Using Pharmacokinetic and Viral Kinetic Modeling To Estimate the Antiviral Effectiveness of Telaprevir, Boceprevir, and Pegylated Interferon during Triple Therapy in Treatment-Experienced Hepatitis C Virus-Infected Cirrhotic Patients

Abstract

Triple therapy combining a protease inhibitor (PI) (telaprevir or boceprevir), pegylated interferon (PEG-IFN), and ribavirin (RBV) has dramatically increased the chance of eradicating hepatitis C virus (HCV). However, the efficacy of this treatment remains suboptimal in cirrhotic treatment-experienced patients. Here, we aimed to better understand the origin of this impaired response by estimating the antiviral effectiveness of each drug. Fifteen HCV genotype 1-infected patients with compensated cirrhosis, who were nonresponders to prior PEG-IFN/RBV therapy, were enrolled in a nonrandomized study. HCV RNA and concentrations of PIs, PEG-IFN, and RBV were frequently assessed in the first 12 weeks of treatment and were analyzed using a pharmacokinetic/viral kinetic model. The two PIs achieved similar levels of molar concentrations (P = 0.5), but there was a significant difference in the 50% effective concentrations (EC₅₀) (P = 0.008), leading to greater effectiveness for telaprevir than for boceprevir in blocking viral production (99.8% versus 99.0%, respectively, P = 0.002). In all patients, the antiviral effectiveness of PEG-IFN was modest (43.4%), and there was no significant contribution of RBV exposure to the total antiviral effectiveness. The second phase of viral decline, which is attributed to the loss rate of infected cells, was slow (0.19 day⁻¹) and was higher in patients who subsequently eradicated HCV (P = 0.03). The two PIs achieved high levels of antiviral effectiveness. However, the suboptimal antiviral effectiveness of PEG-IFN/RBV and the low loss of infected cells suggest that a longer treatment duration might be needed in cirrhotic treatment-experienced patients and that a future IFN-free regimen may be particularly beneficial in these patients.

INTRODUCTION

Chronic infection with hepatitis C virus (HCV) affects approximately 160 million people worldwide (1) and is the leading cause of cirrhosis, liver cancer, and liver transplantation (2). The goal of treatment is to achieve a sustained virological response (SVR), a marker of viral eradication, assessed by the absence of detectable HCV RNA 6 months after treatment discontinuation. The approval in 2011 of two protease inhibitors (PIs), telaprevir and boceprevir, in combination with pegylated interferon (PEG-IFN) and ribavirin (PEG-IFN/RBV) (3), marked an important milestone with SVR rates higher than 70% in HCV genotype 1-infected patients (4, 5). Recently, two new triple therapies involving sofosbuvir, a nucleoside polymerase inhibitor, and simeprevir, a new protease inhibitor, showing even higher SVR rates of 90% in clinical trials, were approved by the European and American regulatory agencies (6). However, the cost of these new treatments, about twice as much as that of telaprevir- or boceprevir-based therapy (7), will make them out of reach for patients in many countries. Therefore, triple therapy with PEG-IFN, RBV, and telaprevir/boceprevir will continue to be widely used in the next few years and will remain the only therapeutic option for many patients.

Although these results suggest that a functional cure might be obtained in a large majority of patients, one should keep in mind that issues remain. In particular, the proportion of patients with advanced liver disease and cirrhosis and/or in whom previous treatment with PEG-IFN/RBV had failed is underrepresented in the patient population in clinical trials (8–11). The evaluation of the triple therapy in this population was precisely the goal of the National Agency for Research on AIDS and Viral Hepatitis (ANRS) CO20-Compassionate Use of Protease Inhibitors in Viral C Cirrhosis (CUPIC) cohort (ClinicalTrials.gov registration number NCT01514890) (12), where 511 HCV genotype 1-infected, treatment-experienced cirrhotic patients were included. In this study, the SVR rates 12 weeks after treatment discontinuation (SVR12) were 52% and 43% in telaprevir- and boceprevir-treated patients, respectively (13). The origin of this impaired response might encompass a variety of factors, in particular, impaired drug pharmacokinetics (PK) or limited sensitivity to PI agents and/or PEG-IFN/RBV in this particular population.

One way to evaluate treatment antiviral effectiveness and to optimize therapy is to use PK-viral kinetic (VK) models that provide a useful tool to quantitatively describe the relationship between drug exposure and viral response (reviewed in reference 14). However, no such analysis for boceprevir has been published, and results published for telaprevir were mostly based on treatment-naive and/or noncirrhotic patients (15–17).

