DNA regions of a very low but significant partial identity (60–70%) to the known TE were also analyzed in the same way as the insertions described above. As noted before (6), such regions usually belong to highly diverged families of young TEs.

Using the majority rule applied to multiple aligned copies of TEs, we built their consensus sequences. Copies of TEs created by chromosomal duplications or redundant sequencing were discarded on the basis of similarity between the corresponding flanking regions.

Distantly related proteins were identified by using PSI-BLAST (15). Multiple alignments of protein sequences were created by CLUSTAL-W (16) and edited manually by using GENEDOC (17). Multiple alignments of DNA sequences were obtained by using the VMALN2 and PALN2 programs developed at the Genetic Information Research Institute. We used FGENESH (ref. 18; www.softberry.com) for identifying genes encoded by TEs. Phylogenetic analysis was conducted by using MEGA (19). The ages of TEs were calculated by using the formula t = K/ν, where t is the average time elapsed since the TE copies were transposed in the genome, K is the average divergence of the TE copies from their consensus sequence, and ν is the neutral substitution rate, 0.016 substitution per site per million years (Myr) (20). Ages of retroviruses were estimated on the basis of pairwise divergences of their LTRs by using the formula t = K/2ν, where K is the divergence between LTRs that flanked the same provirus.

The Transib families are highly divergent from each other; their consensus sequences are only 60–62% identical to each other. At the same time, the identities between the corresponding consensus sequences and copies from the same family are over 90%. The Transib2-4 consensus sequences are 2,844, 2,883, and 2,656 bp long, including 42-, 45-, and 40-bp TIRs, respectively. Analogously to Transib1, Transib2-4 elements have generated 5-bp TSDs upon their insertions into the genome. The identified TSDs and TIRs indicate that Transib1–4 elements are DNA transposons. Moreover, the Transib1–4 consensus sequences encode hypothetical proteins, 461-aa Transib1p, 690-aa Transib2p, 685-aa Transib3p, and 533-aa Transib4p, respectively, which are 32–46% identical to each other (Table 4, which is published as supporting information on the PNAS web site). Such a high identity between the proteins, encoded by the Transib1–4 DNA sequences that are only ≈60% identical to each other, indicates that Transib1–4p proteins were necessary for the Transibs transposition. Because there are no other proteins encoded by these elements, we consider Transib1–4p as predicted transposases. Based on a TBLASTN search, there are about 100 elements in the DMG that code for proteins similar to Transib1p-4p. Autonomous DNA transposons, i.e., elements that encode transposases, usually constitute a minor fraction of DNA transposons fixed in eukaryotic genomes. The major fraction is composed of nonautonomous elements that do not encode the transposase but share common termini with autonomous elements from which they have descended. The same pattern is seen for Transibs (Fig. 1).

We noted a distant ≈60% identity between Transib1 and Hopper, a 1,435-bp nonautonomous DNA transposon identified as a de novo insertion associated with the Sxl^M4 allele (21). Hopper has 33-bp TIRs similar to these in Transib, and its copies are also flanked by 5-bp TSDs (21) (Fig. 6). Most likely, Hopper is the youngest member of the Transib clade. Hopper elements identified in the sequenced genome do not encode proteins. Presumably, the autonomous Hopper encoding a Transib-like transposase is not fixed in the DMG.

Currently, Anopheles gambiae (AG) is the only other species whose sequenced portion of the genome (22) encodes proteins similar to the Transib transposases. Using WU BLAST, we extracted all DNA fragments encoding proteins similar to the Transib transposases from the public set of assembled AG scaffolds. These fragments were split into several homogenous clusters. On the basis of the multiple alignment of sequences from the most abundant cluster, we reconstructed their 4,303-bp consensus sequence. This consensus sequence is a “recovered” mosquito Transib-like DNA transposon, called Transib1_AG, which encodes a 708-aa transposase (35% identity to Transib2p) and has 9-bp TIRs similar to those in the fly Transib elements (Fig. 7, which is published as supporting information on the PNAS web site). Moreover, Transib1_AG copies also are flanked by the 5-bp TSDs.

Given the multiple copies, TIRs, TSDs, and insertions of some Transib copies into other TEs, there is no doubt that Transib-like elements are molecular fossils of recently active DNA transposons. As there is no significant similarity of the Transib transposases to any known proteins (PSI-BLAST, BLASTP, E < 1), it is certain that Transib1p–Transib4p and Transib1_AGp belong to a previously unrecognized class of DNA transposases, catalysts of Transib “cut and paste” transpositions. The alignment of Transib1–4p, Transib1_AGp, and Transib2_AGp revealed seven conserved motifs, presumably most important for the transposition (Fig. 2). A putative catalytic D(34)D(35)E signal found in many cut-and-paste transposases is also present in the fly Transib transposases (Fig. 2). However, this signal is not present in the mosquito transposases.

