The L. chalumnae genome was downloaded from the Ensembl server (http://www.ensembl.org/; accession LatCha1 GCA_000225785.1). L. menadoensis BAC sequences were obtained from NCBI (http://www.ncbi.nlm.nih.gov/) with following accession numbers: GI:164698640, GI:170514516, GI:189459217, GI:190886531, GI:193083250, GI:237406519, GI:239735715, GI:239835822, GI:239835823, GI:239835824, GI:239835825, GI:239835826, GI:239835827, GI:239835828, GI:239835829, GI:239835830, GI:305644147, GI:305644148 [Birren 2009, NCBI direct submissions], GI:166987259 [28], GI:220898172 (HoxA gene cluster), GI:220898186 (HoxB gene cluster), GI:220898198 (HoxC gene cluster), GI:220898210 (HoxD gene cluster) [13], GI:296011776 [29], GI:407080572 (IgW2 locus), GI:407080573 (IgW1 locus) [Saha 2012, NCBI direct submissions], GI:40789109, GI:50253612, GI:50253613, GI:50284579, GI:50284580, GI:50284581, GI:52077680 [Grimwood 2004, NCBI direct submissions], GI:66912372 [Lau 2010, NCBI direct submission].

In order to determine orthology relationships between L. menadoensis BAC clones and the L. chalumnae genome, sequence comparison was performed using MegaBlast [30]. Best hits were selected and the main alignment diagonals were used to define the maximum match coordinates along with their orientation. Orthologous fragments are listed in Table S1. L. chalumnae and L. menadoensis TEs were localized using the RepeatMasker software [31] with a TE library specifically built for the L. chalumnae genome [6], [26]. TEs were selected according to the following criteria: length ≥100 nucleotides (nt) and divergence to the consensus sequence from the library ≤20%. Low complexity sequences, simple repeats as well as tRNA and rRNA (pseudo)genes were discarded. Remaining elements located in corresponding L. menadoensis BACs and L. chalumnae genome scaffolds were then listed “face to face” as shown in Figure S1, and further manually aligned to visualize orthologous insertions in both species. Species-specific insertions were inspected manually by extracting and comparing both “empty” and “filled” sites using the Muscle alignment software [32].

Using RepeatMasker, masking of L. menadoensis with the L. chalumnae TE library might lead to artifactual split insertions in L. menadoensis due to discrete interspecific sequence differences with L. chalumnae TEs. This would lead to an overestimation of the number of common insertions. To avoid this problem, the number of common insertions between both Latimeria species was estimated on L. chalumnae data. Close hits matching contiguous parts of the same TE sequence from the database and indexed by a common element number in RepeatMasker outputs were counted as a unique insertion. The same procedure was applied on the whole genome of L. chalumnae to estimate its global TE content.

TE insertions were assigned to known TE superfamilies based on Wicker's classification by combining comparative and predictive approaches [16]. In order to look for putative coding regions, TE sequences were submitted to the de novo gene prediction program Genscan [33] and to a BlastX search [34] on the NCBI website with default parameters against the Genbank non-redundant protein sequence database. Insertions were also analyzed with the Censor software [35] that compares sequences with the Repbase repeat database [36]. Structural features such as Long Terminal Repeats (LTRs) and Terminal Inverted Repeats (TIRs) were identified with the Blast bl2seq utility [34]. Target Site Duplications (TSDs), which correspond to the duplication of few nucleotides from the insertion site, were searched at the extremities of insertions. TE expression was analyzed by sequence similarity analysis of a transcriptome of L. menadoensis testis [6] using the Blast algorithm with default parameters. Copy number of the identified TEs was determined in compared genomic sequences by sequence similarity using the same procedure, with filters of minimal length (80% of the considered TE length) and minimal nt sequence identity (80%) classically used to define TE families [16].