Here, we aimed to get new insights into the determinants of the response to triple therapy by analyzing in detail, within a subset of 15 patients enrolled in the ANRS CO20-CUPIC study, the relationship between drug concentrations and the early virological response. We used the techniques of PK-VK modeling in order to tease out the relative antiviral effectiveness of each of the agents involved in the triple therapy (i.e., boceprevir or telaprevir, PEG-IFN, and RBV) and to investigate a possible association with the long-term virological response.

MATERIALS AND METHODS

Patients and data.

MODCUPIC is a substudy of the French multicenter prospective ANRS CO20-CUPIC cohort. In four centers from September 2011 to September 2012, patients who were chronically monoinfected with HCV genotype 1, who had compensated cirrhosis (Child-Pugh class A), who were nonresponders to a prior IFN-based therapy, and who had started triple therapy were recruited. The diagnosis of cirrhosis was made by liver biopsy specimens or noninvasive tests (Fibrotest, Fibroscan, Fibrometer, or Hepascore) at the discretion of the investigator, according to the French recommendations (18). The choice between telaprevir- or boceprevir-based therapies was at the investigator's discretion without randomization. Telaprevir-based therapy included 12 weeks of telaprevir (750 mg/8 h) in combination with PEG-IFN-α-2a (180 μg/week) and RBV (1,000 or 1,200 mg/day, depending on body weight) and then 36 weeks of PEG-IFN-α-2a/RBV (named the telaprevir group in the following). Boceprevir-based therapy included 4 weeks (lead-in phase) of PEG-IFN-α-2b (1.5 μg/kg body weight/week) or PEG-IFN-α-2a (180 μg/week) and RBV (800 or 1,400 mg/day, depending on body weight) and then 44 weeks of PEG-IFN-α-2b/RBV and boceprevir (800 mg/8 h) (named the boceprevir group in the following). Patients were followed for up to 6 months after treatment discontinuation to assess the SVR.

Written informed consent was obtained before enrollment. The protocol was conducted in accordance with the Declaration of Helsinki and was approved by the Ile-de-France IX Ethics Committee (Créteil, France).

Bioanalytical methods.

HCV RNA and drug concentrations were measured after PI initiation at hours 0 and 8, days 1, 2, and 3, and weeks 1, 2, 3, 4, 8, and 12. Patients treated with boceprevir had two additional viral load (VL) and concentration measurements during the lead-in phase. Blood samples were collected early in the morning before the first daily dose of PIs and RBV, and therefore only trough predose drug concentrations were measured. All samples were collected in SST (serum) vacutainers, kept at 4°C until centrifuged at 3,000 rpm for 10 min in a 4°C centrifuge within 1 h after collection, aliquoted, and kept at −80°C until analysis.

PI concentrations in serum were determined using ultraperformance liquid chromatography coupled with tandem mass spectrometry with lower limits of quantification (LOQ) of 5 ng/ml and 10 ng/ml for boceprevir and telaprevir, respectively (19). The PI concentrations were converted to μmol/liter for analysis using molar masses of 519.68 g/mol and 679.85 g/mol for boceprevir and telaprevir, respectively. The RBV concentrations in serum were determined using ultraperformance liquid chromatography coupled with UV detection with an LOQ of 100 ng/ml (20). PEG-IFN-α-2a and -α-2b in serum were determined with a bioassay which was chosen because the objective was to quantify not only the concentration but also the antiviral activity of PEG-IFN-α. The immunoassay measures the physical quantity of material but does not differentiate between active and inactive molecules, while the bioassay for IFN-α is based on the protection of cultured cells against the cytopathic effect of a challenge virus and also was suitable for assaying both PEG-IFN-α-2a and PEG-IFN-α-2b. The reference solutions contained 2.8 to 180 ng/ml of PEG-IFN-α-2a (Roche Diagnostics, Germany) (21).

HCV RNA levels were measured with a real-time PCR-based assay (Cobas AmpliPrep/Cobas TaqMan; Roche Diagnostics), with a lower limit of detection (LOD) of 15 IU/ml. DNA samples were genotyped for the interleukin 28B (IL28B) rs12979860 polymorphism (AmpliTaq gold DNA polymerase and BigDye terminator cycle sequencing kit; Applied Biosystems, United Kingdom).

Drug pharmacokinetic modeling.

All drug concentrations were fitted separately in telaprevir and boceprevir treatment groups. For both PEG-IFN and RBV, the trough serum concentrations, denoted C^PEG-IFN(t) and C^RBV(t), respectively, were fitted using an exponential model to reflect the progressive increase in trough drug concentrations over time:

$$C^{\text{PEG}\mathrm{-}\mathrm{I}\mathrm{F}\mathrm{N}}(t)=C_{\mathrm{s}\mathrm{s}}^{\text{PEG}\mathrm{-}\mathrm{I}\mathrm{F}\mathrm{N}}\times (1-e^{-kt})$$ (1)

$$C^{\mathrm{R}\mathrm{B}\mathrm{V}}(t)=C_{\text{ss}}^{\mathrm{R}\mathrm{B}\mathrm{V}}\times (1-e^{-kt})$$ (2)

where C_ss is the trough concentration at steady state and k is the rate constant of elimination which reflects the progressive increase in C(t) over time.

