INTRODUCTION

Top Abstract Introduction Materials and Methods Results Discussion References

The survival of species depends on a favorable balance between the beneficial and harmful aspects of mutation. There is loss of the information content of nucleotide sequences once mutation rates are increased beyond a tolerable error threshold (15). Some RNA viruses seem to tolerate mutation rates near this threshold, existing, not as a specific sequence, but as a quasispecies (12, 15, 26). When viral mutation rates approach the maximum value compatible with viability (14), what results is an intolerance to even small increases in mutation rates (27). How surprising, then, that for some of the same viruses with the highest mutation rates, another mutation process has been described in which remarkable levels of one specific type of nucleotide substitution are observed. These sequences are referred to as hypermutants and are a product of one specific mutation at rates far beyond viability. Two main types of hypermutation, which differ with respect to the type of substitution observed, have been described for viral sequences. A-to-G hypermutation occurs mostly in measles virus and vesicular stomatitis virus (8, 9, 49), and three cases in nonlentiviral retroviruses (avian leukosis virus and spleen necrosis virus) have been reported (16, 23, 33). G-to-A hypermutation is found primarily in the lentivirus family of retroviruses, along with two other examples in satellite tobacco mosaic virus (37) and hepatitis B virus (22). G-to-A hypermutation in the lentivirus human immunodeficiency virus type 1 (HIV-1) is the subject of this report.

G-to-A hypermutation is defined as a mutational process in which G-to-A transitions far exceed all other mutations in viral sequences (60). In published reports, up to 60% of G's may be replaced by A's along a hypermutated sequence (64). G-to-A hypermutation occurs specifically within the GpA or GpG dinucleotide context (18, 60). Although G-to-A hypermutation was first described for spleen necrosis virus (SNV) by Pathak and Temin (50), G-to-A hypermutation has been mostly found among the lentiviral group of retroviruses, including HIV-1 (4, 11, 18, 21, 39, 40, 47, 60), HIV-2 (19), simian immunodeficiency virus (SIV) (30), equine infectious anemia virus (EIAV) (52), and caprine arthritis-encephalitis virus (CAEV) (64).

The susceptibility of lentiviruses to hypermutation is thought to be a property of their reverse transcriptase (RT), which, compared to nonlentiviral RTs, has a great capacity to elongate beyond nucleotide mismatches (51) and an increased ability to hypermutate in the presence of unbalanced nucleotide pools in vitro (43). This susceptibility is further evidenced by the elevated A content of lentivirus genomes (7). Retroviral genomes form a bimodal distribution with respect to base composition that follows taxonomic groups (71).

Hypermutation in HIV-1 was first detected during propagation of HIV-1 virus populations in vitro (11, 21, 60). Since then, several groups have recovered hypermutated HIV-1 sequences from clinical samples, confirming that they occur in vivo (4, 18, 40, 47). Typically, mixtures of hypermutated and normal sequences are found, with hypermutated sequences in the minority (4, 18, 19, 30, 39, 52, 67). With respect to genome region, hypermutation was initially found in small subregions of gag, env, nef, and U3/R elements of the long terminal repeat (LTR) (11, 21, 60), but others have found hypermutation throughout the genome (4).

Some progress has been made in elucidating the biochemical basis for hypermutation. First thought to be the work of a mutant RT (18, 50), hypermutation has now been shown to be a property of wild-type RT (42) operating under suboptimal conditions. G-to-A hypermutation is thought to occur when HIV minus-strand DNA synthesis takes place simultaneously with an increased intracellular concentration of dTTP relative to dCTP (60). Attempts have been made to generate hypermutants in a cell-free system in vitro with RNA, purified RT, and strongly biased deoxynucleoside triphosphate (dNTP) pools (43, 44, 59, 61). While G-to-A hypermutants were produced, they lacked the marked preference for the GpA and GpG dinucleotide contexts that typify hypermutants produced in vivo or during virus cultivation in vitro. An intermediate approach where RT in virus particles was allowed to complete first-strand DNA synthesis in the presence of exogenously provided, highly biased dNTP pools also resulted in loss of the GpA and GpG dinucleotide preference (62). Hypermutants were also recovered at low frequency (0.5%) from cell cultures after the addition of deoxythymidine to supernatants, which resulted in an elevated intracellular [dTTP]/[dCTP] ratio (62), but because all possible dinucleotide contexts were not represented in the sequence that was studied, a complete analysis of context preference was not possible. The cell-free experiments described above were conducted using relative concentrations of dTTP and dCTP that differed by 3 to 4 orders of magnitude. More subtle perturbations in the intracellular environment may trigger dNTP pool biases sufficient to generate hypermutation while maintaining the proper dinucleotide context (67). Indeed, a 40- fold [dTTP]/[dCTP] ratio produced hypermutants in which the GpA context preference was maintained (62), and hypermutants were recovered from 1 to 2% of unstimulated and phytohemagglutinin (PHA)-stimulated peripheral blood mononuclear cells (PBMC) without manipulation of dNTP pools (62).

