Introduction

The vast majority (~90 %) of TE insertions in the human genome correspond to retrotransposons (Lander et al. 2001), whose proliferation involves reverse transcription of an RNA intermediate into a cDNA that integrates in the host cell DNA. A small part of retrotransposons (1.5 %) results from the endogenization of retroviruses that once infected the germ line of our ancestors, generally at least 50 million years ago (Mya) (Bannert and Kurth 2006; Lander et al. 2001). Upon integration into the host genome as a provirus, ERV insertions are thereafter vertically transmitted following Mendelian rules. Altogether, the human genome contains ~200,000 ERV insertions (Lander et al. 2001), distributed among 3 classes and at least 30 families (Bannert and Kurth 2006; Gifford and Tristem 2003). Most of these insertions are solitary long terminal repeats (LTRs), produced by recombination between the flanking LTRs of a full-length provirus (Lander et al. 2001). The rest of them, with very few exceptions (de Parseval and Heidmann 2005), correspond to “fossils” that many Mya lost their capacity to transpose (Bannert and Kurth 2006; Feschotte and Gilbert 2012), although they may occasionally be transcribed (Seifarth et al. 2005) and translated with relevant consequences for their hosts (Cordaux and Batzer 2009; Jern and Coffin 2008; Lee et al. 2008). Among these families, only ERVK (HML-2) is considered likely to be still active in humans, based on the finding of several human-specific copies (~7 % of all the insertions of this family), either fixed or still segregating in our species (Buzdin 2007; Mills et al. 2007; Jha et al. 2011; Shin et al. 2013). However, not a single active ERVK copy has been found so far, although its artificial reconstruction from existing members has been possible (Dewannieux et al. 2006; Lee and Bieniasz 2007).

ERVE is a C-type endogenous retrovirus, whose prototypic element (M10976) is a 8.8 kb defective provirus bounded by LTRs of 495 nt, with overall 40 % amino acid similarity with Moloney murine leukemia virus in the gag and pol regions (Repaske et al. 1985). Results from Southern blotting analyses (Repaske et al. 1983; Steele et al. 1984) and BLAST screening of human DNA databases, suggest that the human genome contains up to 50 full-length insertions of this family, mostly confined to CpG- and Alu-rich early replicating regions, preferentially near breakpoints and fragile sites (Taruscio et al. 2002). Perhaps due in part to this bias, ERVE insertions contribute to a disproportionately large fraction of all retroelement-derived promoter or enhancer sequences of cellular genes (Bannert and Kurth 2004). On the other hand, although this family has been repeatedly associated to several forms of cancer and autoimmune diseases (Dolei 2006; Ogasawara et al. 2000; Piotrowski et al. 2005; Prusty et al. 2008; Takahashi et al. 2008; Voisset et al. 2008), its precise role in the pathogenesis remains to be elucidated. In 2002 a complete genome of a “novel” retrovirus isolated from blood cells (genomic DNA) of a patient with chronic myeloid leukemia was submitted to GenBank (HCML-ARV, AF499232—R. Z. Xu and S. Zheng, direct submission; see also AY208746 for the corresponding retroviral RNA). It has all the features of a potentially active provirus, bounded by two LTRs and with apparently intact coding sequences of all retroviral genes: gag, encoding the internal structural protein of the retrovirus, with matrix (MA), capsid (CA), and nucleocapsid (NC) domains; pro, encoding a protease (PR); pol, encoding a polyprotein with reverse transcriptase (RT), ribonuclease H (RH), and integrase (IN) domains; and env, encoding the complex that interacts with cellular receptors, with surface (SU) and transmembrane (TM) domains (standardized nomenclature proposed by Leis et al. 1988). HCML-ARV was later identified as a member of the ERVE family (Prusty et al. 2008); interestingly, its complete sequence was not detected in the published human genome, although part of it was apparently localized on chromosome 2q37. The potential of HCML-ARV to reveal new aspects of the evolution of ERVs connected with their possible involvement in human pathogenesis, moved us to carry out a genome-wide analysis that unveiled several salient features of the retroelement, such as molecular signatures of recombination with relatively old proviral insertions at the human genome and of positive selection on the sequence encoding its envelope region.

