We found strong evidence that γ1 ERVs from Eurasian and African species based on the pol and env phylogenies belong to two different lineages. This pattern was also observed for LTR phylogenies sampled from Eurasian and African host species . Host species from Eurasia are more similar to one another, showing shorter branch lengths compared to the more differentiated African host species with longer branch lengths, consistent with traditional taxonomy which puts all Eurasian species in the genus Sus, whereas there are several genera recognized in Africa [32–34]. A similar pattern was generally observed in γ1 ERV class A and B env phylogenies and for the LTRs, suggesting that proviruses might be codiverging with their respective host species.
The resolution of the pol phylogeny into distinct Eurasian and African lineages, which was not found with gag, is not a consequence of analysis of longer alignments. Bayesian analysis based on 885 bp of pol sequence (the same size as the gag alignment) showed an almost identical topology to that observed in Figure 5, showing a single African clade of ERVs (including the one ERV sequence of the draft pig genome). Similarly, when ca. 400 bp of env is analysed, an almost identical topology of that of Figure 3 is also observed .
The gag phylogeny is different from the other viral gene phylogenies and to the host phylogeny [22, 23]. This may be explained by the gene tree versus species tree problem in which gene phylogenies sometimes conflict with a species tree because individual gene sequences can generate different topologies and consequently conflicting results [35–37], but is also likely to be due simply to the lack of resolution in the gag phylogeny which cannot be attributed to the length of base pairs analysed compared to the other genes. The lack of resolution of this gene was also observed by other authors when analysing other retroviruses [for an example see [38, 39]]. Alternatively, the DNA sequence conservation of gag gene suggests that Gag proteins are important for viral replication. Experimental analysis of Moloney murine leukemia virus showed that mutations of several portions of gag were incompatible with viral replication , implying strong selection for conservation of the gag gene sequence. In this case, while other viral domains might have evolved distinct African and Eurasian sequences, gag might have been constrained from doing so.
Class A and B env phylogenies showed two likely subpopulations in African suids, one in Phacochoerus species, and another in Potamochoerus and Hylochoerus species more related to ERVs from Sus species, (Figures 3 and 4). Amplification of orthologous sequences as well as full length proviral sequences would provide a better basis for comparison, but for reasons of practicality and convenience paralogous sequences amplified by PCR have been routinely used to determine the evolutionary relationship of ERVs in different host species [41–43].
Recombination was apparent in several γ1 and γ2 sequences, presumably as a result of co-packaging of different RNAs in the same viral particle. In Sus scrofa, recombination between highly variable env sequences is easier to detect than for gag and pol. Moreover, selection would be also more likely to be operating on env than on the other genes because the viral envelope must evolve to escape immune detection, thus making recombination a likely source of adaptative phenotypes. ERVs from African species showed more evidence of recombination than ERVs from Eurasian species, although this apparent difference may be an artifact of the higher differentiation (longer branch lengths) for pol and for class A and B env in African species. Furthermore, the RDP 3 software identified a higher number of recombinants than the PHI-NNet algorithm for γ1 genes, a fact also observed in HIV by Lamers et al. (2009) and attributed to the molecular models implemented in some RDP 3 methods, to the number of sequences in each analysis and to variation within each subpopulation altering the number of identified recombinants . This might be why the γ1 pol sequence from the draft pig genome was identified as recombinant by RDP 3 and non-recombinant by the PHI-NNet algorithm. These discordances obviously result from the different methods of analyses, pointing to the need of improving their consistency and power especially in detecting recombinants between ERV sequences. This improvement is also important to better characterize the inter subpopulation recombinants observed in this study and confirm that they represent genuine recombinant sequences.
We also found evidence for past purifying selection on all γ1 genes. This suggested that retroviruses that remained functional were initially selected and that transposition by reinfection was relevant, while loss of function would have likely resulted in neutral evolution within the host genome ω ≈ 1. Furthermore, the rarity of stop codons and/or indels also suggested that complementation in trans and retrotransposition in cis were not relevant mechanisms of transposition [17, 18].
Evolutionary comparisons of γ1 and γ2 env genes
In contrast to the γ1 phylogeny, γ2 ERV env gene showed a bush-like unresolved tree, (Figure 7) confirmed by the likelihood-mapping, either resulting from inadequate sampling or data (soft polytomy) or reflecting the actual evolutionary history of γ2 ERVs in the Suidae (hard polytomy) [45–47]. The bush-like pattern might reflect an incomplete sampling because env from ten other suid species is missing in the phylogeny. However, the γ1 ERV env phylogenies (Figures 3 and 4) using fewer species showed a better resolved tree with species from Eurasia grouping apart from African suids, suggesting that the number of viral sequences samples from each suid host would not be an explanation for the poor resolution of γ2 ERV env phylogeny.
An unresolved phylogeny, like that observed for γ2 ERVs, may also result from data saturation, but the env sequences showed very low saturation. Although we cannot rule out the possibility that the γ2 star-like phylogeny resulted from inadequate sampling, this seems unlikely as samples used here were available from ERVs from three different suid lineages observed by Nascimento [22, 23]. In this case, a hard polytomy seems more plausible. Although Vandamme  emphasised the difficulty of proving the existence of hard polytomies in real life, Poe and Chubb , Barth et al. , Willerslev et al.  and several others [47, 50] have observed some apparent examples in different taxa, including birds, mammals, plants and protozoa, and this pattern was also observed for human ERVs .
This hard polytomy may then represent a rapid radiation from a single common ancestor giving rise to multiple distinct retroviral lineages almost at the same time [19, 45]. These ERVs have subsequently become inactivated resulting in the loss of the last active lineage (also suggested by the presence of several stop codons), following a long period of inactivity leading to the star shaped phylogeny .
The reason why distinct evolutionary histories are observed for γ1 and γ2 ERVs is not known. It would be reasonable to attribute it to the evolution of the host defensive mechanisms. However, Katzourakis et al.  by modelling ERV evolutionary dynamics demonstrated that such a star shaped phylogeny can be generated by a "null model" in which all parameters are set constant through out time.