three studied genes
Full size table

Small pieces (<0.5 cm3) of tissue were fixed in 95% ethanol with daily changes over 5 days and then kept at −20°C until extraction. For genomic DNA extraction we used the 2× CTAB protocol with phenol/chloroform phases and precipitation with ethanol (Coffroth et al. 1992). DNA was resuspended in 10–40 μl of TE buffer (according to pellet size) and kept at −20°C until amplification. For gene amplification we used primers designed by McFadden et al. (in press) for ND6 and ND2 (ND6-1487F: 5′-TTTGGTTAGTTATTGCCTTT-3′; ND3-2126R: 5′-CACATTCATAGACCGACACTT-3′; 16S-647F: 5′-ACACAGCTCGGTTTCTATCTACAA-3′; ND2-1418R: 5′-ACATCGGGAGCCCACATA-3′) and by France and Hoover (2001, 2002) for the 5′ end of msh1 (ND42599F: 5′-GCCATTATGGTTAACTATTAC-3′; Mut3458R: 5′-TSGAGCAAAAGCCACTCC-3′). All the primers were anchored in an adjacent mitochondrial gene to prevent amplification of genes from nuclear or symbiont (zooxanthella) genomes (Fig. 2). All zooxanthella and blank controls were negative except for ND2, for which amplifications sometimes produced a blurry ~1,000 bp band from the symbiont that did not occur in the presence of octocoral DNA. We added ~20 bp M13 tails to all primers, a step that was necessary for sequencing with the Li-Cor Global IR2 DNA sequencing system (primers Li-Cor, M13 reverse: 5′-GGATAACAATTTCACACAGG-3′; M13 forward: 5′-CACGACGTTGTAAAACGAC-3′). Template DNA for sequencing was obtained from the combination of four PCR (polymerase chain reaction) reaction tubes containing 28 μl total (volume completed with ddH2O) with 1 μl of DNA template (1:10 dilutions of genomic DNA extractions), ~3 U Taq polymerase, 3 μl of 10× buffer [10 μM Tris-HCl (pH 8.3), 50 μM KCl, 0.001% gelatin, and 3.1 μM MgCl2], 0.21 μM dNTPs mix, and 0.16 μM of each primer. These reaction conditions yielded the highest template quality for cycle-sequencing. The PCR conditions for all sequenced genes and species were one initial period of 2.0 min at 94°C; followed by 35 cycles of 1.5 min at 94°C, 1.5 min at 58°C, and 1.0 min at 72°C; and a final extension step for 5.0 min at 72°C. The PCR product was precipitated with an equal volume of PEG (20% w/v PEG 8000 in 2.5 M NaCl), briefly (15 min) incubated at 37°C, and centrifuged at 14,000 rpm (15 min). Supernatant was discarded, and the DNA pellet was washed with 125 μl of 80% ethanol, followed by 5 min of centrifugation; the supernatant was again discarded, and the pellet was air dried. The pellet was resuspended in 10 μl of ddH2O at 65°C for 5 min and finally run in a 1.5% 1× TAE-agarose gel (the TAE containing 1/10 of the normal EDTA). The DNA band in the gel was excised using a razor and separated from the agarose using a cellulose acetate membrane (Fisher-Corning Costar Spin-X columns).

Fig. 2.
figure 2

Diagram of sequenced portion of octocoral mitochondrial genome and primers used (modified from Beaton et al. 1998)

Full size image

PCR products were cycle-sequenced in both directions using a Global IR2 DNA sequencing system (Li-Cor). We used M13-labeled primers and the Epicentre-Li-Cor protocol (SequiTherm EXCEL II DNA sequencing kit—LC for 66 cm gels). Consensus sequences were obtained automatically by assembling the two complementary DNA chromatograms. Sequences with several ambiguities at sites other than the ends were repeated from the initial amplification step. Two to four different specimens from four species (Pseudopterogorgia elisabethae, Muriceopsis flavida, Pterogorgia citrina, and Pseudopterogorgia bipinnata) were sequenced to control for procedural error. No differences were found within species other than at the ends that were trimmed for the analyses. Several Eunicea spp. also had identical sequences, which indicated a low probability of procedural error. GenBank accessions for the sequenced genes and species are provided in Table 1.

