Implications from a 28S rRNA gene fragment for the phylogenetic relationships of halichondrid sponges (Porifera: Demospongiae)

Abstract

Halichondrida is a pivotal demosponge order of which the classification underwent major changes in the recent history. The monophyly of this order and its intra-ordinal phylogeny cannot be reliably determined on the basis of morphology. Here we present a 28Sr RNA gene tree of selected halichondrids, which supports the hypothesis of halichondrid non-monophyly and elucidates further inter-ordinal relationships. We enlarged the analysis by previously published sequences, discuss how previous analyses suffer from taxon bias and analyse the resulting phylogenetic implications. Most halichondrid families (in particular Axinellidae und Dictyonellidae) cluster polyphyletic and the molecular classification of several genera does not agree with the current (morphological) system. Die Systematik der Ordnung Halichondrida ist von Bedeutung für die Phylogenie aller Demospongien und hat in den letzten Jahren große Veränderungen vollzogen. Dennoch konnte die Monophylie dieser Ordnung und ihrer Familien noch nicht mit morphologischen Merkmalen deutlich gemacht werden. Hier präsentieren und diskutieren wir eine 28SrDNA-Phylogenie repräsentativer Halichondrien. Teils wurde diese Stammbaumrekonstruktion durch Sequenzen aus anderen Arbeiten erweitert und der Einfluss des Taxonsets auf das Resultat analysiert. Die Monophylie der Ordnung wird durch diese Fragmentanalyse nicht gestützt, und mehrere Familien (insbesondere Axinellidae und Dictyonellidae) gruppieren polyphyletisch. Die Position mehrerer Gattungen, insbesondere Axinyssa und Stylissa, ist nicht im Einklang mit dem aktuellen (morphologischen) System. The phylogeny of the demosponge order Halichondrida is pivotal in demosponge systematics as its composition and affinities in history reflect the major changes in demosponge classification as proposed in the second half of the last century. Gray (1867) erected the taxon ‘Halichondriadae’. Its composition was modified based on morphological features by Vosmaer (1886 [1887]); Ridley and Dendy (1887); Topsent (1928) and de Laubenfels (1936). Later, a new demosponge classification entirely based on reproductive features, was founded by Lévi (1953). In this new classification demosponges were divided in two major subclasses ‘Tetractinomorpha’ (oviparous taxa) and ‘Ceractionomorpha’ (viviparous taxa) as these features matched with large demosponge groups. It resulted in a reallocation of formerly recognized orders and families in separate clades. The order Halichondrida sensude Laubenfels (1936) was split up in the ceractinomorph ‘Halichondrida s.s.’ (including the currently recognized families Halichondriidae and Dictyonellidae) and the tetractinomorph ‘Axinellida’ (including the currently recognized families Axinellidae, Desmoxyidae and Bubaridae). Lévi's classification was elaborated by Bergquist (1980) and Hartman (1982). It formed the backbone of demosponge systematics in the second half of the 20th century. With the introduction of cladistic character comparison in sponge systematics (van Soest 1990) parsimony-inconsistencies of Lévi's classification were criticized (van Soest 1987; Hooper 1990). Evidence for a paraphyletic nature of the two major demosponge subclasses could be shown and an alternative classification was suggested (van Soest 1991), which gained broader acceptance in the recent years (Lévi 1997). The taxon Axinellida sensu Lévi was abandoned and two of its families (Axinellidae and Desmoxyidae) merged with Halichondrida (van Soest et al. 1990). Currently, van Soest and Hooper (2002) assign five families to the order: Halichondriidae, Dictyonellidae, Desmoxyidae, Bubaridae, and Axinellidae. Nevertheless, morphological synapomorphies have neither been found for those five families, nor for the entire order. Halichondrida is still defined by character combinations and underlying synapomorphies. In particular the family Dictyonellidae sensuvan Soest et al. (2002) is primarily based on assumed secondary losses than on observable characters. As a consequence, intra-ordinal relationships of this important order remain highly unresolved. van Soest et al. (1990) proposed a morphological phylogeny for Halichondrida and its family Halichondriidae, but could not provide statistical support. Such supported phylogeny of the Halichondrida is hampered (as for many other demosponge taxa) by either the lack of phylogenetically informative characters and diverging views of their