Effect of albumin and CYP2B6 polymorphisms on exposure of efavirenz: A population pharmacokinetic analysis in Chinese HIV-infected adults

Xian-min Meng a,#, Zi-ran Li b,#, Xin-yin Zheng b, Yi-xi Liu b, Wan-jie Niu b, Xiao-yan Qiu b,*,
Hong-zhou Lu a,*
a Shanghai Public Health Clinical Center, Fudan University, Shanghai 201508, China
b Huashan Hospital, Fudan University, Shanghai 200040, China



Background: Efavirenz is a vital component used to treat HIV-1 infection. Nevertheless, it shows large between- subject variability, which affects both its therapeutic response and adverse effects.

Objective: To investigate the impact of gene polymorphisms and non-genetic factors on the variability of efavirenz pharmacokinetics and to propose the optimal dose regimens.

Methods: A total of 769 plasma samples from 376 HIV-infected Han Chinese outpatients were collected to develop a population pharmacokinetic model using NONMEM software. The impact of patient demographics, laboratory tests, concomitant medication, and genetic polymorphisms of CYP2B6 and ABCB1 on efavirenz pharmacokinetics were explored. According to the final model, the model-informed dose optimization was conducted.

Results: The pharmacokinetics of efavirenz was characterized by a one-compartment model with first-order ab- sorption and elimination. The typical values of the estimated apparent oral clearance, volume of distribution, and absorption rate constant in the final model were 9.44 L/h, 200 L, and 0.727 h — 1, respectively. Efavirenz clearance was significantly influenced by CYP2B6 variants, including rs2099361, rs3745274, and rs2279343, along with albumin and weight. The volume of distribution was affected by albumin and weight. Based on the CYP2B6 polymorphisms of patients, the recommended daily doses of efavirenz were 100 mg for CYP2B6 slow metabolizers, 400 or 600 mg for intermediate metabolizers, and 800 or 1000 mg for extensive metabolizers. Conclusions: Polymorphisms of CYP2B6, along with albumin and weight, resulted as the predictors of efavirenz pharmacokinetic variability, which could be used in prescribing optimal efavirenz doses.

1. Introduction

Efavirenz (EFV), a nonnucleoside reverse transcriptase inhibitor, exerts its action by binding directly to reverse transcriptase and inhib- iting viral RNA- and DNA-dependent DNA polymerase activities (Arts and Hazuda, 2012; Vo and Gupta, 2016; Rakhmanina and van den Anker, 2010). EFV has been approved for the treatment of type I human immunodeficiency virus (HIV-1) infections since 1998. In many low-income nations, EFV is still one of the most commonly prescribed antiretroviral drugs due to its proven efficacy and low cost (Sheran, 2005, 5).
EFV is characterized by large within- and between-subject variability (BSV) in plasma concentrations (Csajka et al., 2003; Luo et al., 2016; Habtewold et al., 2017). To ensure a balance between adverse drug reactions and possible treatment failure, therapeutic drug monitoring (TDM) has been suggested as a crucial part of patient management during EFV treatment (Guti´errez et al., 2005). Several studies reported that steady-state plasma concentrations of EFV below 1 μg/mL were associated with an increased risk for virological failure and drug resis- tance, while concentrations above 4 μg/mL have been associated with an increased risk for the development of central nervous system adverse effects (Langmann et al., 2002; Marzolini et al., 2001).

EFV reaches a peak plasma concentration (Cmax) in approximately 3 to 5 h after oral administration (Vrouenraets et al., 2007). It is highly bound to human plasma proteins (99.5 – 99.75%), predominantly al- bumin (ALB). In one in vitro study, an increase in extracellular ALB has found to be associated with a significant reduction in intracellular penetration and antiviral effect of EFV (Avery et al., 2013). However,
the in vivo quantitative influence of ALB on EFV pharmacokinetics (PK) and whether the EFV dose should be adjusted according to ALB levels remains unclear.

Hepatic metabolism by the cytochrome P450 (CYP450) system is the principal mechanism of EFV apparent oral clearance (CL/F), and CYP2B6 is reported to be the major isozyme responsible for this process (Ward et al., 2003). Previous studies have shown that screening CYP2B6 functional variants has high predictability for EFV plasma levels and could be used to optimize EFV doses (Arab-Alameddine et al., 2009; Dhoro et al., 2015; Hui et al., 2016; Olagunju et al., 2018). Clinical Pharmacogenetics Implementation Consortium (CPIC) has recom- mended initiating EFV dosage based on the CYP2B6 phenotype of pa- tients (Desta et al., 2019). In addition, some studies reported that several common single nucleotide polymorphisms (SNPs) in the protein region of ABCB1 were associated with elevated plasma EFV concentrations (Swart et al., 2012; Ngaimisi et al., 2013; Mukonzo et al., 2009). How- ever, other studies failed to replicate this association (Dhoro et al., 2015; Duarte et al., 2017; Elens et al., 2010). The effects of SNPs in the ABCB1 gene on EFV PK remain controversial.