For both PI drugs, no significant increases in trough concentrations over time were observed, consistent with the fact that they have short elimination half-lives (22). Therefore, concentrations for both telaprevir and boceprevir were fitted using a constant model, where C_ss is the trough concentration:

$$C^{\text{PI}}(t)=C_{\text{ss}}^{\text{PI}}$$ (3)

Viral kinetic modeling.

The following model of HCV viral kinetics (VK) was used to fit the changes in HCV RNA (23):

$$\frac{dI}{dt}=bVT-\delta I$$

$$\frac{dV}{dt}=p(1-\epsilon (t))I-cV$$ (4)

where T represents the target cells that can be infected by virus, V, with rate b. Infected cells, I, are lost with rate δ and produce p virions per day, which are cleared from serum with rate c. The target cell level is assumed to be constant throughout the study period (12 weeks) and remains at its pretreatment value, T₀ = cδ/pβ. Treatment is assumed to reduce the average rate of viral production per cell from p to p(1 − ε), where ε represents the drug antiviral effectiveness, i.e., ε = 0.99, implying that the drug is 99% effective in blocking viral production. This model predicts that VL will fall in a biphasic manner, with a rapid first phase that reduces the VL with a magnitude equal to log₁₀(1 − ε) and lasts for a few days, followed by a second slower but persistent phase of viral decline with a rate of εδ. Therefore, a difference between ε = 99.9% and ε = 99.0% corresponds to a 10-fold difference in the viral production under treatment and will lead to a 1-log difference between the two curves of viral decline (24). We fixed p and b to 100 IU/ml/cell/day and 10⁻⁷ (IU/ml)⁻¹/day, respectively, without loss of generality (25).

The effectiveness of each drug in blocking viral production was described by a maximum effect (E_max) model, assuming a maximum inhibition of 100%:

$${\epsilon }^{\text{PI}}(t)=\frac{C^{\text{PI}}(t)}{C^{\text{PI}}(t)+{\text{EC}}_{50}^{\text{PI}}}$$

$${\epsilon }^{\text{PEG}\mathrm{-}\mathrm{I}\mathrm{F}\mathrm{N}}(t)=\frac{C^{\text{PEG}\mathrm{-}\mathrm{I}\mathrm{F}\mathrm{N}}(t)}{C^{\text{PEG}\mathrm{-}\mathrm{I}\mathrm{F}\mathrm{N}}(t)+{\mathrm{E}\mathrm{C}}_{50}^{\text{PEG}\mathrm{-}\mathrm{I}\mathrm{F}\mathrm{N}}}$$ (5)

where EC₅₀^PI (respectively, EC₅₀^PEG-IFN) is the PI (respectively, PEG-IFN) concentration at which the PI (respectively, PEG-IFN) is 50% effective, and C^PI(t) [respectively, C^PEG-IFN(t)] is the individual prediction (see below) given by the PK models (equations 1 and 3).

The combined effect of PIs and PEG-IFN was modeled using a Bliss independent action model (26), and the total efficacy ε(t) is given by

$$(1-\epsilon (t))=(1-{\epsilon }^{\mathrm{P}\mathrm{I}}(t))(1-{\epsilon }^{\text{PEG}\mathrm{-}\mathrm{I}\mathrm{F}\mathrm{N}}(t))$$ (6)

Since the effect of RBV on the early virological response is expected to be modest (27–29), we did not incorporate the effect of RBV into the reference model (equations 4 to 6). In a second step, we tested whether the effectiveness of RBV, also modeled using an E_max model, could enhance the effect of blocking viral production or reduce viral infectivity, as suggested previously (30).

Data analysis and parameter estimation.

The pharmacokinetic/viral kinetic (PK-VK) model given by equations 4 to 6 can be used only to characterize the viral kinetics of a drug-sensitive virus and therefore cannot fit viral rebounds due to the emergence of drug-resistant virus. Therefore, only HCV RNA data until virologic rebounds (with no indication of a lack of compliance) were used to estimate the viral kinetic parameters.

The parameters baseline viral load (V₀), clearance rate of virus from serum (c), clearance rate of virus from serum (δ), EC₅₀^PI, and EC₅₀^PEG-IFN were estimated using nonlinear mixed-effect models (NLMEM). In this approach, each individual parameter θ_i is composed of a fixed part θ, which represents the mean value of the parameter in the population (fixed effects), and a random part η_i chosen from a Gaussian distribution with mean 0 and standard deviation ω_i that accounts for the interindividual variability. Therefore, for all parameters θ = θe^ηi where η_i ∼N(0, ω). Both PK data and log₁₀(HCV RNA) were best described using an additive residual error with constant variance.

