Sequencing of subclone pAMH3C [GenBank: U96295] initially led to the characterization of an env-like gene and the 3' LTR of the SIRE1 endogenous retrovirus belonging to the Ty1-copia group of retroelements 11. The DNA adjacent to the SIRE1 LTR constituted an initially unidentified 544 bp ORF that gave no hits in BLASTn or BLASTx searches. However, when the sequence of the adjacent subclones, pAMH3G and pAMH3D, were addended to pAMH3C and assembled into a contig [Genbank: AF095730], a single, 1383-codon ORF with a nonsense mutation at position 2604, and frameshifts at 2813 and 3139 was generated (Laten and Das, unpublished). The frameshifts occurred in runs of six thymidines and five adenosines, respectively. When the frameshifts were adjusted and the conceptual translation was used to query the Genbank protein database, numerous high scoring hits to retrotransposon reverse transcriptases and integrases were obtained (Laten and Das, unpublished).

The large collection of sequences of reverse transcriptases and integrases that were retrieved, most as contiguous polyproteins, all belonged to the Ty3-gypsy group of LTR retrotransposons. While the BLASTp search identified AF095730 homology to numerous accessions from residue 135 to the carboxyl terminal, no sequences with similarity to the first 134 amino acids were found. The highest scoring hits were identified as Athila-like, related to the Ty3-gypsy group element from A. thaliana 18 that has subsequently been shown to be present in a wide range of plant genomes, including soybean, other dicots, and monocots 10. We have named the new soybean element family Diaspora.

Because the Diaspora DNA in the original genomic clone was truncated at both ends, we initially probed the λFIXII genomic library for additional Diaspora copies. Hybridization detected a few thousand positive plaques, confirming the moderately high copy number of this family. DNAs from a random sample of twenty positive clones were amplified using primers derived from the ends of AF095730 (PDIA01F/PDIA02R and PDIA03F/PDIA04R). No clone produced amplicons with both primer pairs, suggesting that all copies of Diaspora in these clones were either 5'- and/or 3'- truncated, or polymorphic at the primer sites (data not shown). We inferred that the library would not readily yield full-length elements. Retrospectively, this finding would have been anticipated had the unusual length of Diaspora, approaching that of the average insert size in λFIXII, been known (see below). With the availability of BAC soybean genomic libraries with inserts in excess of 100 kb 1920, the possibility of isolating full-length Diaspora copies became a virtual certainty and the λ clones were abandoned in favor of BACs. Filters containing microarrays of BAC clones derived from G. max cv. Forest 20 were probed for the presence of Diaspora. Hundreds of clones hybridized to a prot-rt probe (pAMH3D) and based on the number of hybridizing clones in the library, we estimated that Diaspora represents at least 0.5% of the G. max genome.

DNA was recovered from twenty, randomly chosen BAC clones that hybridized to pAMH3D, and PCR-amplified using the primer pairs derived from the ends of AF095730 (PDIA01F/PDIA02R and PDIA03F/PDIA04R). Surprisingly, only five clones were amplified by both primer pairs, suggesting that many copies were either 5' or 3' truncated or markedly polymorphic. Truncation would be consistent with the characterization of disrupted and nested retrotransposons first reported in maize 21. Of those containing both termini, none successfully served as templates for more than two additional PCR-amplifications using the complete set of AF095730-based primer pairs (PDIA5F through 13R). These findings suggested that Diaspora is a relatively heterogeneous family. This was confirmed by limited sequencing of λFIXII [GenBank: AF095730 and AY656632-AY656653] and BAC [GenBank: AY656654-AY656662] clones using the primers listed in Table 1. A total of 15,433 nucleotides were sequenced, of which 7293 were non-overlapping. It appeared that sequencing individual members of the Diaspora family directly from genomic clones would not lead to satisfactory descriptions of functional coding or regulatory regions, so an in silico strategy for these characteristics was pursued.

