A new classifier-based strategy for in-silico ion-channel cardiac drug safety assessment

Assessing drug safety by identifying the axis of arrhythmia in cardiomyocyte electrophysiology

Many classes of drugs can induce fatal cardiac arrhythmias by disrupting the electrophysiology of cardiomyocytes. Safety guidelines thus require all new drugs to be assessed for pro-arrhythmic risk prior to conducting human trials. The standard safety protocols primarily focus on drug blockade of the delayed-rectifier potassium current (IKr). Yet the risk is better assessed using four key ion currents (IKr, ICaL, INaL, IKs). We simulated 100,000 phenotypically diverse cardiomyocytes to identify the underlying relationship between the blockade of those currents and the emergence of ectopic beats in the action potential. We call that relationship the axis of arrhythmia. It serves as a yardstick for quantifying the arrhythmogenic risk of any drug from its profile of multi-channel block alone. We tested it on 109 drugs and found that it predicted the clinical risk labels with an accuracy of 88.1-90.8%. Pharmacologists can use our method to assess the safety of novel drugs without resorting to animal testing or unwieldy computer simulations.

Use of high throughput ion channel profiling and statistical modeling to predict off-target arrhythmia risk - One pharma's experience and perspective

INTRODUCTION

The use of high throughput patch clamp profiling to determine mixed ion channel-mediated arrhythmia risk was assessed using profiling data generated using proprietary internal and clinical reference compounds. We define the reproducibility of the platform and highlight inherent platform issues. The data generated was used to develop predictive models for cardiac arrhythmia risk, specifically Torsades de Pointes (TdP).

METHODS

A retrospective analysis was performed using profiling data generated over a 3-year period, including patch clamp data from hERG, Ca_v1.2, and Na_v1.5 (peak/late), together with hERG binding.

RESULTS

Assay reproducibility was robust over the 3-year period examined. High throughput hERG patch IC₅₀ values correlated well with GLP-hERG data (Pearson = 0.87). A disconnect between hERG binding and patch was observed for ∼10% compounds and trended with passive cellular permeability. hERG and Ca_v1.2 potency did not correlate for proprietary compounds, with more potent hERG compounds showing selectivity versus Ca_v1.2. For clinical compounds where hERG and Ca_v1.2 activity was more balanced, an analysis of TdP risk versus hERG/Ca_v1.2 ratio demonstrated low TdP probability when the hERG/Ca_v1.2 potency ratios were < 1. Modeling of clinical compound data revealed a lack of impact of the Na_v1.5 (late) current in predicting TdP. Moreover, models using hERG binding data (ROC AUC = 0.876) showed an improved ability to predict TdP risk versus hERG patch clamp (ROC AUC = 0.787).

DISCUSSION

The data highlight the value of high throughput patch clamp data in the prediction of TdP risk, as well as some potential limitations with this approach.

A Quantitative Systems Pharmacology Perspective on the Importance of Parameter Identifiability

There is an inherent tension in Quantitative Systems Pharmacology (QSP) between the need to incorporate mathematical descriptions of complex physiology and drug targets with the necessity of developing robust, predictive and well-constrained models. In addition to this, there is no “gold standard” for model development and assessment in QSP. Moreover, there can be confusion over terminology such as model and parameter identifiability; complex and simple models; virtual populations; and other concepts, which leads to potential miscommunication and misapplication of methodologies within modeling communities, both the QSP community and related disciplines. This perspective article highlights the pros and cons of using simple (often identifiable) vs. complex (more physiologically detailed but often non-identifiable) models, as well as aspects of parameter identifiability, sensitivity and inference methodologies for model development and analysis. The paper distills the central themes of the issue of identifiability and optimal model size and discusses open challenges.

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.

