In an early study, field inversion gel analysis of the GiBV viral genome estimated the presence of 13 segments and a cumulative viral genome size of approximately 250 kbp 30. At the start of our project the viral genome characteristics of GfBV were unknown, but presumed to be similar to those of GiBV. We undertook a whole genome shotgun (WGS) sequencing approach to sequence the GiBV and GfBV viral genomes. Viral DNA was sequenced to approximately 8× coverage using purified virions pooled from the calyx fluid of approximately 400 and 50 female wasps from G. indiensis and G. flavicoxis, respectively (see Materials and methods). Following a manual effort to close sequence and physical gaps, we were able to derive a complete consensus viral segment sequence for 21 GiBV viral segments (8 of which were described previously 15); 4 segments (numbers 17, 19, 21 and 29) remained as partial sequences (Figure 1) due to technical difficulties primarily associated with sequence repeats. GiBV viral segment 25 corresponded to the GiBV genome 'segment F' previously sequenced and characterized in detail 3132. For GfBV, we were able to complete sequence analysis of 27 viral segments and 1 (number 13) remained a partial sequence (Figure 1). Nucleotide sequence polymorphisms occurred more frequently in GiBV than in GfBV, and we presume the higher sequence success rate for GfBV was due to sampling a more homogeneous viral population relative to GiBV. The discrepancy in GiBV viral genome statistics with earlier estimates 30 is due to finding multiple GiBV viral segments of similar size, which would have co-migrated on the agarose gel. The 25 GiBV and 28 GfBV viral segment sequences totaled 489 kbp and 581 kbp, respectively. These aggregate genome sizes and viral segment numbers have been further revised based on proviral sequence data (described below). Individual WGS sequence reads were deposited in the NCBI Trace Archive [1643848625-1643870960, 1813616562-1813617310]. Consensus sequence for viral segments, which ranged in length from 9.7 to 39.0 kbp for GiBV and 3.8 to 50.7 kbp for GfBV have been deposited in GenBank [GenBank:EU001243-EU001285].

We previously reported that a radioactive probe from total GiBV viral DNA hybridized at varying intensity to 127 clones from a bacterial artificial chromosome (BAC) library of 9,216 clones derived from G. indiensis larvae; 17 GiBV viral segment sequences for which probes were available mapped to BAC clones that segregated into 7 distinct non-overlapping groups 15. A similar size BAC library from G. flavicoxis yielded 154 clones that hybridized with total GfBV viral DNA. A combination of BAC end sequencing of 60 BACs and 10 GfBV segment-specific PCRs also segregated these BAC clones into 7 groups. GiBV and GfBV proviral segment sequences not present in the selected BAC clones were isolated by rescreening the arrayed BAC libraries with radioactive viral segment-specific PCR products. Probes for GiBV viral segment number 29 and 30 and GfBV viral segment number 29 and 31 did not hybridize to any BAC clones, while hybridizations of a probe for GiBV segment 28 proved to be false positives. The former may be due to representational bias in the BAC libraries that were screened.

A total of 11 BAC clones from G. indiensis (2 BAC clones were from a prior study 15) and 9 BAC clones from G. flavicoxis have been sequenced. Clone selection was based on a combination of proviral segment sequence composition, BAC fingerprinting, and BAC end sequencing to try to ensure that the BAC clones chosen for sequencing contained the greatest possible coverage of proviral segment sequences. The G. indiensis and G. flavicoxis BAC sequence data totaled 1.21 and 1.16 Mbp, respectively. The location of proviral segments within the BAC clones was determined by sequence alignment of viral segment sequences, as well as by searching for individual segment excision motifs (see below). Overlapping BAC sequences were fused, if possible within an inter-segmental region, to create pseudo-molecule sequences reducing unique sequence data for G. indiensis and G. flavicoxis to 1.08 Mbp and 1.00 Mbp, respectively. The length of contiguous sequences ranged in size from 48.8 to 233.6 kbp for G. indiensis and 91.7 to 279.3 kbp for G. flavicoxis. Sequences were annotated as described in Materials and methods. These data have been deposited in GenBank [GenBank:EF710644-EF710659; AC191960] was previously submitted 15.