interpretation (Carballo et al. 1996). To gain additional characters in sponge systematics, chemical compounds have been recruited (e.g. van Soest and Braekman 1999) but until now these approaches suffer from a taxon bias and ambiguities in pathway homology and metabolite origin. The use of cytological features in sponge systematics has also been examined (Boury-Esnault et al. 1994), but these are technically demanding to observe, prone to preparation artefacts, and their phylogenetic information content is not fully estimated. The main progress in recent sponge systematics lies in the gain of DNA sequence data. Primarily the subunits of the cytoplasmatic ribosome (18S and 28SrDNA) have been used to reconstruct phylogenetic trees (Borchiellini et al. 2000; Alvarez et al. 2002; Manuel et al. 2003; Borchiellini et al. 2004). Some of these 28SrDNA analyses yielded evidence against a monophyletic relationship of the Halichondrida. An analysis of the 28S C1–D2 region found the family Halichondriidae in a shared clade with the hadromerid family Suberitidae (Chombard and Boury-Esnault 1999). The resulting taxon, called ‘Suberitina’ still lacks acceptance as it fails to show unambiguous morphological synapomorphies notably regarding the spicule geometry. An analysis of the 28S D3–D5 region (McCormack and Kelly 2002) favoured a close relationship of the genus Topsentia (order Halichondrida) with the genus Tethya (order Hadromerida) and additionally the genus Hymeniacidon (order Halichondrida) with Aaptos and Suberites (both order Hadromerida). Such constellation of Halichondrida with Hadromerida lacks verification from independent genes and potential morphological or biochemical synapomorphies. Other 28SrDNA studies found a close relationship between the order Halichondrida (genus Axinella) and the order Agelasida (genus Agelas, e.g. Lafay et al. 1992). A Halichondrida–Agelasida constellation can only be explained morphologically with multiple a priori assumed analogies, but in contrary to the Halichondrida–Hadromerida constellation as described above, such an Axinellidae–Agelasida constellation finds biochemical support by the common possession of the pyrrole-2-carboxylic acid derivatives (Braekman et al. 1992) and glycosyl ceramides (Costantino et al. 1996). However, all these molecular analyses comprised only a subset of halichondrids and made a comprehensive study of all major families necessary to elucidate their composition and relationships to non-halichondrids. In this study, we present a subsequent comprehensive 28SrDNA study from the halichondrid perspective. Erpenbeck et al. (2004) observed that the 28S gene structure and evolutionary rate in sponges underlie significant differences on order level. Based on those findings, we could narrow down the taxon set to a homogeneous and comparable character set without long branches, which mask phylogenetic information and lead to an erroneous signal. Here we present the results of a combined taxon set, which comprises representative Halichondrida, Agelasida and Hadromerida. Our aim is to test the monophyly of Halichondrida sensuvan Soest et al. (1990) in an enlarged taxon set and to get insight in phylogenetic relationships between and inside their families. Furthermore, we analyse the evidence of a ‘Suberitina’ clade as proposed by Chombard and Bouˇry-Esnauˇlt (1999) in this enlarged taxon set with a different 28SrDNA fragment. The sponge tissue was either freshly collected with SCUBA diving, or sampled during the SYMBIOSPONGE project (EU-MAS3CT 97-0144). A complete list of species studied is given in Table 1. Up to three specimens per species were examined. DNA extraction and PCR setup, cloning and sequencing was carried out as described in Erpenbeck et al. (2002). PCR primers employed were taken from McCormack and Kelly (2002), primers: RD3A: GACCCGTCTTGAAACACGA and RD5B2: ACACACTCCTTAGCGGA, temperature regime: 94° 3 min during which the Taq polymerase is added, 35× (94° 30 s; 50° 20 s; 72° 60 s), 72° 10 min). Sequence management was performed using MacClade 4.03 (Maddison and Maddison 1992). DAMBE (Xia and Xie 2001) was used to fetch and splice these sequences from Genbank (http://www.ncbi.nlm.nih.gov/), which were included in an enlarged data set for comparison reasons. We received a first alignment by Clustal ×1.8 (Jeanmougin et al. 1998) under default settings and optimized it by eye under usage of secondary structure information. This secondary structure information was obtained by annealing to 23S-like secondary structures published by Hancock et