Imbalanced and fluctuating nucleotide pools are a key element of many types of mutation, including hypermutation. Normal dNTP pools are highly asymmetric in mammalian cells (reviewed in reference 45). dNTP pool biases are mutagenic during DNA replication (34-36, 46) and are able to affect RT error rates in vitro and retroviral mutation rates in vivo (2, 29, 32, 53). Since the composition of dNTP pools changes as cells progress through the cell cycle (5, 20, 45), the timing of HIV-1 infection with respect to T-cell activation and entry into the cell cycle could be an important factor determining the generation of hypermutants. [dTTP]/[dCTP] ratios at different stages of cell activation varied from 1.3:1 to 6:1 (3, 5, 10, 20, 58). In a retrovirus-based shuttle vector, G-to-A transitions predominated (31, 32) and were presumably a result of even these modest fluctuations in pools during the cell cycle.

At this point understanding of hypermutation is still limited, in part because of the lack of a successful systematic screening method. For example, a method based on the blue/white -galactosidase complementation assay yielded a low recovery of hypermutants (0.5 to 2%) (62), not higher than that obtained when clones were randomly picked and sequenced. Indeed, many hypermutated sequences have been submitted to databases without being perceived as such (63). Given the lack of a specific genome location for hypermutation, the fact that hypermutants are almost always buried in a large excess of normal sequences, and the difficulty of precisely reproducing hypermutation in vitro, it is not surprising that this mutational process has been regarded as erratic, rare, and of minor importance in the HIV-1 life cycle. Here we describe the design and application of powerful new methods for systematic detection and recovery of hypermutants and their application to clinical samples and to the products of HIV-1 infection in cell culture. The results call for a reassessment of the frequency of hypermutation in vivo, clarify the conditions that generate hypermutants in cell culture, and importantly, highlight a vulnerability of HIV-1 that could be exploited for clinical benefit.

	MATERIALS AND METHODS

Top Abstract Introduction Materials and Methods Results Discussion References

Clinical samples. Purified PBMC from 53 HIV-1-positive individuals were the source of DNA for analysis. Twenty-six samples were from patients hospitalized in Tanzania in 1996 with symptoms compatible with AIDS (25). Fourteen others were from individuals who seroconverted to HIV-1 while in the U.S. Military between 1997 and 1998 (6). Thirteen people were participants in the San Francisco Men's Health Study between 1985 and 1988 (38, 66). The latter two groups were in the early, asymptomatic stage of HIV-1 infection, and the sample was typically drawn within 6 months of HIV-1 seroconversion.

PCR. DNA extracted from patient PBMC or virus cultures was the template for PCR amplification of HIV-1 protease sequences. Nested PCR primers (28) were used either without modification, for amplification of normal sequences, or with modifications designed to increase their homology to hypermutated sequences. The primers for amplification of hypermutants contained either mixed bases (hyp primers) or G-to-A replacements (hypa primers) at some of the sites (GpA or GpG) susceptible to hypermutation. Each PCR amplification was run in duplicate, once with normal primers and a second time with an equal mixture of hyp and hypa primers. The first round of PCR was conducted with 2 mM dNTPs, 50 mM KCl, 10 mM Tris-HCl (pH 8.3), 2.5 mM MgCl₂, 0.4 µM each primer or primer mixture, and 1.25 U of Amplitaq.

