In total, 33 PRD-class homeobox genes were previously identified in the Nematostella genome 32 by conducting BLAST searches of a draft genome assembly (available at StellaBase; http://www.stellabase.org/ 33 ). Phylogenetic analysis of homeodomain sequences by Bayesian and neighbor-joining methods identified single representatives of the homeobrain (HBN_Nv079), orthopedia (OTP_Nv047) and rx (RX_Nv129) families in Nematostella 32 . Subsequently, we designed gene-specific primers to amplify the 3' and 5' ends of the hbn, otp and rx transcripts from cDNA (Table 1). Overlapping 5' and 3' RACE (rapid amplification of cDNA ends) fragments were conceptually spliced to reconstruct complete transcripts. In addition, hbn, otp and rx transcripts were also identified among 150,000 Nematostella expressed sequence tags (ESTs) sequenced as part of the genome-sequencing project recently completed by the Joint Genome Institute 34 . Once the full-length transcripts were assembled, the longest open reading frames (ORFs) in frame with the highly conserved homeodomain were inferred using MacVector V.7.2.3 (Accelrys Inc., San Diego, CA, USA).

Human and fruit fly (Drosophila melanogaster) homeodomains from the PRD class were taken from a previous study 32 . We searched for hbn, rx and otp homeodomains in the partially or wholly sequenced genomes of four additional vertebrates (Mus musculus, Gallus gallus, Xenopus tropicalis and Danio rerio), four non-vertebrate deuterostomes (Ciona intestinalis, Branchiostoma floridae, Saccoglossus kowalevskii and Strongylocentrotus purpuratus), three additional insects (Anopheles gambiae, Apis mellifera and Tribolium castaneum), one nematode (Caenorhabditis elegans), one mollusk (Lottia gigantea), one annelid (Capitella teleta) and a placozoan (Trichoplax adhaerens). Database searches were conducted via Entrez http://www.ncbi.nlm.nih.gov/gquery/gquery.fcgi using gene names (orthopedia, rx/rax, homeobrain), and similarity searches were performed using the tBLASTn and BLASTp search algorithms. The homeodomains of hbn, otp and rx from Nematostella, human and fruit fly were used as the query sequences. Sequences that matched the query sequence were then subjected to reciprocal tBLASTn and/or reciprocal BLASTp similarity searches against the genome from which the query sequence was derived. We formally tested for orthology with phylogenetic analyses (see below).

Predicted homeobox sequences for Nematostella hbn, otp and rx were mapped to the publicly available assembly of the Nematostella genome (DOE-Joint Genome Institute (JGI) Nematostella vectensis genome assembly 1.0; http://genome.jgi-psf.org/Nemve1/Nemve1.home.html) using BLASTn to determine whether the hbn, rx and otp loci of Nematostella are linked as in Drosophila 31 . Gene structures were determined by mapping transcripts for hbn, rx and otp against the genome. To identify possible non-homeodomain genes that might be closely linked to hbn, rx and otp, the regions flanking these genes were compared against the RefSeq database using BLASTx (V.2.2.15).

The genomic organization of the hbn, rx and otp cluster in Drosophila has been described previously 31 . In other species in which two or three of these PRD-class genes were identified, the genes were mapped to the corresponding assembled genome to determine if they might be linked. The level of coverage for different assemblies varies widely, from completed genome assemblies for human, mouse, fruit fly and nematode, to assemblies at the level of scaffolds, contigs and linkage groups for numerous other animals. Therefore, we have uncertainty for linkage of some genes that are present in the genome because current assemblies limit absolute characterization of location. For species in which at least two of the three genes matched the same scaffold, linkage group or chromosome, we determined the relative position and transcriptional direction of each gene from the cluster.

