Although Fenner and Comben first described recombination between coinfecting poxviruses 50 years ago (16
), the enzyme(s) that catalyzes this process and the mechanism remain obscure. Our previous studies had suggested that poxvirus recombination reactions used some form of SSA mechanism but that these reactions exhibit the unusual property of using a 3′-to-5′ exonuclease to resect duplex ends (63
). VAC encodes only one known exonuclease, and that is the 3′-to-5′ exonuclease of the viral DNA polymerase. When these observations are combined with evidence that highly purified VAC E9 can catalyze duplex-DNA-joining reactions in vitro and is seemingly required for recombination in vivo, they provide strong support for the hypothesis that poxviruses use the DNA polymerase proofreading activity to catalyze genetic exchange.
Mutant viruses provide an ideal way of testing this hypothesis. However, our attempts to rescue mutant VAC strains lacking the 3′-to-5′ exonuclease function were unsuccessful. Although two temperature-sensitive E9L alleles serve as very useful selectable markers, one sees a strong selection against coconversion of these sites and three sites encoding aspartic acid residues that should be critical for exonuclease function. This is illustrated by a dramatic reduction in the recovery of recombinant virus (Fig. ), which we presume is due to the loss of recombinant viruses encoding exonuclease mutations. This bias is exacerbated as the linkage gets tighter, presumably due to the reduced likelihood of recombination separating the two markers. The strongly biased selection against the coconversion of these markers is illustrated by the fact that although the D166A and Dts83 (H185Y) mutations are located only ~60 bp apart, none of the temperature-resistant viruses encoded a novel AluI site that is diagnostic for the D166A substitution (Fig. ). The fact that other silent mutations can be introduced into the sites encoding Exo II and III motifs shows that these effects are not caused by some peculiar impediment to recombination (Fig. ). The selection pressure specifically disfavors the production of viruses encoding D-to-A substitutions at sites predicted to be critical for exonuclease activity, and this leads us to conclude that the 3′-to-5′ proofreading exonuclease is an essential virus function.
Because one cannot generate a VAC strain constitutively lacking 3′-to-5′ exonuclease activity, we decided to use an alternative, antiviral-based approach to study the links between proofreading and recombination. As a starting point, we determined that when CDV is added to the culture medium, it inhibits recombination between circular plasmid substrates in cells infected with wild-type VAC (Fig. ). Plasmid substrates are useful tools for studying poxvirus replication and recombination, because the DNA accumulates at sites of viral replication in the cytoplasm of infected cells, and plasmid replication utilizes all five of the virus-encoded proteins needed for viral genome replication (12
). A previous report has shown that treating VAC-infected cells with DNA polymerase inhibitors such as cytosine arabinoside and aphidicolin also inhibits the recombination of transfected plasmid substrates, emphasizing the need for a functional E9 in VAC recombination (6
). Furthermore, this study found that the rate of recombination of these plasmid substrates by different mutant VAC strains was independent of the rate of replication of these substrates by each strain (6
). This suggested that the VAC DNA polymerase was participating in homologous recombination primarily at some level other than just DNA synthesis (6
). CDV also inhibited plasmid recombination, but the amount of recombination was affected differently by different CDVr
mutations (Fig. ). Viruses encoding the A314T mutation were less susceptible to CDV than A684V-encoding or wild-type viruses (Fig. ), even when the drug doses were chosen to have comparable effects on DNA replication and virus yield. Although one can never exclude the possibility that the different drug doses are in some unknown way modulating host effects on virus recombination, the simplest conclusion that can be drawn from these experiments is that the 3′-to-5′ exonuclease activity likely serves a different role in virus recombination than does the 5′-to-3′ polymerase activity.
