Estimate of the Frequency of Human Immunodeficiency Virus Type 1 Protease Inhibitor Resistance within Unselected Virus Populations In Vitro

Abstract

The frequency of drug-resistant human immunodeficiency virus type 1 (HIV-1) variants in virus populations not previously exposed to drug was determined in vitro by using HIV-1_RF and the protease inhibitor SC-55389A. Two variants with single mutations responsible for drug resistance (V82A and N88S) were quantifiably isolated after only one round of replication, yielding a crude frequency estimate of at least 1 SC-55389A-resistant variant per 3.5 × 10⁵ wild-type infectious units.

Retroviruses, including human immunodeficiency virus type 1 (HIV-1), encode a unique aspartyl protease, consisting of a homodimer of two 99-amino-acid polypeptides, that is responsible for cleavage events resulting in the maturation of essential viral enzymes and structural components. Since a functional protease is required for viral infectivity (10, 11), this enzyme represents an excellent target for the development of antivirals (5, 9–11), and a considerable number of effective protease inhibitors have been discovered (14, 23). The selection of drug-resistant variants is an unfortunate, but foreseen, consequence of the use of compounds which inhibit the replication of microorganisms, such as HIV, that exist in large, genotypically variable populations. Accordingly, numerous HIV variants that exhibit reduced sensitivity to a variety of protease inhibitors have been selected in vitro and isolated from patients treated with protease inhibitors (4, 14).

The high rate of viral replication, the large population size, and the error-prone nature of reverse transcription imply that HIV-1 variants resistant to antiviral agents will arise comparatively frequently in infected patients. However, although such variants will arise, they are likely to exhibit variable viability. Therefore, the absolute titer of a given viral variant will be a function of the fitness, or selective advantage, of the variant in comparison to the predominant, or wild-type, species (3). Mutations known to confer resistance to a variety of reverse transcriptase (RT) and protease inhibitors have been found in isolates from untreated patients (1, 8, 12, 13, 15). However, the approaches used were unable to account for the occurrence of genotypic changes that have yet to be correlated with phenotypic drug resistance (i.e., novel mutations that confer resistance and have yet to be identified experimentally). Measurement of the steady-state level of drug-resistant variants within unselected populations may be of predictive benefit in the choice of compounds for clinical trials and/or the design of therapeutic strategies. We have taken a simple approach to estimate the frequency of phenotypic drug resistance in unselected populations using a laboratory-adapted strain of HIV-1.

Twenty million CEMT4 cells (AIDS Research and Reference Reagent Program, Division of AIDS, National Institute of Allergy and Infectious Diseases) were infected with 10⁶ 50% tissue culture infective dose (TCID₅₀) units of HIV-1_RF to give a multiplicity of infection of 0.05 TCID₅₀ units per cell. The culture was divided into 960 subcultures in 10 96-well tissue culture plates (primary plates), which were incubated in the presence or absence of 0.5 μg of SC-55389A/ml (2, 7). A 0.5-μg/ml concentration of SC-55389A was used because standard dose-response assays (22) revealed that this concentration completely inhibited viral cytopathic effect (CPE) and supernatant RT activity above background. (All assays, including viral-titer assays, involved the identical cell line and virus stocks that were used in the drug selection experiment. TCID₅₀ assays were scored by CPE and RT on day 7 postinfection [22] and several days thereafter to ensure accurate titers.) The cultures were split 1:2 and were fed with medium containing fresh drug weekly. Regular monitoring of the primary drug-positive subcultures for CPE during the 1st week postinfection revealed a low background level of CPE in all cases (usually first manifested at 4 to 7 days postinfection as small syncytia which progressed slowly on prolonged culture), presumably as a consequence of cell fusion and killing due to the approximately 1,000 TCID₅₀ units of input virus, and 32 cultures which exhibited a significantly more-extensive CPE. On day 7 postinfection, the contents of these 32 “positive” wells were transferred to secondary plates and supplemented with additional media, drug, and 2 × 10⁴ uninfected CEMT4 cells/well. These cocultures were incubated for an additional 7 days in order to expand the population of infected cells and provide sufficient material for subsequent assays. The level of viral replication was then determined by visual scoring of CPE and by RT assay of supernatant samples. Genomic DNA was also extracted from a sample of infected cells at this time. A region of pol encompassing the protease gene was amplified from the genomic DNA of virus-positive samples by PCR and sequenced by direct-cycle sequencing of the PCR product (18, 22). To obtain quasiclonal viral stocks for dose-response assays, cell-free supernatants from the cocultures were subjected to three rounds of growth at limiting dilution in the absence of drug. A stock culture of each virus (termed “selected variant”) was subsequently grown in the absence of drug, titered, and used for phenotypic characterization. The infected CEMT4 cells used to grow the selected-variant and wild-type control stock cultures were retained and used to confirm the viral genotypes as described above.

