Abstract.- Because calcareous sponges are triggering renewed interest with respect to basal metazoan evolution, a phylogenetic framework of their internal relationships is needed to clarify the evolutionary history of key morphological characters. Morphological variation was scored at the suprageneric level within Calcispongia, but little phylogenetic information could be retrieved from morphological characters. For the main subdivision of Calcispongia, the analysis of morphological data weakly supports a classification based upon cytological and embryological characters (Calcinea/ Calcaronea) rather than the older classification scheme based upon the aquiferous system (Homocoela/Heterocoela). The 18S ribosomal RNA data were then analyzed, both alone and in combination with morphological characters. The monophyly of Calcispongia is highly supported, but the position of this group with respect to other sponge lineages and to eumetazoan taxa is not resolved. The monophyly of both Calcinea and Calcaronea is retrieved, and the data strongly rejected the competing Homocoela/Heterocoela hypothesis. The phylogeny implies that characters of the skeleton architecture are highly homoplastic, as are characters of the aquiferous system. However, axial symmetry seems to be primitive for all Calcispongia, a conclusion that has potentially far-reaching implications for hypotheses of early body plan evolution in Metazoa. [Axial symmetry; Calcarea; Calcispongia; evolution; Metazoa; phylogeny; Porifera; 18S rRNA.]

The calcareous sponges (traditionally referred as class Calcarea of the phylum Porifera) are unique with respect to all other sponges (classes Hexactinellida and Demospongiae, the siliceous sponges) in their possession of a mineral skeleton made of calcium carbonate spicules. Although numerically a rather small group (approximately 500 species have been described, representing <5% of known sponge species), they are present on marine hard substrates in all latitudes, and they display a great diversity of organization levels (Borojevic et al., 1990, 2000; Manuel et al., 2002).

More than a century ago, Haeckel (1870a, 1870b, 1872), the famous promoter of the concept of recapitulation, emphasized the interest of calcareous sponges from the point of view of sponge and metazoan evolution. The comparison between the organization of calcareous sponges and coral polyps led him to propose a direct homology of germ layers between sponges and cnidarians (Haeckel, 1870a). Although such homology remains highly controversial, this work was historically crucial in suggesting the inclusion of sponges within a monophyletic Metazoa. Later, Haeckel (1874:151) mentioned that the study of calcareous sponge embryology (Haeckel, 1872) had been the starting point for his considerations on metazoan origins, as he formalized in the Gastraea theory (Haeckel, 1874). According to this theory, all living metazoans, including the sponges, are derived from a hypothetical ancestor, the Gastraea, which was morphologically similar to the gastrula embryological stage in modern animals.

The embryology of calcareous sponges has been studied further during the 20th century and has been largely popularized in zoological courses. Especially well known is the remarkable embryonic development of some calcareous sponges such as Sycon, involving inversion of surfaces (Duboscq and Tuzet, 1935) of an internally flagellated stomoblastula, leading to a bipolar amphiblastula larva (Fig. 1) (Duboscq and Tuzet, 1937, 1942; Tuzet, 1973).

In recent years, there has been a revival of interest in calcareous sponges with respect to early metazoan evolution. In several molecular phylogenetic studies, Calcispongia have been placed as the sister group of nonsponge metazoans (the Eumetazoa: Ctenophora, Cnidaria, Placozoa, and Bilateria) (Lafay et al., 1992; Cavalier-Smith et al., 1996; Collins, 1998; Kruse et al., 1998; Zrzavy et al., 1998; Adams et al., 1999; Borchiellini et al., 2001; Medina et al., 2001). The implications of sponge paraphyly for scenarios of early metazoan evolution have been discussed (Borchiellini et al., 2001). Calcareous sponges were upgraded to the phylum level by Zrzavy et al. (1998), under the name Calcispongia, which we follow here. A search for homologues of eumetazoan key regulators of morphogenetic processes has revealed an unexpected diversification of homeobox genes in a calcareous sponge (Manuel and Le Parco, 2000).

Because Calcispongia is a key taxon for the understanding of basal metazoan phylogeny and evolution, a well-supported hypothesis about the internal phylogenetic relationships of this group is particularly desirable, especially as a support for hypotheses of morphological character evolution. The reconstitution of a character set ancestral for all Calcispongia is a prerequisite for comparison with other basal metazoan lineages. No previous studies of internal calcareous sponge morphology have been conducted using modern phylogenetic tools and concepts.

The relevant anatomical variation within Calcispongia is illustrated in Figure 1. This variation includes the famous morphocline in the organization of the aquiferous system, with three main types: asconoid (a unique atrial cavity lined with choanocytes, i.e., cells bearing a collar of microvilli with a flagellum at its center; Fig. 1A), syconoid (elongated choanocyte chambers radially arranged around the atrial cavity; Fig. 1B), and leuconoid (numerous spherical and small choanocyte chambers scattered throughout the body wall; Fig. 1C). A fourth type, sylleibid (not illustrated), is intermediate between syconoid and leuconoid (i.e., spherical choanocyte chambers not scattered without order but grouped near exhalant canals, which are radially arranged around the atrial cavity). When all internal body cavities are lined with choanocytes, the aquiferous system is called homocoel (Fig. 1A); when some internal cavities are lined with pinacocytes (flattened epithelial cells generally devoid of a flagellum), the aquiferous system is called heterocoel (Fig. 1D). Asconoid sponges are homocoel, syconoid sponges generally are heterocoel, and sylleibid and leuconoid sponges always are heterocoel. Much of the morphological variation in sponges concerns the architecture of the skeleton (Fig. 1E-H), which has long been the basis for diagnoses of the families (Dendy and Row, 1913; Borojevic et al, 1990, 2000; Manuel et al, 2002). In many calcareous sponges (e.g., Sycon; Fig. 1B), the skeleton displays axial or radial symmetry around an axis along which the body is more or less morphologically polarized, but in others there is no visible axial symmetry (or only around the oscula). This character is particularly relevant with respect to body plan evolution in the Metazoa.

