Plasma and Intracellular Population Pharmacokinetic Analysis of Tenofovir in HIV-1-Infected Patients

Abstract

The relationships among the dose of tenofovir disoproxil fumarate (TDF), tenofovir (TFV) plasma concentrations, and intracellular TFV diphosphate (TFV-DP) concentrations are poorly understood. Our objective was to characterize TFV and TFV-DP relationships. Data were pooled from two studies in HIV-infected persons (n = 55) on stable antiretroviral therapy. TFV and TFV-DP were measured with validated liquid chromatography/tandem mass spectrometry (LC/MS/MS) methods. Nonlinear mixed effects modeling (NONMEM 7) was used to develop the population model and explore the influence of covariates on TFV. A sequential analysis approach was utilized. A two-compartment model with first-order absorption best described TFV PK (FOCEI). An indirect stimulation of response model best described TFV-DP, where formation of TFV-DP was driven by plasma TFV concentration. Final plasma population estimates were as follows: absorption rate constant, 1.03 h⁻¹; apparent clearance (CL/F), 42 liters/h (33.5% interindividual variability [IIV]); intercompartment clearance, 181 liters/h; apparent central distribution volume (Vc/F), 273 liters (64.8% IIV); and apparent peripheral distribution volume (Vp/F), 440 liters (46.5% IIV). Creatinine clearance was the most significant covariate on CL/F and Vc/F. The correlation between CL/F and Vc/F was 0.553. The indirect response model for TFV-DP resulted in estimates of the maximal intracellular concentration (E_max), the TFV concentration producing 50% of E_max (EC₅₀), and the intracellular elimination rate constant (k_out) of 300 fmol/10⁶ cells (82% IIV), 100 ng/ml (106% IIV), and 0.008 h⁻¹, respectively. The estimated k_out gave an 87-h TFV-DP half-life. A predictive check assessment indicated satisfactory model performance. This model links formation of TFV-DP with plasma TFV concentrations and should facilitate more informed investigations of TFV clinical pharmacology.

INTRODUCTION

Tenofovir disoproxil fumarate (TDF) is a nucleotide analogue used as a component of combination therapy with other antiretrovirals for the treatment of human immunodeficiency virus type 1 (HIV-1) infection. All nucleoside/nucleotide reverse transcriptase inhibitors (NRTIs) require intracellular conversion to the triphosphate anabolite, which inhibits the viral reverse transcriptase. Unlike nucleosides that require sequential phosphorylation to the mono-, di-, and finally triphosphate for activation, nucleotides such as tenofovir (TFV) require only two such steps because they are monophosphate analogs and do not require the initial phosphorylation reaction. Conventionally, the triphosphate anabolite of TFV is referred to as tenofovir diphosphate (TFV-DP) because only two phosphate groups are added. As the initial phosphorylation step is potentially a limiting factor in the activation of NRTIs in resting CD4⁺ cells and macrophages, TFV may result in better antiviral activity than other NRTIs in cells that have limited proliferative and phosphorylative capacities. Since TFV is activated to TFV-DP in both active and resting cells (20), it has the property of inhibiting HIV replication in macrophages and other nondividing cells not found uniformly among the NRTI class (14).

A central feature of the clinical pharmacology of NRTIs is the dependence of virologic response upon the pharmacokinetics of the intracellular triphosphate anabolite. The lack of knowledge of intracellular NRTI-triphosphate pharmacokinetics contributed to the initial use of higher doses and more frequent dosing and the eventual dose de-escalation of early NRTIs such as didanosine, stavudine, and zidovudine. The lack of knowledge of relationships between extracellular concentrations of the NRTIs and intracellular amounts of their active metabolites remains a significant gap in our knowledge of these agents and limits our ability to fully understand exposure-response relationships and use that information to predict responses to alternative dosing regimens, concomitant disease states, and other drugs. The objective of the present work was to develop a model to characterize the plasma and intracellular pharmacokinetics of TFV and TFV-DP in HIV-infected patients on stable antiretroviral therapy.

MATERIALS AND METHODS

Study design, subjects, and bioanalytical methods.