al. (1988) and Alvarez et al. (2000) under comparison with the common eukaryote model (Schnare et al. 1996). Prior to phylogenetic reconstructions we performed several a priori analyses to check the quality of character and taxon set. In sponges more than in most other organisms the true sponge origin of PCR amplification products should be verified as contaminations by symbionts and food particles can occur. We performed a BLAST search (Altschul et al. 1990) against Genbank sequences and additionally checked the phylogenetic position of our sequences against other eukaryotes in a phylogenetic tree as described in Erpenbeck et al. (2002). This approach was used to verify that our sequences cluster monophyletically and in close relationship to other diploblastic Metazoa. We employed a PTP test (Faith 1991) as implemented in PAUP*4.0b10 (Swofford 2002) with 100 replicates based on heuristic search to determine whether the information of our data sets arose by chance. The base frequencies have been tested for homogeneity in a χ2 test of all taxa as implemented in PAUP*. Several sponge sequences fetched from Genbank were shorter than our own sequences. The combination would have resulted in a data matrix with a majority of missing characters. To reduce the resulting noise we performed two different approaches on two different data sets: A long (D3–D5) data set of taxa with sequences of the full length (including D3 + D4 + D5 subdomains) to receive the strongest possible phylogenetic signal. It resulted in a data set of 53 taxa and 760 characters. An extended taxon set (D3) of all taxa including several shorter sequences from Alvarez et al. (2000). It resulted in a data set of 75 taxa and 455 characters of the D3 region only. Phylogenetic reconstructions were based on Bayesian analyses performed by MrBayes v3b4 (Huelsenbeck and Ronquist 2001). The likelihood and prior probability parameters were estimated using MrModeltest v1.1 (Nylander 2002) under PAUP*. All bayesian analyses were performed with four Metropolis-Coupled Markow Chains and two million generations. The chains were ‘burned in’ until the posterior probabilities reached a stable plateau. Different approaches were analysed regarding application of the relatively best-fitting evolutionary model. The data set was split in several partitions: partition ‘stem’ contained all helix regions sequenced. Partition ‘loop’ contained all not pairing nucleotides. Furthermore, partition ‘excluded’ comprised ambiguous alignable positions. Characters of this partition were not included in the analysis. We performed the bayesian analyses as a two-partition (‘stem’ and ‘loop’, see Erpenbeck et al. 2004) approach and compared the results with a maximum-likelihood reconstruction. We did not apply a nucleotide substitution model based on a 16 × 16 nucleotide doublet matrix (Schöniger and von Haeseler 1995) for the helices, because it reduced the resolution power in smaller character sets as observed in previous analyses (Erpenbeck et al. 2004). For all analyses, we chose a non-demosponge outgroup. As phylogenetic relationships of demosponges are still mostly unresolved a calcareous sponge outgroup seemed more suitable than a demosponge that might lead to uncertain tree polarity. We chose the calcareous sponge Leucosolenia sp. as sequenced by Medina et al. (2001). Branch length differences to other demosponges were previously regarded to be moderate (Erpenbeck et al. 2004). Some 190 of the 760 characters were excluded from the long (D3–D5) data set due to alignment ambiguities. Sequence length varied from 574 to 591 base pairs. The alignment revealed no exceptional secondary structural features as found among other demosponge orders (Erpenbeck et al. 2004). MrModeltest estimated based on the likelihood-ratio-test the following nucleotide substitution models as the relative best-fitting for the included characters of the partitions: partition ‘stem’: HKY + I + G (Hasegawa et al. 1985), partition ‘loop’ GTR + G (Rodríguez et al. 1990). Maximum-likelihood (one partition): SYM + G + I (Zharkikh 1994). Some 167 characters of this data set were excluded from the phylogenetic analyses because they could not be aligned unambiguously. Based on the likelihood-ratio-test we used the following models for the partitions: Stem: HKY + I + G (Hasegawa et al. 1985), loop: GTR + I + G (Rodríguez et al. 1990), Maximum-likelihood (one partition): TrN + I + G with even frequencies (TrNef + I + G, Tamura and Nei, 1993). We retained the gene trees of both analyses with clades of less than 