The length of the Diaspora consensus is 11,737 bp (Fig. 2), far longer than all but a handful of retrotransposons. The exceptional length of Diaspora is due primarily to the unusual length of its LTRs and the long gap between the upstream LTR and the gag start codon (Fig. 2). Like nearly all other retroelements, the LTRs terminate in TG...CA. The element is characterized by a contiguous 1892-codon ORF whose conceptual translation yields a single gag-pol polyprotein (Fig. 3) characteristic of Ty3-gypsy-like retrotransposons 23. Not surprisingly, the consensus contains neither nonsense codons nor frameshifts. This translated ORF possesses core domains for gag (CDD17379), reverse transcriptase (CDD16610) and integrase (CDD25582). There is also a CX₂CX₄HX₄C zinc finger motif in gag and a conserved protease catalytic domain motif, AMLDLGAS (Fig. 3). Interestingly, the first thirty amino acids of the translated ORF are not similar to the amino termini of any gag proteins in Genbank. Similarities to several gag proteins begin at position 31. Translation of the other five reading frames yielded no lengthy ORFs nor any similarities to any sequences in BLASTP searches.

As in the Calypso group 10, the tRNA primer binding site (PBS) begins 5 bp beyond the 3' end of the LTR and is perfectly complementary to the 3' terminal 18 bases of tRNA^Asp from Glycine max 24 (Fig. 4). At 873 bp, the distance between the LTR and the putative gag start codon is unusually long and not shared by related elements. This region contains no extended ORFs and neither BLASTn nor tBLASTn searches of the nr database retrieved significant hits.

Five potential splice donor sites, all in the LTR between 1400 and 2200 bp upstream of the gag-pol ORF, were predicted with medium confidence and two potential acceptor sites flanked the start codon. Although without splicing, the 5'UTR of any Diaspora transcript would be exceptionally long, the biological relevance of these sites is not known, and there are no reported examples of introns upstream of gag for any retrovirus or LTR retrotransposon.

The pol stop codon is 128 bp upstream of the polypurine tract (PPT) that abuts the 3' LTR. Thus Diaspora contains no envelope-like coding sequence beyond pol unlike those reported for its closest relatives, including members of the Athila and Calypso families (Wright and Voytas, 2002), BAGY-2 (Vicient et al., 2001), and Cyclops-1 81013. When this short region was used in BLASTn and tBLASTx searches, no additional sequences with significant probabilities were recovered. Interestingly, translation of this short region yields a strongly predicted transmembrane domain, although it is interrupted by two stop codons (data not shown).

The Diaspora LTR is 2524 bp in length (Fig. 2), making it one of the longest among retrotransposons and contributing to its unusual length. By comparison, the RIRE3 LTR is 2316 25, BARE-1 is 1829 26, Athila1-1 is 1539, and Cyclops-1 is 1504 8. Only the LTRs from Ogre and BAGY-1, at over 5,000 and 4200 bp, respectively 2728, are longer.

The length of the Diaspora LTR made it impossible to construct unique 5' or 3' LTRs by the in silico method employed. In addition, the absence of a contiguous element prohibited characterization of target site duplications. However, we identified eight accessions from the database that contained the tRNA PBS and part of the adjacent upstream LTR. The longest of these extended 491 bp upstream of the PBS. Twenty-two sequences contained the PPT and part of the adjacent downstream LTR. The longest of these extended 596 bp into the LTR. Thus, the central 1437 bp could not be uniquely assigned to either LTR. As a consequence, the available LTR sequences were merged to generate a single, consensus LTR that was affixed to both Diaspora ends.

