In an extended genome walk directed from the chromosome end inwards, one of the telomeres from the bdelloid rotifer A. vaga (telomere O.4; 19 ) was found to carry a very long chain of telomere-associated retrotransposons (EF485020; Figure 1a). In addition to two consecutive Athena retroelements, the most proximal of which was 3'-truncated by fusion with the oppositely-oriented retrovirus-like LTR retrotransposon Juno, the head-to-tail retrotransposon chain continued with a non-LTR retrotransposon encoding two open reading frames (ORFs). The first ORF contained three Zn-knuckle motifs (Figure 1a and 1c; Additional File 1) and appeared most similar to the gag-like ORFs from two Drosophila telomere-associated retrotransposons, TART and HeT-A, while the second ORF had homology to the apurinic/apyrimidinic (AP) endonuclease and RT domains from other representatives of the jockey clade, with the highest degree of similarity to ORF2 of the retrotransposon Syrinx from the putatively asexual ostracod crustacean, Darwinula stevensoni 22 . We named this element Hebe, as it appeared to be relatively young and capable of giving rise to new copies.

In order to determine the exact boundaries of the Hebe element, we designed a 1-kb PCR probe spanning the entire core RT domain (Figure 1b) and used it to screen the A. vaga genomic library to obtain additional Hebe copies. A second 0.6-kb probe spanning the 3' untranslated region (UTR) was also used in subsequent library screens in order to find out whether Hebe might give rise to a large number of short 3' truncated copies which would have been missed by the RT probe. Interestingly, hybridization with the second probe yielded few, if any, additional hybridizing spots, indicating a lack of copies which are 5' truncated in the region between the two probes. The number of fosmids obtained from screening ~ 4 A. vaga genome equivalents is shown in Table 1.

All hybridizing fosmids were first sequenced with the primer located at the C-terminus of ORF2 and directed outwards, in order to find out how many independent insertions (flanked by differing genomic sequences) can be identified on these fosmids. The sequences fall into 13 groups defined by the adjacent flanking regions, which correspond to 13 independent insertion events. One group (L/M) consists of two subgroups which share the same deletions and flanking sequences, but differ by six point mutations, indicating that one was recently copied from the other (for example, in the course of segmental duplication, break-induced replication or gene conversion between two members of a co-linear pair). Another group (F/G) shares the 5' flank and 100% identity in sequence, but differs by a deletion involving the 3' end of copy F. Eleven of the 15 sequenced copies do not exhibit 3' truncation and have a characteristic poly(A) tail, which varies in length between 5 and 25 nucleotides and is located 11 bp downstream of the AATAAA signal. A peculiar feature of the 3' UTR is the presence of a 72-bp tandem repeat, the copy number of which varies from 1 to 4 between different copies, yielding variation in the 3' UTR length between 0.8 and 1 kb (Figure 1b).

In order to obtain an independent estimate of the copy number, we performed a Southern analysis of A. vaga genomic DNA digested with restriction endonucleases SpeI and SacII. The latter does not have a recognition site in any of the sequenced Hebe copies, while the former cuts only once (Figure 1b). We used two enzymes to achieve a better resolution of the individual bands on the gel, by digesting away larger amounts of flanking sequences. The use of the 1-kb probe spanning the RT domain yielded a set of bands corresponding to each genomic Hebe copy plus variable amounts of adjacent flanking sequences (Figure 2). The results are in excellent agreement with the estimates obtained from genomic library screening: there are 11 hybridizing bands on the gel, four of which are of double intensity, therefore yielding a total of 15 different copies.

The number of fosmids in each group, corresponding to a single genomic copy, turned out to be lower than that expected on the basis of genome coverage (typically between 1 and 4, with the exception of groups F/G) (Table 1). In comparison, for the same membranes, the corresponding number of fosmids carrying the single-copy hsp82 gene was between four and six for each haplotype 10 23 and the number of fosmids carrying the histone cluster was similarly high (five to seven for each haplotype) 24 . Such under-representation is highly indicative of subterminal localization of the Hebe-containing fosmids, which would be present in the genomic library in lower numbers as the size-selection step in the library construction protocol puts the terminal regions at a disadvantage (see 19 ).