The model parameters were estimated using the stochastic approximation expectation minimization (SAEM) algorithm in MONOLIX, v4.2 (available at http://www.lixoft.eu). Of note, this approach is based on maximum likelihood estimation, which takes into account the information brought by data under the LOD as left-censored data (31, 32).

The model selection was done using the Bayesian information criterion (BIC), a fitting criterion derived for each model from the computation of likelihood that takes into account the number of estimated parameters used (the lower the better) (33). The model evaluation was performed using goodness-of-fit plots, as well as the individual weighted residuals (IWRES) and the normalized prediction distribution errors (NPDE) over time.

Differences in PK-VK model parameters between telaprevir and boceprevir treatment groups.

A Wald test on the PK-VK model parameters (c, δ, and EC₅₀^PI) was used to assess the differences in population parameters between the two groups. Because we previously showed that this approach could lead to an inflation of the type I error in the case of a small sample size (n = <20/group) (34), a permutation test was performed to confirm the statistical significance when the Wald test was significant at the level of 5%. In brief, 1,000 data sets were simulated by randomly allocating patients to the telaprevir or boceprevir group, maintaining a similar proportion of patients allocated to each group as in the original data set. Then the P value of the Wald test was calculated for each simulated data set. Finally, the corrected P value of the permutation test is equal to the proportion of simulated data sets having a P value lower than the one found in the original data set.

Because the genetic barrier to resistance of PI (i.e., the number of changes in amino acids needed to generate mutants with a high level of resistance) depends on the HCV subgenotype and therefore leads to different SVR rates, we also estimated the effect of HCV subgenotype (1a versus non-1a) on viral kinetic parameters. The IL28B polymorphism, which is also associated with the response to IFN-based therapy, was not investigated because a previous bitherapy had failed in all these patients.

Prediction and comparison of individual parameters.

Individual empirical Bayesian estimates (EBE) parameters for both PK and VK were obtained by computing for each patient the maximum a posteriori (MAP) estimate. The individual antiviral effectiveness at steady state, ε_ss, of each agent was defined by

$${\epsilon }_{\text{ss}}^{\text{PI}}=\frac{C_{\text{ss}}^{\text{PI}}}{C_{\text{ss}}^{\text{PI}}+{\text{EC}}_{50}^{\text{PI}}}$$

$${\epsilon }_{\mathrm{s}\mathrm{s}}^{\text{PEG}\mathrm{-}\mathrm{I}\mathrm{F}\mathrm{N}}=\frac{C_{\text{ss}}^{\text{PEG}\mathrm{-}\mathrm{I}\mathrm{F}\mathrm{N}}}{C_{\mathrm{s}\mathrm{s}}^{\text{PEG}\mathrm{-}\mathrm{I}\mathrm{F}\mathrm{N}}+{\mathrm{E}\mathrm{C}}_{50}^{\text{PEG}\mathrm{-}\mathrm{I}\mathrm{F}\mathrm{N}}}$$ (7)

Nonparametric two-sided tests (Wilcoxon test) were used to compare (i) individual EBE PK parameters between patients who received telaprevir versus boceprevir and between patients who received PEG-IFN-α-2a versus -α-2b and (ii) individual EBE PK parameters between patients who achieved a SVR and those who did not. Because all patients were nonresponders to PEG-IFN, the effect of the IL28B genotype on PK and VK parameters was not tested.

RESULTS

Fifteen HCV genotype 1-infected patients were included, 9 receiving telaprevir and 6 receiving boceprevir. Twelve (80%) were men, with a median (minimum [min] and maximum [max]) age of 55 years (44 and 64 years). Seven (47%) patients were infected with subgenotype 1a, 2 (22%) in the telaprevir group and 5 (83%) in the boceprevir group. Prior treatment responses were as follows: partial response, 2 patients; null response, 5 patients; relapse, 6 patients; and early discontinuation for adverse events, 2 patients. Only 2 patients had the most favorable IL28B CC genotype (35). The main characteristics of the patients are presented in Table 1.

TABLE 1.

