Pegylated interferon fractal pharmacokinetics: individualized dosing for hepatitis C virus

24 Background : Despite recent advances in hepatitis C virus (HCV) therapeutics, the combination 25 of pegylated interferon (PEGIFN) and ribavirin (RBV) remains the cornerstone of treatment. 26 Optimization and individualization of PEGIFN dosing could improve outcomes. 27 Methods: Week one PEGIFN serum concentrations in 42 HCV genotype 1 infected patients 28 treated with conventional PEGIFN/RBV were analyzed using multi-compartmental 29 pharmacokinetic models. For each patient, pharmacokinetic parameter estimates, weight, age, 30 IL-28B single nucleotide polymorphism, CD4 count, baseline HCV RNA, gender, race, and HIV 31 status were examined using classification and regression tree analysis to identify factors 32 predictive of sustained viral response (SVR). Survival analysis was performed to compare the 33 time to undetectable viral load in patients with and without the highest scoring predictor. 34 Results: PEGIFN concentrations varied at least 87-fold. Pharmacokinetics were best described 35 by a two-compartment model with an 8.4hr absorption lag. Patient weight correlated with 36 PEGIFN systemic clearance based on fractal geometry relationships. SVR was achieved in 36% 37 of patients; PEGIFN cumulative one week area-under the curve (AUC) ≤0.79 mg*h/L scored 38 highest in predicting poor response, followed by weight ≥93.7 kg. Patients with a PEGIFN 39 AUC>0.79 mg*h/L achieved undetectable viral load more rapidly than those with a lower AUC 40 (Hazard ratio=1.63; 95% confidence interval 1.21-2.04). 41 Conclusions: PEGIFN exhibits wide pharmacokinetic variability, mainly driven by patient 42 weight, so that the standard dose may not reach levels needed to achieve SVR. Optimizing dose 43 to patient weight and PEGIFN AUC in the first week offers a solution to improve SVR, and to 44 potentially shorten duration of therapy. 45 46 on N ovem er 2, 2017 by gest httpaac.asm .rg/ D ow nladed fom INTRODUCTION 47 Treatments for hepatitis C virus (HCV) are rapidly changing. The combination of 48 pegylated interferon (PEGIFN), ribavirin (PEGIFN/RBV) and a directly acting antiviral (DAA) 49 is becoming the standard of care for HCV genotype 1 infected patients. PEGIFN/RBV 50 combined with the recently licensed NS3/4A protease inhibitors (PIs) such as telaprevir and 51 boceprevir, achieves an overall response rates of ~70% in treatment naïve patients (1-3). When 52 these PIs are administered as monotherapy or as dual therapy with an NS5A replication complex 53 inhibitor, there is rapid emergence of drug resistance (4-6). Therefore, combination therapies 54 will continue to include PEGIFN for the near future although “interferon–free” regimens are 55 currently being studied. Finally, in resource poor countries, the PEGIFN/RBV regimen is likely 56 to continue as the main regimen for all HCV infection for the foreseeable future. 57 Recently, single nucleotide polymorphisms (SNPs) of the IL-28B gene have been shown 58 to predict response to PEGIFN (7). Patients with the IL-28B rs12979860 CC genotype 59 demonstrated a 2-fold higher likelihood of sustained virologic response (SVR) than those with 60 CT/ TT SNPs (8). Although HCV RNA and IL-28B SNPs can predict likelihood of response, 61 they are not modifiable. We sought to determine if PEGIFN pharmacokinetics could have similar 62 predictive power; if so these are modifiable to improve SVR. To test this, we performed a 63 prospective pharmacokinetic-pharmacodynamic (PK/PD) study in patients, given the success of 64 this approach in optimizing other anti-infective agents (9). Uniquely, we examined 65 compartmental pharmacokinetics of PEGIFN and examined the effect of patient weight (body 66 mass) based on fractal geometry methods. Fractal mathematics describes the non-linear 67 relationship between processes such as metabolic rates and weight over several scales of 68 magnitude as well as to non-regular, rough, and continuously branching surfaces and edges such 69 on N ovem er 2, 2017 by gest httpaac.asm .rg/ D ow nladed fom as human organs’ often based on the “3/4 power law” (10-13). It is an analytic tool that is better 70 at identifying the effect of weight compared to standard methods. Here, we apply this to the 71 relationship between PEGIFN pharmacokinetics and body weight, and to SVR. 72 73 MATERIALS AND METHODS 74 Regulatory compliance. The study was approved by the Institutional Review Board 75 (IRB) of the University of Texas (UT) Southwestern Medical Center (IRB#102005-009). The 76 study was registered at www.clinicaltrials.gov (NCT00703560). 