The average levels of sequence variation among the elements within each species are shown in Table 1. Because R1 insertions were found in genomic locations outside the 28S gene, we focused our analysis on the first and last 400 bp of each element and 100 bp of their flanking sequence to insure that all sequences were derived from copies located in the 28S rRNA genes. Except in the specific examples described below, the R1 and R2 elements in each species were extremely uniform, averaging less than 0.2% divergence from the consensus sequence. Because R2 elements are seldom present outside the locus, we also monitored nucleotide variation within internal regions of R2 elements in some species. Sequence divergence for central, coding regions of R2 were estimated at less than 0.1%, similar to or slightly lower than the 5' and 3' untranslated regions (UTRs; not shown).

In Figure 3 the level of nucleotide variation for the 5' and 3' ends of R1 and R2 shown in Table 1 are compared to the levels of nucleotide variation previously found in the 28S genes and internal transcribed spacer (ITS)1 regions of the rDNA units 18. The levels of variation present in R1 and R2 were much higher than that of the 28S gene, and similar to that of the ITS1 region. We have previously shown that the level of nucleotide variation for different regions of the rDNA unit was proportional to the rate at which each region diverged between species 18. This correlation is expected if all regions of the transcribed rDNA unit undergo similar levels of concerted evolution, because increased selective constraints on a sequence removes more variants that arise by mutation, which in turn enables fewer neutral variants to become fixed in all rDNA units (diverge over time). Also shown in Figure 3 (gray bars) are the nucleotide divergence rates of R1 and R2 compared to those for the 28S gene and the ITS1 region. These divergence rates were determined by comparing the consensus sequences of each region from D. melanogaster, D. sechellia, D. simulans and D. yakuba. The relationship between the levels of variation within a species and divergence rates between species that was observed for regions of the rDNA unit was not observed for the R1 and R2 sequences. For example, the 5' end of R1 evolved at four times the rate of the R1 3' end, yet had similar levels of nucleotide variation. The 5' and 3' ends of R2 evolved at one-half the rate of the ITS1 sequences, yet had two to four times the level of nucleotide variation within a species. In this latter example, the slower rate of divergence suggests that the R2 sequences are under greater selective pressure than the ITS1 sequences. Therefore, the finding that the R2 sequences have greater levels of variation suggests that they are not undergoing concerted evolution as effectively as the ITS1 sequences.

In addition to the many highly uniform copies of R1 and R2, five Drosophila species had one or more copies of R1 or R2 with nucleotide divergence of 1% to 7% from the consensus, clearly outside the range of divergences seen for the remaining R1 and R2 copies. The number and level of divergence of these atypical copies are listed in Table 1. Among these copies two had premature stop codons, indicating that they were inactive, while the remaining copies appeared to have intact ORFs. Because most divergent copies were not inserted into 28S genes that were also divergent, these R1 and R2 copies could represent distinct retrotranspositionally competent lineages of elements. However, the number of trace reads suggested these divergent R1s and R2s were at single copy levels, and thus it was likely that they had not recently been active.

Stronger evidence for the formation of nascent subfamilies was found in the examples of distinct 5' ends for the R1 elements of three species ('Variant 5' ends' column in Table 1). In D. simulans there were five distinct sequence classes of R1 5' ends, with each class representing from 11% to 26% of the total number of copies. There was no nucleotide divergence within each class, while divergence between classes ranged from 1% to 3%. In the case of D. pseudoobscura there were two distinct 5' ends. One-third of the R1 copies had 5' ends with over 4% nucleotide divergence from the remaining two-thirds of the R1 copies. Finally, R1 elements in D. ananassae showed the greatest tendency to diverge into subclasses with distinct 5' ends. The R1A elements could be separated into two classes that differed by 10% in nucleotide sequence, while the R1B elements could be separated into four classes that differed by 15% to 22% in sequence.

The separate lineages of the R1 5' ends observed in these three species were not apparent at the 3' ends of the R1 elements (that is, there was one class of 3' ends with mean levels of divergence less than 0.2%). Previous authors have suggested that new sublineages of transposable elements can arise within the same species by the acquisition of new promoter sequences 2930. Thus, one possibility is that the different 5' ends of R1 elements in a species correspond to rapidly evolving promoter sequences. R1 elements have been suggested to contain their own promoters because in some insect lineages R1 inserts in the opposite orientation in the 28S gene, or even outside the rDNA locus 931. R2 elements on the other hand appear to be co-transcribed with the 28S gene and thus do not have their own promoter 3233. No evidence of divergent 5' ends was found for the R2 elements of any Drosophila species.

