Use of Patient Health Records to Quantify Drug-Related Pro-arrhythmic Risk.

There is an increasing expectation that computational approaches may supplement existing human decision-making. Frontloading of models for cardiac safety prediction is no exception to this trend, and ongoing regulatory initiatives propose use of high-throughput in vitro data combined with computational models for calculating proarrhythmic risk. Evaluation of these models requires robust assessment of the outcomes. Using FDA Adverse Event Reporting System reports and electronic healthcare claims data from the Truven-MarketScan US claims database, we quantify the incidence rate of arrhythmia in patients and how this changes depending on patient characteristics. First, we propose that such datasets are a complementary resource for determining relative drug risk and assessing the performance of cardiac safety models for regulatory use. Second, the results suggest important determinants for appropriate stratification of patients and evaluation of additional drug risk in prescribing and clinical support algorithms and for precision health.

Model-Informed Artificial Intelligence: Reinforcement Learning for Precision Dosing.

The availability of multidimensional data together with the development of modern techniques for data analysis represent an exceptional opportunity for clinical pharmacology. Data science-defined in this special issue as the novel approaches to the collection, aggregation, and analysis of data-can significantly contribute to characterize drug-response variability at the individual level, thus enabling clinical pharmacology to become a critical contributor to personalized healthcare through precision dosing. We propose a minireview of methodologies for achieving precision dosing with a focus on an artificial intelligence technique called reinforcement learning, which is currently used for individualizing dosing regimen in patients with life-threatening diseases. We highlight the interplay of such techniques with conventional pharmacokinetic/pharmacodynamic approaches and discuss applicability in drug research and early development.

Importance of target-mediated drug disposition for small molecules.

Target concentration is typically not considered in drug discovery. However, if targets are expressed at relatively high concentrations and compounds have high affinity, such that most of the drug is bound to its target, in vitro screens can give unreliable information on compound affinity. In vivo, a similar situation will generate pharmacokinetic (PK) profiles that deviate greatly from those normally expected, owing to target binding affecting drug distribution and clearance. Such target-mediated drug disposition (TMDD) effects on small molecules have received little attention and might only become apparent during clinical trials, with the potential for data misinterpretation. TMDD also confounds human microdosing approaches by providing therapeutically unrepresentative PK profiles. Being aware of these phenomena will improve the likelihood of successful drug discovery and development.

PKPD modeling of acquired resistance to anti-cancer drug treatment.

Non-small cell lung cancer (NSCLC) patients greatly benefit from the treatment with tyrosine kinase inhibitors (TKIs) targeting the epidermal growth factor receptor (EGFR). However, emergence of acquired resistance inevitable occurs after long-term treatment in most patients and limits clinical improvement. In the present study, resistance to drug treatment in patient-derived NSCLC xenograft mice was assessed and modeling and simulation was applied to understand the dynamics of drug resistance as a basis to explore more beneficial drug regimen. Two semi-mechanistic models were fitted to tumor growth inhibition profiles during and after treatment with erlotinib or gefitinib. The base model proposes that as a result of drug treatment, tumor cells stop proliferating and undergo several stages of damage before they eventually die. The acquired resistance model adds a resistance term to the base model which assumes that resistant cells are emerging from the pool of damaged tumor cells. As a result, tumor cells sensitive to drug treatment will either die or be converted to a drug resistant cell population which is proliferating at a slower growth rate as compared to the sensitive cells. The observed tumor growth profiles were better described by the resistance model and emergence of resistance was concluded. In simulation studies, the selection of resistant cells was explored as well as the time-variant fraction of resistant over sensitive cells. The proposed model provides insight into the dynamic processes of emerging resistance. It predicts tumor regrowth during treatment driven by the selection of resistant cells and explains why faster tumor regrowth may occur after discontinuation of TKI treatment. Finally, it is shown how the semi-mechanistic model can be used to explore different scenarios and to identify optimal treatment schedules in clinical trials.

Metabolic Profiling of Human Long-Term Liver Models and Hepatic Clearance Predictions from In Vitro Data Using Nonlinear Mixed-Effects Modeling.

