Organization and Expression Strategy of the Ambisense Genome of Densonucleosis Virus of Galleria mellonella

Abstract

The expression strategy of parvoviruses of the Densovirus genus has as yet not been reported. Clones were obtained from the densonucleosis virus of Galleria mellonella (GmDNV) that yielded infectious virus upon transfection into LD652 cells. Its genome was found to be the longest (6,039 nucleotides [nt]), with the largest inverted terminal repeats (ITRs) (550 nt) among all parvoviruses. The distal 136 nt could be folded into hairpins with flop or flip sequence orientations. In contrast to vertebrate parvoviruses, the gene cassettes for the nonstructural (NS) and structural (VP) proteins were found on the 5′ halves of the opposite strands. The transcripts for both cassettes started 23 nt downstream of the ITRs. The TATA boxes, as well as all upstream promoter elements, were localized in the ITRs and, therefore, identical for the NS and VP transcripts. These transcripts overlapped for 60 nt at the 3′ ends (antisense RNAs) at 50 m.u. The NS cassette consisted of three genes of which NS2 was contained completely within NS1 but from a different reading frame. Most of the NS transcripts were spliced to remove the upstream NS3, allowing leaky scanning translation of NS1 and NS2, similar to the genes of RNA-6 of influenza B virus. NS3 could be translated from the unspliced transcript. The VP transcript was not spliced and generated four VPs by a leaky scanning mechanism. The 5′-untranslated region of the VP transcript was only 5 nt long. Despite the transcription and translation strategies being radically different from those of vertebrate parvoviruses, the capsid was found to have phospholipase A₂ activity, a feature thus far unique for parvoviruses.

Densonucleosis viruses (densoviruses or DNVs) are parvoviruses of invertebrates (30). Thus far, about 25 DNVs have been isolated from different species of insects belonging to at least six insect orders, shrimps, and possibly crabs (6, 13, 27). All characterized DNVs package complementary, linear single strands of the DNA into separate virions, as do several parvoviruses of vertebrates (e.g., B19 and AAV [11]), and replicate autonomously. DNVs have little sequence identity with the vertebrate parvoviruses (9), and, consequently, they have been classified as a separate subfamily, the Densovirinae, in the family of Parvoviridae (34). Further studies indicated that there are at least three distinct DNV groups, one genus (Densovirus), exemplified by Junonia coenia DNV (JcDNV) (12, 19), a second genus (Brevidensovirus) by the Aedes (1, 8) and shrimp (27) DNVs, and a third genus (Iteravirus) by the Casphalia DNV (14) and Bombyx mori DNV type 1 (22). Most DNVs remain unclassified, since few are characterized with respect to their molecular biology.

The DNV from the greater waxmoth, Galleria mellonella (GmDNV), was the first DNV to be isolated (25). Its particles yield four clearly distinct polypeptide bands upon sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) (between 90 and 45 kDa and relative amounts increasing in the order (from least to greatest) VP1, VP3, VP2, and VP4 [32]). These structural proteins have a strong identity upon peptide mapping, and this suggested a common gene (31). Restriction mapping confirmed a close relationship between GmDNV and JcDNV (19). Sequencing of the JcDNV genome demonstrated large open reading frames (ORFs) in the 5′ halves of opposite strands (12), but the expression strategy of this genus of viruses has, as yet, not been published.

The near-atomic structure of GmDNV was solved by X-ray crystallography (28). Since this is the only DNV thus characterized, the structure cannot be compared to that of other DNVs, but some striking differences with the structure of vertebrate parvoviruses were observed. Although the β-barrel of the capsid proteins is highly conserved among virus structures, the GmDNV capsid protein β-barrel must be rotated by 7.4 Å and translated radially inwards by 9.7 Å to superimpose it on the β-barrel of the capsid protein of canine parvovirus when the rotational symmetry axes are superimposed. Among other, surprising differences with the vertebrate parvoviruses was the absence of loop 4 in the GmDNV capsid protein, resulting in a β-annulus type structure instead of a spike around the threefold axis. In GmDNV, the βA strand is linearly extended across the twofold axis from the βB strand, allowing it to form hydrogen bonds with the βB strand of the neighboring subunit. In contrast, the βA strand folds back to its own subunit in vertebrate parvovirus structures thus far solved. The outside of the GmDNV capsid is much smoother than that of vertebrate parvoviruses (28), perhaps as a result of a different evolutionary pressure.