FIGURE 1. A few relevant morphological characters in the Calcispongia. (A-D) Organization of the aquiferous system. (A) Leucosolenia variabilis (Marseille, France), transverse section showing the asconoid and homocoel aquiferous system. (B) Sycon raphanus (Marseille, France), transverse section showing the syconoid aquiferous system. (C) Leuconia johnstoni (Roscoff, France), transverse section showing the leuconoid aquiferous system. (D) Sycon raphanus, detail of a region of the atrial wall illustrating the heterocoel organization of the aquiferous system (atrium lined with pinacocytes). (e) Schematic representation of the skeleton architecture in Sycon, showing in particular the articulate type of choanoskeleton (several rows of parallel triactines; see character 14). (F) Schematic representation of the skeleton architecture in the Jenkinidae, showing the inarticulate type of choanoskeleton (thin body wall and choanoderm supported by the unpaired actine of subatrial triactines; see character 14). (G) Paraleucilla sp. (Cannes, France; C61), transverse section showing the subcortical skeleton of giant tetractines shared by the Amphoriscidae (zone 1). In zone 2, the choanoskeleton consists in numerous triactines and tetractines scattered without order. (H) Sycettusa sp.

FIGURE 1. (Continued) (Yemen; C59), showing the inarticulate choanoskeleton and the subcortical skeleton of pseudosagittal triactines diagnostic for the Heteropiidae (see comments, character 17). (I) A pugiole from Leuconia sp. (Roscoff, France). This particular type of tetractine spicule is found in the walls of exhalant canals in some Baeriida (see comment, character 20). (J, K) Position of the nucleus in the choanocyte: basal (J) and apical (K) (L, M) Larval types: coeloblastula (L) and amphiblastula (M; in the amphiblastula the flagellated pole is anterior). According to the system of Bidder (1898), drawings J and L illustrate the characters of Calcinea and drawings K and M illustrate the characters of Calcaronea. Abbreviations: atr- atrial cavity; at sk = atrial skeleton; cc = choanocyte chamber; cho = choanocytes; ch sk = choanoskeleton; co = cortex; dc = distal cone; ext = exterior; pin = pinacocytes; subat sk = subatrial skeleton; subco pse = subcortical pseudosagittal triactine; subco tet = subcortical tetractine. Specimen numbers for G and H are from the collection at the Station Marine d'Endoume. Illustrations in E and F adapted from Borojevic et al. (2000), in J and K adapted from Minchin (1900), and in L and M adapted from Hartman (1958); other pictures are original.

The history of calcareous sponge taxonomy is characterized by a long-standing controversy about the broadest taxonomic subdivisions within the group, lasting even into the second half of the 20th century (Manuel et al., 2002). In his comprehensive monograph Die Kalkschwamme, Haeckel (1872) divided the calcareous sponges into three "natural" families based upon the organization of the aquiferous system: Ascones, Leucones, and Sycones. This system was soon criticized and rejected because, ironically, it proved to be artificial (Polejaeff, 1883; Dendy, 1891, 1893; Minchin, 1896). Particularly, several groups of species (latter recognized as families) displaying similar skeletal characters were split between Sycones and Leucones in the system of Haeckel. An alternative scheme, dividing calcareous sponge into Homocoela and Heterocoela, was later proposed by Polejaeff (1883) and was used by many authors during the 20th century. The basic idea underlying this classification was the primitive status of the homocoel aquiferous system (see Dendy and Row, 1913).

A few years later, Bidder (1898), following the work of Minchin (1896), proposed a radically different system in which the Calcispongia were subdivided in two subclasses, Calcinea and Calcaronea, distinguished primarily by the position of the nucleus in the choanocytes (basal with no relation to the nucleus in Calcinea, apical and arising from the nucleus in Calcaronea; see Figs. 1J, 1K) and deep embryological differences (e.g., coeloblastula in Calcinea, amphiblastula in Calcaronea; see Figs. 1L, 1M). Other proposed characters were the shape of triactine spicules (regular in Calcinea, sagittal and inequiangular in Calcaronea), the position of the optic axes of the spicules with respect to the walls of the sponge tubes (perpendicular in Calcinea, oblique and varying along the main body axis in Calcaronea), and the type of spicules to appear first during metamorphosis (triactines in Calcinea, diactines in Calcaronea). As advocated by Bidder (1898) and later by Hartman (1958), the strength of this new system was that it relied upon a combination of several characters; its major weakness was that most proposed characters were only known for a few taxa (Dendy and Row, 1913).

The Calcinea/Calcaronea subdivision was embraced during the 20th century by Hartman (1958) and Borojevic (1970, 1979; Borojevic et al., 1990, 2000) but was not universally accepted among sponge researchers. For example, Burton (1963) and Tuzet (1973) still used the older subdivision Homocoela/Heterocoela, considering the position of the nucleus in choanocytes and embryological characters to be homoplastic. The conflict was between a classification based upon characters of the aquiferous system and a classification based upon cytological and embryological characters.

Few phylogenetic schemes have been explicitly proposed for calcareous sponges (Haeckel, 1872; Polejaeff, 1883; Dendy and Row, 1913: Fig. 133; Borojevic, 1979). All of these schemes remain highly speculative because of the limited number of morphological characters available and because they rest upon preconceptions (Anderson, 1982), such as the primitiveness of the asconoid and homocoel aquiferous system. Until now, opinions about the classification of Calcispongia have been argued on pre-Hennigian grounds.