Materials and methods

Data mining

ERVE sequences from the human genome were retrieved by means of an automated nucleotide and protein sequence homology combined iterative approach (Bartolome et al. 2009). The sequence of the internal region of a full potentially active ERVE element (AF499232) was used as query, and the human genome reference sequence (GRCh37/hg19) as subject. All the sequence hits that displayed ≥60 % nucleotide homology over at least 80 % of the length of the query were regarded as full-length insertions of the ERVE family. Nucleotide sequence alignments were initially obtained with MUSCLE (Edgar 2004) and subsequently corrected by hand with the aid of Bioedit (Hall 1999). ERVE copies bearing long deletions at particular retroviral genes were not used for the corresponding phylogenetic analyses; also, when two insertions appeared to be products of segmental duplication and not of independent transposition, only one of them was kept for further analyses (Table S1). Map positions of the different insertions were obtained with the BLAT alignment tool (Kent 2002) of UCSC Genome Bioinformatics, using genome assembly Feb. 2009 (GRCh37/hg19), and do not include the LTRs.

Phylogenetic analysis

Reconstructions of the phylogenetic relationships among ERVE insertions were carried out with MEGA (Tamura et al. 2007). For neighbor-joining, distances were computed under the maximum composite likelihood model, with the pairwise deletion option; the reliability of the inferred tree was evaluated using the bootstrap resampling technique (1,000 replicates). For maximum parsimony, searches were conducted by close neighbor interchange (level = 1), with initial tree by random addition (20 replications), using all positions; a composite tree was finally obtained using a majority-rule consensus of equally most parsimonious trees.

Evolutionary and statistical analysis

Evidence of positive selection in the phylogenetic tree of ERVE was assessed by maximum likelihood ratio tests (Yang 1998), using the Codeml program in the PAML package. The program produced log likelihood values and maximum likelihood estimates of parameters for each retroviral gene, under different hierarchical models of d N /d S heterogeneity among lineages. To test whether a given model fits the data significantly better than the preceding one in the hierarchy, a likelihood ratio test was carried out, consisting of comparing the corresponding 2Δl (twice the increment in log likelihood) with a χ2 distribution with df = df2 − df1, where df2 and df1 represent the number of free parameters of the models. To assess the potential confounding effect of biased gene conversion on the molecular signature of positive selection, we compared the distribution of strong (S) and weak (W) substitutions for the different genes of HCML-ARV, after their alignment with the reconstructed ancestral sequences of lineage VIIa, with the correspondent distribution obtained from a reference database of alignments of 10,238 human genes to their orthologues in chimpanzee and macaque (Berglund et al. 2009), using Fisher’s exact tests (1-tail).

Characterization of recombination events

Analysis of mosaicism was carried out using SimPlot (Lole et al. 1999). This method is based on the distribution of phylogenetic signals supporting alternative tree topologies among four taxa: the putative mosaic sequence, one representative of each of the two “parental” lineages, and a known outgroup. We used the neighbor-joining method based on the Kimura two-parameter distance model, with a transition/transversion bias of 2.0, after generating 1,000 bootstrap replicates. SimPlot calculates and graphically plots the phylogenetic identity (bootscanning) of the query sequence to the “parental” sequences, in a sliding window of 200 nucleotides that we programmed to move along the nucleotide alignment in steps of 20.

Protein structure prediction

Predicted structures of Env protein domains were obtained using the Protein Homology/analogY Recognition Engine (Kelley and Sternberg 2009). From the ranking of possible templates obtained with the engine, the first hit—the crystal structure of FLV receptor-binding domain from Feline leukemia virus (PDB: d1lcsa)—was chosen as the best template (20 % id; E-value, 9.2e−14; estimated precision, 100 %). Visualization was accomplished with FirstGlance in Jmol (http://molvis.sdsc.edu/fgij).