Previously, using MEME 33, we discovered a highly conserved 6 bp DNA sequence embedded within an approximately 60 bp motif that appears to function as the site of excision and circularization of 8 GiBV proviral segment sequences 15. The analysis of such motifs was expanded to encompass all available viral and proviral sequence data for GiBV and GfBV. Viral, 5' proviral and 3' proviral excision motifs are shown in Figure 2. There are some differences, primarily outside the hexamer repeat, between the new consensus excision motifs for GiBV and those described previously 15. Additionally, the GfBV excision motif is highly similar to that of GiBV. As we demonstrated before for GiBV, differences between the extended sequence motifs at each end of a GfBV proviral segment indicate there is directionality to the mechanism of segment excision from the site of integration.

Using the extended motifs and MAST 34 we searched BAC sequences in order to detect the boundaries of proviral segment sequences for which we had partial viral segment sequence data, and we extended this search to all BACs. In addition to detecting some of the missing viral segment sequences, MAST predicted the presence of four novel proviral segments for GiBV (numbers 9, 12, 16 and 18) and one (number 8) for GfBV (Figure 1). All these potential proviral segments had exact sequence matches to unassembled sequences in the viral WGS sequence data. Sequence coverage in all cases was below one-fold and viral segment sequence versions of the novel proviral segments could not be assembled from the WGS sequence data alone. To determine whether these sequences represented potential false positives, we searched all inter-segmental and flanking non-segmental regions against the unassembled viral WGS sequence data and found no evidence for them in the WGS data. Thus, a combination of the viral and proviral sequence data predicts that GiBV and GfBV viral genomes each contain 29 segments, although 3 GiBV and 1 GfBV viral segments remain partially sequenced (Figure 1).

Based on GiBV and GfBV proviral segment location and conservation of gene order and sequence similarity within proviral segments, 27 of the 29 proviral segments were classified into pairs of segment homologs (Figures 1 and 3). Each member of a segment pair was assigned the same number. Segments 16 and 30 appear to be unique to GiBV, while segments 27 and 31 appear to be unique to GfBV (Figure 1). In addition to synteny between GiBV and GfBV proviral segments, there is clear evidence for conservation of gene order and identity of genes in wasp DNA flanking proviral segment sequences (Figure 3). These data depict the near entirety of the GiBV and GfBV proviral segments; loci encoding three GiBV and three GfBV proviral segments remain to be identified (Figure 1).

The largest region encoding proviral DNA in both species is represented by two linked proviral loci, 1 and 2. GiBV proviral locus 1 15 and GfBV locus 1 consist of 8 proviral segments each, while GiBV and GfBV proviral locus 2 consists of 12 and 13 proviral segments, respectively (Table 1). Although the BAC clones we sequenced did not overlap at these loci, consideration of synteny suggests that they are linked (Figure 3). Synthetic oligonucleotide primers were designed near the ends of GiBV proviral segment 17 and 18 to close the existing gap between BAC clones within GiBV locus 2. However, PCR analysis of pupal stage G. indiensis parasitoid DNA failed, possibly due the large size of the estimated gap of approximately 20 kbp. Additional primers were designed based on an inter-segmental parasitoid sodium:neurotransmitter symporter gene present in the homologous GfBV region (Figure 3), which were then used in appropriate combination with GiBV proviral segment 17 and 18 primers. An amplicon of approximately 3 kbp (symporter to GiBV proviral segment 18) was obtained and end-sequenced (data not shown), verifying the predicted linkage. We were not able to link the large gap from GiBV proviral segment 17 to the symporter, but the data strongly support linkage of GiBV proviral segment 17 and 18 as occurs in GfBV. Synthetic oligonucleotides designed to either side of the gap separating GiBV proviral locus 1 amplified an approximately 5 kbp product and end-sequence data verified physical linkage of the two loci (data not shown). The strong syntenic relationships between GiBV and GfBV proviral genomes argue for the correct assembly and representation of GiBV proviral locus 2 and its linkage to proviral locus 1 (Figure 3).

