Population Pharmacokinetics of Tenofovir in Human Immunodeficiency Virus-Infected Patients Taking Highly Active Antiretroviral Therapy

The influence of renal function on tenofovir pharmacokineticswas investigated in 193 human immunodeficiency virus (HIV)-infectedpatients by the use of a population approach performed withthe nonlinear mixed effects modeling program NONMEM. Tenofovirpharmacokinetics was well described by a two-compartment openmodel in which the absorption and the distribution rate constantsare equal. Typical population estimates of apparent centraldistribution volume (V_c/F), peripheral distribution volume (V_p/F),intercompartmental clearance (Q/F), and plasma clearance (CL/F)were 534 liters, 1,530 liters, 144 liters/h and 90.9 liters/h,respectively. Apparent plasma clearance was related to bodyweight/serum creatinine ratio (BW/S_CR) and to the existenceof a tubular dysfunction. Concomitant treatment with lopinavir/ritonavirwas found to decrease tenofovir clearance. Individual Bayesianestimates of CL/F were used to calculate the tenofovir areaunder the concentration-time curve from time zero to 24 h (AUC_0-24).In patients without tubular dysfunction, AUC_0-24 values markedlydecreased from 6.7 to 1.4 mg · h/liter for BW/S_CR increasingfrom 0.44 to 1.73. The relevance of a dosage adjustment basedon BW/S_CR should be further evaluated.

INTRODUCTION

Top

Introduction

Tenofovir is a nucleotide analogue used as a combination therapywith other antiretrovirals for the treatment of human immunodeficiencyvirus type 1 (HIV-1) infection. To be active, tenofovir needsto be converted at the intracellular level to tenofovir diphosphate,which inhibits the viral reverse transcriptase (10). A pharmacokineticstudy performed after intravenous administrations of tenofovirin HIV-infected adults (5) has shown that 70 to 80% of the dosewas recovered unchanged in urine and that tubular secretionwas an important pathway for tenofovir clearance. A significantincrease in tenofovir exposure was also demonstrated in adultpatients with creatinine clearance (CL_CR) less than 50 ml/min,which required a decrease in tenofovir dosage regimen for thissubpopulation (10). For adult patients with CL_CR greater than50 ml/min, the same dosage regimen of 300 mg once a day (QD)of tenofovir disoproxil fumarate (TDF), an oral prodrug of tenofovir,is recommended. However, the influence of renal function ontenofovir exposure has still not been studied in patients withCL_CR greater than 50 ml/min.

Thus, in order to evaluate whether a dose adjustment based onrenal function could be considered for tenofovir therapy, evenin patients receiving the recommended 300-mg QD dose, the influenceof the renal function on tenofovir pharmacokinetics was investigatedusing a population approach performed retrospectively on routinetherapeutic drug monitoring data.

MATERIALS AND METHODS

Top

Materials and Methods

Patients and treatment. The population comprised adult patients receiving TDF, as 300-mgtablets equivalent to 245 mg of tenofovir, at its recommended300-mg QD dose for the treatment of HIV infection and monitoredaccording to the plasma concentrations of antiretroviral drugson a routine basis. For each patient, time elapsed between administrationand sampling times, gender, body weight (BW), and age were carefullyrecorded, as well as combined treatments, particularly antiretroviraldrugs. Corresponding serum creatinine (S_CR), phosphorus concentration,and viral load were obtained from routine monitoring data performedin Cochin-Saint-Vincent-de-Paul Hospital Biochemistry and VirologyDepartments respectively. Creatinine clearance was calculatedfor each sample according to the Cockroft-Gault formula (3)and the Jelliffe formula (8).

Analytical methods. The tenofovir assay was performed according to a previouslypublished method (9) with a quantification limit, interassayprecision, and bias of 0.005 mg/liter, 9.6%, and 11.4%, respectively.

Population pharmacokinetic modeling. Concentration-time data were analyzed using the first-orderconditional estimation (FOCE) method of the nonlinear mixedeffects modeling program NONMEM (2) (version V, level 1.1, doubleprecision). Several structural pharmacokinetic models were investigated.Classical one, two, and three-compartment models were firstevaluated. A two-compartment pharmacokinetic model in whichthe absorption (k_a) and distribution rate ( ${alpha}$ ) constants are equalwas also tried (14). The explicit solutions for this pharmacokineticmodel were coded in the $PRED section of the control stream,and its parameters were tenofovir apparent total and intercompartmentalclearance (CL/F and Q/F) and central and peripheral distributionvolume (V_c/F and V_p/F), respectively.