Data for this analysis were pooled from two different studies in HIV-infected persons. The first was a single-center two-group study (known as 1427; n = 30) of TFV pharmacokinetics and renal clearance in persons aged 25 to 60 years taking lopinavir/ritonavir versus those not receiving a protease inhibitor (PI) (16). The second study was conducted by the Adolescent Trials Network for HIV/AIDS Interventions (protocol ATN056; n = 25) and characterized TFV and ritonavir-boosted atazanavir pharmacokinetics in adolescents and young adults aged 18 to 25 years (17). In both studies, the dose of TDF was 300 mg once daily, which was given following a meal. Steady-state plasma concentrations of TFV were measured at 8 or 11 different sampling times based on the study (1 sample before TDF administration and 7 or 10 for ATN056 and 1427, respectively, at specified times throughout the 24-h dosing interval, including a 24-h postdose sample) from 55 patients (36 male, 19 female). Among the 55 patients, 51 also had measurements of intracellular TFV-DP concentrations at three different sampling times, either predose and 5 and 24 h postdose (1427), or 1, 4, and 24 h postdose (ATN056). A total of 529 plasma TFV and 151 intracellular TFV-DP samples were collected. The complete study methods and primary results have been published (16, 17). The institutional review boards at each site recruiting subjects approved both studies, and all subjects provided written informed consent.

For both the studies, plasma concentrations of TFV and intracellular TFV-DP in peripheral blood mononuclear cells (PBMCs) were quantified by validated liquid chromatography/tandem mass spectrometry (LC/MS/MS) procedures, as previously described, in the same laboratory (4, 5, 15). For plasma, the method was linear from 10 to 750 ng/ml, with a minimum quantifiable limit of 10 ng/ml when 0.25-ml aliquots were analyzed. The accuracy and precision were within ±15%. The assay for quantification of intracellular TFV-DP concentrations in PBMCs was linear in the range of 50 fmol to 10,000 fmol per sample. The minimal quantifiable limit was 10 fmol/million cells when 5 million cells were analyzed. The accuracy and precision were within ±15%.

Population pharmacokinetic modeling.

Pharmacokinetic modeling was accomplished using a nonlinear mixed effects approach. A first-order conditional estimation method with interaction (FOCEI) using the nonlinear mixed effects modeling (NONMEM) system (NONMEM Version 7 and PDx-Pop Version 4.1; GloboMax LLC, Hanover, MD) (1) was used. Xpose (12) and R (http://www.r-project.org/) were used for goodness-of-fit assessment and model evaluation. Several structural models were investigated to characterize TFV plasma concentrations. Classical one-, two-, and three-compartment models were evaluated. Several error models were also investigated (i.e., proportional, exponential, and additive random effects models) to describe interpatient and residual variability. Selection of the structural pharmacokinetic model and residual error model was driven by the data and based on goodness-of-fit plots, successful convergence, plausibility, precision of parameter estimates, and the objective function value.

A model building process was employed to examine the influence of patient covariates on the estimates of pharmacokinetic parameters following identification of the basic model. The effects of the following patient covariates on TFV pharmacokinetics were evaluated: demographic covariates (age, sex, weight, and race), estimated creatinine clearance (CrCL) (method of Cockcroft and Gault [2]), total bilirubin (TBIL), and concomitant PI therapy. Potential covariates were selected by univariate analysis, testing the effect of each covariate on each of the relevant pharmacokinetic parameters. A decrease in the minimum value of objective function of 3.841 or greater following introduction of a single covariate into the “base” model was considered statistically significant (P < 0.05 with 1 degree of freedom), using the χ² distribution, if the 95% confidence intervals (CI) for the estimate did not include the null value. All significant covariate-parameter relationships were entered in the model in a stepwise fashion (stepwise forward addition) retaining the most significant covariate and adding the next most significant covariate that can result in a decrease in objective function value of at least 6.64 points (χ²; P < 0.01 with 1 degree of freedom). A backward elimination process was then employed to eliminate covariates from the full model in order to develop the final model. Backward elimination was performed by removal of a covariate from the full model one at a time; an increase in the objective function of 7.864 or greater (P < 0.05 with 1 degree of freedom) on removal of a covariate from the full model signified the variable was important, and that covariate was retained in the final model. For continuous covariates, either a centered-around-median or a power function approach was used to add the covariate to the respective pharmacokinetic parameter. For categorical covariates, a linear function was used.