The remaining GiBV and GfBV proviral loci (Table 1, Figure 3) contain either a single proviral segment (loci 4-6) or two proviral segments in tandem array (loci 3 and 7). Locus 6 and locus 7 sequences are only available for GiBV and GfBV, respectively, but must exist for both genomes based on viral segment sequence data (Figure 1). BAC clones containing loci 3-7 do not overlap with each other or with those encoding loci 1 and 2. The genomic context surrounding three proviral segment sequences from each virus remains unknown, and they could occur as single or tandemly linked sequences. Unless these proviral segments are linked to the loci we have already defined (in which case there would be seven loci), our current data indicate that GiBV and GfBV proviral segment sequences will occupy eight or nine loci. It remains possible that sequence analysis of the missing proviral loci may identify additional viral segment sequences.

We previously described two different long (6-7 kbp) tandem repeats flanking GiBV proviral locus 1 (named L1R1 and L1R2) 15. These repeats were not found to universally flank the remaining GiBV proviral loci or to flank GfBV proviral locus 1. A long (7.5-8 kbp) tandem repeat similar to L1R1 was found in DNA adjacent to GiBV proviral loci 4 and 5 (Figure 3, A repeats) but these repeats were not found in the homologous GfBV proviral loci. An additional smaller (1 kbp) repeat similar to L1R1 was found on the 3' side of GiBV proviral locus 2. The L1R2 repeat was not found at any other GiBV proviral loci, although L1R2-like repeats (approximately 1.5 kbp) were found near GfBV proviral loci 4 and 7 (Figure 3, B repeats). GiBV proviral locus 6 does have a long (approximately 6 kbp) tandem repeat on the 3' side (Figure 3, C repeat), although it does not share sequence similarity to either L1R1 or L2R2.

As found in GiBV proviral locus 1 15, inter-segmental regions separating tandemly arrayed proviral segments in both the GiBV and GfBV proviral genomes are generally small (<1 kbp) and do not code for proteins. One exception is the approximately 9 kbp region separating GfBV proviral segments 17 and 18, which codes for a parasitoid sodium:neurotransmitter symporter. As described above, a similar host gene is likely to exist in the homologous region in G. indiensis.

A summary of the major characteristics of the viral genomes of GiBV and GfBV is given in Table 2, and the segment sequences are visually depicted in Figure 1. These data are based on the combined sequences derived from BV viral and proviral segments, which predicts that both BV viral genomes contain 29 segments, with an aggregated genome size of approximately 503 and 594 kbp for GiBV and GfBV, respectively. GfBV viral segments are, on average, 12% larger than their homologous segment in GiBV, which appears to be due to GfBV-specific tandem duplications of gene clusters. Three GiBV and one GfBV viral segment sequence could not be finished with either viral WGS or BAC sequence data, and assuming a 12% size difference between GiBV and GfBV segment homologs, we estimate that we are missing approximately 14 kbp of GiBV and <1 kbp GfBV viral genome sequence data. We thus estimate a cumulative viral genome size to be 517 kbp for GiBV and 594 kbp for GfBV.

The GiBV and GfBV viral genomes are predicted to encode similar numbers of proteins, 197 versus 193, respectively, and 58% and 63%, respectively, of the genes are predicted to contain introns. Both genomes have a similar average G+C content (35-36%) and protein coding density (32-33%). The genomes contain many gene families found in other bracoviruses, including protein tyrosine phosphatases (PTPs), viral ankyrins, C-type lectins, and cystatins; features of these gene families are listed in Table 3. Encoded PTPs, ankyrins, and cystatins are mostly predicted to be single exon genes while C-type lectins are always predicted to be two-exon genes. PTPs and ankyrins are not predicted to encode signal peptides while C-type lectins and cystatins are often predicted to encode them. Additionally, as reported for CcBV and MdBV, the GfBV and GiBV genomes are predicted to encode a small number of tRNAs (Table 2).

A high degree of partitioning is seen between viral segment gene content when the segments are grouped by proviral locus membership (Table 1). The majority of genes in locus 1 are predicted to encode signal peptides, while few genes in any other loci are. Loci 1 and 2 both contain many proviral segments, most of which encode hypothetical proteins of unknown function and contain one intron. In contrast, the remaining loci contain only one or two proviral segments and predominantly encode PTPs and viral ankyrins that are not predicted to contain introns. A recent study of CiBV showed that genes on the same viral segment were generally expressed in the secondary host in the same time period 35. Since BV viral segments are present in virion DNA preparations in different abundances, the locus-specific partitioning of GiBV and GfBV proviral segments may represent some additional form of control over BV viral gene delivery.