Finally, Figure 4 summarizes two examples where the formation of nascent families involves sequence divergence of the entire R1 and R2 elements. In D. grimshawi two equally represented groups of R1B elements were detected that had 21% nucleotide divergence at their 5' ends and lower levels in other regions of the element (Figure 4a; Additional data files 12 and 22). The level of divergence for most regions of the two families was less than the divergence between the R1 elements of D. melanogaster and D. simulans, also shown in Figure 4a, suggesting the two R1B subfamilies in D. grimshawi are not as old as the estimated 3 million year separation between D. melanogaster and D. simulans 34. The 5' ends of the subfamilies have undergone accelerated rates of divergence, similar to that described for the different 5' ends of R1 elements in D. ananassae, D. simulans and D. pseudoobscura (Table 1).

A second example of the presence of subfamilies within a species was found for the R2 elements in D. willistoni. In this case one subfamily, R2.1, was highly abundant while the R2.2 subfamily (Additional data file 23) was present in only a few copies. As shown in Figure 4b the subfamilies have diverged by 4.7% in their 5' UTRs and 3.6% in their 3' UTRs, similar to the divergence between the R2 elements of D. melanogaster and D. simulans. The amino acid divergence of the ORF from the two subfamilies (2.6%) was also similar to the divergence between D. simulans and D. melanogaster (2.4%), suggesting the divergence time between the D. willistoni subfamilies is similar to the time of divergence of D. melanogaster and D. simulans.

Unequal crossover events occurring within R1 (or R2) elements would homogenize their sequences, thus preventing the separation of two distinct lineages. Because the nucleotide divergence for most of the region encoding ORF2 of the R1B.1 and R1B.2 elements in D. grimshawi was less than 1%, and thus could still undergo recombination, we looked for evidence of such events. Blast searches were conducted using a query from the end of each subfamily. We then examined the sequence trace from the other end of each approximately 3.5 kb plasmid to determine whether it contained sequence from the same or opposite subfamily. Of 115 informative plasmid ends examined, only one pair indicated recombination between the subfamilies. This paucity of recombination can explain how these nascent subfamilies are able to avoid concerted evolution and remain independent lineages.

The availability of whole genome shotgun sequences has enabled us to evaluate the level of sequence variation of the R1 and R2 elements in 12 species of Drosophila. The level of nucleotide divergence for most copies was typically less than 0.2%, suggesting either the elements are subject to the same concerted evolution mechanisms that enable the rRNA genes themselves to remain nearly identical, or that the R1 and R2 elements are gained and lost rapidly from the locus (that is, all copies are recent insertions). All previous analysis suggested the latter explanation. Analyses of the 5' truncated copies of R1 and R2 in several species have suggested that these elements do turnover rapidly. Different animals from the same population were found to have different collections of 5' truncated copies 4445 and most 5' truncated elements within an animal had not undergone duplication by recombination 4647. Thus, the individual copies of R1 and R2 do not appear to remain in the rDNA locus for long enough periods to be substantially influenced by the recombinations leading to the concerted evolution of the locus.

One puzzling aspect concerning the evolution of R1 and R2 was the presence of multiple families in some species, yet no evidence for the origin of these lineages by horizontal transfer 2848. The rapid turnover of individual R1 and R2 elements suggests that each active copy can generate its own lineage, which over time should accumulate sequence variation. Thus, while separate lineages of R1 and R2 should be able to arise, the question remained as to whether they could be maintained within a species. In this report we have for the first time detected these nascent lineages of R1 and R2 elements within a species. Two distinct subfamilies of R1B elements were detected in D. grimshawi and two distinct subfamilies of R2 elements in D. willistoni (Figure 4). In addition to these distinct lineages, other species contained individual copies of R1 or R2 that had from 1% to 7% nucleotide divergence from the majority of elements in the species. Many of these divergent copies had intact ORFs, and thus were potentially active. Another frequent finding was the presence of distinct 5' UTRs for the R1 elements (Table 1). This was most widespread in D. ananassae where the R1A elements could be divided into two groups that diverged 10% in their 5' UTR sequences, while the R1B elements could be divided into four groups with 5' UTRs that diverged from 15% to 22%. We suggest this accelerated divergence of the 5' UTRs represents the evolution of new promoter sequences driving the transcription of the elements. Once a new promoter is formed, copies containing this new promoter may differ in their expression, thus giving rise to new lineages of elements.

Similar examples of the rapid evolution of the 5' UTRs of R2s were not detected. We have previously suggested that R2 elements do not contain their own promoter but are co-transcribed with the 28S gene 3233. This co-transcription may make it more difficult for independent R2 lineages to evolve. A single lineage of R2 elements has been found in the Drosophila genus 25, while four distinct lineages of Drosophila R1 elements exist 28. It is also possible that the more frequent loss of R2 from a Drosophila species (3 out of 12 species) compared to R1 (no losses among the 12 species) could also be the result of R2's reliance upon co-transcription with the rDNA units in which they reside.