Given the unique conserved transposase, the 5-bp TSD, the specific TIRs, and their presence in the fly and mosquito genomes, we conclude that Transib-like TEs constitute an eighth superfamily of cut-and-paste DNA transposons (Table 1).

Interestingly, the Transib codon composition is atypical for the DMG. As a result, commonly applied programs, including those used for systematic annotation of the DMG (18), fail to predict Transib1p-4p unless one presumes that the fly genome codes for genes characterized by human-, plant-, or worm-specific features, including their codon composition. Presumably, the atypical codon composition is related to the horizontal transfer of Transibs.

ProtoP. The ≈1100-bp Hoppel (23) or “element 1360” (24) is one of the most abundant TEs in the DMG (25). On the basis of its ≈30-bp TIR, this element was classified as a putative DNA transposon (23, 24). However, its relationship to known superfamilies of DNA transposons was obscure. We found that an ≈2-kb internal portion of one unusually long Hoppel element (GenBank gi 5656725, 634-4052) encoded a protein distantly related to the P transposase (BLASTX, 25–29% similarities to known P transposases, E < 10^–20, two stop codons). Surprisingly, Hoppel was the only TE linked to the identified internal portion (Fig. 3). Moreover, the internal sequence does not have any hallmarks of a TE inserted accidentally into a Hoppel element. Therefore, we concluded that the identified long Hoppel element was a fossilized autonomous DNA transposon that gave birth to the nonautonomous Hoppel/1360 elements. Given its antiquity (described below) and the similarity to all known P transposases, we called this transposon ProtoP. The 1,105-bp Hoppel consensus sequence is ≈97% identical to Hoppel elements, indicating that this family is ≈2 Myr old. ProtoP_B is a second family of nonautonomous ProtoP-like transposons described here. ProtoP_B copies are ≈95% identical to the 1,153-bp ProtoP_B consensus sequence (≈3 Myr old). There is only a 5% divergence between the Hoppel and ProtoP_B consensus sequences that contain 191-bp and 136-bp internal regions, respectively, ≈95% identical to different portions of ProtoP (Fig. 3). The 4,480-bp consensus sequence of ProtoP was reconstructed on the basis of a multiple alignment of its 28 copies, which are ≈95% identical to the consensus (Fig. 3). Hoppel and ProtoP_B elements were excluded from the last set of sequences. The ProtoP consensus sequence is 97% and 96% identical to the Hoppel and ProtoP_B consensus sequences, respectively. Therefore, the peak of ProtoP, Hoppel, and ProtoP_B transpositional activity occurred 3–5 Myr ago.

The last estimate is supported by studies of TEs inserted into ProtoP. For example, a ProtoP element identified in the AE002983 GenBank sequence (positions 8879–1919) harbors a copy of the Gypsy6A endogenous retrovirus (unpublished data; RU), including its flanking LTRs (positions 2325–8774). The 14% divergence between these LTRs corresponds to ≈5 Myr elapsed since the insertion of the retrovirus into the ProtoP element. A ProtoP_B copy (AE002743, positions 1–7347) harbors the Invader5 retrovirus (unpublished data; RU). The 15% divergence between its LTRs also indicates that ProtoP transposons colonized the DMG ≈5 Myr ago. Therefore, P-like elements, which were thought to be recent invaders of the DMG (26), colonized this genome several Myr ago. The ProtoP consensus sequence encodes an 864-aa ProtoP1 transposase (positions 1194–3788). Phylogenetic analysis (Fig. 4) shows that the ProtoP transposase separates a cluster formed by insect P elements and a putative human gene derived from a P-like transposase (27). Computer-assisted analysis of ≈30 ProtoP, Hoppel, and ProtoP_B copies and their target sites shows that ProtoP-like elements are flanked by 7-bp TSDs (not shown). However, all studied P elements, including the P1_AG, P2_AG, and P3_AG families present in AG (ref. 22; unpublished data; RU) are characterized by 8-bp TSDs. Usually, the TSD size is one of a few hallmarks of different superfamilies of eukaryotic DNA transposons. For example, 8-bp TSD is a characteristic of hAT-like transposons; 4-bp, piggyBac; 5-bp, Transib; 9-bp, MuDr; 3-bp, En/Spm and Harbinger. mariner/Tc1 is the only superfamily whose members are characterized by either 2-bp or 3-bp TSDs. Presumably, the phylogenetic place of the ProtoP transposase as an “outlier” among other P transposases (Fig. 4) may be related to the TSD size. While this paper was in review, we became aware of a similar report (28) describing a 3,410-bp autonomous Hoppel transposon (nearly identical to ProtoP deposited in RU in 1999). The authors suggested that the intronless P-like transposase is a result of retrotransposition. However, they did not show any evidence supporting this hypothesis. Alternatively, gain or loss of introns is relatively common and may be independent of retrotransposition. For example, the mosquito Helitron transposases are intronless (described below), whereas similar transposases in plants and worms are encoded by multiple exons (7). Similarly, accidental intron gains were reported for plant mariners (29).

