As already indicated above, the evolutionary analyses performed so far on Athila elements have been focused on comparing RT sequences. However, when we deeply examined the diversity of Athila elements, we detected that the analysis of pol-derived sequences may offer at most a partial view of the patterns of evolution of these elements. We found that many Athila elements are characterized by either of two alternative structures, typical of non-autonomous retrotransposons: 1) LTRs plus a single ORF encoding Gag proteins, or, 2) LTRs plus two ORFs, encoding Gag and Env proteins. We also found that all potentially autonomous Athilas, those with pol sequences (including RTs), also have gag sequences, although they may or may not have env sequences.

These results led us to the idea of reassessing Athila evolution from the point of view of their gag sequences. We reasoned that gag sequences, common to all types of both complete and non-autonomous elements, would provide the most precise picture of the evolutionary history of Athila retrotransposons. We thus built phylogenetic trees based on Athila gag sequences. We must note here that in a previous study, based on RT sequences, Athila and the closest relative of Athila, the env-lacking retrotransposon that we named Little Athila 18 were confounded 20. However, the recent addition of many novel sequences allowed us to confirm that Athila and Little Athila elements are not only often structurally different (Athilas often contain env sequences, while Little Athilas always lack env), but also possess very different sequences and thus are better defined as two different elements. Particularly, we found that they appear as two separate lineages not only in Arabidopsis, but also in species of the Brassica genus. This result demonstrates that Athila and Little Athila split at least 15–20 millions of years ago (our results are summarized in 22). This result was also found by Zhang and Wessler 23 in their general comparison of the elements present in Arabidopsis and Brassica. Those authors also considered Athila and Little Athila as two different elements. Thus, all the subsequent results shown here refer solely to Athila elements, as defined by Marín and Lloréns 18 and Zhang and Wessler 23.

Neighbor-joining and maximum parsimony phylogenetic trees based on gag sequences confirmed that Athila is a complex ensemble, formed by highly differentiated families. Twelve monophyletic, divergent Athila families became apparent in those trees (Figure 1; data for the elements can be found in [Additional file 1]). One of them ("Family 0" in Figure 1) included only highly defective copies and was not further analyzed. Table 1 contains the description of canonical copies for the other 11 families. Apart from the gag sequences of different families being very different, further demonstration for the high degree of diversification among Athila lineages was provided by the fact that LTRs of elements of different families were in general highly divergent and only partially alignable. We obtained estimates for the time when families diverged in the only two cases in which the full sequences of their LTRs could be compared (see Methods). Thus, families IVa and IVb were estimated to split about 2.7 millions of years ago, while families IIIa and IIIb diverged about 3.0 millions of years ago. These results agree well with the upper estimates for the time of insertion within a family, again according to LTR divergence, which reaches 2.4 millions of years (Table 1). If we now extrapolate from the data shown in Figure 1, assuming that divergence in gag sequences is roughly proportional to divergence time, an age of 6 to 10 millions of years for the most ancient splits among Athila families can be estimated. Thus, we can conclude that Athila families, as defined according to gag sequences, are ancient evolutionary lineages. An upper limit of less than 15–20 millions of years can be deduced from the fact that these eleven Arabidopsis families appear as a monophyletic group separated from all Brassica Athilas (data not shown).

Most significantly, only four of the eleven gag-defined families (I, IIIb, IVb and VII) contained elements with pol sequences (see Table 1). These results suggest that most Athila elements, in fact complete families, are non-autonomous, and that they propagate by using the enzymatic machinery provided by elements of other families. Comparative analyses of LTRs demonstrated that non-autonomous families have been multiplying in the genome for periods of time of up to 2 millions of years (Table 1). Obviously, these results also show that all accounts of Athila element evolution published so far, based on RT sequences, offered a very incomplete view of the evolutionary dynamics of this complex ensemble of retroelements.

If we assume that the available sequences correctly represent the diversity of the Athila elements present in Arabidopsis thaliana, we may infer the degree of activity of the different families of elements by their number of active copies. Significantly, most Athila sequences are non functional. Out of the almost 200 sequences of Athila elements analyzed, we detected only 10 potentially active elements, which contained ORFs without any frameshifts or stop codons. These elements also contain all characteristic conserved amino acids of Athila Gag proteins and, those that contain pol sequences, also contain the typical motifs of the active centers of reverse transcriptases and integrases. The 10 elements belonged to seven different families, as follows: 1 element from family II, 1 element from family IVb, 2 elements from family IVc, 1 element from family Va, 1 element from family Vb, 3 elements from family VI and 1 element from family VII codons (see arrows in Figure 1). Interestingly, no element in four of those families (II, IVc, Va and VI) has pol sequences, they just contain gag or gag + env sequences. Thus, only three copies among all Athila elements found so far are potentially autonomous, pol-containing copies. Two of them are from family IVb – corresponding to the "Athila4" element already characterized as potentially autonomous by Marín and Lloréns 18 and Wright and Voytas 20 – and family VII, respectively. These two families contain other elements with pol sequences, albeit defective. The third one belongs, according to its gag sequence, to family Vb, but, surprisingly, all but two elements in this family lack pol sequences. These peculiar elements, named Va-rec in Figure 1, will be discussed in detail in the next section.

