A typical autonomous Ginger1 element encodes a single approximately 500 to 800 amino acid long TPase that includes an approximately 400 amino acid N-terminal region highly similar to the integrase (INT) encoded by Gypsy LTR retrotransposon (Figure 1). The homologous regions include the H2C2 zinc finger domain, DDE catalytic domain, and the GPY/F motif that mediates multimerization 17. The latter is a hallmark of Gypsy and ERV integrases 18. The amino acid identity between Ginger1 TPases and Gypsy integrases is up to 40% in the approximately 170 amino acid long DDE catalytic region (Figure 1c). Remarkably, in addition to the domains mentioned above, Ginger1 TPases and some Gypsy integrases further share an approximate 40 amino acid motif immediately upstream of the zinc finger domain (Figure 1a, b). This motif is not universal in Gypsy integrases; it is only present in a limited number of integrases belonging to the Athila/Tat group, Cer1 group and a few Athila/Tat related groups containing Gypsy-6-I_HM, Gypsy-15-I_NV, Gypsy-1-I_RO, Gypsy-1-I_DD and DGLT-A1_I (Figures 1b and 2). This motif is designated as YPYY for the four conserved amino acids: Y/F-P-Y/F-Y/F (Figure 1b). So far, we have identified dozens of Ginger1 families in four animal species, including hydra (Hydra magnipapillata), gastropod (Aplysia californica), lancelet (Branchiostoma floridae) and aphid (Acyrthosiphon pisum). The hydra genome harbors the most diverse and abundant Ginger1 elements; a total of 12 Ginger1 families were identified in this species (Table 1). In the Ginger1-2_HM family, the divergence of some elements from the consensus is less than 1%, suggesting that Ginger1-2_HM elements are still active in H. magnipapillata genome. In the current release of B. floridae genome sequences, Ginger-1_BF is found as a single autonomous copy (Table 1), but a few non-autonomous copies carrying the same 5' and 3' end sequences are also found. In the genome of A. pisum, only degenerated Ginger1 elements are recognizable, but complete elements are identified in the remaining three species. The full length Ginger1 elements vary from approximately 2.6 kb to 7 kb, and their terminal inverted repeats (TIRs) are approximately 40 to 270 base pairs (bp) long (Table 1). The target site duplication (TSD) sequences of Ginger1 are 4-bp long and the sequences are highly specific: CCGG (75%) or CCGT/ACGG (24%) (Figure 1a). The 5' ends of Ginger1 elements show the same conserved TGTNR pattern as those of Gypsy LTR retrotransposons.

In addition to the INT domains, extra domains are also found at the C-terminus of all Ginger1 TPases with the exception of Ginger1-1_BF TPase. These domains include the ovarian tumor (OTU) cysteine protease domain (pfam02338), the C-terminal catalytic domain of Ulp1 protease (pfam02902) and the plant homeodomain (PHD) finger motif (smart00249) from the Conserved Domain Database http://www.ncbi.nlm.nih.gov/sites/entrez?db=cdd (Table 1, Figure 1a, e, f). Either OTU or Ulp1 is present in a particular Ginger1 TPase, but not both. Ginger-1_BF TPase does not contain these extra C-terminal domains, however, it is not clear whether Ginger-1_BF TPase itself lacks these C-terminal domains or this particular single copy element contains an internal deletion. It is also worth noting that some Ginger1 TPase encoding sequences are interrupted by 1 to 4 introns (Table 1). Except for the first intron of Ginger-5_HM (GC-AG) and the first intron of Ginger-1_AC (GT-TG), all these introns conform to the canonical GT-AG intron type 19.