The HA yellow gel system. PCR products were analyzed in parallel 1% agarose gels, with or without HA yellow, a compound which consists of the DNA ligand bisbenzamide covalently linked to polyethylene glycol (PEG) (Hanse Analytik, Bremen, Germany). Bisbenzamide binds preferentially to AT-rich regions in DNA (in order from highest to lowest preference, AATT and AAAA, TAAT, ATAT, TATA, and TTAA) (1) and, when coupled to PEG, retards DNA mobility during gel electrophoresis according to AT content (41, 48, 65). HA yellow was incorporated at 1 U/ml after the gel solution was cooled to 65°C. Electrophoresis was at 80 V in 1× Tris-borate-EDTA (TBE) for 90 min. Products were visualized by poststaining with ethidium bromide.

Isolation and molecular cloning of normal and hypermutated sequences. We attempted to recover both normal and hypermutated sequences from every sample, but different approaches were used depending on the HA yellow gel profile of the bulk PCR product. The PCR product was cloned directly if it was separated into normal and hypermutated bands of nearly equal intensity, or if there was only one band. If the hypermutated and normal bands were of unequal intensity, the two populations were excised from the gel, extracted using the Qiaquick Spin Purification Gel Extraction Kit from Qiagen (Valencia, Calif.), and cloned separately. When hypermutated bands were faint, the two populations were excised and extracted as above, but the hypermutated sequences were reamplified with second-round primers DP16 and DP17 prior to cloning.

DNA sequencing. Eight hundred nanograms of purified plasmids was digested using 20 U of EcoR1 (New England Biolabs, Beverly, Mass.) for 1 h at 37°C to release the cloned protease gene, which was then rescreened on agarose gels with HA yellow. At least one clone, and often several clones, representing all of the different mobilities found within each sample was selected for sequencing.

Determination of HIV-1 subtype. The genetic subtypes of normal protease sequences from the 53 patients were established by phylogenetic analysis (see Table 2). Sequences were manually aligned with reference sequences of HIV-1 subtypes A through J. Components of the PHYLIP package (17), including SEQBOOT, DNADIST, NEIGHBOR, and CONSENSE, were used to assign each sequence to a known subtype, if possible. All of the asymptomatic patients from the United States harbored subtype B. Among the AIDS patients more than half harbored subtype C; the remainder harbored mostly subtype A or D. One patient was dually infected with subtypes A and D, and one other harbored an unclassified strain, possibly an AD recombinant (data not shown).

Preparation and characterization of virus stocks. Virus stocks generated from an infectious molecular clone of HIV-1 subtype B isolate NL4-3 (56, 57) were used to infect PBMC from a seronegative donor. Cells were activated with 1 µg of PHA/ml for 72 h, infected for 24 h, washed twice, and maintained in RPMI 1640 with 15% fetal calf serum (FCS), 1% PenStrep (Quality Biological, Gaithersburg, Md.), 1% L-glutamine, and 20 U of interleukin-2 (IL-2)/ml. Culture supernatants were assayed for p24 antigen (HIV-1 p24 Antigen Assay Kit; Coulter Corporation, Miami, Fla.) at 3-day intervals. Culture supernatants collected near the peak of p24 antigen production were treated with 50 U of DNAse I/ml (Boehringer Mannheim) for 30 min at room temperature and filtered sterilized (pore size, 0.22 µm). Viral stocks of at least 20,000 50% tissue culture infective doses (TCID₅₀)/ml were stored in liquid nitrogen until use.

Cell culture. Fresh leucopacks were obtained from HIV-1-seronegative donors (RH Laboratories, Baltimore, Md.). PBMC were isolated by density gradient centrifugation over Ficoll-Hypaque (Pharmacia, Stockholm, Sweden). CD4⁺ T cells were purified by negative selection on magnetic beads with the MACS CD4⁺ T Cell Isolation Kit (Miltenyi Biotec, Auburn, Calif.). Prior to infection, an aliquot of the purified CD4⁺ T cells was fixed with 2% formaldehyde and stained with fluorescein isothiocyanate (FITC)- or phycoerythrin (PE)-conjugated monoclonal antibodies specific for cell surface molecules CD3, CD4, CD8, CD14, CD20, CD56, and HLA-DR (Becton Dickinson, San Jose, Calif.). Cell populations were analyzed by flow cytometry on a BD FACScan (Becton Dickinson). The majority of the cells in each experiment were CD4⁺ T cells (mean, 81%; range, 69 to 99%) representing most of the CD3⁺ T-cell subset (mean, 83%; range, 76 to 100%). The other major cell type was B cells (CD20⁺; mean, 10.8%; range, 4.5 to 14.9%). Monocytes (CD14⁺), CD8⁺ T cells (CD8⁺), and NK cells (CD56⁺) were, on average, 2 to 4% of the population. Between 14.6 and 17.7% of the cells were HLA-DR positive.

