Background
Homeobox genes constitute an ancient superclass of regulatory genes with diverse developmental functions 1. The homeobox, which encodes a helix-turn-helix DNA-binding motif known as the homeodomain, originated prior to the evolutionary split between plants, fungi, and metazoans 2. The homeodomain is commonly 60 amino acids in length, though recognizable homeodomains may be as long as 97 or as short as 54 amino acids (reviewed in 3).
Based on phylogenetic analyses and chromosomal mapping studies, animal homeodomains can be divided among ten distinct classes: ANTP, CUT, HNF, LIM, POU, PRD, PROS, SINE, TALE, and ZF 345678910111213141516. The ANTP and PRD classes are substantially larger than the other classes, and these two classes are thought to be sister clades 57. Within the ANTP class, there is evidence for a monophyletic subclass comprising Hox-related genes 47. The PRD class can be divided into subclasses based on the amino acid present at position 50 of the homeodomain (Q50, K50, or S50), but these subclasses do no not appear to represent monophyletic groups 57. The remaining eight homeodomain classes are significantly smaller than the ANTP and PRD classes, and they are thought to have emerged as a series of lineages basal to an ANTP-PRD clade 6. To this point, the HNF class has only been reported from vertebrates 6. Structural and functional properties of the homeodomain appear largely conserved within these homeodomain classes 4. The homeodomain sequences encoded by orthologous homeobox genes are often so highly conserved that orthology between protostomes and deuterostomes, and even between bilaterians and non-bilaterians, is readily apparent 17.
The ANTP, PRD, CUT, LIM, POU, PROS, SINE, TALE, and ZF classes are known from both protostome and deuterostome metazoans 3. Therefore, we can trace their origins to the protostome-deuterostome ancestor, which a recent estimate places at some 579 to 700 million years ago (Figure 1) 18. Identification of these homeobox classes in outgroup taxa would indicate even greater antiquity. For example, molecular clock estimates based on maximum likelihood and minimum evolution suggest that the cnidarian-bilaterian divergence predated the protostome-deuterostome divergence by 25 to 48 million years 18.
<p>Figure 1</p>
Phylogenetic relationships among major metazoan lineages
Phylogenetic relationships among major metazoan lineages. The topology of the tree is consistent with several recent molecular phylogenetic analyses [100-106]. Estimated divergence times for Cnidaria versus Bilateria, protostomes versus deuterostomes, and lophotrochozoans versus ecdysozoans are indicated in the white boxes [18]. The origin of the homeobox gene superclass must have predated the split between animals, plants, and fungi.
![]()
Establishing the antiquity of homeobox genes is critical to understanding the role of these genes in metazoan evolution. The functional diversification of homeobox genes, by gene duplication and divergence, or by cis-regulatory evolution, has been touted as an important mechanism in the evolution of diverse body plans and organs in bilaterian metazoans 619202122232425. The Cnidaria is the likely sister group of the Bilateria 2627, and since their divergence from a common ancestor, these two lineages have undergone very different evolutionary trajectories (Figure 1). The bilaterian ancestor has spawned over 30 distinct phyla comprising more than one million extant species; the cnidarian ancestor has spawned some 10,000 extant species, all comfortably housed in a single phylum 28. The maximum complexity and morphological diversity of cnidarian body plans (for example, sea anemones, sea pens, corals, hydras, and jellyfishes) is modest when compared to the maximum complexity and morphological diversity of bilaterian body plans (for example, vertebrates, sea squirts, sea urchins, insects, nematodes, octopi, and phoronids 2529). Taking into account the presumed importance of homeobox genes in the morphological diversification of bilaterians, the close evolutionary relationship between the Bilateria and the Cnidaria, and the contrasting evolutionary trajectories of these two lineages, a comparison of cnidarians and bilaterians becomes critical for understanding the significance of homeobox genes in the morphological diversification of animal body plans.
Here, we seek to identify homeobox genes that were present in the cnidarian-bilaterian ancestor using phylogenetic analysis of homeodomains from bilaterians and cnidarians. Our analysis takes advantage of the curated genomic datasets of the fruit fly Drosophila melanogaster 3031323334 and Homo sapiens 3536 as well as the recently completed rough draft of the sea anemone Nematostella vectensis, a representative cnidarian (Joint Genome Institute; D Rokhsar, principal investigator).
The phylogenetic analyses presented here reveal the extent to which the homeobox gene superclass had radiated prior to the evolutionary split between Cnidaria and Bilateria. For example, at one extreme, the Cnidaria could have diverged from the Bilateria prior to the origin of the aforementioned homeobox classes (ANTP, PRD, LIM, POU, and so on). If so, then the cnidarian homeobox genes and the bilaterian homeobox genes would constitute independent radiations on the phylogeny (Figure 2a). This possibility is ruled out by published studies that have identified distinct ANTP, POU, PRD, and SINE homeodomains in the Cnidaria 517373839404142434445. Alternatively, the Cnidaria could have diverged from the Bilateria after the origin of the class founder genes (for example, the ancestral ANTP class gene, the ancestral PRD class gene, and so on), but prior to the subsequent radiations of these classes. In this case, the cnidarian and bilaterian class radiations would constitute mutually exclusive monophyletic groups (Figure 2b). However, if the homeobox classes had undergone extensive radiations prior to the cnidarian-bilaterian divergence, then the same homeobox families would be represented in cnidarian and bilaterian genomes (Figure 2c). Finally, it might also be the case that some homeobox classes had radiated prior to the cnidarian-bilaterian radiation, while other classes had not (Figure 2d).