77 Patients and setting. The study was performed at UT Southwestern Medical Center 78 teaching hospitals in Dallas, Texas. The first patient was enrolled March 1, 2004 and the last 79 patient enrolled January 25, 2010. All subjects gave written informed consent after the study was 80 explained to the participant in detail. 81 Inclusion criteria was HCV infection by serum anti-HCV enzyme immunoassay, an HCV 82 viral load of >1000 IU/mL, documented HCV genotype by a CLIA certified lab, age>18 to 65 83 years, and willingness to use two or more methods of birth control in women of childbearing age. 84 Subjects were excluded if they had uncontrolled diabetes mellitus, psychiatric illness, 85 autoimmune disease, decompensated liver disease, or prior interferon therapy. Patients with other 86 liver diseases such as alcohol, hepatitis B virus infection, Wilson’s disease, hemochromatosis, or 87 alpha-1 antitrypsin, were excluded. Patients were excluded from the study if they were unwilling 88 to be admitted for a 48-hour period for serial blood draws. Patients who did not have serum 89 samples at needed time points were excluded. HIV-infected patients had either to be anti90 retroviral therapy (ART) naïve, or if on ART they had to be on a stable ART regimen for at least 91 12 weeks. The ART regimen could not include didanosine due the risk of fatal hyperlactatemia. 92 on N ovem er 2, 2017 by gest httpaac.asm .rg/ D ow nladed fom Patients had to have a CD4 T-cell count >200 cells/mm within the 12 weeks prior to enrollment. 93 Subjects were excluded if they had symptomatic HIV disease including an AIDS-defining 94 illness. 95 Study procedures. This was an open label observational study of HCV infected patients 96 who were otherwise candidates for treatment with PEGIFN/RBV. Patients that met inclusion 97 criteria and consented to participate were admitted for 48 hours to initiate treatment. All patients 98 received subcutaneous pegylated interferon –α-2a 180 μg/week plus ribavirin (PEGIFN/RBV), 99 which was dosed based on weight: if <75 kg 1000 mg daily or if > 75 kg 1200 mg daily for 48 100 weeks. Ribavirin doses were administered as a twice-daily regimen. Blood was drawn at 0, 4, 24, 101 96, and 168hrs after the initial subcutaneous injection, and then spun at room temperature at 102 1500 rpm for 10 minutes within 4 hours of collection. The plasma was decanted into freezer 103 tubes and stored in a -80°C freezer for subsequent HCV RNA quantification and for drug assays. 104 Blood was also collected on day 3, 5, 9, 11, 14, 21, 28, 42, and 56 days for additional HCV RNA 105 quantification, which was performed using the VERSANT 3.0 branched DNA technology 106 (Siemens Medical Solutions Diagnostics, Tarrytown, N.Y.), with a dynamic range of 615107 7,692,310 IU/mL. 108 Standard safety assessments, and urine pregnancy tests in women, were conducted at 109 each study visit. As PEGIFN/RBV was the standard of care, adverse events that were known 110 side effects of the medications were monitored and managed by study investigators as dictated 111 by best practice. 112 Genotype studies. Human genomic DNA was extracted from whole blood or serum 113 samples with QIAamp DNA Blood Mini Kit (Qiagen). The targeted region, including SNP 114 rs12979860, was amplified by PCR and PCR products were purified and sequenced (14). 115 on N ovem er 2, 2017 by gest httpaac.asm .rg/ D ow nladed fom Interferon assay. Serum samples were kept frozen at -80°C after processing, and then 116 sent to Hoffman-La Roche, Inc., laboratories in Nutley, NJ, on dry ice. PEGIFN concentrations 117 were determined using a sandwich ELISA, in which capture antibody (affinity purified rabbit 118 polyclonal anti-PEG-IFN) was coated onto micro-titer plates. Serum samples were added to the 119 plate and incubated. The plate was washed and detection antibody (mouse monoclonal anti-PEG) 120 was added and then further incubated. The plate was washed again and peroxidase conjugated 121 goat anti-mouse IgM) was added and incubated. After a final wash, a substrate solution was 122 added. Color developed in proportion to the amount of PEGIFN present in the sample. This 123 assay has a range of 250-5,000 pg/mL. The intra-batch precision was 2.5-8.4%, and the accuracy 124 was -5.5-13.2%. The inter-batch precision was 11.7-16.6% and the accuracy 0.4-9.0%. 