Sequence analysis. Sequences were examined for base composition, open reading frames, and the composition of the translated protein sequences with EDITSEQ (DNAStar, Madison, Wis.). Parameters of hypermutation were evaluated with the Hypermut Program Package (54) (http://www.hiv.lanl.gov/HYPERMUT/hypermut.html) as implemented at the Los Alamos HIV Sequence Database, Los Alamos, N. Mex. This program identifies mutations and their dinucleotide context with respect to a reference sequence that is provided with each alignment. For the patient samples, we used a sequence with the lowest percent A+T, a normal base composition, and an open reading frame as the normal reference for all sequences from that patient. For virus cultures, the sequence of NL4-3, obtained by RT-PCR and sequencing of the virus stock, was used as the reference.

Nucleotide sequence accession numbers. All the HIV-1 sequences related to this work have been submitted to GenBank and were given accession numbers A Y036228 through A Y036577.

	RESULTS

Top Abstract Introduction Materials and Methods Results Discussion References

Amplification, detection, and recovery of hypermutated DNA sequences. The replacement of G by A in hypermutated sequences increases their A+T content and decreases their homology with PCR primers based on normal sequences. Unless primer mismatching is addressed, the recovery of hypermutants will be severely limited. Once amplified, however, hypermutants should be easily detectable by their increased A+T content. Thus we addressed the issue of primer mismatching first. Table 1 shows the design of a PCR strategy specifically designed to permit more efficient recovery of hypermutated sequences. Nested primers which amplify a 297-bp segment encoding HIV-1 protease were examined for GA and GG dinucleotides, which are susceptible to hypermutation. New primers were designed with either incorporated GA or CT mixtures at some of these sites (hyp primers) or replaced G with A or C with T (hypa primers). The hyp and hypa primers were also shortened to bring potential mispairs closer to the 3' end of the primer, where they could exert the maximal effect in destabilizing interaction with normal sequences.

View larger version (70K): [in this window] [in a new window]   FIG. 1.   Calibration of the HA yellow gel system. HIV-1 protease sequences amplified from patient PBMC and representing a range of G-to-A hypermutations were used to explore the performance of 1% agarose gels containing HA yellow, a dye that preferentially binds to AT-rich regions in DNA. The alignment of the 297-bp sequences, amplified with primers DP16 and DP17 and ranging from normal (62% A+T) to maximally hypermutated (78% A+T) is shown at the top. The GA and GG dinucleotides that are susceptible to hypermutation are shaded. The gel on the left shows the migration of these PCR products as a single band at 297 bp without HA yellow. On the right, a gel incorporating HA yellow is shown, illustrating the direct relationship between AT content and mobility. Sequences with as little as 10% G-to-A substitution (2nd lane) migrated differently than normal sequences. PCR products representing the normal and maximally hypermutated sequences were used as migration standards in subsequent HA yellow gels.

View larger version (44K): [in this window] [in a new window]   FIG. 2.   Relative recovery of hypermutated sequences with normal and hypermutated PCR primers. Primers were designed to incorporate G-to-A substitutions in GA or GG dinucleotides (Materials and Methods) and compared to primers matched to normal sequences for their ability to amplify hypermutated sequences from the PBMC of 10 patients. A side-by side comparison of the PCR products using HA yellow gels is shown. N, normal primers; H, primers hyp and hypa. Hypermutated bands detected with primers hyp and hypa but not with normal primers are indicated for patients 3, 4, 5, 6, and 8 (asterisks). The positions of normal and fully hypermutated standards are indicated by arrows.

Hypermutation in patient PBMC. HIV-1 proviral DNA sequences from 53 HIV-1-positive patients were investigated (Table 2). The appearance of protease genes PCR amplified from patient PBMC is illustrated in Fig. 3. While all of the protease genes were of uniform length without HA yellow, they often separated into multiple bands of different A+T content on HA yellow gels. Sometimes all of the PCR product migrated at the normal position (Fig. 3, right, lane 1), but other PCR products split into multiple bands (lanes 2 to 7, 9, and 10). Notably, some samples had most of the PCR product retarded in the gel (lanes 2, 6, and 10). Other samples migrated mostly at the normal position but were smeared upward without distinct bands (Fig. 3, right, lane 8).