<p>Figure 2</p>
Hypothetical scenarios for the evolution and diversification of homeodomain classes relative to the cnidarian-bilaterian divergence
Hypothetical scenarios for the evolution and diversification of homeodomain classes relative to the cnidarian-bilaterian divergence. The timing of the cnidarian-bilaterian divergence is indicated by an arrow and a dashed vertical line. Cnidarian homeobox genes are indicated by red lines. Protostome (for example, Drosophila) homeobox genes are indicated by green lines. Deuterostome (for example, human) homeobox genes are indicated by blue lines. (a) Cnidaria diverges from Bilateria prior to origin of the major homeodomain classes (ANTP, PRD, LIM, POU, SINE, TALE). (b) Cnidaria diverges from Bilateria after the origin of homeodomain classes but before their diversification. (c) Cnidaria diverges from Bilateria after the diversification of homeobox classes. (d) At the time of the cnidarian-bilaterian divergence, some homeobox classes have not yet originated (ANTP, PRD) whereas others have diversified extensively (POU, SINE).
![]()
The phylogenetic analyses presented here reveal that the ANTP, PRD, LIM, SINE, and POU classes had radiated extensively prior to the divergence of the Cnidaria and the Bilateria. The HNF class, formerly known only from vertebrates, is also represented in the Nematostella genome. In addition, we identify a putative CUT class gene in Nematostella by searching the predicted gene database at StellaBase 4647. Our analyses fail to identify ZF or PROS homeodomains in Nematostella. The phylogenetic analyses reveal 56 distinct homeodomain families that appear to be shared by Nematostella and one or both of the bilaterian taxa.
Results
Metazoan homeodomains
We retrieved 455 distinct homeodomains from the three metazoan taxa under study, including 130 from the genome of Nematostella, a representative non-bilaterian, 228 from Homo, a representative deuterostome bilaterian, and 97 from Drosophila, a representative protostome bilaterian. An alignment of all homeodomains (with accession numbers) is presented in Additional data file 1. The number of homeodomains we identified in the human and fruit fly genomes is comparable to a recent analysis of bilaterian homeodomains that identified 102 in Drosophila and 257 in humans 48. The present analysis includes fewer homeodomains from human and fruit fly because we eliminated hypothetical or computationally predicted homeodomains that introduced new gaps or extended existing gaps in the alignment. Like the aforementioned analysis, we treated individual homeodomains from multi-homeodomain genes as separate taxa in our phylogenetic analysis - lower case letters appended to the gene name distinguish different homeodomains that derive from a single protein.
Because the human and Drosophila genomes are still in the process of being annotated, and because our criteria for homeodomain inclusion were stringent, this dataset cannot be considered exhaustive. However, most sequences excluded from this study represent rapidly evolving and highly divergent sequences that would not have a significant bearing on the conclusions. The Nematostella dataset consists of first-pass predictions from a draft-quality genomic sequence. It is possible that a number of Nematostella homeodomains may have been missed, and it is also possible that homeodomains from one or more pseudogenes have been included. Nevertheless, these data are more than sufficient for the purpose of the analyses performed here: to obtain a qualitatively accurate assessment of the homeobox-gene complement present in the cnidarian-bilaterian ancestor.
Overall tree topologies and classification of animal homeodomains
The homeodomain phylogeny produced by Bayesian analysis agrees substantially with the phylogeny produced by neighbor-joining (fully labeled neighbor-joining and Bayesian phylogenies are contained in Additional data files 2 and 3, respectively; Figure 3 depicts the neighbor-joining topology without individual gene names). Both trees recover nearly all of the accepted bilaterian homeodomain families with high statistical support. Throughout this paper, we emphasize phylogenetic inferences that are supported by both methods, especially those homeodomain families that receive robust statistical support from both methods, as judged by bootstrap proportions in the neighbor-joining analysis (BP) and log-likelihood values in the Bayesian analyses (LnL).