Identifying the sources of EFV PK variability could improve the therapeutic efficacy by using patient-specific factors like genetic infor- mation and patient demographics to guide dosing. Population pharma- cokinetic (PPK) modeling now has an established role in identifying patient-specific predictors that determine BSV and rationally deriving individualized dose regimens (Li et al., 2021). Previous PPK studies on EFV mainly focused on common genetic determinants of CL/F and the influence of concomitant medication (Arab-Alameddine et al., 2009; Dhoro et al., 2015; Hui et al., 2016; Olagunju et al., 2018). However, the impact of other potential SNPs of CYP2B6 and laboratory tests like ALB on EFV PK were rarely explored.

In the present study, we investigated the contribution of 9 potential SNPs of CYP2B6 and 2 SNPs of ABCB1 to the EFV PK variability using the PPK modeling approach in Han Chinese HIV-infected adults. Moreover, the influence of patient demographics, laboratory tests including ALB as well as concomitant medication were also investigated. Finally, based on the final PPK model, dose optimization of EFV based on significant covariates was simulated to maintain the EFV plasma concentration within the therapeutic target range.

2. Material and methods
2.1. Study design and population

The current study enrolled Chinese HIV-infected outpatients aged ≥18 years who had been receiving EFV for at least 1 month at Shanghai Public Health Clinical Center between January 2012 and June 2013. Meanwhile, patients were excluded if: 1) they were pregnant or breastfeeding; 2) they showed poor compliance with drug therapy; 3) the information about their medication or sampling time was missing.
When patients went to the outpatient clinic for follow-up, 5 mL venous whole blood was collected using EDTA anticoagulant tubes and centrifuged at 3500 RPM for 5 min to remove the blood cells. Then, all samples were heat-deactivated in a 56 ◦C water bath for 60 min and stored at —80 ◦C before analysis.

Patients’ demographics, including age, weight, gender, specific cART regimens, concomitant medication, EFV dose, time of EFV administra- tion, and sampling time, were collected. In addition, laboratory test data were also extracted from the hospital information system, including total bilirubin (TBIL), direct bilirubin (DBIL), ALB, alanine amino- transferase (ALT), alkaline phosphatase (ALP), γ-glutamyl trans- peptidase (GGT), creatinine clearance rate (CCR), and hemoglobin (HB). This study followed the principles of the Declaration of Helsinki, and approval was granted by the Ethics Committee of Shanghai Public Health Clinical Center. Written informed consent was obtained from all

2.2. Determination of EFV concentration

The total plasma concentrations of EFV were determined using reverse-phase high-performance liquid chromatography (RP-HPLC) with ultraviolet detection based on a standard protocol (Villani et al., 1999). The linear range was 0.1–20 μg/mL, with intraday/interday co- efficient of variation of 1.9/7.2%, 2.4/2.2% and 2.6/2.2% at concen- trations of 0.3, 3.0, 10.0 μg/mL, respectively.

2.3. Genotyping analysis

SNPs were selected primarily based on 1) data on the CYP2B6 and ABCB1 gene among the Han Chinese population obtained from http://www.ncbi.nlm.nih.gov/SNP and filtered using Haploview4.2 software (Broad Institute, Cambridge, MA, USA); 2) SNPs of CYP2B6 that had a significant influence on EFV plasma concentrations in pre- vious reports (Dhoro et al., 2015; Meng et al., 2015). In total, 9 SNPs of CYP2B6 and 2 SNPs of ABCB1 were selected, including CYP2B6 171+967C>A (rs2099361), 516G>T (rs3745274), 785A>G (rs2279343), 171+3212C>T (rs4803415), 171+4335T>C (rs1872125), 1295–913G>A (rs7260329), 1355A>G (rs707265), 1421T>C (rs1042389) and 171+6023A>C (rs7250601), along with 2 SNPs of ABCB1 including 3435 C>T (rs1045642) and 2677 G>T/A (rs2032582).

All genotyping experiments were performed by Shanghai BioWing Applied Biotechnology (www.biowing.com.cn). Genomic DNA was isolated from peripheral blood using an AxyPrep-96 (AXYGEN) kit, and target DNA sequences amplified using a multiplex polymerase chain reaction (PCR) method. After PCR, genotyping was carried out using an oligonucleotide ligation detection reaction (LDR)-fluorescent micro- sphere assay. LDR conditions were as follows: 95 ◦C for 2 min, 94 ◦C for 30 s, and 50 ◦C for 2 min (35 cycles). Fluorescent LDR products were differentiated using ABI Sequencer 377. Templates containing two al- leles in each SNP were synthesized as positive controls. PCR-LDR find- ings were confirmed by sequencing the PCR products of 30 samples for each SNP.