Early prediction of human clearance is often challenging, in particular for the growing number of low-clearance compounds. Long-term in vitro models have been developed which enable sophisticated hepatic drug disposition studies and improved clearance predictions. Here, the cell line HepG2, iPSC-derived hepatocytes (iCell®), the hepatic stem cell line HepaRG™, and human hepatocyte co-cultures (HµREL™ and HepatoPac®) were compared to primary hepatocyte suspension cultures with respect to their key metabolic activities. Similar metabolic activities were found for the long-term models HepaRG™, HµREL™, and HepatoPac® and the short-term suspension cultures when averaged across all 11 enzyme markers, although differences were seen in the activities of CYP2D6 and non-CYP enzymes. For iCell® and HepG2, the metabolic activity was more than tenfold lower. The micropatterned HepatoPac® model was further evaluated with respect to clearance prediction. To assess the in vitro parameters, pharmacokinetic modeling was applied. The determination of intrinsic clearance by nonlinear mixed-effects modeling in a long-term model significantly increased the confidence in the parameter estimation and extended the sensitive range towards 3% of liver blood flow, i.e., >10-fold lower as compared to suspension cultures. For in vitro to in vivo extrapolation, the well-stirred model was used. The micropatterned model gave rise to clearance prediction in man within a twofold error for the majority of low-clearance compounds. Further research is needed to understand whether transporter activity and drug metabolism by non-CYP enzymes, such as UGTs, SULTs, AO, and FMO, is comparable to the in vivo situation in these long-term culture models.

Translational PK/PD modeling to increase probability of success in drug discovery and early development.

In this review we present ways in which translational PK/PD modeling can address opportunities to enhance probability of success in drug discovery and early development. This is achieved by impacting efficacy and safety-driven attrition rates, through increased focus on the quantitative understanding and modeling of translational PK/PD. Application of the proposed principles early in the discovery and development phases is anticipated to bolster confidence of successfully evaluating proof of mechanism in humans and ultimately improve Phase II success. The present review is centered on the application of predictive modeling and simulation approaches during drug discovery and early development, and more specifically of mechanism-based PK/PD modeling. Case studies are presented, focused on the relevance of M&S contributions to real-world questions and the impact on decision making.

Combining Nonclinical Experiments with Translational PKPD Modeling to Differentiate Erlotinib and Gefitinib.

We quantitatively compare the efficacy of two approved EGFR tyrosine kinase inhibitors, erlotinib and gefitinib, based on in vivo and in vitro data and show how a modeling approach can be used to scale from animal to humans. Gefitinib shows a higher tumor uptake in cancer patients, and we explored the potential impact on pharmacologic and antitumor activity in in vitro and in xenograft mice. Tumor growth inhibition was monitored, and the pharmacokinetics (PK) in plasma and tumor, as well as temporal changes of phospho-Erk (pErk) signals were examined in patient-derived tumor xenograft mice. These data were integrated in a translational PKPD model, allowing us to project an efficacious human dose, which we retrospectively compared with prescribed doses for cancer patients. In vitro experiments showed that cell-cycle arrest was similar for erlotinib and gefitinib. Similar pERK biomarker responses were obtained despite a 6.6-fold higher total tumor exposure for gefitinib. The PKPD model revealed a 3.7-fold higher in vivo potency for gefitinib, which did not translate into a lower anticipated efficacious dose in humans. The model-based dose prediction matched the recommended clinical doses well. These results suggest that despite having lower total tumor-to-plasma ratios, active drug exposure at target site is higher for erlotinib. Considering the PK properties, this translates in a 50% lower recommended daily dose of erlotinib in cancer patients. In summary, total exposure at target site is not suitable to rank compounds, and an integrated modeling and experimental approach can assess efficacy more accurately. Mol Cancer Ther; 15(12); 3110-9. ©2016 AACR.

Recent developments in using mechanistic cardiac modelling for drug safety evaluation.

On the tenth anniversary of two key International Conference on Harmonisation (ICH) guidelines relating to cardiac proarrhythmic safety, an initiative aims to consider the implementation of a new paradigm that combines in vitro and in silico technologies to improve risk assessment. The Comprehensive In Vitro Proarrhythmia Assay (CiPA) initiative (co-sponsored by the Cardiac Safety Research Consortium, Health and Environmental Sciences Institute, Safety Pharmacology Society and FDA) is a bold and welcome step in using computational tools for regulatory decision making. This review compares and contrasts the state-of-the-art tools from empirical to mechanistic models of cardiac electrophysiology, and how they can and should be used in combination with experimental tests for compound decision making.

Simulation-Based Evaluation of PK/PD Indices for Meropenem Across Patient Groups and Experimental Designs.

