The genomic basis of cnidarian evolution has so far been viewed from the perspective of an anthozoan, the sea anemoneNematostella vectensis⁶.Hydra is a medusozoan that diverged from anthozoans at least 540 millions year ago. Features ofHydra andNematostella are compared inSupplementary Table 1. We generated draft assemblies of theHydra magnipapillata genome using a whole-genome shotgun approach (Supplementary Information sections 1–3 and Supplementary Figs 1–3). TheHydra genome is (A+T)-rich (71% A+T), and includes ~57% transposable elements (see below). Although the sequenced strain reproduces clonally in the laboratory by asexual budding, it is diploid with substantial heterozygosity (~0.7% single nucleotide polymorphism between alleles), which we find is distributed along the genome as expected if it were drawn from a randomly mating population (Supplementary Information section 3). These features complicate shotgun sequencing and assembly. Two complementary assemblies (CA and RP) were generated (Supplementary Information section 3) and deposited in GenBank. The CA assembly (1.5 gigabases (Gb)) has contig and scaffold N50 values of 12.8 kilobases (kb) and 63.4 kb, respectively. The RP assembly (1.0 Gb) has a contig N50 length of 9.7 kb and a scaffold N50 length of 92.5 kb. The CA assembly gives an estimated non-redundant genome size of 1.05 Gb. The RP assembly gives an estimated non-redundant genome size of 0.9Gb (seeSupplementary Information section 3 for a discussion of genome size calculations). For analysis, we chose the assembly that minimized sequence redundancy owing to the separate assembly of haplotypes (seeSupplementary Information section 3 for further discussion). Approximately 99% of knownHydra genes are found in both assemblies, attesting to their completeness with respect to protein-coding genes.

Although the presentHydra assembly is too fragmented for a chromosome-scale analysis, we found evidence for synteny with other metazoans. Of the 33 longest gene-richHydra scaffolds (that is, those containing genes from at least 10Hydra/Nematostella orthologue groups), 15 (45%) were significantly enriched (P < 0.01) for genes from specific eumetazoan linkage groups⁶, indicating that vestiges of the ancestral eumetazoan genome organization persist inHydra. This is in contrast to the highly diverged genomes ofDrosophila andCaenorhabditis elegans, which show no synteny with other metazoans by these methods.

We estimate that theHydra genome contains ~20,000 bona fide protein-coding genes (excluding transposable elements), based on expressed sequence tags (ESTs), homology andab initio gene prediction (Supplementary Information section 6). The amino acid substitution rate in theHydra lineage is enhanced relative to theNematostella lineage; the sequence divergence between aHydra peptide and itshuman orthologue is typically greater than the sequence divergence betweenNematostella and human (Supplementary Information section 8 and Supplementary Fig. 4) as expected based on the longer branch leading toHydra in peptide-based phylogenies⁶. Similarly, the rate of intron loss has been higher in theHydra lineage; we find that 22%(126 out of 575) of the introns shared byNematostella and human in well-aligned coding regions have been lost inHydra. Conversely, only 6% (28 out of 476) of the introns shared byHydra and human are absent inNematostella.

Transposable elements make up ~57% of theHydra genome and represent over 500 different families (Supplementary Information section 9). The most abundant element, comprising ~15% of the genome (Fig. 1 andSupplementary Table 3), is a non-long-terminal-repeat (non-LTR) retroelement of the chicken repeat 1 (CR1) family. To our knowledge, elements of this family are more abundant in theHydra genome than in any other sequenced animal genome (in comparison, the CR1 family occupies only ~1%of theNematostella assembly and 3% of the chicken assembly). This retrotransposon is still active inHydra, as indicated by its representation in 105 ESTs. We also found 789 cases of intronless genes that were derived recently from multi-exon genes, most probably through retrotransposition. DNA transposons (predominantly ‘cut-and-paste’ elements of the mariner, Transib and hAT (hobo-Ac-Tam3) types) occupy ~20% of both theHydra andNematostella genomes, and are also active inHydra based on the presence of ESTs.