Annotation of non-encapsidated DNA sequences flanking GiBV and GfBV proviral segment sequences predicted the presence of 78 and 71 genes, respectively. Nine genes that flank GiBV proviral locus 1 were described previously 15. Very few flanking genes showed similarity to known virus (including PDV) genes; those that did are discussed below. However, many had a high sequence similarity with insect genes (78% for flanking genes and 29% for segment genes; BLASTP, E < e-10). Additionally, flanking genes tended to have more exons per gene than proviral segment genes, although the difference was not statistically significant (2 ± 1 for segment genes and 3 ± 2 for flanking genes).

Proviral segment and flanking DNA are distinct not only at the gene level, but on a nucleotide composition level as well. Previously, we demonstrated that for GiBV proviral locus 1, proviral segment sequences had nucleotide compositions similar to each other and distinct from flanking sequences 15. We extended the analysis of trinucleotide frequencies to encompass all available GiBV and GfBV proviral segment, flanking, and inter-segmental sequence data. The results are shown in Figure 4. The majority of proviral segment and flanking sequences cluster into distinct groups, and the short terminal-branch lengths indicate highly similar composition within these groups. Inter-segmental regions also tend to cluster together, although with a higher degree of variation, indicated by the longer terminal-branch lengths. This was expected as the generally shorter lengths of inter-segmental sequences make calculating trinucleotide frequencies less accurate. These results were consistent across both the GiBV and GfBV sequence data.

As koinobiont endoparasitoids, PDV-carrying wasps develop as larvae within caterpillars that are still undergoing development. Therefore, they are in an arms race with the caterpillars and PDVs contribute to survival against attacks by the immune system of a living caterpillar. Given this situation, it is plausible to hypothesize that PDV genes on a whole must evolve rapidly, particularly since conserved genes that typically control viral replication and particle formation appear to be absent from the viral genome. We utilized sequence divergence between GiBV and GfBV genes to test this hypothesis. All genes described here were divided into two sets: one encompassing genes encoded by proviral segments, and one including flanking genes. Genes were considered orthologs if they had reciprocal best BLAST hits to each other and appeared in syntenic positions. Using these criteria, we identified 72 orthologous gene pairs in the proviral segment gene set and 41 in the flanking gene set.

In order to examine the strength and direction of selection acting on these genes, the ratio of non-synonymous to synonymous substitutions (dN/dS) was calculated and a histogram summarizing the results is shown in Figure 5. Detailed results for each gene pair are shown in Additional data file 1. Flanking wasp genes predominantly have dN/dS values near 0 (median = 0.23, average = 0.38), indicating that they are under strong negative selection, which is expected for genes involved in essential cellular processes. On the other hand, proviral segment genes have dN/dS values centered near 1 (median = 0.92, average = 0.96), suggesting that many of these genes are under neutral or positive selection. Analysis with the Mann-Whitney U test shows that, overall, segment genes are evolving at significantly different rates than flanking genes (p < 0.001).

An alternative explanation for the dN/dS distribution of proviral segment genes centered around 1 could be that we predicted a large number of pseudo-genes, which would be evolving neutrally; this would spike the dN/dS distribution near 1. To test this possibility and more accurately assess positive selection, we conducted a more powerful test of selection, the McDonald-Kreitman (M-K) test 36, by comparing divergence between the 72 orthologous gene pairs to polymorphisms within those genes in the GiBV viral shotgun sequence data. Results of the test (Table 4) show significant evidence for positive selection across GiBV genes as a whole (p = 3 × 10^-8), indicating that most of our gene models are not false predictions and that many GiBV genes are under positive selection. Criticism of the M-K test suggests that a spurious significant result can be produced by purifying selection acting on slightly deleterious mutations within a species rather than positive selection between species 37. However, it is highly unlikely that slightly deleterious mutations alone could account for the level of significance found in this analysis.