To our knowledge, only two families of eukaryotic DNA transposons, TDD-4 and TDD-5, have been reported to code for a Gypsy-like integrase. They are present in the protist Dictyostelium discoideum, and are characterized by 5-bp TSD 1415. Their amino acid identity to Ginger1 TPases is only approximately 24% in the DDE catalytic domain. To find out more members of this potentially new group of Ginger elements, we used the DDE core domain of the TDD-4 TPase as a query in Tblastn or Blastp searches against the available National Center for Biotechnology Information (NCBI) databases http://blast.ncbi.nlm.nih.gov/Blast.cgi. As a result, we identified >6 families of homologous DNA transposons from different metazoan species, including hydra (H. magnipapillata), sea anemone (Nematostella vectensis), aphid (A. pisum), nematode (Trichinella spiralis), sea snails (Littorina saxatilis) and lancelet (B. floridae) (Table 2). All these transposons contain approximately 50 to 180 bp long TIRs and produce 4 bp TSDs, instead of the 5 bp TSDs of TDD-4 and TDD-5. However, as shown below, all these elements and TDD-4 or TDD-5 belong to the same group (Figure 2), referred to as Ginger2. In the fungus Malassezia globosa, we also found a protein named XP_001728957.1 closely related to the Ginger2 TPases. Neither of the two 5-kb regions flanking the XP_001728957.1-coding region encodes the ribonuclease H or reverse transcriptase. Therefore, even if the XP_001728957.1-coding region is not flanked by TIRs, we classify it as a Ginger2-1_MG DNA transposon vestige (Figure 2) rather than an LTR retrotransposon. Like Ginger1 and the vast majority of Gypsy LTR retrotransposons, the termini of Ginger2 follow the same conserved pattern, TGTNR. However, in contrast with the GC-rich 4-bp Ginger1 target sequences, Ginger2 elements preferentially target 6-bp AT-rich sequences, RTATAY (Figure 1a). Ginger2 TPases contain the H2C2 zinc finger domain and the DDE catalytic domain, but they lack the GPY/F and YPYY motifs present in Ginger1 TPases. The lowest pairwise amino acid identity in the DDE catalytic core region of the Ginger2 elements is approximately 30%, compared with the 36% identity within the same region of the Ginger1 group. This suggests that the Ginger2 group is more divergent than the Ginger1 group and it is consistent with the observation that Ginger2 elements are present in protists, fungi and animals, whereas Ginger1 elements were identified only in four animal species.

To better understand the relationships between Ginger1, Ginger2, and the Gypsy lineages, we performed phylogenetic analyses of a wide collection of integrases from Gypsy LTR retrotransposons, exogenous/endogenous retroviruses, and Polinton/Maverick DNA transposons. Copia and BEL integrases and some integrase-like transposases from bacteria and protozoan Trichomonas vaginalis were included as outgroups. These bacterial transposases are from the IS3 and IS481 elements, the protozoan integrase-like TPases are closely related to them. TPase from eukaryotic Mariner/Tc1 and Pogo elements and bacterial IS630 elements are not included for analysis because they are phylogenetically closer to each other 8, and are more distantly related to the integrases encoded by LTR retrotransposons and Ginger than those encoded by IS3 and IS481 elements (data not shown). Although our phylogenetic tree is based on the limited sequence information from the zinc finger domain and the DDE core domain (see Additional file 1), most Gypsy integrases are clustered together away from other older groups, such as IS3/IS481, Copia, BEL and retrovirus 1, with the exception of two lineages ofGypsy-like integrase from fungal species, clustering with Polinton group and retrovirus groups, respectively (Figure 2). Except for the Woot element being separated from the Osvaldo clade, the clades and the polytomy distribution of all known Gypsy lineages are consistent with the other studies based on the analysis of multiple domains 2021. In addition to the known lineages, some extra Gypsy clades also appear in our phylogeny, probably due to a larger data set; some of them might represent new lineages of the Gypsy LTR retrotransposons. Remarkably, the Ginger1 and Ginger2 groups are distinctly separated (Figure 2): Ginger2 integrases tend to group with the integrases of Polinton/Maverick DNA transposons, while Ginger1 are closely grouped with Athila/Tat lineage of Gypsy LTR retrotransposons. Although the YPYY motif is not included in the sequence information used to build the tree, the majority of YPYY motif-containing Gypsy integrases apparently cluster together with the Ginger1 TPases that also contain the YPYY motif (Figure 2), indicating the YPYY motif is genetically significant. However, no Gypsy lineages are found coclustering with Ginger1 or Ginger2 TPases with significant bootstrap values. A similar pattern is also observed in the tree constructed using the different neighbor-joining method (see Additional file 2).