To avoid any impact of these hypothetical host effects and further elucidate the effect of CDV on poxvirus recombination, we prepared linear recombination substrates containing CDV as the penultimate residue on the 3′ ends (Fig. ). DNAs bearing these structures are resistant to exonuclease attack (34
). This alteration clearly inhibited joint-molecule formation catalyzed by wild-type E9 in vitro (Fig. ), as well as recombinant production in cells infected with wild-type and A684V-encoding VAC strains in vivo (Fig. ). In contrast, a mutant E9 protein incorporating the A314T substitution could still catalyze joint-molecule formation in vitro (Fig. ), and viruses encoding the A314T mutations still efficiently catalyzed the recombination of CDV-containing substrates in vivo (Fig. ). Biochemical studies showed that these effects correlate with an enhanced capacity of the A314T-encoding E9 enzyme to excise CDV from DNA (Fig. ). A complicating factor is that one can never completely fill in the ends of these substrates with CDV. Using densitometry, we estimated that ~17% of the XhoI-cut substrates that were end filled with CDV, shown in Fig. , lacked 3′ CDV residues. This finding corresponds well with the 13 to 20% yield of recombinant molecules typically produced by wild-type E9 using CDV-bearing substrates (Fig. and ). Of course, these limitations in end-filling efficiency could also explain why some recombination was still detected in cells infected with wild-type and A684V-encoding viruses and transfected with substrates into which CDV had been incorporated (Fig. ).
The 2- to 3.5-fold differences in the absolute recombination frequencies between our luciferase (Fig. ) and Southern blot-based (Fig. ) assays should be noted. Southern blotting represents the method of choice for providing insights into the structure and quantity of the DNA recovered from infected and transfected cells. However, these techniques are subject to high experimental variation due to possible differences in infection and transfection efficiency. Luciferase-based assays are convenient and are normalized to β-galactosidase expression from a cotransfected plasmid, thus providing more control for infection/transfection variations and differences in plasmid DNA replication. However, these assays tend to underestimate the Rf, because a functional luciferase can be expressed only after substrate recombination has created an intact gene and transcription and translation has occurred. Despite these differences, both methods provided a consistent finding: CDV residues inhibit recombination, and viruses encoding the A314T substitution are far more resistant than other strains.
The enhanced ability of A314T-encoding E9 to excise CDV from DNA ends likely explains why this mutation arose during repeated passage of VAC in CDV-containing media (1
). Interestingly, the same mutation is recovered when VAC is passaged in media containing the related compound (S
)-9-[3-hydroxy-2-(phosphonomethoxy)propyl]adenine (HPMPA) (1
). We have also found that HPMPA, when located in the penultimate position in a primer strand, is resistant to VAC E9 proofreading activity (34
) and that HPMPA residues also block joint-molecule formation in vitro (D. Gammon, unpublished data). The A314T mutation is the first and most common mutation recovered when poxviruses are exposed to nucleoside phosphonate drugs, and the strong selection pressure acting at this site suggests that exonuclease activity is essential either because of a need to excise drugs that would otherwise inhibit replication or because of a requirement for gapped DNA molecules in some process such as viral recombination, or both. We have previously suggested that the A314 residue would likely be located in a β-hairpin structure that is highly conserved among B-family DNA polymerases (1
). A recent study of RB69 and T4 phage DNA polymerases either containing mutations in this structure or lacking the β-hairpin structure completely have suggested that the β-hairpin plays an important role in separating primer and template strands and thus facilitates the removal of mismatched bases (24
). RB69 polymerase mutants lacking the β-hairpin can degrade single-stranded DNA, as well as duplex DNA containing three terminal mismatches, as efficiently as the wild-type enzyme but are hindered in their ability to excise a single 3′-terminal mismatched nucleotide (52
). We have noted that 3′-terminal CDV and HPMPA molecules are as readily excised by wild-type E9 as are dCMP and dAMP (34
). However, the incorporation of one additional nucleotide renders CDV and HPMPA residues quite resistant to the exonuclease. We suspect that the A314T substitution confers on VAC E9 an enhanced ability to destabilize a strand bearing a nucleoside phosphonate drug in the penultimate 3′ position, leading to strand separation, excision of the drug residue, and 3′-to-5′ resection of duplex DNA. These resected molecules could then serve as recombination substrates in an annealing reaction stimulated by the VAC single-strand binding protein I3. We have been shown previously that I3 promotes Mg2+
-dependent DNA aggregation in vitro (53
) as well as increasing the efficiency of joint-molecule formation in reactions catalyzed by purified VAC DNA polymerase (61
). Single-strand DNA-binding proteins are known to play an important role in many repair, recombination, and replication systems. The fact that one can inhibit in vivo recombination with a siRNA targeting I3 mRNA (Fig. ) is consistent with the protein also serving some recombination-related function(s) in VAC-infected cells. However, knocking down I3 levels also inhibits DNA replication (Fig. ), an important fact that is discussed in more detail below.