PURPOSE: Antibiotic dose predictions based on PK/PD indices rely on that the index type and magnitude is insensitive to the pharmacokinetics (PK), the dosing regimen, and bacterial susceptibility. In this work we perform simulations to challenge these assumptions for meropenem and Pseudomonas aeruginosa. METHODS: A published murine dose fractionation study was replicated in silico. The sensitivity of the PK/PD index towards experimental design, drug susceptibility, uncertainty in MIC and different PK profiles was evaluated. RESULTS: The previous murine study data were well replicated with fT > MIC selected as the best predictor. However, for increased dosing frequencies fAUC/MIC was found to be more predictive and the magnitude of the index was sensitive to drug susceptibility. With human PK fT > MIC and fAUC/MIC had similar predictive capacities with preference for fT > MIC when short t1/2 and fAUC/MIC when long t1/2. CONCLUSIONS: A longitudinal PKPD model based on in vitro data successfully predicted a previous in vivo study of meropenem. The type and magnitude of the PK/PD index were sensitive to the experimental design, the MIC and the PK. Therefore, it may be preferable to perform simulations for dose selection based on an integrated PK-PKPD model rather than using a fixed PK/PD index target.

Influence of molecular size on tissue distribution of antibody fragments.

Biodistribution coefficients (BC) allow estimation of the tissue concentrations of proteins based on the plasma pharmacokinetics. We have previously established the BC values for monoclonal antibodies. Here, this concept is extended by development of a relationship between protein size and BC values. The relationship was built by deriving the BC values for various antibody fragments of known molecular weight from published biodistribution studies. We found that there exists a simple exponential relationship between molecular weight and BC values that allows the prediction of tissue distribution of proteins based on molecular weight alone. The relationship was validated by a priori predicting BC values of 4 antibody fragments that were not used in building the relationship. The relationship was also used to derive BC50 values for all the tissues, which is the molecular weight increase that would result in 50% reduction in tissue uptake of a protein. The BC50 values for most tissues were found to be ~35 kDa. An ability to estimate tissue distribution of antibody fragments based on the BC vs. molecular size relationship established here may allow better understanding of the biologics concentrations in tissues responsible for efficacy or toxicity. This relationship can also be applied for rational development of new biotherapeutic modalities with optimal biodistribution properties to target (or avoid) specific tissues.

Model-based assessment of erlotinib effect in vitro measured by real-time cell analysis.

Real time cell analysis (RTCA) is an impedance-based technology which tracks various living cell characteristics over time, such as their number, morphology or adhesion to the extra cellular matrix. However, there is no consensus about how RTCA data should be used to quantitatively evaluate pharmacodynamic parameters which describe drug efficacy or toxicity. The purpose of this work was to determine how RTCA data can be analyzed with mathematical modeling to explore and quantify drug effect in vitro. The pharmacokinetic-pharmacodynamic erlotinib concentration profile predicted by the model and its effect on the human epidermoïd carcinoma cell line A431 in vitro was measured through RTCA output, designated as cell index. A population approach was used to estimate model parameter values, considering a plate well as the statistical unit. The model related the cell index to the number of cells by means of a proportionality factor. Cell growth was described by an exponential model. A delay between erlotinib pharmacokinetics and cell killing was described by a transit compartment model, and the effect potency, by an E max function of erlotinib concentration. The modeling analysis performed on RTCA data distinguished drug effects in vitro on cell number from other effects likely to modify the relationship between cell index and cell number. It also revealed a time-dependent decrease of erlotinib concentration over time, described by a mono-exponential pharmacokinetic model with nonspecific binding.

Translational PK/PD modeling for cardiovascular safety assessment of drug candidates: Methods and examples in drug development.

INTRODUCTION: Cardiovascular toxicity is a significant cause of candidate failure in drug development. Pharmacokinetic/pharmacodynamic (PK/PD) modeling may reduce attrition by improving the understanding of the relationship between drug exposure and changes in cardiovascular endpoints. Diverse examples are discussed that elucidate how modeling can facilitate the interpretation of cardiovascular safety data in animals and enable quantitative translation of preclinical findings to man. METHODS: Twelve compounds under development in diverse therapeutic areas were tested in cardiovascular safety studies in the telemetered beagle dog and cynomolgus monkey. Drug-induced changes observed in different cardiovascular endpoints (QRS complex and QTc interval of the ECG, heart rate, blood pressure, and myocardial contractility) were described by means of PK/PD modeling. A range of direct and indirect effect models were employed to characterize the plasma concentration-cardiovascular effect relationship for each compound. RESULTS: For every drug candidate the proposed PK/PD models appropriately described the cardiovascular effects observed in dog and monkey. Two of the compounds subsequently reached clinical development and cardiovascular data were generated in first-in-human clinical trials. For one drug candidate, a threshold model was used to describe QTc prolongation in the monkey and man. Blood pressure changes induced by the second compound were linked to plasma exposure in dog and human via an indirect response model. In both cases it was found that translational modeling accurately predicted the human response observed during clinical development. DISCUSSION: In this article, a range of PK/PD models are discussed that successfully described cardiovascular safety findings in the preclinical setting. Where clinical data were available, it was found that translational modeling enabled the accurate prediction of outcomes in man and facilitated the description of the therapeutic index. PK/PD modeling is thus demonstrated as a powerful tool to aid in the quantitative cardiovascular safety assessment of drug candidates and the optimization of early clinical study protocols.