In this context, we first evaluated classification schemes proposed in the literature using morphological evidence through explicit coding and analysis of morphological characters. A molecular data set consisting of 18S ribosomal RNA (rRNA) sequences (of which 11 are new) was then analyzed independently and in combination with the morphological data set. Our goals were primarily to test the monophyly of Calcispongia, to test the two competing hypotheses for their broadest subdivision, and to formulate hypotheses about morphological character evolution (e.g., aquiferous system, axial symmetry) within the group.

This work has been made possible by the recent generic revisions of Calcinea (Borojevic et al, 1990) and Calcaronea (Borojevic et al., 2000), to which some of us have contributed. This revised taxonomic treatment has been completed by the same authors that contributed chapters on the calcareous sponges in Systema Porifera (Hooper and van Soest, 2002).

MATERIALS AND METHODS

Morphological Data Set

Virtually all available and potentially informative morphological variation above the generic level in Calcispongia have been scored cladistically (Appendix). The terminal taxa were 37 genera of Calcispongia, which we assumed a priori to be monophyletic. The taxonomic sampling included 8 of the 18 genera and 6 of the 10 families recognized within Calcinea by Borojevic et al. (1990) and 29 of the 56 genera and 12 of the 14 families recognized within Calcaronea by Borojevic et al. (2000). Table 1 includes taxonomic information about the genera included in the data set. As a general rule, we attempted to achieve a balanced representation of maximally separated taxa, as currently recognized in the Calcispongia, and we favored inclusion within the data set of genera for which detailed morphological descriptions and/or material was available.

Because no phylogenetic data are available at any level within Calcispongia, nonmonospecific genera were scored in a conservative way: characters varying within a genus were scored as polymorphic. Autapomorphies of terminal taxa, although noninformative, were kept in the present study because they represent essential information for future studies at lower taxonomic levels. In general, characters for which data were available in <60% of terminal taxa were not included in the data set, with the exception of character 22 because this recently discovered character (Worheide and Hooper, 1999) was particularly relevant with respect to the scope of the present study (Worheide and Hooper presented this character as strong new evidence in favor of the Calcinea/Calcaronea hypothesis).

No outgroup taxa were included in the morphological data set because the great majority of the characters are inapplicable outside the Calcispongia. This is particularly true for characters dealing with spicules and the architecture of the skeleton. Calcareous spicules of Calcispongia are clearly not homologous with siliceous spicules of Hexactinellida and Demospongiae (in addition to the difference in mineral composition, the siliceous spicules are secreted intracellularly and calcareous spicules are produced extracellularly), and their homology with calcareous spicules found in eumetazoans (e.g., octocorallians) is highly dubious (see Borchiellini et al., 2001). In consequence, trees derived from morphological data alone are unrooted.

TABLE I. Classification of the genera of Calcispongia included in the morphological data set, according to Borojevic et al. (1990, 2000, 2002a, 2002b, 2002c) and Vacelet et al. (2002a, 2002b). Genera for which 18S rRNA sequences are available are marked with an asterisk. Correspondence with the alternative classification scheme: Hom = Homocoela; Het = Heterocoela.

TABLE 2. Matrix of morphological data. Characters are described in the Appendix. A dash = inapplicable; ? = unknown; letters indicate polymorphism: a = 1/2; b = 0/1; c = 2/3; d = 0/1/2; e = 0/ 2.

The data sources, unless otherwise specified, are the taxonomic revisions at the generic level by Borojevic et al. (1990, 2000), added to by observations from a large number of histological preparations of specimens of the BMNH collection and from field collections. The nomenclature is the same as in Systema Porifera (Hooper and Soest, 2002). We followed the terminology of the Thesaurus of Sponge Morphology (Boury-Esnault and Rutzler, 1997). The annotated character list is given in the Appendix. The data matrix is given in Table 2.

Molecular Data Set

Sample collection and identification.-The species sampled and their collecting sites are listed in Table 3. Four species were collected at Heron Island (Great Barrier Reef, Australia) in July 1998 during the field excursion following the 5th International Sponge Symposium in Brisbane. After collection, samples were cleaned for removal of epibionts, and approximately 0.5 g was cut off from the interior of the specimen to avoid possible subsurface unicellular algal contamination. Tissues were fixed in 80% ethanol and stored at -20[degrees]C.

Specimens were systematically embedded in araldite after staining in acid fuschin and dehydration and sectioned to a thickness of about 20 [mu]m. The sections were then stained with toluidine blue and mounted. These preparations were used for identification at the genus level (mainly based on the work of Borojevic et al., 1990, 2000) and when possible at the species level (based on the work of Worheide and Hooper, 1999, for specimens from the Great Barrier Reef).

Genomic DNA extraction.-Before genomic DNA extraction, tissue samples were dehydrated and frozen in liquid nitrogen, and small pieces of frozen samples were ground to a powder in a precooled mortar with liquid nitrogen. Powder was poured into 500 [mu]l lysis buffer (10 mM Tris-HCl, pH 8, 0.1 M EDTA, pH 8, 20 [mu]/ml RNase DNase free, 0.5% SDS) and incubated for 1 h at 37[degrees]C. The mixture was then digested with Proteinase K (100 [mu]g/ml) for 3 h at 50[degrees]C. After digestion, the aqueous lysate was extracted with water-saturated ultrapure phenol, followed by a single chloroform extraction of the aqueous phase. Genomic DNA was recovered by a standard precipitation procedure with 0.1 volume of 3 M ammonium acetate (pH 7) and 2.5 volumes of absolute ethanol. Genomic DNA was finally resuspended in sterile water at 1 [mu]g/ [mu]l after optic density measurement at 260/280

m.