Helitrons in Insects. A TBLASTN search for fly proteins similar to those encoded by the plant and worm Helitron rolling-circle DNA transposons (7) did not reveal any evident matches. However, using an approach similar to that described in ref. 7, we identified in AG the insect Helitron transposon, called Helitron1_AG. Its 8,200-bp consensus sequence encodes an intronless 1,700-aa Helitron1_AGp protein composed of the canonical Helitron-like Rep and SF1 helicase domains. Analogously to their plant and worm relatives (7), insect Helitron transposons are characterized by the 5′-TC and CTAG-3′ termini, the 3′-terminal ≈12-bp hairpin, and TSD-free insertions between the target A-3′ and 5′-T nucleotides. The Helitron1_AG consensus sequence is only ≈5% divergent from several Helitron1_AG copies, indicating that these elements were transposed in AG during the last few million years. The AG genome harbors more than 100 copies of different Helitron transposons that belong to more than 10 different families (not shown), characterized by a high (>90%) intra- and low (<70%) interfamily nucleotide identities. Additionally to the Helitron1_AG, we also reconstructed the consensus sequence of a Helitron2_AG family. Surprisingly, we found that a 649-bp portion of the Helitron2_AG (positions 972-1620) was 75% identical to a ≈600-bp fly sequence (AE002840, 9827–9264) (Fig. 7). Because this Helitron-like element is flanked by other DM-specific TEs, sequencing-related artifacts can be discarded. The identified element encodes a portion of the Rep/Helicase protein, and its coding sequence is interrupted by a few stop codons. Therefore, this element is a remnant of a Helitron. Ancestral lineages of DM and AG are thought to have separated ≈250 Myr ago (30). The neutral substitution rate in AG is even higher than in DM (31). Therefore, transposable elements that were present in a lineage ancestral to both AG and DM would not be recognizable anymore. For example, exons from ≈6,000 orthologous genes identified in the mosquito and fly genomes show only a ≈56% nucleotide identity (32). The 75% identity between the mosquito and fly Helitrons indicates that these elements were transferred horizontally rather than transmitted vertically.

piggyBac. The 1,881-bp consensus sequence of TE called Looper1_DM, reconstructed from five copies, encodes a 358-aa protein, which is more similar to the piggyBac-like transposase from the human Looper (7) than to the transposase encoded by the moth piggyBac (33, 34). DM was the third species where piggyBac DNA transposons were identified. Recently, piggyBac TEs were also identified in mosquito (22) and fishes (35, 36). These piggyBacs are characterized by ≈15-bp TIRs, including 5′-CCC and GGG-3′, and the TTAA TSDs. Using PSI-BLAST,we found that two hypothetical genes annotated in the DMG (gi 7303440, 21355395) are encoding proteins similar to the piggyBac transposase (PSI-BLAST, E < 10^–3). However, because the biological function of these genes is unknown, it is not clear whether they are “host genes” derived from the transposases or remnants of young piggyBacs. Analogously, several hypothetical DM genes are former Harbinger transposases (ref. 6; RU).