Hbn, otp, rx and other PRD-class homeodomain sequences were aligned using a web implementation of the computer program Muscle 35 . The default parameters were used, and the alignment algorithm did not introduce any alignment gaps. Phylogenetic analysis was performed using the neighbor-joining method as implemented in the computer package Phylip (V.3.61; 36 ). In total, 111 PRD-class homeodomains were used in the phylogenetic analysis, including 33 from Nematostella, 39 from human and 24 from fruit fly, which had been identified in a previous study 32 . Human and fruit fly homeodomain families comprising multiple invariant or nearly invariant homeodomains were pared to a single representative. Select homeodomains were added to provide broader representation from deuterostomes and protostomes (Deuterostomia: Branchiostoma floridae (otp, rx), Strongylocentrotus purpuratus (otp, rx, hbn), Saccoglossus kowalevskii (otp, rx); Protostomia: Caenorhabditis elegans (rx), Patella vulgata (otp), Platynereis dumerilii (rx) and Capitella teleta (hbn). Distalless and hox1 homeodomains from Drosophila melanogaster and Branchiostoma floridae were included as outgroups. Homeodomain sequences and accession numbers for homeobrain, otp and rx are provided in Figure 1. Sequences and accession numbers for the other PRD-class homeodomains used in the phylogenetic analysis are available from Ryan et al. 32 . Pairwise distances between homeodomains were calculated using the PAM distance matrix. Support for individual nodes was assessed using 1,000 replicates of the bootstrap 36 . Amino acid substitutions were localized to particular branches on the phylogeny using MacClade V.4 37 .

To confirm the existence and determine the geographic distribution of two unusual polymorphisms at positions 52 and 59 within the homeodomain of Nematostella Rx, we designed primers to amplify a 585 bp fragment of the gene (Table 1). PCR was carried out using 0.5 U Taq DNA polymerase, 1 × Mg-free PCR buffer, 1.7 mM MgCl₂, 0.67 mM dNTPs, 1 μm of each primer and approximately 5 ng genomic DNA in 20 μl reactions. Thermal cycling consisted of denaturation for 5 minutes at 94LC, followed by 30 cycles of 30 seconds at 94°C, 30 seconds at 60°C and 1 minute at 72°C, with a final extension of 15 minutes at 72°C. PCR products were directly sequenced (RX_for; Macrogen, Seoul, South Korea). All polymorphisms were verified by eye.

Digoxygenin-labeled riboprobes were generated from 3' RACE products for each gene (homeobrain, 749 nt; orthopedia, 1556 nt; rx, 1419 nt). Each probe included a portion of the homeobox, plus a region of highly divergent coding sequence downstream of the homeobox and the 3' untranslated region (UTR). Whole-mount in situ hybridization was performed on embryos and larvae using published protocols 38 . Stringent hybridization conditions were employed (65°C for 20-44 hours using probe concentrations of 1.0 ng/μl).

Using RACE PCR, we cloned and annotated full transcripts for otp, rx and hbn from Nematostella (accession numbers HM004556-8). In addition, we queried the EST databases for Nematostella to further annotate these genes and to identify non-synonymous polymorphisms in each gene.

The otp transcript is 1901 nucleotides long, and encodes a predicted protein of 291 amino acids. When mapped to the genomic sequence, this transcript spans 9587 nucleotides, including four exons (see Additional file 1). We identified two putative full-length otp transcripts that encode different splice variants in Nematostella. Comparing the two splice variants of the Nematostella otp gene, one lacked a highly conserved transcriptional repression domain that was present in the other and that is typical of otp orthologs from other taxa. Neither splice variant that we have identified encodes an OAR (otp, aristaless and rax) domain. However, a highly conserved version of this domain is predicted to be encoded by a stretch of nucleotides that is located immediately downstream and in frame with exon 3 (see Additional file 1, boxed sequence). The predicted OAR motif of Nematostella Otp is highly conserved relative to deuterostome sequences; in fact, it is identical to both the human and sea urchin sequences at 15 of 16 residues. By contrast, in protostome orthopedia proteins, only the central core of the OAR domain appears to be conserved (for example, SIAALRRRA in Drosophila), or the OAR domain appears to have been lost entirely (for example, in Aedes aegypti or Anopheles gambiae). Given this degree of sequence conservation, we predict that Nematostella expresses at least one additional Otp splice variant that includes this OAR domain. Whereas other reported orthopedia proteins lack an octapeptide domain upstream of the homeodomain, the Nematostella gene encodes a stretch of eight residues upstream of the homeodomain that shares five out of eight residues with the octapeptide of deuterostome Rx proteins (HSIxxILx; Figure 1).