Physiologically based pharmacokinetic modeling of CYP3A4 induction by rifampicin in human: influence of time between substrate and inducer administration.

The induction of cytochrome P450 enzymes (CYPs) is an important source of drug-drug interaction (DDI) and can result in pronounced changes in pharmacokinetics (PK). Rifampicin (RIF) is a potent inducer of CYP3A4 and also acts as a competitive inhibitor which can partially mask the induction. The objective of this study was to determine a clinical DDI study design for RIF resulting in maximum CYP3A4 induction. A physiologically based pharmacokinetic (PBPK) model was developed to project the dynamics and magnitude of CYP3A4 induction in vivo from in vitro data generated with primary human hepatocytes. The interaction model included both inductive and inhibitory effects of RIF on CYP3A4 in gut and liver and accounting for the observed RIF auto-induction. The model has been verified for 4 CYP3A4 substrates: midazolam, triazolam, alfentanil and nifedipine using plasma concentration data from 20 clinical study designs with intravenous (n=7) and oral (n=13) administrations. Finally, the influence of the time between RIF and substrate administration was explored for the interaction between midazolam and RIF. The model integrating in vitro induction parameters correctly predicted intravenous induction but underestimated oral induction with 30% of simulated concentrations more than 2-fold higher than of observed data. The use of a 1.6-fold higher value for the maximum induction effect (Emax) improved significantly the accuracy and precision of oral induction with 82% of simulated concentrations and all predicted PK parameters within 2-fold of observed data. Our simulations suggested that a concomitant administration of RIF and midazolam resulted in significant competitive inhibition limited to intestinal enzyme. Accordingly, a maximum induction effect could be achieved with a RIF pretreatment of 600 mg/day during 5 days and a substrate administration at least 2 h after the last RIF dose. A period of 2 weeks after RIF removal was found sufficient to allow return to baseline levels of enzyme.

PK/PD assessment in CNS drug discovery: Prediction of CSF concentration in rodents for P-glycoprotein substrates and application to in vivo potency estimation.

The unbound drug concentration in brain parenchyma is considered to be the relevant driver for interaction with central nervous system (CNS) biological targets. Drug levels in cerebrospinal fluid (C_CSF) are frequently used surrogates for the unbound concentrations in brain. For drugs actively transported across the blood-brain barrier (BBB), C_CSF differs from unbound plasma concentration (Cu_p) to an extent that is commonly unknown. In this study, the relationship between CSF-to-unbound plasma drug partitioning in rats and the mouse Pgp (Mdr1a) efflux ratio (ER) obtained from in vitro transcellular studies has been investigated for a set of 61 CNS compounds exhibiting substantial diversity in chemical structure and physico-chemical properties. In order to understand the in vitro-in vivo extrapolation of Pgp efflux, a mechanistic model was derived relating in vivo CNS distribution kinetics to in vitro active transport. The model was applied to predict C_CSF from Cu_p and ER data for 19 proprietary Roche CNS drug candidates. The calculated CSF concentrations were correlated with CNS pharmacodynamic responses observed in rodent models. The correlation between in vitro and in vivo potency for different pharmacological endpoints indicated that the predicted C_CSF is a valuable surrogate of the concentration at the target site. Overall, C_CSF proved superior description of PK/PD data than unbound plasma or total brain concentration for Mdr1a substrates. Predicted C_CSF can be used as a default approach to understand the PK/PD relationships in CNS efficacy models and can support the extrapolation of efficacious brain exposure for new drug candidates from rodent to man.

Pharmacokinetic/Pharmacodynamic predictors of clinical potency for hepatitis C virus nonnucleoside polymerase and protease inhibitors.

