Alignment of homeodomains included in the phylogenetic analyses. All sequences are aligned to the Drosophila Antennapedia homeodomain. Each Nematostella sequence is grouped with putative bilaterian homologs. The degree of statistical support (bootstrap proporation [BP] or posterior probability [PP]) for each of these homology assignments is indicated separately for the neighbor-joining (NJ), Bayesian (Bayes), and maximum-likelihood (ML) trees. The average statistical support for each grouping is indicated in the far right column (NJ-BP+Bayes-PP+ML-BP/3). The dataset is available in Phylip format as Figure S12.

Homeodomain phylogeny based on neighbor-joining. The cladogram is rooted using the dll sequences. Nematostella sequences are shown in red. Bilaterian sequences are shown in black. Bootstrap proportions are presented at each node. Open circles depict implied gene losses for Nematostella (red) and Branchiostoma (black). Closed circles depict implied lineage-specific gene duplications for Nematostella (red) and Branchiostoma (black). The dataset used in this analysis is available as Figure S12.

Clusters of Hox-related homeobox genes in the Nematostella genome. Based on the current genomic assemblies, thirty ANTP class genes of Nematostella are distributed among seven homeobox clusters [52]. The location of the PRD class gene Dmbx is also shown. The arrangement of the Nematostella genes is shown in relation to the hypothesized "extended Hox cluster," that is presumed to have existed in the most recent common ancestor of protostomes and deuterostomes (gray box; [127], [128]). Horizontal lines connecting Nematostella genes indicate known genomic linkage. Double-arrows connect Nematostella homeodomains to their putative bilaterian homologs based on phylogenetic analyses of homeodomain sequences (Figure 1; supp figs. 1–2). Detailed diagrams of each of the eight Nematostella homeodomain clusters are presented in supplemental figures 3–10.

Developmental Expression of Hox-ParaHox related genes in Nematostella. Gene expression was assayed throughout embryonic and larval development using in situ hybridization. All images are optical sections that permit visualization of the endodermal tissue layer. Panels M and V are transverse sections, but all other images are longitudinal sections with the future oral end of the animal facing left. The blastopore (site of the future mouth) is indicated by an asterisk. Abbreviations are as follows: apical tuft (at); coelenterone (coe); bodywall ectoderm (ecbw); pharyngeal ectoderm (ecph); bodywall endoderm (enbw); pharyngeal endoderm (enph); mesentery (mes); pharynx (pha); tentacle (tn).

Developmental Expression of Hox and ParaHox homologs in Nematostella. Gene expression was assayed throughout embryonic and larval development using in situ hybridization. All images are optical sections that permit visualization of the endodermal tissue layer. Panels A, J, M, P, S, and Y are transverse sections, but all other images are longitudinal sections with the future oral end of the animal facing left. The blastopore, which becomes the mouth, is indicated by an asterisk. Abbreviations are as follows: apical tuft (at); coelenterone (coe); bodywall ectoderm (ecbw); pharyngeal ectoderm (ecph); bodywall endoderm (enbw); pharyngeal endoderm (enph); mesentery (mes); pharynx (pha); tentacle (tn).

Mean statistical support for select phylogenetic groupings. The mean statistical support for hypothetical clades linking each of six Nematostella Hox/ParaHox lineages with potential homologs in the Bilateria is depicted graphically. The mean statistical support is the average of the neighbor-joining bootstrap proportion, the Bayesian posterior probability, and the maximum-likelihood bootstrap proportion. It is expressed as a percent of trees in which the given grouping was recovered. Individual Nematostella genes were grouped into lineages (for example, anthox1-anthox1a) when the mean statistical support for the clade uniting them exceeded the mean statistical support for any other competing relationship. The final column depicts the highest support obtained for any pairwise relationship with another cnidarian homeodomain.

Hox/ParaHox evolutionary scenarios. The phylogenies drawn here depict six mutually exclusive scenarios regarding the evolution of the Hox and ParaHox genes. Ten distinct Hox and ParaHox lineages are thought to have been present in the ancestral bilaterian (Hox1, Hox2, Hox3, Hox4, Hox5, Hox6–8, Hox9+, Cdx, Gsx, and Xlox). Five distinct Hox/ParaHox lineages are recognized for Nematostella. (Nematostella homeodomains that tend to cluster together in the phylogenetic analyses are grouped together here: anthox1/1a; anthox2/9; anthox6/6a; anthox7/8a/8b.) Assuming no gene loss in the Cnidaria, then the existence of five Hox/ParaHox lineages in Nematostella implies that the cnidarian-bilaterian ancestor (CBA) could have possessed as few as one Hox/ParaHox gene (scenario A) or as many as five (scenario E). There is some indication that a central class Hox gene was lost in the Cnidaria [47], and that the CBA may have possessed six distinct Hox/ParaHox genes (scenario F). The ancestral Hox/ParaHox genes present in the CBA are indicated by solid squares. If a particular hypothetical clade is recovered on one or more of the phylogenetic analyses presented here, this is indicated above the relevant branch (NJ = neighbor-joining, Figure 2; Ba = Bayesian inference, Figure S1; ML = maximum- likelihood, Figure S2; φ = none). Below each branch, the average statistical support is indicated (NJ-bootstrap proportion+Bayes-posterior probability+ML-boostrap proportion/3). Each scenario implies a different number of lineage-specific gene losses.