Main patient characteristics

Characteristic	Results for patients receiving:		Results for all patients (n = 15)
	PEG-IFN/RBV + telaprevir (n = 9)	PEG-IFN/RBV + boceprevir (n = 6)
Age (median [min, max]) (yr)	55 (49, 59)	53 (44, 64)	55 (44, 64)
Males (no. [%])	8 (89)	4 (67)	12 (80)
HCV RNA (log10 IU/ml) (median [min, max])	6.5 (6.0, 6.8)	5.4 (4.9, 6.60)	6.2 (4.9, 6.8)
HCV genotype (no. [%])
1a	2 (22)	5 (83)	7 (47)
Non-1a	7 (78)	1 (17)	8 (53)
IL28B genotype (rs12979860) (no. [%])
C/C	2 (22)		2 (13)
C/T	6 (67)	6 (100)	12 (80)
T/T	1 (11)		1 (7)
Response to previous bitherapy (no. [%])
Partial responder		2 (33)	2 (13)
Null responder	4 (44)	1 (17)	5 (33)
Relapser	3 (33)	3 (50)	6 (40)
Early discontinuation for adverse event	2 (22)		2 (13)

Two patients had a viral breakthrough (at weeks 3 and 8). Eleven patients received PEG-IFN-α-2a (8 in the telaprevir group and 3 in the boceprevir group), 3 patients received PEG-IFN-α-2b (all in the boceprevir group), and 1 patient in the telaprevir group did not receive any injection of PEG-IFN (and this patient had a viral breakthrough at week 3).

Figure 1 shows the observed drug concentrations versus time, and Table 2 gives the estimated steady-state trough concentrations (C_ss) for all drugs. There was no significant difference in the molar median steady-state concentrations of telaprevir and boceprevir (C_ss^telaprevir = 3.77 μmol/liter [min and max, 2.68 and 5.98 μmol/liter], i.e., 2,563.0 ng/ml [min and max, 1,822.0 and 4,065.5 ng/ml] and C_ss^boceprevir = 3.92 [min and max, 3.22 and 7.64 μmol/liter], i.e., 2037.1 ng/ml [min and max, 1,673.4 and 3,970.4 ng/ml], P = 0.5). There was no significant difference in the median steady-state concentrations of PEG-IFN-α-2a and -α-2b (C_ss^{PEG-IFN-α-2a} = 89.6 ng/ml [min and max, 52.8 and 110.4 ng/ml] and C_ss^{PEG-IFN-α-2b} = 55.4 ng/ml [min and max, 55.3 and 57.9 ng/ml], P = 0.2). The concentrations of RBV increased over time in all patients and were well captured by our model (equation 2) with median k = 0.10 day⁻¹, corresponding to a half-life of increase of about 7 days. At equilibrium, median C_ss^RBV was 2,860 ng/ml (min and max, 2,428 and 3,874 ng/ml).

FIG 1.

Observed concentrations over time. (a) Telaprevir in 9 patients (black line, μmol/ml) and boceprevir in 6 patients (gray line, μmol/ml); (b) PEG-IFN in the telaprevir group (black line, ng/ml) and in the boceprevir group (gray line, ng/ml); (c) RBV in the telaprevir group (black line, ng/ml) and in the boceprevir group (gray line, ng/ml). Patients who received boceprevir-based therapy had only two blood samples during the lead-in phase at baseline and week 2.

TABLE 2.

Individual predicted trough concentrations at steady state

Parametera	No. of samples	Results (median [min, max])b
Csstelaprevir (μmol/liter)	9	3.77 (2.68, 5.98)
Cssboceprevir (μmol/liter)	6	3.92 (3.22, 7.64)
CssPEG-IFN-α-2a (ng/ml)	11	89.6 (52.8, 110.4)
CssPEG-IFN-α-2b (ng/ml)	3	55.4 (55.3, 57.9)
CssRBV (ng/ml)	15	2,860 (2,428, 3,874)

^aC_ss, trough concentration at steady state.

^bmin, minimum; max, maximum.

After the PK parameters were estimated, the predicted individual PK time courses were plugged into the PK-VK model (see Materials and Methods). The baseline VL was higher in the telaprevir group than in the boceprevir group; thus, a treatment group effect was added on the baseline VL (V₀^telaprevir = 6.43 log₁₀ IU/ml, versus V₀^boceprevir = 5.52 log₁₀ IU/ml, P = 0.0001). A greater proportion of patients who received boceprevir were HCV genotype 1a-infected relative to those who received telaprevir (P = 0.04). The subgenotype is an important predictor of the response to treatment, in particular with telaprevir, which has a lower genetic barrier to resistance with genotype 1a than 1b (only one nucleotide change in the genotype 1a viral genome is required to generate mutations V36M and R155K/T versus two nucleotide changes in genotype 1b) (36). This may explain why genotype 1a-infected patients were preferentially treated with boceprevir. We did not find any significant effect of subgenotype on any of the parameters.

The model described well the kinetics of HCV decline observed both during the lead-in phase (in the boceprevir group) and after the initiation of the PIs (in both groups) (Fig. 2). There was no evidence of model misspecification as showed by the goodness-of-fit plot (Fig. 3), and all parameters could be estimated with good precision (Table 3).