Several error models were also investigated (i.e., proportional,exponential, and additive random effects model) to describeinterpatient and residual variability.

The influence of continuous covariates on pharmacokinetic parameterswas systematically tested via a generalized modeling accordingto the following equation, using CL/F and BW for example:

where TV(CL/F) is the typical value of apparentclearance for a patient with the median covariate value and ${theta}$ _BW is the influential factor for body weight. Binary covariates(gender, combined treatment, and tubular dysfunction) were investigatedas follows:

The effect of a covariatewas assessed by the chi-square test of the difference betweenthe objective functions of the basic model (without covariate)and the model including the covariate. A covariate was retainedin the model if it produced a minimum decrease in the objectivefunction of 4 units (P < 0.05, 1 degree of freedom) and ifone of the following criteria was satisfied: (i) it led to areduction of the interindividual variability ( ${eta}$ ) of the associatedpharmacokinetic parameter or (ii) its effect was biologicallyplausible. An intermediate multivariate model was then obtained,including all selected covariates. A covariate was retainedin the final multivariate model if its deletion from the intermediatemodel led to a 7-point increase in the objective function (P< 0.01, 1 degree of freedom). At each step, the goodnessof fit was evaluated by the weighted residuals (WRES) versuspredicted concentrations (PRED) or time-after-dose graphs.

The accuracy and robustness of the final population model wereassessed using a bootstrap method, consisting of repeated randomsampling with replacement from the original data set. This resamplingwas repeated 1,000 times, and the values of the parameters estimatedfrom the bootstrap set were compared to the estimates obtainedfrom the original data set. The entire procedure was performedin an automated fashion using Wings for NONMEM (12).

Individual Bayesian estimates of the pharmacokinetic parameterswere used to calculate individual AUC_0-24 and minimal concentration(C_min).

Relationship between tenofovir exposure and phosphatemia/viral load. Possible relationship between AUC_0-24 and phosphatemia was evaluatedby linear regression. Patients for whom the viral load at samplingtime was known were also divided into four increasing AUC_0-24range groups. The limits of these groups were the 25th, 50th,and 75th percentiles of the calculated AUC_0-24 values in theentire population of the study. The percentage of patients witha viral load lower than the reference cutoff value (i.e., 400copies/ml) was calculated, and a chi-square test for trend wasused to investigate a possible effect of tenofovir AUC_0-24 onthis percentage.

RESULTS

Top

Results

Demographic data. One hundred ninety-three patients, ranging in age from 16.1to 70.9 years, were available for pharmacokinetic evaluation.All adult patients received TDF at the recommended 300-mg once-dailydose. Their characteristics are listed in Table 1. Viral loadat sampling time was available for 79 patients, and a viralload of <400 copies/ml was observed for 50 of these patients.Figure 1 represents the distributions of CL_CR (calculated accordingto the Cockroft-Gault formula) and BW/S_CR ratio in the populationstudy. Only four patients had a CL_CR slightly less than 50 ml/min(37, 42, 47, and 48 ml/min). They were nonetheless kept in thedatabase as they received the TDF 300-mg QD regimen recommendedfor patients with CL_CR greater than 50 ml/min. TDF was combinedwith at least one nucleoside reverse transcriptase inhibitor,one protease inhibitor (PI), and a nonnucleoside reverse transcriptaseinhibitor in 98, 77, and 22% of the samples, respectively. Lopinavir/ritonavirwas the most frequently PI combined with tenofovir as this concomitanttreatment involved 53% of the samples. Hypophosphatemia (i.e.,phosphorus <0.8 mmol/liter) was observed in 17% (n = 33)of the patients. A tubular dysfunction induced by tenofovirwas observed in four patients. The diagnosis was based on clinical(renal failure, bodyweight loss) as well as biological signs(normoglycemic glycosuria, proteinuria, phosphate wasting, andincrease in serum creatinine concentration). The CL_CRs of thesefour patients were 65, 59, 72.5, and 52.6 ml/min and were associatedwith phosphorus concentrations of 0.43, 0.46, 0.79, and 0.35mmol/liter, respectively.

