Cells.

293T is a human embryonic kidney cell line that expresses adenoviral E1A and simian virus 40 (SV40) large T antigen. 293T cells were grown from a master cell bank (40) and maintained in Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum. A 293T cell clone containing a single integrated retroviral DNA was produced by transducing 293T cells with the retroviral vector described below, selecting clones in puromycin, and screening colonies for high lacZ expression.

Plasmids.

pHEF-VSVG (6) is a vesicular stomatitis virus (VSV) G protein expression construct that was obtained through the AIDS Research and Reference Reagent Program, Division of AIDS, NIAID, NIH, from Lung-Ji Chang. pCMVΔR8.2 (22) encodes all HIV-1 proteins except the envelope glycoproteins and Vpu.

p_CMVLac_SV40Puro, which encodes an HIV-1-based vector that expresses lacZ from the cytomegalovirus (CMV) immediate-early promoter and puromycin resistance from an SV40 promoter, was previously called pHIVlac (1). The SV40 promoter in p_CMVLac_SV40Puro was replaced with the Rous sarcoma virus (RSV) promoter (nucleotides 483 to 806 from pREP8; Invitrogen) to produce p_CMVLac_RSVPuro. To generate p_{CMV S/D-} Lac_RSVPuro, the CMV promoter in p_CMVLac_RSVPuro was altered by site-directed mutagenesis so that the putative splice donor site (ACGGTAAAT) was changed to TTCCGTCTC.

A minimal HIV-based “empty” vector, pEmpty, was produced from p_CMVLac_SV40Puro by replacing the internal promoters and genes plus downstream long terminal repeat (LTR) with an altered 3′-LTR. The 3′-LTR fragment was produced by overlapping PCR using the primers 5′-CCATCGATAAGACAAGATATCCTTGATCTGTGGA-3′, 5′-GCGGAAAGTCCCTTGTAGCA-3′, 5′-AGGGACTTTCCGCACTAGTTCTGGTTAGACCAGATCTGAG-3′, and 5′-CTGCAGACTTGAAGCACTCAAGGCAAGCT-3′. The resulting PCR fragment was digested with ClaI plus PstI and used to replace the corresponding fragment in p_CMVLac_SV40Puro. As a result, the 3′ end of the RNA encoded by pEmpty lacked the polypurine tract and most of the 3′ LTR, including R sequences required to accept minus-strand transfer products. An additional retroviral vector, called p_CMVLac_SV40, was produced from p_CMVLac_SV40Puro by replacing the puromycin resistance gene and downstream LTR with the altered 3′-LTR described above.

Virus production and infection.

293T cells were transfected with expression clones by calcium phosphate precipitation (40) to produce replication-defective HIV particles bearing VSV G envelope protein. Virus-containing medium was harvested from 293T cells at 48 and 72 hours posttransfection, passed through 0.2-μm filters, and stored at −70°C.

Fresh 293T cells were infected with retroviral vectors and selected in 1 μg/ml puromycin at 48 hours postinfection. Twelve days later, puromycin-resistant colonies were fixed and stained with X-Gal (5-bromo-4-chloro-3-indolyl-β-d-galactopyranoside) for β-galactosidase activity as described previously (3). Viral vectors were produced for serial passage by transfecting pools of puromycin-resistant clones with helper plasmids (pHEF-VSVG and pCMVΔR8.2), and the resulting virions were harvested for further infections at 2 days posttransfection. To infect hydroxyurea (HU)-treated cells, 293T cells were pretreated with the indicated concentrations of HU (Sigma-Aldrich, St. Louis, MO) for 2 hours and the medium was replaced with medium containing fresh HU and retroviral particles. At 12 hours postinfection, the medium was replaced with medium containing 200 μM zidovudine (AZT), and colonies were selected in puromycin and stained with X-Gal as described above. Where indicated, cells were exposed to 200 μM AZT at 1.5 h postinfection to limit the time available to complete reverse transcription.

Integrated provirus structure analysis.