TABLE 3. Sequence sources and GenBank accession numbers. Sequences from this work are indicated with an asterisk.

Cloning of 18S rRNA genes,-The 18S rRNA genes were amplified by polymerase chain reaction (PCR) using primers 18S1 (5'- AACCTGGTTGATCCTGCCA-3') and 18S2 (5'-TGCAGGTTCACCTACAGAA-3'). The PCR protocol was the same as that of Borchiellini et al. (2001). PCR fragments were cloned into pGEM-T Easy Vector System I (Promega). Plasmid DNA was recovered using the nucleobond AX purification kit (MachereyNagel) and sequenced using the dideoxy-nucleotide chain termination method (Sanger et al., 1977). The 18S rRNA gene sequences resulting from this work have been deposited in the GenBank database (accession numbers listed in Table 3).

Sequence alignment and taxonomic sampling.Sequences from this work were aligned together with a number of 18S rRNA sequences available in GenBank (Table 3). All available 18S rRNA sequences from Calcispongia were included, with the exception of a sequence from Leucetta sp. (GenBank accession number AF084240), which is incomplete. Additionally, the 18S rRNA gene was sequenced from a specimen of Sycon ciliatum determined by us (from Roscoff, France). The sequence was identical (as expected) to that deposited in GenBank under the synonymous name Scypha ciliata, thus confirming the correct identification of the species. Also included in the alignment were outgroup sequences from Eumycota, Choanoflagelatta, Hexactinellida, Demospongiae, Cnidaria, Ctenophora, and Bilateria (see Table 3).

Alignments were first done automatically using the GeneWork program (Intelligenetics) and then manually for optimization of the alignment using the Sea View program package (Galtier et al., 1996). We followed the secondary structure model for Mnemiopsis of Podar et al. (2001). Indels were coded as binary characters in a separate matrix, using the simple indel coding method described by Simmons and Ochoterena (2000). With this method, all gaps that have different 5' and/or 3'termini are treated as separate presence/ absence characters. For a given gap character, sequences that have a longer over-lapping gap are coded as inapplicable for the character. Indels were coded only when primary homology of the 5' and 3' flanking sequences was unambiguous. Indels autapomorphic for one terminal taxon were not coded, and indels specific for the Bilateria were not taken into account. A total of 31 gap characters were retrieved by this method.

A recurrent problem in phylogenetic analyses of noncoding nucleotide sequences is the treatment of ambiguously aligned regions. Total exclusion of these regions has several disadvantages, as discussed by Lutzoni et al. (2000), the most serious being the loss of substantial phylogenetic information. Here we adopted the following strategy. First, when part of the sequence of a terminal taxon was not alignable with the other sequences, it was replaced by missing data codes (this occurred mostly with the sequences from fungi, choanoflagellates, and hexactinellids). second, in ambiguously aligned regions, when one or more subsets of the terminal taxa were unambiguously alignable, the sequences of each one of these subsets were conserved as a distinct block and the corresponding positions in the remaining sequences were replaced by missing data codes. Only blocks containing at least three terminal taxa were kept in the final alignment, and blocks containing only sequences from Bilateria were removed. In practice, this treatment was applied to variable domains V1, V2, V4, and V9 (corresponding to portions of the final alignment at positions: 70-86, 231-269, 691- 785, and 1726-1808, respectively). These blocks always contained all terminal taxa belonging to at least one of the higher level taxa included in the data set (e.g. Demospongiae, Calcispongia, Ctenophora, Cnidaria, Bilateria). Third, in alternative cases, ambiguously aligned regions were simply removed.

Using this alignment strategy, only positions where primary homology is unambiguous were kept, while minimizing the loss of information. The obvious drawback of this strategy resides in the introduction of missing data codes, which arguably could influence the topology. However, according to our observations this influence was negligible (when domains V1, V2, V4, and V9 were completely excluded, the general topology was not affected).

The final alignment contains 1,885 positions, of which 1,034 are constant and 636 are parsimony informative. The alignment is accessible at the Systematic Biology website.

Combined Molecular and Morphological Data Sets

Combined data set 1 was constructed by adding the 18S rRNA sequence data, the gap characters, and the morphological characters for calcareous taxa included in the molecular data set and 10 outgroup taxa (the demosponges Suberites and Spongilla, the two hexactinellids, the ctenophorans Hormiphora and Beroe, the cnidarians Cirripathes and Parazoanthus, and the bilaterians Antedon and Nerita). The three Sycon species were treated here as distinct terminal taxa (morphologically, they differ only with respect to character 9-relation between radial tubes-with state 1 in S. ciliatum and state 2 in S. calcaravis and S. raphanus). In the outgroup taxa, for demosponges and hexactinellids, all morphological characters were inapplicable except characters 1 (a special state 2 for Demospongiae because the nucleus is not clearly apical or basal but rather occupies a large volume more or less in the center of the cell, and a special state 3 for choanoblaste in hexactinellids), 2 (unknown for Suberites, a special state 2 for parenchymella in Spongilla, and a special state 3 for the hexactinellid larva), 3 (state 0), 5 (state 0), 6 (state 3), and 10 (state 2 for hexactinellids and state 3 for demosponges). For ctenophorans and cnidarians, all morphological characters were inapplicable excepted character 2 (special state 3 for planula in cnidarians and special state 4 for cydippid in ctenophorans). For bilaterians, all morphological characters were inapplicable.

