Transposable elements (TEs) are short DNA sequences (usually <15 kb) that behave as intragenomic parasites, vertically transmitted through generations 1. According to their molecular structure and life cycle, they are classified into DNA transposons (type 1) and retrotransposons (RTs; type 2), reflecting the absence or presence, respectively, of an RNA intermediate in the transposition process. The latter are further divided into two major classes according to whether or not they are flanked by long terminal repeats (LTRs): LTR RTs and non-LTR RTs 234. TEs have been linked to fundamental genomic features 5 such as size 678, chromosome structure 910 and chromatin organization 11, and their abundance is determined by an equilibrium between their ability to replicate by transposition and the opposed effects of natural selection 112 and host-defense mechanisms 13.

The possibility of stochastic loss means that TEs should be progressively eliminated from the genomes until their extinction, but this contrasts with the fact that they are found in all life forms 2. Horizontal transfer (HT) between species is the most likely means by which TEs can escape vertical extinction 14151617, and an increasing amount of evidence for HT of eukaryote TEs has accumulated over the years, from the classic examples of the P and Mariner elements of Drosophila 1819, to more recent cases described in other dipterans 2021, invertebrates 22, vertebrates - including fish 23 and mammals 24 - and plants 25. Drosophila is the genus whose TEs have been most thoroughly studied. In a recent review, Loreto et al. 26 gathered evidence for over 100 cases of HT of TEs across Drosophila species. However, methodological issues such as ascertainment bias (for example, the use of TE detection methods based on sequence homology, such as PCR or nucleotide sequence comparisons, or the preferential study of young active TE families) mean that this catalogue of HT cases cannot be used as a reference for the relative importance of such events in the evolutionary biology of the pool of active elements in a given genome.

To directly address this issue, we extracted and compared the DNA sequences of all autonomous TEs in the genome sequences of D. melanogaster, D. simulans and D. yakuba. These species were selected on the grounds of the large differences in their relative genomic TE content - 5%, 2% and 12%, respectively 27 - our previous knowledge of their TE repertoire (D. melanogaster), the quality of their genome assemblies, and their phylogenetic relationships, in order to ensure the optimal performance of the TE detection strategies used (see Materials and methods).

The best proof that a DNA fragment shared by two species originated by HT is that the level of nucleotide divergence at its neutrally evolving sites is much lower than the average neutral divergence between the two species' vertically transmitted genomes. Provided that TE sequences are subject to similar evolutionary forces as those that operate over the genomes that host them, this can be used to study the HT of TEs across species (Figure 1) 142628. Using this approach, we compared the patterns of neutral divergence of TEs with those of a comprehensive set of 10,150 nuclear genes from the genomes of the same species 29. Synonymous sites were used as a proxy for neutrally evolving sites. Thus, TE families without coding capacity - non-autonomous - were not included in this study. Our results suggest that a significant fraction of TEs have experienced HT, and allowed us to estimate the genomic rate of HT of TEs amongst these Drosophila species.

We used a combined strategy (see Materials and methods) to retrieve the sequences of all potentially active insertions of autonomous TEs (that is, insertions that covered >80% of the canonical length of any TE with the capacity to encode the enzymes responsible for their transposition) in the genomes of D. melanogaster, D. simulans and D. yakuba.

We considered as members of the same family all insertions generated by transposition of one or various closely related elements - that is, those that displayed 80% or higher sequence homology in at least 80% of the canonical sequence 34. For between-species comparisons, we needed to distinguish 'orthologous' families - that is, those derived from a single family that was active in the two species' most recent common ancestor by the time of their split, or later transmitted by HT between the two species - from 'paralogous' families, originated by differentiation of TE lineages in the species' common ancestor prior to their split, or by HT from species other than those included in this study. To do this, we compared the estimates of synonymous divergence between TEs and nuclear genes from each species and established a threshold above which two TEs would be considered as paralogous. Considering the extra rounds of DNA replication during transposition and the lower fidelity of retrotranscriptases, the rate of neutral evolution of TE-derived sequences is expected to be the same or slightly higher than that typical of neutral sites of the host genomes. Thus, we arbitrarily considered as orthologous all families that displayed a level of synonymous divergence (K_S) below the 97.5% quantile of the distribution of synonymous divergence values for the set of 10,150 nuclear genes between the host species 29 (see below).

In total, we obtained 1,436 insertions and grouped them into 141 orthologous families (Table 1). LTR RTs are the most abundant major type of TE, followed by non-LTR RTs and DNA transposons, although non-LTR RTs are the most abundant in D. simulans. D. melanogaster and D. yakuba display a similar diversity of families, with 97 and 87, respectively, nearly twice as many as the 57 of D. simulans. These results are broadly consistent with the observed fractions of repetitive DNA in the genomes of these species 27. It should be noted that the DINE-1 family was not included in this study as no coding region has been identified; this is by far the most abundant TE in these species, particularly in D. yakuba 3031. Insertions of 72 families were found in more than one species, 28 of which are present in all three species (Figure 2). For four families we were unable to find any insertion covering at least 85% of the coding sequence and these were excluded from the analyses (see Materials and methods).

