Role of efavirenz plasma concentrations on long-term HIV suppression and immune restoration in HIV-infected children

Background To access the long term relationship between efavirenz plasma concentrations and evolution of HIV RNA loads and CD4 cell counts in children. Methods Retrospective analysis of data from HIV-infected children on first line efavirenz-containing regimen. A population pharmacokinetic-pharmacodynamic (PK-PD) model was developed to describe the evolution of HIV RNA load and CD4 cell count (efficacy outcomes) in relation to efavirenz plasma concentration. Individual CYP2B6 516 G>T genotype data were not available for this analysis. A score (ISEFV) quantifying the effect of efavirenz concentrations on the long-term HIV replication was calculated from efavirenz concentrations and PD parameters and, a value of ISEFV below which HIV replication is likely not suppressed was determined. Cox proportional hazards regression models were used to assess the association of the risk of viral replication with ISEFV, and with efavirenz mid-dose concentration(C12). Results At treatment initiation, median (interquartile range, IQR) age was 8 years (5 to 10), body weight 17 kg (14 to 23), HIV RNA load 5.1 log10 copies/mL (4.6 to 5.4), and CD4 cell count 71 cells/mm3. A model of PK-PD viral dynamics assuming that efavirenz decreases the rate of infected host cells adequately described the relationship of interest. After adjusting for age, baseline HIV RNA load and CD4 cell counts an ISEFV <85% was significantly associated with a higher risk of viral replication (p-value <0.001) while no significant association was observed with C12 <1.0 mg/L. Conclusion The ISEFV score was a good predictor of viral replication in children on efavirenz-based treatment.


Background
To access the long term relationship between efavirenz plasma concentrations and evolution of HIV RNA loads and CD4 cell counts in children.

Methods
Retrospective analysis of data from HIV-infected children on first line efavirenz-containing regimen. A population pharmacokinetic-pharmacodynamic (PK-PD) model was developed to describe the evolution of HIV RNA load and CD4 cell count (efficacy outcomes) in relation to efavirenz plasma concentration. Individual CYP2B6 516 G>T genotype data were not available for this analysis. A score (IS EFV ) quantifying the effect of efavirenz concentrations on the long-term HIV replication was calculated from efavirenz concentrations and PD parameters and, a value of IS EFV below which HIV replication is likely not suppressed was determined. Cox proportional hazards regression models were used to assess the association of the risk of viral replication with IS EFV , and with efavirenz mid-dose concentration (C 12 ).

Introduction
Efavirenz in combination with a dual-nucleoside reverse transcriptase inhibitor (NRTIs) backbone is the preferred first line antiretroviral therapy (ART) for HIV-infected children >3 years of age and weighing >10 kg [1]. The use of efavirenz-based ART has been shown very effective in suppressing viral replication in children and restoring immune function [2][3]. Efavirenz has been shown associated with a lower rate of virological failure compared to nevirapine in children [4][5]. However, a variability in response to efavirenz has been ascribed to differences in pharmacologic, virologic, immunologic, and behavioral characteristics [6]. To achieve target drug exposure, dose adjustment strategies have been designed using therapeutic drug monitoring (TDM), which individualizes drug dosing in order to maintain plasma drug concentrations within a target range (therapeutic window) [7][8]. Among children initiating first line efavirenz-based ART in Africa, a higher percentage of those with a minimum efavirenz plasma concentration >1.1 mg/L had an HIV RNA load decrease greater than 2 log 10 copies/mL after 3 months compared to children below this threshold [9].
A retrospective study in adult Marzolini et al. [10], reported that 50% of patients with efavirenz plasma concentration (below 1.0 mg/L) had viral replication (HIV RNA load >400 copies/mL). Among children, lower efavirenz concentrations were associated with higher risk of viral replication >400 copies/mL [11]. Recent data indicates that the efavirenz plasma concentrations may not provide an efficient tool to accurately predict the risk of viral replication [12][13].
Our aim was to predict with more accuracy than efavirenz concentrations alone the risk of viral replication using model taking into account the relationship between exposure to efavirenz and evolution of HIV-1 RNA load and CD4 cell count.

