Computational Prediction of Drug-Induced Arrhythmias

Limitations of Current Methods for Predicting Drug-Induced Torsades De Pointes

Our understanding of the ionic currents responsible for ventricular depolarization and repolarization allows us to develop a straightforward model of drug-induced TdP. Because the rapid delayed rectifier K ⁺ current, I _Kr , is one of the major ionic currents responsible for membrane repolarization, it follows that blocking I _Kr will lead to action potential (AP) prolongation at the cellular level, prolongation of the QT interval at the organ level, and an increased risk of arrhythmia. The α-subunit of the I _Kr channel is encoded by the KCNH2 gene, also known as hERG by analogy with its Drosophila homologue ( hERG = human ether-a-go-go-related gene). Based on this paradigm, current guidelines for establishing the safety of new chemical entities rely on a combination of channel-level and organ-level screening. These guidelines, established by the International Conference of Harmonization, call for measurements of hERG block in a patch-clamp assay, followed by in vivo screening for potential QT prolongation in patients. The latter is often assessed in a “thorough QT” study, in which drug-induced changes in QT interval are measured in a large cohort of healthy volunteers.

Although fewer proarrhythmic drugs have been approved since these guidelines were first implemented in 2005, there are nonetheless many limitations of the current approach. First, although the present guidelines have been effective at identifying potentially TdP-inducing drug candidates, the stringent cutoffs applied in practice have likely resulted in the termination of promising drug candidates that would have been safe in practice. Second, I _Kr , although certainly important for repolarization, is only one current among many that collectively orchestrate ventricular APs, and it stands to reason that concomitant block of additional ion channels can influence TdP risk. Third, the present guidelines require only the in vitro hERG assay, a relatively simple patch-clamp measurement performed in an expression system, followed by studies in humans, which are extremely expensive. Complementary information could presumably be provided by additional preclinical assays such as experiments with animal preparations, human-induced pluripotent stem cell–derived cardiomyocytes (iPSC-CMs) and computational predictions. However, these approaches play no role under the present guidelines.

Fourth, the current paradigm relies on the correlation between QT prolongation and TdP risk, which is imperfect at best. Drugs that cause dramatic QT prolongation usually prove to be proarrhythmic, but drugs that cause moderate QT prolongation can be either proarrhythmic or safe, and it is not always clear what additional features, along with QT prolongation, may make a drug dangerous. Fifth, even for drugs that confer some TdP risk, we do not always know which patient populations may be most at risk from taking that drug. Although clinical experience shows that certain risk factors, including preexisting heart disease, female sex, and diabetes, are associated with a higher incidence of drug-induced TdP, it is not known whether these risk factors hold for all drugs.

This long list of limitations highlights the need for novel approaches to detect TdP risk early in drug development. Accordingly, the second decade of the 21st century has seen the emergence of a new paradigm, one in which simulations with mechanistic mathematical models play a central role. ^, The remainder of this chapter describes recent studies that have advanced these ideas, concluding with unresolved issues that can be solved in future years with creative computational approaches.

Proposed Use of Mathematical Models For Drug-Induced Arrhythmia Prediction

Because of the limitations of current standards for assessing TdP risk, a novel approach has been proposed in recent years. This has taken the form of the Comprehensive In Vitro Proarrhythmia Assay (CiPA), launched by the US Food and Drug Administration (FDA) in partnership with academic researchers and pharmaceutical companies, which proposes to improve testing of drugs for TdP risk. CiPA aims to provide novel methods to assess drugs for TdP risk in a cost-effective way, early in the drug development process, avoiding the substantial financial and human costs of last-stage failures. CiPA consists of three major initiatives to be conducted in preclinical, in vitro assays: (1) patch-clamp measurements of ion channel block in cells expressing particular cardiac ion channels; (2) simulations with mechanistic models to predict, over a range of concentrations, the effects of drugs that may block multiple ion channels; and (3) physiologic measurements of drug-induced changes in iPSC-CMs to assess correspondence with modeling predictions. Mathematical modeling is therefore a central component of CiPA, and the remainder of this chapter will discuss studies that have advanced these goals. As CiPA was formally introduced in 2014, the studies discussed all date from 2016 or more recently, indicating the dynamic and evolving nature of this field of research.

The use of mathematical models in CiPA has been made possible by the fact that cardiac electrophysiology modeling is a mature area of research, with dozens of models having been developed over the last several decades. These models capture the interplay between various ionic currents that collectively produce the cardiac AP, and contemporary models also include representations of calcium release from the sarcoplasmic reticulum. Over many years of research, these models have refined their representations of individual ionic currents, and variants have been developed to represent both atrial and ventricular myocytes from a variety of species. New methods have also been developed to calibrate and tune parameter values in these models based on quantitative comparisons with experimental data. ^, These basic science advances, over decades of work, make it relatively straightforward to simulate the effects of drugs on APs and intracellular [Ca ²⁺ ]. Patch-clamp and cellular physiology experiments can often isolate how a particular drug may influence a molecularly defined ionic current, summarizing the effects through a half-maximal inhibitory concentration (IC ₅₀ value) and Hill coefficient. With this knowledge in hand, concentration-dependent effects on cellular physiology can readily be simulated.

These considerations all suggest that mathematical modeling can be productively applied for the study of drug-induced arrhythmias. Importantly, groups pursuing this research have not had to develop new mathematical models de novo but could instead leverage existing tools. In addition to these points, there are several advantages to simulating drug effects on cardiac electrophysiology using mechanistic mathematical models:

• Simulation studies can generate predictions quickly and in a cost-effective manner.

• When in vitro assays produce inconclusive or seemingly inconsistent results, simulations can suggest future experiments to resolve conflicts and may uncover mechanisms underlying the discrepancies.

• Drug combinations can be systematically tested through simulations in a way that is extremely difficult to do experimentally, and nearly impossible clinically, because of combinatorial complexity.

• Studies of heterogeneous populations allow us to understand and predict how the effects of drugs may be different between individuals with similar characteristics, or between subpopulations of patients, such as males versus females, or patients with versus without existing cardiac disease.

These advantages of mathematical modeling were a major reason why CiPA was designed with this approach proposed as a fundamental component. The remaining sections describe advances over the last few years that have driven the field forward.

Recent Studies Have Demonstrated the Value Of Computational Modeling

The idea of using mechanistic mathematical models for simulations of cardiac drug effects has been suggested for some time, with important studies on both the efficacy of antiarrhythmic drugs and potential side effects published in 2011. The idea of large-scale screening of drug effects using models, however, gained considerable momentum during the period 2016–19 with the publication of several notable papers that demonstrated important principles and pointed the direction toward future advances in this area. Together these studies emphasize the important role that mechanistic modeling can play, not only for synthesizing experimental results, but also for identifying gaps in our knowledge and suggesting new experiments to address these deficiencies. Another important principle emphasized by these studies is the potential for acceleration of research advances through collaborative interactions between academic laboratories, pharmaceutical industry scientists, and regulatory agencies such as the FDA.