DNAREP1_DM. On the basis of a multiple alignment of ≈100 expanded copies of the 300-bp ARS320 repetitive element identified in fly (GenBank J01068), we reconstructed a 600-bp consensus sequence. The ARS320 element matches an internal portion of a 600-bp repetitive element called DNAREP1_DM (positions 100–400). Some DNAREP1_DM copies are inserted into other TEs (not shown). Therefore, DNAREP1_DM is a transposable element. It is the most abundant TE in the DMG. Several thousand copies of DNAREP1_DM are present in the sequenced genome. They are ≈85% identical to the consensus sequence. Therefore, DNAREP1_DM is an old family of TEs, transposed mainly ≈10 Myr ago. Currently, there are no young subfamilies of DNAREP1_DM, i.e., with copies at least 95% identical to each other. It indicates that DNAREP1_DM elements lost their mobility >3 Myr ago. Surprisingly, members of the DNAREP1_DM family do not reveal any hallmarks of DNA transposons and retrotransposons (TIRs, TSDs, 3′-microsatellites, etc). In 1999, we characterized DNAREP1_DM as a putative DNA transposon (unpublished work; RU) because of its conserved termini and multiple internal deletions. Based on the most recent data, another mechanism of DNAREP1_DM transpositions can be suggested. We found that a ≈200-bp 3′ untranslated region (UTR) portion of Penelope, an active Drosophila virilis retrotransposon (37), is ≈72% identical to the DNAREP1_DM consensus sequence (Fig. 8, which is published as supporting information on the PNAS web site). Moreover, the 3′ UTR of Penelope is required for its successful transposition (38). Therefore, we consider DNAREP1_DM as a remnant of Penelope-like retrotransposon that was highly active in the ancestral lineage of DM ≈10 Myr ago. We consider Penelope as a non-LTR retrotransposon. Alternatively, Penelope is also considered as a separate class of retroelements (38). Given that the DNAREP1_DM consensus sequence cannot be expanded, it is unlikely that it is a solo LTR.

Non-LTR Retrotransposons. Members of seven different clades of non-LTR retrotransposons populate the fly genome (Table 2). Remarkably, of all sequenced genomes, no genome contains elements from as many different clades of non-LTR retrotransposons as the DMG does. Among the seven different clades, Jockey is the most diverse and abundant in the DMG. We identified 15 previously undescribed families, bringing the total number to 25 known families (Table 2). Like other TEs, families of non-LTR retrotransposons that belong to the same clade are characterized by <75% inter- and >85% intrafamily DNA identities.

LTR Retrotransposons. We identified and characterized 28 families of LTR retrotransposons in DM (Table 3). All fly LTR retrotransposons belong to Gypsy, Copia, or Bel superfamilies (39), of which Gypsy is the most diverse and abundant. Surprisingly, the DMG does not contain nonautonomous LTR retrotransposons, present in mammals (40) and plants (6). On the basis of sequence identities and structural commonalities, Gypsy can by divided further into four major groups (Table 3), called Gypsy, MDG1, MDG3, and Osvaldo. The best-studied Gypsy group is composed of canonical Gypsy-like families, including Gypsy, 297, 17.6, and Zam. The Osvaldo group includes retrotransposons similar to Osvaldo and Ulysses, discovered in Drosophila buzzatii (41) and D. virilis (42). Gypsy12 is an Osvaldo family we found in DM. The ≈2,300-bp Gypsy12 LTR is the longest LTR identified in DM. Some members of the Gypsy and Osvaldo groups encode env, the surface envelope proteins, characteristic for animal retroviruses. Independently, the Gypsy superfamily was split into four groups (43), on the basis of the phylogenetic study of reverse transcriptase, RNase H, and integrase domains, and by us (unpublished work; RU 1999) on the basis of the structural commonalities (Table 3). In RU, the Accord, Tabor, and Invader groups were introduced instead of Gypsy, MDG1, and MDG3, respectively (43). Members of the Bel superfamily (44) encode one long polyprotein composed of gag, protease, reverse transcriptase, RNase H, and endonuclease domains. This superfamily was also called BCDRP (after Bel–Catch–Diver–Roo–Pao) and included Diver in fly and Catch in fugu (unpublished work; RU 1999).

Abundance of TEs. TEs account for 6% and 60% of the sequenced DM euchromatin and heterochromatin regions, respectively. Given that heterochromatin constitutes ≈30% of the DMG (5), TEs make up ≈22% of the whole genome. Our estimates are three times higher than those reported recently (25). The difference can be attributed to the different sets of TEs used as query sequences. Whereas the previous report (25) included TEs from 39 families, we used elements representing ≈150 families. Also, CENSOR and WU BLAST, used in our studies, are more sensitive than BLAST used previously (25). As a result, we identified numerous families of TEs that are only 60% identical to known elements and are represented by only a few truncated copies. There is a strong increase in the TE density on the acrocentric chromosomes 2 (Fig. 5) and 3 (not shown) in ≈500-kb regions separating euchromatin and paracentromeric heterochromatin, which is mostly not sequenced. This chromosomal distribution of TEs is very similar to that observed in AT (6). Given the fact that ≈80% of the paracentromeric/centromeric heterochromatin in DMG has not yet been sequenced, the percentage of TEs may be even higher.