The purpose of this study was (i) to obtain an infectious clone of the virus and a tissue culture host that supports it, (ii) to determine the genome organization of GmDNV, and (iii) to investigate the strategy of transcription and expression. It was found that this virus has an ambisense genome organization, confirming results obtained by in vitro translation of hybrid-selected mRNA (17), with long inverted terminal repeats (ITRs). The expression strategy was found to differ from that of all other parvoviruses analyzed thus far. The capsids were also found to contain a domain with phospholipase A₂ activity. Finally, the roadmap of its capsid surface was compared with that of JcDNV, since other studies (e.g., see references 16 and 29) have indicated that parvovirus allotropic determinants reside on or near the virion surface.

MATERIALS AND METHODS

Production and purification of GmDNV and isolation of viral DNA.

GmDNV was obtained by infection of G. mellonella larvae and purification in CsCl gradients as described previously (31). DNA was extracted from virus particles banding at 1.40 g/ml after dialysis against a buffer containing 50 mM Tris-HCl (pH 8.0), 100 mM NaCl, and 10 mM EDTA. In the original DNA extraction method, proteinase K was added to a final concentration of 50 μg/ml and SDS was added to a final concentration of 0.5%. The sample was incubated for 1 h at 56°C and then diluted with an equal volume of cold water and extracted three times with phenol (equilibrated with 50 mM Tris-HCl [pH 8.0]), once with phenol-chloroform (1:1), and once with an equal volume of chloroform. The phenol phase with interphase was back extracted several times with the buffer. The DNA was precipitated with ethanol from the combined suspensions, resuspended in 1× TE (10 mM Tris-HCl [pH 8.0], plus 1 mM EDTA), and stored at −20°C. Alternatively, the High Pure plasmid isolation kit from Roche Molecular Biochemicals (catalog no. 1754785; Laval, Canada) was adapted to improve quantitatively and qualitatively the extraction of viral DNA. Lysis buffer (300 μl of 6 M guanidine-HCl, 10 mM urea, 10 mM Tris-HCl, 20% Triton X-100 [pH 4.4] containing 80 μg of poly(A) carrier RNA/ml) and 200 μl of sample were mixed and incubated for 10 min at 70°C. The sample was vortexed after adding 125 μl of isopropanol, and the DNA was then purified on the spin columns according to the supplier's instructions.

Cloning of viral DNA.

Complete DNA was blunt ended using Klenow and T4 DNA polymerase (10 U each with 300 ng of DNA and 35 μM deoxynucleoside triphosphates in 50 mM Tris-HCl [pH 8.0], 5 mM MgCl₂, and 10 mM dithiothreitol), heated for 15 min at 70°C, and then ethanol precipitated. Either complete DNA or restriction fragments were cloned by standard methods. Terminal fragments (of about 1 kb) were obtained by digestion with HindIII and cloned into HindIII/SmaI-digested pUC19 (pGmH series of clones). A BamHI restriction site was present at about 285 nucleotides (nt) from either end, and the large middle fragment of about 5.5 kb was also cloned in pUC19 (pGm3 and pGm4, respectively, forward and reverse orientation). These two sets of clones served to obtain pBluescript II(KS) clones, which contained the complete viral genome (pGm1 and pGm2, respectively, forward and reverse orientation).

Sequencing and analysis of GmDNV DNA.

Originally, M13 subclones, which formed overlapping nested sets, were sequenced by the dideoxy method (Sanger) using the Sequenase (U.S. Biochemicals) or T7 sequencing (Pharmacia) kits. Recent resequencing was by primer-walking using an ABI310, according to the manufacturer's instructions. Both directions, from independent clones, were sequenced. The terminal hairpins yielded compressions which were difficult to read; however, prior treatment with the restriction enzymes (AlwNI and BstUI in separate experiments) that cut in that region yielded clean sequence reads.

