The normalized inhibitory quotient of boosted protease inhibitors is predictive of viral load response in treatment-experienced HIV-1-infected individuals

Objective: The normalized inhibitory quotient (NIQ) has been proposed as a measure for refining the precision of HIV resistance testing when selecting antiretroviral therapy (ART). We undertook an assessment of NIQ and 48-week virological outcome in patients commencing ritonavir-boosted protease inhibitor (PI) regimens.

Design: A cohort of 87 HIV-infected individuals who all had extensive prior exposure to ART were assigned a new boosted PI regimen following resistance testing. PI therapy consisted of lopinavir, indinavir, saquinavir and amprenavir at 50, 32, 11 and 6%, respectively. Fold change (FC) for each PI was determined from the resistance test at baseline. Trough drug concentration (C_min) was determined at week 4.

Methods: NIQ was derived individually by taking the logarithm of the ratio of C_min/FC divided by the fixed ratio of population mean trough drug concentration/clinical cut off. Associations between viral load (VL) response over 48 weeks with baseline VL, FC, C_min, NIQ and selected PI were assessed.

Results: Mean change from baseline VL reduced by 0.83 log at week 48. In multivariate analyses, baseline VL and NIQ were the parameters most associated with change from baseline VL at week 48 (P = 0.012 and 0.003, respectively). FC, C_min and selected PI were not significantly associated with VL changes.

Conclusion: In this cohort of highly treatment-experienced individuals treated with boosted PI regimens, baseline VL and NIQ were significantly predictive of virological response over 48 weeks whereas FC and C_min were not. These results support the use of a NIQ at week 4, as a tool for predicting response to therapy in this setting.

Introduction

The development of HIV-1 resistance and subsequent virological failure occur in a substantial proportion of HIV-infected individuals receiving highly active antiretroviral therapy (HAART) ^[1]. Resistance represents an increase in the concentration of an antimicrobial agent required to inhibit a given level of microbial infection. This implies overcoming drug resistance by increasing drug exposure may be possible.

Inherent variability in the plasma concentration of the HIV protease inhibitors (PI) have been demonstrated in controlled pharmacokinetic studies in individuals who are HIV infected. In these studies, coefficients of variation for the area under concentration time curves range between 37 and 91% among all the available protease inhibitors ^[2,3]. Using ritonavir as a pharmacokinetic enhancer may reduce, but not eliminate, this variability ^[4].

The use of HIV drug resistance testing has been shown to provide virological benefit in randomized controlled trials ^[5-7]. A significant relationship between plasma drug concentrations and virological response has been described for several agents ^[8-10]. The ratio of drug concentration at trough (C_min) to a measure of susceptibility to a drug, the inhibitory quotient (IQ), was first suggested as a means to assess the response to antibiotics over 20 years ago ^[11]. More recently IQs have been proposed as a predictive measure of virological response to HIV protease inhibitors. To date limited clinical data have shown correlations between IQs for individual PIs and virological response ^[12-15].

A lack of standardization in measuring IQ has prevented direct comparisons of different antiretroviral agents. Several methods to calculate IQ have been proposed including the use of genotypic resistance testing and virtual phenotypic resistance testing (virtual IQ; VIQ). Recently the use of normalized IQ (NIQ) has been suggested; namely the normalizing of IQ values using population average C_min and fold changes in sensitivity ^[16]. This method overcomes differences in protein binding and C_min between different PIs. In a recent study of 55 treatment-experienced individuals, NIQ was significantly associated with virological response to lopinavir-ritonavir ^[17].

The aim of this study was to describe the relationship between a range of covariates, including NIQ, and 48-week virological response within a cohort of highly treatment-experienced individuals treated with a range of boosted single-PI-containing regimens.

Methods

Study design

The CREST Study (Can Resistance Enhance Selection of Therapy?) was a large multi-centered, randomized study comparing genotypic resistance testing with virtual phenotypic (Viragen II) resistance testing in treatment-experienced HIV-1-positive individuals failing antiretroviral therapy ^[18]. Individuals currently taking antiretroviral medication and with a plasma HIV RNA above 2000 copies/ml were eligible to enter this study.

For each individual, investigators chose a new antiretroviral regimen based on the results of one of the above randomly assigned baseline resistance tests. For individuals assigned to a PI-containing regimen, a trough PI level was performed after 4 weeks of treatment on the newly assigned regimen.

Individuals enrolled in the CREST study, who commenced a boosted single-PI regimen and for whom a trough plasma PI level was successfully collected at week 4, were included in this exploratory analysis.