Of the 32 cultures transferred to the secondary plate, 9 showed evidence of further virus replication by CPE and/or RT activity (Table 1), implying the presence of drug-resistant variants. Sequencing of PCR-amplified proviral DNA derived from the drug-treated cultures confirmed the presence of nucleotide changes encoding amino acid substitutions in the protease which were absent from the wild-type controls grown in the absence of drug (Table 1). Of the changes observed, N88S (2 of 9) and V82A (2 of 9) have previously been correlated with resistance to SC-55389A (17, 18, 21, 22). To confirm that the mutations observed were responsible for drug resistance, each was re-created by oligonucleotide site-directed mutagenesis in an HIV-1_HXB2 proviral clone (18, 22). Progeny viruses were derived from the mutated proviral DNA and were used to establish virus stocks (termed “recombinant variants”). The presence of the mutated residue in the recombinant variants was confirmed by PCR and cycle sequencing as described above. The growth kinetics and drug sensitivity of each recombinant variant and selected variant were then determined (18, 22). Recombinant viruses bearing an I62M substitution in protease exhibited growth kinetics (Fig. 1) that were almost indistinguishable from those of wild-type HIV-1_HXB2. The I62M M46I RF selected variant exhibited only a moderate shift in SC-55389A sensitivity, which was not replicated by the I62M HXB2 recombinant variant (Table 2). The V82A recombinant variant also grew well (Fig. 1), but in this case resistance to SC-55389A was observed (Table 2). The N83S recombinant virus grew to titers comparable to those of wild-type HIV-1_HXB2 (Fig. 1) and remained susceptible to SC-55389A (Table 2). In contrast, the N88S substitution resulted in a more-marked growth defect (Fig. 1) and a significant increase in resistance to SC-55389A for both the recombinant HXB2 and selected RF variants (Table 2).

TABLE 1.

Drug resistance in HIV-1_RF selected variants, as manifested by CPE and RT activity, and related mutations in protease

Virus isolate	CPE scorea	RT activity (cpm/20 μl)b	Mutationc
1C9	+/−	527	wt (AAT to A[A/g]T)d
1H5	ND	20,622	N88S (AAT to AGT)
5B11	ND	32,701	N83S (AAC to AGC)
7A7	+	36,686	V82A (GTC to G[C/t]C)e
7H3	+	29,948	V82A (GTC to GCT)
8B11	+++	51,366	N88S (AAT to AGT)
8F4	+	28,052	wt (AAT to A[A/g]T)d
8H5	+	419	N83S (AAC to AGC)
10G1	+	16,556	M46I (ATG to AT[T/A/G]); I62M (ATA to ATG)f

^aND, not determined. 

^bWells with negative CPE scores were found to have RT activities of 140 to 510 cpm/20 μl of culture. 

^cAmino acid substitution (nucleotide substitution). Minor fractions of a mixed sequence are indicated by lowercase. Sequences were determined by direct-cycle sequencing of pol PCR product derived from total-cell DNA corresponding to viral stocks. wt (wild type), sequence corresponding to that derived from identical control subcultures incubated in the absence of drug. 

^dMixed sequence at nucleotides corresponding to N88. Minor fraction is N88S. 

^eMixed sequence at nucleotides corresponding to V82A. Minor fraction corresponds to wild-type sequence. 

^fMixed sequence at nucleotides corresponding to M46I. Approximately 50% wild-type sequence. 

FIG. 1.

Growth kinetics of HIV-1_HXB2 variants created by site-directed mutagenesis of a proviral clone (pHXB2gpt2). p24 core antigen was measured in cell-free medium samples by enzyme-linked immunosorbent assay (Coulter Corp.) (16).

TABLE 2.