Phylogenetic reconstruction

Phylogenetic analyses were conducted using coding regions and two different matrices: nucleotides and amino acids. Assembly of the three genes (ordered ND6–ND2–msh1), matrix editing, translation, and alignment were done using BioEdit (Hall 1999) and CLUSTALW (Higgins et al. 1996). Non-coding regions were removed from the analysis. Corrections for the genetic code were made according to Beaton et al. (1998), and the mold, protozoan, and coelenterate mitochondrial code (NCBI genetic code 4) was used to translate nucleotide sequences. The complete alignments for the coding regions and amino acid positions were deposited in TREEBASE (University at Buffalo-SUNY, http://www.treebase.org; accessions S833, M1334, M1335). Both maximum-parsimony and maximum-likelihood approaches were used to analyze both nucleotide and amino acid sequences. Maximum-parsimony analyses were carried out in PAUP* for Windows (Swofford 2002). Nexus matrices were translated from NBRI/PIR format using ProSeq 2.8 (Filatov 2001). Alcyonium digitatum sequences (McFadden et al., unpublished data) were used as the outgroup for the analyses. We initially included six outgroup species from different orders of octocorals, but they had quite high divergence from Caribbean gorgonian octocorals, including the presence of indels in coding regions. Since the analyses did not change with inclusion of one versus six outgroups, we decided to use only the closest species that did not add indels to the alignment. Analyses of the other outgroup sequences and indels will be published elsewhere.

Tree reconstruction with maximum parsimony was conducted using the branch-and-bound search algorithm in PAUP* with default parameters. As measures of phylogenetic content the skewness of the tree-length distribution of 1,000 randomly generated trees in PAUP* was assessed with the g1-test (Hillis and Huelsenbeck 1992). To examine the noise (e.g. homoplasy) in the data sets (Fu and Murphy 1999), the permutation tail probability (PTP) was also determined using PAUP*. Character re-sampling by means of non-parametric bootstrapping (10,000 replicates) was used to visualize the branch/node support of the trees in 50%-majority-rule consensus in PAUP* (e.g. Hillis and Bull 1993).

We used several likelihood approaches for reconstructing phylogenetic relationships. Initially, we searched for the best-fit models of DNA evolution using MODELTEST (Posada and Crandall 1998), which evaluates the goodness of fit of the data alignment by a series of hierarchical likelihood-ratio tests (Posada and Crandall 2001). Using the likelihood settings recommended by MODELTEST, a maximum-likelihood tree was reconstructed using a heuristic search and random addition of sequences in PAUP*. Only 100 bootstrap replicates were done using the maximum-likelihood tree due to logistic/time constraints. Additionally, using the best-fit model from MODELTEST, Bayesian-likelihood analysis for the estimation of phylogeny was assessed using the program MrBayes (Huelsenbeck and Ronquist 2001) and the recommended block command for coding regions (Hall 2001). Finally, the program Tree-Puzzle was used to choose the best-fit model of evolution for amino acid substitution and phylogeny reconstruction using a quartet-puzzling, maximum-likelihood approach (Strimmer and von Haeseler 1996). Nucleotide/amino acid maximum-likelihood hypotheses were compared to trees reflecting the taxonomic classification scheme using the Shimodaira–Hasegawa test in PAUP* (S–H test; Goldman et al. 2000).

