Mechanisms of Virologic Failure in Previously Untreated HIV-Infected Patients From a Trial of Induction-Maintenance Therapy FREE

Induction treatment of short duration followed by less intensive maintenance regimens proved unable to maintain viral load below the detection limit in recent clinical trials.^1- 4 The Trilège (Agence Nationale de Recherches sur le SIDA 072) study involved antiretroviral-naive patients eligible for the maintenance phase if, after a 3-month induction period with zidovudine, lamivudine, and indinavir (triple-drug therapy), their plasma human immunodeficiency virus (HIV) RNA levels were below 500 copies/mL. During the maintenance phase, 58 of the 279 randomized patients reached the primary end point (viral rebound >500 copies/mL in 2 consecutive samples). The proportion of patients reaching the primary end point was significantly higher in the zidovudine-lamivudine (29/93, P<.001) and zidovudine-indinavir groups (21/94, P = .01) than in the continued triple-drug maintenance arm (8/92). In multivariate analysis, baseline HIV RNA level was the only factor predictive of virologic failure.¹

Antiretroviral therapy failure has been attributed to low antiviral potency, selection of drug-resistant variants, poor treatment adherence, and pharmacokinetic interactions. The main aim of this study was to identify mechanisms of virologic failure during the maintenance phase of the trial by screening for genotypic resistance, estimating treatment adherence, and assessing degree of viral rebound.

When the Trilège trial was stopped in December 1997, the 58 patients with virologic failure (2 consecutive HIV RNA levels >500 copies/mL) were individually and randomly matched in a case-control study with 58 patients with sustained reduction in viral load of below 500 copies/mL. Cases and controls were matched for the randomization arm, maintenance therapy duration, baseline CD4⁺ cell count (±0.10 × 10⁹/L), baseline plasma HIV RNA level (±0.8 log copies/mL), and plasma HIV RNA level at the end of the induction phase (month 3; ≤ or >50 copies/mL). The current study took place in Paris, France, involving 3 urban hospitals (Bichat-Claude Bernard, Pitié-Salpétrière, and Paul Brousse Hospitals), and was conducted from February to October 1998.

During the maintenance phase, clinical and biologic assessments were performed every 6 weeks from randomization. Virologic and pharmacological studies of cases and controls were done in plasma sample 1 (S1), corresponding to initial virologic failure, and plasma sample 2 (S2), taken 6 weeks after S1 for confirmation of viral rebound. In controls, S1 and S2 samples were collected at times corresponding to those of samples for matched cases. Trial approval was given by institutional review boards of participating centers and patients provided written informed consent.

Genotypic studies were done on masked plasma samples from cases and controls. The reverse transcriptase gene was analyzed using the line-probe reverse-transcriptase assay,⁵ or by sequencing⁶ when the line-probe assay failed to give a signal after hybridization. The protease gene was analyzed by sequencing.⁷ Reverse transcriptase and protease genes were analyzed at baseline and at the time of S1 and S2 in cases.

The reverse transcriptase gene M184V mutation confers high-level resistance to lamivudine and appears rapidly when viral suppression is not maintained in patients receiving lamivudine-containing regimens.⁸ The M184V mutation was screened for by a restriction fragment digestion assay⁹ in samples obtained at the end of the induction from 50 of the 58 cases, and in S1 and S2 samples from zidovudine-lamivudine controls. This method amplified a short reverse transcriptase gene fragment and has higher sensitivity than sequencing when applied to plasma containing fewer than 500 copies/mL.⁹

Trilège trial protocol included treatment adherence assessment via pill count and plasma drug measurement. The number of tablets dispensed at each visit (weeks 2, 4, 8, and 12, then every 6 weeks) was recorded as was the number returned at the following visit; average between-visit adherence rate was calculated for each patient and each drug. During induction, patients received 100% of the pills corresponding to the interval before the next scheduled visit. During maintenance, they received 133% of prescribed tablets at the first of 2 visits, and 67% at the second visit. For example, if a subject was prescribed 2 pills of indinavir 3 times a day for 6 weeks (252 pills), 336 pills were dispensed at the first visit. If 100 pills were returned, the adherence rate was [(336-100)/252] × 100 = 93.7%. Thus, ranges may exceed 100%.