Individual puromycin-resistant colonies were expanded and divided to generate duplicate samples of cloned cells for X-Gal staining and for DNA preparation. Total cell DNA was prepared using the DNeasy tissue kit (Qiagen, Valencia, CA). Integrated HIV vectors were amplified by PCR using the following primer pairs (numbering refers to coordinates of the _CMVLac_SV40Puro proviral sequence and F or R to forward or reverse primers, respectively): 205F, CGACAGGCCCGAAGGAATA; 955R, GTAAAACGACGGGATCTAGCATG; 931F, TCCATGCTAGATCCCGTCG; 1683R, TAGGTAGTCACGCAACTCGCC; 1563F, GCTGCATAAACCGACTACACAAA; 2313R, GTGGCCTGATTCATTCCCC; 2259F, CGATCGTAATCACCCGAGTGT; 3014R, ACTGTGAGCCAGAGTTGCCC; 2784F, ACACCAGCAGCAGTTTTTCCA; 3539R, GCCACTTCAACATCAACGGTAAT; 5221F, GGATTGATGGTAGTGGTCAAATG; 6075R, CCTCACTACTTCTGGAATAGCTCAG; 6348F, GGAGAGCGTCGAAGCG; and 6623R, TAGAAGGGGAGGTTGCGG. Primers were removed from the amplified fragments using the QIAquick PCR purification kit (Qiagen, Valencia, CA) or by agarose gel electrophoresis and purification using the QIAquick gel extraction kit (Qiagen, Valencia, CA). All puromycin-resistant cell clones yielded puro-specific PCR products. When lack of another PCR product indicated a proviral deletion, alternate primer pairs were used to generate deletion junction-containing fragments. The nucleotide sequences of PCR fragments were determined by the University of Michigan DNA Sequencing Core Facility.

Establishing baseline replication error rates with lacZ-puro vectors.

The HIV-1-based vectors used for this study were expressed from the plasmid p_CMVLac_SV40Puro, which contains lacZ driven by a CMV promoter and the puromycin resistance gene expressed from an SV40 promoter, or from its derivatives (Fig. 1). Coexpression of p_CMVLac_SV40Puro with HIV-1 helper function plasmids in 293T cells produced viral particles that each contained two copies of the _CMVLac_SV40Puro vector RNA. These particles were used to infect fresh 293T cells, the infected cells resulting from a single round of reverse transcription and integration were selected in medium containing puromycin, and proviral lacZ mutation frequencies (measured by the loss of functional β-galactosidase) were determined by white-per-total (blue plus white) puromycin-resistant colony counts observed upon X-Gal staining.

Such screens for loss of phenotype are commonly used to monitor retroviral mutation frequencies. However, nonstaining white provirus-containing colonies could result either from lacZ inactivating mutations or from other factors. For example, because gene expression levels vary by integration site, some white colonies may result if integration site position effects limit lacZ expression to below levels detectable by X-Gal staining.

To calibrate the assay used here and address the relative contributions of lacZ inactivation and of positional effects to total white-colony formation, a population of _CMVLac_SV40Puro vectors was passaged serially for five sequential single rounds of replication in 293T cells (Fig. 2). This was achieved as follows. After one round of vector reverse transcription and integration, the population of infected cells was selected in puromycin. One plate of puromycin-resistant cells was X-Gal stained to generate the white-colony percentage data presented in Fig. 2, while the colonies on a duplicate plate were pooled. The integrated viral vectors in this infected cell pool were then remobilized by transfection with helper function plasmids. To perform a subsequent round of reverse transcription, the resulting viral particles were used to infect duplicate plates of fresh cells: one to score for X-Gal staining and the other as a source of integrated vectors for a subsequent round of remobilization.

The percentage of colonies that were white due to position effects was then estimated by plotting white-colony proportions over sequential viral generations. Total white-colony frequencies were assumed to reflect the sum of white colonies due to integration position effects plus those due to mutations arising in lacZ during viral replication. The percentage that are white due to position effects should remain fairly constant from one viral generation to the next, while those that are white due to lacZ mutation should increase with each round of reverse transcription. Thus, the percentage of white colonies not accounted for by lacZ mutations can be predicted by extrapolating the data to the y intercept to represent the number of white colonies present before viral replication (zero passages).