LTR comparative analyses showed that the youngest elements in three of the families without potentially active copies, IIIa, IIIb and IVa, retrotransposed 0.8, 1.8 and 1.0 millions of years ago respectively (Table 1). This result may imply that these families are currently extinct. However, the presence of active Athila elements of these families in other Arabidopsis genomes cannot be excluded. For the fourth family without active copies (family I), a very recent insertion (estimated to have occurred 0.07 ± 0.03 millions of years ago; Table 1) was detected, suggesting that this family is still active. On the other hand, the most recent copies of the families with potentially active elements are in general quite young (average: 0.28 ± 0.09 millions of years) suggesting that most or perhaps all of them are still currently replicating.

The low number of potentially active copies and of recently inserted retrotransposons suggest that Athila elements, at least in the strains from which the examined sequences were derived, have in general a very low level of activity. We specifically searched for Athila cDNAs in order to obtain further evidence for the level of activity of these elements. A total of 169 ESTs corresponding to Athila cDNAs were found in the NCBI EST database and 121 of them could be unambiguously assigned to one of the defined Athila families (Table 2). Of them, just 48 ESTs could be assigned to pol-containing elements. However, none of them derived from the pol gene of those elements. Moreover, only 2 ESTs derived from one of the potentially active elements described above (the one in sequence AL163975; family IVc). If we consider that finding ESTs does not necessarily mean that these elements can transpose, that most of these ESTs derive from experimental conditions in which Athilas are known to be derepressed (see review by 24) and that the NCBI database currently contains 1.3 millions of ESTs (an average of more than 40 per gene), it is clear that the level of Athila transcription and in general its ability to replicate must indeed be very limited.

As just detailed, one of the potentially active elements (Va-rec) contained a gag sequence that was included in family Vb in our phylogenetic tree (Acc. No. AB073163; Figure 1). However, closer inspection of this particular copy demonstrated that it was not a typical Vb element. We found that, while no element in family Vb has env sequences (Table 1), the Va-rec element contained an env sequence very similar to those found in family Va. In addition and as already indicated above, Va-rec and another very similar but defective element (Acc. No. AB046433; Figure 1), were the only two elements having Vb-like gag sequences but containing also pol sequences. We noticed that these pol sequences were actually very similar to those found in family VII elements. These strange results, suggestive of a recombination process, led us to consider in more detail the relationships among gag, pol and env sequences for the whole set of Athila retrotransposons (see Figures 2, 3 and [Additional files 2, 3] for the details of the RT and env sequences).

When trees generated with the gag (Figure 1), RT (pol-derived; Figure 2 and see [Additional file 2] for details) and env sequences (Figure 3 and [Additional file 3]) were compared, it was found that they were often congruent (Figures 4A, 4B). However, we observed several significant differences (outlined also in Figures 4A and 4B). They could only be explained by four independent recombination events among elements of distant families. Comparisons among LTRs and coding sequences allowed us to determine the evolutionary histories for those events:

1) Origin of Va-rec elements: We were able to determine that although these elements appeared in the gag-based phylogenetic trees as members of the Vb family, this was an artifact caused by them having a gag sequence of mixed origin. They emerged by the acquisition by an element of family Va of part of the gag gene and a complete pol gene derived from a family VII element (Figure 5A). Va-rec is therefore an element of recombinant origin, generated from a non-autonomous progenitor of the Va family that lacked pol. As we already mentioned, a second Va-rec element, but with stop codons and frameshifts was also detected. This finding demonstrates that Va-rec elements have been active after the recombination process that originated them. The only apparently active Va-rec element found has a relatively recent origin, becoming inserted 0.27 ± 0.07 millions of years ago.

2) Acquisition by a family IIIb element of env sequences originated from a family Va element. This event explains the shift in the position of family IIIb elements in the gag- and env-based trees (Figures 4A, 5B).

3) Recombination between elements of the IVc and VII families, to give rise to family VI. Family VI elements have LTRs and part of the gag sequences that are extremely similar to family VII elements, while the rest of the gag and the env sequences are very similar to those in family IVc (Figure 5C)

4) Acquisition of some family IIIa elements of an env of uncertain origin, generating an additional branch of elements in the env-based tree, that we named IIIa-rec (Figure 5D).