In vitro data and computational models can assist with calculating pro-arrhythmic risk

We use patient health records and FDA Adverse Event Reporting System reports

Use of such datasets helps assess relative drug risk and cardiac safety models

We quantify how patient characteristics can affect arrhythmia incidence

A general procedure to select calibration drugs for lab-specific validation and calibration of proarrhythmia risk prediction models: An illustrative example using the CiPA model

INTRODUCTION

In response to the ongoing shift of the regulatory cardiac safety paradigm, a recent White Paper proposed general principles for developing and implementing proarrhythmia risk prediction models. These principles included development strategies to validate models, and implementation strategies to ensure a model developed by one lab can be used by other labs in a consistent manner in the presence of lab-to-lab experimental variability. While the development strategies were illustrated through the validation of the model under the Comprehensive In vitro Proarrhythmia Assay (CiPA), the implementation strategies have not been adopted yet.

METHODS

The proposed implementation strategies were applied to the CiPA model by performing a sensitivity analysis to identify a subset of calibration drugs that were most critical in determining the classification thresholds for proarrhythmia risk prediction.

RESULTS

The selected calibration drugs were able to recapitulate classification thresholds close to those calculated from the full list of CiPA drugs. Using an illustrative dataset it was shown that a new lab could use these calibration drugs to establish its own classification thresholds (lab-specific calibration), and verify that the model prediction accuracy in the new lab is comparable to that in the original lab where the model was developed (lab-specific validation).

DISCUSSION

This work used the CiPA model as an example to illustrate how to adopt the proposed model implementation strategies to select calibration drugs and perform lab-specific calibration and lab-specific validation. Generic in nature, these strategies could be generally applied to different proarrhythmia risk prediction models using various experimental systems under the new paradigm.

Classifying Drugs by their Arrhythmogenic Risk Using Machine Learning

All medications have adverse effects. Among the most serious of these are cardiac arrhythmias. Current paradigms for drug safety evaluation are costly, lengthy, conservative, and impede efficient drug development. Here, we combine multiscale experiment and simulation, high-performance computing, and machine learning to create a risk estimator to stratify new and existing drugs according to their proarrhythmic potential. We capitalize on recent developments in machine learning and integrate information across 10 orders of magnitude in space and time to provide a holistic picture of the effects of drugs, either individually or in combination with other drugs. We show, both experimentally and computationally, that drug-induced arrhythmias are dominated by the interplay between two currents with opposing effects: the rapid delayed rectifier potassium current and the L-type calcium current. Using Gaussian process classification, we create a classifier that stratifies drugs into safe and arrhythmic domains for any combinations of these two currents. We demonstrate that our classifier correctly identifies the risk categories of 22 common drugs exclusively on the basis of their concentrations at 50% current block. Our new risk assessment tool explains under which conditions blocking the L-type calcium current can delay or even entirely suppress arrhythmogenic events. Using machine learning in drug safety evaluation can provide a more accurate and comprehensive mechanistic assessment of the proarrhythmic potential of new drugs. Our study paves the way toward establishing science-based criteria to accelerate drug development, design safer drugs, and reduce heart rhythm disorders.

Classifying Drugs by their Arrhythmogenic Risk Using Machine Learning

All medications have adverse effects. Among the most serious of these are cardiac arrhythmias. Current paradigms for drug safety evaluation are costly, lengthy, conservative, and impede efficient drug development. Here, we combine multiscale experiment and simulation, high-performance computing, and machine learning to create a risk estimator to stratify new and existing drugs according to their proarrhythmic potential. We capitalize on recent developments in machine learning and integrate information across 10 orders of magnitude in space and time to provide a holistic picture of the effects of drugs, either individually or in combination with other drugs. We show, both experimentally and computationally, that drug-induced arrhythmias are dominated by the interplay between two currents with opposing effects: the rapid delayed rectifier potassium current and the L-type calcium current. Using Gaussian process classification, we create a classifier that stratifies drugs into safe and arrhythmic domains for any combinations of these two currents. We demonstrate that our classifier correctly identifies the risk categories of 22 common drugs exclusively on the basis of their concentrations at 50% current block. Our new risk assessment tool explains under which conditions blocking the L-type calcium current can delay or even entirely suppress arrhythmogenic events. Using machine learning in drug safety evaluation can provide a more accurate and comprehensive mechanistic assessment of the proarrhythmic potential of new drugs. Our study paves the way toward establishing science-based criteria to accelerate drug development, design safer drugs, and reduce heart rhythm disorders.