Synonymous divergence values for pairwise comparisons of the sample of 10,150 nuclear genes from the three host species 29 are nearly normally distributed (mean [2.5%-97.5% quantiles]): 0.126 [0.037-0.230], 0.303 [0.096-0.531] and 0.284 [0.083-0.505], for D. melanogaster versus D. simulans, D. melanogaster versus D. yakuba and D. simulans versus D. yakuba comparisons, respectively. In contrast, the distributions of synonymous divergence estimates for orthologous TEs differ significantly from those for the nuclear genes (Figure 3; P < 0.001, two-tailed Kolmogorov-Smirnov tests). In fact, the probability of randomly drawing a sample from the nuclear genes' K_S values not significantly different from the corresponding sample of TE values was smaller than 0.01 for the three between-species comparisons (Materials and methods). TE divergence estimates display multimodal distributions, with a large fraction of lowly diverged TEs, and two minor peaks of families with K_S values close to the nuclear gene averages and, in the comparisons involving D. yakuba (with a deeper phylogenetic resolution), of highly diverged families.

In a previous study, experimental data obtained for a reduced sample of 14 TE families from the same species by means of PCR amplification and DNA sequencing provided evidence for unexpectedly low K_S values for orthologous TEs from the same species 17. That dataset can be used as an external quality control: out of the 28 possible between species comparisons (14 D. melanogaster TEs compared with their orthologues from D. simulans and D. yakuba) we found five minor discrepancies between the two approaches, which do not affect the overall results. Both studies detected elements representative of the same overall number of families in D. simulans and D. yakuba. However, two families, HMS-Beagle and roo, were PCR-amplified from D. simulans, but have not been detected in the bioinformatic analysis. On the other hand, 412 and F were detected in D. yakuba in the bioinformatic study only. These differences can be attributed to the properties of the techniques used, for the following reasons. First, PCR primers in the study of Sanchez-Gracia et al. 17 were designed to amplify an approximately 1.5 kb fragment of coding DNA from each family. Thus, the only requisite for a TE to be detected by PCR was the presence of a single intact copy of the amplicon region. This means that the PCR technique cannot discriminate defective from potentially active elements, so that PCR amplifications could be mistakenly taken as evidence for the presence of active copies. This could explain the results for HMS-Beagle and roo. Second, PCR primers in the study of Sanchez-Gracia et al17 were designed using D. melanogaster TE sequences as a reference. Considering the large dependency on sequence homology at the priming sites for PCR amplification success, moderately diverged TEs in the other species may have remained undetected by this method. This could explain the failure to amplify some families from D. yakuba DNA (412 and F). Third, it is also conceivable that some of the TE insertions might not have been fully assembled in the complete genome sequences, so that there is a chance that some families with potentially active copies are not represented in the genome sequences. Fourth is the use of different Drosophila strains in the two studies: two isofemale lines from African natural populations of D. simulans and D. yakuba in the study of Sanchez-Gracia et al. 17, and laboratory strains D. simulans w501 and D. yakuba Tai18E2 in the whole genome sequencing projects 27. It is well known that most active TEs segregate at low frequencies in natural populations of Drosophila 13233 and that most families are represented by only a few copies in each genome 3435, so that a certain amount of variation in the number of families represented by full-length copies across individuals of the same species would not be unexpected.

The other discrepancy concerns the opus family. PCR data suggested reduced divergence between D. melanogaster and D. simulans copies (K_S = 0.003), which conflicts with the results from the bioinformatic analysis (K_S = 0.13; Table S1 in Additional data file 1). A closer look at the sequences obtained in the present analysis revealed that three opus sequences were detected in D. simulans but two of them did not fit the length requirements and were excluded. One of these sequences overlaps a 634 bp region of the amplicon obtained by PCR. Interestingly, these D. simulans opus sequences display high sequence homology with the PCR amplicon produced in the study of Sanchez-Gracia et al. 17 (K_S = 0.006), as well as with the canonical sequence of D. melanogaster (K_S = 0.006). It is likely, therefore, that there are at least two lineages of opus elements in D. simulans, one of which displays high homology with D. melanogaster opus sequences. Both of them were detected by our bioinformatics analysis, but the one more similar to the D. melanogaster sequences does not seem to be represented by any intact copy in the sequenced genome. In summary, the comparison of these two independent sets of data confirms that both TE detection methods produce equivalent results regarding the number of detected families and overall patterns of synonymous diversity, and that the bioinformatics approach used here has a better resolution than the PCR method.