View larger version (44K): [in this window] [in a new window]   FIG. 3.   Abundance and variety of hypermutated PCR products amplified from patient PBMC. DNA extracted from patient PBMC was amplified in a touchdown, nested PCR with primers hyp and hypa as described in Materials and Methods. PCR products were compared to normal and hypermutated standards on HA yellow gels. While all products migrated as a single band at 297 bp without HA yellow (left), they were often split into several bands in the presence of HA yellow (right). Products were visualized by staining with ethidium bromide.

View larger version (26K): [in this window] [in a new window]   FIG. 4.   Classification of sequences from patient PBMC. HIV-1 protease genes were amplified using DNA extracted from the PBMC of 53 patients ranging from the early, asymptomatic stage of HIV-1 infection to late-stage AIDS, and including infections with HIV-1 subtypes A through D (Table 2). Products PCR amplified with primers hyp and hypa were molecularly cloned and sequenced. In total, 287 sequences were obtained, of which 219 were unique. Among the 219 sequences, the percent A+T distribution is shown in the top graph. Using the sequence with the lowest percent A+T from each patient as a normal reference, the percent G in the GA or GG context that was mutated to A was calculated for each sequence, and the distribution of percent G-to-A mutation was examined (bottom graph). Based on the bimodal distribution of these parameters, 149 sequences were classified as normal (less than or equal to 66% A+T and less than 20% G-to-A mutation) and 70 were classified as hypermutated (67% A+T and more than 21% G-to-A mutation).

Parameters of hypermutation. Seventy hypermutated sequences were recovered directly from patient PBMC, together with normal reference sequences from each patient, by a specific and consistent method. This permitted, for the first time, a systematic evaluation of the parameters of HIV-1 hypermutation in vivo. The specificity of the hypermutations was extraordinary (Fig. 5, top). The sequences showed elevated A content and reduced G content compared to normal sequences, while their C and T content was completely unaltered. In the 23 patients from whom both normal and hypermutated sequences were recovered, the normal sequences had the same base composition as those of the 30 patients whose recovered sequences were entirely normal (data not shown).

View larger version (49K): [in this window] [in a new window]   FIG. 5.   Parameters of hypermutation in sequences from patient PBMC. One hundred forty-nine normal and 70 hypermutated sequences from patient PBMC were examined for base composition (top). The normal sequences had a narrow distribution of percent G and percent A, centered on 22 and 37%, respectively. Hypermutated sequences showed a broad range of G content from 5 to 20% and a range of A content from 40 to 54%. The distribution of percent C and percent T was the same in normal and hypermutated sequences. Hypermutated sequences were arranged in order of increasing hypermutation (bottom two graphs). While G-to-A mutations in the GA or GG context increased 10-fold, other mutations (G to A in the GC and GT context and all other mutations) showed no discernible increase. The percentages of available G replaced by A in the GA and GG contexts were calculated separately and compared over the range of G-to-A substitution (bottom). More of the G residues in the GA context than in the GG context were replaced by A at all levels of hypermutation.

Loss of informational content in hypermutated sequences. Hypermutated and normal DNA sequences were translated into protein sequences and evaluated for the integrity of the genetic information. Virtually all of the hypermutated sequences from patient PBMC had lost their protein-coding potential (Table 3); 69 of 70 had an in-frame stop codon, compared to 3 of 149 normal sequences. In addition, there was a net accumulation of positively charged amino acids, mostly from decreased Gly, Asp, and Arg and increased Lys and Asn in hypermutants. The proteases encoded by hypermutated sequences had, on average, a fivefold increase in net charge (Table 3). It is noteworthy that hypermutated sequences with the lowest levels of G-to-A substitutions sustained inactivating in-frame stop codons and many nonsynonymous nucleotide substitutions. These combined effects make it highly unlikely that functional proteases, and, consequently, viable viruses can be derived from hypermutated sequences.