Combined data set 2 was made by adding to combined data set 1 all genera of Baeriida and Lithonida (sensu Borojevic et al., 2000) that were not included in the former, data set 1, i.e., Leucopsila, Trichogypsia, and Lamontia (Baeriida) and Minchinella and Plectroninia (Lithonida). Molecular characters were coded as unknown for these five taxa.

Phylogenetic Analyses

All phylogenetic analyses were performed using PAUP* 4.0b6 (Swofford, 1998). For maximum parsimony (MP) analyses, characters were always treated as unordered. For MP analyses of molecular and combined data sets, characters were always treated as equally weighted; and for MP analyses of morphological data, characters were either treated as equally weighted or they were reweighted. Reweighting was performed (using the Reweight command in PAUP*) according to the maximal RC (rescaled consistency index) over 100,000 minimal trees (56 steps) from an initial analysis with equally weighted characters, of which the strict consensus was a complete polytomy involving the 37 terminal taxa. Characters with weights other than 1 (in parentheses) were 4 (0.25), 5 (0.0), 6 (0.4375), 12 (0.4166), 13 (0.45), 17 (0.33), 18 (0.33), and 24 (0.166).

MP trees were computed using heuristic searches with 50 replicates of random taxon addition sequence and tree bisection- reconnection branch swapping. For the maximum likelihood (ML) analysis of nucleotide data we used a general time reversible + G + I model because it is one of the most general models available for phylogenetic reconstruction from nucleotide data using ML. Among- site variation was estimated using a discrete approximation to the gamma distribution with eight rate categories. Parameters were estimated with PAUP* 4.0b6 from the result of an MP heuristic search. The estimated parameters were shape = 0.3726; pinvar = 0.1653; R-matrix = (1.25, 2.39, 0.97, 1.53, 4.99). The heuristic search was done with 20 replicates of random taxon addition sequence.

Gaps in molecular sequences and inapplicable morphological character states were treated as missing data (but see above for separate indel scoring). The analysis of molecular data using MP was done for the combination of the 1,885 sequence characters and the 31 presence/absence gap characters. Gap characters were not considered for the ML analysis. Branch support was tested with bootstrapping (Felsenstein, 1985) (500 replicates for MP, 100 replicates for ML) and computation of Bremer values (MP) (Bremer, 1988). Reconstitutions of character evolution were done using MacClade 3.0 (Maddison and Maddison, 1992).

Phylogenetic Nomenclature

Clade names are proposed here in accordance with the principles and rules of the Phylocode (provisional draft, http://www.ohiou.edu/ phylocode/). Only well-supported clades were named, and only preexisting names were used (conversion in the sense of the Phylocode, provisional draft, art. 9) to avoid the proliferation of names (see art. 2.2 and recommendation 9B). Names are employed in a sense close to their classical sense in the Linnean system (recommendation 10A). Stem-based definitions are employed when the resolution outside the clade is better than that inside, and node- based definitions are used when the resolution is better inside than outside the clade. Specifiers are species names.

RESULTS

Morphology Alone

To test previous taxonomic hypotheses against morphological evidence, we performed cladistic analyses of the morphological data. Because characters are treated as unordered and no outgroup taxa have been included (because most characters are inapplicable outside the Calcispongia), trees are unrooted. The analyses thus did not permit testing for monophyly of any grouping, but unrooted trees can be considered compatible or not with a given hypothesis.

Limited resolution was expected a priori from the data set because it co

tains only 25 characters (18 are parsimony informative) for 37 terminal taxa, and many character states are inapplicable or unknown. The lack of resolution results in a huge number of equally parsimonious shortest trees, and it was not possible to complete the heuristic search. With equally weighted characters, a strict consensus tree computed from 100,000 shortest trees (after search was aborted) shows an entire lack of resolution, with all 37 taxa falling in the same general polytomy.

Nevertheless, we continued to explore the structure of the data by character reweighting (according to the amount of homoplastic changes) and/or by reducing the taxonomic sampling (Figs. 2, 3). With characters reweighted according to the best RC among the 100,000 initial shortest trees, eight branches were resolved (strict consensus of 100,000 shortest trees from aborted heuristic search with all 37 terminal taxa included) (Fig. 2A). The same topology was retrieved from 20 distinct analyses starting from a random addition sequence of taxa, each one being stopped after 100,000 trees were saved. The unrooted tree is compatible with the Calcinea/Calcaronea hypothesis but not with the Homocoela/Heterocoela hypothesis. However, the Bremer value of the branch separating Calcinea from Calcaronea is weak (0.55), and an increase in tree length of only 0.73 is required to obtain topologies compatible with the Homocoela/ Heterocoela hypothesis.

Taxonomic sampling was then reduced to be able to compute complete heuristic searches. Starting from the preceding tree (Fig. 2A), terminal taxa were eliminated from the data set when (1) they branched to a polytomy and (2) they did not belong to the 18S rRNA taxonomic sample. This process resulted in the elimination of 17 terminal taxa. With this new taxonomic sample, heuristic searches could be completed, both from reweighted (Fig. 2B) and equally weighted (Fig. 2C) characters, and the tree topologies obtained from both types of analyses are the same except for one branch, which is found in the reweighted analysis (but with a very low Bremer value: 0.08) but not in the equally weighted analysis. These two analyses with reduced taxonomic sampling again support the separation between Calcinea and Calcaronea (with a Bremer value of 1 in both analyses), and the number of supplementary steps required for topologies compatible with a separation between Homocoela and Heterocoela are 1.18 and 1 in the reweighted and the equally weighted analyses, respectively.