The Nematostella rx transcript we assembled using RACE is 1792 nucleotides long and it encodes a predicted protein 266 amino acids long (see Additional file 2). It extends over a 4972-nucleotide region of the genome, comprising three exons (see Additional file 2). There are three noteworthy regions of similarity between the rx proteins of Nematostella and bilaterians: the homeodomain, the octapeptide and the OAR domain (Figure 1). In the homeodomain, Nematostella is identical to both human and fruit fly at 90% of residues (54/60). Among PRD-class homeodomains, the presence of a valine at residue 43 appears to be a synapomorphy of the rx family, and this trait is shared by all the rx sequences in our dataset. The rx family is also characterized by arginine-alanine at positions 18-19 of the homeodomain. This is seen in all taxa except the coral Acropora millepora, which has a glutamine at position 18 29 32 39 40 . The sea anemone is also identical to both human and fruit fly at six of seven residues in the octapeptide. Within the OAR domain, Nematostella and human are identical at 13 of 16 residues, and Nematostella and Drosophila are identical at 14 of 16 residues (Figure 1).

We identified four additional EST sequences for Nematostella rx, two that were previously deposited at the National Center for Biotechnology Information (NCBI) (CAGN10625 and DV088198) and two that were generated by the Joint Genome Institute (JGI) as part of the Nematostella genome sequencing initiative (2664141-1 and 2664141-2). One of the EST sequences (DV088198) does not encode the complete OAR motif, instead producing a predicted protein 24 residues shorter than the other transcripts (see Additional file 2). Comparison of the genome assembly, our RACE product and these four ESTs revealed 28 single-nucleotide polymorphisms, including six that result in a change to the protein sequence (Additional file 2). Two of the polymorphic amino acid positions reside within helix 3 of the homeodomain, the so-called 'DNA-binding helix' (Figure 1; see Additional file 2): an A/G polymorphism that results in an amino acid change from arginine (R) to lysine (K) at position 52, and a C/G polymorphism results in an amino acid substitution from glutamine (Q) to glutamic acid (E) at position 59, an amide to acidic amino acid. Both the possession of lysine at position 52 and glutamic acid at position 59 are unique within the Rx family (Figure 1). In fact, the possession of a lysine at position 52 is very unusual for PRD-class homeodomains in general; of the 111 PRD-class homeodomains included in our phylogenetic analysis, the only other sequence with a lysine at this position is the Nematostella DMBXd homeodomain. Over the entire PRD class, the possession of glutamic acid at position 59 is not as uncommon. The Pax4/6 family exhibits glutamic acid at this position, as do three other Nematostella PRD-class homeodomains (DMBXb, NVHD_101 and NVHD_148).

To verify the existence of the two nonsynonymous polymorphisms identified in EST sequences (see above) in natural populations and to begin characterizing their geographic distribution, we sequenced a fragment of the rx gene from 95 individual animals collected throughout the range of the species (see Additional file 3). At position 52, the phylogenetically unusual lysine variant is relatively rare, accounting for only 15.16% of all alleles, but it does exhibit a broad geographic distribution, being found in seven different estuaries from both the Atlantic and Pacific coasts of North America. However, despite the wide distribution of the K allele, KK homozygotes were only recovered in a single location, Kingsport in Nova Scotia. The apparently limited geographic distribution of this genotype is interesting considering that KR heterozygotes accounted for more than a quarter of all individual animals assayed (RR = 71.58%, RK = 26.16%, KK = 2.11%). At position 59, the glutamine variant accounted for 96.28% of all alleles. Despite its overall rarity (3.72%), the glutamic acid variant was recovered in four widely separated estuaries in New Jersey, Maryland, California and Washington, although no EE homozygotes were recovered in any individual (QQ = 92.55%, QE = 7.45%, EE = 0.00%).

The Nematostella homeobrain transcript that we assembled using RACE encompasses three exons (see Additional file 4). Similar to otp and rx, an intron interrupts the homeodomain between positions 46 and 47. The third exon, 647 nucleotides in length, encodes the final 14 amino acids of the homeodomain plus an OAR motif, which is located near the carboxy terminus of the predicted protein. The Nematostella homeobrain protein has high similarity to insect orthologs within the homeodomain region but very little similarity could be detected elsewhere. For example, the homeodomain is identical to its Drosophila ortholog at 95% of residues (57/60). Interestingly, the sequence identity between Nematostella and Drosophila exceeds the sequence identity between Nematostella and its fellow cnidarian, Hydra magnipapillata (83%; 53 of 60 identical residues). This suggests that the Hydra homeodomain has evolved relatively quickly.