2.4. Population pharmacokinetic modeling
2.4.1. Structure model building

The PK data of EFV were analyzed using the nonlinear mixed-effects modeling software NONMEM, version 7.4.1 (Icon Development Solu- tions, Hanover, MD, USA) and R, version 3.6.1. The first-order condi- tional estimation method with interaction (FOCE-I) was used in the overall process of model development. The one- and two-compartment models with first-order absorption and elimination were evaluated for the structure model development. The BSV was accessed using the exponential model. The residual unexplained variability (RUV) was sequentially tested using additive, proportional, and combined models. Throughout the modeling process, the minimal objective function value (OFV), goodness-of-fit plots, precision in parameter estimates, and scientific plausibility were comprehensively considered in the model

2.4.2. Covariate model building

After the structure model was established, the following covariates were investigated on CL/F and apparent volume of distribution (V/F) of EFV: weight, age, sex, ALT, ALP, TBIL, DBIL, ALB, GGT, CCR, HB,concomitant medication, and 9 SNPs of CYP2B6 including rs2099361, rs3745274, rs2279343, rs4803415, rs1872125, rs7260329, rs707265,rs1042389 and rs7250601, along with 2 SNPs of ABCB1 including rs1045642 and rs2032582. Meanwhile, the correlation among all covariates were tested, and none of them showed obvious correlation.For continuous covariates, including age, ALT, ALP, TBIL, DBIL, ALB, GGT, CCR, and HB, linear (Eq.1), power (Eq.2), and exponential (Eq.3) relations were tested, respectively. For categorical covariates, including gender, concomitant medication, and SNPs, a linear relation was eval-where P is the PK parameter estimate for a subject(s) with COVi, PTV is the typical value of parameter P, COVi is the continuous covariate of the ith subject, COVmedian is the median value of continuous covariate, COV is the categorical covariate, j is the category of COV and θ is the coef- ficient term to be estimated. For the most common category of COV, the θ is set to zero as the reference.
Every covariate was identified separately into the model using stepwise forward inclusion and backward elimination approach with the likelihood ratio test. Since the OFV follows a chi-square distribution, a covariate was considered significant when the addition of it decreased the OFV > 3.84 (p < 0.05; df = 1), and the backward elimination of it increased the OFV > 6.63 (p < 0.01; df = 1). The effects of each significant covariate on model parameters in the full covariate model were summarized using a forest plot. The forest plot was implemented with the ‘forestplot’ add-on package in R, version 1.9 and Adobe Illustrator, version CC 2018 (Adobe Systems, Inc., San Jose, CA, USA). In the final model, only covariates whose effect on the typical pop- ulation parameter was > 20% were retained. The 20% effect for a cat- egorical covariate was defined as the parameter estimate of the alternate value as compared with the reference; for a continuous covariate, the effect was defined as the point estimate of the parameter at either the 5th or 95th percentile of the tested covariate (Leil et al., 2010).

2.4.3. Model evaluation

Apart from goodness-of-fit plots, the final PPK model was evaluated using bootstrap, visual predictive check (VPC), and normalized predic- tion distribution error (NPDE) plots. The calculations of bootstrap and graphical visualizations of VPC were performed with PsN, version 4.9.0 and Pirana, version 2.9.9 (Certara USA, Inc., Princeton, NJ, USA). The NPDE was implemented with the NPDE add-on package in R, version 2.0.

To check the stability of the final model, a nonparametric bootstrap method of 1000 iterations was conducted. The 95% confidence intervals (CIs) for the bootstrap replicates were obtained and compared with parameter estimates from the final model.
For the VPC, we performed 2000 simulations and assessed the 5th, median, and 95th percentiles of the distributions of the simulated con- centration values with the observed data.

NPDE with 2000 simulations was performed for each observation in the original dataset from the final model. NPDEs and their distributions were evaluated to determine if the model adequately described the observed data. Moreover, plots of NPDEs versus observations and time were also evaluated to determine whether trends were present.

2.4.4. Simulations for dose optimization

To propose dose optimization, the final model was used to simulate steady-state EFV mid-dose concentrations at 12 h after dosing (C12) on 1000 individuals with different laboratorial and genetic data (Csajka et al., 2003; Olagunju et al., 2018; Alghamdi et al., 2019). Different dose regimens including 100, 200, 400, 600, 800, and 1000 mg once daily were simulated. Since the steady-state EFV concentrations below 1 μg/mL could increase the risk for virological failure and drug resistance, while concentrations above 4 μg/mL could increase the risk for the central nervous system adverse effects (Langmann et al., 2002; Marzo- lini et al., 2001), the optimal EFV dose should maintain the 25th to 75th of the steady-state C12 within 1–4 μg/mL (Csajka et al., 2003; Olagunju mg EFV once-daily for HIV therapy, and 23% of them were co-infected with tuberculosis and co-administrated with RFP.