Another line of indirect evidence pointing at localization in subterminal regions is the nature of the surrounding flanking sequences: fosmid end-sequences and the immediately adjacent genomic flanking regions are characterized by features previously found in other bdelloid telomeric fosmids, such as tandem repeats, other TEs, ORFs coding for proteins of repetitive nature or of foreign origin and short stretches of telomeric repeats (Table 1). Finally, telomeric retrotransposons often possess 3' UTRs which are prone to formation of tandem repeats 25 26 and this is also the case for the Hebe element (Figure 1b; Additional File 2).

We also sought to confirm that the 5' and 3' ends of Hebe are equally represented in the genomic library, as was indicated by our initial screening. To this end, we probed two additional membranes with the 5' and 3' Hebe probes of approximately equal length (Figure 1b) and counted the number of hybridizing spots: a total of 101 spots were shared between the two probes; the 5' probe revealed 15 additional spots not detected by the 3' probe; and the 3' probe revealed 26 additional spots not detected by the 5' probe. Thus, there is no significant excess of Hebe copies containing only the 3' end and additional spots in both directions can be explained either by deletions involving one of the termini, as seen in Figure 1b, or by the presence of incomplete copies truncated by cloning.

In order to evaluate the intactness of Hebe copies and the degree of divergence between them, we sequenced these copies by primer walking. The results are shown in Figure 1b (see also Additional File 2). Hebe exhibits a number of peculiar features which are not in agreement with the currently prevailing views on proliferation of non-LTR retrotransposons. First, we could not find any copies exhibiting 5'-terminal truncation, which normally results in formation of a large number of inactive copies and is believed to occur due to the premature dissociation of RT from its template. Second, it is evident that inactivation of individual copies occurred mostly via deletions. Nine copies (D, F/G, J, H, L/M, N, O) carried internal deletions 12-200 bp in length affecting the integrity of their ORFs, and comparison of the sequences at deletion boundaries (Figure 3c) reveals that most of them contain characteristic microhomologies (5-14 bp), which are typically regarded as a hallmark of non-homologous end-joining events resulting in imprecise repair of double strand DNA breaks (DSBs) (reviewed in 27 ). Four copies (F, H, L/M) exhibit 3' terminal truncation, which could have also occurred by deletion, although in this case it is not possible to compare the sequence with its original non-deleted version to reveal the presence of microhomologies. Copies D, F and G carry in-frame stop codons. Copy K contains a 2.3-kb insertion of unknown nature 26 bp upstream from the poly(A) tract. Overall, three copies (A, B and E) may be considered intact, because they carry no obvious defects in their ORFs and possess intact 5' and 3' termini. These copies are flanked by 7-12 bp target site duplications (Figure 3a and 3b) and differ from each other by 29-33 nucleotide substitutions.

Notably, the Hebe element begins with the sequence CATT, which is the canonical initiator motif in many eukaryotes and is characteristic of internal RNA pol II promoters found in Drosophila non-LTR retrotransposons which ensure that a full-length copy does not lose its promoter after retrotransposition 28 29 30 . The 5' UTR is rather short, being only 220 bp in length. There are two more copies (C and I) which we could not sequence in its entirety because fosmids containing these copies were truncated by cloning, but they appeared to be intact in their sequenced part. If these copies are given the benefit of the doubt, this may bring up to five the number of potentially intact Hebe copies in the genome.

The genealogy of Hebe genomic copies is depicted in Figure 4a. Overall, this pattern is characteristic for non-LTR elements, with inactive copies yielding long terminal branches, as they accumulate numerous point mutations in addition to deletions and the potentially active elements yielding much shorter branches, which are indicative of relatively recent activity. Pairwise all-by-all comparison of ORF1 and ORF2 coding sequences reveals an overall excess of synonymous substitutions over non-synonymous ones, which gradually fades away as copies become more decayed (Additional File 3). Again, this is in agreement with relatively recent activity of the element, although no two independent insertions were found which differed by less than 10 nucleotide substitutions (copies A+J and C+E, which potentially represent the most recent retrotransposition events; F/G and L/M are not independent insertions).