Reconstruction of Hox2 evolution. Portions of the neighbor-joining tree, the maximum-likelihood tree, and the Bayesian tree are redrawn here. The neighbor-joining tree implies that anthox7, 8a, and 8b are direct descendants of the ancestral Hox2 gene in the cnidarian-bilaterian ancestor. No gene loss is required. The maximum-likelihood tree implies that both a Hox2 precursor (square) and an anthox7/8a/8b precursor (circle) were present in the cnidarian-bilaterian ancestor. Hox2 was lost in the line leading to Nematostella, while anthox7/8a/8b ortholog was lost in the line leading to Bilateria. The Bayesian tree implies that a Hox2 precursor (square), a Hox1 parecursor (triangle), and an anthox6/6a precursor were present in the cnidarian-bilaterian ancestor. Hox2 was lost in the line leading to Nematostella, while anthox6/6a was lost in the line leading to Bilateria.

Re-analysis of the Kamm et al. maximum-likelihood phylogeny. (A) The redrawn maximum-likelihood phylogeny presented by Kamm et al. [49]. The tree is unrooted. Based on the Dayhoff-PAM1 substitution matrix [129] the tree's overall likelihood is −2288.08809. Bootstrap proportions determined in the original study are shown above the relevant nodes. Bootstrap proportions determined in the current study are shown below the relevant nodes. Inferred lineage specific gene losses are represented by open circles (black for Bilateria and red for Cnidaria). Inferred lineage specific gene duplications are represented by solid circles (black for Bilateria and red for Cnidaria). Sequences inferred to have been present in the common ancestor are indicated by lettered squares. The names of gene families that contain cnidarian representatives are enclosed by red lines. (B) A tree we identified using global instead of local rearrangements with the Kamm et al. data and the Dayhoff substitution matrix. The tree's overall likelihood is −2280.76406. Bootstrap proportions determined in the present study are shown below the relevant nodes. The dataset used in this analysis is available as Figure S13.

Phylogenetic mapping of Hox expression. The neighbor-joining and Bayesian phylogenies (Figure 2 and S1) were pared to remove all bilaterian sequences. The strict consensus topology shown here depicts the relative relationships among Nematostella sequences. Each of the Nematostella Hox-related sequences is coded according to whether its expression is restricted along the primary (O/A) body axis or the secondary (directive) body axis (Y = yes; N = no). A yellow Y in the directive column signifies that the expression is bilateral (both sides of the directive axis), and a red Y indicates that the expression is unilateral. The character state found in the terminal taxon is indicated in the colored boxes. The internal nodes are shaded to indicate the character states found in hypothetical ancestors. For each gene, the spatial expression is depicted on a diagram of the juvenile polyp. In the case of Dlx, anthox6a and anthox1, the expression pattern that is depicted actually occurs earlier, in the larval stage, but it is represented on a diagram of the polyp to facilitate spatial comparisons with the other genes. The polyp is drawn in lateral view with the overlying ectoderm (dark gray) partially peeled away to reveal the underlying endoderm of the body column (light gray) and the lumen of the pharynx (white). The pharynx is drawn as though everted. Only one representative tentacle is shown. The mesoglea, a largely acellular layer of connective tissue that separates the endoderm from the ectoderm, is depicted as a thin black line. Gene expression is depicted as black shading in the endoderm or ectoderm. The major regions along the primary body axis are demarcated with dotted lines: Ph = pharynx; H = head; C = column; F = foot. Cross-sectional views through the body column (at the arrowheads) are shown for Gbx, anthox7, anthox8a, anthox8b, anthox6a, anthox1a, and NVHD065.

Origins of the Hox and ParaHox clusters. (A) The tandem duplication model of Hox and ParaHox clusters. Our evidence conflicts with the current theory that a multigene duplication event formed the Hox and ParaHox cluster. Under this alternative model, each Hox and ParaHox gene was formed through a series of tandem duplications either in the 3′ or 5′ direction and a translocation event separated the two ParaHox genes from the Hox cluster. Hox3 and Central Hox genes are not included in this model because it is not clear whether they were present in the cnidarian-bilaterian ancestor and subsequently lost in the Cnidaria or alternatively, were derived in the Bilateria from subsequent tandem duplication events of another Hox/ParaHox gene (for example anthox7/8a/8b or anthox1/1a). The general scenario presented here is robust enough to easily accommodate either event. (B) In this scenario adapted from Chourrout et al. 2006 Figure 3, a two-gene ProtoHox cluster is duplicated to form two sister clusters, the Hox and ParaHox clusters [51]. Chourrout et al. found that NVHD065 (Xlox/Cdx in their analysis) showed conflicting affinities for Xlox and Cdx; similarly our neighbor-joining and Bayesian trees had NVHD065 grouping with Cdx while the maximum-likelihood tree had NVHD065 grouping with Xlox. Unlike our study, Chourrout et al. considered anthox1 and anthox1a (HoxE and HoxF in their analysis) to be non-Hox/ParaHox lineages, hence their omission from their model [51].