The maximum-likelihood (ML) and neighbor-joining (NJ) analyses performed here reinforce the conclusions of Ryan et al. 32 : Nematostella hbn, otp and rx sequences group with putative orthologs from bilaterian taxa with a moderate to high level of bootstrap support (Figure 2). The Nematostella hbn homeodomain is nested within a clade that also includes hbn sequences from multiple animals including protostomes (for example, Capitella and Drosophila), Trichoplax and two deuterostomes (Strongylocentrotus and Saccoglossus). Bootstrap support for this clade is moderate (ML = 77, NJ = 64). We were unable to recover a clear hbn ortholog from any chordate model system 41 . Our phylogenetic analyses recovered a monophyletic grouping for predicted rx orthologs from all species sampled except the most basal member queried, the placozoan Trichoplax. Similar to the hbn clade, bootstrap support was moderate for rx orthologs (ML = 43, NJ = 80). Finally, we identified putative otp orthologs from all species expect the nematode C. elegans (ML = 66, NJ = 98).

The Nematostella hbn, otp and rx transcripts map to a single scaffold in the publicly available assembly of the Nematostella genome (jgi scaffold 62; DOE-JGI Nematostella vectensis genome assembly V.1.0; http://genome.jgi-psf.org/Nemve1/Nemve1.home.html). The entire scaffold is 1,036,593 nucleotides long, and the cluster is located roughly 225,000 nucleotides from the nearest end of the scaffold. The rx locus intervenes between the otp and hbn loci. Otp and rx are encoded on one strand, whereas hbn is encoded on the opposite strand (Figure 3). The entire three-gene cluster spans 34,246 nucleotides (from the first nucleotide in the predicted 5' exon of orthopedia to the first nucleotide in the predicted 5' exon of homeobrain). The distance between the predicted stop codon of hbn and the predicted stop codon of rx is only 3,437 nucleotides (Figure 3). The intergenic distance between otp and the predicted transcription start site of rx is 12,158 nucleotides. Using BLASTx we searched for predicted genes positioned in the intergenic regions within the otp-rx-hbn cluster. No significant hits were recovered in the smaller intergenic region between rx and hbn (E < 10). We identified one position of moderate similarity to other proteins in the region between otp and rx (~ 1 × 10^-6). Upon closer inspection there were no consistent proteins, regions or domains identified from these results, and the hits appeared as spurious matches. Based on the current assembly of the Nematostella genome, the nearest annotated genes bracketing the cluster are a predicted ferredoxin (JGI: 101894) located 16.6 kb from otp and a predicted gene similar to RIKEN (JGI: 242384) located 5.8 kb from hbn.

The genomic organization of the Drosophila homeobrain cluster has been described previously 31 . The fruit fly cluster exhibits the same relative gene order as the Nematostella cluster, but the intergenic distances are greater and the transcriptional orientation of otp is reversed. The intergenic distance between hbn and rx is approximately 17.0 kb in the Drosophila cluster, whereas the intergenic distance between rx and otp is approximately 21.0 kb (Figure 4).

Other animal genomes in which otp, rx and hbn genes have been identified or predicted were analyzed for evidence of gene linkage. A homeobrain cluster was identified in the sequenced genomes of three additional insects: Anopheles gambiae (mosquito), Apis mellifera (honeybee) and Tribolium castaneum (flour beetle). In honeybee and flour beetle, the cluster organization is similar to that of Drosophila: the clusters have the same gene order (otp-rx-hbn), the genes exhibit the same transcriptional orientations and the intergenic distances are comparable (Figure 4). In the mosquito cluster, the gene order and scale are different, with rx linked to otp across a span of approximately 109 kb, and otp linked to hbn across a span of approximately 386 kb. In addition, there are a number of genes located in the intergenic spaces between these Paired class genes. For example, a predicted ATP-synthase-like gene is located between rx and otp, and a calcium/calmodulin protein kinase is located between otp and hbn. We also identified the three-member homeobrain cluster in the limpet Lottia. Interestingly, the gene order (otp-hbn-rx) differs from the insects and Nematostella, suggesting that an inversion occurred in the lophotrochozoan lineage. For the annelid Capitella we were only able to confirm linkage of otp-hbn, which has a comparable intergenic distance and the same transcriptional orientation as Lottia, but rx was located on a different scaffold in the current assembly, precluding a confident assessment of linkage. Rx is located near the end of the scaffold; however, there is one annotated gene (globin precursor, JGI 21023) between rx and the end of scaffold. Furthermore, if the orientation of the Capitella cluster is the same as that for Lottia (the most closely related species in our analysis), there would be > 400 kb of sequence separating these genes. We identified a number of genes within 40 kb of hbn, none of which were rx or any homeobox gene.