125 Population pharmacokinetic analyses. PEGIFN concentrations were modeled using 126 the ADAPT 5 program (15). One compartment model, one compartment model with lag, two127 compartment model, two-compartment model with lag, and a three compartment model, were 128 examined. First, initial guesses of the population pharmacokinetic parameter estimates for each 129 model were identified using the standard two-stage approach. The estimates were then used in 130 further pharmacokinetic analysis using the maximum likelihood solution via the expectation131 maximization (MLEM) algorithm. Four criteria were used to choose the best compartment 132 model: Akaike’s Information criteria, Bayesian Information Criteria, negative log likelihoods, 133 and the law of parsimony (16-18). Next, the relationship between pharmacokinetic parameter and 134 the following covariates were examined in log-log scatter plots for the selected model: self135 identified race (African-American versus other) given the known poor treatment response in this 136 group, IL-28B SNPs, HIV status, weight, CD4 count, gender and age. Fractal relationships are 137 easiest to spot with log-transformed data, i.e., in log-log plots [11]. When a relationship was 138 on N ovem er 2, 2017 by gest httpaac.asm .rg/ D ow nladed fom identified, the slope of the relationship between covariate and pharmacokinetic parameter was 139 calculated. Those covariates and initial estimates of slope were then added in the subroutine 140 COVMOD of ADAPT. The relationship between covariate and pharmacokinetic parameter was 141 then determined using MLEM to yield the final model estimates. 142 Classification and regression tree (CART) and survival analysis 143 CART analysis was used to identify the best predictors of virologic outcome, and to identify the 144 threshold cut-off values for such predictors. CART analysis models are very accurate at 145 identifying and estimating complex high-order nonlinear interactions. The outcomes of this non146 parametric and recursive partitioning analysis is predictive accuracy, as opposed to statistical 147 association with the standard statistical approaches. The outcomes examined were SVR, defined 148 as HCV RNA <615 IU/ml at least 24 weeks after termination of antiviral therapy, as well as 149 rapid virologic response (RVR), defined as (HCV RNA <615 IU/mL) at week 4. Variables 150 examined for prediction of RVR and SVR included PEGIFN peak concentration, trough 151 concentration, 168h (1 week) area under the concentration-time curve (AUC), patient weight, 152 age, IL-28B SNPs, CD4 count, initial HCV viral load, gender, race, and HIV status. All 153 outcomes were examined so that they could be ranked by the most important determinant of 154 outcome. First, CART analysis was used for rank variables. Significant variables were chosen if 155 the CART score was ≥20%. The analysis also identified clinically meaningful cut-offs points for 156 the selected continuous variables. The splitting criterion was based on the Gini index (19). 157 Several trees were constructed, pruned and went through 10-fold cross-validation (20). The 158 optimum tree was selected based on relative misclassification costs, PK/PD considerations and 159 biological plausibility. CART was performed using the Salford Predictive Miner System 160 software, San Diego, CA (20). 161 on N ovem er 2, 2017 by gest httpaac.asm .rg/ D ow nladed fom Next, we validated the CART analysis findings using standard statistical approaches 162 familiar to most clinicians. First, the main predictor of SVR (highest decision node) in CART 163 was examined in a survivor analysis to determine time to undetectable viral load in patients 164 above the cut-off value versus those below. Since patients who failed therapy were taken off 165 PEGIFN/RBV after week 12 and had no viral loads done after that, data for these patients was 166 censored at the 12-week time point. In addition, since the ratio of hazard functions was not the 167 same at all time points, the Gehan-Breslow-Wilcoxon method was used to compare the survival 168 curves. Next, the CART outputs, including data obtained by examination of surrogate and 169 competitor variables, were used as inputs for parametric univariate analysis and multivariate 170 logistic stepwise analyses. Modifiable and clinically important variables were initially added in 171 the several models examined and then sequentially (stepwise) removed using backward 172 regression if p was <0.1 based on the Wald statistic. The survival analysis and the multivariate 173 analyses were performed in SPSS version 12. 174 RESULTS 175 Clinical Features 176 Of 58 HCV genotype 1 patients eligible for the study, 42 (72%) met inclusion criteria and 177 are included in the present analyses. The patients’ demographic, clinical and laboratory 178 characteristics are shown in Table 1. The IL-28B SNP could not be determined in 4 (9.5%) 179 patients. Table 1 shows that self-identified “race”/ethnicity was associated with IL-28 genotype 180 allele distribution. Liver disease severity prior to start of therapy did not significantly differ 181 between the race/ethnic groups (Table 1). The three most common adverse events among the 42 182 patients were neutropenia (83%), depression (79%) and anemia (62%); one patient was admitted 183 on N ovem er 2, 2017 by gest httpaac.asm .rg/ D ow nladed fom to the hospital for pulmonary related issues not thought to be due to HCV treatment. These 184 adverse events, including hospitalization for anemia, were at expected rates. 