Among the 119 pairwise comparisons, we detected 37 families with K_S values lower than the lower 2.5% quantile of the nuclear genes' K_S distributions (Table 2 and Figure 4). LTR RTs display the largest fraction of lowly diverged families (41%), and there is also consistent evidence for lower than expected K_S values for 40% of the comparisons involving DNA-transposons (although the sample size of the latter (N = 5) is too small for strong conclusions to be made), but only for 6% of those involving non-LTR elements. These differences between the main TE groups are statistically significant (P < 0.0001, G_H test). The fraction of shared TEs that display lower than expected divergence does not differ significantly across species (40%, 36% and 36% for D. melanogaster, D. simulans and D. yakuba, respectively).

If synonymous sites from TEs and host nuclear genes evolve at similar rates, these results can only be explained if an unexpectedly high fraction of the TEs analyzed have recently experienced HT among these species. It might be argued that other processes that reduce the levels of variation among homologous TE sequences, such as higher selective constraints, or recurrent gene conversion between insertions of the same family, could slow down the rate of evolution of TEs. However, it is difficult to see how these could explain such low levels of divergence. High selective constraints on TE sequences - for example, to elude host silencing mechanisms - would have the same effect on all sites of the element, such that K_A/K_S values would be expected to be close to one. But this contrasts with the low average K_A/K_S value for the studied TE open reading frames (ORFs; 0.41; 95% confidence interval (CI) 0.27-0.55; Table S1 in Additional data file 1), consistent with purifying selection operating on TE amino acid changes, similar to most host nuclear genes. Selection on codon usage is unlikely because codon bias is very weak for TEs 36 compared with host genes. The relatively larger effective population size of TEs 37 would not greatly increase the efficacy of selection at TE synonymous sites, given that the median numbers of potentially active copies per family in these species are not very large (5.5, 1.0 and 2.5 for families in D. melanogaster, D. simulans and D. yakuba, respectively). Indeed, codon usage in TEs is less biased than in host nuclear genes of these species (mean effective number of codons (ENC) = 54.0 versus 47.1, respectively); similarly, the GC content at third-codon positions in TEs (0.43) is much lower than that of nuclear genes (0.68), and close to the expected equilibrium GC content (0.40) for unconstrained sequences in Drosophila 383940. This suggests a lower effectiveness of selection on synonymous sites of TEs than on host nuclear genes.

Unbiased gene conversion is expected to have a relatively small effect on silent within-species diversity among members of the same family 41, and cannot affect divergence between species that has arisen since the species split. It is possible that AT-biased gene conversion, or GC to AT mutational bias, could reduce the rate of evolution of AT-rich sequences such as synonymous sites in TEs. However, unconstrained intergenic DNA sequences in the D. melanogaster genome are also AT-rich and evolve at a similar rate to synonymous sites in nuclear genes 42, and there is no reason to believe that AT-rich synonymous TE sites should evolve at a slower rate than these.

The ratio of TE K_S values to the mean K_S for nuclear genes of the hosts can be used as an estimate of the time since the most recent common ancestor of orthologous TEs and, thus, to date putative HT events. Assuming vertical transfer, these ratios should be distributed around one, or slightly above one if TEs experience a larger mutation rate than nuclear genes (for example, as a consequence of extra rounds of replication during transposition and lower fidelity of TE replication enzymes). The distributions of these ratios do not vary significantly across the three between-species comparisons (P > 0.05; Kolmogorov-Smirnov tests; Figure S1a in Additional data file 1). They reflect an excess of young TEs that have diverged little as compared with expectations assuming vertical transfer, and are consistent with the observation that Drosophila TEs are much younger than the genomes that harbor them. This is further supported by the fact that the levels of variation among insertions of a given family are much lower within the three species than expected assuming copy number equilibrium. On average, they display one-fifth of the expected diversity assuming equilibrium (Table S2 in Additional data file 1). This is also in good agreement with previous results for D. melanogaster TEs 174344. In addition, nucleotide variants are at lower frequencies (that is, present in fewer insertions) than would be expected under copy number equilibrium, as revealed by the consistently negative results of Tajima's D test 45 (Figure 5; Table S2 in Additional data file 1). This is expected if most insertions have been generated recently from a single or a few active copies for each family, so that most nucleotide changes are found in a new insertion.

There are significant differences in the relative age distributions across the major classes of elements (P < 0.001; χ² heterogeneity test; Figure S1b in Additional data file 1). LTR RTs and DNA transposons are, on average, significantly less-diverged than non-LTR RTs (P< 0.001; χ² heterogeneity test). Overall, LTR RTs contribute to 89% of the putative cases of HT detected, a fraction twice that previously reported in Drosophila 26. Our results also support the notion that HT is rare amongst non-LTR RTs 121626.