Morphological characters

Several morphological characters from the axis and sclerites of the studied octocorals were examined in order to map them onto the molecular phylogenies. Characters from the axis and base mineralization were mostly obtained from Bayer and McIntyre (2001). To extend those data some species were prepared using similar methods and examined under scanning electron microscopy (SEM). SEM analyses were conducted at the Instrumentation Center (Hitachi S-800), SUNY at Buffalo, and the National Museum of Natural History, SEM laboratory (AMRAY 1810), Smithsonian Institution, Washington, D.C. The sclerites were examined using the protocols of Bayer (1961, 1992) and Sánchez (2001). First we examined the complete sclerite structure (80–1,500×), and then we scanned details of warts/decorations or irregularities of the sclerites (500–5,000×). At least two sclerites of each sclerite type were observed from each species. Carbon double-sided tape was used to mount samples, and gold and carbon was used for SEM coating. Only characters that were amenable to discernable qualitative and discrete coding were retained from the initial survey. Direct measurements of prepared sclerites were made with a compound microscope in order to have at least ten measurements per species. Maximum sclerite lengths were determined from the literature (Bayer 1961; Sánchez et al. 1998; Sánchez 2001). Only measurements that could be separated by gap coding were considered informative characters (Mickevich and Johnson 1976; see also Swiderski et al. 1998).

Results

Our sequence data comprised ~366 bp of the ND6, 43 bp of the intergenic non-coding region, and the first 21 bp of the ND3 coding region (positions 1,590–2,020 compared to the sequence of Sarcophyton glaucum, Pont-Kingdon et al. 1998); the last ~87 bp from the 3′ end of the lsu-rRNA (16S), and ~550 bp of the ND2 (positions 722–1,358, Beaton et al. 1998); the last ~6 coding positions of ND4L, ~14 bp of non-coding intergenic spacer, and 747 bp at the 5′ end of msh1 (positions 2,644–3,411, Pont-Kingdon et al. 1998). The final database comprised 1,633 (544 AA) coding positions including, in this order, 360 bp of ND6 (1–360), 525 bp of ND2 (361–885), and 747 bp of msh1 (886–1,633). Only two gaps were present in the matrix (two and three amino acid positions in the msh1 from Pacifigorgia stenobrochis and the outgroup Alcyonium digitatum, respectively), which did not have any weight in the analysis. Since different combinations of genes, or each gene separately, yielded the same overall result we conducted the phylogenetic reconstruction using the three genes combined, which produced the best supported relationships.

Nucleotide substitutions

Using maximum parsimony, we obtained four partially resolved most-parsimonious trees using 279 parsimony-informative characters (Fig. 3A, B; see also supplementary electronic material, Appendix 1). Most informative characters came from msh1 (176; 63%), followed by ND2 (61; 21%), and ND6 (42; 16%). The length-frequency distribution from 1,000 random trees (mean=1,444.7, SD=64.7) was significantly skewed (g1=−0.54, P<0.01) and the unpermuted most-parsimonious trees were significantly shorter than 1,000 permutation replicates (>1,322, PTP: P<0.001). Both of these analyses indicate prevalence of phylogenetic signal over noise and homoplasy. The most-parsimonious trees contained clades that mixed members of Plexauridae and Gorgoniidae (Fig. 3). Some members of Plexauridae, such as the Plexaurella spp., grouped basally to some gorgoniids, and some gorgoniids (Pterogorgia spp.) grouped with plexaurids (Muriceopsis spp.). The trees differed in the placement of two clades. Muricea spp. were either basal to Eunicea spp.–Plexaura spp.–Pseudoplexaura spp. (Fig. 3B) or the sister group of Pseudoplexaura crucis (Fig. 3A). The other was a basal node including the sister relationship Muriceopsis spp. and Pterogorgia spp. that was either basal to the whole ingroup or basal to most plexaurids (Plexaura spp., Eunicea spp., Muricea spp., and Pseudoplexaura spp.) (Fig. 3; see also supplementary electronic material, Appendix 1). Other differences among most-parsimonious trees are mainly in the placement of Eunicea spp. and Plexaura spp. (Figs. 3, 4), but the number of substitutions at those nodes is low, and the clade could be considered unresolved. Topological relationships among most gorgoniids (Leptogorgia spp., Pacifigorgia spp., Pseudopterogorgia spp., Gorgonia spp., and Phyllogorgia spp.) and Plexaurella spp. remained unchanged in all most-parsimonious trees. Bootstrap analysis showed that most basal nodes were supported with values >94%, with the exception of Muriceopsis spp.–Pterogorgia spp. (54%), Pseudoplexaura spp.–Muricea spp. (58%) and two nodes within Pseudopterogorgia spp.–Gorgonia spp. (see supplementary electronic information, Appendix 2). There was no meaningful bootstrap support among species of Plexaura spp. and Eunicea spp. (Plexauridae), whereas most nodes within the Gorgoniidae were resolved and supported with high values. Pseudopterogorgia spp. and Gorgonia spp. exhibited mixed and polyphyletic relationships.