FIG 2.

Individual fits of the viral decline (log₁₀, IU/ml). Nine patients in telaprevir group (black curve) and 6 patients in boceprevir group (gray curve). Black crosses represent the observed viral load, and gray stars represent the viral load under the limit of detection.

FIG 3.

Goodness of fit of the viral kinetic-pharmacokinetic model. Residuals (weighted residuals calculated using individual predictions [IWRES] and normalized prediction distribution errors [NPDE]) versus time and versus prediction plots. Residuals seem to distribute homogeneously around 0. The observed viral loads are plotted as black crosses and viral load under the limit of detection as gray stars.

TABLE 3.

Parameter estimates and relative standard errors

Parametera	Estimate	RSEb (%)
V0telaprevir (log10 IU/ml)	6.43	2
V0boceprevir (log10 IU/ml)	5.52	3
c (day−1)	3.98	12
δ (day−1)	0.18	11
EC50PEG-IFN (ng/ml)	106	40
EC50telaprevir (μmol/liter)	0.009	30
EC50boceprevir (μmol/liter)	0.04	43
ωV0	0.07	20
ωc	0.47	19
ωδ	0.42	16
ωEC50PEG-IFN	0.67	30
ωEC50PI	0.61	32
σ	0.27	7

^aV₀, baseline viral load; c, clearance rate of virus from serum; δ, loss rate of infected cells; EC₅₀, 50% effective concentration; ω, interindividual variability; PI, protease inhibitor; σ, standard deviation of residual error.

^bRSE, relative standard error of parameter estimate.

The model predicted a mean EC₅₀^PEG-IFN of 106 ng/ml, leading to low antiviral effectiveness of PEG-IFN at a steady state of 43.4% (min and max, 0.0 and 52.7%), consistent with the modest 0.67 log₁₀ IU/ml drop observed during the 4-week lead-in phase in patients treated with boceprevir (Fig. 2).

After PI initiation, VL declined in a biphasic manner in all patients, where a rapid first phase was followed by a second slower phase. The rapid first phase was attributed to a clearance rate of virus, c, of 3.98 day⁻¹ and to high levels of antiviral effectiveness for both PIs. The intrinsic potency of the two molecules, as measured by the EC₅₀^PI, was significantly higher for telaprevir than for boceprevir (EC₅₀^telaprevir = 0.009 μmol/liter, versus EC₅₀^boceprevir = 0.04 μmol/liter, P = 0.008). Importantly, the statistical significance of this difference was obtained after taking into account the small sample size (see Materials and Methods) and adjustment on the baseline VL. Since telaprevir had a lower EC₅₀ than boceprevir and the two drugs achieved similar molar concentration levels, the model predicted that the median individual antiviral effectiveness of the PI agent in blocking viral production was significantly higher in patients who received telaprevir than in those who received boceprevir (ε_ss^telaprevir = 99.8% [min and max, 99.3 and 99.9%] and ε_ss^boceprevir = 99.0% [min and max, 98.0 and 99.6%], P = 0.002). Interestingly, this model captured well the relationship between the serum exposure and its antiviral effectiveness, demonstrating that the variability in drug exposure needs to be taken into account to understand the between-subject variability in PI antiviral effectiveness (Fig. 4a). Lastly, because the effectiveness of either PI was much greater than that of PEG-IFN (Fig. 4b), the total antiviral effectiveness obtained by the combination of either PI and PEG-IFN was largely similar to that obtained with the each PI only.

FIG 4.

Relationship between predicted trough concentration at steady state (C_ss) and predicted antiviral effectivenesses (ε_ss). (a) Protease inhibitor (telaprevir in black and boceprevir in gray, μmol/liter); (b) PEG-IFN (PEG-IFN-α-2a in black and PEG-IFN-α-2b in gray, ng/ml). The solid lines denote the predictions with the mean antiviral effectiveness, and the dotted lines denote the 95% confidence intervals computed with the standard errors predicted by the Fisher information matrix.

After the VL was rapidly reduced as a result of the strong antiviral effectiveness of the two PIs, the model predicted that a second slower phase of viral decline ensued, driven by the loss rate of infected cells, δ. We estimated δ to be 0.18 day⁻¹, corresponding to a half-life of infected cells of 3.9 days, with no significant differences between patients receiving telaprevir or boceprevir (P = 0.5).

Next we investigated the relationship between the PK-VK parameters and SVR. Among the 7 patients (47%) who achieved a SVR, 5 received telaprevir and 2 received boceprevir (56% versus 33%, respectively, P = 0.6). As shown in Fig. 5, neither the antiviral effectiveness of the PI nor that of PEG-IFN was significantly associated with the long-term virological response. However, the loss rate of infected cells, δ, was significantly higher in patients who subsequently achieved a SVR (median δ^SVR = 0.27 day⁻¹, versus median δ^non-SVR = 0.14 day⁻¹, P = 0.03).