The results showed that serial passage of _CMVLac_SV40Puro resulted in a nearly linear increase in white colonies over sequential rounds of viral replication (Fig. 2). Approximately 10% of all puromycin-resistant colonies showed no detectable β-galactosidase activity after one round of infection, with this number increasing to 35% white colonies following five rounds of reverse transcription. These data suggest that _CMVLac_SV40Puro vectors accumulated lacZ-inactivating mutations at a rate of approximately 6% per replication cycle (linear slope), with a frequency of white colonies due to positional effects of about 3% (y intercept). Note that relatively few products of reverse transcription conditions where this background appears significant (e.g., pseudodiploid genomes from untreated cells and single replication cycles) were characterized in this study, and whether or not detected point mutations inactivated lacZ was not rigorously explored. Thus, the sequencing analyses described below lack the power to confirm this deduced position effect value.

Comparing defective product frequencies with one or two copies of lacZ template RNA.

To test the hypothesis that template switching contributes to replication fidelity, a skewed vector expression approach was used to produce viral particles containing predominantly single copies of lacZ (26). When two different HIV-1-based RNAs are coexpressed in individual virus-producing cells, the vector RNAs segregate into virions at random, generating encapsidated RNA homo- and heterodimers in proportions predictable by the Hardy-Weinberg equation (10). Thus, experimentally, RNA dimer proportions in HIV-1 populations can be controlled by altering vector coexpression ratios.

Here, generating haploid virions containing a single RNA copy of lacZ involved coexpressing p_CMVLac_SV40Puro with a molar excess of an additional vector plasmid called pEmpty (Fig. 1). pEmpty was produced by deleting lacZ, the puromycin resistance gene, and their respective promoters from p_CMVLac_SV40Puro, as well as ablating sequences needed for reverse transcription from the downstream LTR. As a result, pEmpty generated a short RNA vector that could heterodimerize with _CMVLac_SV40Puro RNA to facilitate its packaging but that could not contribute, either directly or via homologous recombination, to the production of puromycin resistance-conferring proviruses. RNase protection assays showed that vectors produced by transfection with a 20-fold molar excess of pEmpty compared to p_CMVLac_SV40Puro contained an average of 25-fold more Empty than _CMVLac_SV40Puro RNA copies, thus confirming that pEmpty was efficiently expressed and encapsidated (data not shown). In addition, the molecular analysis of _CMVLac_SV40Puro proviral products of heterozygous virions confirmed that recombination with Empty RNAs was not detected in puromycin-resistant proviruses (see below).

In the skewed vector approach used here, producer cells were transfected with a 20-fold molar excess of pEmpty over p_CMVLac_SV40Puro, the resulting virions were harvested, and fresh cells were infected and selected in puromycin. The puromycin resistance colony-forming titer that was observed under these conditions was reduced 100-fold relative to that for virions produced without pEmpty cotransfection. From a 20:1 cotransfection ratio, virions containing _CMVLac_SV40Puro homodimers, _CMVLac_SV40Puro/Empty heterodimers, or Empty RNA homodimers are predicted to arise at 0.2%, 9.1%, and 90.7%, respectively, according to the Hardy-Weinberg equation. Thus, much but not all of the observed titer reduction could be explained by predicted proportions of Empty vector homodimers. Of the predicted 9.3% of all particles that should be capable of producing puromycin resistance-conferring proviruses from a 20:1 transfection, 98% should contain a single copy of lacZ RNA.

Using this approach, the frequencies of lacZ-inactivating mutations arising during reverse transcription of virions with RNA genome populations comprised either of principally a single lacZ template or of two gene copies were compared (Fig. 3A). Initial results comparing white-colony titers for single-copy lacZ vectors to those for two lacZ templates revealed a modest, albeit consistent, twofold increase in apparent lacZ inactivation rates in the absence of a second lacZ gene copy (Fig. 3A [data from experiments with 0 mM HU]).