Isolation and characterization of viral RNA.

Total RNA was isolated from infected larvae and transfected, LD652 cells, 48 h postinfection, by the method of Chomczynski and Sacchi (10). Total RNA extracted was subjected to an mRNA purification using an mRNA isolation kit (Roche).

Northern blots.

The mRNA was then submitted to electrophoresis on a 1% formaldehyde-agarose gel and blotted on positively charged nylon membranes (Roche). The blotted membranes were incubated overnight at 65°C with the digoxigenin-labeled probes (the 2-kb nonstructural (NS) probe was obtained by PCR amplication between nt 635 and 2658, whereas the 1.3-kb VP probe was amplified between nt 3151 and 4422 of the GmDNV genome). Hybrids were identified with the chemiluminescence kit (CSAPD; Roche) and exposure for 3 h of BioMax film (Kodak) to the membrane.

Mapping of 5′ and 3′ ends of viral transcripts.

From the open reading frames obtained by sequence analysis, the most probable location of the transcripts was predicted. The 3′-rapid amplification of cDNA ends system (15) was used to characterize the 3′ ends of the polyadenylated transcripts. A primer (RNA-TAG, 5′-GGG TCT AGA GCT CGA GT [17]) was synthesized which would recognize poly(A) tails [all GmDNV transcripts bear poly(A) tails (18)] and serve as the primer for reverse transcription. Primers upstream of putative AATAAA polyadenylation sites and the ADAP primer (as RNA-TAG but lacking 3′-terminal 16 T) were then used to amplify the 3′ ends of the viral transcripts. For the 5′ end of the transcripts, primers at different distances from the putative transcription starts were used to generate mRNA using AMV reverse transcriptase. After degradation of the RNA by RNaseH, the cDNA was dA tailed using terminal transferase. Using internal DNV-specific primers, reagents, and oligonucleotides and the ADAP primer, amplification by PCR was carried out according to standard methods (15). Restriction sites appended to some primers (such as in RNA-TAG) and just downstream of other primers were used to clone the amplicons. The cloned segments were then submitted to dideoxy sequencing as described above.

Expression of structural proteins by the baculovirus system.

In the sequence (below), the potential initiation codons were identified for the VPs (the first five AUGs in ORF4). Each of the potential coding sequences was cloned into the Autographica californica nuclear polyhedrosis virus downstream of the polyhedrin promoter by the Bac-To-Bac Baculovirus expression system (GIBCO-BRL) developed by Luckow et al. (23) according to the supplier's instructions.

The following primers were synthesized in order to obtain by PCR amplification the potential structural protein genes (homologous viral sequence are in caps): DNVP1 gtgggatccaaTTAGTCACTATGCTTTCTTCAAAAATC, DNVP2 gtgggatccacaATGTCCCGTCATATTAAT, DNVP3 gtgggatccacaATGCAAGAAGCAACGAAACG, DNVP4a gtgggatccGCTATGTCATTACCAGGAACTGGC, and DNV4b gtgggatccACAATGGCTATaTCATTACCAGGA as forward primers (italics indicate initiation codons) and DNVPR CACGTCGACTTGCTTTTCAAGAAGCTCCG as the reverse primer. The expressed proteins were analyzed by SDS-PAGE (21) and Western blotting (33) and compared to the VPs from the virus. The ability of the individually expressed capsid proteins to form intact viral capsids was assessed by electron microscopy.

Amino-acid sequencing.

Structural proteins from GmDNV were separated by SDS-PAGE (21) on 10% polyacrylamide gels and in the presence of thioglycolate, electroblotted onto polyvinylidene difluoride membranes (Westran; Schleicher & Schuell, Keene, N.H.) (33), and sequenced according to the method of Matsudaira (24).

Transfection of GmDNV clones.