Relationship of hypermutation to stage of disease and genetic subtype. The relationships between hypermutation, clinical stage, and HIV-1 subtype were investigated (Table 2). Hypermutated sequences were recovered at essentially the same rate from AIDS patients (42%) and early asymptomatic individuals (44%). Hypermutation was found in 57% of subtype A, 67% of subtype D, and 44% of subtype B infections, as well as in A+D dually infected and unclassified samples. Although subtype C was the most frequently represented subtype among AIDS patients, hypermutation was detected only in 3 of 14 (21%) patients infected with this subtype. Subtype C hypermutated sequences also tended to have lower levels of G-to-A mutation than the other subtypes, but in this small sample the differences did not reach statistical significance (data not shown).

Investigation of hypermutation in cell culture. The link between unbalanced dNTP pools and hypermutation in vivo is probable but not yet conclusive. Experiments with cell-free systems have shown that unbalances in dNTP pools during reverse transcription can lead to hypermutation, but the specificity for the GA and GG dinucleotide context has been difficult to reproduce. On the other hand, hypermutants in the correct dinucleotide context have been recovered from HIV-1 cultures in PBMC, but the exact nature of the cell population that produced them and the status of the dNTP pools in this population have not been elucidated. In CD4⁺ T cells, which are a main target of HIV 1 replication in vivo, dNTP pools are elevated when the cells are activated and in cycle (20) but are unbalanced and fluctuating in resting T cells and during the transition from a resting to an activated state (5, 10, 20, 58). We designed an experiment in which the timing of HIV-1 infection was varied with respect to activation of resting CD4⁺ T cells by PHA; the outcome measure was recovery of hypermutated sequences whose characteristics corresponded exactly to those found in vivo.

View larger version (37K): [in this window] [in a new window]   FIG. 6.   Relationship of hypermutation to T-cell activation in virus cultures. CD4+ T cells isolated from the PBMC of normal donors were infected with an NL4-3 virus stock for 2 h at a multiplicity of infection of 0.02 (stippled gray bar) under varying conditions of PHA activation (gray bars). Condition I, infection without addition of PHA; condition II, simultaneous infection and PHA addition; conditions III to VIII, infection with prior PHA activation for 3 to 72 h respectively; condition IX, infection and addition of PHA 24 h later. Cultures were sampled at intervals and assayed for the presence of p24 antigen in the culture supernatant (solid lines) and for the presence of hypermutated HIV-1 DNA sequences. Filled stars, samples where hypermutated sequences were abundant; open stars, samples with rare hypermutated sequences. The appearance of PCR products on HA yellow gels is shown on the right. Time points are labeled "a" through "n" at the tops of the gels and in the diagram. Samples a, b, c, d, e, g, and h are bulk PCR products. Samples from time points at which hypermutation was rare (f, i, j, k, and l) are shown both as bulk PCR products (B) and after pre-enrichment by ScrF1 and AvaII digestion (A). Time points not marked by stars or asterisks yielded only normal sequences both before and after enrichment (not shown).

View larger version (48K): [in this window] [in a new window]   FIG. 7.   Parameters of hypermutation in virus cultures. A total of 227 HIV-1 protease sequences, 131 of which were unique, were obtained from virus cultures. The unique sequences were compared for their distribution of percent A+T (A). Seventy-three were normal and 58 were hypermutated. (B) Distribution of G-to-A mutation in sequences from different culture conditions. Roman numerals refer to the culture conditions described in the legend to Fig. 6. Filled symbols, normal sequences; open symbols, hypermutated sequences. The cultures where abundant hypermutants covering the full range of G-to-A replacement were found are shaded. Below, the average ratio of percent G-to-A substitutions in the GA versus GG context is plotted for the sequences (all except six) in which both contexts were used. (C) Base composition of normal and hypermutated sequences from virus cultures. (D) Specificity for G-to-A substitutions in the GA or GG context in contrast to all other mutations, and the preferential use of GA over GG over the range of hypermutation.

	DISCUSSION

Top Abstract Introduction Materials and Methods Results Discussion References

This study represents the first systematic evaluation of patient PBMC and virus cultures with a consistently applied methodology optimized for detection of hypermutated HIV-1 sequences. Hypermutated sequences could be detected in 43% of patients and in virus cultures when T-cell stimulation was initiated at the time of, or shortly before, virus infection. The nearly identical quantitative parameters of the hypermutants recovered from patients and from virus cultures suggest that T cells in the process of becoming activated are also the main source of hypermutation in vivo.