<p>Figure 3</p>
Phylogenetic relationships among homedomains from Nematostella (red lines), human (blue lines), and fruitfly (green lines) determined by neighbor-joining [95]
Phylogenetic relationships among homedomains from Nematostella (red lines), human (blue lines), and fruitfly (green lines) determined by neighbor-joining [95]. Gene names are not provided in this condensed version of the tree, which is intended to convey an overview of the homeodomain radiation in metazoans. A fully labeled version of this tree is provided in Additional data file 2. All homeodomain classes that are known to be shared among cnidarians and bilaterians are indicated by colored bars (ANTP, HNF, LIM, POU, PRD, SINE, and TALE). Histograms to the right of the tree indicate the number of sequences from each species that fall within a given class (Hs, Homo sapiens; Dm, Drosophila melanogaster; Nv, Nematostella vectensis). The gray bars on the histograms provide a conservative estimate for the size of each homeodomain class in the cnidarian-bilaterian ancestor (CBA). The homeodomain tallies shown here are based solely on the phylogenetic analyses performed in this study. Additional data sources, cited in the text, would lead us to adjust the tallies for Nematostella and the CBA slightly upward.
![]()
The neighbor-joining analysis supports the monophyly of the ANTP class overall, and the monophyly of a Hox-related subclass within the ANTP class. The Bayesian analysis also supports the monophyly of the Hox-related subclass. However, on the Bayesian tree, there is an unresolved polytomy at the base of the ANTP class that includes a number of non-ANTP class homeodomains. This polytomy could be resolved in a manner that is compatible or incompatible with the monophyly of the ANTP class. The HNF, POU, PRD, and SINE classes appear monophyletic on both neighbor-joining and Bayesian trees. The CUT, LIM, and ZF classes do not appear monophyletic on either the neighbor-joining or Bayesian trees (Additional data files 2 and 3).
The Bayesian and neighbor-joining trees agree on the class-level relationships of 126 out of 130 of the Nematostella homeodomains (96.2%). According to both trees, 72 Nematostella homeodomains belong to the ANTP class, one to the HNF class, four to the LIM class, five to the POU class, 33 to the PRD class, five to the SINE class, and six to the TALE class (Table 1). This represents the first report of cnidarian HNF, LIM and TALE homeodomains. Four of the Nematostella homeodomains group with different classes on the Bayesian and neighbor-joining trees. None of Nematostella sequences groups with bilaterian homeodomains of the CUT class, the PROS class, or the ZF class. However, in a subsequent search of predicted Nematostella genes, we were able to identify a single protein that exhibits significant similarity to bilaterian CUT genes. The extensive intermingling of homeodomains from Nematostella, human, and fly on the phylogeny (Figure 3) reveals that the ANTP, CUT, LIM, POU, PRD, SINE, and TALE classes had undergone substantial radiations prior to the split between Cnidaria and Bilateria.
<p>Table 1</p>
Number of homeodomain proteins by class, family, and species
Hs
Dm
Nv
CBA
ANTP class/Hox-related
CDX*
3
1
1
1
EVX*
2
1
1
1
EXEX*
1
1
1
1
GBX*
2
1
1
1
GSX*
2
1
1
1
HOX1*
3
1
2
1
HOX2*
2
1
3
1
HOX3*
3
3
0
0
HOX4*
4
1
0
0
HOX5*
3
1
0
0
HOX6-8*
8
3
0
0
HOX9-13*
16
1
0
0
IPF*
1
0
†
0
0
MOX*
2
1
4
1
ROUGH*
0
1
1
1
Unknown family
0
1
3
n/a
Total
52
19
18
9
ANTP class/other
BARH*
2
2
0
0
BARX
2
0
0
0
BSH*
1
1
0
0
CG13424*
0
1
2
1
DLX*
6
1
1
1
EMX*
2
2
2
1
EMXLX*
0
1
2
1
EN*
2
1
0
0
HHEX*
1
1
1
1
HLX*
2
2
7
1
HMX*
3
1
1
1
LBX*
2
2
1
1
MSX*
2
1
1
1
MSXLX*
0
1
2
1
NK1*
1
1
1
1
NK2*
7
2
5
1
NK3*
2
1
1
1
NK6*
2
1
1
1
NK7*
0
1
1
1
TLX*
3
1
1
1
VAX*
1
0
2
1
Unknown family
3
0
22
n/a
Total
44
24
54
17
CUT class
COMPASS
0
2
0
0
CUTL*
2
1
0
0
ONECUT*
3
1
0
0
SATB
2
0
0
0
Total
7
4
0
0
HNF
HNF1/2*
2
0
1
1
LIM class
AP*
2
1
0
0
ISLET*
2
1
1
1
LHX1/5*
2
1
1
1
LXH3/4*
2
1
0
0
LHX6/8*
2
1
1
1
LMX*
1
2
0
0
Unknown family
0
0
1
n/a
Total
11
7
4
3
POU class
POU1*
1
0
1
1
POU2*
3
2
0
0
POU3*
4
1
2
1
POU4*
3
1
1
1
POU5
2
0
0
0
POU6*
2
1
1
1
Total
15
5
5
4
PRD
AL*
1
2
1
1
ALX*
3
0
1
1
ANF
1
0
0
0
ARIX*
2
1
0
0
CEH10*
2
2
1
1
DMBX*
1
0
6
1
DUX*
20
0
3
1
GSC*
2
1
1
1
HB*
0
1
1
1
MIX
1
0
0
0
OTP*
1
1
1
1
OTX*
3
1
3
1
PAX3/7*
2
3
2
1
PAX4/6*
3
4
2
1
PRX*
2
1
0
0
PTX*
3
1
1
1
REPO*
0
1
1
1
RX*
2
1
1
1
SHOX
2
0
0
0
UNC4*
0
2
1
1
Unknown family
2
2
7
n/a
Total
53
24
33
15
PROS class
PROS
1
0
0
0
Total
1
0
0
0
SINE class
SIX1/2*
2
1
1
1
SIX3/6*
2
1
1
1
SIX4/5*
2
1
2
1
Unknown family
0
0
1
n/a
Total
6
3
5
3
TALE
IRX*
7
3
1
1
MEIS*
3
1
1
1
PBX*
4
1
1
1
TGIF*
1
2
1
1
Unknown family
1
0
2
n/a
TOTAL
16
7
6
4
ZF class
ZFHX2
2
0
0
0
ZFH4
2
0
0
0
ZHX
5
0
0
0
zfh1
0
1
0
0
zfh2*
1
1
0
0
Unknown family
1
0
0
n/a
Total
11
2
0
0
Unknown class
Total
10
2
4
n/a
*Counted as a shared family in Table 2. †Absence of IPF in Drosophila is due to secondary loss. CBA, cnidarian-bilaterian ancestor; Hs, Homo sapiens; Dm, Drosophila melanogaster; Nv, Nematostella vectensis.