Acceptable population models resulted in successful minimization, a successful estimation of the covariance, and the absolute value of last iteration gradients greater than 0 but smaller than 100. Confidence intervals of structural parameters did not include zero; absolute value of correlation between two structural parameters was not greater than 0.95. Acceptable models did not lead to trends in the distribution of weighted residuals versus model predictions and versus independent variables. The predicted versus observed data were evenly distributed around the unit line. If the weighted residual was greater than 5, that datum point was excluded from analysis.

Several models for sequential or simultaneous analysis of plasma and intracellular concentrations of TFV were tested using a population modeling approach, including an effect compartment model, a plasma and intracellular link model, and an indirect response model. A sequential analysis approach was ultimately selected because no quantitative information on cellular rate constants of TFV transport, anabolism, and catabolism in humans existed (prior to this work), and we only had information on TFV-DP and not, for example, TFV-monophosphate concentrations. The sequential analysis approach allowed an accurate determination of the individual post hoc pharmacokinetic parameters from the final plasma model (which could be compared with historical data) and then used these fixed values to estimate parameters of the intracellular model. The first-order conditional estimation method (FOCE) in NONMEM 7 with $DES block was used (ADVAN8 TRANS1). The schematic presentation of the two-compartment open pharmacokinetic model for TFV in plasma linked with a stimulatory indirect response model for intracellular TFV-DP is shown in Fig. 1.

Fig. 1.

Schematic presentation of a two-compartment open pharmacokinetic model coupled with a stimulatory indirect pharmacodynamic response model. Cp, plasma TFV concentration; k_a, first-order rate constant for TFV absorption; k_in, apparent zero-order rate constant for the production of intracellular TFV-DP concentration; k_out, first-order rate constant for the loss of TFV-DP; E_max, maximal intracellular concentration; EC₅₀, plasma TFV concentration that produces 50% of E_max.

An indirect stimulation of response model (3, 21) was ultimately selected because of its biologic relevance and the ability to simplify the intracellular pathway as described above. The basic premise of the indirect response model is that a measured response (R) to a drug may be produced by indirect mechanisms. Factors controlling the production (k_in) of the response variable are stimulated, or the determinants of loss (k_out) of the response variable may be stimulated. For our purpose, the response variable (R) measure is intracellular TFV-DP concentration. The rate of change of intracellular concentration with time can be described by the following:

$$\frac{dR}{dT}=k_{\mathrm{in}}-k_{\mathrm{out}}·R$$

where k_in represents the zero-order constant for production of the response and k_out defines the first-order rate constant for loss of the response. It is assumed that k_in and k_out fully account for production and loss of the response. According to this model, the formation of intracellular TFV-DP concentrations may be obtained by assuming that plasma TFV concentrations stimulate k_in and hence an indirect response whereby TFV plasma concentrations increase intracellular TFV-DP formation by stimulating k_in according to the stimulatory function S(t):

$$S\left(t\right)=1+\frac{E_{\max }·\mathrm{Cp}}{{\mathrm{EC}}_{50}+\mathrm{Cp}}$$

In this model, E_max represents the maximal intracellular concentration of TFV-DP, EC₅₀ represents the plasma concentration producing 50% of E_max, and Cp is the plasma TFV concentration. No conversion of TFV-DP concentrations to the same units as plasma was made because of the assumptions and parameterization of the model, the similar measurement scales, and because such a conversion would require application of a standard cell volume across all patients and thus would only scale and not impact estimation of k_in. The equations for the change in intracellular concentrations are provided above.

Model evaluation.