Patients
This analysis includes data from the prospective Program for HIV Prevention and Treatment (PHPT) Observational Cohort (NCT00433030). We selected all antiretroviral HIV-infected children who started as first line an efavirenz-containing ART in between May 1, 2002 and March 30, 2010, had HIV-1 RNA load and CD4 cell count measurements, and at least one efavirenz plasma concentration measurement available at any time point. Efavirenz was prescribed following US FDA approved weight-based dosing guidelines, as capsules or tablets to be taken once a day in the evening without regard for food. Children demographics (sex, age, body weight and height), antiretroviral treatment, adverse events, and laboratory data, including lipid plasma concentrations, alanine transaminase levels, HIV-1 RNA loads and CD4 cell counts were extracted from the cohort study database. Children receiving drugs known to interact with efavirenz, such as rifampicin, were excluded from the analysis.
The PHPT Cohort Study was approved by the ethical committees at the Thai Ministry of Public Health, local hospital ethics committees, and the Faculty of Associated Medical Sciences, Chiang Mai University, Thailand. Written informed consents were obtained from parents/legal guardians, and assents as appropriate.

HIV-1 RNA load and CD4 cell count measurements
HIV-1 RNA loads and CD4 cell counts were performed before treatment initiation and every 6 months after starting or switching ART. Plasma HIV-1 RNA loads were measured using the Cobas Amplicor HIV-1 Monitor RNA test version 1. 5

Population pharmacokinetic analysis
Efavirenz plasma concentration data were predicted using a previously published population PK model generated from the same study population [11]. Efavirenz plasma concentrations over 24 hours were described using a one-compartment model with delayed absorption via two transit compartments. Children were separated into two groups as either 'fast' (95% of children) or 'slow' metabolizers using a mixture model for efavirenz clearance (CL/F). Allometric scaling best described the relationship between body weight, apparent oral CL/F and volume of distribution.
HIV-1 RNA load and CD4 cell count data according to treatment duration were fitted using a nonlinear mixed effects modelling software program Monolix (Version 4.1.3, wfn. software.monolix.org). The population means and variances for the PD parameters were estimated by computing the maximum likelihood estimator without any approximation of the model (no linearization) using the stochastic approximation expectation maximization (SAEM) algorithm combined to a Markov Chain Monte Carlo (MCMC) procedure [14] (number of chains = 5 for all estimations). Residual variability (ε) was investigated using proportional, constant or combined error models. The between-subject variability (BSV or η) was ascribed to an exponential model. The Bayesian information criterion (BIC) was used to compare different models, i.e. structural models, covariate effect(s) on PD parameter(s). residual variability models, and structure of the variance-covariance matrix for the BSV parameters. Age, body weight, height, body mass index z-score (continuous or categorized), alanine transaminase levels, sex and type of NRTI backbone were individually tested as covariate effects. Categorical covariates (CA), e.g. sex, were modelled according to the equation where B CA is the estimated influential factor for the categorical covariate (CA = 0,1). Continuous covariates (CO) were systematically tested via generalized additive modelling versus the basic model. For example, for IC 50 , the equation where y IC 50 is the typical value of IC 50 for an individual with the median covariate value and B CO is the estimated influential factor for the continuous covariates. A covariate was retained in the model if it (i) was biologically plausible, (ii) produced a BIC decrease, and (iii) produced a reduction in the variability of the PD parameter, assessed be the associated BSV. Goodness-offit were assessed graphically.
The PK-PD model was similar to that reported to describe the relationship between the exposure to three antiretroviral drugs (efavirenz, didanosine and lamivudine) and the evolution of HIV-1 RNA load and CD4 cell count in Africa [15]. Following Ribeiro and Perelson [16], three differential equations described the viral dynamics where efavirenz inhibits the production of infected cells: These equations represent three compartments including (i) uninfected T lymphocyte target cells (T), (ii) HIV-infected cells (T � ) and (iii) free virions (V). The model assumes that T lymphocyte target cells are produced at a constant rate λ and die at rate d. Parameters β and δ are infection and death rates of infected cells. The virus is produced at rate p per infected cell and eliminated at rate c. Efavirenz, which reduces the ability of HIV virus to infect new target cells by inhibiting the reverse transcriptase enzyme [17], was assumed to decrease the production of infected cells via the inhibition of the parameter β (Fig 1). The corresponding inhibition function I(t) is defined according to the following an E max model [18]; where I max is the maximum fractional inhibition, IC 50 is the efavirenz plasma concentrations producing 50% of the maximal inhibitory effect (mg/L) and C av is the individual average efavirenz plasma concentration (mg/L). C av = (dose/dosing interval)/CL i , where CL i is the individual efavirenz clearance (L/h). The individual efavirenz clearances for each child were estimated using our previously published population PK model described above [11]. At steady state, before initiating antiretroviral treatment, the initial conditions of the system are T 0 = (c×δ)/(β×p), [15]. Initial conditions were then re-formulated in order to estimate HIV-1 RNA load (VL 0 ) and CD4 cell counts (CD 0 ) as baseline parameters, allowing the determination of the corresponding between-subject variability. Thus, λ, d, β, δ, p and c were derived from HIV-1 RNA load and CD4 cell count at baseline.