This analysis was conducted to determine whether the hepatitis C virus (HCV) viral kinetics (VK) model can predict viral load (VL) decreases for nonnucleoside polymerase inhibitors (NNPolIs) and protease inhibitors (PIs) after 3-day monotherapy studies of patients infected with genotype 1 chronic HCV. This analysis includes data for 8 NNPolIs and 14 PIs, including VL decreases from 3-day monotherapy, total plasma trough concentrations on day 3 (C(min)), replicon data (50% effective concentration [EC(50)] and protein-shifted EC(50) [EC(50,PS)]), and for PIs, liver-to-plasma ratios (LPRs) measured in vivo in preclinical species. VK model simulations suggested that achieving additional log(10) VL decreases greater than one required 10-fold increases in the C(min). NNPolI and PI data further supported this result. The VK model was successfully used to predict VL decreases in 3-day monotherapy for NNPolIs based on the EC(50,PS) and the day 3 C(min). For PIs, however, predicting VL decreases using the same model and the EC(50,PS) and day 3 C(min) was not successful; a model including LPR values and the EC(50) instead of the EC(50,PS) provided a better prediction of VL decrease. These results are useful for designing phase 1 monotherapy studies for NNPolIs and PIs by clarifying factors driving VL decreases, such as the day 3 C(min) and the EC(50,PS) for NNPolIs or the EC(50) and LPR for PIs. This work provides a framework for understanding the pharmacokinetic/pharmacodynamic relationship for other HCV drug classes. The availability of mechanistic data on processes driving the target concentration, such as liver uptake transporters, should help to improve the predictive power of the approach.

Mass spectrometry-based quantification of CYP enzymes to establish in vitro/in vivo scaling factors for intestinal and hepatic metabolism in beagle dog.

PURPOSE: Physiologically based models, when verified in pre-clinical species, optimally predict human pharmacokinetics. However, modeling of intestinal metabolism has been a gap. To establish in vitro/in vivo scaling factors for metabolism, the expression and activity of CYP enzymes were characterized in the intestine and liver of beagle dog. METHODS: Microsomal protein abundance in dog tissues was determined using testosterone-6ß-hydroxylation and 7-hydroxycoumarin-glucuronidation as markers for microsomal protein recovery. Expressions of 7 CYP enzymes were estimated based on quantification of proteotypic tryptic peptides using multiple reaction monitoring mass spectrometry. CYP3A12 and CYP2B11 activity was evaluated using selective marker reactions. RESULTS: The geometric mean of total microsomal protein was 51 mg/g in liver and 13 mg/cm in intestine, without significant differences between intestinal segments. CYP3A12, followed by CYP2B11, were the most abundant CYP enzymes in intestine. Abundance and activity were higher in liver than intestine and declined from small intestine to colon. CONCLUSIONS: CYP expression in dog liver and intestine was characterized, providing a basis for in vitro/in vivo scaling of intestinal and hepatic metabolism.

Development of a physiologically based model for oseltamivir and simulation of pharmacokinetics in neonates and infants.

BACKGROUND: Physiologically based pharmacokinetic (PBPK) modelling can assist in the development of drug therapies and regimens suitable for challenging patient populations such as very young children. This study describes a strategy employing PBPK models to investigate the intravenous use of the neuraminidase inhibitor oseltamivir in infants and neonates with influenza. METHODS: Models of marmoset monkeys and humans were constructed for oseltamivir and its active metabolite oseltamivir carboxylate (OC). These models incorporated physicochemical properties and in vitro metabolism data into mechanistic representations of pharmacokinetic processes. Modelled processes included absorption, whole-body distribution, renal clearance, metabolic conversion of the pro-drug, permeability-limited hepatic disposition of OC and age dependencies for all of these processes. Models were refined after comparison of simulations in monkeys with plasma and liver concentrations measured in adult and newborn marmosets after intravenous and oral dosing. Then simulations with a human model were compared with clinical data taken from intravenous and oral studies in healthy adults and oral studies in infants and neonates. Finally, exposures after intravenous dosing in neonates were predicted. RESULTS: Good simulations in adult marmosets could be obtained after model optimizations for pro-drug conversion, hepatic disposition of OC and renal clearance. After adjustment for age dependencies, including reductions in liver enzyme expression and renal function, the model simulations matched the trend for increased exposures in newborn marmosets compared with those in adults. For adult humans, simulated and observed data after both intravenous and oral dosing showed good agreement and although the data are currently limited, simulations in 1-year-olds and neonates are in reasonable agreement with published results for oral doses. Simulated intravenous infusion plasma profiles in neonates deliver therapeutic concentrations of OC that closely mimic the oral profiles, with 3-fold higher exposures of oseltamivir than those observed with the same oral dose. CONCLUSIONS: This work exemplifies the utility of PBPK models in predicting pharmacokinetics in the very young. Simulations showed agreement with a wide range of observational data, indicating that the processes determining the age-dependent pharmacokinetics of oseltamivir are well described.