The bootstrap and visual predictive check (VPC) methods were used for model evaluation. The bootstrap approach was applied to assess the reliability of the final TFV plasma parameter estimates and their 95% confidence interval (CI). One thousand data sets were simulated using the final plasma parameter estimates to obtain the lower and upper boundaries of the bootstrap 95% CI and were then compared with the values obtained from NONMEM. For the VPC, a 90% prediction interval was determined from the 5th and 95th percentiles of the simulated dependent data at each time point and compared with the original data. A total of 100 simulations were done for VPC for both plasma and intracellular models.

RESULTS

Population pharmacokinetics of TFV.

The observed plasma TFV concentrations are shown in Fig. 2. The demographic distribution of patient population is presented in Table 1. A two-compartment model with first-order absorption best described plasma TFV concentrations. An exponential error model described interpatient random effects. Due to data limitations, the variance component for apparent intercompartmental clearance (Q/F) and absorption rate (k_a) was not estimated. Residual error was modeled as a proportional error model. This model was implemented using the PREDPP subroutine ADVAN4 TRANS4. The model-building steps for the covariate analysis in NONMEM are summarized in Table S1 in the supplemental material. The influence of covariates was not tested on Q/F and k_a. As presented in Table S1, age, CrCL, and PI on apparent clearance (CL/F), age, CrCL, and PI on apparent central distribution volume (Vc/F), and age and PI on apparent peripheral distribution volume (Vp/F) were significant covariates. Per the modeling methodology, stepwise forward addition was employed where the most significant covariate (age∼CL/F) was retained and other covariates were added. Among the significant relationships, age and CrCL were significant on CL/F and for final model development only CrCL on CL/F was utilized, as age is a variable in calculation of CrCL. Following backward elimination, the significant covariates were CrCL on CL/F and Vc/F, and they were retained for the final model. The results from the final model with the results for interindividual variability (IIV), the correlation between parameters, and the residuals are presented in Table 2. The final model did result in better fit than the base model and also resulted in lower variability for CL/F and Vc/F. The residual variability did not change in the final model and remained at 18%. Due to correlations between IIVs related to key pharmacokinetic parameters following the final covariate model, the off-diagonal elements of the covariance matrix were estimated. The correlation between CL/F and Vc/F was 0.553. Goodness-of-fit plots from the base model and the final model are presented in Fig. S1 in the supplemental material. The final model shows an improvement of fit over that of the base model.

Fig. 2.

Plasma concentrations of TFV and the visual predictive check (VPC) for the final model for plasma TFV.

Table 1.

Characteristics of study patients

Parameter	Result for group
	Overall (n = 55)	1427 (n = 30)	ATN056 (n = 25)
Continuous variablesa
Age (yr)	33 (32), 18.6–60	41.5 (39.5), 25–60	22.8 (22.9), 18.6–25
Body wt (kg)	79.1 (74.7), 38.7–131.6	79.6 (78), 38.7–121.4	78.6 (71.4), 46.9–131.6
CrCL (ml/min)b	112.5 (108), 43.2–227.1	97.8 (95.1), 43.2–154.9	130.2 (127.1), 65.7–227.1
Total bilirubin (mg/dl)	1.1 (0.9), 0.2–4.7	0.7 (0.6), 0.2–1.4	1.7 (1.4), 0.4–4.7
Categorical variablesc
Gender
Male	36 (65%)	22 (73%)	14 (56%)
Female	19 (35%)	8 (27%)	11 (44%)
Race/ethnicity
Black/African American	23 (42%)	6 (20%)	17 (68%)
White	25 (45%)	23 (77%)	2 (8%)
Asian	1 (2%)	1 (3%)	0
Other	5 (9%)	0	5 (20%)
Unknown	1 (2%)	0	1 (4%)
Hispanic/Latino	7	1	6
PId
ATV	25 (50%)	0	25 (100%)
LPV	15 (25%)	15 (50%)	0
No PI	15 (25%)	15 (50%)	0

^aMean (median), range.

^bEstimated by the method of Cockcroft and Gault (2).

^cn (%).

^dPI, protease inhibitor; ATV, atazanavir; LPV, lopinavir.

Table 2.