80% posterior probability collapsed. The trees are in agreement with their particular maximum-likelihood trees. We reconstructed the ‘combinable component’ consensus of both bayesian trees (Fig. 1). The tree of the ‘Long (D3–D5) data set’ consists of a subset of taxa and is plotted with bold branches on the ‘Extended taxon set (D3)’ tree. Combined tree of the ‘long (D3–D5) data set’ (bold lines, posterior probabilities below the branches) and the ‘extended taxon set (D3)’ (bold lines and thin lines, posterior probabilities above the branches). Clades with posterior probabilities lower than 80% are collapsed. Dashes indicate a posterior probability <80% in that particular analysis. The asterisk indicates one clade not supported by the long (D3–D5) data set as shown, favouring Mycale flagellifera more basal (dotted lines) The two trees differed in the relationships of the two Mycale spp. (see Fig. 1) and Liosina paradoxa to Axinella polypoides: The (shorter) ‘extended taxon set (D3)’ favoured a sister-group relationship and the ‘Long (D3–D5) data set’ a paraphyletic and more derived position of A. polypoides. Fig. 1 displays a polyphyletic order Halichondrida (sensuvan Soest and Hooper 2002). Except for the desmoxyid samples, none of the halichondrid families falls monophyletic either. The family Halichondriidae with the exception of Axinyssa spp. forms a clade with the Suberitidae (Suberites and Aaptos, order Hadromerida). The family Axinellidae is split up in multiple clades: (1) an Axinella-clade which is formed with all species of Stylissa (Dictyonellidae) and the Agelasida; (2) a paraphyletic clade which consists of Ptilocaulis and Reniochalina together with the Desmoxyidae samples; (3) a Dragmacidon clade and (4) a Cymbastela spp. clade clustering with some Dictyonellidae. Axinella is not monopyletic as Axinella damicornis clusters basal to Stylissa and A. verrucosa in a sister-group relationship to the Agelasida samples (Agelas and Astrosclera). Axinella polypoides is even more basal. The family Dictyonellidae also split up in multiple clades: Stylissa groups separately. Acanthella and Dictyonella cluster with Axinyssa (Halichondriidae) and Cymbastela (Axinellidae). Affinities of Scopalina and Svenzea to other dictyonellids are not supported as well as the position of L. paradoxa. We restrict ourselves to regard Bayesian results supported by posterior probabilities of 80 and higher as bayesian posterior probabilities in general tend to be higher than bootstrap support values (Huelsenbeck et al. 2002). This stringency affects the resolution power negatively, which is certainly more acceptable than a higher resolved tree under erroneous phylogenetic signal. Here we combine the consensus trees of different approaches to keep the amount of ambiguities as low as possible. This analysis can be regarded as an investigation on the influence of the taxon set on the views of phylogenetic relationships in a difficult systematic taxon such as the demosponges. The overall topologies of the thorough morphological and molecular studies on Axinellidae by Alvarez de Glasby (1996) and Alvarez et al. (2000) can be re-found in Fig. 1, but the inclusion of the additional taxa splits up clades that were previously recognized by those authors. Taxa of the order Hadromerida mingle with Halichondrida in this 28S tree. The genera Suberites, Aaptos (both Suberitidae) and Spheciospongia (Clionaidae) are representatives of two hadromerid families, which cluster with Halichondrida. These results are similar to the molecular findings of Chombard and Boury-Esnault (1999) and McCormack and Kelly (2002) in which the sequences of the hadromerid genera Suberites, Tethya and Spheciospongia group with the halichondrid Axinella, Topsentia and Hymeniacidon. Further Halichondrida/Hadromerida clusters might be detected at least on 28SrDNA-gene fragment level. Nevertheless, only if further independent data confirm these findings, the ‘Suberitina’ (Chombard and Boury-Esnault 1999) should be expanded. Similarly, more Halichondrida/Poecilosclerida clusters might be expected with 28SrDNA. Poecilosclerida, the taxon-richest sponge order, is clearly underrepresented in our current taxon set. In our analysis, the sequences of Mycale flagellifera and Mycale fibrexilis do not form a supported sister group and remain at uncertain positions in the gene tree but both cluster with Halichondrida. Mycale fibrexilis is assigned to the subgenus Carmia, M. flagellifera (Vacelet and Vasseur 1971) to the subgenus Naviculina. Their morphological differences, such