Human samples

Two sets of samples were used in this study. One consisted of 18 anonymous DNA samples from chronic myeloid leukemia patients of the Hematological Diseases Service of the Complexo Hospitalario Universitario de Santiago de Compostela (Spain). The second was the HGDP-CEPH Human Genome Diversity Panel, which includes DNA samples from 1,063 individuals representing 52 world populations (Cann et al. 2002), and is available from the Foundation Jean Dausset-CEPH in Paris.

PCR analysis and DNA sequencing

The nucleotide sequence of HCML-ARV (AF499232) and 49 other full length ERVE element insertions identified by us in the human genome were aligned with the aid of MUSCLE (Edgar 2004) and corrected by hand in Bioedit (v. 7.1.3.0) (Hall 1999). The resulting alignment was used to identify sequence motifs exclusive for HCML-ARV, which allowed us to design PCR primers specific for this insertion: ERVE-F, 5′AGC TTC ATC AAC CCA CAA GG 3′, beginning at nucleotide 3,632 (see alignment in Fig. S2); and ERVE-R, 5′ GCT GCT TTT CGA GCG TCA 3′, beginning at nucleotide 4,035. This pair of primers should amplify a 398 bp sequence corresponding to positions 4,591–4,988 in AF499232. Aliquots of every four samples of the HGDP-CEPH panel were combined in pooled samples (266 pools in total). PCR reactions were performed using FastStart High Fidelity PCR System (Roche Diagnostics) under manufacturer’s conditions. An initial denaturation step (95 °C for 3 min) was followed by 35 cycles (denaturation at 95 °C for 1′; annealing 68 °C for 1′; elongation at 72 °C for 1′), and a final elongation step at 72 °C for 1′. A PCR assay with a lower annealing temperature (59 °C), using an anonymous human DNA sample, amplified a fragment of an ERVE insertion different from AF499232 (it did not present any of its distinctive sequence motifs; data not shown). As a result of the PCR amplification process, the 5′ and 3′ ends of this amplicon are exact matches of the two primers. This amplicon was used to set positive PCR controls. PCRs were performed in 96-well plates, including 90 pooled samples, three positive, and three negative controls randomly placed in each plate.

URLs

MUSCLE: :

http://www.ebi.ac.uk/Tools/msa/muscle/.

PAML: :

http://abacus.gene.ucl.ac.uk/software/paml.html.

SimPlot: :

http://sray.med.som.jhmi.edu/SCRoftware/simplot/.

UCSC Genome Bioinformatics, Blat Search Human Genome Tool: :

http://genome.ucsc.edu/cgi-bin/hgBlat.

PHYRE: :

http://www.sbg.bio.ic.ac.uk/~phyre/.

Jmol project: :

http://www.jmol.org.