Timing of transposable element activity using sequence divergence of extant copies reveals at least three periods of element expansion (at ~5%, ~20% and ~40% nucleotide substitutions;Fig. 1 andSupplementary Figs 5 and 6). In marked contrast, comparable expansions are absent from theNematostella genome (Supplementary Fig. 7). Most individualHydra transposable element families show discrete bursts of expansion (Fig. 1b, c) that are possibly associated with population bottlenecks⁷. The correspondence between speciation times in the genusHydra and the timing of transposon activity may have been associated with the approximately threefold increase in genome size (Fig. 1a) inH. magnipapillata,H. vulgaris andH. oligactis relative toH. viridissima (380 megabases (Mb))⁸.

Addition of short RNA leader sequences to the 5′ ends of messenger RNAs bytrans-splicing occurs in a subset of metazoans and unicellular eukaryotes⁹. Transcripts from at least one-third of EST-supported genes inHydra undergotrans-spliced leader addition (Supplementary Information section 10).Hydra has multiple spliced leader genes (Supplementary Table 9), and a given transcript may betrans-spliced with several different spliced leaders. Notably,trans-splicing is absent fromNematostella (Supplementary Information section 10). It now seems likely thattrans-splicing has evolved multiple times independently⁹.

Trans-splicing occurs inHydra viridissima (N. A. Stover and R.E.S., unpublished data; GenBank accession numberDQ092354) and in several other hydrozoans (Supplementary Table 10), and may be an ancestral feature of the class. Spliced leader addition gives a eukaryotic cell the opportunity to combine genes into operons, the multi-cistronic transcripts of which can be resolved into individual mRNAs bytrans-splicing. We found 32 potentialHydra operons (Supplementary Information section 10, Supplementary Table 11 and Supplementary Fig. 9), but no obvious evidence for functional relationships between genes in these operons.

Bacteria are stably associated withHydra¹⁰. Electron micrographs reveal bacterial cells underneath the glycocalyx, the coat that overlies the apical surface of the ectodermal epithelial layer ofHydra (Supplementary Fig. 10). Our assembly yielded eight large putative bacterial scaffolds as evidenced by: (1) high G+C content (in contrast to the low G+C content of theHydra genome); (2) no high-copy repeat sequences typical ofHydra scaffolds; and (3) closely spaced single-exon open reading frames with best hits to bacterial genes (Supplementary Information section 11, Supplementary Fig. 11 and Supplementary Table 12). These scaffolds span a total of 4Mb encoding 3,782 single-exon genes and represent an estimated 98% of the bacterial chromosome. Phylogenetic analysis of 16S rRNA (Supplementary Fig. 12) and conserved clusters of orthologous groups of proteins (COGs) indicate that this bacterium is a novelCurvibacter species belonging to the family Comamonadaceae (order Burkholderiales)¹¹. About 60% of annotatedCurvibacter sp. genes have an orthologue in another species of Comamonadaceae (Supplementary Table 13). Notably, theCurvibacter sp. genome encodes nine different ABC sugar transporters, compared to only one or two in other species of Comamonadaceae (Supplementary Table 14), possibly reflecting an adaptation to life in association withHydra.

Non-metazoan genes among cnidarian ESTs have been reported previously¹², and we have now found further examples of such genes in theHydra genome assembly. These genes are candidates for horizontal gene transfer (HGT) (Supplementary Information section 12). Seventy-oneHydra gene models showed closer relationships to bacterial genes than to metazoan genes based on sequence similarity and phylogenetic analysis (Supplementary Table 15). Of these, 51 have no blast hits to other metazoans, except in a few cases toNematostella. Potential donors of these HGT candidates are widely distributed among different bacterial phyla (Supplementary Table 15) and show no enrichment for close relatives ofCurvibacter. Approximately 70% of the HGT candidates have EST support, and transcripts from 30% of the genes have spliced leaders, indicating unambiguously that they are derived fromHydra and not from associated bacteria (Supplementary Table 15). The HGT candidates generally have fewer introns thanHydra genes and nearly one-half are single-exon genes (Supplementary Fig. 14), as expected if they were relatively recently acquired byHydra. A number of the HGT candidates encode sugar-modifying enzymes. Three genes encode enzymes in the branch of the bacterial lipopolysaccharide synthesis pathway that leads to formation of the activated heptose precursor of the lipopolysaccharide inner core (Supplementary Fig. 13). This pathway could modify endogenous glycoproteins or proteoglycans inHydra.

We also identified 90 transposable elements that were potentially horizontally transferred into theHydra genome. These elements have expanded recently (less than 10% nucleotide divergence from their consensus) and have no older copies in the genome. The most frequent element class consists of hAT transposons with 34 different families, although all major classes of transposable element (DNA transposon, LTR and non-LTR elements) are represented. Transposable elements have been shown previously to be horizontally transferred inmetazoans¹³.