Closer examination of specific GiBV and GfBV gene families also revealed evidence of potential positive selection, particularly within PTPs, cystatins, and C-type lectins (Table 3). PTPs showed a highly significant pattern of positive selection (p < 0.0001) using the M-K test, despite having an average dN/dS below 1. Additionally, the C-type lectin and cystatin analyzed had dN/dS ratios above 1, suggesting that these genes are also under positive selection. While it is plausible that gene families that had uniformly low dN/dS values might be involved in core viral functions such as replication or encapsidation, the only gene family found with such a dN/dS distribution was the sugar transporters (average dN/dS = 0.23). As these genes are specific to GiBV and GfBV and not found in other sequenced BVs, it seems unlikely that they serve a core viral function.

In contrast to the high levels of positive selection detected within proviral segment protein coding sequences, we previously reported that, within GiBV proviral locus 1, non-coding segment DNA appeared to be under purifying selection 15. However, when the analysis was extended to include all sequence data, non-coding sites were evolving at comparable rates to synonymous sites (data not shown). This suggests that this conservation of non-coding DNA is either restricted to GiBV proviral locus 1 or that previous results were affected by limited sample size.

Little is known about the origins of PDV genes. Genes that have sequence similarity to known eukaryotic genes, such as those encoding PTPs, are thought to have eukaryotic origins (see 38 for review). However, these gene families have such diverse sequences within a single PDV genome that phylogenetic analyses have thus far been unable to determine an exact origin of these genes. Recently, in an analysis of genes from previously published BV genomes, Bezier et al. 39 found that no BV genes were more similar to their insect homologs than their vertebrate homologs, and phylogenetic analysis did not suggest an insect (or any other) origin for any BV genes.

Annotation of two pairs of GiBV and GfBV segment homologs (segments 17 and 18) revealed a novel family of 8 genes with a surprisingly complex intron-exon structure, consisting of 6-8 exons in each gene as opposed to an average of 2 ± 1 exons in other viral genes. The eight genes are predicted to encode major facilitator superfamily (MFS) transporter proteins, which in general transport a wide variety of small solutes across membranes 40. The best BLASTP hits for all eight proteins were matches to predicted insect sugar transporters, particularly one from the honey bee Apis mellifera and one from the parasitic wasp Nasonia vitripennis (approximately 45% amino acid identity to both). Such genes have not been described within PDV genomes sequenced to date.

Bayesian phylogenetic analysis using a GTR+gamma+I model was conducted on the 8 GiBV and GiBV sugar transporters and the 16 most similar transporters in insects. Three sugar transporters from vertebrates were chosen as outgroups, as no other arthropod sugar transporters were present in the GenBank protein database. The resulting phylogram is shown in Figure 6, with posterior probabilities given above branches. The GiBV and GfBV sugar transporters group strongly with an orthologous pair of sugar transporters from Nasonia and Apis, suggesting that these genes share a common hymenopteran ancestry. The sugar transporters did not group with that of the silk moth, Bombyx mori, suggesting that these genes were not acquired from the secondary host genome. The similarity of these genes to hymenopteran sugar transporters is likely not due to convergence, as if the genes were acquired from the secondary host genome, they would most likely be expressed in the caterpillar host, and therefore convergence should not drive the genes toward wasp sugar transporters. The low dN/dS values calculated for the sugar transporters (Table 3) suggest that they are under purifying selection. We hypothesize that this, coupled with a recent acquisition event, is why these genes still so closely resemble their ancestral wasp genes. It is unclear why the BV transporters would be under different selective pressures than other described virus gene families such as PTPs. Nevertheless, this finding provides support for the hypothesis that a large number of PDV genes have been acquired from their primary host, but have diverged to such an extent through positive selection that limited similarity remains with their ancestral wasp genes.

The GiBV proviral locus 2 sequence was generated by merging sequences from overlapping BAC clones. One BAC clone contained an 11,578 bp insertion in proviral segment 11 [GenBank:EU822800], which was not present in the other BAC clone in the region of overlap. The insertion sequence in BAC clone 1C20 is predicted to encode a single gene with sequence similarity to transposase and was bounded by a perfect 8 bp direct repeat with perfect 49 bp inverted repeats internal to the direct repeats, suggesting that the sequence is a TE. Analysis of the transposase using ProteinRepeatMask 41 and CENSOR 42 classified this transposase as Drosophila p-element-like. However, the 8 bp direct repeat (AAATTCCA) is different from that typical of p-elements (GGCCAGAC), and the TE is much larger, overall, than typical p-elements, which are only approximately 3 kbp in size. A single copy of the 8 bp direct repeat exists in BAC clone 2C5, which lacks the TE and most likely represents a point of replicative insertion for this class II transposon. The TE insertion does not disrupt a gene. Inclusion of the TE in the nucleotide composition analysis shows that the TE strongly groups with flanking sequences, rather than the segment sequence in which it is inserted (Figure 4).