Final plasma parameter estimates^a

Parameter	Result for parameter estimated (95% CI)
	Base model	Final model	Final model bootstrap
CL/F (liters/h)	42.1 (37.9–46.3)	42.0 (38.2–45.8)	41.8 (38.0–45.8)
Vc/F (liters)	224 (131–317)	277 (183–370)	279 (181–407)
Q/F (liters/h)	175 (141–209)	182 (150–213)	176 (140–213)
Vp/F (liters)	512 (426–598)	436 (348–523)	434 (331–534)
ka (h−1)	0.822 (0.491–1.15)	1.05 (0.618–1.481)	1.11 (0.696–2.04)
CrCL∼CL/F	NA	0.489 (0.273–0.704)	0.506 (0.257–0.771)
CrCL∼Vc/F	NA	1.01 (0.58–1.43)	1.08 (0.521–1.74)
IIVCL/F	35.9% (28.6%–42.0%)	33.5% (27.5%–38.5%)	32.8% (27.4%–38.5%)
CovCL/F∼Vc/F	NA	0.118 (R = 0.553)	0.115
IIVVc/F	79.4% (53%–98.9%)	64.8% (41.1%–81.9%)	64.3% (43.1–85.7%)
CovVc/F∼Vp/F	NA	−0.0415 (R = −0.139)	−0.0398
IIVVp/F	41.7% (22.8%–54.4%)	46.5% (29.0%–59.0%)	47.6% (32.6%–63.5%)
CovCL/F∼Vp/F	NA	0.113 (R = 0.731)	0.111
RUV (% CV)	18.3% (15.9%–20.3%)	18.3% (16.0%–20.3%)	18.3% (16.0%–20.5%)

^aCI, confidence interval; CL/F, apparent clearance; Cov, covariance; % CV, percentage coefficient of variation; CrCL, estimated creatinine clearance; k_a, absorption rate constant; IIV, interindividual variability; NA, not applicable; Q/F, apparent intercompartmental clearance; RUV, residual unexplained variability; R, correlation; Vc/F, apparent central compartment volume of distribution; Vp/F, apparent peripheral compartment volume of distribution; ∼, influence of, for example CrCL∼CL/F means the influence of creatinine clearance on apparent clearance.

The visual predictive check (VPC) and a stratified nonparametric bootstrap indicated that the final model for plasma TFV provided a reliable description of the data with good precision of structural model and variance parameter estimates. The VPC demonstrated that 12% of the observed TFV concentrations were outside the 90% CI. As shown in Fig. 2, the percentage of observations outside the 90% CI is due to the inability of the model to predict relatively high and low concentrations, particularly around the time of maximum concentration. The reliability of the final population pharmacokinetic model was also confirmed by reestimating the model parameter estimates and their 95% CI using a nonparametric bootstrap approach. Bootstrap analysis (902 converged runs) resulted in close to normal distributions of the parameter estimates, which were in good agreement with NONMEM for all parameter estimates. The parameter estimates for the fixed effects and the random effects from the bootstrap procedure are comparable and within approximately 5% of the estimates from NONMEM, which indicate that the final model is robust (6). On the basis of these results, we conclude the model describes the plasma TFV data well.

Population pharmacokinetics of TFV-DP.

The distribution of observed intracellular TFV-DP concentrations is presented in Fig. 3. Interindividual variability was described only for EC₅₀ and E_max. The residual error model was a proportional model. Table 3 lists the model parameter estimates and the 95% CIs of the structural parameters based on 100 bootstrap replicates, as the NONMEM estimates included zero. The 95% CIs for the variance terms are also estimated by bootstrap and presented in the table. Covariate analysis was not performed for the intracellular model.

Fig. 3.

Intracellular concentrations of TFV-DP and the visual predictive check (VPC) for the final model for intracellular TFV-DP.

Table 3.