View this table:
[in this window]
[in a new window]
TABLE 1. Characteristics of adult patients in this study^a

View larger version (16K):
[in this window]
[in a new window]
FIG. 1. CL_CR (A) and BW/S_CR (B) distributions in the population of the study.

Population pharmacokinetics. The best fit was obtained with the two-compartment model inwhich k_a and ${alpha}$ are equal. This model provided a further 73-pointdecrease in the objective function compared to the one-compartmentmodel. The classical two-compartment model (subroutines ADVAN4TRANS4) was first tried, but no value could be estimated fork_a. As no previously published value was available for k_a, severalattempts to fix this parameter to arbitrary values were made.The pharmacokinetic parameter estimates obtained were then characterizedby important standard errors or by typical values in disagreementwith published data (not shown). The three-compartment modelseemed to be overparameterized for our data as it systematicallyled to convergence failure. The graph of observed concentrationsas a function of time after dosing with the typical pharmacokineticcurve is displayed in Fig. 2. Interpatient variability was describedby an exponential error model, whereas residual variabilitywas described by an additive error model. Interindividual variabilityof V_p/F and Q/F could not be estimated, and no covariate wastested on these parameters. A significant covariance was foundbetween CL/F and V_c/F.

View larger version (26K):
[in this window]
[in a new window]
FIG. 2. Observed tenofovir concentrations (OBS) versus time after dose and typical pharmacokinetic curve.

Three different markers of the renal function, S_CR, CL_CR, andthe BW/S_CR ratio were found to improve the fit. The relationshipbetween CL/F and CL_CR was independent in the calculation formulaused for CL_CR (i.e., Cockroft-Gault or Jelliffe), but only thedeletion of BW/S_CR from the intermediate model significantlyincreased the objective function. Among the concomitant antiretroviralagents, a significant interaction was found only with lopinavir/ritonavir,which decreased tenofovir CL/F. Though only four patients wereinvolved, the presence of a tubular dysfunction explained 37%of CL/F interindividual variability. The effects of the differenttested covariates on the objective function and the interindividualvariability of the pharmacokinetic parameters are summarizedin Table 2.

View this table:
[in this window]
[in a new window]
TABLE 2. Effect of the tested covariates on the objective function^a

The final covariate submodel was then

withL = 0.86 if lopinavir/ritonavir were combined with tenofovirand T = 2.3 if a tubulopathy was observed.

Table 3 summarizes the population parameter estimates. The improvementof the fit provided by the covariates is visually evaluatedby the PRED versus observed (OBS) concentration plots obtainedwith the covariate-free and the final model (Fig. 3A and B).The goodness-of-fit was also evaluated graphically by the gooddistribution of the points on the weighted residuals (WRES)versus PRED and time plots (Fig. 4A and 4B). A correlation coefficientof 0.764 was obtained between observed and model-predicted concentrations.

View this table:
[in this window]
[in a new window]
TABLE 3. Population pharmacokinetic parameters of tenofovir in 193 adult patients and bootstrap validation^a

View larger version (22K):
[in this window]
[in a new window]
FIG. 3. Improvement of the fit from the covariate-free model (A) to the final model (B) visualized on the observed (OBS) versus model-predicted (PRED) concentrations (log-log scale).

View larger version (18K):
[in this window]
[in a new window]
FIG. 4. Goodness of fit evaluated by weighted residuals (WRES) versus time after dose (A) and WRES versus model predicted (PRED) tenofovir plasma concentrations (B).

Bootstrap validation. The final model obtained with the original data set was subjectedto a bootstrap analysis. As shown in Table 3, the mean parameterestimates obtained from the bootstrap process, 1,000 runs, werenot statistically different from the estimates previously obtainedwith the original data set.