3.2. Population pharmacokinetic modeling
3.2.1. Structure model building

As only sparse samples were collected in the absorption and distri- bution phase, the steady-state EFV plasma concentrations were described by the one-compartment model with first-order absorption and elimination without lag-time. The BSV was described by an expo- nential model supported for CL/F and V/F. The proportional error model revealed the RUV to be a well fit. The inclusion of an additive error did not result in significant model optimization, but it decreased model stability instead. The population parameter estimates of the structure model were summarized in Table 2. The mean population estimates for absorption rate constant (Ka), CL/F, and V/F (% relative standard error [RSE]) were 0.815 h — 1 (8.2%), 9.46 L/h (3.8%), and 185 L (4.1%),respectively. Meanwhile, the BSV on CL/F and V/F expressed by the coefficient of variation were 58.1% and 55.7%, respectively, and the RUV was 22.3%.

3.2.2. Covariate model building

The full covariate model was constructed after comprehensively assessing all the screened covariates. Weight-based allometric scaling (0.75 for CL/F; 1 for V/F) was added to the model first, which decreased the OFV by 10.483 points.CYP2B6 rs3745274 was secondly included in the model on CL/F as a categorical covariate, which caused a reduction of 129.453 in the OFV and a decrease of BSV on CL/F by 14.6%. Then, we subsequently iden- tified the other five covariates on CL/F, namely rs2099361 (ΔOFV= —34.408), rs2279343 (ΔOFV= —12.426), ALB (ΔOFV= —8.79), GGT (ΔOFV= —9.541) and RFP (ΔOFV= —7.776). These five covariates decreased the BSV of CL/F from 58.1% to 35.5%. Meanwhile, ALB was also identified as a significant covariate on V/F with a decrease of OFV about 14.342 and a decrease of BSV from 55.7% to 49.2%. The model building steps for full covariates model were summarized in Table S1. The full covariate model is shown below (Eq.5–7).
Full covariates model: Ka h—1) = 0.703 (5) BSV between subject variability, CL/F apparent clearance (L/h), Ka absorption rate constant (h — 1), RSE% the relative standard error (%), V/F apparent volume of distribution (L).

The population estimates of the final model were summarized in Table 2. The typical population parameter estimates of Ka, CL/F, and V/ F in the final model were similar to the estimates in the structure model. But the BSV values on CL/F and V/F in the final model were lower than that in the structure model.

3.2.3. Model evaluation

The goodness-of-fit plots of the structure model and the final model were shown in Figure S1 and Fig. 2, respectively. The distribution of conditional weighted residuals (CWRES) was generally uniform across the range of predicted concentrations and the time after dosing. For the final model, the observations vs. population and individual predictions of the final model were densely distributed on either side of the identity line, and the CWRES were also evenly distributed around the x-axis, which implied that the estimated model fitted the data well.

Fig 1. Covariate effects on PK parameters in the full covariate model. Categorical covariate effects (95% CI from bootstrap) are represented by open symbols (horizontal orange lines). Continuous covariate effects are represented by horizontal boxes and the 95% CI from bootstrap at the 5th/95th percentiles of the covariate are represented by the horizontal orange lines at the end of boxes. The open area of boxes represents the range of covariate effects from the 5th percentiles to 50th of the covariate. The blue area of boxes represents the range of covariate effects from the 50th to 95th percentiles of the covariate. For continuous covariates with blue boxes on the left of the reference line (100%), with values higher than the reference, the PK parameters were decreased (eg. for albumin with 53.7 g/L, the CL/F is 91.6% as referred at the left of blue box, while for albumin with 30.0 g/L, the CL/F is 121.8% as referred at the right of open box). The reference patient was with 70 kg, rs3745274 = GG, rs2099361 = GG, rs2279343 = AA, albumin = 46.2 g/L, γ-glutamyl transpeptidase = 50 g/L and not concomitant with rifampin. Parameter estimate in reference patient was considered to be 100% (vertical solid line), and dashed vertical lines are at 80% and 120% of this value.

The bootstrap analysis was successful in 99.6% of the 1000 runs, and the median estimated values of the bootstrap were within 10% of the values obtained in the final model (Table 2), which indicated that the final model was stable and robust when it was fitted to various combi- nations of concentration-time datasets.VPC for the final model showed that the predicted and observed data were in adequate agreement (Fig. 3). The 5th, median, and 95th per- centiles of the observed data generally fell within the respective 95% CIs.