The TE sequence is not depicted in the proviral locus 2 sequence presented here (Figure 3) as BAC clone 2C5 contained a complete sequence of proviral segment 11 while clone 1C20 did not. In addition, the consensus sequence derived for GiBV viral segment 11 did not contain the TE. A BLAST search of the TE against the unassembled GiBV viral WGS sequence reads revealed four matching reads. These reads matched internal parts of the TE and the junction between one of the TE ends and viral segment 11, suggesting the TE is found in some GiBV viral genomes at a low frequency. We hypothesize that this TE was recently acquired from the wasp genome, and has not yet been lost or gone to fixation in the GiBV genome. If the TE was highly deleterious to the BV genome, it would be unlikely to be present in the population. While this TE does not encode any genes other than a transposase, it does suggest that TEs can enter a BV genome and become packaged successfully. No evidence of this TE was found in either viral or proviral GfBV sequence data.

GfBV proviral locus 7 contains proviral segment 28 and a partial sequence of proviral segment 27, and is flanked on one side by approximately 95 kbp of non-segment DNA (Figures 3 and 7). The latter encodes some common TE elements (Gypsy retrotransposon and Mariner-like) and, remarkably, genes with sequence similarity to genes identified in the CcBV viral genome (Figure 7). Additionally, this region encodes a PTP gene, which is identical in nucleotide sequence to a PTP gene at the 3' end of GfBV proviral segment 28 (asterisks in Figure 7). MAST searches of this region using the GfBV proviral excision motifs did not reveal any potential excision motifs, and BLAST searches of the region against the unassembled GfBV viral sequence reads did not reveal any matching hits, suggesting that this region does not contain novel GfBV proviral segments.

The genes similar to those in CcBV include one that has strong sequence similarity to CcBV hypothetical protein 30.5 and a block of three genes that are homologous to, and in the same order as, CcBV segment 31 genes 31.1, 31.2, and 31.3 (Figure 7). The gene homologous to CcBV 31.1 contains a DNA Pol B2 domain, which in CcBV was annotated as a pseudogene due to a premature translational stop codon 1143. The second and third genes in the block contain an integrase core domain and a poxvirus-like DNA packaging domain, respectively. The CcBV genes were previously considered 'virus-like' and were hypothesized to represent ancestral virus genes within the BV genome 43.

However, ProteinRepeatMasker 41 and CENSOR 42 identified the block of three genes and the next two downstream genes (Figure 7) as components of a Maverick TE. Mavericks are a novel class of giant TEs 44, also known as Polintons 45. BLAST searches revealed homologs of all five genes to be common components of Maverick TEs present in the recently released genome sequence of the parasitic wasp Nasonia vitripennis. Nasonia belongs to a lineage of wasps separate from those carrying PDVs, suggesting that these proteins are not of BV origin. Based on these results, we hypothesize that the tandem array of three genes in CcBV segment 31 represents remnants of a Maverick TE that was transferred to the CcBV proviral genome and has since become fixed.

The Maverick-like array of proteins in the GfBV flanking DNA is missing two components common to all Maverick TEs: a coiled-coiled domain protein and terminal repeats. The complete absence of terminal repeats coupled with the maintenance of open reading frames is particularly unusual. Since TEs cannot function without terminal repeats, TE open reading frames typically degrade rapidly after their terminal repeats are lost. This suggests that selection on the wasp genome is maintaining these open reading frames, although their potential function is unknown. The abundance of both TEs and BV-like genes encoded in DNA flanking GfBV proviral locus 7 may indicate that this region is a 'hotspot' for the movement of DNA between proviral and flanking sequences. It will be of interest to determine if similar sequences are present in the homologous GiBV locus.