The sea urchin Strongylocentrotus purpuratus and the hemichordate Saccoglossus kowalevskii are the only deuterostomes in which we could identify a homeobrain gene. Because a public assembly of the Saccoglossus genome is not yet available, we were unable to assess linkage and can only report that orthologs for all three members of the cluster are present. In the sea urchin, hbn and rx are closely linked on the same genomic scaffold (Scaffold_v2_14510, length 454945 nt). The two genes are separated by a distance of approximately 22 kb. This intergenic distance is similar to that observed in Drosophila, Apis, Tribolium and Lottia. However, unlike these protostomes, in the urchin, both genes are encoded on the same strand of DNA. Otp maps to a different scaffold (Scaffold_v2_10421, length 387609 nt). Because otp resides on a different scaffold, evidence of linkage to hbn and rx could not be established. However, by taking into account the distances flanking otp on its scaffold and hbn and rx on their scaffold, otp can be no closer than 437,964 nucleotides from hbn and 193,112 nucleotides from rx.

We could not identify a homeobrain locus in the genomes of the urochordate Ciona, the cephalochordate Branchiostoma or any of the vertebrate taxa we queried (Homo, Mus, Rattus, Gallus, Xenopus and Danio). In all of the vertebrates and Ciona, otp and rx are located on different chromosomes, so no remnant of the hbn-rx-otp cluster remains. In the current Branchiostoma assembly, the otp and rx homeodomains are separated by ~162 kb on the same genomic scaffold. There are seven annotated genes within the intergenic region (JGI: 102897, 136227, 286088, 102900, 130204, 102902 and 270692) with diverse functions (for example, SCAMP (secretory carrier membrane protein), N-acetyltransferase, tubulin chaperone).

We identified orthologs of otp and hbn, but not rx, in the placozoan Trichoplax. Both of these genes are located on the same scaffold, are separated by approximately 15 kb and have the same transcriptional orientation. Because Trichoplax represents the earliest diverging taxa in our study, the close proximity of otp and hbn strongly suggests that these genes have been clustered since very early in animal evolution.

Whole-mount in situ hybridization using an antisense NvOtp riboprobe of 1,556 bp revealed that expression begins in mid planula stages (Figure 5e), well after gastrulation is complete. Expression is confined to the oral pole and the oral end of the pharyngeal ectoderm. Expression appears strongest at the oral opening but extends into the pharyngeal ectoderm. Expression is also apparent in scattered cells in the body wall ectoderm, which appear to be neurons based on morphological criteria, namely their position deep within the ectoderm and their basally located nuclei (Figure 5f, Figure 5g) 42 . Later, Otp expression is also seen in the tentacle ectoderm (Figure 5h, Figure 5i). Oral expression persists well into polyp stages (Figure 5j, Figure 5k). The probe used for in situ hybridization overlaps a substantial portion (< 300 bp) of all three predicted splice variants, so this probe would presumably bind to all three.

Expression of Rx begins before otp expression, during mid gastrulation (Figure 6c) Rx is initially expressed in the aboral hemisphere (Figure 6d), but by the end of gastrulation, expression becomes refined to a band that encircles the aboral side of the midbody but is excluded from the aboral pole, including the position of the presumptive apical tuft. Within this band, expression is spotty; only a fraction of the cells within the boundaries of the expression domain actually express Rx. As with otp and hbn, these rx-expressing cells in the ectoderm have the morphological appearance of neurons (Figure 6f, Figure 6i). Rx expression persists in this aboral band throughout planula and polyp stages (Figure 6g-j). During adult polyp stages, a second domain of expression is seen in individual ectodermal cells in the middle of each tentacle (Figure 6).