Fig. 3A, B.
figure 3

Two of the four maximum-parsimony phylograms using nucleotide substitutions (length=554, consistency index=0.78, retention index=0.89, and homoplasy index=0.22). Black dots depict the main conflicting node among maximum-parsimony trees. Species classified as Plexauridae are underlined. See electronic supplementary material (Appendix 1) for additional trees

Full size image
Fig. 4.
figure 4

Maximum-likelihood tree from best-fit model (HKY+G) selected by hLRT in MODELTEST version 3.06. Black dot depicts the only conflicting node among maximum-parsimony trees. Species classified as Plexauridae are underlined

Full size image

Likelihood-ratio tests showed that the HKY+G model (−lnL=8,648; Gamma shape parameter=0.51) significantly improved (P<0.0001) fit of the data in a maximum-likelihood analysis. Parameters used in the analysis were as follows: base-frequency composition (A=0.289, C=0.1682, G=0.195, T=0.347), transition/transversion ratio (ti/tv=3.26), and equal transition rates among sites. After 3,345 rearrangements using the heuristic search, a maximum-likelihood tree was obtained (log score=1,765.91). The maximum-likelihood tree was identical to one of the most-parsimonious trees (Figs. 3A, 4), with the node Muriceopsis spp.–Pterogorgia spp. basal to Eunicea spp.–Plexaura spp.–Pseudoplexaura spp.–Muricea spp. It is notable in the maximum-likelihood tree that some plexaurid species, particularly in the genera Eunicea and Plexaura, had low or no substitutions per site, whereas some basal gorgoniids (Leptogorgia spp., Pacifigorgia spp., and Pterogorgia spp.), as well as Plexaurella spp., exhibited a larger number of substitutions. Bootstrap analyses from the maximum-likelihood tree (100 replicates, not shown) had nearly identical topological relationships and similar support values as the Bayesian-likelihood estimation of phylogeny (Fig. 5A). The Bayesian analyses use a different maximum-likelihood approach by assuming partitions (third codon position evolving differently). Clade support values using maximum likelihood with Bayesian analysis indicated good support for most nodes found in the maximum-likelihood tree and identified two more resolved nodes compared to maximum parsimony (Fig. 5A; see also supplementary electronic material Appendix 2). Unconstrained trees were significantly more likely than trees constrained by the classification (families and genera), indicating disagreement between the common classification and this molecular hypothesis (H–S test: diff. −lnL=182.7, P=0.0001).

Fig. 5A, B.
figure 5

Branch/node support from different likelihood phylogenetic analyses. A Bayesian-estimated likelihood tree (HKY) using nucleotide partition by positions, 1,000,000 Monte Carlo Markovian chain generations (Bayesian–Monte Carlo simulation by MrBayes sampling every 100 simulations). Above node support is the 50%-majority-rule consensus from the 10,001 sampled trees generated by PAUP*. B Maximum-likelihood tree using the model mtREV24 for mitochondrial amino acid substitution in the program Tree-Puzzle, showing support from 10,000 puzzling quartets. Black dots depict the only conflicting node among maximum-parsimony trees. Species classified as Plexauridae are underlined

Full size image

Amino acid substitutions

Initially, we obtained ten partially resolved most-parsimonious trees using 106 parsimony-informative characters [length=238, consistency index (CI)=0.798, retention index (RI)=0.891, and homoplasy index (HI)=0.202; trees not shown]. The lengths of 1,000 random trees had a mean of 570.8 (SD=20.9), and the tree-length frequency distribution was significantly skewed (g1=−0.51, P<0.01). PTP with 1,000 replicates clearly showed that all replicates had more steps (>485) than the unpermuted tree (P<0.001). These analyses suggest low noise and homoplasy in the most-parsimonious trees. Despite the lower number of parsimony-informative characters using amino acid positions compared to base substitutions (106 and 279, respectively), the results from bootstrap analyses were nearly identical, except for three nodes comprising relationships among closely related species (see supplementary electronic material, Appendix 2).