Sequences not unambiguously associated to particular TE families were classified using phylogenetic analysis. TE sequences were extracted from BACs and genomic segments and aligned using Muscle [32] with default parameters. Phylogenetic trees were constructed with PhyML [37] using Maximum Likelihood and aLRT values (non-parametric bootstraps) on different types of sequences depending on the element analyzed: reverse transcriptase (RT) and integrase core domain protein sequences for Endogenous RetroViruses (ERVs), RT protein sequences for CR1/L2 non-LTR retrotransposons, and nucleotide sequences for SINE elements. Using the same procedure, a molecular phylogeny was also reconstructed for CR1 elements with nucleotide sequences.

Using a comparative approach, we searched for species-specific insertions, i.e. for sites “filled” by a TE insertion in one species orthologous to “empty” sites (i.e. without insertion) in the other (Table 1, Table 2 and Table S2). Of note, for DNA transposons, what we define here as an “insertion” in one species could alternatively be the result of the excision in the other species of an element inserted in the last common ancestor of both species. Furthermore, insertion polymorphism might reflect transposition in one species after speciation, but also insertion polymorphism at allelic positions in the last common ancestor. In any case, we consider these different possibilities as evidence for relatively recent (<10 my old) transposition events.

Manual inspection of candidates for polymorphic insertions allowed us to exclude a number of false positives that corresponded to stretches of “N” in the L. chalumnae draft genome. Generally the length of such N stretches matched almost exactly the length of the insertion at orthologous positions in L. menadoensis, suggesting that they have been produced during the assembly phase. Pairwise alignment of “empty” and “filled” sites allowed us to define insertion boundaries, as shown in Figure 1. Several insertions presented evidence of degeneration (short size, truncated open reading frames, etc.) (Table 2).

Comparison between both coelacanth species led to the identification of 27 species-specific TE insertions, 13 in L. chalumnae and 14 in L. menadoensis (Tables 1 and 2). Insertion length ranged from 225 to 5,091 nt, with a mean of 1,363 nt. Insertions were classified according to TE ontology [16] using different specific characteristics of the superfamily, in particular the similarity with known TE-encoded proteins and the presence of specific structural features such as LTRs and TIRs. On the whole, identified polymorphic insertions mainly corresponded to CR1 non-LTR retrotransposons (6/13 and 9/14 insertions in L. chalumnae and L. menadoensis, respectively). A reverse transcriptase-encoding region, belonging to CR1 ORF2 [38], was present in 3 and 7 of them, respectively, with only 3 cases (one in L. chalumnae and 2 in L. menadoensis) where it was apparently complete. The two longest L. menadoensis CR1 insertions with complete RT sequences showed 95% of nt identity (insertions 7 and 8; Table 2). They further contained, upstream of the RT, an APE (Apurinic/Apyrimidic Endonuclease) domain also belonging to ORF2 (Figure 1). CR1 insertions appeared truncated in their 5' part, since they did not show any significant similarity with known ORF1 sequences [38], [39]. While ORF2 of insertion 8 presented several frameshifts and two stop codons, insertion 7 ORF2 did not show any degeneration of the sequence. With the exception of a N-terminal deletion removing 140 amino acids of the APE domain due to 5'-truncation, the protein product predicted for this copy showed all amino acids thought to be important for its activity [39]. This suggests that this element might have been recently transposition-competent. Several short CR1 insertions corresponded to the extreme 3' part of the CR1 with no coding potential (insertions 4, 5, 6, 14 and 15, Table 2). Polymorphic insertions of other non-LTR retrotransposons from the L1 (two in L. menadoensis) and L2 (one in L. chalumnae) superfamilies were additionally detected, as well as SINEs (three in L. chalumnae and three in L. menadoensis), including the recently described CoeG-SINE family [27]. LTR retroelements were represented by a strongly corrupted copy of a Gypsy-like retrotransposon in L. chalumnae and an endogenous retrovirus in L. menadoensis (see below). For class II elements, only two insertions with a palindromic structure reminiscent of that of Miniature Inverted TEs (MITEs) were identified, one of them possibly derived from a hAT DNA transposon (as predicted using Censor). Finally, insertions 26 and 27 presented a composite structure. Insertion 26 comprised, sequentially, (i) a partial CR1 non-LTR retrotransposon with RT domain, framed by an “AGT” TSD, (ii) a possibly novel SINE element flanked by an “AAGT” TSD, (iii) half of a LF-SINE and (iv) a tRNA-derived SINE. Insertion 27 observed in L. menadoensis (Table 2) is formed by a CoeG-SINE flanked by two LF-SINEs in direct orientation, while a “solo” LF-SINE is found at the orthologous locus in L. chalumnae. This configuration suggests a deletion in L. chalumnae that occurred through a recombination between the two LF-SINEs framing the CoeG-SINE element in L. menadoensis.