Some quantitative parameters of hypermutation have been confirmed. The excessive accumulation of nonsynonymous substitutions and stop codon formation observed in hypermutants reinforces the perception that they can no longer generate viable viruses. However, hypermutation is not accompanied by a general increase in mutation rate; it is completely selective for G residues followed by A or G. We examined the question of whether hypermutation affects GC and GT dinucleotides, although at lower levels, in addition to the preferred GA and GG contexts. Our data are most consistent with the idea that G-to-A substitutions in GC and GT contexts are not part of the hypermutation process, but are only background mutations, for the following reasons: (i) the utilization of GC and GT did not increase proportionally as the saturation of GA and GG increased from 10% to more than 90% (other mutations; Fig. 5) and (ii) the levels of G-to-A substitution in GC and GT were never significantly higher than those of other types of mutations. However, our observations are limited to the HIV-1 protease gene, and examination of full HIV-1 genomes for the context of hypermutation would be required to fortify this conclusion. A consistent preference for saturation of available GA over GG dinucleotides within hypermutated sequences was also noted.

Efficient detection of hypermutated sequences requires the proper combination of PCR variables, including DNA sampling, primer design, annealing temperatures, and a physical method for separation and enrichment of hypermutants. While far more hypermutated sequences were detected here than in other studies, they still may have been underrepresented in our analyses. PCR primers matching all possible hypermutants were not included, in part because of their low annealing temperature. The detection of hypermutants on HA yellow gels, while efficient, may be influenced by the context of the G-to-A transitions and by their distribution within the sequence. Indeed, the DNA-binding preferences of bisbenzamide for runs of AT, particularly AAAA and AATT, would suggest that GA to AA would be more efficiently detected than GG to AG, but the fact that the vast majority of hypermutants we recovered used both contexts suggests that this was not a major obstacle to recovery. Finally, we have examined 297 bp of a 9,200-bp genome. Many hypermutated proviruses may have been missed by our procedure. The common perception that hypermutated sequences are rare in vivo should be revised in light of these considerations.

Careful definition of the quantitative parameters of hypermutation is necessary in order to define its biochemical basis. Here we show that the process is completely independent of the normal accumulation of mutations in HIV-1 and covers a much broader dynamic range. Figure 5 shows that the rate of accumulation of all other mutations, including G-to-A substitution in the GT and GC contexts, remained stable while G-to-A substitutions reached virtual saturation. The preferential utilization of available GA over GG dinucleotides over the whole dynamic range, and the fact that the vast majority of hypermutants utilized both GA and GG contexts, suggests that both contexts are recognized during a single reverse transcription event. It could be that the efficiency of hypermutation in the GA context is intrinsically, but not overwhelmingly, higher, or the frequency of extension past the mismatch could be higher for the GA context. The high error rate of HIV-1 RT is apparently necessary to permit both "normal" mutation and hypermutation, but hypermutation should not be thought of as an elevated error rate for RT; it is clearly an independent process with separate biochemical requirements which must vary over a wide range during HIV-1 replication in vivo.

Could hypermutation be influenced by viral variation, or is it essentially a passive function of the condition of the cell population in which HIV-1 replicates? Evidence that there may be a viral component appears in Table 2. Hypermutated sequences were recovered from a smaller percentage of patients infected with subtype C, compared to other subtypes. While itself not significant in this small sample (P = 0.078), the percent G-to-A substitution was also skewed to the lower end of the observed range in subtype C sequences. Stronger evidence would be needed, including larger numbers of patients with a broad range of clinical stages, to establish a definitive relationship between hypermutation and HIV-1 subtype.

The experiments with virus cultures are illuminating with respect to the conditions under which hypermutation occurs. First, the evidence that true hypermutants were generated in CD4⁺ T cells in culture should be reiterated. The target cell populations were free of HIV-1 because DNA extracted from the cells of mock-infected cultures carried in parallel were HIV-1 PCR negative. The virus stock was molecularly cloned and verified to be free of hypermutation by sequencing of the viral RNA by RT-PCR and by DNase treatment to remove any contaminating proviral DNA remaining from virus propagation. DNA PCR of the culture supernatants after infection was negative. The identity of the target cell population was carefully established; the culture contained principally, if not exclusively, CD4⁺ T cells. The major contaminant was B cells, which do not support HIV-1 replication. Most, if not all, CD4⁺ T cells were quiescent. It has been previously documented that infection of quiescent cells results in the absence of virus progeny (68) and that viral DNA can be detected after infection of unstimulated cells although it is not associated with virus production (69, 70). In our experiments, proviral DNA was established but there was no detectable virus production in the absence of PHA stimulation, which is consistent with the infection of quiescent cells. Most of the cells were HLA-DR negative. It is highly probable that those cells that were HLA-DR⁺, ranging from 15 to 18% in different experiments, represented recently, but not currently, activated cells.