FIG 5.

Relationship between the long-term virological response (SVR) and parameters estimated by the viral kinetic-pharmacokinetic model. (a) Predicted antiviral effectiveness (ε_ss) of the PIs; (b) predicted antiviral effectiveness (ε_ss) of PEG-IFN; (c) δ parameter (loss rate of infected cells). P value from Wilcoxon tests.

Lastly, we verified that incorporating the effect of RBV exposure in the PK-VK model, either on the block of viral production or in the decrease of viral infectivity (data not shown), did not improve the fit of the data. Furthermore, there was no significant association between the predicted C_ss^RBV and the long-term virological response (P = 0.5).

DISCUSSION

Here we used a PK-VK model to provide the first detailed picture of the relationship between the exposure to all drugs involved in triple therapy (PEG-IFN, RBV, and telaprevir or boceprevir) and the early virological response. This novel model provides important insights into the understanding of the response to triple therapy in hard-to-treat patients.

We predicted that both PIs achieved high levels of antiviral effectiveness in blocking viral production (>97.9%) in all patients. However, telaprevir had a higher intrinsic potency than boceprevir, as measured by the EC₅₀ (P = 0.008 after correction for the small sample size), leading to a significantly higher level of antiviral effectiveness than that of boceprevir (ε_ss^telaprevir = 99.8%, versus ε_ss^boceprevir = 99.0%, P = 0.002), i.e., a 5-fold difference in the viral production under treatment. Importantly, the difference in the EC₅₀ was obtained despite the facts that the study was not randomized and that patients who received telaprevir had less favorable baseline characteristics than those who received boceprevir with higher baseline VL (6.43 log₁₀ IU/ml versus 5.52 log₁₀ IU/ml, respectively, P < 10⁻⁴) and a higher proportion of null responders to previous bitherapy (4/9 versus 1/6).

The comparison of the antiviral effectiveness of the drugs should be made with caution because of the small sample size, the absence of randomization, and the fact that only trough concentrations were used to estimate the EC₅₀ of the PIs, which may lead to underestimation. However, these results demonstrate for the first time a significant association between serum exposure to PI agents and the antiviral effectiveness achieved. To confirm the significance of this association, we fitted HCV RNA data to a simplified model where drug exposure was not taken into account (37). Compared to this model, we found that the PK-VK model both improved the fitting criterion (BIC decreases from 181.3 to 176.3, i.e., an improvement of 5 points, which is regarded as positive evidence) and reduced the between-patient parameter variability by 26% (ω_EC50^PI from 0.85 to 0.61), thus demonstrating that serum PK is an important predictor of the antiviral effectiveness of triple therapy.

Our estimate that telaprevir achieves an antiviral effectiveness of 99.8% is largely similar to that found in naive patients (15), suggesting that compensated cirrhosis does not affect the maximal antiviral effectiveness of telaprevir. Whether this is also true for boceprevir is not known, as to our knowledge there is no published viral kinetic modeling study evaluating the in vivo antiviral effectiveness of boceprevir.

In contrast to the high effectiveness achieved by both PIs, PEG-IFN was found to have a modest contribution in blocking viral production, with a mean value of 43.4%. Of note, including the patient who did not receive PEG-IFN in our analysis allowed us to add information on telaprevir antiviral effectiveness. Further RBV exposure had no significant contribution on the early viral kinetics. Together these results indicate that PEG-IFN and RBV have a minimal contribution to the early virologic response, at least in this population of previous nonresponders to PEG-IFN/RBV therapy.

In order to attain a rapid viral decline, it is important to achieve not only a high level of effectiveness but also a rapid second phase of viral decline. Here, the latter was rather slow in both treatment groups compared to what had been found in telaprevir-treated patients, and this was attributed in our model to a low loss rate of infected cells, δ, about 3 times smaller than that in noncirrhotic naive patients (δ = 0.18 day⁻¹ versus 0.60 day⁻¹) (15, 16). Those lower values may be related to several factors, such a lower penetration of PIs into infected cells in a highly scarred liver. Because the loss rate of infected cells is strongly related to the treatment duration needed to achieve SVR (15), our results suggest that the time to achieve SVR in this population could be longer than what had been predicted from clinical trials (15). Consistent with this prediction, the relapse rate in the ANRS CO20-CUPIC trial was 41% in both treatment groups (13), i.e., much higher than that reported in treatment-experienced patients in phase 3 clinical trials (12% to 27%) (9, 11, 22).