Blinded In Silico Drug Trial Reveals the Minimum Set of Ion Channels for Torsades de Pointes Risk Assessment

Torsades de Pointes (TdP) is a type of ventricular arrhythmia which could be observed as an unwanted drug-induced cardiac side effect, and it is associated with repolarization abnormalities in single cells. The pharmacological evaluations of TdP risk in previous years mainly focused on the hERG channel due to its vital role in the repolarization of cardiomyocytes. However, only considering drug effects on hERG led to false positive predictions since the drug action on other ion channels can also have crucial regulatory effects on repolarization. To address the limitation of only evaluating hERG, the Comprehensive in Vitro Proarrhythmia Assay initiative has proposed to systematically integrate drug effects on multiple ion channels into in silico drug trial to improve TdP risk assessment. It is not clear how many ion channels are sufficient for reliable TdP risk predictions, and whether differences in IC₅₀ and Hill coefficient values from independent sources can lead to divergent in silico prediction outcomes. The rationale of this work is to investigate the above two questions using a computationally efficient population of human ventricular cells optimized to favor repolarization abnormality. Our blinded results based on two independent data sources confirm that simulations with the optimized population of human ventricular cell models enable efficient in silico drug screening, and also provide direct observation and mechanistic analysis of repolarization abnormality. Our results show that 1) the minimum set of ion channels required for reliable TdP risk predictions are Nav1.5 (peak), Cav1.2, and hERG; 2) for drugs with multiple ion channel blockage effects, moderate IC₅₀ variations combined with variable Hill coefficients can affect the accuracy of in silico predictions.

Usefulness of Bnet, a Simple Linear Metric in Discerning Torsades De Pointes Risks in 28 CiPA Drugs

The Comprehensive in vitro Proarrhythmia Assay (CiPA) project suggested the torsade metric score (TMS) which requires substantial computing resources as a useful biomarker to predict proarrhythmic risk from human ether-à-go-go-related gene (hERG) and a few other ion channel block data. The TMS was useful to predict low TdP risks of drugs blocking Na+ (ranolazine) and Ca2+ (verapamil) channels as well as the hERG channel. However, Mistry asserted that the simple linear metric, Bnet reflecting net blockade of a few influential ion channels has similar predictive power. Here we compared the predictability of Bnet and TMS for the 12 training and 16 validation CiPA drugs which were pre-classified into three categories according to the known TdP risks (low, intermediate, and high risk) by CiPA. Bnet at 5×Cmax (Bnet5×Cmax) was calculated using the ion-channel IC50 and Hill coefficients of CiPA drugs collected from previous reports by the CiPA team and others. The receiver operating characteristic curve area under curve (ROC AUC) values for TMS and Bnet5×Cmax as performance metrics in discerning low versus intermediate/high risk categories for the 28 CiPA drugs were similar. However, Bnet5×Cmax was much inferior to TMS at discerning between intermediate- and high-risk drugs. Dynamic Bnet, which used in silico hERG dynamic parameters unlike conventional Bnet, improved the misspecification. Thus, we propose that Bnet5×Cmax is used for quick screening of TdP risks of drug candidates and if the "intermediate/high" risk is predicted by Bnet5×Cmax, in silico approaches, such as dynamic Bnet or TMS, may be further considered.