Phylogenetic analysis of the 896-aa ORF2 (including EN and RT domains) places Hebe as a basal member of the jockey clade of non-LTR retrotransposons, which also includes TART and HeT-A/TAHRE (Figure 4b), while its closest known RT relative is the Syrinx element from D. stevensoni 22 . The 485-aa ORF1, which codes for a gag-like protein with three zinc knuckle motifs (CCHC; Figure 1b and 1c), is expected to evolve faster than ORF2 and, indeed, it exhibits much lower levels of sequence identity (≤25%, compared to 34% for RT and 32% for EN) to gag-like proteins from other non-LTR retrotransposons (Figure 4c). It is intriguing that the top BLASTP hits included the corresponding gag-like ORFs of Drosophila telomere-associated retrotransposons, some of which have evolved special properties targeting them to telomeric heterochromatin 31 . However, telomeric targeting is known to have evolved independently in members of other non-LTR clades such as R1 32 .

The A. vaga genomic fosmid library 23 was screened with the ³²P-labelled 1-kb RT domain fragment amplified by PCR using a pair of primers F1 (CCAGTGGTTTGATGATGGTGT) and R1 (CTGCTGATACGTTGCCACTTC), and the 0.6-kb 3'UTR fragment amplified with primers 3'UTR-F1 (ATGTCACATACAATCCAGCTTC) and 3'UTR-R1 (GTAACATAAAGTCAACGGAAGG). Selected fosmids were end-sequenced with standard T7 and ccFos primers, split into different groups with the primer seq1 (CAACAAACAACGACATTACACTG) directed into the flanking host sequences, and several fosmids from each group were sequenced by genome walking with custom primers (F1; R1; seq3, AGCCTTTTTCCACATTGCTGG; seq 4, AAAGTTGGACTATCATCTTCG; seq5, GTTGGTGCAAGTCATGGAAAT; seq6, TCGATCTTCTTGATCTTCTGATG; seq7, TGTCATGGATATTGACTTCAGCA). The 5' probe for additional membrane screening to compare representation of 5' and 3' ends was obtained using primers 5p (GATCAGTCGCATTCGTCCAA) and seq7, and the 3' probe - using primers seq1 and 3'UTR-R1. Entire fosmid sequences were obtained by shotgun subcloning into pBluescript II SK- and sequenced on the ABI3730XL at the W M Keck Ecological and Evolutionary Genetics Facility at the Josephine Bay Paul Center for Comparative Molecular Biology and Evolution, Marine Biological Laboratory. Sequences were deposited in GenBank under accession numbers EF485020 and GU176366-GU176379.

A. vaga genomic DNA was sequentially digested with restriction endonucleases SpeI and SacII, fractionated on 0.7% agarose gel, and transferred to Hybond+ membrane (Amersham). The RT probe (about 1 kbp) was amplified from A. vaga genomic DNA using the primer pair F1/R1 described above, gel-purified using the Qiagen Gel Extraction Kit and labelled with ³²P-dCTP using random primers (Invitrogen, CA, USA). The probe was hybridized at high stringency (2×SSC, 65°C overnight).

Alignment (ClustalW) and phylogenetic analysis was done with MEGA4 38 , using either nucleotide sequences (maximum composite likelihood; pairwise deletion; 1000 bootstrap replications) or amino acid sequences (neighbour-joining or minimum evolution; Poisson correction or P-distance; pairwise deletion; 1000 bootstrap replications). Amino acid sequence alignments in BoxShade format are presented in Additional File 1. Pairwise Ka/Ks ratios were calculated by the program DIVERGE from the Wisconsin package (Accelrys Inc., San Diego, CA, USA).