The defects associated with haploid genome reverse transcription were subsequently magnified by carrying out reverse transcription under mutagenic conditions. The above findings (Fig. 2) showing that integration position can contribute to white-colony proportions suggested that the apparent magnitude of haploid genome-associated defects might be masked by the presence of position effect products and thus would become more evident under conditions where this background was reduced relative to the total number of white colonies. This was achieved by increasing viral mutation rates by infecting cells that had been treated with the concentrations of HU indicated in Fig. 3A. HU inhibits ribonucleotide reductase and rapidly depletes intracellular deoxynucleoside triphosphate (dNTP) pools, resulting in increased retroviral mutation rates (30). Because HU treatment raises retroviral mutation rates but should not affect integration site-dependent white-colony background levels, we anticipated that HU treatment would increase the “signal-to-noise” ratio and allow a clearer assessment of effects on reverse transcription. Results from infections of HU-treated cells revealed that reverse transcription products of virions containing two copies of lacZ contained significantly fewer lacZ-inactivating mutations than those of virions bearing only a single lacZ template (Fig. 3A). Factoring in 3% as the integration site-dependent background white-colony frequency, lacZ inactivation was more than threefold more frequent when a single copy of lacZ was reverse transcribed than when proviruses were templated by homodimeric genomes.

Confirming that lacZ gene copy number, and not RNA heterodimerization, affects white-colony product frequency.

An additional test was performed to ensure that the skewed RNA expression approach used to generate vectors with a single lacZ template did not cause the observed high lacZ inactivation frequency. For this approach, a new vector called p_CMVLac_SV40 was created, in which the intact lacZ gene was retained but from which the puromycin resistance coding sequences of p_CMVLac_SV40Puro were deleted (Fig. 1). Most virions capable of conferring puromycin resistance that were generated in the presence of excess _CMVLac_SV40 RNA (which replaces Empty RNA in the experiments here) and limiting _CMVLac_SV40Puro RNA would be predicted to contain RNA heterodimers with two copies of lacZ but only one puro gene copy. Fresh cells infected with virions produced under these conditions displayed low puromycin resistance titers similar to those of _CMVLac_SV40Puro/Empty heterodimers. However, as shown in Fig. 3B, in either the presence or absence of HU, the lacZ inactivation rate among products of _CMVLac_SV40Puro/ _CMVLac_SV40 heterodimeric virions was indistinguishable from rates with _CMVLac_SV40Puro homodimers. These results contrasted with the marked increase in white colonies observed for _CMVLac_SV40Puro/Empty heterodimer products and support the notion that the higher white colony frequency observed during infection with single-lacZ-copy virions was primarily dependent on the lacZ copy number.

Addressing contributions of packaged subgenomic RNAs to haploid genome-associated replication defects.

Extracting proviral sequences by PCR from infected cell clones revealed that one 4-kilobase deletion arose at a high frequency (Fig. 4A). This common deletion, which was observed in nearly 60% of all analyzed _CMVLac_SV40Puro white proviruses, was generated by haploid and by pseudodiploid RNA genomes and whether or not cells were treated with HU.

Analysis of this product's deletion endpoints revealed strong splice site consensus sequence similarity (37), suggesting that the use of cryptic splice sites had generated spliced RNAs that were encapsidated within the virion population and served as templates for these deletion products (Fig. 4A). To provide confirmatory evidence for the likelihood that the putative splice donor and acceptor sites were involved in deletion product formation, additional vectors, called p_CMVLac_RSVPuro or p_{CMV S/D−} Lac_RSVPuro, were designed that lacked one or both of the splice junctions, respectively (Fig. 1). The potential 3′ splice signal (designated S/A for “splice acceptor” in Fig. 4A) was removed by replacing the SV40 promoter with the RSV promoter to generate p_CMVLac_RSVPuro, and the 5′ splice site (S/D in Fig. 4A) was altered by site-directed mutagenesis, yielding p_{CMV S/D−} Lac_RSVPuro.