The program Tree-Puzzle chose the model for amino acid substitutions "mtREV24," which is a model for amino acid substitution in mitochondrial genes (Adachi and Hasegawa 1996). Amino acid frequencies ranged from 1.4% (Cys) to 9.9% (Ser), with a notable bias for leucine (17.4%). Quartet puzzling was used to choose from possible maximum-likelihood trees using amino acid substitutions (mtREV24) and simultaneously calculate branch support values (Strimmer and von Haeseler 1996). The phylogenetic reconstruction included 10,000 puzzling steps and 23,751 analyzed quartets (738, 3.1% unresolved quartets). Species topological relationships were very similar to the maximum-parsimony results, with the exception of two additional nodes that were well supported with this analysis (Fig. 5B; supplementary electronic material, Appendix 2). Nonetheless, there was no significant difference between the mtREV24 hypothesis (amino acids) and the HKY+G model using nucleotide substitutions (S–H test: diff. −lnL=5.93, P=0.187).

Morphological characters and the molecular hypotheses

A limited number of homologous morphological characters (6) were found among the axis/base mineralization and sclerite characteristics (Table 2). Mineral axial composition (if present) is discrete with a clear species distribution (Figs. 6D, 7C, E; see also supplementary electronic material, Appendix 3). Though sclerites exhibit many morphological characters, only a few were clearly different among species. Only one character (base mineralization) supported the Plexauridae and Gorgoniidae classification (Fig. 7C). Five characters were congruent with the molecular phylogenetic hypotheses presented here. Length, CI, and RI were not different for the six morphological characters when mapping them onto the maximum-likelihood trees from nucleotide and amino acid substitutions (Fig. 7). Most character state changes occurred in basal clades and/or gorgoniids, which agreed with the distances and substitution rates found with the molecular analyses. Interestingly, all characters supported the position of Plexaurella spp. as basal to gorgoniids. Plexaurella spp. have unique axis mineralization characters (calcite and MgCO3, Fig. 6D) and surface sclerites (antler-like spindles, Fig. 6E). Plexaurella spp. have intermediate-sized sclerites, between the tiny sclerites of some gorgoniids (<0.3 mm) and the large spindles of some plexaurids (up to several millimeters, Fig. 6N). Plexaurella spp. shared important characters with Gorgoniidae, such as small polyp rods and the absence of armature in the polyps (Figs. 6C, 7A, B), and complex (many internal geometrical/fractal subwarts) warts (or tubercles) disposed in transverse girdles, usually <20 μm wide (Figs. 6A, 7A, B). These traits were also present in Muriceopsis spp., suggesting a separation from Plexauridae where they have been classified.

Table 2. Morphological characters and states for Caribbean gorgonian corals (for species distribution see Fig. 7)
Full size table
Fig. 6A–N.
figure 6