Homeobrain is first detected in the late blastula, when it is expressed throughout most of the blastoderm except in the region surrounding the blastopore (the presumptive gastrodermis; Figure 7c,d). With the onset of gastrulation, expression becomes excluded from the aboral pole, where the apical tuft of sensory cilia will form (Figure 7d,e). As gastrulation proceeds, expression persists in populations of ectodermal cells around the oral pole and later in those that encircle the base of each developing tentacle (Figure 7g,h). Expression is also seen in individual cells that are scattered throughout the mid body (Figure 7g). These cells have the morphological appearance of neurons. During polyp stages, expression is confined primarily to the ectoderm at the base of the tentacles (Figure 7i-k).

Homeobrain was originally identified in Drosophila 43 , and mapped to a region of chromosome 2 that contained two additional PRD-class homeobox genes, Orthopedia and rx 44 45 . Homeobrain is expressed in the fly embryonic brain and ventral nerve cord 31 . Recently the first lophotrochozan homeobrain-like gene was reported 41 , and found to be expressed in a restricted region of the anterior brain. Hbn expression has also been reported in the sea urchin, where it is expressed in oral ganglia of the animal pole 46 .

Otp is associated with neural development in a phylogenetically diverse collection of animals, and this may represent an ancestral role for this homeodomain family. Orthopedia-related genes in mouse are involved in brain patterning and development 47 48 49 50 51 52 . In the hemichordate Saccoglossus, otp has punctate, ectodermal expression in neural domains of the prosome (proboscis) and mesocome (collar) 53 . In planaria, orthopedia is implicated in patterning the branch structure of the brain 54 55 . In the limpet, otp is involved in the development of the larval apical sensory organ 56 , and in flies it is involved in the developing CNS and hindgut and anal pads 51 . However, a connection to neural development is not obvious in the sea urchin, where otp appears to be involved in larval skeletal morphogenesis and the establishment of oral ectodermal cell fate 57 58 59 60 61 .

Rx (retinal associated homeobox) genes have been extensively studied in many deuterostome taxa including human, mouse, Xenopus, chicken, zebrafish, medaka and tunicate 45 62 63 64 65 66 67 68 69 70 . In all of these organisms, rx is involved in brain development and the formation of retinal territories and associated neural structures. In the hemichordate, rx is expressed throughout the prosome ectoderm, the most anterior region of this organism, but absent from the most apical pole 53 . Rx is expressed more broadly during sea urchin development, with punctate expression in the animal pole and developing gut 46 . A fruit fly rx gene has also been identified, and studies suggest that it is necessary for proper brain development, but it is not required for eye development 45 71 .

In addition to a homeodomain and a PRD domain, two additional motifs have been identified in some PRD-like genes. The octapeptide motif 72 73 , located towards the N-terminus, is involved in transcriptional repression 74 . The OAR domain, found in otp, aristaless and rx 40 51 67 75 is typically located at the C-terminus and is known to function as a transactivator in Otp 40 51 .

We have identified alternative transcripts for Nematostella orthopedia and rx. For rx, some transcripts encode a highly conserved OAR domain and one transcript does not. For orthopedia, none of our RACE products nor any ESTs in publicly available databases included the OAR domain. However, a well-conserved OAR domain was identified downstream and in-frame of these sequences in predicted genomic gene models, suggesting that this domain is probably expressed in an as yet unidentified splice variant. The OAR domain has been described as an intramolecular switch, which acts to reduce the affinity of the homeodomain transcription factor for its binding site. In the mouse, ectopically expressed mutant forms of the Alx3 and Cart1 proteins lacking the OAR domain exhibit increased binding to their DNA targets 76 .