Many of the initial steps involved in the cleavage of the target site bottom strand and its use to prime reverse transcription have been characterized using purified R2 protein in vitro 103849. The recent discoveries that sequences near the 5' end of the RNA transcript regulate top-strand cleavage 11, and that the R2 polymerase can efficiently displace an RNA strand while using a DNA strand as template 50, suggest that the R2 protein also synthesizes the second DNA strand. However, many questions remain concerning whether and how top-strand cleavage of the target DNA is used to prime second-strand synthesis.

Because of an unusual template jumping ability of the R2 polymerase 39, we previously suggested that during reverse transcription the R2 polymerase jumps from the R2 transcript onto the upstream DNA target sequences 53637. The hundreds of R2 5' junctions analyzed here suggest a different model. In Drosophila we propose that the 5' ends of the R2 RNA transcripts contain terminal G residues that, after reverse transcription and top-strand cleavage, enable the terminal C residues to anneal to the G residues of the top DNA strand after cleavage (Figure 5c). This model would explain the most common full-length R2 insertions in all Drosophila species. Similarly, microhomologies between internal R2 sequences and the upstream 28S gene sequences can explain the priming of second-strand DNA synthesis in many 5' truncated R2 insertions. When annealing of microhomologies does not occur, non-templated residues can be observed that were either added during first-strand synthesis when the R2 polymerase runs off an RNA template 39 or during second-strand synthesis before the R2 polymerase engages the cDNA 38.

The only aspect of this integration model that does not agree with previous biochemical studies is the location of the top-strand cleavage site. The experiments using the B. mori R2 protein suggested this cleavage was two nucleotides upstream of the bottom-strand site 10 while our analysis of Drosophila R2 junctions suggest it is opposite the bottom-strand site (Figure 5c). We suggest this represents an evolved difference between the R2 elements in these divergent insects. We have analyzed the 5' junctions of over 40 B. mori R2 insertions and found their structure is consistent with the location of top-strand cleavage determined with the purified protein 37. We suggest that cleavage of the top strand by the R2 endonuclease is not rigidly determined, and thus its location can vary.

Much less was previously known about the R1 retrotransposition mechanism. In vitro studies of the R1 endonuclease, again from B. mori, revealed a bottom-strand cleavage site in a position consistent with its use to prime reverse transcription 1213. All Drosophila R1 3' junctions, as well as 3' junctions in other insect species 19, are consistent with this cleavage site. As in the case of R2, the R1 5' junctions suggest cleavage of the top strand is less precise and subject to evolutionary changes. For many Drosophila R1 insertions top-strand cleavage was proposed to be at a site 14 bp downstream of the bottom strand, again consistent with the biochemical studies. Use of this site to prime second-strand synthesis results in a 14 bp TSD. Minor changes in this cleavage location were found in D. ananassae, where the R1B insertions have 11 or 13 bp TSDs, and in D. willistoni, where some ancestral type R1A insertions have an 8 bp TSD. Annealing of cDNA sequences to the upstream target site presumably does not occur as significant nucleotide variation is observed at the junctions.

A radical change in top-strand cleavage site preference is proposed for the R1A elements of the melanogaster species group. In these species, many R1 insertions result in a 9 bp target site deletion, suggesting cleavage frequently occurs 9 bp upstream of the bottom-strand site (Figure 10). Associated with this remarkable change in cleavage site preference also appears to be a change in the nature of the R1 RNA transcript used for retrotransposition. Based on the structure of the ISR full-length insertions and of non-28S insertions, these melanogaster-type R1 transcripts contain from 6 to 27 nucleotides of 28S sequence ending 9 bp upstream of the integration site. These upstream 28S sequences on full length R1 transcripts appear to anneal to the target site, giving rise to highly precise 5' junctions. Interestingly, these melanogaster-type R1s retain the ability to cleave downstream of the integration site, resulting in the ISRs seen in some full-length insertions and the TSDs observed for all 5' truncated insertions. Because annealing of the upstream 28S sequences is not possible with these downstream cleavage sites, significant variation is observed at these junctions. We searched for evidence of these additional 28S sequence tags among available expressed sequence tag sequences from D. melanogaster. However, all R1 sequences contained extensive upstream 28S sequences or were transcripts of R1 tandem arrays (data not shown).

