Effects of Varying Sequence Similarity on the Frequency of Repeat Deletion during Reverse Transcription of a Human Immunodeficiency Virus Type 1 Vector

We generated an HIV-1-based transducing vector (pHIVlacWT) that contained expression cassettes for both the puromycin resistance selectable marker gene and for the β-galactosidase (lacZ) gene (Fig. 1A). pHIVlacWT was a derivative of pHIVlac (1), with silent mutations that generated EcoRI and XhoI restriction sites flanking a 156-bp region of lacZ. Derivatives of pHIVlacWT in which the156-bp sequence was 95, 91, 82, 73, 63, or 58% homologous to the wild-type sequence were then constructed. For each derivative, a pair of lengthy sense and antisense oligonucleotides with complementary 3′-terminal ends were annealed, and standard procedures for Escherichia coli DNA polymerase I Klenow fragment were used to mutually prime and fill out the oligonucleotides. EcoRI and XhoI restriction sites engineered into the 5′ ends of the oligonucleotides were used to generate replacement fragments, forming pHIVlac95%, pHIVlac91%, pHIVlac82%, pHIVlac73%, pHIVlac63%, and pHIVlac58% (Fig. 1B). Sequence modifications introduced into this region followed three criteria, as follows: (i) the amino acid coding was not altered; (ii) the mutations were distributed evenly over the region; and (iii) mutations in the variants were introduced based on the codon frequency of the parental lacZ gene (to minimize any effect of codon usage on translational efficiency) wherever possible. The integrity of the lacZ variants was confirmed by transfecting individual vectors into ET cells—the ET cell line is a 293T derivative that constitutively expresses ecotropic envelope (30)—and examining lacZ expression with 5-bromo-4-chloro-3-indolyl-β-d-galactopyranoside (X-Gal; Sigma) staining as previously described (30) (results not shown).

Repeat vectors were then constructed by joining a plasmid half that contained the 156-bp portion of lacZ and upstream sequences from one pHIVlac derivative with a plasmid half that included the 156-bp segment and downstream sequences from a second vector, using the XhoI and EcoRI sites that flanked the varied portion of lacZ and the plasmid backbone site, BclI. These two larger fragments were ligated with a 1.5-kb XhoI-MfeI linker fragment that contained murine leukemia virus coding sequence from pNCA (5). In this way, repeat vectors with various combinations of 5′ and 3′ repeats were constructed. For example, pHIVlaac95%-WT contained a 156-bp segment that differed by 5% from that of the wild type, followed by the linker segment and then a wild type 156-bp segment (Fig. 1C). It has been shown for murine leukemia virus that deletion rates increase when repeated sequences are separated by a spacer sequence, presumably because the distance between repeats increases the chance of template switching (6).

It has been suggested that reverse transcriptase template switches do not occur at uniform frequencies and are highly context dependent (2, 40). For example, sequences that potentially form secondary structures have been shown to be hot spots for both mutation and recombination (12, 13, 25). Thus, to address potential sequence-specific effects of the introduced variation on template switching rates, the initial deletion vectors tested contained each of the156-nucleotide (nt) region sequence variants in tandem with an identical repeat of itself (Fig. 2). Because the repeats in each of these vectors were identical to one another, differences among vectors' deletion rates would reveal whether or not particular 156-nt variants displayed significant differences in intrinsic template switching properties.

HIV-1 vector virions were generated by transiently cotransfecting ET cells with the HIV-1 helper plasmid, pCMVΔR8.2 (23), and each of the identical repeat vectors (Fig. 2). The resulting virions were harvested, and mATRC1/293 cells (17) were infected, selected in puromycin, and assayed for β-galactosidase by blue-white screening as previously reported (30). For each vector, three independent infections were performed in triplicate to assess deletion rates during single replication cycles. The numbers of blue and white colonies were summed for each infection experiment, and deletion rates were calculated as ratios of blue colonies to the total number of puromycin-resistant colonies. To correct for the lacZ inactivation rate (5.0% for pHIVlacWT [data not shown]), the deletion rates calculated as described above were divided by 0.95 (i.e., 1 − 0.05). The corrected deletion rates were then averaged from three independent infections for each vector. For the 156-nt wild-type tandem repeat (wt-wt), the measured deletion rate was 15.6% (average calculated from data shown in Fig. 3A and B). Deletion rates for vectors containing identical repeats of 95, 91, 82, 73, 63, and 58% homology to the wild-type sequence were all similar in value and were observed to be 18.6, 16.8, 20.8, 15.6, 18.2, and 17.5%, respectively (Fig. 2). Note that pairwise χ² tests suggest that deletion rates for certain pairs of vectors, such as the 82%-82% and 73%-73% vectors, may differ significantly (P < 0.05). However, most pairwise comparisons did not result in statistically different values; thus, observed deletion rate variation among identical repeat vectors should not affect interpretation of the trends reported here.

Vectors with two different repeats were constructed to test the effects of sequence divergence on repeat deletion (Fig. 3). It is believed that repeat deletion takes place mostly during retroviral DNA minus strand synthesis when the growing point of the newly synthesized minus strand DNA dissociates from the 3′-repeat template RNA (template switch donor) and reassociates with the homologous 5′ repeat (template switch acceptor) (44). Thus, deletion rates were first determined for vector viruses containing the wild-type lacZ sequence as their 3′ repeat, because identical sequences would serve as the template switch donor for each vector under these conditions, provided that template switching occurred during minus strand recombination. As shown in Fig. 3A, deletion rates were 9.6, 3.9, 0.77, <0.04, <0.05, and <0.08% for repeats of 95, 91, 82, 73, 63, and 58% homologies, respectively.