Final intracellular parameter estimates^a

Parameter	Parameter estimated (95% CI),b final model
kin (h−1)	0.276 (0.0145–1.56)
kout (h−1)	0.00808 (0.0007–0.0372)
EC50 (ng/ml)	99.9 (1.000–403)
Emax (fmol/106 cells)	300 (4.000–484)
IIVEC50	106% (11%–635%)
IIVEmax	82.3% (0.54%–153%)
RUV (% CV)	56.7% (46%–63%)

^aCI, confidence interval; % CV, percentage coefficient of variation; E_max, maximal intracellular concentration of TFV-DP; EC₅₀, plasma concentration of TFV producing 50% of E_max; k_in, zero-order constant for production of the response; k_out, first-order rate constant for loss of the response; IIV, interindividual variability; RUV, residual unexplained variability.

^b95% CI estimated by bootstrap simulation.

The indirect response model for the intracellular TFV-DP concentrations resulted in estimates of E_max, EC₅₀, k_out, and k_in of 300 fmol/10⁶ cells (82% IIV), 100 ng/ml (106% IIV), 0.008 h⁻¹, and 0.276 h⁻¹, respectively. The residual variability was 56%. The estimates of interindividual variability for EC₅₀ and E_max were high. The estimated typical intracellular half-life of TFV-DP was 87 h [ln(2)/k_out].

Model evaluation using VPC revealed that the final model provided a reliable description of the data with good precision of structural model and variance parameter estimates. Data sets were simulated based on the fixed- and random-effect estimates from the final model. The predictive check was adequate because the majority of the observations were shown to lie within the 90% prediction interval (Fig. 3).

DISCUSSION

We have presented a description of the plasma pharmacokinetics of TFV and the intracellular pharmacokinetics of TFV-DP. The population pharmacokinetic parameters for TFV compared well with published values in the literature, and they showed that TFV CL/F is dependent upon renal function and age. The formation of intracellular concentrations of TFV-DP was modeled as a function of plasma concentrations. The estimated plasma TFV concentration that drives TFV-DP formation at half maximum and the estimated TFV-DP maximum are plausible; the elimination half-life of TFV-DP was 87 h, consistent with in vitro and in vivo data (10, 18–20). The statistical evaluation of the plasma and intracellular models provided strong support for a reliable description of the data.

The plasma concentrations of TFV were best described by a two-compartment model with first-order absorption similar to that in the model presented before (7, 13). Overall, the estimates for CL/F and Vc/F and the variability associated with them compared very well with literature values. For example, Jullien et al. reported a population estimate for TFV CL/F of 50.5 liters/h (after correction of TFV dose from TFV disoproxil to TFV) in 193 adult HIV-infected persons who had a mean CrCL of 89.8 ml/min (13). In our study CrCL had a significant influence on TFV CL/F and Vc/F. This finding is expected. TFV is eliminated primarily by renal excretion through a combination of glomerular filtration and tubular secretion. CrCL reasonably approximates glomerular filtration and can be estimated from serum creatinine, with consideration for age, sex, ethnicity, and body size because they are associated with creatinine excretion. The effect of age on TFV CL/F can be seen in the data from the two populations in this study. The mean age in the ATN056 participants was 22.8 years, and their estimated CrCL was 130.2 ml/min, versus the 1427 population, who had a mean age of 41.5 years and a mean CrCL of 97.8 ml/min. The mean TFV CL/F in the ATN056 subjects was 54.1 liters/h and was 37.4 liters/h in the 1427 study. This age-associated decline in TFV CL/F underscores the importance of monitoring renal function in patients who receive TDF and dose adjustments if CrCL is reduced.

The primary objective of the present work was to investigate the relationship between plasma TFV and intracellular TFV-DP concentrations. A total of 151 TFV-DP concentrations from 51 subjects were pooled for this analysis. TFV-DP concentrations ranged from 10.6 to 414 fmol/10⁶ cells with a mean value of 90.7 fmol/10⁶ cells. An indirect, stimulation of response, E_max model was found to describe well the formation of TFV-DP as a function of plasma TFV. The plasma concentration driving the reaction at half maximum was 99.9 ng/ml, and the TFV-DP E_max was 300 fmol/10⁶ cells. The elimination half-life of TFV-DP was 87 h. These modeled parameters have biologic and pharmacologic plausibility. The observed TFV-DP concentrations are comparable to those reported in other smaller data sets, where the average TFV-DP concentrations were in the range of 80 to 160 fmol/10⁶ cells (8, 10, 18, 19, 22). The estimated parameters of our model would predict TFV-DP concentrations of 128 to 174 fmol/10⁶ cells using a typical trough (72 ng/ml) or steady-state average (135 ng/ml) plasma concentration of TFV, also consistent with the values reported by other investigators. The model would allow for an increase in systemic exposure to increase intracellular concentrations, as has been suggested by Pruvost on the basis of work investigating an intracellular interaction of TFV-DP in patients who received lopinavir (19). The elimination half-life is consistent with in vitro data that reported a half-life of 50 h in resting lymphocytes and with in vivo data from small numbers of patients that found values from 60 to 180 h (10, 18–20).