Resistance of selected and recombinant HIV-1 variants to antiviral drugs^a

Drug	Fold shift in EC50 against variant with the indicated mutation(s)b
	RF selected variantc			HXB2 recombinant variant
	N88S (8B11)	wt (8F4)	M46I I62M (10G1)	I62M	V82A	N83S	N88S
AZT	1	0.4	1	0.3	0.4	0.6	0.8
SC-55389A	>6	2	4	0.5	4.6	1.1	19.2

^aSequences of viruses were determined by direct-cycle sequencing of pol PCR product derived from total-cell DNA corresponding to viral stocks. 

^bCompared to the 50% effective concentration (EC₅₀) for the wild type. EC₅₀s were determined as previously described (16, 18). The EC₅₀ for SC-55389A against HIV-1_RF in CEMT4 cells, derived from multiple independent experiments, is 35 ± 6 ng/ml. The V82A and N83S HIV-1_RF variants did not grow to sufficient titers to allow for the performance of a dose-response assay. 

^cIsolate designations are given in parentheses after mutations. wt (wild type), sequence corresponding to that derived from identical control subcultures incubated in the absence of drug. 

These data are consistent with the hypothesis that the N83S and I62M mutations are not responsible for drug resistance (assuming that such mutations do not have an HIV-1_RF protease context specificity [20]) and presumably represent false positives from the visual-screening step. Notably, positions 83 and 62 are reported to be naturally heterogeneous (1, 12). The comparatively high RT values in the secondary plates in some cases (Table 1) may indicate the presence of mutations associated with drug resistance that lie outside of the protease gene, or the replication of an unrelated resistant virus(es) that was insufficiently represented in the proviral DNA that was ultimately sequenced. Indeed, the RT values associated with isolate 8F4 (wild type) were probably due to the growth of an N88S variant that comprised a minor portion of the total population (Table 1). Since the viruses were subsequently cloned by limiting dilution in the absence of protease inhibitors, such variants may have been lost from the population due to their lower titer or an inability to compete with the wild type in the mixed population, before their effect upon drug susceptibility was determined.

The V82A and N88S substitutions conferred SC-55389A resistance on HXB2 recombinant variants (Table 2) and were observed in four of the nine primary virus isolates that grew in the presence of SC-55389A (Table 1). It is therefore likely that these variants were isolated because of their increased tolerance of SC-55389A. The N88S and V82A drug-resistant viruses may be derived from two possible sources. It is possible that they existed in the viral population at the start of the experiment or arose during the course of the experiment. Since these isolates were obtained after only a single passage (i.e., only one instance of cell-free inoculation) in the presence of sufficient SC-55389A to essentially abolish the replication of wild-type virus, it is likely that in the majority of cases only a single cycle of reverse transcription and integration took place. The frequency with which the V82A and N88S mutants were isolated in this experiment may therefore reflect the frequency of their occurrence in unselected populations. The population size from which the resistant variants were selected was 10⁶ TCID₅₀ units. The distribution of infected wells in a TCID₅₀ assay is described by the Poisson distribution, e^{−TCID₅₀} = 0.5, yielding a value of 0.7 infectious units per TCID₅₀ unit (6), or 7 × 10⁵ infectious units in the present experiment. Assuming equivalent infectivity and a random distribution of the variants between subcultures, the frequency of the V82A or N88S mutation in the population of HIV-1_RF can be crudely estimated to be 2 per 7 × 10⁵ infectious units, or approximately 1 in 3.5 × 10⁵. Although this estimate indicates that such mutants are comparatively common, because of the many assumptions underlying it, the absolute value is unlikely to be particularly accurate. For example, due to experimental limitations, the approach we have taken is likely to overlook drug-resistant variants which grow poorly or fail to elicit significant CPE. Indeed, because viruses bearing nucleotide substitutions corresponding to the N88S change were also apparent as minor fractions of two other isolates (Table 1), it is likely that, at least for N88S, this frequency is an underestimate. It is expected that additional experiments will provide insight onto the reproducibility of our results and will improve the precision of the estimate.

These studies demonstrate that it is possible to isolate HIV-1 variants resistant to protease inhibitors from comparatively small viral populations that have never been exposed to the inhibitor. Such variants are, therefore, likely to be present in HIV-1-infected patients. Indeed, positions 82 and 88 (19) are reported to be variable in natural populations isolated from patients who have not been treated with protease inhibitors. These and similar data may prove useful in the development of novel protease inhibitors and drug combinations that select for a drug resistance mutation(s) that occurs with substantially reduced frequency. Since frequency is a function of selective advantage, such an approach may yield a quantitative method for favoring the selection of drug-resistant variants that are severely debilitated and possibly less pathogenic.