Indinavir concentrations were measured in masked plasma S1 and S2 samples from cases and controls in both indinavir-containing regimens, using high-performance liquid chromatography¹⁰ with a quantification limit of 5 ng/mL of plasma. Interval between last drug ingestion and sampling was recorded. Measured indinavir concentrations were interpreted relative to expected concentrations derived from mean 24-hour pharmacokinetic profiles in 24 healthy volunteers (area under the curve _{SS 0-24h} = 23.029 ng/mL per hour; C_max = 1.463 ng/mL; half-life = 1.85 hours). Because the coefficient of variation at various sampling times was close to 40% in the volunteers, a ratio of 0.6 between the study sample concentrations and expected indinavir concentration was taken as the lowest value compatible with expected level.¹¹

Plasma HIV RNA levels were determined using the Amplicor HIV assay (Roche Laboratories, Alameda, Calif) (detection limit, 200 copies/mL). Median viral load in S1 and S2 samples from patients with virologic failure was compared with baseline value to assess degree of rebound.

We used an intent-to-treat approach. All the case-control study observations were included in the analysis, even when the study treatments were discontinued prematurely. Wilcoxon rank-sum test was used to compare groups regarding continuous variables (ie, pill count, plasma indinavir level, and degree of viral load rebound).

Plasma indinavir ratios in cases and controls were compared separately at S1 and S2. Pill count data were censored at the time of the S2 sample. Adherence rates were calculated for the induction(enrollment to month 3), maintenance (month 3 to sample S2), and entire study periods. Case and control adherence rates were compared by prescribed drug, study period, treatment group, and time to virologic failure. In cases, early and late failures during the maintenance phase were respectively defined as those occurring at month 4.5 (patients were evaluated at 6-week intervals) and at month 6 or later. Degree of viral rebound in the two 2-drug arms was analyzed by pairwise comparisons against the triple-drug arm. Correlation analysis was used to determine strength of association between plasma indinavir ratio and corresponding degree of viral rebound, individually in S1 and S2 samples. P values were adjusted for multiple comparisons using the Hochberg method.¹² All reported P values are 2-sided. Analyses were performed using SAS software (Version 6.12; SAS Institute Inc, Cary, NC).

We included all cases available in December 1997. The choice of 1 control per case was dictated more for feasibility and study duration reasons than for power of comparisons. However, this case-control study is based on a relatively large number of patients, enabling us to reliably check our assumptions.

No mutations associated with reverse transcriptase inhibitor resistance were found at baseline. At the end of induction, the M184V mutation was not detected in any of the 22 successfully amplified case samples (of 50 samples available). In S1 and S2 case samples, the only primary resistance mutation detected in the reverse transcriptase gene was the M184V substitution, present in most patients receiving a lamivudine-containing maintenance regimen (Table 1). The key codon 215 zidovudine resistance mutation was not detected. Two patients developed minor codon 41 or 70 zidovudine resistance mutations.

In comparison with baseline data, the protease genotype in S1 and S2 samples showed only secondary indinavir-associated mutations, and in only 7 patients, 4 of whom were in the zidovudine-lamivudine maintenance arm.¹³ A set of 4 secondary mutations (codons 20, 36, 71, and 77) was found in sample S2 from 1 case in the triple-drug maintenance arm; these can be selected before the primary mutations at codon 46 or 82, conferring resistance to indinavir.¹⁴

There were 22 S1 and 21 S2 samples available from zidovudine-lamivudine controls for M184V testing; of the successfully amplified specimens, the M184V mutation was found in 11 of 13 S1 and in 10 of 11 S2 samples.

Plasma indinavir levels in S1 and S2 samples from cases and controls with an indinavir-containing regimen are shown in Figure 1. Controls had indinavir levels compatible with expected values, except for 1 patient in the zidovudine-indinavir arm. Indinavir was undetectable in all samples from cases in the 3-drug maintenance arm, except for 2 S1 samples, and levels were significantly lower than in controls (S1, P = .02; S2, P<.01).

In the zidovudine-indinavir arm, cases fell into 2 subgroups according to indinavir concentration ratios. Eleven of 18 and 5 of 12 cases had indinavir levels compatible with expected values in samples S1 and S2, respectively, while the remainder had values below those expected. All but 3 cases had indinavir concentration ratios on the same side of the cutoff in samples S1 and S2. No statistical difference between case and control values was found at S1 (P = .35), whereas values at S2 were significantly lower in cases than controls (P = .04).