Examples of some of the morphological characters using scanning electron microscopy that were mapped on the molecular hypotheses (see character states in Table 2). A, B Sclerite ornamentation (warts). A Details of warts from (top to bottom): Plexaurella grisea (scale bar: 20 μm), Muriceopsis flavida (20 μm), Pacifigorgia stenobrochis (10 μm), and Pterogorgia citrina (10 μm). B Details for (top to bottom): Plexaura kuna (20 μm), P. flexuosa (10 μm), Eunicea sp. (40 μm), and Muricea muricata (200 μm). C Polyp rod sclerite from Eunicea sp. (left) and rods from Pseudopterogorgia elisabethae (right, 60 μm). D Examples of axial/base mineralization from (top to bottom): P. elisabethae holdfast showing carbonate hydroxiapatite (CHAp) (2 μm), calcite-MgCO3 from Plexaurella nutans axial loculi (6 μm), and aragonite from P. flexuosa base (2 μm). E–H Different surface sclerite types. E Surface sclerites from (top to bottom): P. nutans (100 μm) and P. grisea (antler-like, 60 μm); F sclerites (scaphoids) from (top to bottom): Phyllogorgia dilatata (20 μm), Pseudopterogorgia acerosa (10 μm), Gorgonia ventalina (10 μm), and Pseudopterogorgia bipinnata (10 μm); G sclerites from (top to bottom): Muriceopsis bayeri (100 μm) and M. flavida (100 μm); H sclerite (disk-spindle) from Leptogorgia virgulata (20 μm); I sclerite from Pacifigorgia elegans (20 μm); J sclerite from M. muricata (100 μm); K club-sclerites from (left to right): Plexaura homomalla (80 μm), P. flexuosa (40 μm), Pseudoplexaura crucis (100 μm), and P. kuna (40 μm). L–N Examples showing size differences among three size-classes of spindle sclerites. L Spindles from (left to right): Muricea pinnata (400 μm), P. kuna (200 μm), and P. flexuosa (200 μm); M spindles from (left to right): Pacifigorgia stenobrochis (20 μm), P. acerosa (10 μm), Pseudopterogorgia americana (20 μm), Gorgonia mariae (10 μm), Pterogorgia anceps (100 μm), and P. bipinnata (20 μm); N spindles from (left to right): M. flavida (100 μm) and P. grisea (100 μm). See Fig. 7 and Table 2 for character state descriptions and species distribution

Full size image
Fig. 7A–F.
figure 7

Morphological characters mapped onto Bayesian-estimated likelihood tree (as in Fig. 5A). Characters (A–F) and character states (indicated by numbers and shading) defined in Table 2 [A polyp sclerites (Fig. 6C); B sclerite warts (or tubercles) (Fig. 6A, B); C base/holdfast mineralization; D maximum spindle length (middle or internal cortex layers) (Fig. 6L–N); E axis mineralization (Fig. 6D); F surface layer cortex sclerites (Fig. 6E–K)]. Mapping characters on the tree: length=14, consistency index=0.78, and retention index=0.94. Species classified as Plexauridae are underlined

Full size image

Discussion

Phylogenetic analyses of partial sequences from the NADH-dehydrogenase (subunits 6 and 2) and the 5′ end of the mutS homologue (msh1) produced well-supported phylogenetic relationships for Caribbean gorgonian octocorals. Maximum-parsimony and -likelihood analyses of both base and amino acid substitutions produced very similar results, with high supporting values for most nodes. The only conflicting node was the clade Pterogorgia spp.–Muriceopsis spp., which had low support values in all of the analyses. These molecular results differ significantly from the common classification scheme, particularly regarding separation of the families Plexauridae and Gorgoniidae. Plexaurella spp., nominal plexaurids, were basal with respect to gorgoniids and shared many morphological characters with the gorgoniids. This result supports Gerhart's (1983) hypothesis that Plexaurella spp. are related to gorgoniids rather than to plexaurids. Similarly, Muriceopsis spp. also grouped with gorgoniids (Pterogorgia spp.) instead of plexaurids.

The sea fans Gorgonia spp. and Pacifigorgia spp. had non-sea fan relatives, supporting Bayer's (1953) hypothesis of convergent evolution. Convergent evolution in colony form was also observed among the pinnate gorgonians Muriceopsis spp. and Pseudopterogorgia spp. The pattern of convergent evolution observed among these gorgonian families seems to follow the overall pattern of convergence in branching observed among higher taxa of octocorals. Sánchez et al. (in press) hypothesize that branching in octocorals has evolved at least twice among suborders of octocorals, particularly Calcaxonia and Holaxonia.