Evaluation of the predictive performance using the NPDE analysis was shown in Fig. 4. Deviations from the identity line showed minimal departures from the expected distribution. The mean of NPDE was 0.196 (Wilcoxon signed-rank test: p<0.0001), the variance of NPDE was 1.04 (Fisher’s variance test: p = 0.385), and the NPDE density of the predictive model discrepancies followed a theoretical normal distribution (Shapiro–Wilk normality test: p<0.0001). The global adjusted p-value was smaller than 0.0001. No obvious trend was observed in the scatter plots, confirming that the final PPK model adequately described the observed data. 3.2.4. Simulations for dose optimization To provide optimal EFV dose according to CYP2B6 polymorphisms, patients were divided into three CYP2B6 metabolizer phenotypes based on their genotypes including extensive (rs3745274=GG, rs2099361=GG, rs2279343=GG), intermediate (rs3745274=GT, rs2099361=GT, rs2279343=GA), and slow metabolizer (rs3745274=TT, rs2099361=TT, rs2279343 = AA). Meanwhile, according to the ALB range (26.8–73.9 g/L) of patients enrolled in our study, simulated patients were set into different ALB levels including 30, 50, and 70 g/L. The distributions of simulated EFV C12 were presented in Fig. 5. The optimized dosage to maintain C12 within 1–4 μg/mL were 800 mg or 1000 mg once daily for CYP2B6 extensive phenotype patients, 400 mg or 600 mg once daily for intermediate phenotype patients, and 100 mg once daily for slow phenotype patients, regardless of their ALB levels. In addition, for extensive phenotype patients with ALB of 70 g/L and slow phenotype patients with ALB of 30 g/L, 600 mg once daily and 200 mg once daily were also recommended, respectively. 4. Discussion In this study, we identified three SNPs in CYP2B6, including rs2099361, rs3745274, and rs2279343, along with ALB and weight on EFV PK in Chinese HIV-infected adults by using the PPK approach. Among them, the influence of CYP2B6 rs2099361 and ALB on EFV PK was quantitatively explored for the first time. In addition, steady-state EFV C12 based on patient phenotypes and ALB levels was simulated to provide dose optimization. Fig 2. Diagnostic goodness-of-fit plots for the final model. A observed versus individual predicted concentration; B observed versus population predicted concen- tration; C conditional weighted residuals (CWRES) versus population predictions of the final model; D CWRES versus time after the last dose. The solid lines in A and B are identity lines, and the solid lines in C and D are zero lines. Since over 99% of EFV is bound to serum ALB, lower ALB levels are supposed to increase the distribution of EFV into periphery organs including the brain, and metabolic organs like liver, which may increase both the V/F and CL/F of EFV. In our study, ALB was identified as a significant covariate both on CL/F and V/F. For patients with ALB of 30 g/L, their CL/F was 24.3% higher, and V/F was 53.5% higher than the typical value with ALB of 46.2 g/L, while for patients with ALB of 70 g/ L, their CL/F and V/F were about 18.9% and 33.2% lower than the typical value, respectively. According to the simulation results in Fig. 5, patients with slow phenotype are recommended to take 100 mg once daily. But for these with low ALB level of 30 g/L, both 100 mg and 200 mg once daily were recommended. Therefore, for patients with low levels of ALB, such as cirrhosis of liver or nephropathy, the EFV dose regimen may need to be adjusted. The effects of CYP2B6 SNPs on CL/F of EFV have been investigated in previous studies (Alghamdi et al., 2019; Bienczak et al., 2016; Robarge et al., 2017; Salem et al., 2014; Sinxadi et al., 2015). However, the in- fluence of rs2099361 on EFV PK was never explored. The rs2099361 SNP is an intronic variant in CYP2B6, which has been reported to in- fluence nevirapine exposure (Bertrand et al., 2012; Vardhanabhuti et al., 2013). By using linear regression analyses, Bertrand et al. revealed that the rs2099361 showed a significant influence on the CL/F of nevirapine (Bertrand et al., 2012). Vardhanabhuti et al. used parametric regression models and revealed that rs2099361 showed a significant association with nevirapine exposure (Vardhanabhuti et al., 2013). In our study, we found that CL/F of EFV for GT/GG carriers of rs2099361 increased by 43.5% compared with TT carriers. Therefore, the mutation from G to T of rs2099361 impeded the metabolism of the EFV, which may be related to the false splicing due to intron mutation by rs2099361. As to the other two identified SNPs, the mutation of rs3745274 from G to T could cause aberrant splicing to reduce the protein expression of CYP2B6 (Hofmann et al., 2008), while the mutation of rs2279343 from A to G has been reported to increase the enzyme activity of CYP2B6 (Desta et al., 2019; Duarte et al., 2017). In our study, the CL/F for GT and TT carriers of rs3745274 was 41.5% and 74.6% lower than GG carriers, respectively, while