ANTP class
Hox-related subclass
Genes from the Hox-related subclass have played a prominent role in the evolution and diversification of the primary body axis in animals 22394950. The phylogenetic analyses indicate 52 Hox-related homeodomains in human, 19 in fruit fly, and 18 in Nematostella. All 89 of these genes constitute a monophyletic group on both Bayesian and neighbor-joining trees (Additional data files 2 and 3). Within this large clade of Hox related genes, we can identify 15 distinct monophyletic families (Additional data file 1; Table 1). On both the Bayesian and neighbor-joining trees, eight of these families appear to have Nematostella representatives: CDX, EVX, EXEX, GBX, GSX, HOX1, MOX, and ROUGH. Previous studies have reported CDX, EVX, GBX, GSX, HOX1, and MOX genes in cnidarians 173738394051, but EXEX and ROUGH homeodomains have not previously been identified in this phylum. According to the neighbor-joining tree, the HOX2 family may also be represented in Nematostella, which would be consistent with previously published homeodomain phylogenies that have identified putative anterior Hox genes (HOX1 and HOX2 families) in the Cnidaria 17383951. No Nematostella sequences group with the HOX3, HOX4, HOX5, HOX6-8, or HOX9-13 families. The apparent absence of 'central' Hox genes (HOX4-HOX8) in cnidarians, has been a consistent finding of recent phylogenetic analyses, but these same studies have supported the existence of 'posterior' Hox genes in cnidarians (HOX9-HOX13) 17383951. For example, in published neighbor-joining and maximum likelihood analyses, the Nematostella homeodomains anthox1 and anthox1a have grouped with posterior Hox genes in bilaterians 172238. In the present analysis, these same homeodomain sequences (known as NVHD099 and NVHD106) either fall basal to a clade containing both posterior and central genes (Bayes), or they fall basal to a clade comprising all the central Hox genes (neighbor-joining).
While previous studies have reported multiple Hox-related ANTP genes from individual cnidarian species, including EVX, MOX, GSX, and Hox genes 173738394051, the present study is unique in terms of its scope and the thoroughness with which the Hox-related homeodomains have been sampled from a single cnidarian genome. No previous study has reported as many as 18 Hox-related genes from a member of this phylum. The inclusion of numerous additional sequences has resulted in the identification of previously unreported families (EXEX and ROUGH), and it has caused us to question the previously hypothesized relationships of NVHD099 and NVHD106. The current analysis does not support the designation of these genes as posterior Hox genes. The Bayes tree suggests an interesting alternative hypothesis - that these two Nematostella homeodomains could be direct descendants of the common ancestor of central and posterior Hox genes. This could explain the apparent absence of central Hox genes without the need to invoke gene loss 1252. More detailed phylogenetic and gene linkage studies of Nematostella and other basal metazoan lineages may help to elucidate the early evolution of Hox-related genes.