Model validation
The final population PK-PD model was evaluated using a visual predictive check (VPC). For the VPC, prediction-corrected (Uppsala prediction correction) visual check were used as informative diagnostic tools to allow inspection of model appropriateness across time as well as across covariate values.

Impact of efavirenz plasma concentrations on viral replication
We defined virological failure as viral replication (HIV-1 RNA load >200 copies/mL) at least after 6 months of treatment and confirmed by a consecutive measurement after 6 months of treatment. The C 12 and C 24 were estimated using our population PK model [11]. From the final model, an inhibitory score (IS EFV ) was calculated from C av , the I max , and IC 50 values as follows: The IS EFV stands for the resulting effect of efavirenz on HIV replication. A higher IS EFV indicates a higher effect of efavirenz on HIV replication. Sex, baseline characteristics (age, body mass index Z-score, alanine transaminase levels, type of NRTI combination, HDL, LDL, total cholesterol and triglycerides), C 12 , C 24 , C av , IC 50 and IS EFV during the period of observation were compared between children with or without viral replication using Fisher's exact test for categorical variables and the Wilcoxon rank-sum test for continuous variables.
A value of the IS EFV below which HIV replication is likely not to remain suppressed was determined using three methods as Youden's index, a closest-to-(0,1) criterion and Lui method [19].
The time to confirmed viral replication during the first three years of treatment was estimated by Kaplan-Meier method, and time-to-event distributions were compared using a logrank test. We estimated the hazard ratio of viral replication between children with C 12 above/ below 1.0 mg/L and with IS EFV above/below cut-off value, after adjusting for potential confounders (HIV-1 RNA load, CD4 cell count and age at treatment initiation) using Cox proportional hazard regression models. The Cox proportional hazards assumption of each model was assessed using Schoenfeld residuals.

Study population and follow up
Eighty-seven children (51% boys) met the selection criteria for analysis. Their first-line regimens contained zidovudine (n = 57) or stavudine (n = 30), plus lamivudine (n = 84) or didanosine (n = 3) in combination with efavirenz (dosed according to US FDA approved weight-  Table 1. The median duration of follow up was 24 months (12 to 34) and a total of 443 HIV-1 RNA load and 447 CD4 cell count measurements were available.