The antiretroviral activity of all NRTIs is a function of the intracellular triphosphate anabolite, but quantification is analytically challenging. Thus, the ability to predict intracellular TFV-DP concentrations based upon knowledge of plasma TFV concentrations would aid in understanding exposure-response relationships. Additionally, interactions with NRTIs may arise at both the systemic and cellular levels and an ability to predict the effect of an increase or decrease in plasma concentrations on TFV-DP concentrations and the associated response may improve dose recommendation guidance and patient management. For example, the PIs atazanavir, darunavir, and lopinavir all increase TFV plasma concentrations between 22 and 32%. Among 146 HIV-infected persons receiving treatment with TDF, coadministration with a ritonavir-boosted PI was associated with a greater decrease in renal function than was coadministration with a nonnucleoside reverse transcriptase inhibitor (NNRTI) (9). The ability to model intracellular TFV-DP concentrations based on knowledge of plasma TFV concentrations would allow exploration of a dose adjustment strategy to determine if patient outcomes could be improved.

There are some limitations to this work. Our study was not designed to look for a relationship between TFV-DP exposure and clinical markers of antiretroviral efficacy (viral load and CD4 count), as all but six participants had HIV RNA of <400 copies/ml at the time of pharmacokinetic evaluation. The lack of knowledge of an intracellular TFV-DP concentration associated with virologic response remains a critical gap in the clinical pharmacologic profile of TFV. The estimate of the intracellular half-life of TFV-DP arises from samples collected predose to 24 h postdose, within the usual once-daily dosing interval. Because the sampling interval is less than the estimated half-life, the value we report, even though it is consistent with other data, may suffer some imprecision. Ideally, samples would be collected over a time period longer than the half-life, although this is not feasible in subjects such as those in these studies, who were doing well on stable antiretroviral therapy. Finally, we used an indirect model to overcome the lack of data on other TFV intracellular phosphate anabolites and kinase activity, as well as mathematical and biological difficulties of mass transfer. Other clinical investigators have also used simplified models to describe the intracellular pharmacology of TFV, such as that described for the study of TFV-DP in neonates (11). In both our approaches, a single rate constant was used to describe the conversion of TFV to TFV-DP. Important differences included that we modeled TFV-DP formation from TFV as an indirect response and estimated the elimination rate constant of TFV-DP, while an effect compartment was used to model TFV-DP in neonates and the elimination rate constant was fixed to adult values. These practical approaches come at some cost to a complete understanding of intracellular phosphorylation, intra- and interpatient sources of variability in the processes such as those that may arise from cell activation state and host genetic variability in membrane transporters, and relationships to systemic exposure.

In conclusion, we have developed a model that describes the plasma pharmacokinetics of TFV and linked these plasma concentrations to the formation of the intracellular, pharmacologically active moiety TFV-DP. The population PK model satisfactorily described TFV plasma behavior, including effects of known covariates such as estimated CrCL. Sex, weight, TBIL and concomitant PI medications did not affect TFV CL/F or V/F. The indirect formation model adequately describes the intracellular concentrations of TFV-DP at steady state. The modeled parameters of E_max, EC₅₀, and intracellular half-life of TFV-DP are plausible and consistent with in vitro and in vivo data. This linked plasma and intracellular model should facilitate more informed investigations of the clinical pharmacology of TFV through an ability to predict intracellular concentrations based on knowledge of plasma concentrations.