Individual exposure to tenofovir. Individual Bayesian clearances were used to calculate individualAUC_0-24 and C_min values in adult patients. For the 189 patientswith no tubular dysfunction at sampling time, overall mean ±standard deviation (SD) AUC_0-24 was 3.0 ± 0.8 mg ·h/liter. Calculated AUC_0-24 and C_min values markedly decreasedfrom 6.7 to 1.4 mg · h/liter (Fig. 5) and from 0.198to 0.020 mg/liter, respectively, when BW/S_CR increased from0.44 to 1.73.

View larger version (24K):
[in this window]
[in a new window]
FIG. 5. Tenofovir AUC_0-24 achieved for a 300-mg TDF once-a-day dose in the 189 adult patients with no tubular dysfunction versus BW/S_CR.

Relationship between tenofovir pharmacokinetics and phosphatemia. No relationship was found between tenofovir AUC_0-24 and phosphatemia.The mean ± SD CL_CR and CL/F of patients with hypophosphatemia(n = 33) were 90 ± 27 ml/min and 89 ± 39 liters/h,respectively, whereas the mean ± SD CL_CR and CL/F were89 ± 23 ml/min and 88 ± 26 liters/h for the otherpatients (n = 160). No significant difference was found betweenthe CL_CR (P = 0.8, Mann-Whitney test) or the CL/F (P = 0.6,Mann-Whitney test) of these two groups.

Four patients with hypophosphatemia presented a tenofovir-inducedtubular dysfunction. Their phosphorus concentration, CL_CR, CL/F,and C_min are displayed in Table 4. Their CL_CR comprised between52.6 and 72.5 ml/min. Forty-nine patients without tubulopathyhad a CL_CR value comprised in this range, and their mean ±SD AUC_0-24 was 3.7 ± 0.8 mg · h/liter. The C_minvalues of these four patients were superior to 0.2 mg/liter.No patient with no tubular dysfunction had a C_min value higherthan 0.2 mg/liter.

View this table:
[in this window]
[in a new window]
TABLE 4. Biologic and pharmacokinetic parameters of the four patients suffering from a tubular dysfunction at sampling time

Relationship between tenofovir pharmacokinetics and viral load. The mean ± SD tenofovir AUC_0-24s were 3.0 ± 1.5mg · h/liter for patients with a viral load of >400copies/ml and 3.0 ± 1.1 mg · h/liter for patientswith a viral load of <400 copies/ml. No significant increasein the percentage of patients achieving a viral load of <400copies/ml (Table 5) was observed as a function of tenofovirAUC_0-24 (P = 0.35, chi-square test for trend).

View this table:
[in this window]
[in a new window]
TABLE 5. Percentage of patients with viral load of <400 copies/ml as a function of tenofovir AUC_0-24h

DISCUSSION

Top

Discussion

Tenofovir plasma pharmacokinetics were well described by a two-compartmentmodel. However, the lack of points in the absorption phase didnot allow us to estimate a value for k_a. As fixing this parameterwas not possible, an alternative model in which a common valuewas estimated for k_a and ${alpha}$ was preferred. Our mean CL/F estimate(90.9 liters/h) was nevertheless close to the previously reportedvalue in healthy volunteers (88.1 liters/h) (6). Furthermore,a pharmacokinetic study performed after intravenous administrationsof tenofovir in HIV patients (5) reported CL and V_ss valuesof approximately 18 liters/h and 800 liters, respectively. Astenofovir bioavailability is thought to be about 25 to 30% (1),these results are also in agreement with our own data (i.e.,90.9 liters/h for CL/F and 2,064 liters for V_ss). The mean calculatedAUC_0-24 obtained in the present study (3.0 mg · h/liter)was also similar to the previously described values (3.6 mg· h/liter) (6), 2.94 mg · h/liter (1). Our resultsalso confirmed the increase in tenofovir AUC_0-24 induced byconcomitant treatment with lopinavir/ritonavir that has beenreported previously (10). A possible explanation for this pharmacokineticinteraction is the inhibition of the Mrp-2-mediated transportof tenofovir outside renal tubules (13). Our data did not allowus to identify the similar interaction with atazanavir (10)as this association involved only two patients. Didanosine wascombined with tenofovir for 60 patients. If this associationis known to increase didanosine exposure (10), an effect ofdidanosine on tenofovir pharmacokinetics was not found in thepresent study.