In summary, these results demonstrate that recombination between elements of different families has occurred frequently in the past: at least 4 of the 13 lineages observed in this study (i. e. the 11 families described in Table 1 plus the IIIa-rec and Va-rec lineages, which cannot be detected in gag-based trees), are of recombinant origin. This is probably an understimate, because ancient recombination events or those involving short sequences would remain undetected with our methods. In any case, recombination has been so frequent that none of the phylogenetic trees obtained properly reflected the diversity of Athila retrotransposons. Only tree comparisons allowed us to understand the evolution of these elements.

The fact that 70% of the potentially active copies encode for Gag or Env proteins but not for Pol proteins raises the question of whether the pol-less elements are simply parasites of the pol-containing copies or, alternatively, they may be contributing to their own propagation or to the propagation of other Athila elements. This contribution would require the production of active Gag or Env proteins by the non-autonomous elements. Of course, to conclude that these elements may contribute functional proteins is not enough to find out that the non-autonomous copies contain potentially coding ORFs or finding ESTs derived from these elements. Even then, all they could be propagating strictly in a parasitic way, i. e. depending solely on proteins provided in trans by other elements with their own genes being non-functional or fully repressed.

The question of whether particular elements contain functionally relevant sequences can be tackled by considering the selective pressures acting on them. Particularly, if they are functionally irrelevant, we would expect the coding regions of non-autonomous copies to evolve at a neutral rate. If, on the other hand, we find that evolution in certain coding regions of non-autonomous copies are under negative, purifying selection, then this would be a strong evidence for them being functionally significant. However, a problem that arises in this type of studies is the potential confusion caused by the past effects of negative selection. Imagine that, after an element replicates, there follows a period of time in which the coding sequences of both copies remain active. In that period, selection on their coding regions would generate an accumulation of changes in permissive positions (e. g. third positions of codons) respect to more constrained positions. If later one of these copies becomes inactive, and starts evolving neutrally, changes will accumulate at random. However, if, after some time, we compare both sequences, we may still find evidence for negative selection, due to the imprint of pass negative selection being still detectable. To avoid this problem, we followed the strategy of comparing the ORFs of potentially active Athila elements only with copies that were very closely related. More precisely, we analyzed trios of sequences, consisting on the sequence that we want to analyze, one of its closest relatives and a third, more distant relative. With these trios, for which phylogenetic relationships are obvious, we performed codon-based analyses of the relevant ORFs to establish the specific rates of non-synonymous vs. synonymous changes (ω) for the different branches (see details in Methods). All the relevant results are shown in Table 3 and the whole set of analyses is detailed in [Additional file 4]. The summary of these results is very simple: in all cases, the best model implies strong negative selection on the branch that corresponds to the elements with potentially active ORFs (branch "z" in the schematic tree shown at the top of Table 3). These results strongly reinforce the idea that these ORFs indeed encode functional proteins, which may contribute to either the replication of the copies that carry them or of other elements in trans.

We built databases of Gag, reverse transcriptase and Env proteins using BlastP and TblastN searches against the databases available at the National Center for Biotechnology Information (NCBI). We used as queries multiple representative Athila elements until the searches become saturated. After each search, we aligned the sequences obtained and removed duplicates and partial sequences. All alignments were performed with ClustalX 1.83 39 using default parameters. Alignments were manually corrected when necessary with GeneDoc 2.6 40. We used two methods of phylogenetic inference, neighbor-joining and maximum parsimony, implemented in MEGA2 41 following the methods described in 42. For both methods, statistical support for the branches was assessed performing 1000 bootstrap replicates.

To determine the structure of Athila elements, we first used Blast2sequences searches 43, comparing each element with known gag, RT, integrase and env Athila sequences. ORFs were detected with ORF finder 44. LTR locations were determined by looking for similarity within an element, also with Blast2sequences.

We estimated the insertion time for an element or the divergence time for elements of two different families following the strategy described by San Miguel et al. 45. The nucleotide sequences of either both LTRs of each element or single LTRs of two different elements were aligned and the Kimura two-parameter distance 46 was estimated using MEGA2. The distance obtained was then divided by two (because it refers to changes accumulated in both LTRs) and then again divided by the substitution rate at synonymous sites estimated for the brassicaceae Chs and Adh genes 47, that is, 1.5 10^-8 per site per year.

Recombination events were deduced from incongruent phylogenetic positions of two proteins of a same element or group of elements 48. We searched for the recombination breakpoints by analyzing pairwise alignments of amino acidic sequences and also, at the nucleotide level, following a sliding-window approach implemented in SimPlot 49, which utilizes the DNAPARS and NEIGHBOR programs of the Phylip package 50.

The rate of synonymous and nonsynonymous substitution were estimated using the PAML3.1 package 51, following the strategy of branch-dependent analyses described previously in 52. Three models were analyzed. M0 refers to a model in which all branches are assumed to have the same rate. M1 is a model in which all branches are assumed to evolve at different rates. Finally, M2 is a model in which the branch that leads to the active element ("z" in Table 2) is assumed to evolve at a different rate than the rest.