Results and discussion

The few existing studies on the evolutionary history of ERVE (Taruscio et al. 2002; Yi and Kim 2006, 2007) reported no human-specific insertions, but exhaustive analyses of primate genome databases through efficient methods of retroviral insertion identification have not been conducted for this family yet. Using human genome build GRCh37/hg19, we obtained a set of 49 proviral nucleotide sequences (Table S1), whose phylogeny was reconstructed independently for each of several retroviral gene regions: gag, pro, rt (encoding RT), rh (encoding RH), in (encoding IN), and env. We also included in our analysis the sequence of HCML-ARV and an insertion from Pan troglodytes (P_022319), for reasons that will soon become apparent. Irrespectively of the gene or the method of reconstruction (neighbor-joining or maximum parsimony), ERVE copies can be classified into at least eight subfamilies (I through VIIb, according to rules described in López-Sánchez et al. 2005; Fig. 1; supplementary Fig. S1), grouped into two major clusters, deeply rooted in the phylogeny (I–II–III vs IV–V–VI–VIIa–VIIb). The second cluster contains previously characterized ERVE members (Taruscio et al. 2002; Yi and Kim 2007), whereas the first one contains sequences submitted to Repbase as HERV-E(a) (Smit 2008). Sequences of subfamily VIIb are part of a recently described giant chimeric composite transposon (DA Type II/III, Ji and Zhao 2008), that coamplified ERVH and ERVE variants (Lindeskog et al. 1998) most intensely in an interval that extends from 14 to 7 Mya (Li et al. 2009). Using average nucleotide distances within this subfamily as a rough yardstick, it can be estimated that the two major clusters of ERVE initiated their divergence not sooner than 58 Mya. HCML-ARV (abbreviated to HCML in Fig. 1) is included in subfamily VIIa. It should be borne in mind that AF499232 reports a genomic DNA with the characteristic LTRs of a proviral insertion, but it says nothing about the integration site, and not knowing whether there is a target site duplication or not, we can not be certain whether this copy resulted from a new insertion or from a chromosome rearrangement involving other proviral copies of the genome. However, the source of the sequence definitely indicates that it is endogenous, not a replication-competent exogenous virus. In addition, the relatively low identity (97.2 %) between its flanking LTRs clearly precludes the possibility of the copy being the result of a recent new insertion unique to one patient. Besides HCML, the subfamily VIIa includes a most peculiar sequence, H_022319, which contains the putative “part” of HCML-ARV that had been previously detected in the human genome (Prusty et al. 2008). Actually, H_022319 corresponds to an 8.1 kb defective provirus that inserted on chromosome 2 (2q37, map positions 232,265,423–232,273,554, including the 495nt LTRs, in human genome build 37.1) before the split of gibbons and great apes, 18–20 Mya. All orthologous copies in man, chimpanzee, gorilla, orangutan and gibbon share a diagnostic long deletion of most of the in region. The orthologous copy in P. troglodytes is on chromosome 2b (8.4 kb extending from 237,619,424 to 237,627,894 in genome assembly CGSC 2.1/panTro2), denoted as P_022319 in our phylogenetic trees, and shows several private differences and a few shared stop codons and small indels with H_022319 (see alignment in supplementary Fig. S2). The phylogenetic analysis of env (Fig. 1) places HCML-ARV as the closest relative of these two sequences. However, as shown in supplementary Fig. S1, there is a lack of congruence in the phylogenetic signal of the different retroviral regions on this respect. A plausible explanation for this finding is supplied in Fig. 2, which depicts the results of a bootscan test of recombination, showing that the human sequence H_022319 is most likely to be the product of ectopic recombination between the original ERVE, residing in 2q37, and a proviral insertion of HCML-ARV located elsewhere in the genome, similarly to what has been already reported for ERVK (Hughes and Coffin 2005). Three double crossovers would have left a small part of gag, nearly the whole pro, and the majority of env as the only remnants of the original ERVE insertion. According to the present day divergence between H_022319 and HCML-ARV (0.002, excluding sequence tracts from the original insertion), that recombination event dates back to approximately 0.4 Mya. What is more, the close similarity between the 5′ and 3′ flanking LTRs of H_022319 and the corresponding ones reported in AF499232 (four differences out of 987 nucleotide positions) strongly suggests that the recombination could have taken place precisely with this proviral insertion of HCML-ARV. The possibility that there is a primate ERVE sequence, which could be the source of HCML-ARV, seems to be extremely unlikely. Exhaustive searches of primate genomes using different regions from HCML-ARV as queries consistently produce the highest scores with either H022319 or the cluster formed by H022319 and P022319 (Fig. 2), with a single exception. The sequence comprising the end of RH and most of IN (4022–4693 in the alignment shown in Fig. S2), which is missing at H022319 and all its primate orthologs, only differs by two indels (otherwise, 100 % nucleotide identity) from H172358, the human locus of a proviral insertion on chromosome 17 (see also Genbank report AB062274) that we only share with chimpanzees. This strongly indicates that HCML-ARV originated in the human lineage, and that it was involved in ectopic recombination events with two other different ERVE insertions.