Transfection of insect cells (LD652) was carried out using the DOTAP transfection kit (Roche). For this purpose, 2.5 μg of viral DNA was precipitated with 10 μg of DOTAP in 1 ml of serum-free medium for each well of six-well plates containing the same cell densities as were used for infection.

Nucleotide sequence accession number.

The GenBank sequence accession number for GmDNV is L32896.

RESULTS

Purification of GmDNV and isolation of DNA.

The yield of virus was about 5 μg of DNV per larva. The widely used SDS-proteinase K method for extraction of DNA yielded rarely more than 5 to 10% maximum recovery, even when combined with several back extractions. The extraction conditions were chosen so that the complementary single-stranded DNA annealed upon release from the virion. Electrophoretic analysis of the extracted DNA showed, in addition to the expected 6-kb band, a fuzzy band corresponding to about 4 kb, several bands at high molecular weights, and a general strong background over the whole lane. Most of the DNA stayed in the sample well and may have formed large complexes. The final yield of monomeric double-stranded DNA (<1 μg of DNA/mg of virus) was reproducibly much lower than for comparable DNVs from Pseudoplusia includens, Mythimna loreyi, Acheta domesticus, Casphalia extranea, Spodoptera littoralis, and Chilo agamemnon (unpublished observations). In contrast, the recovery of DNA from GmDNV was virtually complete with the spin column method. Electrophoretic analysis of this DNA did not demonstrate extra visible bands or background, while very little DNA stayed in the sample well.

Cloning and stability of clones.

Preliminary single or multiple digestions with restriction enzymes revealed useful sites for cloning. Since a BamHI site was found only near both ends, first a large BamHI-BamHI fragment (5.48 kb) from the viral genome was cloned into pUC19 (two orientations, pGm3 and pGm4). Both clones served for further restriction and sequencing analysis. Cloning of complete viral DNA was unsuccessful. Therefore, viral DNA was blunt ended, digested with HindIII, and cloned into pUC19 vectors digested with SmaI and HindIII in order to obtain the terminal fragments of the viral genome. The bacterial host and the temperature of incubation of the cultures determined the stability of the inserts. Sure cells, but not JM101, DH5α, XL-1 blue, or Stbl2 cells, with an incubation at 30°C (instead of 37°C), were found to confer stability to the clones over at least five successive overnight cultures. After PvuII digestion, clones obtained using XL-1 blue or Stbl2 bacteria usually yielded inserts that had undergone recombination. These clones yielded a doublet containing the expected 602-nt and an additional 505 (deleted) nt fragments indicating that only a fraction of the population of that clone had not yet undergone recombination). After a few overnight passages all clones had undergone deletions in the hairpins.

Sequencing of DNA.

The complete sequence of the viral genome was determined for both strands (Fig. 1). The total length of the genome was 6,039 nt (the longest parvovirus genome described to date) and contained ITRs of 550 nt, although many clones were found up to 4 nucleotides shorter from either end. The convention of Armentrout et al. (2) to present parvoviral sequences with the 3′ end of the viral genome (minus strand) to the left could not be adopted since both strands are packaged and both contain large open reading frames (ambisense genome, see below). Since all parvoviruses following this convention have the genes coding for the nonstructural proteins in the left half of the genome, it was decided to define the strand of the GmDNV genome containing the ORFs for the NS genes as the “+” strand so that the genes would be similarly located.

FIG. 1.

Sequence of the GmDNV genome. The first 136 nt in the 550-nt ITR can be folded in a typical parvovirus Y-shaped hairpin in which the middle 32 nt occurred in both forward and reverse complementary orientations (flip/flop). The transcripts started 23 nt downstream of the ITRs with the TATA box just inboard of the ITRs. The forward transcription and translation domains are indicated above the sequence, whereas those from the complementary strand are underneath the sequence. The NS and VP transcripts were found to be antisense between positions 2954 and 3013. Domains conserved with other DNVs are shaded (I, II, putative replication initiation domains; A, B, and C, putative DNA-dependent ATPase motifs; and PLA2, phospholipase A₂ motif).