Mapping morphological characters onto molecular results improves our understanding of evolutionary relationships among Caribbean octocorals. Ancestral characters (e.g. placed basal in the phylogenetic tree) seemed to be present in Gorgoniidae (e.g. CHAp mineralization, smallest spindles, and complex warts). Plexaurids, excluding Plexaurella spp. and Muriceopsis spp., all had large spindle sizes, lacked branch mineralization, contained aragonite in the base, and had club-like sclerites in the surface cortex layer. The morphologically intermediate genus Plexaurella (traditionally placed in the Plexauridae), was more closely related and occupied a basal position relative to gorgoniids. These molecular results made Gorgoniidae paraphyletic. Alternatively, Plexaurella spp. could be reassigned to a different family with respect to Plexauridae and Gorgoniidae. Independent sources of phylogenetic information, such as single-copy nuclear genes, as well as additional non-Caribbean species are needed to resolve the placement of the clade Muriceopsis spp.–Pterogorgia spp. because it had two possible topological positions, both with low branch support, in all the analyses.

Caribbean gorgonian octocorals: presence of recently evolved species?

Among Caribbean octocorals mitochondrial DNA (mtDNA) divergence was lower among some plexaurids (Eunicea spp. and Plexaura spp.) compared to gorgoniids and other plexaurids. The biogeography of the groups present in the Caribbean may provide an explanation for this. For instance, the genus with the lowest divergence, Eunicea, is distributed only in the Caribbean Sea. Gorgoniids, on the other hand, are distributed worldwide (Bayer 1953). The genus Leptogorgia has many species in the Pacific Ocean (Bayer 2000) as well as on the African Atlantic coast (Grasshoff 1988). Pacifigorgia spp. are mostly distributed in the eastern Pacific (e.g. Breedy 2001), but P. elegans occurs in the Caribbean. Muriceopsis spp., which grouped with gorgoniids in this analysis (Pterogorgia spp.), are also distributed in the western and eastern Atlantic, and the phylogeny of this genus seems to be related to a post-Pangean distribution pattern (Sánchez 2001). Pseudopterogorgia spp. are also present in Indo-Pacific reefs, although the Atlantic species are likely an endemic clade (Williams and Vennam 2001).

Plexaura spp. and Eunicea spp., on the other hand, exhibit a strikingly different biogeographic pattern. Eunicea is one of the more diverse genera in the Caribbean, and all 16 species occur sympatrically (Sánchez 1998). Closely related species within the genera Eunicea and Plexaura are reproductively isolated, and their genetics and morphology agree with the hypothesized species boundaries (Lasker et al. 1996; Sánchez 1998). Neither genus is found on Brazilian Atlantic reefs, where many gorgoniids occur (Bayer 1961). Eunicea and Plexaura appear to have undergone rapid speciation, as indicated by the low genetic divergence among species within their restricted geographical range.

Mitochondrial DNA sequences as phylogenetic characters for octocorals

MtDNA regions, such as genes for cytochrome oxidase I and cytochrome b, have been uninformative for low-level phylogenetic studies (e.g. intraspecific and closely related species) in other anthozoans such as scleractinian corals (e.g. Medina et al. 1999; Van Oppen et al. 1999). These genes also have low variation among octocorals (France and Hoover 2001, 2002). However, sequences from the NADH-dehydrogenase subunits and msh1 had sufficient variability to reconstruct holaxonian relationships. Comparisons of these Caribbean octocoral sequences with species from suborders Calcaxonia and Alcyoniina and the order Pennatulacea identified many indels in coding regions (authors' unpublished data), which suggests a great deal of phylogenetic information is also present in these genes for comparison of octocoral orders. The use of NADH-dehydrogenase subunits also provides the opportunity to include octocorals in broader phylogenies, since this gene is present in most metazoan mtDNA (Boore 1999). It is worth noting that although NADH-dehydrogenase and msh1 are among the most variable genes in the octocoral mtDNA, they provided limited information for presumably recently evolved species, such as Eunicea spp. and Plexaura spp. These genes, in fact, are invariant within species populations (LePard and France, unpublished data). Though regrettable for population and phylogeographic analyses, this low variation could be considered an advantage for delineating species boundaries. Saturation in these genes is very unlikely due to the low levels of mutation (France and Hoover 2001), which may be related to the function of the msh1 gene, which codes for a DNA mismatch repair protein (Pont-Kingdon et al. 1995; Culligan et al. 2000; Malik and Henikoff 2000).