Other ANTP class families
We identified 122 ANTP class homeodomains that fall outside the Hox-related clade: 44 from human, 24 from fruit fly, and 54 from sea anemone. Of these 122 homeodomains, 98 can be classified into one of 21 different gene families (Additional data file 1; Table 1). According to both trees, Nematostella appears to possess representatives from 17 of these 21 families (Additional data files 2 to 3). Single Nematostella homeodomains group with each of the following families: DLX, HHEX, HMX, LBX, MSX, NK-1 (slouch), NK-3, NK-6, NK-7, and TLX. The statistical support for these groupings is very robust, with neighbor-joining bootstrap proportions and Bayesian log-likelihood values in excess of 0.88 in all cases. Multiple Nematostella homeodomains group with each of the following families: EMX (two sequences), EMXLX (two sequences), HLX (seven sequences), MSLX (two sequences), NK-2 (five sequences), and VAX (two sequences). Two Nematostella homeodomains also group with the predicted Drosophila homeodomain CG13424 in what appears to be a very ancient, but not formally recognized family of ANTP-class homeodomains. While CG13424 appears missing in the human genome, two CG13424-related proteins have been described in another deuterostome, the appendicularian urochordate Oikopleura dioica 53. None of the Nematostella homeodomains groups with the following four families on either of the trees: BARH, BARX, BSH, and EN. Twenty-two of the Nematostella sequences could not be assigned to a specific family. The results presented here, bolstered by previous studies that have reported BARX, DLX, EMX, HHEX, MSX, NK-2, and TLX genes from other cnidarians 3944545556, make it clear that the ANTP class had radiated extensively prior to the cnidarian-bilaterian split.
CUT class
The genes of the Cut class 3, also known as the Cut superclass 657, typically encode two different types of DNA-binding domains: homeodomains as well as cut domains 585960. Cut domains are roughly 80 amino acids long, and they are typically located upstream of the homeodomain 6. Cut proteins may possess only a single cut domain (as in Onecut), two cut domains (as in the SATB genes), or three cut domains, (as in the Drosophila gene Cut 58). Genes of the Compass family lack a Cut domain altogether, but they are placed within this class on the basis of their shared possession with the SATB genes of a conserved COMPASS domain at the amino terminus 6. The Cut class is believed to be monophyletic on the basis of the shared possession of the cut domain (in all but the Compass family) and on the basis of phylogenetic analyses of homeodomain and cut domain sequences 59.
On both the neighbor-joining and Bayesian phylogenies produced here, each of the four previously recognized subgroups of Cut genes appears monophyletic (COMPASS, CUTL, ONECUT, and SATB 6). However, the class as a whole does not appear monophyletic on either tree. On the Bayesian tree, the ONECUT family appears closely related to the CUTL family, but the COMPASS and SATB families emerge as independent lineages. On the neighbor-joining tree, all four Cut families emerge as distantly related independent lineages. Clearly, when a broad representation of homeodomain proteins is considered, phylogenetic analysis of the homeodomain does not support the monophyly of the Cut class. On the Bayesian tree, none of the Nematostella homeodomains groups with Cut class homeodomains. On the neighbor-joining tree, two Nematostella homeodomains do group with the SATB genes in a weakly supported clade (BP = 0.14). The phylogenetic analyses clearly imply that the CUT class had not diversified prior to the cnidarian-bilaterian split.
However, an independent analysis suggests that the primordial CUT gene did originate prior to the split between Cnidaria and Bilateria, and that this gene most resembled the ONECUT family, as previously predicted 6. We have identified a single putative CUT gene in the Nematostella genome by searching the database of predicted genes at StellaBase 4647 for CUT domains (query conditions: Protein Family Name: CUT; E-value threshold: 1e-6). The single gene returned by this search (StellaBase ID: 14839) encodes both a Cut domain and a homeodomain. The top 50 hits in a BLASTp search of the non-redundant protein database using this protein as the query are all CUT class proteins, specifically members of the ONECUT family.
HNF class
The HNF class is a small class of homeodomain proteins that was erected to accommodate HNF1, a liver-specific transcription factor (hepatic nuclear factor) with a highly atypical homeodomain 61. The homeodomains of the HNF class are unusual in that they possess a large number of extra residues between helix 2 and helix 3 6. So far, this homeodomain class has not been reported outside of vertebrates. On both the neighbor-joining and Bayesian trees, there is robust support for a clade uniting two human HNF homeodomains (HNF1a, HNF1b) with the Nematostella sequence NVHD070 (Additional data files 1 to 3). No Drosophila sequence groups with this HNF clade.
LIM class
The LIM homeobox genes are characterized by two protein-binding zinc fingers called LIM domains, which are located upstream of the homeodomain 62. LIM homeodomain proteins are widely implicated in neural patterning throughout the animal kingdom 6263. Recently, a LIM-domain containing gene was reported in Nematostella 64, but this gene does not encode a homeodomain. No LIM-class homeodomains have yet been described for the phylum Cnidaria.
The phylogenetic analysis presented here identifies 11 LIM homeodomains in human, 7 in fruit fly, and 4 in Nematostella (Table 1; Additional data files 1 to 3). The LIM class is divided into six distinct groups: APTEROUS, ISLET, LIN-11, LHX3/4, LHX6/8, and LMX 62. In our trees, all six of these groups represent discrete clades. Here, we refer to the LIN-11 class as the LHX1/5 group based on the names of the human and fruit fly genes that belong to it. If we limit the membership of the LIM class to these six groups, then the LIM class appears paraphyletic on the neighbor-joining and Bayesian trees (Additional data files 2 and 3). In both the Bayesian and neighbor-joining trees, a number of zinc-finger homeodomains disrupt the monophyly of the LIM class. On both neighbor-joining and Bayesian trees, the ISLET, LIM1/5, and LHX6/8 clades each contain a single Nematostella gene. The Nematostella homeodomain NVHD055 appears as the sister to a clade comprising the LHX1/5 and LHX3/4 families on both the neighbor-joining tree and the Bayes tree.