We identified 51 unique non-tRNA/non-rRNA transcripts that correspond to putative non-coding RNA genes based on 454 sequencing of short transcripts fromHydra (Supplementary Information section 13 and Supplementary Table 16). At least 17 of these are microRNAs (miRNAs), compared to 40 identified miRNAs inNematostella¹⁴. Surprisingly, only a single miRNA gene in the available data sets,miR-2022, is common to both cnidarian species.

Hox and ParaHox gene families arose from a megacluster that included a number of other homeobox genes (for example, NK genes)¹⁵. With the exception of engrailed, descendants of all of the classes of homeobox genes in the megacluster are found inNematostella^16,17.Hydra is missing a substantial fraction of megacluster descendants¹⁶, indicating secondary loss. For example, theeve andemx genes are absent fromHydra, although they are present inNematostella and several hydrozoans (Supplementary Table 17). The loss of these genes fromHydra is therefore recent in relation to the diversification of hydrozoans. These genes are expressed in a cell-type-specific manner in larvae and adults ofNematostella¹⁷ andHydractinia¹⁸; it is intriguing that the loss of these genes correlates with the absence of a larval stage inHydra (Supplementary Table 17). The absence of these genes inHydra indicates that despite their near-universal presence in animals, it is possible to construct a metazoan without either of them. In addition to the loss ofemx andeve genes,Hydra has undergone several other marked gene losses; for example, it lacks fluorescent protein genes and key circadian rhythm genes (Supplementary Information section 14).

All major bilaterian signalling pathways, including Wnt, transforming growth factor-β, Hedgehog, receptor tyrosine kinase and Notch, are present inHydra andNematostella. An important signalling centre inHydra is the head organizer, which uses the Wnt signalling pathway to establish positional values along the body column^19,20. The head organizer, which is located at the apical tip of the adult polyp, is derived from the gastrula blastopore in cnidarians. A transplanted head organizer has the capacity to induce axis formation²¹, similar to the Spemann–Mangold organizer inXenopus. Orthologues of a number of genes known to act in the Spemann–Mangold organizer inXenopus are present in theHydra andNematostella genomes. Moreover, several of the secreted signalling molecules and transcription factors encoded by these genes are expressed specifically in theHydra head organizer and the blastopore organizer in theNematostella gastrula (Supplementary Information section 15 and Supplementary Table 18). Thus, theHydra head organizer and theXenopus Spemann–Mangold organizer may share common descent from an organizer in the ancestor of cnidarians and bilaterians.

The extracellular portions of twoHydra receptor tyrosine kinases^22,23 contain a novel protein domain, sweet tooth (SWT). The SWT domain is also present in ESTs from the hydrozoanClytia, but is absent from all other sequenced genomes, including that ofNematostella (Supplementary Fig. 15). SWT is among the most abundant protein domains encoded in theHydra genome. The SWT domain is present in one or more copies in predicted secreted proteins. Given its presence in receptors and secreted proteins, we deduce that the SWT domain defines a large, diverse and novel set of signalling proteins.

Hydra contains a pluripotent stem cell type that gives rise to germ cells, nerve cells, nematocytes and secretory cells⁴. Of the five genes that have been shown to induce pluripotency in differentiated somatic cells of mammals (Myc,Nanog,Klf4,Oct4 andSox2)²⁴, homologues of three (Nanog,Klf4 andOct4) are clearly not present in theHydra genome.Hydra has fourMyc homologues. There are two members of the Sox B group inHydra. The Sox B group includesSox2, but the evolutionary relationship between vertebrateSox2 genes andHydra Sox B genes is not clear²⁵. We conclude that the stem cell genetic network inHydra probably has an evolutionary origin independent from the network used in mammalian stem cells. Studies of diverse cnidarians support this scenario (seeSupplementary Information section 14 for details).