Thirteen LTR sequences were 5' junctions and sixteen were 3', based on the complete absence of sequence similarity beyond the 5' or 3' ends, respectively, of the aligned LTR sequences. When the flanking DNAs of these 29 sequences, were used in BLASTn searches to query the GSS database, all but three generated dozens of hits (data not shown), and thus constituted repetitive elements themselves. Of the repetitive flanking DNAs, 75% represented Diaspora insertions into the coding regions of other retrotransposons. In addition, insertions into the coding regions of transposons related to En/Spm and Tam3 were also found. The identity of the three low- or single-copy sequences could not be ascertained. The Diaspora family therefore appears to be embedded in retrotransposon and transposon-rich regions. We have made similar observations for the SIRE1 retroelement (unpublished). Searches focused on the region upstream of the PPT failed to uncover any Diaspora copies with additional DNA between pol and the PPT.

Among the sequences used to assemble the contig, non-coding regions contained a variety of short indels, especially in homonucleotide runs and dinucleotide repeats, presumably from replication slippage. The sequences in the GSS collection from which the consensus was built represented unedited submissions, and excluding single base indels that might have been the result of unedited miscalls, most of the indels in the ORF retained the correct reading frame. Eight accessions: BH023632, CG813336, CG820702, CG821179, CG822466, AY656639, AY656648, and AY656656, were chimeric and probably represented truncated copies. All but two of these sequences were within gag or the putative 5'UTR. Since chimeric sequences in GSS would invariably produce lower bit scores, they were generally excluded form the contig. Consequently, many of the slightly lower scoring sequences initially retrieved in our BLASTn search were also chimeric, but were not retained for the assembly and were not further characterized.

The conserved region of RT was translated and the region representing peptide domains 2 though 7 29 was used in a BLASTp search to retrieve closely related accessions from Genbank. All of the sequences retrieved were from higher plants and their distribution among species reflected, to a large extent, the current progress of genome sequencing projects. The sequences were aligned (see Additional file 1) and a neighbor joining tree was generated (Fig. 5) and was rooted to the RT from gypsy.

The tree resolves two major clades, designated A and B (Fig. 5). With respect to coding potential beyond pol, clade B members have none, and in all cases, the pol stop codon is closely followed by a PPT and the LTR. In contrast, with the exception of Diaspora, all members of clade A for which sequences downstream of pol are available contain a putative env-like pseudogene. The major structural difference between Diaspora and other members of this group is illustrated in Fig. 6. Clade A is further partitioned into sister clades AI and AII with 94% bootstrap support. Clade AI is further divided, with 100% bootstrap support, into AIa/b and AIc. The bifurcation of AIa and AIb in Fig. 5 is only weakly supported (40%) and may not be significant. Clades AIa and AIb are populated exclusively with env-containing members, including the Athila and Calypso families. The only full length members of AIc are Diaspora, DiasporaLc, and Tpb1-1. The other members of the AIc lineage are sequences from PCR-amplified rt fragments from sycamore and cotton 10, and three representative rt-containing genomic clones from Fritillaria. DNA downstream of the Fritillaria rt has not been characterized (C. Baysdorfer, personal communication). In contrast to Diaspora in G. max and its close relative in L. corniculatus, 1479 bp separate the pol stop codon from the LTR in Tpb1-1. While this region contains no identifiable or extended ORFs, there is a proximal 19-codon ORF whose conceptual translation is predicted with very high confidence to be a transmembrane domain (data not shown). There is also a 47-base polyA segment in the middle of this region, suggesting the interval contains an integrated cDNA. Thus, the AIIc lineage is not monophyletic for the absence of a long pol-LTR interval, but whether this region in Tpb1-1 represents a degenerate env cannot be determined.

In conclusion, the nesting of clade AIc within a much larger group of elements with an env-like gene suggests that at least Diaspora suffered a complete and nearly precise loss of a coding region, rather than failed to acquire one. To more exhaustively search for other members of this group, the RT from members of lineage AIc were used in tBLASTn searches. All hits, however, were already in the tree.