Morphological characters and the molecular hypotheses

Using maximum parsimony, the distribution of morphological characters, such as polyp sclerites, sclerite warts, and axis mineralization, was completely consistent with the two more likely molecular phylogenetic hypotheses. Base mineralization, surface layer sclerites, and maximum spindle length, however, all required at least one homoplasy when mapped onto the molecular tree (Fig. 7, see also supplementary electronic material, Appendix 3). Sclerites and axial structures have been the most important characters for the taxonomy and classification of octocorals (e.g. Bayer et al. 1983), but an evaluation of their homology among species, genera, and families has not been made. Mapping sclerite characters onto the molecular results was consistent with genus but not family boundaries. The most variable character found was the type of surface sclerite. This character coincided with seven different clades with both phylogenetic and taxonomic meaning (Plexaurella spp., Leptogorgia spp., Pacifigorgia spp., Pseudopterogorgia spp.–Gorgonia spp.–Phyllogorgia spp.–Pterogorgia spp., Muriceopsis spp., Muricea spp., and Eunicea spp.–Plexaura spp.–Pseudoplexaura spp.). The phylogenetic utility of this character was first realized by Bayer (1953), and based on differences in surface sclerites he suggested convergent evolution for the sea fan forms of Pacifigorgia spp. and Gorgonia spp. Within each type there are also sub-types of sclerites. For instance, gorgoniid genera harboring scaphoids can be readily distinguished by modifications of the girdles and ornaments of these structures (Bayer 1961).

The explanation for such diversity of sclerite forms could be due to their functional morphology. Surface sclerites modify colony flexion by preventing compressibility of the outer cortex when they contact each other, which helps the colony to resist waves and flow forces (Lewis and Von Wallis 1991). This strong interaction with the environment could select for differences in sclerite type as a function of colony morphology. Since internal sclerites such as axial spindles are more conserved among species, it could be hypothesized that external sclerites are more likely to represent adaptive features.

Other sclerite characters such as sclerite warts and degree of polyp armature provide additional support for the molecular phylogenetic hypotheses. These sclerite traits had the same distribution among species and provided a clear delineation between derived plexaurids (Muricea spp.–Pseudoplexaura spp.–Eunicea spp.–Plexaura spp.) and basal "gorgoniids" (Plexaurella spp.–Muriceopsis spp.–gorgoniids). Derived plexaurids harbor enormous spindles that can exceed 1 mm in length. Ornamentations on their surfaces were also characteristic. Sclerites within the polyps of these species were also robust, ornamented, and frequently formed a sclerite armature in the base of the polyp. Basal "gorgoniids," on the other hand, had the tiniest spindles and polyp rods. Their ornamentation was more regular, and the wart girdles were very symmetrical, and they often exhibited smaller, modular warts. Differences in these characters may be explained by the nature of sclerite growth in the two groups. Sclerite construction mechanisms are only known for a few species, but sclerite characters appear to be correlated to extra- versus intra-cellular stages of formation. Sclerite formation in plexaurids involves both intra- and extra-cellular mechanisms as well as crystallization stages that interact with other cellular structures (Goldberg and Benayahu 1987). In contrast, gorgoniid octocoral sclerites mature within single scleroblasts (Kinsley and Watabe 1982). Gorgoniid sclerites are about an order of magnitude smaller than plexaurid sclerites. Although more work is necessary to understand the complex morphological evolution of octocorals, a combination of molecular phylogenetics and morphology (sclerites and axial structures) is a very promising approach for determining phylogenetic relationships among octocoral families and, perhaps, sub-orders.