Table 2 compares adherence rates based on pill counts in cases and controls for each drug throughout the study period. Globally, adherence to zidovudine (P = .02) and indinavir (P = .02) was significantly lower in cases than controls, with no statistical differences for lamivudine (P = .16). During induction, there was no significant difference in adherence rates between cases and controls. During the maintenance period, adherence to zidovudine (P = .05) and indinavir (P = .05) regimens was significantly lower in cases than controls.

Table 3 provides P values for differences in adherence rates between cases and controls in each maintenance arm. In the triple-drug maintenance arm, there was no difference during both induction and maintenance phase between patients with and without subsequent virologic failure. However, sample sizes in this group were small. In the zidovudine-lamivudine group, there was no significant difference in adherence rate for either drug during induction or maintenance periods. In the zidovudine-indinavir arm, cases had a lower adherence rate than controls for both drugs during the total period, and for the induction phase, it was significantly different (P = .04). Regarding whether possible poor indinavir adherence in the triple-drug arm (Figure 1) was associated with poor zidovudine-lamivudine adherence, a comparison between triple-drug and zidovudine-lamivudine cases showed the former had lower adherence to zidovudine (P = .04) and lamivudine (P = .03).

The significantly lower adherence rate for indinavir (P = .04) during induction in cases randomized to zidovudine-indinavir maintenance was associated with a large number of virologic failures at month 4.5 (early failure). Failures occurred early in 22 of the 58 cases (4 receiving triple-drug maintenance; 6, zidovudine-lamivudine; and 12, zidovudine-indinavir) and later (after month 6) in 36 patients (4 receiving triple-drug maintenance; 23, zidovudine-lamivudine; and 9, zidovudine-indinavir). Adherence rates in cases and controls are shown in Table 4 according to time of virologic failure. Globally, there was a significant difference in adherence rates of zidovudine, lamivudine, and indinavir during induction between cases and controls with early failure. We emphasize that for these cases, the difference in adherence rates during maintenance could not be reliably estimated because of the short follow-up. In late virologic failure, zidovudine, lamivudine, and indinavir adherence rates did not differ between cases and controls during induction or maintenance.

Baseline median viral load was 53,000 copies/mL in the 58 cases, with no statistical difference among the maintenance arms. At the time of virologic failure, the median viral load for each arm was 21,000 copies/mL for triple drug; 1100 copies/mL for zidovudine-lamivudine; and 1800 copies/mL for zidovudine-indinavir. Median time to virologic failure after maintenance therapy initiation was 3 months. Viral rebound at S1 and S2 in the maintenance arms is shown in Figure 2. Patients randomized to the zidovudine-indinavir arm were divided in 2 subgroups according to plasma indinavir levels: patients having at least 1 indinavir ratio below 0.6 were classified as subgroup indinavir-0 with the others classified as subgroup indinavir-1.

In the triple-drug and zidovudine-lamivudine maintenance arms, median viral load at S1 was 0.35 and 1.56 log lower than at baseline, respectively. In the zidovudine-indinavir arm, it was 0.73 and 1.66 log lower, respectively, than at baseline in the cases with indinavir levels below and in keeping with expected values. In S2 samples, the corresponding values were −0.39, −1.38, −0.31, and −0.31 log.

Correlation of degree of viral rebound with indinavir levels at S1 was r = 0.44 (P = .03) and at S2, r = 0.40 (P = .09) in patients randomized to triple-drug and zidovudine-indinavir maintenance arms.

This study shows that virologic failure during the Trilège trial maintenance phase was not associated with key zidovudine or indinavir resistance mutations. No such mutations were found at viral rebound or baseline, consistent with the patients' antiretroviral-naive status. In the 2 lamivudine-containing maintenance arms, the M184V mutation in the reverse transcriptase gene was seen in most successfully amplified samples from cases at viral rebound and from matched zidovudine-lamivudine controls (about 50% of which could be evaluated). Similar results have been seen in patients receiving lamivudine as part of a 2–nucleoside analog regimen.^8,15 However, as the M184V mutation is associated with persistent low-level viral replication, its significance may be minimal in patients with low viral load (<500 copies/mL). Our data suggest that this mutation does not play a major role, at least short-term, in virologic failure in patients with a zidovudine-lamivudine regimen. Virologic failure in the Trilège trial was not associated with resistant-mutant selection. Absence of resistance mutations, except for M184V, in this study may partly be explained by the fact that we assessed samples obtained shortly after virologic failure. These results prompted the hypothesis that lamivudine-resistant variants may arise more rapidly than indinavir-resistant variants since the M184V mutation confers greater selective advantage (in the presence of the drug) than any single protease mutation.