TSDs, which are hallmarks of insertions consisting in few duplicated target site nucleotides (Figure 1), could be clearly identified in 20 out of 27 insertions. Two other insertions occurred in poly-A or AT-rich regions. Thirteen out of 16 CR1 insertions were flanked by TSDs, with no obvious sequence specificity for integration.

Altogether, polymorphic TE insertions covered a total of 13.3 kb in L. chalumnae and 23.5 kb in L. menadoensis, corresponding to approximately 0.23% and 0.41% of the genomic regions analyzed in the two coelacanth species. Hence, transposon activity is responsible for ca. 0.64% of genome divergence between both Latimeria species in the regions considered.

Among the 27 polymorphic sequences identified, six were found to be intronic, the others being located in intergenic regions (Table 2). While half of insertions were located more than 6 kb away from gene exons, five were closer than 1 kb from the next exon; in L. menadoensis, a CR1 element was inserted in an intron of the SRA1 gene, about 200 base pairs next to the closest exon. Hence, some TE insertions are closely linked to coelacanth genes. Additional experiments will be required to determine if these insertions influence the function and evolution of neighbor genes.

The ERV insertion in L. menadoensis (insertion 23) corresponded to the largest of all identified polymorphic insertions (5,091 nt). This sequence showed a significant BlastX similarity (alignment of 113 amino acids with E-value  = 2.2e-10) with an integrase protein encoded by elements from the ERV1 family, which belongs to Epsilon retroviruses [40]. The ERV copy was delimited by two almost identical LTRs (462 nt, 99% of identity) and flanked by TSDs (“AGAT”) (Figure 2B). The orthologous region in L. chalumnae corresponded to a sequence of 462 nt almost identical to the LTRs of the ERV in L. menadoensis and flanked by the AGAT TSD sequences (Figure 2A). Hence, the L. chalumnae sequence corresponds to a so-called solo-LTR, formed through homologous recombination between the LTRs of the original retrovirus element, this eliminating one copy of the LTR as well as the intervening retrovirus sequence.

In order to better characterize the new coelacanth endogenous retrovirus, which was called CoeERV1-1, a consensus sequence was reconstructed from different copies found in the L. chalumnae genome (Figure 2C). This sequence contains LTRs (475 nt long) and a central region encoding Gag (viral capsid), Pol (polyprotein responsible for the synthesis of the viral DNA and its integration into host genome, including protease [Pro], reverse transcriptase [RT], ribonuclease H [RH] and integrase [Int]), and Env (envelope). Compared to this reconstructed sequence, the identified insertion (number 23 in Table 1) is strongly deleted, lacking major parts of the Gag and Pol domains (Figure 2 B–C). Most of the internal part of the insertion does not share any similarity with CoeERV1-1 and other known sequences. We detected 258 fragments of variable sizes similar to CoeERV1-1 in the whole L. chalumnae genome (identity >82%, E-value <10e-20), including four larger copies (ranging from 6,804 to 8,045 nt in size) with LTRs of variable sizes (from 444 to 551 nt) and all partial or complete ORFs. At least 10 of these 258 fragments corresponded to solo-LTRs, 8 being 475 nt long and presenting TSDs. Inspection of sixteen identified TSD sequences of CoeERV1-1 copies (ranging from 3 to 5 nt) suggested preference of insertion into target sites containing the “GT” (in 7 TSDs) or “AC/G” (in 10 TSDs) nucleotide motif (Figure 2C). Two TSDs contained both motifs while one did not contain any of them.