Internal relationships within the two subclades are poorly resolved. In the tree shown in Figure 2A, the ratio (number of branches/number of taxa) is 0.25 in Calcinea and 0.17 in Calcaronea. Nevertheless, the classification scheme currently in use for the Calcispongia (see Table 1) can be evaluated in light of these results. No orders were monophyletic. Within Calcinea, the unrooted topology is compatible however with the monophyly of Clathrinida (Fig. 2A), the remaining order Murrayonida being represented by only one terminal taxon, Murrayona. Within Calcaronea, the monophyly of Baeriida was rejected in the reweighted analysis of the full data set (Fig. 2A; monophyly could not be rested in other analyses), but the topology is compatible with the monophyly of an extended Baeriida including Petrobiona. This position of Petrobiona implies that Lithonida is not monophyletic (Figs. 2A-C). In analyses with reweighted characters (Figs. 2A, 2B), this region of the trees (containing the representatives of Baeriida and Lithonida) is where branches display the highest Bremer values. The monophyly of Leucosoleniida was rejected in Figure 2B but not in Figures 2A and 2C. At the family level, the analyses supported the monophyly of Minchinellidae (Figs. 2A-C), and Staurorrhaphidae (Figs. 2A-C) (provided that the root is located outside of these clusters). However, the monophyly of Baeriidae was rejected (Fig. 2A; not tested in other analyses), and the monophyly of Grantiidae and Heteropiidae was rejected in Figure 2B but not in other analyses. For all remaining families, monophyly was neither supported nor rejected.

FIGURE 2. Analyses of morphological data. Dots indicate nodes that are strictly compatible with the monophyly of taxa recognized in the current system (see Table 1); 1 = Minchinellidae; 2 = Baeriida + Petrobiona; 3 = Staurorrhaphidae. (A) Strict consensus of 100,000 shortest trees resulting from an analysis including all terminal taxa, with characters reweighted according to the best RC from the initial analysis: 36.99 steps; consistency index (CI) = 0.752; homoplasy index (HI) = 0.248; retention index (RI) = 0.867. Terminal taxa branching to the main polytomy have not been labeled (Lelapia, Grantiopsis, Grantia, Ute, Leucandra, Aphroceras, Jenkina, Leucascandra, Anamixilla, Uteopsis, Heteropia, Sycettusa, Vosmaeropsis, Leucilla, Paraleucilla, Amphoriscus). Numbers on the branches in A and B are Bremer values (which are not integer values because of character reweighting). (B) Strict consensus of the 81 shortest trees obtained after elimination of terminal taxa that branch to a polytomy in the preceding tree and were not sampled in the 18S rRNA data set; characters have been reweighted as in the preceding analysis (25.08 steps; CI = 0.830; HI = 0.170; RI = 0.873). (C) Strict consensus of the 162 shortest trees obtained with the same taxonomic sampling as in B but with characters equally weighted (36 steps; CI = 0.722; HI = 0.278; RI = 0.787). All branches have a Bremer value of 1.

FIGURE 3. Tree resulting from the analysis of morphological data alone, with only those taxa for which 18S rRNA data were available. Characters were equally weighted. This is the strict consensus of 10,258 shortest trees (31 steps). Numbers on the branches are Bremer values (CI = 0.7742; HI = 0.2258; RI = 0.8000).

We also computed a tree including only those taxa for which 18S rRNA sequences were available (Fig. 3). The three Sycon species have been treated here as three distinct terminal taxa, as in the combined data sets. Only two branches were resolved in this tree (Fig. 3), including the branch linking Leuconia and Petrobiona. The topology is compatible with both the Homocoela/Heterocoela and the Calcinea/Calcaronea hypotheses. It is also compatible with the trees in Figure 2.

18SrRNA Alone

New complete 18S rDNA sequences were obtained from 11 species of calcareous sponges belonging to 11 distinct genera and 9 distinct families, increasing the available data set to 17 species belonging to 13 families (see Tables 1, 3). The taxonomic representation is better in the subclass Calcaronea (9 sampled families over 15 existing) than in the subclass Calcinea (4 sampled families over 9 existing).

The results of the analyses of our 18S rRNA data set, including the 17 calcareous sponge sequences and 29 outgroup sequences, is shown in Figure 4. The monophyly of Calcispongia was strongly supported by bootstrap values (100% with both ML and MP) and by the Bremer value (20) of the branch leading to the Calcispongia in the MP tree (Fig. 4). By comparison, the monophyly of Demospongiae was supported by only 56% (ML) and 73% (MP) of bootstrap values and a Bremer value of 2; the monophyly of Cnidaria was not supported by MP and weakly supported by ML. Hexactinellida, Ctenophora, and Bilateria have the highest Bremer values (54, 68, and 68, respectively).

In the optimal tree generated by ML, the sponges are paraphyletic, with the Calcispongia clustering as the sister group of a monophyletic Eumetazoa, but this grouping was not well supported (bootstrap value <50). In contrast, in the strict consensus of shortest MP trees (Fig. 5A), the sponges (Porifera) are monophyletic, and the Calcispongia is the sister group of the Silicispongia (Hexactinellida + Demospongiae). This monophyly of sponges is however only weakly supported by a bootstrap value of 55 and a Bremer value of 1. The Eumetazoa (Ctenophora, Cnidaria, Bilateria) are not monophyletic in this tree. Ctenophorans are the sister group of Porifera, with very weak statistical support, and Cnidaria + Bilateria is the sister group of Ctenophora + Porifera. The sister-group relationship between Cnidaria and Bilateria was strongly supported in both analyses. The topology of the MP tree was exactly the same when the gap-coded characters were excluded (data not shown).