Outbred populations of G. indiensis and G. flavicoxis, solitary and gregarious endoparasitoids of gypsy moths (Lymantria dispar), respectively, were maintained at the USDA-ARS-Beneficial Insects Introduction Research Unit, Newark, Delaware, until they were moved to Beltsville, Maryland in 2007. The maintenance protocol and history of the G. indiensis colony was described previously 15. G. flavicoxis was reared under the same protocol with the exception that G. flavicoxis parasitizes late third instar gypsy moth larvae. Cocoons formed from parasitized hosts were stored at 24°C until adult parasitoid emergence and then separated by sex. For BAC library material, G. flavicoxis larvae were dissected from parasitized gypsy moth larvae 10 days post-parasitization, briefly rinsed in phosphate buffered saline, flash frozen in liquid nitrogen and stored frozen at -80°C.

Virions were purified from G. indiensis and G. flavicoxis females using established protocols 48. Briefly, female wasps were anaesthetized in 75% ethanol and rinsed in phosphate buffered saline. Ovaries were dissected from the females in a drop of phosphate buffered saline and ruptured, draining the calyx fluid. Pooled calyx fluid was subsequently filtered through a 0.45 μm filter to remove eggs and cellular debris 49. Viral DNA was extracted according to established protocols 30.

BAC libraries of G. indiensis and G. flavicoxis with a 120 kb average insert size were constructed by Amplicon Express (Pullman, WA, USA), using a partial BamHI digest inserted into an MboI site of a pECBAC1 vector. A nylon filter arrayed with 9,216 BAC clones was created from each library. In order to identify BAC clones containing proviral segment DNA, BV encapsidated viral DNA from each was radioactively labeled with ³²P-labeled α-dCTP (NEN/Perkin-Elmer, Waltham, MA, USA) using the Redi-prime II DNA labeling kit (Amersham Biosciences, Piscataway, NJ, USA). Labeled DNA was then purified using a QIAquick PCR purification kit (Qiagen, Valencia, CA, USA). The filter was pre-hybridized at 65°C for at least 3 hours with Rapid-hyb Buffer (Amersham Biosciences) and 500 μg of salmon testes DNA (denatured at 100°C; Sigma-Aldritch, St. Louis, MO, USA). The probe was added and allowed to hybridize overnight at 65°C. The filter was then washed 2 times for 60 minutes each at 65°C with a 0.1 × SSC/0.1% SDS solution, wrapped in plastic wrap, and autoradiographed using Kodac BioMax MS film.

Approximately 7.5 μg of BV encapsidated DNA was sheared and DNA fragments in the size range 3.5-4.5 kbp purified after separation by agarose gel electrophoresis. The fragments were blunt ended and, after addition of BstXI adaptors, cloned into the BstXI site of pHOS2. Shotgun libraries were similarly made for each BAC clone. Celera Assembler 50 and TIGR Assembler 51 were used to assemble random sequence data for BV and BAC clones. Gap closure was assisted by a closure editor tool called Cloe that also permits the manual inspection and editing of sequence data. A variety of methods were used to close gaps, including re-sequencing the ends of random clones, transposon assisted sequencing (GPS, New England Biolabs™, Ipswitch, MA, USA) or 'micro-library' construction of single or pooled templates, and conversion of physical gaps to sequence gaps using 'POMP' (pipette optimal multiplex PCR) 52 and or/a 'Genome Walker' kit (Invitrogen™).

Primers were developed specific to individual identified GiBV and GfBV viral segment sequences as described in 15. PCR was performed in a 10 μl solution that included 0.1 μl template DNA, 0.3 μl 50 mM MgCl₂, 1 μl 10× PCR buffer, 0.2 μl 10 mM dNTPs, 7.9 μl H₂O, 0.1 μl Platinum Taq (Invitrogen), 0.2 μl F primer (20 pm/μl), and 0.2 μl RC primer (20 pm/μl). The PCR protocol was 94° for 2 minutes; 35 cycles of 94° for 30 s, 58° for 30 s, 72° for 45 s; followed by 72° for 7 minutes. PCR products to be used for hybridizations were purified using a QIAquick PCR purification kit (Qiagen). Segment-specific hybridizations were done as described above for total viral DNA hybridizations.

Because individual sequence reads could not be associated with individual wasps, a conical consensus sequence was generated for each BV segment using the SliceTools package 53. At a given position in a conical consensus, all bases with a cumulative quality value within 50% of the highest cumulative quality value were assigned to that position.