The distal 136 nt of the ITRs could be folded into the Y-shaped hairpin terminals typical of many parvoviruses (Fig. 2A). In the stem, the first 52 nt are perfectly complementary to nt 85 to 136, whereas nt 53 to 84 can be folded into two asymmetrical GC-rich arms and are present in either of two alternative sequences of 32 nucleotides, “flip” (unpaired pyrimidines in tips of arms in 5′ to 3′ orientation and unpaired purines in the 3′ to 5′ orientation) and its reverse complement, “flop” (Fig. 2A).

FIG. 2.

Organization of the GmDNV genome. (A) Folding of the genome termini of GmDNV into a flip and flop (its reverse complement) orientation. The bipartite AlwNI recognition site is underlined. Since one of its nucleotides is located in the flip/flop side-arm, digestion with this enzyme readily distinguishes flip and flop orientations. (B) Structure of the JcDNV hairpin. (C) Northern blots with NS and VP probes. (D) Summary of the GmDNV genome organization.

Large ORFs.

Both complementary strands contain large ORFs in their 5′ halves, one strand with three ORFs (ORF1 to ORF3) of which two are overlapping and one strand with a 2.5-kb ORF (ORF4 [Fig. 1]). Excluding frameshifting, only the latter would be large enough to code for the largest structural protein, VP1. Since all structural proteins have common sequences (29), this suggested that the other structural proteins, VP-2, VP-3, and VP-4, are coded by ORF4 as well. ORF1 to ORF3 would then code for the nonstructural proteins. This hypothesis was reinforced by the observation that ORF1 contains the NTPase motif found in the NS1 of other parvoviruses (3) and that ORF4 contains the phospholipase A₂ motif recently identified in the structural proteins of most parvoviruses (14, 22, 36).

Mapping of viral transcripts.

The initiation codon ATG for VP1 was found to start 28 nt after the ITR, suggesting that most of the upstream promoter elements would be located within the ITR. Consequently, the identical promoter elements would be present in the ITR upstream of the NS transcripts. The inboard ends of the ITRs contained potential TATA boxes (Fig. 1). Northern blotting revealed two NS transcripts (about 2.5 and 1.8 kb) and 1 VP transcript (about 2.6 kb) (Fig. 2C). Both NS transcripts were sequenced to further study the mode of expression.

The VP transcript was found to start 23 nt upstream of the 3′ ITR at nt 5467, and the starts of both NS transcripts were at nt 573, also 23 nt downstream of the 5′ ITR. The sequence context of both starts corresponded reasonably well with the consensus sequence for Inr boxes (TCAGTG). The 5′-leader sequence in the VP mRNA to the putative VP1 AUG was therefore only 5 nt long, whereas for the two NS transcripts the 5′-untranslated regions were 82 (1.8-kb transcript) and 84 nt (2.5-kb transcript) long.

The 3′ ends of the transcripts were identified at nt 2954 for the VP transcript and at nt 3013 for both NS transcripts (Fig. 1), i.e., both 13 nt downstream of their canonical AAUAAA polyadenylation signals that start at nt 2971 and nt 2995, respectively. Consequently, the VP transcript overlapped the NS transcript by 60 nt, but it was not clear whether this antisense RNA had any regulatory function. Recently we have observed this phenomenon also with transcripts from other DNVs in the same genus (unpublished observations).

No splicing was observed in the VP transcripts. However, the 1.8-kb NS transcript was found to be a spliced form of the 2.5-kb unspliced NS mRNA (Fig. 1). As estimated by Northern blotting, about 50 to 75% of the NS transcripts were spliced (Fig. 2C). The first initiation codon in the unspliced 2.5-kb NS transcript at nt 657 was in a reasonably favorable context and would yield the 232-amino-acid NS3 protein. This NS3 coding sequence was completely removed in the spliced NS transcript. A single 707-nt splice was observed between nt G(654), 2 nt upstream of the AUG start codon of NS3, and A(1362) of the AUG initiation codon of NS1, 7 nt downstream of the stop codon for NS3 (Fig. 1). The two large downstream ORFs could thus be translated by a leaky scanning mechanism from the same transcript as a result of this splicing. The first initiation codon in this transcript, in the NS1 ORF, starting at nt 1362, was, after splicing, not in a favorable context, which might lead to leaky scanning and explain additional translation from the initiation codon starting at nt 1369, which has a consensus Kozak sequence. The two largest NS proteins (544 and 274 amino acids, respectively) would therefore be generated from the smaller NS transcript.