We also explored the effect of topological constraints (Fig. 5B) on the MP tree. When constrained for the monophyly of Eumetazoa (Fig. 5B, bootstrap values above the branches), the monophyly of sponges was not retrieved: Calcispongia is the sister group of Eumetazoa (but this relation was weakly supported by 56% of bootstraps), and Hexactinellida is the sister group of the remaining metazoans (but also with low bootstrap values). When constrained for the monophyly of Calcispongia + Ctenophora + Cnidaria + Bilateria, the Eumetazoa are monophyletic with relatively high support (87% of bootstraps), and the Hexactinellida is the sister group of a clade containing all other metazoans, supported by a bootstrap value of 81%. Both of these constraints generate shortest trees only one step longer than the shortest tree with no constraint.

The Calcispongia was subdivided into two subclades (Fig. 4), consistent with the two subclasses Calcinea and Calcaronea (see Table 1). The Calcinea subclade was supported by 100% (ML) and 98% (MP) of bootstraps and by a Bremer value of 7. The Calcaronea subclade was supported by 100% (ML and MP) of bootstraps and by a Bremer value of 24. Constraining the MP search for monophyly of both Homocoela (Clathrina, Soleneiscus, and Leucosolenia) and Heterocoela (Leucetta, Leucaltis, Sycon spp., Vosmaeropsis, Sycettusa, Leuconia, Petrobiona, Grantiopsis, Paraleucilla, Grantia, Leucandra, and Anamixilla) required 71 additional steps. Constraint for the monophyly of Homocoela alone required 65 steps, and constraint for the monophyly of Heterocoela alone required 69 steps. Thus, the molecular data unambiguously support the Calcinea/Calcaronea hypothesis and reject the Homocoela/Heterocoela hypothesis. In the ML tree (Fig. 4), the branches leading to Calcispongia, Calcinea, and Calcaronea are longer than those inside the clades Calcinea and Calcaronea.

FIGURE 4. Analyses of the 18S rRNA data set. The tree shown is the ML tree (-In likelihood = 15159.18). The scale bar indicates the number of nucleotide substitutions per position in the sequence. ML bootstrap values (when >50%) are given below the branches. MP bootstrap values (when >50%) and Bremer values (in parentheses) are given above the branches. Bremer values have been estimated for the monophyly of only high-level groups. For clarity purpose, bootstrap values are not indicated within Calcinea and Calcaronea (see Fig. 6).

FIGURE 5. Summary of the results from MP analyses of the 18S rRNA data set. (A) Unconstrained analysis (strict consensus of 24 shortest trees, 2,599 steps; CI = 0.527; Hl = 0.473; RI = 0.732). Bootstrap values are indicated above the branches. (B) Analyses constrained for monophyly of Eumetazoa (bootstrap values above the branches) and for monophyly of Calcispongia + Eumetazoa (bootstrap values below the branches). Both of the constrained analyses yielded the same topology (strict consensus of 24 shortest trees; 2,600 steps; CI = 0.527; HI = 0.473; RI = 0.731).

Figure 6 shows in more detail the relationships among terminal taxa in the Calcispongia based on MP and ML analyses. In the 18S rRNA data set, 113 positions were informative within Calcispongia, 8 within Calcinea, and 41 within Calcaronea.

There is disagreement between MP and ML analyses about the relationships within Calcinea. In the MP analysis, Soleneiscus is basal to a clade consisting of the three remaining genera supported by a bootstrap value of 82% and a Bremer value of 3, whereas in the ML analysis, Leucetta is the basalmost terminal taxon in CaIcinea (but without significant support). Within the CaIcaronea, there is total topological congruence between the results of MP and ML analyses. The only asconoid representative, Leucosolenia, does not appear basal. The most well-supported monophyletic groups are Heteropiidae (Vosmaeropsis + Sycettusa), (Leuconia + Petrobiona), (Grantiopsis + Paraleucilla), and (Anamixilla + Leucandra + Sycon raphanus). The genus Sycon is polyphyletic according to these results. The species Sycon ciliatum groups with the Heteropiidae but with only moderate support. The family Grantiidae (Grantia and Leucandra) is not monophyletic nor is the order Leucosoleniida (all Calcaronea in this data set excepted Leuconia and Petrobiona).

FIGURE 6. Relationships within Calcispongia as inferred from 18S rRNA. (A) MP tree. The outgroup taxa (not shown) are the same as in the combined analyses. This is the strict consensus of 48 shortest trees (1,322 steps). Bootstrap values (when >50%) are indicated above the branches; Bremer values are indicated below the branches (CI = 0.713; HI = 0.287; RI = 0.746). (B) ML tree (same analysis as in Fig. 4). Bootstrap values (when >50%) are indicated above the branches.

Combined Morphology and 18S rRNA

The morphological data set reduced to the taxa sampled for 18S rRNA contains only two parsimony-informative characters within the Calcinea and eight parsimony-informative characters within the Calcaronea. To determine whether these few characters could influence topology and/or branch support when added to the 8 (Calcinea) and 41 (Calcaronea) parsimony-informative positions in the 18S rRNA data set, we performed an analysis of combined data set 1 (Fig. 7). This tree should be compared with the consensus MP tree resulting from 18S rRNA alone with the same taxon sample (Fig. 6A). The resolution within both Calcinea and Calcaronea is better in the combined than in the molecular strict consensus tree, i.e., four additional clades, absent in the molecular tree, were recovered in the combined tree. Within Calcaronea, the topology obtained from the analysis of combined data set 1 is identical to that of the optimal ML tree from the molecular data (Fig. 6B).