As expected, the four patients with tubular dysfunction hadmarkedly reduced concomitant CL/F compared to patients withequivalent CL_CR values but without tubulopathy. This discrepancybetween CL_CR and tenofovir CL/F can be explained by the importanceof tubular secretion in the renal elimination of tenofovir.For these patients, tubular dysfunction was imputed to tenofovirtherapy as the renal function normalized after the tenofovirtreatment was stopped.

Tenofovir is indeed known to induce tubular dysfunction thatcan lead to renal impairment in a limited number of patients(13). Hypophosphatemia is a biological sign that is thoughtto occur in 100% of tenofovir-induced tubular dysfunction (7).No relationship was found between phosphatemia and tenofovirCL/F in the present study. However, this result can be explainedby the fact that hypophosphatemia is a parameter of poor specificity(4) and does not mean that this relationship does not existin patients with tenofovir-induced tubular dysfunction. However,one interesting point that remains to be assessed is the possiblerelationship between a high tenofovir exposure at the beginningof tenofovir therapy and the incidence of tenofovir-inducedtubulopathy.

A strong relationship was found between tenofovir CL/F and BW/S_CR.Serum creatinine alone did not provide such an improvement ofthe fit, probably because it is not an accurate marker of therenal function as it does not take into account creatinine productionfrom muscle mass, a parameter that can be reflected by bodyweight. The effect of CL_CR was also less pronounced, maybe becausethe Cockroft-Gault and Jelliffe formulas also take gender andage into account, two covariates that showed no significanteffect on CL/F. Thus, and as it was previously suggested (11),BW/S_CR could appear as a surrogate marker of the renal function,which explains its significant effect on tenofovir CL/F.

Tenofovir exposure decreased when BW/S_CR increased, the highestBW/S_CR values corresponding to an AUC_0-24 markedly lower thanthe mean AUC_0-24 reported for the recommended 300-mg TDF dose(i.e., 2.94 mg · h/liter) (1). This decrease in tenofovirAUC_0-24 could have consequences upon treatment efficacy. Indeed,the phase II study cited above (1) demonstrated that mean viralload decrease and AUC_0-24 were –1.2 log₁₀ copies/ml and2.94 mg · h/liter, respectively, for the 300-mg dose,whereas the mean viral load decrease and AUC_0-24 were –0.44log₁₀ copies/ml and 1.6 mg · h/liter, respectively, fora 150-mg dose. Furthermore, another phase II study showed theexistence of a correlation between tenofovir AUC_0- at the firstday of treatment and the antiviral response at the 14th day(5). Thus, though tenofovir is an inactive prodrug, these twostudies strongly suggested the existence of a relationship betweentenofovir exposure and the treatment efficacy.

Our results did not allow us to identify such a relationship.However, this does not mean that such a relationship does notexist as our study was not designed for this virological endpoint.A rigorous methodology would have necessitated virological aswell as pharmacological selection criteria. Because of the lackof such criteria, many biases (i.e., unknown viral load for114 patients, viral load at the beginning of tenofovir therapy,tenofovir therapy duration at sampling time, combined antiretroviraldrugs, treatment history, compliance, etc.) can explain thelack of correlation between viral load and tenofovir AUC_0-24that we observed. In consequence, it seems important to investigatethe possible influence of the renal function of patients receivingTDF at 300 mg QD on treatment efficacy in a prospective trial.

In conclusion, this study showed that tenofovir plasma clearancewas related to the body weight/serum creatinine ratio and thathigh ratio values corresponded to an important decrease in tenofovirAUC_0-24. The virological consequence of this decrease in tenofovirexposure should be prospectively investigated, and the usefulnessof a dosage adjustment based on body weight/serum creatinineratio should be evaluated in further studies.

ACKNOWLEDGMENTS

This work was supported by l'Institut National de la Santéet de la Recherche Médicale (Contrat de recherche stratégique2002).

FOOTNOTES

* Corresponding author. Mailing address: Service de Pharmacologie Clinique, Hôpital Cochin-Saint-Vincent-de-Paul, 74-82 Avenue Denfert-Rochereau, 75674 Paris Cedex 14, France. Phone: 33 1 40488209. Fax: 33 1 40488223. E-mail: Vincent.jullien{at}svp.aphp.fr.

REFERENCES

Top