Predictions of the molecular masses of the four structural proteins from the first four in-frame AUGs in the VP ORF corresponded well with the values seen in SDS-PAGE (32) and suggested a leaky scanning mechanism for this gene as well. The very short untranslated 5′-leader sequence and the less favorable contexts of VP2 and VP3 AUG initiation codons would support such a mechanism. Nevertheless, it would be possible that the set of VPs is generated by proteolysis of the largest VP (VP1), translation downstream of (an) internal ribosome entry site(s), or ribosomal shunting.

Translation products.

The capsid proteins were submitted to N-terminal sequencing in order to confirm that these proteins start with the methionines predicted for translation initiation. This was successful only for VP3, since the other N termini were blocked. The sequence obtained for VP3, MQEATK*KADSP (asterisk indicates ambiguous amino acid), indicated that its translation started at AUG3 at nt 4493. Previously, Dumas et al. (12) reported for JcDNV that VP4 also starts with an AUG (starting at 4340). This suggested that VP4 initiates at the downstream AUG of the AUGCTGAUG sequence. It remains possible, however, that the N terminus of another VP4 translation product, initiated at the upstream AUG codon, would be blocked and undetected.

In order to investigate further the translation process, the ORF for the VP and 5′-truncated forms that started just upstream of the different candidate initiation codons were inserted into a baculovirus expression vector in order to eliminate regions that could be involved in internal ribosome entry site or shunting activities. The expression products were analyzed by electron microscopy and Western blotting (Fig. 3). Expression of the VP-1 yielded the same four proteins as for the virus, with similar relative amounts, despite a much longer 5′ untranslated leader sequence in this mRNA than in the case of the viral mRNA (91 versus 5 nt). These proteins auto-assembled into virus-like particles that could not be distinguished from virus particles. The context of the candidate AUGs for VP2 and VP3 initiation was improved by changing them to the context found for VP4. Whereas for VP4 hardly any leaky scanning was observed, both VP2 and VP3 initiation codons still showed some leakiness in the baculovirus expression system, although the large majority of translation products now started at the 5′-proximal initiation codon. The size of the proteins that were obtained corresponded to that obtained for proteins from the virus. VP4 was expressed using either of the potential initiation codons between nt 4340 and 4349 by mutating the other. Western blotting showed that strong initiation was obtained from both AUG codons, and electron microscopy revealed that all expression experiments yielded virus-like particles.

FIG. 3.

Expression in baculovirus of GmDNV structural protein from the first five initiation codons of the ORF (3 to 7). Their size corresponded well with the size in the native virus (1). Nonrecombinant baculovirus was used as a control (2). All expressions yielded virus-like particles (bars, 20 nm).

Phospholipase A₂ activity.

Previously we identified a phospholipase A₂ (motif between amino acids 176 and 230 of the unique region of VP1, and we demonstrated that the expressed protein domain (amino acids 2 to 378) has weak phospholipase A₂ activity, i.e., a (k_cat/K_m)_app of about 4 × 10³ M⁻¹ · s⁻¹, i.e., about 30% of the enzyme activity from DNVs from the Iteravirus genus (14, 22, 36). To investigate whether the domain that was chosen had an impact on the activity, proteins were expressed with domains covering amino acids 1 to 276, 1 to 323, 176 to 323, and 176 to 276,. The activities obtained in the mixed micelles assay at 30°C (highest activity) were similar for all proteins and decreased within 2 weeks to virtually undetectable, whereas the phospholipase A₂ activity of the vertebrate parvoviruses remained stable. The proteins were not degraded, suggesting that an incorrect folding caused this low activity.