Gypsy integrase-1 gene (Gin-1), encoding a Gypsy integrase-like protein, was thought to have evolved from a Gypsy LTR retrotransposon related to the 412/Mdg1 lineage 22. However, Gin-1 genes and a number of other homologous genes actually evolved from the Ginger1 elements. The most parsimonious scenario is that two independent exaptations of Ginger1 took place during the evolution of vertebrates, which gave rise to two sets of orthologous genes designated here as Gin-1 and Gin-2 (Figure 3). The first exaptation event gave rise to Gin-2 genes and probably happened in the common ancestor of vertebrates, while the second gave rise to Gin-1 genes and happened more recently in the common ancestor of reptiles and mammals. For example, Gin-1 genes are present in lizard Anolis carolinensis (FG759656.1), chicken Gallus gallus (XP_424858.2), opossum Monodelphis domestica (XP_001380076.1) and human (NP_060146.2); Gin-2 genes were found in fish Danio rerio (CAM46974.1), frog Xenopus tropicalis (AAI69154.1), lizard Anolis carolinensis (FG723791.1) and chicken Gallus gallus (XP_415124.1). Gin-2 genes are not found in any currently sequenced mammalian genomes and it is likely that Gin-2 gene was lost in the early stage of mammal evolution. Consistent with this scenario, Gin-1 and Gin-2 encoded proteins form two clusters in the phylogenetic tree, and both cocluster within the Ginger1 families rather than any other Gypsy lineages (Figure 3a, Additional file 3).

Of the six to eight introns in each of the host genes, only three are universally conserved: they are found at the same positions and have the same intron phases (Figure 3b). Strikingly, all the three conserved introns are found in Ginger1-5_HM TPase gene (Figure 3b), which has four introns in total. In addition, the first conserved intron is also present in the Ginger1-6,7,8_HM TPase genes (data not shown). The data strongly indicate that Gin-1 and Gin-2 genes are derived from a Ginger1-5_HM-like element. In invertebrate tunicate Ciona intestinalis, two genes (XM_002130131.1 and FF869668.1) may also be host genes derived from Ginger1-5_HM-like elements. They also contain the three conserved introns (Figure 3b), and their upstream and downstream genes are only within the range of approximately 0.5 kb to 2.5 kb, within which no flanking TIRs are found.

Genomic sequences and mRNA sequences of various species were mainly taken from NCBI GenBank. X. tropicalis genome sequences (release 4.1 assembled scaffolds) were downloaded from Department of Energy (DOE) Joint Genome Institute (JGI) http://www.jgi.doe.gov/. For the phylogenetic analyses, integrase sequences were widely selected from the Gypsy Database (GyDB) http://gydb.uv.es/index.php/Main_Page20, Repbase Update database http://www.girinst.org/repbase/index.html37 and GenBank. Other transposable elements without specification of source in this paper are from the Repbase. Additionally, the sequences of TEs reported in this work have been deposited in Repbase with the same family names listed in Table 1 and Table 2.

The consensus sequence of each Ginger1 or Ginger2 family, if possible, was rebuilt for analyses. The protein coding region of individual DNA transposons or host genes was either deduced from corresponding mRNA sequences, or manually predicted based on the sequence similarities between homologous proteins; exons and introns were determined accordingly. Multiple protein sequence alignments were carried out using MUSCLE 38 and were adjusted manually. Sequence alignments were edited and illustrated with BioEdit 39. Logo representation of the TSD sequences was created by the WebLogo 40. The phylogenetic tree was constructed using neighbor-joining (NJ) method and minimum evolution (ME) method (Poisson correction model, pairwise deletion, 1,000 bootstrap replicates) implemented in the MEGA4 software 41. The gamma parameter for the phylogenetic tree was estimated using PhyMl 42.