The distributions of K_S values among the little-diverged TEs display a peak within the range 0.03-0.05 (Figure 3). If we assume a mutational clock of 0.011 substitutions per nucleotide per million years 46, this suggests that most HT has occurred over a broad period of time centered between 30,000 and 40,000 years ago and prior to the world-wide expansion of D. melanogaster and D. simulans from their ancestral African distribution range, around 15,000 years ago 47.

Among the 48 TE families shared by D. melanogaster and D. simulans, 15 putative cases of HT were detected. Considering that they diverged 5.4 million years ago 46, this yields a rate of 0.058 HT events per family per million years (95% CI, 0.032-0.095, assuming a Poisson distribution). This is twice that observed between either of these species and D. yakuba (0.027 (95% CI, 0.015-0.045) and 0.019 (95% CI, 0.008-0.040), respectively), which suggests a negative association between HT rate and host genetic differentiation. However, longer divergence times between species mean larger probabilities of stochastic loss of TEs from a lineage and lower power of detection (see below). These differences should, therefore, be taken with caution.

Accordingly, with the observed differences described above, the average HT rates for LTR RTs and DNA transposons (mean ± standard error: 0.046 ± 0.015 and 0.047 ± 0.024, respectively) are nearly seven times larger than for non-LTR RTs (0.007 ± 0.004). Overall, our results suggest a rate of 0.035 ± 0.012 HT events per family per million years across these Drosophila species. It should be noted, however, that HT of a TE could happen anytime after the host species split, but the power to identify such events decreases as the time to speciation and the HT events approach each other, so that the possibility that a fraction of little-diverged elements might have been misclassified as vertically transmitted - that is, their K_S values are above the 2.5% quantile of the distribution of K_S values for nuclear genes - cannot be discarded, and this would make our estimates slightly conservative.

These differences between HT rates across TE classes raise the possibility that the current relative abundances of the major groups of elements in these genomes reflect only their very recent history, so that the over-abundance of LTR RTs in D. melanogaster and D. yakuba is a recent phenomenon produced by their currently higher HT rate. Assuming that TE infection of a new host is followed by a period of high transposition activity (Figure 1), this could also explain the discrepancies between direct estimates of the TE transposition rate from mutation accumulation experiments 484950515253 and those based on genome sequence data 44, as the former could reflect higher current transposition rates of recently horizontally transferred elements. However, this would apply only if the rate of HT of new elements to a given species varied widely over time, but the fact that we did not detect significant differences in the fractions of horizontally transferred elements across species argues against this scenario.

One could also speculate on the possibility that the arrival of new active autonomous families to a naïve genome could prompt the mobilization of extant dormant non-autonomous TEs and, thus, be associated with large between species variation in transpositional activity and copy number of non-automous elements, such as is observed for DINE-1 elements across Drosophila species 30.

It would be tempting to invoke the ability of some LTR RTs to produce potentially infectious virus-like particles to explain their higher genomic HT rate 54, but LTR RTs with an env gene (essential for virus-like particle synthesis) do not display a significantly greater HT rate than those that lack it (P = 0.75 in a Fisher exact test; data not shown). Other mechanisms, probably involving the role of a vector, such as a DNA virus 55, bacteria, parasitoids 56 or mites 57, must also play important roles in the HT of TEs among these Drosophila species (reviewed in 1626).

We have identified 1,436 potentially active TEs that represent 141 families in the genomes of D. melanogaster, D. simulans and D. yakuba. The genome-wide patterns of sequence diversity of these TEs are consistent with the hypothesis that HT plays an essential role in the natural history of TEs. Nearly one-third of the autonomous families have originated by recent HT between these species. This process is more common amongst LTR RTs and DNA transposons than amongst non-LTR RTs. The fraction of TEs generated by HT does not seem to vary significantly across species. Overall, we estimate a HT rate of 0.035 events per TE family per million years.

CI: confidence interval; ENC: effective number of codons; HT: horizontal transfer; LTR: long terminal repeat; ORF: open reading frame; RT: retrotransposon; TE: transposable element.

CB and XM designed the research; CB, XB and XM performed the research; CB, XB and XM wrote the paper.

The following additional data are available with the online version of this paper. Additional data file 1 includes supplementary Tables S1 and S2 and supplementary Figure S1. Table S1: average pairwise nucleotide diversity values at synonymous (K_S) and nonsynonymous (K_A) sites for orthologous TE families from D. melanogaster, D. simulans and D. yakuba. Table S2: genetic diversity values at synonymous sites for transposable elements in the genomes of D. melanogaster, D. simulans and D. yakuba. Figure S1: distribution of the pairwise genetic distances between TE families found in more than one species.