as the different architecture of the skeleton and shape of the anisochelae, could be reflected in their different positions. The monophyly of the family Axinellidae is not supported in this enlarged taxon set. Evidence for polyphyly of the morphologically variable genus Axinella was previously shown in Alvarez et al. (2000). The genus Axinella is in our analysis not distinguishable from the dictyonellid genus Stylissa, with the exception of A. damicornis. Several Stylissa species such as S. massa and S. carteri, both included in our data set, were previously regarded as Axinella because of their vaguely reticulate skeleton of styles, lacking a distinct surface specialization. The distant clustering of the type species A. polypoides is in congruence with the shorter data of Alvarez et al. (2000). With the inclusion of Alvarez et al.'s sequences we provide evidence that the axinellid taxa Reniochalina, Ptilocaulis and Cymbastela are more distant. The phylogenetic position of the Cymbastela coralliophila sequence as published by Alvarez et al. (2000) still remains ambiguous. Cymbastela coralliophila is atypical for its genus as it possesses a paratangential ectosomal skeleton. There is only one genus of the family Halichondriidae not clustering with the other members of that family: Axinyssa species display closer affinities to Dictyonella and Acanthella (Dictyonellidae) in this gene tree. Morphological evidence for a closer relationship between Axinyssa and Dictyonellidae may be the shared absence of a special ectosomal spicule skeleton, which is part of the definition of Dictyonellidae (van Soest et al. 2002). Dictyonellidae sensu van Soest et al. (2002) are polyphyletic in our results. Next to Stylissa, which is discussed above, Liosina is distant from the type genus Dictyonella. This configuration implies upcoming difficulties to keep an integral Dictyonellidae valid in the future. On the other hand, the genus Acanthella (Dictyonellidae) confirms its recent placement in this family. It was previously regarded as axinellid in morphological (Bergquist 1970) and molecular studies (Alvarez et al. 2000). Here, it clusters close to Dictyonella, separate from Axinella. The recent morphological assignment of Svenzea to Dictyonellidae (Alvarez et al. 2002) finds molecular support as both the Carribean S. zeai and the Pacific S. devoogdae cluster as a sister group to Scopalina. Myrmekioderma and Didiscus were both previously regarded as Halichondriidae (van Soest et al. 1990) from which they are clearly distant in our gene tree. Their placement with other Desmoxyidae (van Soest and Lehnert 1997; Hooper 2002) has to be verified by additional molecular data. Our data elucidates a different perspective of halichondrid phylogeny. We do not intend to come up with a new phylogeny of halichondrid demosponges at this stage, or a proposal to abandon this taxon, but present an alternative view on halichondrid relationships. The current analysis is based on only a fragment of one single gene, and the topology should, like every gene tree, be compared and validated with other gene trees, preferably from independent genes such as mitochondrial genes or nuclear proteins. The suitability of 28SrDNA to resolve demosponge relationships appeared to be accepted (McInerney et al. 1999), but major inconsistencies have been shown (e.g. McCormack et al. 2002; Erpenbeck et al. 2004), which supports our view that 28SrDNA data should be compared with other genes. Only a thorough comparison between independent phylogenetic data sets of molecular, morphological and biochemical data can provide the final the existence of morphological un-explainable constellations such as ‘Suberitina’. Nevertheless, the current analyses provide evidence for alternative relationships in demosponges. It elucidates that one should be aware that in such a morphologically rather ‘simple’ taxon as the Porifera character-analogies should never be disregarded. Phylogenies remain ambiguous as long as monophyly is not assessed and not all taxa are investigated (Graybeal 1998). We would like to thank Matthijs van Couwelaar, Raquel Gomez, Bert W. Hoeksema, Mario J. de Kluijver, Peter Kuperus, Frederick R. Schram and Nicole J de Voogd for contributions in various ways to this paper. Part of the material was freshly collected during the SYMBIOSPONGE project (EU-MAS3CT 97-0144). This work was financed by the Dutch organization for scientific research (NWO) Nr: ALW 809.34.003. All experiments comply with the current laws of the Netherlands and the European Union.