These new vectors were then tested to address whether or not mutating these putative splice junctions would eliminate haploid genome-associated increases in lacZ inactivation rates. Each of the three lacZ-puro vectors was expressed either alone or in the presence of excess pEmpty, and the resulting virions were used to direct one round of replication in either untreated or HU-treated cells (Fig. 4B). The results revealed that all three lacZ-puro vectors, when reverse transcribed from haploid genomes produced by transfection with an excess of pEmpty, exhibited higher white-colony frequencies than the corresponding products from homodimeric virions (Fig. 4B). As had been observed for _CMVLac_SV40Puro, increases in white-colony formation were more pronounced in the presence of HU for both the vector derivatives (right half of Fig. 4B). These results confirmed the correlation between lacZ RNA template copy number and proviral error rates and indicated that the high-frequency putative spliced RNA products of _CMVLac_SV40Puro were not solely responsible for the higher white-colony rates observed with single-copy lacZ vectors. Thus, the property of enhanced defective genome synthesis correlated with imbalanced gene copy number, even in the absence of the previous major product.

Assessing the spectra of lacZ-inactivating mutations.

The lacZ genes from many individual β-galactosidase-negative proviruses (each templated by one of the three lacZ-puro vectors, expressed either alone or with excess pEmpty, and reverse transcribed in the presence or absence of HU) were then isolated, mapped, and sequenced (Table 1). Of the total of 82 such white-colony provirus structures summarized in Table 1, 53 contained large deletions in lacZ and 29 differed from the parental sequence only by the possession of one or more small mutations, including point mutations and frameshifts or small deletions revealed only by DNA sequencing.

As introduced above, initial work with p_CMVLac_SV40Puro products revealed that 23 out of 26 large deletions shared the same 5′ and 3′ deletion junctions (S/D1 and S/A5 in Fig. 5A). As predicted by the assumption that these common deletions were products of spliced subgenomic RNAs, analysis of p_CMVLac_RSVPuro products (which were designed to lack S/A5 [Fig. 5A]) revealed the use of the same 5′ junction that was observed for the original _CMVLac_SV40Puro vector (S/D1) but joining to different cryptic 3′ splice sites (S/A1 to -4). In contrast, all these putative spliced RNA products were absent among proviruses generated by p_{CMV S/D−} Lac_RSVPuro (Table 1), which lacked both S/D1 and S/A5. However, a second putative splice donor (S/D2 at vector map position 4673) was revealed by the structures of two additional proviral clones derived from p_CMV Lac_RSVPuro. Both of these deletion products resulted from use of the same potential splice acceptor site (S/A4 in Fig. 5A).

The deletion junctions of three deletion-containing products of p_CMVLac_SV40Puro, about half of those from p_CMVLac_RSVPuro, all characterized clones with deletions generated from p_{CMV S/D-}Lac_RSVPuro, and many early-time-point products (described below) did not display features suggestive of cryptic RNA splicing but instead displayed structures typical of retroviral nonhomologous recombination-generated deletions (Fig. 5B). Most of these products contained 1 to 11 nucleotides of microhomology at the deletion junctions, one clone showed no deletion junction-associated homology (clone 963), and one product contained an 11-base insertion between the deletion junctions (clone 546). Junctions with microhomology, no homology, and insertions within deletions are all common nonhomologous recombination products (9).

Taken together, these data indicate that most if not all of the deletions observed in lacZ-puro vector products can be explained either by the ability of reverse transcriptase to use spliced subgenomic RNA templates or by viral nonhomologous recombination.

In contrast to the deletion derivatives characterized above, 29 of the 82 “white-colony” proviruses did not contain length variation but instead were revealed by sequencing to differ from the parental sequence by point mutations in lacZ. Although point mutations appeared to contribute to lacZ inactivation less frequently than deletions, some clones contained multiple changes, with as many as 10 point mutations in a single mutant lacZ gene (data not shown). Interestingly, 27 of the total of 51 observed nucleotide substitutions were G-to-A transitions.