POU class
POU genes are characterized by an approximately 75 amino acid DNA binding domain upstream of the homeodomain. During development, their expression is known to be spatially and temporally restricted, and they have been implicated in cell-fate determination, early embryonic development and neuronal determination 65. The POU class comprises six different families 65. POU I genes have been reported from non-Bilateria such as sponges 66 and cnidarians (D Jacobs, personal communication). POU IV and VI genes have also been described in a cnidarian 67.
Nematostella has five putative POU genes, including single representatives from the POU I, IV, and VI families, and potentially two representatives from the POU III family (Additional data files 1 to 4). Class II and class V genes appear lacking in Nematostella. Drosophila, like Nematostella, is missing a class V gene, which suggests that this class may be a vertebrate invention. On the other hand, Drosophila is missing a class I gene. Its absence in the fruit fly and presence in sea anemone and human suggests a possible gene loss in the line leading to Drosophila. We can surmise that at least four POU homeodomains were present in the cnidarian-bilaterian ancestor, including single representatives of classes I, III, IV, and VI. Class II may be a bilaterian invention.
PRD class
Both the neighbor-joining and Bayes trees support the monophyly of a PRD clade comprising 53 human homeodomains, 24 fruit fly homeodomains, and 33 Nematostella homeodomains (Additional data files 1 to 3). A previous phylogenetic analysis of PRD homeodomains delineated the following distinct evolutionary lineages: Al, Anf (HESX1), Arix, Cart1 (ALX3/4), Ceh10, Gsc, Mix, Og12 (SHOX), Otp, Otx, Pax3/7, Pax4/6, Prx, Ptx, Rx, Siamois (DUX), and Unc4 5. All but two of these lineages appear monophyletic on both Bayesian and neighbor-joining trees - the Bayesian tree does not support the monophyly of the ALX3/4 and AL families. Three additional homeodomain families reside within the PRD radiation on the Bayesian and neighbor-joining trees, bringing the total number of PRD families to 20 - the DMBX, HB (Homeobrain), and REPO families are each represented in both Nematostella and the Bilateria, and they cannot be subsumed within the 17 PRD lineages that were defined previously 568.
On both the Bayesian and neighbor-joining trees, 15 of the 20 PRD families harbor Nematostella sequences, including several families not previously reported in the Cnidaria: AL, ALX, CEH-10, DMBX, DUX, GSX, HB, OTP, OTX, PAX3/7, PAX4/6, PTX, REPO, RX and UNC4 (Additional data files 1 to 3; Table 1). Nematostella appears to lack a representative from the ARIX and PRX families, which are found in fruit fly and human, and from the ANF and MIX families, which are found only in human. The fruit fly appears to lack representatives of the ALX, DMBX, and DUX families, all of which are represented in the human and sea anemone. Likewise, three of the groups found in fruit fly and sea anemone appear to lack a human representative: HB, REPO, and UNC4.
The phylogenetic analyses suggest that the cnidarian-bilaterian ancestor may have possessed representatives of 15 PRD homeodomain families. The ANF, ARIX and PRX families may have originated within the Bilateria. Three PRD families may have been lost in the line leading to Drosophila (ALX, DMBX, DUX), while three different PRD families may have been lost in the line leading to human (HB, REPO, and UNC4).
The DUX family is home to several human genes with double and triple homeodomains. Interestingly, three closely linked Nematostella homeodomains group with the human DUX homeodomains. These Nematostella homeodomains may be part of the same locus. If all three homeodomains are expressed as part of a single protein, it would be the first reported triple-homeodomain gene in a cnidarian. However, the statistical support for the branches uniting human DUX homeodomains with these potential Nematostella DUX homeodomains is low (BP = 0.21; LnL = 0.35), and the existence of a single transcript comprising all three homeodomains has not been demonstrated experimentally in Nematostella, so this homology assignment must be regarded as tentative pending additional evidence. Also, the two most closely linked of these putative DUX homeoboxes (DuxA and DuxC) are extremely similar at the nucleotide level, both within the homeobox itself and in an intron that interrupts the homeobox. This is a region of the assembly rife with repeated sequence, a condition that would be consistent with either a very recent tandem duplication or a false duplication caused by an error in the assembly. A molecular analysis of this region will be required to verify the assembly.
SINE class
SINE class genes (for example, Drosophila sine oculis and vertebrate six genes) possess a highly distinctive homeodomain in addition to a conserved Six/so domain, 120 amino acids in length, that is located upstream of the homeodomain. Three families are recognized (SIX1/2, SIX3/6, and SIX4/5) 6. All three families have been reported from the Cnidaria previously 4569. A single SIX1/2 class gene has also been recovered from sponges 45.