When comparing branch support estimates between the 18S rRNA alone and the morphology + 18S rRNA trees, most clades have similar support in the two analyses. However, branch support is significantly lower with combined data than with molecular data alone for clades (Leucosolenia + Sycon calcaravis + S. ciliatum + Vosmaeropsis + Sycettusa + Leuconia + Petrobiona), (S. ciliatum + Vosmaeropis + Sycettusa), and (S. raphanus + Anamixilla + Leucandra), but for the latter clade it remains high in the combined analysis (81%). In contrast, there is strong increase in branch support for Leuconia + Petrobiona with the addition of morphological data: bootstrap value increased from 84% to 98% and Bremer value increased from 2 to 6.

In the strict consensus tree from the combined analysis, as in the molecular trees, Leucosolenia is not in a basal position among Calcaronea, the genus Sycon is polyphyletic, the family Grantiidae is not monophyletic (at least paraphyletic), and the order Leucosoleniida is paraphyletic.

Given that the sister-group relationship between Leuconia (order Baeriida) and Petrobiona (order Lithonida) was the most well- supported result from the combined analysis, we further analyzed relationships between these two orders and tested the monophyly of the recently created order Baeriida (Borojevic et al, 2000) by including in the combined data set the remaining taxa of Baeriida (Leucopsila, Trichogypsia, and Lamontia) and Lithonida (Minchinella and Plectroninia), which were not sampled for 18S rRNA (combined data set 2). The result is shown in Figure 8. Taxa belonging to Baeriida and Lithonida altogether form a monophyletic unit. Monophyly of Baeriida was recovered in the strict consensus tree, but Petrobiona was included in this clade. The analysis also resolved the clade Leuconia + Petrobiona + Lamontia within this extended Baeriida, leaving Baeriidae not monophyletic. Monophyly of Minchinella + Plectroninia (Minchinellidae) was recovered. These results are consistent with the results from the analysis of morphological data alone (Fig. 2A).

Evolution of Morphological Characters within Calcispongia

Morphological synapomorphies as reconstructed by MacClade were plotted on the strict consensus trees derived from combined morphological and molecular analyses (Figs. 7, 8). Diagnostic characters for Calcinea and Calcaronea cannot be polarized because the outgroup criterion is inapplicable. Hence, it is not possible to propose unambiguous morphological synapomorphies for Calcinea or Calcaronea.

FIGURE 7. Result of the combined morphology and 18S rRNA analysis (combined data set 1). This tree is the strict consensus of three shortest trees (1,371 steps). Bootstrap values (when >50%) are indicated above the branches; Bremer values are indicated below the branches. Synapomorphies (character followed by character state in parentheses) are indicated on the branches. Autapomorphies are not indicated (CI = 0.713; HI = 0.287; RI = 0.746).

Also within Calcinea and Calcaronea, very few morphological synapomorphies can be proposed (Fig. 7). Most of the clades have no unambiguous morphological synapomorphies; this is true in particular for Grantiopsis + Paraleucilla and Anamixilla + Leucandm + Sycon raphanus, both of which are well-supported in the combined analysis. In Calcinea, the clade Leucaltis + Leucetta has one synapomorphy: the presence of a cortex (character 12, state 1, also present in numerous calcaronean taxa). In Calcaronea, the clade Vosmaeropsis + Sycettusa (the Heteropiidae) has one unambiguous synapomorphy: a continuous layer of pseudosagittal triactines located beneath the body surface (character 17, state 1, a diagnostic character). The clade Leuconia + Petrobiona has four synapomorphies in this tree.

As reconstructed on the tree resulting from the analysis of combined data set 2 (Fig. 8), the clade Baeriida + Lithonida has two synapomorphies: absence of a spacious atrial cavity (character 10, state 3) and absence of axial symmetry (character 13, state 0). The extended Baeriida has one synapomorphy: microdiactines abundant in the whole body or at least in the choanoskeleton (character 21, state 1, a diagnostic character). The clade Leuconia + Petrobiona + Lamontia has one synapomorphy: exhalant aquiferous system lined with pugioles (character 20, state 1, a diagnostic character). The emended Lithonida has two synapomorphies: a reinforced choanoskeleton of fused spicules (character 14, state 4, diagnostic) and the presence of diapasons (character 24, state 1, also present in Petrobiona, Lelapia, and Murrayona). Petrobiona has three autapomorphies (tracts of storage cells, character 3, state 1; an aspicular calcareous skeleton consisting of a basal mass of calcite, character 15, state 2; and diapasons, character 24, state 1), of which the latter is homoplastic with Lithonida.

The history of four important morphological characters as optimized by MacClade under the principle of parsimony is shown in Figure 9. It is not possible to determine whether the coeloblastula and the amphiblastula originated independently (from an unknown ancestral larval type) or whether one is derived from the other (Fig. 9A). However these characters are diagnostic for Calcinea and Calcaronea, respectively. It is not possible to infer the primitive state of the organization of the aquiferous system (Fig. 9B). For this character, coding asconoid or syconoid as the ancestral state costs eight steps over the tree, whereas coding leuconoid as the ancestral state costs one step more (but the outgroup criterion would favor the leuconoid aquiferous system as ancestral). If the leuconoid state is not coded as ancestral, the topology implies that it was acquired five times independently,three times unambiguously from the syconoid state (in the branches leading to Vosmaeropsis, Paraleucilla, and Leucandra). The topology does not allow polarization of the presence of a cortex (Fig. 9C). If the cortex were ancestrally absent, it must have been acquired at least four times independently (one time in Calcinea and three times in Calcaronea) and then secondarily lost (reversion) in the branch leading to Sycon raphanus. If the cortex were ancestrally present, then it must have been lost at least five times (including in the branch leading to S. raphanus). Ax