Although the numbers of sequences considered here limit the statistical power of these interpretations, when all products of the three vectors were considered together, the following trends emerged. Of 12 clones generated by pseudodiploid virions in the absence of HU, 8 were inactivated by point mutations (Table 1, rows 1, 5, and 9), while 4 contained deletions. Among 14 proviruses templated by homodimerized RNAs in cells treated with HU, 8 contained nucleotide substitutions and 6 contained deletions (Table 1, rows 2, 6, and 10). Thus, the defective reverse transcription products generated by vectors containing two lacZ templates included approximately equal proportions of alleles with lacZ-inactivating point mutations and with deletions.

In contrast, clones resulting from infection with single-gene-copy lacZ vectors predominantly contained deletions (Table 1, rows 3, 4, 7, 8, 11, and 12). Thirteen analyzed clones generated by haploid vectors in untreated cells contained deletions, while only three were disrupted by point mutations. Among 41 analyzed products of haploid vectors reverse transcribed in HU-treated cells, 31 contained deletions and only 10 contained point mutations (Table 1, rows 4, 8, and 12).

Assessing underrepresentation of intact genome products at early time points during reverse transcription.

Cataloging of defective reverse transcription products (Table 1 and discussed above) showed that the spliced-RNA-templated 4-kb _CMVLac_SV40Puro deletion was observed among products generated with or without HU treatment and among products of both haploid and pseudodiploid homozygous virions. This suggested that the spliced RNA that templated the deletion product was constitutively present in virion RNA populations. These observations also raised the possibility that the reason the spliced RNA's proviral form was overrepresented among haploid genome products was due to disadvantages of full-length genome synthesis in the absence of a second RNA gene copy.

Some of the increased proportions of lacZ inactivation observed in the presence of HU (Fig. 3) may reflect a selective advantage of deletion products, which could be completed faster than full-length lacZ vectors. The half-time to completion of HIV DNA synthesis is about 4 h under optimal intracellular conditions (24), and reverse transcription in the presence of HU takes even longer due to limiting substrate availability (15, 23, 30). Here, single-copy lacZ gene vectors were more prone to generating lacZ inactivation products than were homozygous genomes, and many defective products contained deletions. Thus, we hypothesized that part of the reason that virions containing a single copy of the assayed lacZ gene might be less capable of generating full-length DNAs than were homozygous genomes was that reverse transcriptase elongation for single-gene-copy genomes proceeded less efficiently than when a second copy was present.

To test this hypothesis, proviruses generated during limited durations of DNA synthesis were examined. This approach addressed whether or not the same spectra of deletion products that were enriched by skewed vector ratios also were overrepresented under conditions known to disfavor full-length provirus synthesis. Reverse transcription was disrupted prematurely by treating cells with inhibitory concentrations of AZT at 1.5 h after initiation of infection (Fig. 6). This resulted in roughly 70-fold decreases in proviral product yields, as determined by puromycin resistance titers (not shown). Note that no HU was used in these experiments.

Consistent with the HU treatment results, haploid vectors had higher lacZ inactivation rates than pseudodiploid vectors when the time of reverse transcription was limited (right half of Fig. 6). Structural analysis of defective products of pseudodiploid particles revealed that the same common 5′ junction observed above among skewed vector products was also predominant among clones derived from limited reverse transcription duration (1.5 h), representing 9 of 10 deletions and 2 of 6 deletions for p_CMVLac_SV40Puro and p_CMVLac_RSVPuro, respectively (Table 2). Thus, the same spliced-RNA-templated deletion products became overrepresented both by limiting durations of reverse transcription and by infection with haploid virions. These observations demonstrate that products of late-time-point synthesis for haploid vectors resemble those of much earlier time points for pseudodiploid genomes, thus suggesting that the presence of two intact viral genomes is more favorable for full-length viral DNA synthesis.