Transfection with cloned DNV.

GmDNV was found to infect LD652 cells from Lymantria dispar but not SPC-SL52 from S. littoralis. Transfection of LD652 cells with either pGm1 (containing the complete viral genome) or pGm4 (lacking the ITRs from the BamHI sites to the extremities) resulted in positive immunofluorescence in the nuclei of LD652 cells, indicating VP production, but not after mock transfection. Supernatants were collected from pGm1-, pGm4-, and pBluescript-transfected LD652 cells 3 days after transfection, and 5 μl was injected into larvae. Only the supernatant from the pGm1-transfected cells contained infectious virus, as indicated by the mortality of the injected larvae in a dose-dependent manner and the production of virus in these larvae as confirmed by electron microscopy. Injection of larvae with supernatants from pGm4- or mock-transfected cells did not result in mortality, and these larvae underwent normal moulting.

DISCUSSION

Cloning and sequencing of GmDNV genome.

Cloning of the full-length genome of GmDNV and prevention of rearrangements in bacteria proved to be much more difficult than the preparation of infectious clones from other DNVs (14, 22) or porcine parvoviruses (4, 5). Both the use of Sure bacteria and culturing at 30°C instead of 37°C were found to be essential. The use of recombination-defective bacteria was previously shown to confer increased stability (7). Five successive overnight cultures in other bacterial hosts, even in Stbl2, or at 37°C all led rapidly to deletions by recombination.

When the sequence of the GmDNV genome was initially determined, several important differences with the JcDNV sequence reported by Dumas et al. (12) were observed: (i) the perfect Y-shaped structure of the JcDNV terminal palindrome (no mismatches in side-arms) was not obtained for GmDNV; (ii) the putative TATA boxes for the NS and VP proteins were for GmDNV inside the ITRs, in contrast to JcDNV, and made it difficult to predict how temporal regulation of expression would be achieved; (iii) the ITRs of GmDNV had identical lengths and thus identical hairpins, in contrast to JcDNV, which, compared to GmDNV, lacked 109 nt at the NS end (no terminal hairpin) and 16 nt at the VP end, which raised the question of whether these ITR differences were somehow involved in temporal regulation of transcription; and (iv) the predicted NS3 of GmDNV was substantially longer than the NS3 of JcDNV. We adopted the same genome orientation and position numbering for JcDNV as for GmDNV to facilitate comparison.

Recently we obtained more independent clones, and clean sequencing reads of the hairpins were obtained with the modified procedure instead of the notoriously difficult-to-interpret compressions usually associated with parvoviral ITRs. This showed that the GmDNV sequence that was originally submitted to GenBank (L32896) was correct. The JcDNV ITR at the VP end in the pBRJ clone was also sequenced and was also found to be correct, indicating structural differences between the ITRs of GmDNV and JcDNV (Fig. 2A and B) caused by an insertion of unpaired bases at positions 55 and 67 in the flop orientation of GmDNV. This is surprising, since for other DNVs a very strong conservation is found in the flip/flop region (cf. BmDNV and CeDNV [14, 22]). The ITRs of JcDNV were found to be as long as those of GmDNV but with a deletion of G258 and mispairings at 540 and in the TATA box at 546 and 548 (numbers correspond to those of GmDNV). The ITRs reported for JcDNV are shorter than those of GmDNV, probably because termini lacking hairpins may have been preferentially cloned or deletions may have occurred due the bacterial host or during subcultures at 37°C. For GmDNV, cloning into XL-1 cells at 37°C gave mostly deleted ITRs, whereas most ITRs obtained using Sure cells and at 30°C were complete. Whereas for JcDNV only a flop orientation was obtained at one end, we obtained both flip and flop orientations at both ends.

The difference in lengths of the predicted NS3 proteins was found to be caused by a sequencing error in JcDNV, i.e., an insertion of 1 nt, corresponding to a position between nt 686 and 687 of GmDNV, thus causing a frameshift. The sizes of NS3 of the two viruses were very similar after this correction.