Retroviruses (RV) are classified in seven genera including Alpha-, Beta-, Gamma-, Delta-, Epsilon-, Lenti- and Spuma-viruses [40]. With six identified genera, the mammalian lineage presents the largest diversity of RV among vertebrates. To better understand the origin and evolution of the coelacanth ERV identified in this work, we performed phylogenetic reconstructions based on both RT (ca. 210 amino acids) and integrase core domain sequence alignments (ca. 132 amino acids) (Figure 3 and Figure S2). Coelacanth sequences were found to cluster in the Epsilon-virus group, one of the most spread branches of RV in vertebrates. Interestingly, CoeERV1-1 sequences were closely related to turtle and crocodile RV sequences, a result supported by both RT and integrase-based reconstructions (Figure 3 and Figure S2). Coelacanth ERV sequences share more than 3,000 nucleotides with over 64% of identity with alligator, crocodile and turtle sequences. This strong relatedness could suggest horizontal transfer (HT) between reptiles and coelacanths or infection of both lineages by a same subgroup of related retroviruses.

In order to get more insight into the relative transposition activity of TEs identified as polymorphic in this work, we determined their copy number, including common insertions, by similarity search against the 5.7 Mb of orthologous sequences analyzed in both coelacanth species (filter: length ≥80% of the considered insertion length and sequence identity ≥80%; cf methods and Table 2). In 11 cases, including 4 CR1 non-LTR retrotransposons (insertions 2, 3, 6 and 13, Table 2), both CoeG-SINEs, both LTR elements, one MITE (insertion 25, Table 2) and the 2 composite insertions, we were only able to retrieve the query sequence with these filtering parameters, indicating the absence of other related copies of approximately the same size in the regions analyzed (Table 2). In contrast, other insertions were found reiterated, with copy numbers ranging from 2 to 110 (39 on the average). The highest hit numbers were obtained for CR1 sequences (insertion 14, up to 110 hits) and for the LF-SINE insertion (48 hits). Interestingly, eleven out of 15 CR1 species-specific insertions (insertions 1, 4–7, 9–12, 14 and 15) were very related one to each other: they showed a high degree of nt sequence identity (98–100%, Table S3) and grouped closely together in a CR1 molecular phylogeny (Figure S3). The CR1 subfamily formed by these sequences is responsible for 3.4% (105/3063) of insertions shared by both species, but constituted as much as 40% (11/27) of interspecific polymorphic insertions. Hence, a subfamily of CR1 retrotransposons has been very recently particularly active in coelacanth genomes, strongly contributing to genome divergence between both Latimeria species. L1 and L2 non-LTR retrotransposons as well as a MITE-like element (insertion 24) showed a more modest level of reiteration. The two L1 specific-insertions of L. menadoensis matched against two sets of sequences mutually non-overlapping, indicating that these two L1 correspond to distinct elements.

In order to determine if some of the polymorphic elements identified in this work (or elements closely related to them) might be expressed, and thus to get another clue on their putatively recent activity, TE insertions were used as queries against a L. menadoensis testis transcriptome (see methods). Table 2 presents for each insertion the number of Blast hits obtained with length ≥80 nt and identity ≥95%. Nine elements (2 in L. chalumnae and 7 in L. menadoensis) matched at least 10 times in the transcriptome: five CR1s including the most complete insertions 7 and 8, both L1s, one CoeG-SINE and the ERV, which presents the highest number of hits (41), probably because it is also the longest insertion, the hits being scattered all along its sequence. All nine elements are intergenic, apart from the CoeG-SINE, which is located in an intron. Common hits were obtained for the eleven CR1 elements of the subfamily previously described, with a total number of 29 distinct sequences in the transcriptome. This observation is congruent with the similarity search against genomic sequences suggesting stronger activity of this particular CR1 subfamily.