A combination of SoftBerry's FGENESH 54 using the honey bee (A. mellifera) training set, and the Beijing Genome Institute's BGF 55 trained on the silkmoth (B. mori) were used for gene prediction, in addition to the AAT package 56, which allows spliced alignment of proteins to genomic DNA, thereby revealing potential exon-intron boundaries. Gene models from FGENESH were generally accepted except when multiple other sources of information contradicted those models. SignalP 5758 and tRNAScan-SE 59 were used to predict signal peptides and tRNAs, respectively. Transposable elements were annotated using ProteinRepeatMasker 41 and CENSOR 42.

Excision motifs were generated by cutting out a sequence extending 30 bp upstream and downstream from the GCT excision site at the 5' and 3' boundaries of proviral segments and at the GCT circularization site of viral segments for both GiBV and GfBV. No additional alignment was conducted on the sequences. All motifs were visualized using WebLogo 6061. Proviral excision motifs were also generated with these sequences using MEME 33, and the resulting motifs were used to search BAC sequences for potential additional proviral segments using MAST 34.

Jaccard orthologous gene clusters between GiBV and GfBV were calculated using Sybil 62. Syntenic blocks were defined as two or more adjacent orthologous gene clusters, and the results were visualized using Sybil 62. This information, in addition to conserved location (locus and position within that locus) of proviral segments, was used to define homologous segments between GiBV and GfBV.

For the phylogenetic analysis, GiBV and GfBV MFS transporter genes were searched against the GenBank non-redundant protein database using BLASTP. The top 18 hits from unique organisms were downloaded from GenBank (N. vitripennis, [GenBank:XP_001607065; XP_001602960]; A. mellifera, [GenBank:XP_001120868; XP_395522]; Tribolium castaneum, [GenBank:XP_973694; XP_973659], [GenBank:XP_966705; XP_966524]; Anopheles gambiae, [GenBank:XP_311836]; Aedes egypti, [GenBank:XP_001649205]; Drosophila melanogaster, [GenBank:NP_611451; NP_524479; CAA73031; XP_001361445]; Drosophila pseudoobscura, [GenBank:XP_001358762]; Mus musculus, [GenBank:NP_035525; CAC36405]; and Rattus norvegicus, [GenBank:NP_062103]). Additionally, the single GfBV MFS transporter gene was searched against B. mori expressed sequence tags in GenBank, and the single strong hit (E = e-61; [Genbank:BJ985900]) was downloaded from GenBank and translated. These sequences, in addition to the eight MFS transporters predicted for GiBV and GfBV, were aligned using ClustalW 63, and regions of ambiguous alignment were removed using Seaview 64, resulting in an alignment of 448 amino acids. The phylogenetic analysis was conducted using MrBayes 3.1.2 6566, sampling every 1,000 generations for 1 × 10⁷ generations. The first 50% of generations were discarded as burn-in, and posterior probabilities were calculated from the remaining 501 sampled generations.

For the nucleotide composition analysis, relative trinucleotide frequencies 67 were calculated for all segment, inter-segmental, and flanking sequences. A Euclidean distance matrix was then constructed from those frequencies. The sequences were then clustered using the neighbor-joining algorithm in PAUP* 68 and the resulting tree was visualized with Treeview 69.

All proteins described here for each genome were divided into two sets: those encoded by proviral segments and flanking genes. For each set, BLASTP in WU-BLAST 70 was used to search the GiBV (or Gi) proteins against the GfBV (or Gf) proteins and vice versa. Nucleotide sequences of reciprocal best hit pairs that appeared in syntenic regions (in the same segment for proviral genes and in the same region for flanking genes) were aligned using the cdna_fast_pair method in T-Coffee 71, and pairs with ambiguous or frameshifted alignments were removed. The codeml program in PAML 7273 was used to calculate dN/dS (Ka/Ks) for the remaining gene pair alignments. Pairs of proviral genes were further analyzed by using codeml in PAML 7273 to calculate the number of silent and replacement substitutions. The number of silent and replacement polymorphisms within the GiBV shotgun sequence data for these genes with at least 3× coverage were then calculated using previously described methods 15. The M-K test 36 was then utilized to test gene pairs for evidence of positive or negative selection.