An additional set of vectors was constructed which differed from the first in that the orientation of the repeats was inverted. Thus, the new vectors contained different variants as template switching donors and identical wild-type sequences as template switching acceptors (Fig. 3B). When assayed, deletion rates were determined to be 7.6, 4.9, 0.2, <0.3, <0.2, and <0.1% for repeats of 95, 91, 82, 73, 63, and 58% homologies, respectively (Fig. 3B). Although the total number of colonies examined for each of the vectors in this second set was lower than that of the previous set, deletion rate trends associated with sequence modulation were consistent with those observed for the set of vectors analyzed in Fig. 3A. Statistical analysis revealed no significant difference between the deletion rates between the two sets of vectors (χ² test or Fisher exact test, P > 0.05).

The work in this study tested effects of altering the genetic distance between repeated sequences on HIV-1 template switching. HIV-1 deletion frequencies for repeats of 95% homology were about 65% as high as values for identical repeats. The frequency decreased to 26% of the identical repeat rate for repeats with 91% homology and was about 5% for repeats with 82% homology. Failure to detect any recombinants among thousands of products that resulted from vectors with repeats that differed by 27% (73% homology) suggests that this level of divergence decreased recombination frequency more than 300-fold. Considering that nonhomologous recombination reportedly occurs roughly 1,000-fold less frequently than homologous retroviral recombination (45), this level of recombination suppression appears similar in magnitude to that predicted for two unrelated sequences. The effects of sequence context on repeat deletion rates were tested by placing each variant repeat in tandem with itself. Whether or not there were hotspots for template switching in our 156-nt sequences was not addressed directly. However, the level of similarity in deletion rates observed for the different identical repeat vectors suggests that the magnitude of any possible hotspots was too low to alter the above conclusion.

These results suggest that retroviral recombination is more permissive of genetic variation than is bacterial or mammalian DNA recombination. It has previously been established that homologous recombination is far more frequent during retroviral replication than during cellular replication (10, 15): our new findings suggest that greater levels of sequence divergence are tolerated during retroviral recombination as well. Studies on cellular homology requirements during recombination usually take one of two approaches: either to determine the rate of recombination as a function of the length of shared uninterrupted homology or to test the effects of base pair mismatch on the rate of recombination (see reference 37 and citations therein). The system used here, whereby mismatch was introduced into the 156-base lacZ gene fragment, resulting in variants with differing degrees of sequence homology, could be considered as following the second approach. In bacteria, a 16% mismatch between sequences reduces phage-plasmid recombination rates by a factor of 100 (34), and in mammalian cells, 19% mismatch reduced intrachromosomal recombination over 1,000-fold (38). In comparison, our results show18% mismatch reduced HIV-1 repeat deletion by roughly 20-fold (comparing the observed deletion rate of 0.77% reported here for 82% homologous repeats with the 14.9% deletion rate determined for identical repeats). We realize that this comparison has its limitation because the corresponding rate of cellular recombination was not measured for the 156-bp sequence used in this study. Nevertheless, the findings presented here suggest that HIV-1 recombination may be less sensitive to sequence variation than is the cellular replication machinery. Recent research has demonstrated that DNA mismatch repair systems play critical roles in inhibiting recombination between nonidentical sequences in both bacteria and eukaryotes (7). Mutational inactivation of components of the mismatch repair system can significantly reduce the block to homeologous recombination between similar but nonidentical sequences (22). Whether or not such factors as the host cells' mismatch repair machinery affected our read-out of apparent HIV-1 deletion frequencies was not addressed here.

Here, repeat deletion was used to measure the effect of sequence divergence on HIV-1 template switching, whereas genetic recombination as commonly defined occurs between sequences on different copackaged RNAs. Repeat deletion appears to occur primarily through intramolecular template switching, at least for gamma retroviruses (9, 43). In contrast, recombination between different strains of HIV-1 is an intermolecular event, and our assays did not address parameters that differ between inter- and intramolecular template switching. These parameters include the physical interaction which has been implicated in promoting recombination at the region flanking the dimerization initiation sequence (3, 16, 20, 21). Indirect evidence from melting profiles of nicked genomic RNAs suggests that extensive RNA-RNA interactions exist at other regions of the RNA genome as well (24), and these may also contribute to intermolecular recombination.

Another way our approach differed from intermolecular recombination is that tandem repeats provide a total of four possible templates for the assayed region (two on each copackaged RNA) instead of the two present for intermolecular recombination. Thus, in our system, competitive effects of the homologous repeat on the copackaged RNA may have affected deletion rates. However, since such competition exists for both nonidentical and identical repeat vectors, the observed decreases in deletion clearly reflect effects of decreasing homology on template switching, though they may misrepresent the magnitude of such effects during intermolecular switching.

Not surprisingly, our results suggest that regions with high sequence similarity are more likely to serve as HIV-1 recombination substrates than sequences that differ more significantly. However, as evidenced by the natural occurrence of M/O recombinants, whose parent strains differ by an average of roughly 35%, recombination between highly divergent sequences can take place, even though scattered 27% sequence differences reduced recombination frequencies greater than 300-fold in our assays. Selective pressures lead to differences in extents of sequence variation among isolates in different portions of the HIV-1 genome, with some regions more conserved than others, and it is possible that local regions of similarity serve disproportionately as recombination crossover sites. However, due to the very high replication rate of HIV-1, even recombination products that arise at frequencies no higher than those between unrelated sequences likely contribute to viral populations if they confer a selective advantage (4).

This work was supported by National Institutes of Health grant GM64479.