Alternative splice variants involving the presence or absence of an OAR domain have also been identified the mouse prx1 gene, a member of the PMX family of PRD-class homeobox genes 77 . The carboxy terminus of the Prx1a protein includes an activation domain and an OAR domain, whereas the carboxy terminus of Prx1b encodes a repression domain and lacks an OAR domain 77 78 . The tissue distribution of both transcripts appears to be similar in mice and humans, but different tissues exhibit pronounced differences in the relative ratios of prx1a and prx1b 79 80 . It has been hypothesized that the presence of the OAR domain in prx1a could render it sensitive to modulation via an unidentified partner protein that interacts with the OAR domain itself. In the absence of this cofactor, the OAR domain masks the activation domain and reduces the affinity of prx1a for DNA binding sites. When bound by its co-factor, the activation domain becomes unmasked and the DNA binding affinity increases 76 77 81 . In the case of Nematostella rx, the situation may be somewhat simpler than in the mouse prx1 gene, because except for the presence or absence of the OAR domain itself, the alternative splice variants encode essentially identical proteins.

In the most famous example of a conserved homeobox cluster, the Hox genes, the spatial ordering of Hox expression territories along the body's main axis and the timing of their onset mirrors the physical ordering of linked Hox genes (although not in all taxa). This correspondence is termed colinearity. However, in animals with dispersed Hox clusters, such as the urochordate Oikopleura dioica, some spatial colinearity remains whereas temporal colinearity is absent 82 . This suggests that it is temporal rather than spatial colinearity that is driving the maintenance of these clusters. Although work has been performed in studying temporal colinearity in Hox, ParaHox and NK clusters, the homeobrain cluster could prove to be another supporting example.

In Nematostella, hbn, rx and otp appear to be expressed in a temporally colinear pattern. Hbn is expressed first, in the blastula stages, followed by rx at mid gastrulation and finally otp in the planula. However, in the case of the Drosophila homeobrain cluster, there is no clear evidence of temporal colinearity. Homeobrain is expressed first, in the syncytial blastoderm, then otp is expressed slightly before rx 31 51 65 . Future studies in other animals will help determine whether temporal colinearity is widely conserved among homeobrain clusters.

Although the expression of Nematostella hbn, rx and otp is consistent with temporal colinearity, it is not consistent with spatial colinearity. The three genes are expressed in non-overlapping domains along the oral-aboral axis; however, these domains do not appear to be related to their position in the cluster. NvRx is expressed at the most aboral domain, although expression is not present at the aboral pole, where the apical tuft will form. NvHbn is more broadly expressed in oral ectoderm during early embryogenesis and becomes confined to the most oral ectoderm, mainly around the base of the tentacles. Finally, NvOtp is also expressed in the oral ectoderm, in the domain that will invaginate and form the pharynx. NvOtp and NvHbn are both expressed in close proximity to three paralogs of Otx (NvOtxA, B, C) in the pharyngeal ectoderm and tentacles surrounding the oral pole 30 .

In addition to the broad non-overlapping domains, all these genes are also expressed in individual cells throughout the body column. Based on their cell morphology, it appears that these cells may be neurons. Additionally, the expression of NvOtp and NvOtx(A,B,C) in the oral ectoderm coincides with formation of the oral nerve ring 30 42 . Considering the function of these genes in other animals and the expression patterns seen here, it is likely that hbn, rx and otp play some role in neural development in Nematostella. However, it is interesting that there is no expression in the apical tuft, another strongly neurogenic region in Nematostella.

The literature on homeobox clusters would suggest that the history of the ANTP class has been qualitatively different from that of the histories of other classes of homeobox genes. The most intensively studied and widely conserved homeobox clusters are all composed of ANTP-class homeobox genes (the Antennapedia complex, the Bithorax complex, the Hox cluster, the ParaHox cluster, the NK cluster and the EGH box cluster). Ultimately, all of these ANTP-class clusters may derive from a single ancestral cluster. For example, the ANTP-C and BX-C of Drosophila are clearly derived from a single ancestral Hox cluster that is widely shared by protostomes and deuterostomes, and in simpler form, by cnidarians. The Hox cluster, in turn, appears to have broken off from an ancestral 'mega-Hox' or 'extended Hox' cluster that at one time may have encompassed the Hox cluster, the ParaHox cluster, the EGH cluster and the NK cluster. Over time, the hypothetical ancestral cluster appears to have fragmented, and different remnants of this ancestral cluster may be more highly conserved in different animal lineages 83 .