DNA containing the SIRE1 endogenous retrovirus was recovered from a λFIXII soybean genomic library (Stratagene) by standard plaque hybridization 43, and HindIII-digested fragments were sub-cloned into pSPORT1 (Life Technologies) as described 11. DNA from three contiguous subclones, pAMH3C, pAMH3G, and pAMH3D were isolated and sequenced as described 11. The junctions and contiguity of these subclones were confirmed by direct sequencing of the intact λFIXII genomic clone across the HindIII junctions. These sequences were previously deposited [Genbank: U96295 and AF095730]. A BLASTp search with the conceptual translation of AF095730 (see below) indicated that this accession contained the pol region of an uncharacterized retrotransposon. Several additional positive clones from this library were recovered and segments of the isolated DNAs were sequenced directly or amplified using Taq DNA Polymerase (Promega). For amplifications, reactions were preheated for 3 min. at 94°C, then 30 cycles were run at 94°C for 30 sec., 54°C for 30 sec., and 72°C for 1 to 2 min. Amplicons were spin column-purified (Qiagen) and sequenced as described 11. Sequences were deposited [GenBank: AY656632-AY656653].

pAMH3D, compromising the protease and RT coding domains, was used to probe a soybean BAC library 20 (generously provided by K. Meksem) under moderate stringency 43 for the presence of sequences related to AF095730. Ten clones were chosen arbitrarily for amplification and sequencing. DNAs from BAC clones were recovered using Procipitate (Ligochem) and selected regions were amplified using Taq DNA Polymerase (Promega). After preheating reactions for 3 min. at 94°C, 30 cycles were run at 94°C for 30 sec., 54°C for 30 sec., and 72°C for 1 min. Regions within pAMH3D were first PCR-amplified using primer pairs PDIA01F-02R, PDIA03F-04R, PDIA05F-06R, PDIA07F-08R, PDIA09F-10R, PDIA11F-12R, and PDIA13F-14R (see Table 1). The amplicons were purified on Qiagen spin columns and sequenced directly without cloning as described 11. Sequences for regions beyond the ends of AF095730 were generated directly from BAC DNA using outward facing primers (PDIA02R and 03F) derived from the termini of AF095730, followed by additional outward extensions with primers PDIA16F, 17R, 18R, 19R, 20R, 21F (Table 1). BAC clone sequences have been deposited [GenBank: AY656654-AY656662].

Selected regions of AF095730 and their conceptual translations were used to query all relevant Genbank databases, including nr, Genome Sequence Survey (GSS) and Expressed Sequence Tag (EST), with BLASTn and BLASTp searches 4445. Conserved protein domains were identified with CDD 46.

For assembly of the consensus nucleotide sequence, Glycine max accessions from the GSS and EST databases with bit scores greater than 200 (E values < 10^-52) were added to the consensus construct. These criteria generally reflect >90% DNA sequence identity over at least 200 bp of overlap. The nr nucleotide database contained no significant hits. New additions to the ends of the expanding consensus were used to re-query the databases until the LTR redundancy was recognized and the contig formed a circle. Contigs were assembled using the Seqman program from Lasergene 5 (DNAStar). To locate the LTR, direct repeats greater than 25 bp were first identified using Lasergene GeneQuest (DNAStar) and the termini of the LTRs were confirmed by manual inspection, as were other non-coding features of the sequence. Because LTR junction sequences at both termini contained either internal element DNA or external flanking DNAs, these were carefully examined for consensus DNA (internal) or unique DNAs (external). External DNAs were trimmed from the contig. Potential splice junctions were evaluated using GeneSplicer 47 and NetGene2 48. Transmembrane domains were predicted using TMPred 49

The pol region of the conceptually translated consensus sequence was used to query the Genbank protein database for related sequences. A ClustalW alignment (see Additional file 1) was generated from a contiguous region of RT representing conserved domains two through seven 29 using Lasergene 5 (DNAStar), and a neighbor joining tree using p distances with 1000 bootstrap pseudo-replicates was constructed using MEGA2 50.