Thus, the integration mechanisms used by the R1 and R2 elements are quite similar. Cleavage of the bottom strand and its use to prime first-strand DNA synthesis is rigidly determined, and no variation was observed between species. Cleavage of the top strand and its use to prime second-strand synthesis is flexible, which results in different junctions both within and between species. R1 and R2 are highly successful in the 28S niche. Whether change in the location of top-strand cleavage occurs simply because it is neutral or because it increases insertion efficiency is unclear. Nevertheless, the ability of R1 and R2 to explore the top-strand cleavage site suggests that variations of the retrotransposition mechanism can evolve among arthropods that could affect both the elements and the rDNA loci in which they reside.

Original sequencing reads from the whole genome shotgun sequencing projects were accessed by Blast search (version 2.2.17) in the trace archives at NCBI 51. The 12 Drosophila species and their sequencing fold coverage were: D. ananassae (9-fold), D. erecta (10-fold), D. grimshawi (8-fold), D. melanogaster (12-fold), D. mojavensis (8-fold), D. persimilis (4-fold), D. pseudoobscura (9-fold), D. sechellia (5-fold), D. simulans (4-fold; white 501 strain only), D. virilis (8-fold), D. willistoni (8.4-fold) and D. yakuba (9-fold).

Downstream 28S sequences flanking the known R1 and R2 insertion sites were used as Blast queries using default parameters except for requesting the maximum of 20,000 target sequences. The reads were collected, trimmed upstream of the query and aligned in ClustalX 52. Those sequences upstream of the query that were not 28S rRNA sequences were considered putative R1 or R2 elements and were used as Blast queries in the nucleotide collection database (nr/nt) and VecScreen at NCBI 51. Some putative elements were identified as cloning vector sequence and disregarded. For the remaining majority of sequences, 5' extensions of the sequences were assembled by Blast search, followed by alignment and extraction of a consensus. Iterations were done until the 5' junction of the element with the 28S was reached. Full assemblies were not possible in a few instances because of low element copy number and low coverage of some genome sequences.

Alignments of concatenated 18S and 28S sequences (from Stage and Eickbush 18 and available at the Eickbush lab website 53) were done in ClustalX 52 with minor manipulations done in Jalview 54. Default parameters of phyML version 3.0 55 as implemented through the LIRMM website 5657 were used to construct maximum likelihood trees and visualized using TreeDyn 198 58. Branch support values were the minimum score based on Shimodaira-Hasegawa-like and Chi2-based approximate likelihood-ratio tests 59.

Blast queries were 75 bp long near the 5' or 3' end of each consensus R1 or R2 sequence; default Blast parameters were used except for requesting the maximum of 20,000 target sequences and using a match/mismatch score of 1,-1. The Blast results were trimmed to include 400 bp of the element and 100 bp of flanking sequence. Retention of the flanking sequence enabled those elements present in the 28S gene to be separated from those copies inserted elsewhere in the genome. ClustalX alignment 52 of results from each Blast search was conducted and groups of like sequences were extracted and consensus sequences derived. Quality scores for reads containing differences from the main consensus were examined to verify that the variation observed was not sequencing error (scores less than 40 were considered sequencing error). The Sequence Manipulation Suite: Ident and Sim 60 aided determination of the percent identity of the variants relative to the consensus.

In theory it should be possible to make copy number estimates based on the genome sequence coverage and the number of trace reads recovered. In practice, however, random variation in fold coverage for any given sequence means only an estimate of copy number for repetitive sequences can be inferred. In addition, rDNA loci in many species of Drosophila are located on the X and Y chromosomes, resulting in different sequencing coverage. Finally, by requiring analyzed reads to cover 400 bp to enable a significant and uniform level of information to be recovered from each read, fewer reads were obtained for each analysis. Thus, estimates of element abundance were made relative to the number of rDNA units also recovered in the trace archives (that is, fraction of rDNA units inserted). Copies with small numbers of reads (approximately 2 to 12) and identical sequence variation were interpreted as single elements.

Using the 5' and 3' ends of R1 and R2 elements as Blast queries and analyzing the flanking sequences, we found many examples of R1 elements inserted outside the usual 28S site. Identification of the non-28S sequences was attempted by comparing them to our assembled rDNA-related sequences and by Blast search of the NCBI non-redundant nucleotide database 51. While a few target sites could be identified, the majority of sequences had not been characterized and represented repeated sequences, and thus were presumed to be heterochromatic. We then attempted to find uninserted copies of the sequences using the flanking sequences as Blast queries. In those cases where the target site was repeated, we could identify examples of uninserted target sites, and in some instances the opposite junction of the R1 insertion.

Selection analysis was done using the HyPhy package available at Datamonkey 61. We conducted all the available tests and report here the dN/dS results of the PARRIS test 62, which were similar in all the tests.