We identified six SINE homeodomains in human, three in fly, and five in Nematostella. Both the neighbor-joining and Bayesian trees support the monophyly of the SINE class and the monophyly of each of its constituent families. On both trees, Nematostella homeodomain NVHD073 groups with the SIX1/2 family, NVHD128 groups with the SIX3/6 family, and NVHD030 groups with the SIX4/5 family. Two other Nematostella homeodomains (NVHD061 and NVHD093) fall within the SINE class, but their exact phylogenetic positions differ between trees. All five of these predicted homeodomain sequences are located in close proximity to predicted Six/so domains (data not shown). The findings of this study and previous studies make it very clear that the SINE family had expanded to encompass three distinct members prior to the cnidarian-bilaterian split 4569.
TALE class
Homeodomains of the TALE (three amino acid loop extension) class are characterized by the possession of three extra amino acids in the loop between helix 1 and helix 2 of the homeodomain 6. TALE homeodomains have been recovered from bilaterian animals, plants, and fungi 670. We identified 16 TALE class homeodomains from human, 7 from Drosophila, and 6 from Nematostella. This appears to be the first report of TALE class homeodomains in a non-bilaterian metazoan. On both the neighbor-joining and Bayesian trees, the four recognized families of TALE homeodomains appear monophyletic: IRX, MEIS, PBX, and TGIF 6. All four families are represented in the Nematostella genome. On both trees, Nematostella homeodomain NVHD108 groups with the IRX class, NVHD107 groups with the MEIS class, NVHD040 groups with the PBX class, and NVHD149 groups with the TGIF class. Two Nematostella homeodomain sequences (NVHD036 and NVHD143) fall within the TALE radiation, but their precise position differs between the neighbor-joining and Bayesian trees. Five of the six of the Nematostella TALE homeodomains contain three extra amino acids in the same position as in human and fly. The sixth, NVHD036 actually contains four extra amino acids in this location. In five of six Nematostella TALE homeodomains, the first extra residue is a histidine, just as in bilaterians.
ZF class
Proteins of the ZF class are known to encode as many as 4 homeodomains and 17 zinc fingers 6. The homeodomain sequences are highly divergent. It has been suggested that the large number of DNA-binding domains present per protein might reduce the evolutionary constraints operating on the evolution of each individual DNA-binding domain 6. Presumably, the shared possession of zinc fingers reflects a shared common ancestry of ZF class homeodomains. However, neither of the homeodomain phylogenies supports the monophyly of this class. A few well supported ZF homeodomain families can be recognized on both trees, but none of these families includes a Nematostella representative (Additional data files 1 to 3). At this time, it appears possible that this homeodomain class is specific to bilaterians.
Introns
The presence or absence of introns and their location relative to the homeodomain may provide evidence regarding homeodomain phylogeny. However, in the Bilateria, this trait appears evolutionarily labile, and so the phylogenetic utility of homeodomain introns may be compromised by rampant homoplasy 3. In the Bilateria, homeobox genes from all 10 classes may possess introns that interrupt the homeodomain, and these introns have been found to occur at over 20 different positions within the homeodomain (Additional data file 1) 3.
In contrast to the Bilateria, in Nematostella, the presence and location of homeodomain-interrupting introns appears much more evolutionarily stable (Additional data file 1). In Nematostella, only the HNF, PRD, and TALE class exhibit introns within the homeodomain. Furthermore, the location of introns within the homeodomain is highly consistent. Of the 130 Nematostella homeodomains included in this study, 38 are interrupted by introns (Additional data file 1). Three Nematostella homeodomains are interrupted by two introns each (NVHD170 of the HNF class plus NVHD107 and NVHD036, both of the TALE class). The overwhelming majority of these introns (33/41) are located at nucleotide position 139 of the canonical 180-nucleotide homeobox. Nearly all members of the PRD class in Nematostella (31/33) contain an intron at this location. The only PRD class homeodomains to lack an intron at this location are sequences that cannot be assigned to a particular family (NVHD031 and NVHD052).
The possession of an intron at the identical location in nearly all Nematostella PRD homeodomains reinforces the conclusion that the PRD class is monophyletic. One Nematostella homeodomain of uncertain class affinities (NVHD088) also exhibits an intron in the same location as 31 of the PRD sequences. This sequence is nested within the PRD radiation in the Bayesian tree, but it falls outside of the PRD radiation in the neighbor-joining tree. This sequence may in fact be a member of the PRD class.
Three homeodomains from the TALE class and the lone representative of the HNF class are also interrupted by introns in Nematostella. The TALE class homeodomain of NVHD040 (PBX) is interrupted by a single intron at nucleotide position 133 of its 189-nucleotide homeobox. The homeoboxes of two other TALE class members, NVHD107 (MEIS) and NVHD036 are each interrupted by two introns. Likewise, the homeodomain of NVHD070 (HNF class) is interrupted by two introns. Two homeodomains whose class membership is ambiguous (NVHD045 and NVHD007) are interrupted by a single intron at nucleotide position 133 of their 189-nucleotide homeoboxes, just as in the TALE class homeodomain NVHD040.