Regarding the use of early viral kinetic parameters for treatment prediction, we found that δ was higher in patients who subsequently achieved SVR (median δ^SVR = 0.27 day⁻¹, versus median δ^non-SVR = 0.14 day⁻¹, P = 0.03), suggesting that δ could be a relevant predictor of the outcome of triple therapy, as was the case for PEG-IFN/RBV bitherapy (38). In contrast, there was no significant relationship between the antiviral effectiveness of the PIs on SVR (Fig. 5a). This absence of a relationship is consistent with the hypothesis that in order to achieve SVR, it is necessary to have high antiviral effectiveness not only at treatment initiation, when the viral population is predominantly wild type and drug sensitive, but also at later times, when the viral population is predominantly resistant to PI agents (39, 40). The fact that the effectiveness of neither PEG-IFN nor RBV was associated with SVR is more surprising, as one would expect these agents to be equally active against wild-type and resistant virus. However, our patient population was both treatment experienced and cirrhotic, two major causes of insensitivity to PEG-IFN/RBV.

Clearly, the main limitation of this study was its small size. In a previous study, we evaluated by simulation the power to detect a difference in antiviral effectiveness between two treatment groups for a variety of designs (34). With a design comparable to that of the present study, i.e., 10 patients per group, 7 VL per patient, and an antiviral effectiveness of 99% versus 99.9%, the power to detect this difference was 100% with the same statistical method that we used in this analysis. However, further studies on larger populations are still needed to estimate more precisely the exposure-effect relationship (Fig. 4) and other kinetic parameters involved in the long-term virologic response. A second limitation is that only trough predose drug concentrations were collected and modeled. Thus, C_ss is the steady-state C_trough. Moreover, no information was collected on treatment adherence. The data analysis did not show any signal of a lack of adherence such as viral oscillations, which indicates that missed doses, if they occurred, did not have a major effect on the observed kinetics of decline. Here, we considered that the concentrations of the PIs were constant over time. A detailed pharmacokinetic analysis showed that the steady state of residual concentrations is attained after 2 days of treatment (41). As explained in detail in Guedj et al. (42), the fact that we neglected this initial buildup may explain why our estimate of the viral clearance rate, c, was lower than that previously found in treatment-naive patients (15). Further, the lack of information on the time of PEG-IFN injection also precluded a precise characterization between PEG-IFN exposure and the virological response. The fact that we used empirical models is less problematic for RBV, whose long elimination half-life resulting in a slow increase over time could be well characterized here (27). Moreover, as mentioned previously, in order to attain SVR, it is important for drugs to achieve higher effectiveness against PI-resistant virus. Because no sequencing was done here, we focused only on the early virological response when, presumably, the virus is predominantly drug sensitive. In order to estimate the effectiveness of a PI against a resistant virus, we would need to quantify and follow the proportion of resistant virus over time, as early as possible, for instance, using pyrosequencing (43).

A greater proportion of patients who received boceprevir were HCV genotype 1a infected relative to those who received telaprevir (P = 0.04). It has been well established that subgenotype is an important predictor of the response to treatment, and, for instance, the fact that telaprevir has a higher genetic barrier to resistance with genotype 1b than 1a (36) may explain why genotype non-1a-infected patients were preferentially treated with telaprevir rather than with boceprevir. However, the effect of subgenotype on the early viral kinetics, when most of the virus is drug-sensitive, is unknown and has never been investigated as far as we know. In our study, no significant effect of subgenotype on any of the parameters (c, δ, and EC₅₀^PI) was found.

The effect of RBV was analyzed using serum drug concentrations. Some authors prefer to use erythrocyte RBV concentration (44), which was not measured in the present study. However, a significant relationship was shown between erythrocyte RBV concentrations and serum concentrations (45), suggesting that serum RBV can be used for the assessment of early and sustained virological responses (46, 47).

To summarize, this study provides the first characterization of the relationship between drug concentrations involved in triple therapy and early HCV viral kinetics in patients treated with telaprevir or boceprevir. We found that the median value of antiviral effectiveness for telaprevir was similar to what had been found in treatment-naive patients and significantly larger than that in boceprevir-treated patients. In all patients, the second phase of viral decline was slow and may explain the high relapse rate observed in the ANRS CO20-CUPIC cohort. This suggests that, notwithstanding safety issues, longer treatment duration could improve the treatment efficacy and lead to a higher SVR rate. Lastly, the antiviral effectiveness of PEG-IFN was modest (<50%), suggesting that cirrhotic treatment-experienced patients may particularly benefit from the upcoming IFN-free treatment. Our approach, which shows the importance of PK data to disentangle the effects of drug combinations and to understand the variability in the virological response, is not specific to triple therapy and could also be used to optimize future IFN-free regimens, particularly in hard-to-treat patients.