Global Sensitivity Analysis of Ventricular Myocyte Model-Derived Metrics for Proarrhythmic Risk Assessment

Multiscale computational models of the heart are being extensively investigated for improved assessment of drug-induced torsades de pointes (TdP) risk, a fatal side effect of many drugs. Model-derived metrics such as action potential duration and net charge carried by ionic currents (qNet) have been proposed as potential candidates for TdP risk stratification after being tested on small datasets. Unlike purely statistical approaches, model-derived metrics are thought to provide mechanism-based classification. In particular, qNet has been recently proposed as a surrogate metric for early afterdepolarizations (EADs), which are known to be cellular triggers of TdP. Analysis of critical model components and of the ion channels that have major impact on model-derived metrics can lead to improvements in the confidence of the prediction. In this paper, we analyze large populations of virtual drugs to systematically examine the influence of different ion channels on model-derived metrics that have been proposed for proarrhythmic risk assessment. We demonstrate via global sensitivity analysis (GSA) that model-derived metrics are most sensitive to different sets of input parameters. Similarly, important differences in sensitivity to specific channel blocks are highlighted when classifying drugs into different risk categories by either qNet or a metric directly based on simulated EADs. In particular, the higher sensitivity of qNet to the block of the late sodium channel might explain why its classification accuracy is better than that of the EAD-based metric, as shown for a small set of known drugs. Our results highlight the need for a better mechanistic interpretation of promising metrics like qNet based on a formal analysis of models. GSA should, therefore, constitute an essential component of the in silico workflow for proarrhythmic risk assessment, as an improved understanding of the structure of model-derived metrics could increase confidence in model-predicted risk.

In Silico Pharmacoepidemiologic Evaluation of Drug-Induced Cardiovascular Complications Using Combined Classifiers

Drug-induced cardiovascular complications are the most common adverse drug events and account for the withdrawal or severe restrictions on the use of multitudinous postmarketed drugs. In this study, we developed new in silico models for systematic identification of drug-induced cardiovascular complications in drug discovery and postmarketing surveillance. Specifically, we collected drug-induced cardiovascular complications covering the five most common types of cardiovascular outcomes (hypertension, heart block, arrhythmia, cardiac failure, and myocardial infarction) from four publicly available data resources: Comparative Toxicogenomics Database, SIDER, Offsides, and MetaADEDB. Using these databases, we developed a combined classifier framework through integration of five machine-learning algorithms: logistic regression, random forest, k-nearest neighbors, support vector machine, and neural network. The totality of models included 180 single classifiers with area under receiver operating characteristic curves (AUC) ranging from 0.647 to 0.809 on 5-fold cross-validations. To develop the combined classifiers, we then utilized a neural network algorithm to integrate the best four single classifiers for each cardiovascular outcome. The combined classifiers had higher performance with an AUC range from 0.784 to 0.842 compared to single classifiers. Furthermore, we validated our predicted cardiovascular complications for 63 anticancer agents using experimental data from clinical studies, human pluripotent stem cell-derived cardiomyocyte assays, and literature. The success rate of our combined classifiers reached 87%. In conclusion, this study presents powerful in silico tools for systematic risk assessment of drug-induced cardiovascular complications. This tool is relevant not only in early stages of drug discovery but also throughout the life of a drug including clinical trials and postmarketing surveillance.

In Silico QT and APD Prolongation Assay for Early Screening of Drug-Induced Proarrhythmic Risk