Fig. 1
figure 1

Phylogenetic relationships within the ERVE family based on analysis of env gene nucleotide sequences. The displayed tree was obtained by the neighbor-joining (NJ) method, and it is rooted with provirus H_0X7643. Values above branches indicate bootstrap support (1,000 replications) for NJ; below branches, support in the consensus of equally most parsimonious trees; in both cases, only values higher than 85 % are indicated. ω 2–ω 9  = model-based maximum-likelihood estimates of nonsynonymous/synonymous rate ratios (d N /d S ) along the corresponding branches (see also Table 1)

Full size image
Fig. 2
figure 2

Sliding-window bootscan analysis of the triad formed by HCML-ARV and the proviral insertions H_022319 and P_022319. Nucleotide positions (bp) of H_022319 and bootstrap values (%) are depicted on the x- and y-axis, respectively. The upper right box shows the sequences (HCML-ARV and P_022319) used for comparison to H_022319. Arrows point to bootstrap NJ phylogenetic trees drawn from indicated regions of the bootscan plot. Trees were implemented in MEGA with all the insertions from subfamilies VIIa and VIIb. Bootstrap values 75 % and greater are listed at the respective nodes

Full size image

To investigate the forces that have shaped HCML-ARV, we carried out an analysis of the patterns of selection along the branches of the phylogeny of ERVE by maximum likelihood ratio tests (Yang 1998). Whenever a gene has been important for the proliferation of the retrovirus, an asymmetry is expected between internal and external branches in the ratio of nonsynonymous to synonymous substitutions per site, d N/d S (ω). The former correspond to ancestral, transiently proliferating, active sequences (presumably subject to purifying selection), and the latter to present-day specific proviral insertions (presumably fixed and defective products of neutral evolution). Thus, in principle, the significantly lower average ω estimated for the internal branches of the phylogeny of the env gene, as compared to external branches (0.314 vs. 0.933, P < 0.001, Table 1; Fig. 1), should be taken as evidence that the proliferation of the ERVE family was almost entirely due to germ-line reinfection (Belshaw et al. 2005). However, ω values for active sequences of subfamilies VIIa and VIIb (1.421 and 0.905, respectively, at Table 1) deviate significantly (P < 0.001) from this general pattern, indicating either a loss of selective constraints or the action of positive Darwinian selection. The first hypothesis is most likely to be true for subfamily VIIb (associated to giant transposons whose proliferation no longer depends on mobility between cells), but the case of VIIa seems to be entirely different, since it includes HCML-ARV, a proviral insertion with an apparently intact and functional env gene. Moreover, all the other retroviral genes seem to be intact in HCML-ARV too. This is especially remarkable, since the ratio between intact and interrupted open reading frames for retroviral loci in the human genome is extremely small (Ruprecht et al. 2008) (1.8, 0.06 and 0.4 %, for gag, pro-pol—the polyprotein including PR, RT, RH, and IN—and env, respectively). To investigate the role of selection in the evolution of all these genes, we carried out several additional likelihood ratio tests. The results are shown in Table 2, and clearly hint at the action of positive selection on gag, in and env active sequences of VIIa. However, recombination-driven biased gene conversion is known to produce spurious signals of positive selection (high ω values) (Berglund et al. 2009). This factor cannot be discarded for gag and in, which show significant W → S (weak to strong) biases in their patterns of nucleotide substitution (Table 3), but it apparently does not affect env. The predicted structure of a region extending from residues 87 to 252 of Env, identified as a receptor-binding domain, was obtained for both the reconstruction of an ancestral active copy (the consensus sequence of VIIa), and its modern derivative, HCML-ARV. As shown in Fig. 3, amino acid replacements have produced a few conspicuous differences in their secondary structure (the folding/unfolding of two short β-sheets, and the turning of a α-helix into two). It is tempting to speculate that the force behind these changes was resistance to host-encoded control mechanisms of retroviral proliferation, which may work by interfering the binding of Env proteins to cell surface receptors, often mediated, paradoxically, by the Env products of other ERVs (Arnaud et al. 2007; Best et al. 1997).