Population pharmacokinetic-pharmacodynamic analysis
The PK-PD viral dynamic model adequately described the relationship between efavirenz plasma concentrations, viral replication, and evolution of CD4 cell counts over the first 3 years of treatment. All fixed-effect parameters had a residual standard error (RSE%) below 25%, except for the efavirenz IC 50 (RSE = 47%). All random effect parameters had a RSE below 34%. After the maximum inhibitory effect (I max ) was fixed to 100% (not identifiable), the estimated mean efavirenz IC 50 was 0.1 mg/L. The infected cells (δ) elimination rate constant was fixed to 15.2 per month [20]. Between-subject variability parameters were retained for (i) production and elimination rates of uninfected target cells, (ii) infected rate of infected cells, (iii) production and elimination rates of free virions, and (iv) IC 50 . The residual variability was best described by an additive error model for HIV-1 RNA load and CD4 cell count data. None of the covariates had a significant effect on PD parameters. The final population PD parameters are summarized in Table 2.

Model evaluation
The VPCs of the final model, i.e. median, 25 th and 75 th percentile curves drawn over observed HIV-1 RNA loads and CD4 cell counts as a function of treatment duration, showed that the

Efavirenz plasma concentrations and viral replication
Regardless of metabolizer phenotypes, the median (IQR) C 12 was 2.9 mg/L (2.0 to 3.7), C 24 1.1 mg/L (0.7 to 1.4) and C av 1.7 mg/L (1.3 to 2.2). From Eq 5, the median IS EFV score calculated was 95% (86 to 96). Baseline characteristics and estimated C 12 , C 24 and C av were not significantly different between the 16 children with viral replication >200 copies/mL and the 71 virologically suppressed. However, both IC 50 and IS EFV significantly differed (p-value <0.001) between children with and without viral replication (median IC 50 : 0.60 versus 0.07; median IS EFV : 63% versus 96%, respectively).
Overall, 10 of 87 children had C 12 <1.0 mg/L and 77 children (89%) had C 12 �1.0 mg/L. Among the 10 children with a C 12 <1.0 mg/L (target threshold for efavirenz efficacy), three (30%) had confirmed viral replication compared with 13 (17%) of those 77 with C 12 above 1.0 mg/L (p-value = 0.383). Using the Youden's index, an 85% IS EFV cut-off value provided a  sensitivity of 94% and a specificity of 89% to predict viral replication, and classification accuracy was 90%. An 50% IS EFV cut-off value using the closet-to-(0,1) criterion provided a sensitivity of 7%, a specificity of 99% and a classification accuracy of 82%. Using Lui method, an 78% IS EFV cut-off value provided sensitivity of 63%, a specificity of 92% and a classification accuracy of 86%. The IS EFV cut-off value of 85% provided a high sensitivity, high specificity and high classification accuracy. Using this IS EFV cut-off, 15 of 23 children (65%) with a score below cut-off had viral replication compared with 1 of the 64 children (2%) with a score above cut-off (log-rank, p-value <0.001).
The cumulative risk of viral replication was not different in children with C 12 below and above 1.0 mg/L in the univariate analysis, or adjusting for HIV-1 RNA load, CD4 cell count and age at treatment initiation (adjusted hazard ratio [aHR], 2.74 [95% CI, 0.65 to 11.55], p-value = 0.170) (Fig 2A). In contrast, it was significantly higher in children with IS EFV <85% than children with IS EFV �85% in the univariate analysis or adjusted for HIV-1 RNA load, CD4 cell count and age at treatment initiation (aHR, 67.17 [95% CI, 8.29 to 544.45], p-value < 0.001) (Fig 2B).