Drug-induced proarrhythmicity is a major concern for regulators and pharmaceutical companies. For novel drug candidates, the standard assessment involves the evaluation of the potassium hERG channels block and the in vivo prolongation of the QT interval. However, this method is known to be too restrictive and to stop the development of potentially valuable therapeutic drugs. The aim of this work is to create an in silico tool for early detection of drug-induced proarrhythmic risk. The system is based on simulations of how different compounds affect the action potential duration (APD) of isolated endocardial, midmyocardial, and epicardial cells as well as the QT prolongation in a virtual tissue. Multiple channel-drug interactions and state-of-the-art human ventricular action potential models ( O'Hara , T. , PLos Comput. Biol. 2011 , 7 , e1002061 ) were used in our simulations. Specifically, 206.766 cellular and 7072 tissue simulations were performed by blocking the slow and the fast components of the delayed rectifier current ( I_Ks and I_Kr, respectively) and the L-type calcium current ( I_CaL) at different levels. The performance of our system was validated by classifying the proarrhythmic risk of 84 compounds, 40 of which present torsadogenic properties. On the basis of these results, we propose the use of a new index (Tx) for discriminating torsadogenic compounds, defined as the ratio of the drug concentrations producing 10% prolongation of the cellular endocardial, midmyocardial, and epicardial APDs and the QT interval, over the maximum effective free therapeutic plasma concentration (EFTPC). Our results show that the Tx index outperforms standard methods for early identification of torsadogenic compounds. Indeed, for the analyzed compounds, the Tx tests accuracy was in the range of 87-88% compared with a 73% accuracy of the hERG IC₅₀ based test.

Complex versus simple models: ion-channel cardiac toxicity prediction

There is growing interest in applying detailed mathematical models of the heart for ion-channel related cardiac toxicity prediction. However, a debate as to whether such complex models are required exists. Here an assessment in the predictive performance between two established large-scale biophysical cardiac models and a simple linear model Bnet was conducted. Three ion-channel data-sets were extracted from literature. Each compound was designated a cardiac risk category using two different classification schemes based on information within CredibleMeds. The predictive performance of each model within each data-set for each classification scheme was assessed via a leave-one-out cross validation. Overall the Bnet model performed equally as well as the leading cardiac models in two of the data-sets and outperformed both cardiac models on the latest. These results highlight the importance of benchmarking complex versus simple models but also encourage the development of simple models.

Global Optimization of Ventricular Myocyte Model to Multi-Variable Objective Improves Predictions of Drug-Induced Torsades de Pointes

In silico cardiac myocyte models present powerful tools for drug safety testing and for predicting phenotypical consequences of ion channel mutations, but their accuracy is sometimes limited. For example, several models describing human ventricular electrophysiology perform poorly when simulating effects of long QT mutations. Model optimization represents one way of obtaining models with stronger predictive power. Using a recent human ventricular myocyte model, we demonstrate that model optimization to clinical long QT data, in conjunction with physiologically-based bounds on intracellular calcium and sodium concentrations, better constrains model parameters. To determine if the model optimized to congenital long QT data better predicts risk of drug-induced long QT arrhythmogenesis, in particular Torsades de Pointes risk, we tested the optimized model against a database of known arrhythmogenic and non-arrhythmogenic ion channel blockers. When doing so, the optimized model provided an improved risk assessment. In particular, we demonstrate an elimination of false-positive outcomes generated by the baseline model, in which simulations of non-torsadogenic drugs, in particular verapamil, predict action potential prolongation. Our results underscore the importance of currents beyond those directly impacted by a drug block in determining torsadogenic risk. Our study also highlights the need for rich data in cardiac myocyte model optimization and substantiates such optimization as a method to generate models with higher accuracy of predictions of drug-induced cardiotoxicity.

Novel Two-Step Classifier for Torsades de Pointes Risk Stratification from Direct Features

While pre-clinical Torsades de Pointes (TdP) risk classifiers had initially been based on drug-induced block of hERG potassium channels, it is now well established that improved risk prediction can be achieved by considering block of non-hERG ion channels. The current multi-channel TdP classifiers can be categorized into two classes. First, the classifiers that take as input the values of drug-induced block of ion channels (direct features). Second, the classifiers that are built on features extracted from output of the drug-induced multi-channel blockage simulations in the in-silico models (derived features). The classifiers built on derived features have thus far not consistently provided increased prediction accuracies, and hence casts doubt on the value of such approaches given the cost of including biophysical detail. Here, we propose a new two-step method for TdP risk classification, referred to as Multi-Channel Blockage at Early After Depolarization (MCB@EAD). In the first step, we classified the compound that produced insufficient hERG block as non-torsadogenic. In the second step, the role of non-hERG channels to modulate TdP risk are considered by constructing classifiers based on direct or derived features at critical hERG block concentrations that generates EADs in the computational cardiac cell models. MCB@EAD provides comparable or superior TdP risk classification of the drugs from the direct features in tests against published methods. TdP risk for the drugs highly correlated to the propensity to generate EADs in the model. However, the derived features of the biophysical models did not improve the predictive capability for TdP risk assessment.