Laboratory analyses

The C_min of the selected PI was determined at week 4, using a validated high-performance liquid chromatographic assay (HPLC), by the Department of Pharmacology and Toxicology, St. Vincent's Hospital, Sydney. Assay cut-offs for the lower limit of detection of plasma saquinavir, indinavir, amprenavir and lopinavir levels were 10, 25, 50 and 50 μg/l respectively.

Genotypic resistance tests were performed as previously described in the CREST study ^[18].

Fold change (FC) for each PI was determined from the virtual phenotype result (virco^®TYPE HIV-1 reports) obtained at baseline for all individuals, regardless of allocation of resistance test report physicians received. Reverse transcriptase (RT) and protease (PR) sequences were sent electronically to Virco BVBA (Mechelen, Belgium) and submitted to their database for generation of a predicted FC report. Where there were insufficient matches in the database a Virco-derived, rules-based interpretation of the sequence was made.

Calculation of NIQs

IQs were calculated as the ratio of PI C_min as determined at week 4 to the FC measured at baseline. IQs were corrected to NIQs by taking log₁₀ of the following: IQ divided by the ratio of population C_min to population cut-off fold change (Table 1). Population C_min levels were taken from published historic control data ^[19].

Table 1

Population cut-off fold changes were assessed using two methods. Biological cut-off data (Virco BVBA; virco^®TYPE HIV-1 reports) were used to calculate a biological cut-off NIQ (BCO-NIQ). Recently described clinical cut-off data were used to calculate clinical cut-off NIQ (CCO-NIQ) ^[20]. For this analysis we used the preliminary upper clinical cut-off FC for the corresponding ritonavir-boosted PI (preliminary upper CCO virco^®TYPE). This value corresponds to an 80% reduction in expected virological response from a large clinical database comprising of over 13 000 patients.

Statistical analysis

Predictors of change from baseline to week 48 HIV RNA were determined using linear regression modelling. The potential predictors assessed in univariate analysis were: baseline CD4⁺ cells × 10⁶/l; baseline viral load (VL) log₁₀ copies/ml; resistance test report type (genotypic or genotypic and virtual phenotype); PI; change in antiretroviral therapy during study (yes or no); Centers for Disease Control (CDC) stage; trough PI plasma concentration; fold change; CCO-NIQ; and BCO-NIQ. To allow comparison of trough PI levels for different PIs, the former were reported as the normal standard deviate (z transformation). Parameters with a non-linear distribution were log₁₀-transformed for the purpose of the analysis. Multivariate analyses were performed on parameters in univariate models with P-values less than 0.10, using a step-wise-forward method. As CCO-NIQ, BCO-NIQ, C_min and FC are correlated variables, only the most predictive of these variables was assessed in the multivariate analysis.

Association between predictors of detectable VL (> 400 copies HIV RNA/ml) and NIQ was assessed by logistic regression. NIQ was categorized into inter-quartile groups for this analysis.

All P-values are two-tailed with values less than 0.05 regarded as statistically significant. Data were analysed using STATA statistical software (v. 8.0; STATA Corp., LP).

Results

Patients and baseline characteristics

A total of 338 individuals were randomized into the CREST study between October 2000 and April 2001. Of these individuals a total of 87 individuals were assigned a new boosted single-PI regimen by physician choice and had a trough plasma PI level measured. These individuals represent a heavily pre-treated population with all individuals previously exposed to nucleoside reverse transcriptase inhibitors (NRTI), 68% previously exposed to non-nucleoside reverse transcriptase inhibitors (NNRTI) and 93% previously exposed to PIs. Of these 87 individuals, 44 (50%) commenced lopinavir-r, 28 (32%) indinavir-r, 10 (12%) saquinavir-r and five (6%) amprenavir-r. These regimens were assigned on the basis of a genotypic resistance test result in 52% of individuals and virtual phenotypic report in the remaining 48% of individuals (Table 2).

Table 2

Clinical outcomes

Mean change from baseline HIV RNA at week 48 was -0.83 log₁₀ copies/ml [95% confidence interval (CI), 0.63-1.04] and mean change from baseline CD4+ lymphocyte count 70 × 10⁶ cells/l (95% CI, 29-113).

Of these 87 individuals, 40 (45%) experienced some change in antiretroviral therapy during the 48-week study period. In 24 (28%) patients, the designated PI was stopped (two individuals receiving amprenavir, 15 indinavir, four lopinavir and three saquinavir).