Discussion
A PK-PD viral dynamics model assuming that efavirenz decreases the rate of infected host cells production adequately described the relationship between efavirenz plasma concentrations and efficacy outcomes (HIV-1 RNA load and CD4 cell count evolutions) over the first 3 years of treatment. However, there was no significant difference in the time to viral replication between children with C 12 below and above 1.0 mg/L. This finding is consistent with previous studies in adults and children starting efavirenz-based regimen. In a study in Cambodian adults, efavirenz plasma concentrations (12 to 16 hours after post-dose) below 1.0 mg/L were not significantly associated with virologic failure (HIV RNA load >250 copies/mL) [21]. This was similar to a study in African children, where there was no significant difference in the risk of virologic failure (HIV RNA load >400 copies/mL) according to efavirenz plasma concentration [22]. While, a study A5202 found an interaction between weight and low efavirenz concentration (at least one plasma concentration <1mg/L) was associated with virologic failure (defined as a confirmed HIV-1 RNA load �1,000 copies/mL at or after 16 weeks and before 24 weeks or �200 copies/mL at or after 24 weeks) [23]. Similarly, an adult study in South Africa found higher mid-dose efavirenz plasma concentration was associated with decreased risk of virologic failure (HIV RNA load >400 copies/mL at Week 16 and >40 copies/mL at Week 48), but they found a new threshold of mid-dose efavirenz concentration with 0.7 mg/L was the most predictive of virologic failure [24].
In our study, we defined a novel inhibitory score based on a PD hypothesis, IS EFV , calculated from C av , the maximum possible effect of efavirenz on HIV replication (I max ) and efavirenz plasma concentrations producing 50% of the maximal effect of drug (IC 50 ). The IS EFV accounted for the contribution of individual efavirenz plasma concentration and viral susceptibilities, and appeared to provide a better predictor of viral replication during the first 3 years of treatment than drug plasma concentration or viral susceptibilities alone. The risk of viral replication was significantly higher in children with a IS EFV below 85%.
Of note, an inhibitory quotient (IQ) was calculated from actual or expected trough drug concentration (C 24 ) divided by a measure of viral susceptibilities to that drug, IC 50 . In adults, a study published in 2003 in patients on lopinavir/ritonavir with or without efavirenz found that baseline efavirenz IQ was correlated with the 24 th week virologic response (HIV RNA load below 400 copies/mL) but the logarithmically transformed efavirenz concentration was not significantly correlated with virologic response [25]. Another study in children calculated a composite inhibition score (CIS), based on the same PD hypothesis as in our analysis, using the plasma concentrations of three drugs (efavirenz, didanosine and lamivudine) to estimate the relative potency of each drug, and efavirenz accounted for 65% of the total effect. A low CIS was associated with the risk of viral replication [15].
There are some limitations to our findings. Firstly, our PK-PD model was developed using data from HIV-infected children, median age 8 years, in Thailand, who may differ from populations previously studied in terms of frequencies of gene polymorphisms affecting efavirenz metabolism [26][27]. A study in Thailand found 48%, 44% and 11% of children with CYP2B6 516G/G, G/T, and T/T genotypes, respectively. The CYP2B6 516G>T gene polymorphisms were strongly associated with higher efavirenz plasma concentration (mean ±SD of efavirenz plasma concentrations were 1,604 ±729 ng/mL for children with G/G genotype, 2,635 ±1,199 ng/mL for G/T genotype and 11,582 ±2,972 for T/T genotype; p-value <0.001) [28]. Of note, the individual CYP2B6 516G>T genotype results were not available for the population studied but we took into account the presence of a small proportion of 'slow' metabolizers in the studied population. Secondly, the risk of viral replication may also be related to the NRTI backbone. However, in our model, there was no significant association with the NRTI backbone combinations. Thirdly, our PK-PD model was developed to investigate the effect of efavirenz on HIV-RNA time-course and CD4 cell count based on data from HIV-infected children initiating efavirenz-based combination in Thailand; thus the data may not describe a population with different antiretroviral combinations, races and/or ages.
In conclusion, the target threshold of efavirenz plasma concentration <1.0 mg/L may not be an optimal marker of the risk of viral replication. The IS EFV based on a PD hypothesis was a better predictor of confirmed viral replication during the first 3 years of treatment than C 12 .