This study is the first to provide evidence that a non-ANTP-class homeobox cluster was conserved over hundreds of millions of years of animal evolution. Clearly, the hbn-rx-otp cluster has been fairly well conserved over the evolutionary history of holometabolous insects. A cluster with the same constituent genes, in the same orientation, spanning a comparable distance is found in representatives of three different orders of insects (Coleoptera, Diptera and Hymenoptera; Figure 4). The cluster also appears to date to the ancestral Protostome, as evidenced by conservation of the three-gene cluster in the limpet Lottia, although the order of the genes has changed. In addition, the cluster was also likely to have been present in the cnidarian-bilaterian common ancestor, some 600 million years ago. The cluster in Nematostella involves the same closely linked genes in the same order as the inferred ancestral cluster of holometabolous insects, but the otp locus has been inverted. Similarly, in four other key taxa (cephalochordate, sea urchin, annelid and placozoan) we find evidence of a partial cluster, in which two of the three genes are linked. Further sequencing and assemblies of the echinoderm and hemichordate genomes will reveal the extent of the hbn-rx-otp cluster in the deuterostome ancestor. The placozoan genomic data suggests that a portion of this cluster (hbn and otp) dates back even further to the ancestral eumetazoan. However, it remains to be seen whether rx was present in this ancestor and lost in the placozoan lineage, or whether rx evolved after the split.

Mechanistically, it is easier to envision how a cluster of three closely related genes might remain linked than to envision how three closely related genes, if already dispersed, could independently become so closely juxtaposed in multiple taxa. If the genes arose by tandem duplication, the cluster would have originated as a result of the gene duplication; that is, the starting point would have been a cluster. Subsequently, the cluster could have been maintained over hundreds of millions of years of evolution in multiple animal lineages by stabilizing selection.

Closely linked genes will tend to reside in the same chromosomal territories, and they may come under the influence of shared regulatory elements. For this reason, the proper regulation of linked genes may be related to their physical proximity in the genome. This is the general explanation for why Hox genes have remained clustered for hundreds of millions of years in many animals that have been examined. In the case of Hox genes, the spatial ordering of Hox expression territories along the body's main axis mirrors the physical ordering of linked Hox genes along the chromosome. This correspondence is termed colinearity. In the anterior CNS, rx, hbn and otp are expressed in nested territories, which is somewhat reminiscent of Hox genes. Future bioinformatics studies could test for conserved regulatory elements within the hbn-rx-otp clusters of fly, honeybee and flour beetle. The functionality of these putative enhancer-binding sites could then be studied experimentally. In addition, the effect of cluster disruption can be examined experimentally in Drosophila and in mice, as it has been with Hox genes 84 85 . It may also prove very informative to use evolutionary comparisons of taxa with intact and disrupted clusters to investigate the consequences of cluster disruption, as has recently been carried out for the ParaHox cluster 20 .

The homeobrain gene appears to have been lost in chordates. We could not identify it in the sequenced genomes of human, mouse, chicken, clawed frog, zebrafish, lancelet or tunicate (Figure 4). However, we identified clear orthologs in two other deuterostomes (sea urchin and hemichordate) and throughout the protostomes and 'basal' metazoans. Thus, this gene may have been truly lost from the genome or it may have become so highly modified that it is no longer recognizable as a homeobrain ortholog. The phylogenetic analysis performed here, based on homeodomain sequences, does not strongly suggest another PRD-class gene as a possible ortholog. Unfortunately, we cannot rely on regions outside the 60-amino acid homeodomain to obtain additional phylogenetic signal because there was not sufficient sequence conservation to permit alignments across all genes represented in the phylogeny. The highly conserved 128-amino acid Paired domain, present in many PRD-class homeobox genes, is absent from the homeobrain, orthopedia and rx families, among others 29 39 40 . With the present data, our analyses support the hypothesis that homeobrain was lost early in the chordate lineage.

Despite potential stabilizing selection to maintain the cluster, we observed many polymorphisms in the Nematostella genes comprising the cluster, including nonsynonymous substitutions in the homeodomain of rx that have a patchy distribution in natural populations. The functional role of these polymorphisms awaits future experimental characterization. The presence of few homozygotes for the rare amino acid for each position in our sampling of natural populations is potential evidence for a functional difference between alleles.