185 RVR was achieved in 11 (26%) patients and SVR in 15 (36%) out of 42 patients. There 186 were 4 patients who stopped therapy due to adverse events and 11 (26%) other patients who 187 relapsed after a documented viral response. Twelve (29%) patients were non-responders, failing 188 to achieve a 2-log drop at 12 weeks and for HCV to remain suppressed until 48 weeks. 189 Interferon Levels and CART analysis 190 Altogether, 198 PEGIFN concentrations collected during the first week of therapy were 191 analyzed. The concentration-versus-time profiles for each of the 42 patients are shown in Figure 192 1. The figure demonstrates wide variability in concentrations at any of the time points. As an 193 example, the ratio of the highest to the lowest drug concentration 24hrs after PEGIFN injection 194 was 87, while the 168hr trough concentrations varied 50 fold. Population pharmacokinetic 195 analysis revealed that a two-compartment model with absorption lag best described PEGIFN 196 pharmacokinetics (supplementary Table 1). Scatter plots revealed that only patient weight 197 significantly correlated with pharmacokinetic parameters (Figure 2). The slopes of the log-log 198 plots of clearance and volume encompassed 3⁄4, consistent with the 3⁄4 power law of fractal 199 geometry (11, 12, 21, 22). Therefore, weight was examined as a covariate in ADAPT. The 200 pharmacokinetic parameter estimates for this final model are shown in Table 2. 201 CART analysis revealed that the highest ranked predictor of RVR was a PEGIFN 202 trough≤0.01 mg/L, which outranked the second placed IL-28B SNP (figure 3A). In this analysis, 203 the 168hr AUC was ranked third in importance. However, RVR itself is a surrogate for SVR. On 204 the other hand, PEGIFN AUC had the highest score for predicting SVR, making it the first 205 “decision” node (Figure 3B) for long term outcome; the cut-off point was a 168hr AUC of 0.79 206 on N ovem er 2, 2017 by gest httpaac.asm .rg/ D ow nladed fom mg*h/L. To put this into context, if one used standard statistical approaches, the odds ratio for 207 failure below the AUC cut-off point was 7.0 (p=0.02). The next node for SVR was the patient’s 208 weight, with higher failures above 93.70 kg. Finally, the trough concentration had the third 209 highest score for SVR (Figure 3B). Importantly, the IL-28B genotype scored below 20%. The 210 minimum and maximum predictive accuracy was 68% and 95%, respectively, for all trees that 211 were examined. The overall prediction success for the CART models was 76%. We then 212 compared time to viral load suppression in patients who achieved AUC above the cut-off of 0.79 213 mg*h/L versus those below, with results shown in Figure 4. The median time to viral load 214 suppression with the higher AUC was 56 days compared to 91 days (HR=1.625; 95% CI: 1.207215 2.043). This means that achievement of AUC above the threshold is associated with faster time 216 to viral load suppression. 217 To confirm the CART findings, univariate and multivariate logistic were performed, with 218 the univariate analysis results shown in supplementary Table 1. Weight and IL-28B CC allele 219 remained significantly associated with RVR in several stepwise multivariate logistic models. 220 The adjusted odds ratio for weight was 0.90 (95% CI 0.83–0.98) and that for IL-28B CC allele 221 was 7.91 (95% CI 1.35–46.38). For SVR, IL-28B CC allele was not associated with outcome in 222 multivariate analysis. Weight and the PEGIFN AUC were associated with SVR. The adjusted 223 odds ratio of success with an AUC>0.79 mg*h/L was 9.09 (95% CI 1.23–100). Thus, the 224 multivariate logistic analysis confirmed CART analysis. 225 Discussion 226 Our study revealed two important findings with direct clinical implications. First, the 227 relationship between AUC and SVR superseded that of IL28B SNPs, and should therefore be 228 more useful if dosage corrections can be made based on these parameters. Since the 229 on N ovem er 2, 2017 by gest httpaac.asm .rg/ D ow nladed fom pharmacokinetics were calculated after the first PEGIFN dose, early intervention based on these 230 findings could be undertaken as early as 2 weeks of therapy, once the interferon levels were 231 available. The determination of AUC requires multiple sampling times during the first week, 232 however, given the slow clearance of PEGIFN, three time points during the first 1 week would 233 be adequate, especially if optimal sampling theory is applied. With these, a full AUC for the first 234 week can easily be determined by physicians and higher doses administered during subsequent 235 weeks in order to achieve AUCs above 0.79 mg*h/L. The precise dose will depend on the AUCs 236 achieved during the first week, after which each patient can begin receiving an individualized 237 dose. With faster times to undetectable viral load encountered with the higher AUCs, and in 238 combination with DAAs, treatment duration might be further shortened based on the concept of 239 response-guided therapy (3). 