Hypermutants were recovered principally from virus cultures in which HIV-1 infection was simultaneous with, or briefly preceded by, PHA stimulation (Fig. 6). The parameters of the hypermutants were quantitatively identical to those recovered from patient PBMC (Fig. 7), establishing a direct link between the circumstances established in culture and those leading to hypermutation in vivo. Our results suggest that hypermutation may be the direct and predictable outcome of certain biochemical stages during early PHA stimulation of T cells. Even the dinucleotide context preference is partially parsed out by these experiments; we note that in culture conditions XI, I, and II, where PHA either was not added or was added along with or after the virus, utilization of available GA and GG was almost equivalent. The strong bias for GA appeared once PHA preceded HIV-1 infection by only 3 h (Figure 7B). It is tempting to speculate that GG preference hypermutants derive from resting cells while those with the strong GA preference are generated in cells that are in the process of becoming activated.

A previous study demonstrated that hypermutants were recovered, although at low frequencies, after infection of unstimulated (2%) and stimulated (0.9%) bulk PBMC (62). Such recovery of hypermutants was ascribed to virus infection of a small proportion of PBMC with distorted dNTP pools, reinforcing the concept of an occasional and sporadic event. Because the composition of dNTP pools fluctuates as cells progress through the cell cycle (5, 20, 45), we investigated if the timing of HIV-1 infection with respect to T-cell stimulation could influence the generation of hypermutation. Our results define a window of susceptibility to hypermutation during initial events of T-cell stimulation. The target cell population for hypermutation must be continually renewed in vivo, commensurate with the proportion of resting T cells that become newly activated during active virus replication. Once PHA had been present for 8 h or more, the recovery of hypermutants plummeted. They may be no longer generated once T cells proceed far enough along the activation pathway. Nonetheless, it appears that the loss of informational content observed in hypermutants would preclude their further expansion except as passive passengers in the genomes of T cells. Such sequences should be included in conceptualizations of the latent reservoir of HIV-1 in patients.

Is hypermutation related to the observation that the genomes of HIV-1 and other lentiviruses are AT rich? The total and unrecoverable loss of coding potential of hypermutated HIV sequences (Table 3) would suggest that they are incapable of generating progeny virions and cannot contribute to the HIV-1 gene pool. However, it is important to consider the dynamic range of G-to-A substitution. For purposes of analysis we classified sequences as either normal or hypermutated, but, of course, the frequency of G-to-A substitution is a continuum, and those sequences with the lowest levels of hypermutation may sometimes be viable.

We have shown that hypermutation is much more abundant in patient PBMC than previously recognized. It will be important to determine what fraction of the genome in hypermutants remains unaffected, as well as the potential for hypermutated proviruses to generate viable viruses through recombination. Hypermutated proviruses could represent a difficult-to-eradicate remnant of HIV-1 during highly active antiretroviral therapy. Another aspect is the fact that the vast majority of hypermutated protease sequences present in this study were associated with a disrupted coding potential. Since protease has a crucial role during the HIV-1 replication cycle, the loss of its coding potential integrity should lead to abortive infections. Our data demonstrated that hypermutation may reduce the pool of viable replicating genomes in an infecting HIV-1 population. Detection of HIV-1 hypermutated sequences in at least 43% of patients studied suggests that hypermutation may happen in a systematic way in HIV-1-infected individuals. Based on these two observations one can speculate that hypermutation may be seen as a result of a host-associated mechanism which may decrease virus replication.

A better understanding of the metabolic profile of early proliferating T cells in relation to their expanding dNTP pools would lead to a better interpretation of the correlation between temporal aspects of cell cycle metabolism and HIV-1 hypermutation. This interpretation could develop into new strategies of promoting a hypermutation-inducible state in HIV-1 target cells, leading to irreversible mutagenesis of viral genomes. A better comprehension of the cellular mechanisms involved in HIV-1 hypermutation could open new avenues for virus control.