the CL/F for GG/GA carriers of rs2279343 increased by 33.3% compared to AA carriers, which is consistent with previous studies (Desta et al., 2019; Duarte et al., 2017; Robarge et al., 2017; Salem et al., 2014). Other 6 investigated SNPs in CYP2B6 did not significantly decrease the OFV value and were not included in the covariate model. The effects of common ABCB1 SNPs on EFV PK are conflicting (Mukonzo et al., 2009; Elens et al., 2010; Fellay et al., 2002). Fellay and colleagues reported higher plasma exposure and better immunological outcomes of EFV associated with TT carriers of ABCB1 rs1045642 (Fellay et al., 2002). However, Mukonzo et al. and Laure et al. failed to replicate this association (Mukonzo et al., 2009; Elens et al., 2010). Similarly, our results did not reveal a significant influence of rs1045642 and rs2032582 in ABCB1 on the CL/F of EFV. The effects of ABCB1 SNPs on EFV PK should be further investigated by larger multicenter studies. Body weight was identified as a significant covariate for CL/F and V/ F of EFV in our study, which is consistent with previous studies (Luo et al., 2016; Dhoro et al., 2015; Hui et al., 2016). The weight-based allometric scaling factor for CL/F was fixed to be 0.75, which is similar to 0.67 estimated by Luo et al. (Luo et al., 2016). In serval PPK studies, gender was reported to influence the CL/F of EFV (Dhoro et al., 2015; Haas et al., 2009). But in our study, gender showed no significant influence either on CL/F or V/F, which may result from the unbalanced gender distribution in our study (338 males vs 38 females). The actual effect of gender on EFV PK still needs further exploration. Fig 3. Visual predictive check (VPC) of the final model. The dots represent observed concentrations; solid lines represent the median (red), 5th, and 95th percentiles (blue) of the observations, which are overlapped by the 90% confidence intervals for the median (red area) and the 5th and 95th percentiles (blue areas) of the simulated profile. Fig 4. Visual results of the normalized prediction distribution errors (NPDE) analysis. A Plots of theoretical quantiles versus sample quantiles; B plots of frequency versus NPDE; C plots of NPDE versus time post-dose (h); D plots of NPDE versus predicted concentrations. Fig 5. Boxplots of the distributions of simulated EFV concentrations at 12 h after dosing. HIV infected patients with different albumin levels from 30, 50, 70 g/L, and slow, intermediate, and extensive composite CYP2B6 phenotypes were set for the simulation. Dashed horizontal lines represent the therapeutic target range (1–4 μg/mL). In the full covariate model, the GGT and RFP both showed influence on CL/F. GGT indicates the degree of cholestasis. As EFV was mainly metabolized by the liver and eliminated by choler, higher GGT may indicate lower CL/F of EFV. However, for patients in our study with high level of GGT like the 95th percentage (328.6 U/L), their CL/F is only 16.3% lower than that with median GGT level of 46.2 U/L. RPF is an inducer of CYP2B6 and CYP3A4, which can decrease the exposure of EFV (Lo´pez-Cort´es et al., 2002; Matteelli et al., 2007). In our study, concomitant with RFP could increase the CL/F of EFV by 15.9%. Though GGT and RFP showed statistical significance on CL/F, their influence on CL/F was lower than 20% (Fig. 1); therefore, both were excluded from the final model. The low influence effect of concomitant with RFP on CL/F may result from the limited information collected in this study since only 23% of patients were concomitant with RFP. The exact in- fluence of GGT and concomitant with RFP need to be further investi- gated in clinical studies with a larger population. To provide individualized dosage, determining EFV regimens based on patient characteristics, especially SNPs in CYP2B6 was recommended (Desta et al., 2019). CPIC divided patients into five phenotypes ac- cording to their genotypes of rs3745274, rs2279343, rs28399499, rs34223104 and recommended initiating EFV with a decreased dose of 200 or 400 mg once daily for CYP2B6 poor metabolizer, 400 mg once daily for CYP2B6 intermediate metabolizer, and 600 mg once daily for CYP2B6 normal, rapid, and ultrarapid metabolizer (Desta et al., 2019). In our study, the dose recommended for patients in intermediate phenotype is similar to that by CPIC (Desta et al., 2019), but the range of recommended doses for slow and extensive phenotype patients is wider. The differences of CL/F among slow, intermediate, and extensive pa- tients according to genotypes of three SNPs included in our study may be larger, which may result in the different dose recommendations between our study and CPIC. Moreover, in our study, more individualized dos- ages for patients in slow and extensive phenotypes are recommended according to their ALB levels, which could be used for more precious dose recommendation. Meanwhile, the EFV concentrations of patients with slow phenotype showed higher