240 The second important finding is the relationship between weight and both PEGIFN 241 pharmacokinetics and virologic response. We identified the compartmental pharmacokinetics of 242 PEGIFN, which allowed us to accurately estimate pharmacokinetic parameters and then examine 243 them with relationship to such factors as weight. Clearance and volume of distribution were 244 partly determined by weight, based on fractal geometry dependent 3⁄4 power law (10, 13). While 245 we and others have demonstrated that clearance of xenobiotics such as ethambutol and 246 echinocandins obeys these fractal geometry-determined relationships (20-22), the current study 247 demonstrates this relationship for a type 1 interferon (IFN-α). Type 1 interferons, such as IFN-α, 248 are endogenously produced as part of the body’s response to infection: PEGIFN merely provides 249 higher concentrations that are much more slowly cleared. Our observations suggest that 250 metabolism of this vital component of the innate immune system is affected by body mass. The 251 implication is that even the initial PEGIFN dose should be adjusted according to weight 252 on N ovem er 2, 2017 by gest httpaac.asm .rg/ D ow nladed fom especially in overweight patients, not in a linear fashion, but proportional to (weight/typical 253 weight). Many previous studies have shown that higher body mass index (BMI) and weight 254 are associated with reduced response (23-25). However, early viral kinetics and SVR were not 255 associated with BMI when weight-adjusted dosing of RBV was used (26, 27). Our study 256 suggests that the relationship of weight to response is not a linear relationship, and that the 257 adjustment should be by the factor of (weight/typical weight). 258 There are several limitations to our study. Previous attempts to determine the role of drug 259 concentrations for interferon based compounds in the treatment of HCV have generally met with 260 limited success (28-30). Reasons for previous lack of predictive insights from others’ data might 261 be due to limited precision of pharmacokinetic parameter estimates, use of a single time point 262 drug “concentration”, the type of viral kinetic measurements utilized, or the statistical methods to 263 determine the relationship(26, 27). Second, our sample size was only 42, which could throw the 264 generalizability of our study into question. As an example, studies that established the role of IL265 28B SNPs examined hundreds of patients. However, the multiple approaches we took increased 266 the ability of our study to detect differences. Our study design allowed for intensive 267 pharmacokinetic sampling, and compartmental pharmacokinetic analyses, which allowed for 268 accurate discrimination of pharmacokinetic parameters for each patient. Finally, we utilized 269 CART models, which have the advantage of accuracy for higher order non-linear relationships. 270 In conclusion, this study examined population pharmacokinetics of PEGIFN in HCV 271 infected patients. Wide pharmacokinetic variability was encountered. PEGIFN AUC was 272 determined to be the most predictive factor for SVR, its effect superseding that of IL28B 273 polymorphism. Weight was also associated with both PEGIFN clearance (and hence AUC) and 274 SVR, but the relationship was not linear; rather it followed fractal geometry relationships. The 275 on N ovem er 2, 2017 by gest httpaac.asm .rg/ D ow nladed fom major implications are that PEGIFN dosing can be improved by individualizing dosing based on 276 weight and measured drug concentrations, leading to improved SVR and more rapid virologic 277 response. 278 Funding 279 This work was supported by the National Institute of Health (career development award K23280 AI-065630 and NCTCTSI Pilot Program grant UL1 RR024982 to MKJ. The work was also 281 supported by Clinical Translational Research Center grant UL1-RR-024982. Additional support 282 from the Jeanne Roberts and Rollin and Mary Ella King Funds of the Southwestern Medical 283 Foundation. TG and JGP were supported by the National Institutes of Health (NIH) via the NIH 284 Director New Innovator Award (National Institutes of General Medical Sciences) DP2 285 OD001886 and National Institute of Allergy and Infectious Diseases R01AI079497. 286