In contrast to a previous report (17), both spliced and unspliced NS transcripts served for translation, a strategy reminiscent of that of vertebrate parvoviruses. However, for vertebrate parvoviruses with multiple NS products the splicing occurs within the major NS ORF, resulting, after translation, in proteins with different C-terminal domains, whereas splicing of GmDNV NS transcripts resulted in the expression of alternate genes. Unlike vertebrate parvoviruses, the GmDNV VP and spliced NS transcripts direct the synthesis of proteins from multiple initiation codons via a leaky scanning mechanism.

The spliced, bicistronic NS transcript directs the synthesis of NS1 and NS2, initiated, respectively, at the first (NS1) and second (NS2) AUG codons that are spaced by 4 nt (AUGAACAAUG). Interestingly, RNA-6 of influenza virus B also directs the synthesis of two proteins (integral membrane glycoprotein and neuramidase) from two initiation codons by leaky scanning with an identical spacing sequence of AACA (26). It was suggested that linear scanning by the 40S ribosomal subunit may break down if two AUGs are in close apposition, since increasing the spacing to 46 nt prevented initiation from the second AUG (35). However, Kozak (20) provided evidence of adherence to the first-AUG rule and a context-dependent leaky scanning mechanism. It is also interesting that the context of the first AUG in influenza virus RNA-6, with an A in position −3, is adequate (GenBank no. NC_004284). Although the GmDNV NS1/NS2 transcript has a long leader sequence, it has a suboptimal context of the 5′-proximal AUG codon. The better context of the NS2 initiation codon probably explains why the 30-kDa NS2 was found to be the most abundant translation product after in vitro translation of viral RNAs (17).

Removal of potential internal ribosome entry sites or ribosomal shunting domains by expression of the VP-coding domains without upstream sequences into baculovirus did not change the production of multiple VPs. Translation by a leaky scanning mechanism, in which the 40S ribosomal subunits bypass initiation codons, is rare, in particular for more than two translation products (20). Nevertheless, the DNVs of the Iteravirus (14, 22) as well as the Densovirus genus, as exemplified by GmDNV, use this strategy to generate their sets of structural proteins. In contrast, parvoviruses from vertebrates use alternative splicing. Interestingly, if the very short untranslated 5′ region (5 nt) of the GmDNV VP transcript was replaced by a long untranslated region in the baculovirus system, leaky scanning was not significantly different, suggesting that other factors contribute to this mechanism. The production of the relative amounts of the four viral proteins may, however, differ slightly from batch to batch.

The allotropic determinant of some parvoviruses has been located in the structural proteins (5, 18) and is present on or near the surface of the virion (29). Preliminary results in our laboratory, using chimerae obtained by exchanging restriction fragments between the infectious clones of GmDNV and MlDNV, indicated that the allotropic determinants of these viruses are also located in the VP gene. Some of the limited number of amino acids on the surface of the virion that differ with those on the JcDNV particle, as predicted from the GmDNV structure (28) and the sequence differences between GmDNV and JcDNV, could thus be responsible for the host range of each virus (Fig. 4). This polymorphism could also be responsible for other phenotypic properties that distinguish these closely related DNVs. This roadmap shows that there are only 10 amino acid differences on the surface of each protein of the viral capsid, some of which are adjacent in the tertiary structure but not in the primary structure (e.g., N677 and E755, T568 and D549). Nevertheless, it remains possible that some of the N-terminal extensions of the structural proteins are somehow involved in tropism. Currently, the impact of these amino acid differences on the host-range phenotype is further investigated.

FIG. 4.

Roadmap of amino acids in one of the asymmetric units on the surface of the GmDNV particle. Locations of amino acids that differ from those of JcDNV are surrounded by thick lines and indicated on the virus particle. These highlighted amino acids probably are responsible for the phenotypic differences, such as host range, between these two viruses. The penton indicates the fivefold axis, the triangle indicates the threefold axis, and the biface symbol indicates the twofold axis.