The intron situation in Nematostella contrasts markedly with that in Drosophila and humans. These bilaterian organisms possess many more PRD-class homeodomains that lack introns, many more non-PRD-class homeodomains that contain introns, and the position of introns within the homeodomain is highly variable (Additional data file 1). These data suggest that an intron was introduced at position 139 of the homeobox in the ancestral Paired homeodomain. Subsequently, after the divergence of Cnidaria and Bilateria, there has been a greater constraint on loss or gain of homeodomain introns within the Cnidaria. Additional analyses are needed to determine whether this constraint on intron gain or loss is specific to the homeodomain superfamily or whether it might be a general feature of cnidarian genomes. If intron location proves to be a particularly stable trait in many cnidarian genes, then the Cnidaria may prove extremely valuable for elucidating the early evolution of metazoan gene families.
Discussion
It is clear that a major radiation of homeobox genes occurred prior to the split between the Cnidaria and Bilateria. As expected, human homeodomains substantially outnumber fruit fly or anemone homeodomains. Typically, each homeodomain family contains two to three times as many human representatives as fruit fly representatives. This partly reflects the large scale genomic duplications that are known to have occurred in the history of the deuterostomes 7172. However, it is surprising that the sea anemone, a morphologically simple animal and an outgroup to the Bilateria, would possess substantially more homeodomains than the fruit fly (130 versus 97). This result may be attributed to three factors. The sea anemone inherited a large complement of homeodomains from the cnidarian-bilaterian ancestor, the fruit fly has experienced some apparent homeodomain loss, and the anemone has experienced numerous homeodomain duplications after its divergence from the Bilateria.
Homeodomain families in the cnidarian-bilaterian ancestor
How many homeodomains were present in the cnidarian-bilaterian ancestor? If we infer that every homeodomain family shared by Nematostella and the Bilateria was represented by a single ancestral sequence in their common ancestor, an inference consistent with the phylogenetic analyses, then this ancestor possessed at least 56 homeodomains (Table 1; Figure 3). The phylogenetic affinities of some Nematostella homeodomains are less well supported than others, and it is likely that a few homeodomains are misidentified here. However, our phylogenetic reconstruction seeks to strike a balance between two types of error: misidentifying particular Nematostella homeodomains as orthologs of particular bilaterian homeodomains; and failing to recognize true orthology between particular homeodomains in Nematostella and bilaterians. The latter error forces us to assume evolutionary events (gene duplications) that never actually occurred. The estimate given here for the homeodomain complement of the cnidarian-bilaterian ancestor almost certainly represents an underestimate because more cnidarian homeodomains will be found in the future, and because many sequences that were included in this analysis could not be placed unambiguously into specific families.
Some of these difficult-to-classify sequences may derive directly from ancestral genes that were present in the cnidarian-bilaterian ancestor. For example, on the Bayesian tree, NVHD099 and NVHD106 appear as the sister group to a large clade containing central and posterior Hox families. These cnidarian genes could be directly descended from a single central/posterior ancestral sequence in the cnidarian-bilaterian ancestor. Taking this into account, our estimate for the number of homeoboxes in the genome of the cnidarian-bilaterian ancestor could plausibly be increased from 56 to 57.
Two other factors could cause us to underestimate the number of homeodomains present in the cnidarian-bilaterian ancestor. In some instances, homeodomains derived from a common ancestor may have diverged so substantially in the three lineages represented in this study that they can no longer be recognized as members of the same family. In other instances, gene loss in either Nematostella or the two bilaterian systems could hide the fact that a particular homeodomain was present in the cnidarian-bilaterian ancestor.
Homeodomain families unique to Bilateria
In our dataset, 17 different gene families shared by human and fruit fly appear to be lacking in Nematostella. Five of these are Hox-related homeodomains: HOX3, HOX4, HOX5, HOX6-8, and HOX9-13. Other ANTP class genes that are shared by the bilaterians but missing from Nematostella are BARX, BSH, and EN. Nematostella also appears to lack two CUT families that are shared between human and fruit fly (CUTL and ONECUT), three LIM families (AP, LHX3/4, and LMX), one POU family (POU2), two PRD families (ARIX and PRX), and one ZF family (ZFH2). Additional gene surveys may identify some of these 'missing' genes in the genome of Nematostella or other Cnidaria (for example, the identification of a likely CUT gene in Nematostella that was discussed above). However, if the absence of particular homeodomain families in Cnidaria can be confirmed, then we may one day attribute the evolution of certain bilaterian traits to the origin and diversification of these key developmental regulators. Homeodomain proteins found in Bilateria but apparently lacking in Cnidaria (such as central Hox genes, EN, and BSH) are implicated in the development of important bilaterian body plan features, including segmentation, paired appendages, and brains.
Homeodom