If a viral DNA polymerase plays some role in catalyzing recombination, it follows that any intervention that causes a switch between polymerization and exonuclease activities should also alter recombination rates. We have previously noted that increased dNTP concentrations inhibit E9-catalyzed strand-joining reactions in vitro (22
). To determine whether dNTP pools also affected recombination in vivo, we examined the effects of the ribonucleotide reductase inhibitor HU on VAC recombination. HU has previously been shown to inhibit VAC replication (50
), and at 10 mM concentrations, it can also enhance the recombination of circular plasmids in VAC-infected cells (6
). HU has also been shown to enhance cellular recombination rates (19
). We varied the drug dose and detected a clear trend where the yield of recombinants increased with increasing doses of HU (Fig. ). The highest doses of HU that were tested (0.5 and 5 mM) are known to reduce dATP pools to about 10% of the levels detected in untreated, VAC-infected BSC-40 cells (50
), and they increased recombinant production ~2-fold in our studies. Of course, these experiments assume that HU exclusively inhibits a ribonucleotide reductase activity in VAC-infected cells and does not induce some other, unknown host-dependent recombination-enhancing effect(s). To test this, we have recently generated a VAC strain lacking the F4L
gene, which encodes the small subunit of the viral ribonucleotide reductase. This ΔF4L
virus also exhibits an enhanced-recombination phenotype linked to deficiencies in replication. In particular, this virus recombines transfected linear plasmid DNAs at frequencies ~12% higher than wild-type VAC while producing only ~20% of the DNA seen in cells infected with wild-type virus (D. Gammon, unpublished data).
How one interprets all of these experiments is frustratingly complicated by the intimate links between viral recombination and replication. Where possible, we have used cotransfected plasmids, encoding β-galactosidase, to normalize the levels of luciferase and thus avoid detecting a simple artifactual link between reduced rounds of replication and reduced amounts of replication-associated recombination. Moreover, it should be noted that these experiments show how VAC replication and recombination are not always linked in such a simple manner. Knocking down I3 expression inhibits replication and inhibits recombination (Fig. ), but interfering with the ribonucleotide reductase activity inhibits replication (50
) while enhancing recombination (Fig. ). Furthermore, a previous study of the recombination of transfected plasmid substrates in cells infected with mutant VAC found that the recombination rates measured in these different strains were largely independent of the plasmid DNA replication rates (6
). VAC replication and recombination are undoubtedly tightly linked processes, sharing many common enzymes, but one cannot explain these observations with a simple model that directly links replication rates to recombination frequencies.
Our studies suggest a more complex mechanism by which a DNA polymerase could catalyze recombinational repair in a manner dependent on dNTP regulation of the balance between 5′-to-3′ DNA synthesis and 3′-to-5′ exonucleolytic processing. Poxviruses have recently been shown to encode a DNA primase (11
), suggesting that DNA replication involves leading- and lagging-strand DNA synthesis at a classical replication fork. Under such circumstances, a potentially lethal double-strand break would be created following a collision between the replication fork and a nick located on either strand ahead of the replication complex (Fig. ). By attacking the 3′-ended strand, the 3′-to-5′ proofreading exonuclease could expose sufficient homology to permit the reformation of the original replication fork through I3-assisted SSA. VAC DNA polymerase and DNA ligases could then repair the nick or small gap.