Hydra’s shape is formed by epitheliomuscular cells, a cell type unique to cnidarians. A survey of genes that encode muscle structural and regulatory proteins inHydra andNematostella reveals a conserved eumetazoan core actin-myosin contractile machinery shared with bilaterians (Supplementary Table 19). Both cnidarians, however, lack crucial, specific regulators associated with vertebrate striated (troponin complex) or smooth muscles (caldesmon), indicating that these specializations arose after the cnidarian–bilaterian split.Hydra also shows secondary simplifications relative toNematostella, which has a greater degree of muscle-cell-type specialization, including specialized retractor muscle cells.Hydra lacks several components of the dystroglycan complex (α/ε-sarcoglycan and β-sarcoglycan, α/β-dystroglycan and γ-syntrophin), which may lead to a less robust tethering of actin to the cell membrane than inNematostella. Similarly, the absence of a bona fide myosin light chain kinase and phosphatase inHydra indicates a divergence or loss of regulation by myosin regulatory light chain phosphorylation. The greater degree of muscle-cell-type specialization inNematostella is also mirrored in the higher number of myosin light chain genes in this species. Thus, even among cnidarians, we see substantial variation in muscle-associated components superimposed on the eumetazoan core, with theHydra muscular system representing a secondary simplification from a more complex cnidarian ancestor.

Ultrastructural studies show that nerve cells inHydra form synapses on contractile epitheliomuscular cells (Fig. 2a), and that these synapses contain dense core vesicles, paramembranous densities and cleft filaments²⁶ similar to canonical neuromuscular junctions in bilaterians. Several components of the bilaterian neuromuscular junction (choline transporter, nicotinic acetylcholine receptor) are encoded in theHydra genome (Supplementary Information section 16 and Supplementary Table 20) and their expression is consistent with a role in neuromuscular signalling (Supplementary Figs 16 and 17). Other components, however, are found only in a possibly primitive form (putative carnitine acetyltransferases that lack the diagnostic residues for choline selectivity), and some components are absent (the vesicular acetylcholine transporter;Fig. 2b). Together, these data indicate that a canonical bilaterian neuromuscular junction was probably not present in the last common ancestor of cnidarians and bilaterians.Hydra is known to use neuropeptides for the control of behaviour²⁷, and these may be contained in the dense-core vesicles seen atHydra synapses.

InHydra andNematostella, epitheliomuscular cells have an apical junctional belt in the form of a septate junction, clear apical–basal polarity, and hemidesmosome-like contact sites with the extracellular matrix (mesoglea) on their basal surface (Fig. 3a). TheHydra andNematostella genomes encode almost all of the proteins known from bilaterians to be involved in the establishment of cell–cell and cell–substrate contacts (Fig. 3b andSupplementary Fig. 18). This indicates that the common cnidarian–bilaterian ancestor possessed a genetic inventory for the formation of all types of eumetazoan cell–cell and cell–substrate junctions. The presence of innexin genes in theHydra²⁸ andNematostella genomes (Fig. 3b andSupplementary Fig. 19) combined with the lack of connexin genes in non-chordate genomes clearly support the view that innexin-based gap junctions are an ancestral eumetazoan feature, and that gap junctions formed by connexins²⁹ arose later in animal evolution. Similarly, the lack of occludin genes in cnidarians and other non-chordates (Fig. 4) indicates that occludins and their function in tight junction formation first arose in the deuterostome lineage.

Although some gene families associated with cell–cell and cell– substrate interactions are also found in placozoans, demosponges and choanoflagellates, it is important to note that there are cell-adhesion-associated protein domains specific to cnidarians and bilaterians. For example,Hydra andNematostella have classic cadherins exhibiting a highly conserved, bilaterian-type cytoplasmic (CCD) domain(Fig. 3b andSupplementary Fig. 18) that is able to interact with β- and p120/δ-catenin (Supplementary Information section 17). So far, only one sponge cadherin gene that encodes a cytoplasmic domain with weak similarity to the eumetazoan CCD domain has been detected.

The sequencing of theHydra genome has revealed unexpected relationships between the genetic makeup of the animal and its biology. The genes encoding the proteins that form epithelial junctions in bilaterians are present inHydra yet there are obvious differences in structures of the junctional complexes. Despite the morphological similarity of neuromuscular junctions in bilaterians andHydra, several of the key genes required to make this junction in bilaterians are absent fromHydra.Hydra has a complete set of muscle genes but lacks mesoderm and forms muscles only in epithelial cells. Most of the genes required for stem cell pluripotency in mammals are absent fromHydra, yetHydra has a multipotent stem cell system that functions similarly to stem cell systems in bilaterians. The availability of theHydra genome sequence and methods to manipulate it³⁰ provide an opportunity to understand how this remarkable animal evolved.