Table 1 Maximum likelihood analyses of selection patterns along the phylogeny of the ERVE env gene
Full size table
Table 2 Maximum likelihood estimates of d N/d S of different genes and gene regions on the outer branches leading to HCML-ARV (ω 2 in Fig. 1), under a three-ratio model of heterogeneity among lineages
Full size table
Table 3 Patterns of nucleotide substitution in the different genes of HCML-ARV compared to the average human genome
Full size table
Fig. 3
figure 3

Predicted structure for part of the Env protein (165 residues) of the reconstructed consensus of subfamily VIIa of ERVE (a), compared to HCML-ARV (b). The entire protein chain is shown as a smoothed backbone trace; α-helices are shown as rockets, and β-strands as planks; random coils are white. Colour scheme adapted from the DRuMS Standardized Colour Schemes for Macromolecules (http://www.umass.edu/molvis/drums). Aminoacid differences (position followed by single letter code) between both sequences are indicated on the corresponding models

Full size image

The hard prediction from this work is that the patient with chronic myeloid leukemia, wherefrom HCML-ARV DNA was obtained, carried a functional ERVE copy. Unfortunately, all our attempts to reach the authors of the submission to GenBank have failed. Diagnostic features of the HCML-ARV sequence, e.g. a private deletion of 6 nucleotides at position 448 of rh (supplementary Fig. S2), let us carry out PCR-assisted assays to detect individuals bearing this kind of insertions. Using this strategy, we failed to detect HCML-ARV either in 18 anonymous CML Spanish patients or in the 1,063 individuals of the HGDP-CEPH Human Genome Diversity Panel. This could mean that the former association between CML and the proviral insertion was fortuitous. However, there is still considerable debate over medical diagnostics of CML and related myeloid diseases (Lange and Deininger 2010; Verstovsek 2009), which advises prudence before drawing conclusions on that respect, particularly when considering the many different ways that ERVs might contribute to hematopoietic disorders (Schneider et al. 2009). Different disease-causing alleles may predominate in different populations, so that the outcome of an association study between ERVE and CML could vary between a Spanish and a Chinese sample.

Human endogenous retroviruses (HERVs) have been repeatedly associated to cancer (Ruprecht et al. 2008; Kassiotis 2014), autoimmune diseases (Balada et al. 2009; Nissen et al. 2013; Volkman and Stetson 2014), prion diseases (Lee et al. 2013), and abnormally acute reactions to different kinds of stress (Cho et al. 2008), but the evidence for their precise role in pathogenesis is still quite controversial (Voisset et al. 2008; Young et al. 2012, 2013), pivoting around the impact of variation in ERV insertion profiles on the individual susceptibility to these serious health problems. Since all human beings share the vast majority of TE insertions, the pinpointing of variants that may still be active and segregating in our populations becomes of outmost importance for developing efficient tests of association (Moyes et al. 2007). HCML-ARV, as reported in AF499232, is a proviral insertion of a human endogenous virus in a so far unknown genome location, which could be segregating in some human (Chinese?) populations. The results of this paper identify HCML-ARV as a representative of a potentially active lineage of ERVE, whose origin traces back to relatively recent genetic exchanges with different proviral copies that inserted in the genome of our ancestors several million of years ago, and that bears distinctive molecular signatures of positive selection on its envelope region which could be the key of its success in running away from the control mechanisms of retroviral proliferation encoded in our genome.

ERVE was a prolific family of endogenous retroviruses that displayed at least eight long-lasting bursts of activity in our ancestors, two of them (subfamilies VIIa and VIIb) still in course after the separation of the gorilla from our common trunk. This offers a wide-open field for recombination that should make us cautious in pronouncing its sentence of death.