Associations with virological response

Factors associated with reductions in plasma HIV RNA from baseline to week 48 were assessed (Table 3). In a univariate analysis, baseline CD4+ lymphocyte count, type of resistance test report, changes in PI during study, CDC classification, trough PI level, FC and BCO-NIQ were not significantly associated. Baseline HIV RNA VL, choice of PI in new regimen, and CCO-NIQ were significantly associated. There were no associations between reductions in plasma HIV RNA and individual PIs in new regimens (P-values 0.461, 0.145, 0.649 for indinavir, lopinavir and saquinavir, respectively; reference group amprenavir).

Table 3

In a multivariate analysis, baseline VL, changes in PI therapy during study and CCO-NIQ remained significantly associated with fall in HIV RNA at week 48 (P-values 0.012, 0.034 and 0.003, respectively). When stratifying results for individuals who changed PI therapy during the study period, CCO-NIQ was associated with changes in week 48 HIV RNA only for those who did not change PI therapy (P = 0.015 versus 0.327).

Furthermore, CCO-NIQ was also significantly associated with reductions in plasma HIV RNA from baseline to weeks 12 and 24 in multivariate analyses (P = 0.038 and 0.032, respectively).

Stratifying CCO-NIQ into inter-quartile ranges (-0.760 to 0.590, 0.591 to 1.090, 1.10 to 1.634, 1.635 to 2.70) and assessing numbers of individuals with undetectable HIV RNA to less than 400 copies/ml at week 48 (Fig. 1), indicated that individuals in the higher CCO-NIQ inter-quartile ranges were more likely to be undetectable compared with those in the lower ranges (18, 58, 56 and 76%, respectively; P-value for trend = 0.004). Odds ratios for undetectable VL for each inter-quartile range are shown in Fig. 2. BCO-NIQ inter-quartile ranges were also associated with greater numbers of individuals with undetectable VLs in the higher ranges (P-value for trend = 0.015).

Fig. 1

Fig. 2

Discussion

In this cohort of highly treatment-experienced individuals treated with boosted single-PI regimens, baseline VL and CCO-NIQ were significantly predictive of virological response over 48 weeks whereas FC, C_min and other baseline characteristics were not. These results support the use of a NIQ at week 4, as a parameter which may enhance associations in FC and C_min with virological response that are not observed independently, as a tool in predicting response to therapy in this setting.

Several methods for calculating IQs have been described. The use of NIQ encompasses a correction for population C_min allowing comparisons between different agents. These data have shown a high correlation between NIQ result and virological response which was not dependent on which PI an individual was receiving.

Several methods to assess cut-off fold change can also be utilized. From our data, the use of CCO-fold change to calculate NIQ was associated with a greater correlation with virological response than BCO-fold change. The CCO-fold change we have utilized may be more accurate than BCO as it incorporates a measure of clinical outcome ^[20]. We have utilized the reported upper CCO which correlates to the FC associated with an expected decrease of 80% in the maximal antiviral activity of the drug. CCO data may change over time as further clinical data are collected and analysed. As these results change, and the formula to calculate NIQ is appropriately adjusted, continued work to assess the relationship between this parameter and clinical endpoints are needed.

Our study has several limitations. No measure of adherence to medication was included within our analysis. Individuals enrolled within the CREST study represent a highly treatment-experienced group. The association of NIQ and virological response may differ in less treatment-experienced cohorts. Since this work has been carried out new PIs such as atazanavir, fosamprenavir, tipranavir and TMC-114 have been licensed and are in clinical development. These agents have different pharmacokinetic and resistance profiles from the agents used in this analysis and these may affect the correlation between NIQs and virological response.

The use of boosted double-PI regimens has recently been assessed in a number of small clinical trials ^[21-23]. In one study in which individuals received amprenavir-lopinavir-ritonavir, the VIQs of amprenavir and of lopinavir were both independently associated with trends in VL response to treatment ^[24]. Our data do not assess the use of NIQ within boosted double-PI regimens.

CCO-fold change has been calculated using data from several datasets including clinical trials and cohort data. The CREST trial was included in this analysis and this may have impact on some of the associations we have observed in this analysis.

The calculation of a CCO-NIQ after 4 weeks of a new boosted single-PI regimen using the formula described, is simple, and could be used as a predictor of virological response within clinical practice. Further work is needed to validate these results in other cohorts, with the use of new PIs and in boosted double-PI regimens.

Acknowledgements

The National Centre in HIV Epidemiology and Clinical Research is funded by the Australian Government Department of Health and Ageing, and is affiliated with the Faculty of Medicine, The University of New South Wales.

We would like to thank Lee Bacheler and Bart Winters for their assistance in the interpretation of phenotypic clinical cutoffs.