FIG. 10. Proposed model for poxvirus recombination and its regulation by dNTPs. (A) Diagram of how a collision between the viral replication fork and a preexisting nick will lead to replication fork collapse. We propose that the E9-encoded 3′-to-5′ (more ...)
It is more difficult to speculate on how this process might be regulated, but several unusual features of poxvirus biology suggest a way in which dNTP concentrations could play an important role in regulating DNA synthesis versus degradation. The VAC-encoded ribonucleotide reductase (comprising the I4L
gene products) plays a key role in the biosynthesis of the extraordinary amounts of dNTPs required for viral replication (26
). Interestingly, F4 interacts with I3 (9
), and it has been suggested that the VAC single-stranded DNA-binding protein I3, like the T4 bacteriophage gp32 (57
), can recruit a putative “dNTP synthetase complex” to the replication fork and thus help couple dNTP production and consumption (26
). If this were true, the collapse of a replication fork would disturb this equilibrium, and the resulting change in the dNTP microenvironment could perhaps favor strand gapping over DNA synthesis (Fig. ). To date we have used immunoprecipitation methods to confirm that I3 can interact with F4 in vitro (C. Irwin and D. Evans, unpublished data), but further studies are needed to test this hypothesis.
Our model provides important new insights into the diversity of enzymes and mechanisms that can be used to catalyze double-stranded break repair. This process can be partitioned into pre- and postsynaptic events, and the role that DNA polymerases might play in it reflects the multiplicity of reactions potentially catalyzed by DNA polymerases. It is well established that particular DNA polymerases can play a specialized role in catalyzing the postsynaptic DNA synthesis associated with homologous (25
) and nonhomologous (4
) recombination. It has also been shown that, in yeast, Pol2 proofreading activity can catalyze the postsynaptic processing of imperfect recombinant intermediates (54
) in a manner similar to that of the reactions that we have shown are catalyzed by the E9 3′-to-5′ exonuclease (21
). However, our present studies strongly suggest that VAC can also use the E9 proofreading activity to catalyze a presynaptic step in genetic exchange. It is difficult to prove this with certainty, because one can always suggest more-complex ways in which the enzyme's role might be limited to catalyzing just a postsynaptic step(s) in recombination. For example, we cannot absolutely rule out the possibility that the proofreading activity of E9 is used to remove CDV from the 3′ ends of the DNA strands after they have entered into a process such as a synthesis-dependent strand-annealing (SDSA) reaction. SDSA reactions typically require the presynaptic 5′-to-3′ gapping of recombination substrates, which exposes 3′-ended single-stranded DNAs that can then invade duplex strands, form displacement loops, and prime DNA synthesis (43
). However, this model seems unlikely, because if poxviruses did use SDSA reactions in vivo, it would follow that these viruses should very efficiently catalyze recombination between linear duplex and circular substrates (51
), and they clearly cannot (64
). Furthermore, we have previously examined the fate of mismatch-tagged recombination substrates and demonstrated that >80% of the mature recombinants recovered from VAC-infected cells retain the mismatched nucleotide originally located on the 5′ strand (63
). This shows that double-stranded breaks are processed in a 3′-to-5′ manner during virus recombination, which would be inconsistent at least with standard SDSA models. We suggest that the in vitro biochemical data provide the simplest and most logical explanation for the in vivo data and that one need posit no more complex a model than that the E9L
-encoded proofreading activity also catalyzes presynaptic gapping in VAC recombination. This proposal does differentiate the VAC recombination system from the vast majority of SSA reactions that depend on 5′-to-3′ presynaptic processing by simple exonucleases (e.g., those of herpesviruses [44
], yeast [42
], and mammals [7
]). Whether this unique mechanism reflects some special constraints created by poxvirus biology (e.g., cytoplasmic replication, genome size, the mechanism of dNTP biogenesis, the lack of a RecA-like recombinase), some unusual feature(s) of E9, or the first example of what is actually a more widespread biological process remains a very interesting question.