Mechanistic Systems Modeling to Improve Understanding and Prediction of Cardiotoxicity Caused by Targeted Cancer Therapeutics

Tyrosine kinase inhibitors (TKIs) are highly potent cancer therapeutics that have been linked with serious cardiotoxicity, including left ventricular dysfunction, heart failure, and QT prolongation. TKI-induced cardiotoxicity is thought to result from interference with tyrosine kinase activity in cardiomyocytes, where these signaling pathways help to control critical processes such as survival signaling, energy homeostasis, and excitation–contraction coupling. However, mechanistic understanding is limited at present due to the complexities of tyrosine kinase signaling, and the wide range of targets inhibited by TKIs. Here, we review the use of TKIs in cancer and the cardiotoxicities that have been reported, discuss potential mechanisms underlying cardiotoxicity, and describe recent progress in achieving a more systematic understanding of cardiotoxicity via the use of mechanistic models. In particular, we argue that future advances are likely to be enabled by studies that combine large-scale experimental measurements with Quantitative Systems Pharmacology (QSP) models describing biological mechanisms and dynamics. As such approaches have proven extremely valuable for understanding and predicting other drug toxicities, it is likely that QSP modeling can be successfully applied to cardiotoxicity induced by TKIs. We conclude by discussing a potential strategy for integrating genome-wide expression measurements with models, illustrate initial advances in applying this approach to cardiotoxicity, and describe challenges that must be overcome to truly develop a mechanistic and systematic understanding of cardiotoxicity caused by TKIs.

Requirements for multi-level systems pharmacology models to reach end-usage: the case of type 2 diabetes

We are currently in the middle of a major shift in biomedical research: unprecedented and rapidly growing amounts of data may be obtained today, from in vitro, in vivo and clinical studies, at molecular, physiological and clinical levels. To make use of these large-scale, multi-level datasets, corresponding multi-level mathematical models are needed, i.e. models that simultaneously capture multiple layers of the biological, physiological and disease-level organization (also referred to as quantitative systems pharmacology-QSP-models). However, today's multi-level models are not yet embedded in end-usage applications, neither in drug research and development nor in the clinic. Given the expectations and claims made historically, this seemingly slow adoption may seem surprising. Therefore, we herein consider a specific example-type 2 diabetes-and critically review the current status and identify key remaining steps for these models to become mainstream in the future. This overview reveals how, today, we may use models to ask scientific questions concerning, e.g., the cellular origin of insulin resistance, and how this translates to the whole-body level and short-term meal responses. However, before these multi-level models can become truly useful, they need to be linked with the capabilities of other important existing models, in order to make them 'personalized' (e.g. specific to certain patient phenotypes) and capable of describing long-term disease progression. To be useful in drug development, it is also critical that the developed models and their underlying data and assumptions are easily accessible. For clinical end-usage, in addition, model links to decision-support systems combined with the engagement of other disciplines are needed to create user-friendly and cost-efficient software packages.

Recent developments in using mechanistic cardiac modelling for drug safety evaluation

•

Modelling and simulation can streamline decision making in drug safety testing.

•

Computational cardiac electrophysiology is a mature technology with a long heritage.

•

There are many challenges and opportunities in using in silico techniques in future.

•

We discuss how models can be used at different stages of drug discovery.

•

CiPA will combine screening platforms, human cell assays and in silico predictions.

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.