We used 2 complementary methods to assess adherence. We investigated adherence rates in cases and controls by drug, treatment arm, and phases (induction and maintenance) of the trial despite the small number of virologic failures (especially in the triple-drug arm). When the drug is present, plasma indinavir measurements provide unequivocal evidence of ingestion but only of the last dose. A potential bias is that patients might comply better than usual just prior to visits, while a good adherer missing a dose before the visit may be misclassified. Assays at 2 different visits, as in this study, may attenuate these shortcomings. A low plasma-indinavir concentration despite claimed ingestion within the prior 8 hours might indicate difficulties with adherence and poor digestive absorption, or both. Use of enzyme-inducing drugs, causing pharmacokinetic interactions, was prohibited by the protocol. Malabsorption was unlikely, because HIV RNA levels were reduced to below 500 copies/mL after the induction period. Pill counts are potentially a more objective way of estimating mid-term to long-term adherence and are relatively inexpensive. Pill counts allow estimation of medication quantity probably taken but not dosing frequency, which is important for drugs with a fixed dosing interval such as indinavir. Most important, the number of returned tablets routinely underestimates the extent of poor or partial adherence.¹⁶ Despite these shortcomings, comparison of cases and controls strongly suggested that poor adherence was associated with virologic failure in patients in the triple-drug maintenance arm and in some assigned to zidovudine-indinavir. Among the 8 cases in the triple-drug maintenance arm, 2 abruptly discontinued the study treatment just before virologic failure (due to pregnancy and a personal decision). Cases assigned to the triple-drug arm, who all had low or undetectable plasma indinavir levels both at time of virologic failure and 6 weeks later, were first thought to have discontinued indinavir only. But adherence rate comparison between triple-drug and zidovudine-lamivudine cases showed the former also had low adherence to zidovudine (P = .04) and lamivudine (P = .03). In the triple-drug arm estimated adherence and the fact that viral load rebounded to near-baseline values suggest that patients with virologic failure had low adherence throughout the maintenance period. In this treatment subgroup particularly, a possible limitation may be the likely low statistical power due to a relatively small number of events and subjects. In contrast, adherence rates were not different between cases and controls assigned to zidovudine-lamivudine maintenance. In this group, 79% of virologic failures occurred late (>6 months), with low viral rebound levels, as described with these 2 drugs.¹⁷ Our findings thus suggest that these failures are mainly related to lack of drug potency. Cases randomized to zidovudine-indinavir fell into 2 subgroups according to plasma indinavir levels and degree of rebound. Cases with low indinavir levels and strong rebound were comparable to cases assigned to triple-drug maintenance, low adherence being the main cause of virologic failure. In contrast, in cases with plasma indinavir concentrations in the range of expected values and weak rebound, virologic failure was probably related to suboptimal antiviral activity.

Few studies have simultaneously examined the roles of resistance and adherence in anti–HIV therapeutic failure.¹⁸ It has been stated that not all treatment failure was due to viral resistance, in either naive^19- 20 or heavily pretreated patients.^21- 23 In the Trilège trial, in which the only factor predictive of virologic failure was the baseline HIV RNA level,¹ failure seemed to be related more to low adherence and low antiretroviral potency, or both, than to selection of resistant mutants. In ACTG 343, a similar trial, escaping viruses were also wild-type, except at the reverse transcriptase gene codon 184.²⁴ The possibility that resistance assays used in the 2 trials failed to detect minor resistant species (representing <20% of the total viral population) cannot be dismissed. In the Trilège trial, 58 (20%) of 279 patients failed within 1 year, but this rate cannot be compared with rates obtained in other trials, because subjects with high adherence rates were selected during the 3-month induction period.

These results have a major practical implication for antiretroviral drug management: treatment adherence must be thoroughly investigated in virologic failure before switching therapy. When an initial antiretroviral regimen fails, compounds posing adherence problems must be identified and replaced by simpler alternatives. These results raise the hypothesis that compounds that do not pose adherence problems and for which no resistance mutations are detected might be pursued to maintain the largest possible therapeutic potential. However, if minority resistant virus populations are present, continuing with some part of the regimen may result in viral escape; thus, this strategy cannot be recommended today without supporting data from controlled trials.