differences from these with extensive or intermediate phenotypes in our study. Therefore, dose ad- justments are more important for those with slow phenotype to avoid excessive concentrations. The main limitation of this study is that only sparse concentrations of EFV were collected in each patient. The samplings on the absorption and distribution phase were limited, which may result in insufficient infor- mation for estimation of the lag time in the absorption phase and inability to describe the two-compartment distribution PK of EFV re- ported in the previous study (Robarge et al., 2017). However, the CL/F of EFV estimated in our study is similar to those reported in previous studies (Csajka et al., 2003; Cabrera et al., 2009). Since the CL/F is the most important parameter for dose optimization, one-compartment model is preferable when only sparse samples are available (Wu and Furlanut, 1998). Moreover, unbounded EFV concentrations were not detected in this study. The influence of ALB on unbounded EFV fraction and unbounded EFV clearance were not investigated.In conclusion, this study identified three SNPs in CYP2B6, including rs2099361, rs3745274, and rs2279343, along with ALB and weight as predictors of EFV PK variability. EFV dose individualization could be determined based on these covariates. CRediT authorship contribution statement Xian-min Meng: Writing – original draft, Methodology, Formal analysis. Zi-ran Li: Writing – original draft, Methodology, Formal analysis. Xin-yin Zheng: Writing – original draft, Formal analysis. Yi-xi Liu: Formal analysis. Wan-jie Niu: Formal analysis. Xiao-yan Qiu: Writing – original draft, Methodology. Hong-zhou Lu: Methodology. Declaration of Competing Interest All authors declared no potential conflicts. Funding This study was supported by the program for the 12th Five-year Plan, the People’s Republic of China (NO: 2012ZX10001003), and Shanghai municipal natural science foundation (14ZR1434900). Supplementary materials Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.ejps.2021.105986. References Arts, E.J., Hazuda, D.J., 2012. HIV-1 Antiretroviral Drug Therapy. Cold Spring Harbor Perspectives in Medicine 2 (4). Vo, T.T., Gupta, S.V., 2016. Role of cytochrome P450 2B6 Pharmacogenomics in determining efavirenz-mediated central nervous system toxicity, treatment outcomes, and dosage adjustments in patients with human immunodeficiency virus infection. Pharmacotherapy 36 (12), 1245–1254. Rakhmanina, N.Y., van den Anker, J.N., 2010. Efavirenz in the therapy of HIV infection. Expert Opin. Drug Metab. Toxicol. 6 (1), 95–103. Sheran, M., 2005. The nonnucleoside reverse transcriptase inhibitors efavirenz and nevirapine in the treatment of HIV. HIV Clin. Trials 6 (3), 158–168. World Health Organization, Consolidated Guidelines On the Use of Antiretroviral Drugs For Treating and Preventing HIV infection: What’s New. November 2015. Csajka, C., et al., 2003. Population pharmacokinetics and effects of efavirenz in patients with human immunodeficiency virus infection. Clin. Pharmacol. Ther. 73 (1), 20–30. Luo, M., et al., 2016. Population Pharmacokinetics Analysis To Inform Efavirenz Dosing Recommendations in Pediatric HIV Patients Aged 3 Months to 3 Years. Antimicrob. Agents Chemother. 60 (6), 3676–3686. Habtewold, A., et al., 2017. Population Pharmacokinetic Model Linking Plasma and Peripheral Blood Mononuclear Cell Concentrations of Efavirenz and Its Metabolite, 8-Hydroxy-Efavirenz, in HIV Patients. Antimicrob. Agents Chemother. 61 (8). Guti´errez, F., et al., 2005. Prediction of neuropsychiatric adverse events associated with long-term efavirenz therapy, using plasma drug level monitoring. Clin. Infect. Dis. 41 (11), 1648–1653. Langmann, P., et al., 2002. Efavirenz plasma levels for the prediction of treatment failure in heavily pretreated HIV-1 infected patients. Eur. J. Med. Res. 7 (7), 309–314. Marzolini, C., et al., 2001. Efavirenz plasma levels can predict treatment failure and central nervous system side effects in HIV-1-infected patients. AIDS 15 (1), 71–75. Vrouenraets, S.M., et al., 2007. Efavirenz: a review. Expert Opin. Pharmacother. 8 (6), 851–871. Avery, L.B., et al., 2013. Increasing extracellular protein concentration reduces intracellular antiretroviral drug concentration and antiviral effect. AIDS Res. Hum. Retroviruses 29 (11), 1434–1442. Ward, B.A., et al., 2003. The cytochrome P450 2B6 (CYP2B6) is the main catalyst of efavirenz primary and secondary metabolism: implication for HIV/AIDS therapy and utility of efavirenz as a substrate marker of CYP2B6 catalytic activity. J. Pharmacol. Exp. Ther. 306 (1), 287–300. Arab-Alameddine, M., et al., 2009. Pharmacogenetics-based population pharmacokinetic analysis of efavirenz in HIV-1-infected individuals. Clin. Pharmacol. Ther. 85 (5), 485–494. Dhoro, M., et al., 2015. CYP2B6×6, CYP2B6×18, Body weight and sex are predictors of efavirenz pharmacokinetics and treatment response: population pharmacokinetic modeling in an HIV/AIDS and TB cohort in Zimbabwe. BMC Pharmacol Toxicol 16, 4. Hui, K.H., Lee, S.S., Lam, T.N., 2016. Dose Optimization of Efavirenz Based on Individual CYP2B6 Polymorphisms in Chinese Patients Positive for HIV. CPT Pharmacometrics Syst Pharmacol 5 (4), 182–191. Olagunju, A., et al., 2018. Evaluation of universal versus genotype-guided efavirenz dose reduction in pregnant women using population pharmacokinetic modelling. J. Antimicrob. Chemother. 73 (1), 165–172. Desta, Z., et al., 2019. Clinical Pharmacogenetics Implementation Consortium (CPIC) Guideline for CYP2B6 and Efavirenz-Containing Antiretroviral Therapy. Clin. Pharmacol. Ther. 106 (4), 726–733. Swart, M., et al., 2012. ABCB1 4036A>G and 1236C>T Polymorphisms Affect Plasma Efavirenz Levels in South African HIV/AIDS Patients. Front Genet 3, 236.
Ngaimisi, E., et al., 2013. Importance of ethnicity, CYP2B6 and ABCB1 genotype for efavirenz pharmacokinetics and treatment outcomes: a parallel-group prospective cohort study in two sub-Saharan Africa populations. PLoS One 8 (7), e67946.
Mukonzo, J.K., et al., 2009. A novel polymorphism in ABCB1 gene, CYP2B6×6 and sex
predict single-dose efavirenz population pharmacokinetics in Ugandans. Br. J. Clin. Pharmacol. 68 (5), 690–699.
Duarte, H., et al., 2017. Population Approach to Efavirenz Therapy. J. Pharm. Sci. 106 (10), 3161–3166.
Elens, L., et al., 2010. Influence of host genetic factors on efavirenz plasma and intracellular pharmacokinetics in HIV-1-infected patients. Pharmacogenomics 11 (9), 1223–1234.
Li, Z.R., et al., 2021. Population Pharmacokinetics of Levetiracetam: a Systematic Review. Clin. Pharmacokinet.
Villani, P., et al., 1999. High-performance liquid chromatography method for analyzing the antiretroviral agent efavirenz in human plasma. Ther. Drug Monit. 21 (3), 346–350.
Meng, X., et al., 2015. Effect of CYP2B6 Gene Polymorphisms on Efavirenz Plasma Concentrations in Chinese Patients with HIV Infection. PLoS One 10 (6), e0130583.
Leil, T.A., et al., 2010. Quantification of apixaban’s therapeutic utility in prevention of venous thromboembolism: selection of phase III trial dose. Clin. Pharmacol. Ther. 88 (3), 375–382.
Alghamdi, W.A., et al., 2019. Population pharmacokinetics of efavirenz in HIV and TB/ HIV coinfected children: the significance of genotype-guided dosing. J. Antimicrob. Chemother. 74 (9), 2698–2706.
Bienczak, A., et al., 2016. The impact of genetic polymorphisms on the pharmacokinetics of efavirenz in African children. Br. J. Clin. Pharmacol. 82 (1), 185–198.
Robarge, J.D., et al., 2017. Population Pharmacokinetic Modeling To Estimate the Contributions of Genetic and Nongenetic Factors to Efavirenz Disposition.
Antimicrob. Agents Chemother. 61 (1).
Salem, A.H., Fletcher, C.V., Brundage, R.C., 2014. Pharmacometric characterization of efavirenz developmental pharmacokinetics and pharmacogenetics in HIV-infected children. Antimicrob. Agents Chemother. 58 (1), 136–143.
Sinxadi, P.Z., et al., 2015. Pharmacogenetics of plasma efavirenz exposure in HIV- infected adults and children in South Africa. Br. J. Clin. Pharmacol. 80 (1), 146–156.
Bertrand, J., et al., 2012. Multiple genetic variants predict steady-state nevirapine clearance in HIV-infected Cambodians. Pharmacogenet Genomics 22 (12), 868–876.
Vardhanabhuti, S., et al., 2013. Clinical and genetic determinants of plasma nevirapine exposure following an intrapartum dose to prevent mother-to-child HIV transmission. J. Infect. Dis. 208 (4), 662–671.
Hofmann, M.H., et al., 2008. Aberrant splicing caused by single nucleotide polymorphism c.516G>T [Q172H], a marker of CYP2B6×6, is responsible for decreased expression and activity of CYP2B6 in liver. J. Pharmacol. Exp. Ther. 325
(1), 284–292.
Fellay, J., et al., 2002. Response to antiretroviral treatment in HIV-1-infected individuals with allelic variants of the multidrug resistance transporter 1: a pharmacogenetics study. Lancet 359 (9300), 30–36.
Haas, D.W., et al., 2009. Associations between CYP2B6 polymorphisms and pharmacokinetics after a single dose of nevirapine or efavirenz in African americans.
J. Infect. Dis. 199 (6), 872–880.
Lo´pez-Cort´es, L.F., et al., 2002. Pharmacokinetic interactions between efavirenz and rifampicin in HIV-infected patients with tuberculosis. Clin. Pharmacokinet. 41 (9), 681–690.
Matteelli, A., et al., 2007. Multiple-dose pharmacokinetics of efavirenz with and without the use of rifampicin in HIV-positive patients. Curr. HIV Res. 5 (3), 349–353.
Cabrera, S.E., et al., 2009. Influence of the cytochrome P450 2B6 genotype on population pharmacokinetics of efavirenz in human immunodeficiency virus patients.
Antimicrob. Agents Chemother. 53 (7), 2791–2798.
Wu, G., Furlanut, M., 1998. Prediction of serum vancomycin concentrations using one-, two- and three-compartment models with implemented population pharmacokinetic parameters and with the Bayesian method. J. Pharm. Pharmacol. 50 (8), 851–856.