The Kinetics and State-Dependence of Drug-Binding to Kv11.1

Mark Jonathan Perrin

A thesis submitted for the degree of Doctor of Philosophy in the Faculty of Medicine

University of New South Wales

2009 Abstract

KCNH2 encodes the pore forming α-subunit of the rapid delayed rectifier K+ channel (Kv11.1). Kv11.1 is crucial for timely repolarisation of the cardiac action potential. Loss of channel function, either congenital or acquired, is a cause of the long QT syndrome (LQTS) and is proarrhythmic. Mutations in 12 different genes may cause congenital LQTS. Conversely, drug-block of Kv11.1 is almost the sole cause of drug-acquired LQTS and is the single most common reason for drug withdrawal from market. As such, drug safety is a significant concern of the pharmaceutical industry and regulatory bodies; an early assessment of a drug’s affinity for Kv11.1 is mandated by the FDA. However, the characteristics of Kv11.1 drug-binding are incompletely described. Specif- ically, the effect of potassium concentration on drug-block, the presence or absence of state-dependence, and the mechanism of frequency-dependence, remain controversial. Current accurate testing methods for drug-Kv11.1 affinity are laborious; high-throughput methods are inaccurate. Here we pursue experimental and mathematical modelling methods to describe the kinetics and state-dependence of drug-binding to Kv11.1. We gain insight into the effect of potassium on drug-block of Kv11.1, demonstrate that drug-binding to Kv11.1 is state- dependent with a marked preference for the inactivated state, and show that frequency- dependence drug-block is likely to result from drug-unbinding prior to channel deactivation. These insights are employed to investigate the mechanism of preclinical markers of drug-acquired long QT syndrome. Specifically, the TRIaD (Triangulation, Reverse use- dependence, Instability, and Dispersion) markers of proarrhythmic risk are found to be a product of drug-specific binding kinetics and state-dependent Kv11.1-block. Our findings suggest a new avenue for safer drug development. Further, we demonstrate that the inaccuracy of high-throughput screening methods results from drug-specific state- and frequency-dependent binding properties. An improved method for high-throughput drug-screening is proposed. The findings in this thesis provide insight into the fundamental properties of drug-block of Kv11.1 and have both clinical – by improving prediction of proarrhythmic propensity, and pharmaceutical – by improving high-throughput screening, application. For my dear father and mother, Ken and Mavis

1 Contents

1 Introduction 11 1.1 Preface ...... 11 1.2 Foundation of Cardiac Electrical Activity ...... 13 1.2.1 Ion channels ...... 13 1.2.2 Cardiac action potential ...... 17 1.3 Congenital Disorders of Cardiac Action Potential Repolarisation ...... 23 1.3.1 Congenital long QT syndrome ...... 23 1.3.2 Congenital short QT syndrome ...... 27 1.4 Acquired Disorders of Cardiac Action Potential Repolarisation ...... 29 1.4.1 Acquired long QT syndrome ...... 29 1.4.2 Kv11.1 (hERG) channel ...... 31 1.4.3 Drug-binding to Kv11.1 ...... 34 1.4.4 Safety pharmacology ...... 35 1.5 Aims ...... 40

2 General Methods 41 2.1 Molecular Biology ...... 41 2.2 Electrophysiology ...... 41 2.2.1 Drug-block protocol for high and low potassium, and mutant- and WT-Kv11.1 ...... 42 2.3 Statistics ...... 43

3 Potassium and Inactivation of Kv11.1 44 3.1 Background ...... 44 3.2 The Measurement of Inactivation in Kv11.1 ...... 47 3.2.1 Introduction ...... 47 3.2.2 Experimental protocols ...... 48

2 3.2.3 Results ...... 50 3.2.4 Discussion ...... 50 3.3 External K+ Concentration and Inactivation of Kv11.1 ...... 54 3.3.1 Introduction ...... 54 3.3.2 Results ...... 54 3.3.3 Discussion ...... 55 3.4 External K+ Concentration and Drug Block of Kv11.1 ...... 57 3.4.1 Introduction ...... 57 3.4.2 Results ...... 57 3.4.3 Discussion ...... 57

4 State-Dependent Drug-Binding to Kv11.1 60 4.1 Background ...... 60 4.2 Results ...... 61

4.2.1 V0.5 of steady-state inactivation in WT-Kv11.1, N588E-Kv11.1 and N588K-Kv11.1 expressed in CHO cells...... 61 4.2.2 High affinity drug-binding is modulated by N588 charge mutants . . 62 4.2.3 Low affinity drug-binding to N588 charge mutants ...... 65 4.2.4 Does reduced affinity for N588K-Kv11.1 reflect state-dependent binding? ...... 65 4.2.5 Modelling kinetics of drug-binding to open and inactivated states . . 69 4.3 Discussion ...... 74 4.3.1 State-dependence of drug-binding to Kv11.1 ...... 74 4.3.2 What explains the strong preference for inactivated state binding? . 75 4.3.3 Inactivated state binding is necessary but not sufficient for high affinity binding ...... 76 4.3.4 Relevance for drug-binding in SQTS ...... 76 4.3.5 Relevance for high-throughput assays ...... 77

5 Drug-Kv11.1 Kinetics and Risk of Acquired Long QT Syndrome 78 5.1 Background ...... 78 5.2 Methods ...... 79 5.2.1 Markov state models of cisapride and dofetilide ...... 79 5.2.2 Voltage protocols employed in modelling experiments ...... 82 5.2.3 Ventricular cell modelling ...... 84 5.2.4 Cable simulations ...... 87

3 5.3 Results ...... 88 5.3.1 Markov-state models of Kv11.1 drug-block reproduce voltage-clamp experimental data ...... 88

5.3.2 IC50 from voltage-clamp experiments and action potential duration experiments ...... 90 5.3.3 Cisapride and dofetilide cause triangulation of the cardiac action potential ...... 91 5.3.4 Dofetilide is reverse use-dependent; cisapride is use-dependent . . 92 5.3.5 Cisapride and dofetilide cause instability of the action potential duration ...... 93 5.3.6 Cisapride and dofetilide increase dispersion of repolarization .... 93 5.3.7 Cisapride and dofetilide induce early afterdepolarisations in M cells 95 5.3.8 Cisapride and dofetilide delay repolarization in cable models of the ventricular wall ...... 95 5.4 Discussion ...... 99 5.4.1 Mechanistic insight from Markov models ...... 99 5.4.2 Markov models reproduce TRIaD markers of proarrhythmic risk . . . 99 5.4.3 Predicting arrhythmogenicity ...... 101 5.5 Conclusion ...... 101

6 Improving High-Throughput Screening Assays 102 6.1 Background ...... 102 6.2 Mathematical Model of a High-Throughput Assay ...... 104 6.2.1 Formation of the mathematical model ...... 104 6.3 Current-Clamp Experiments ...... 106 6.3.1 Results ...... 106 6.3.2 Discussion ...... 107 6.4 Modelling an Improved High-Throughput Assay ...... 110

7 Conclusion 112

4 List of Figures

1.1 Calculated action potential waveform of Hodgkin and Huxley ...... 12 1.2 Ribbon representation of KcsA ...... 14 1.3 Schematic diagram of a voltage-gated potassium channel ...... 16 1.4 Cardiac action potential of the left ventricle ...... 18 1.5 Cardiac action potential shapes in different regions of the human heart . . . 20 1.6 ECG recordings from a patient with congenital long QT syndrome...... 25 1.7 The kinetics of Kv11.1 gating ...... 33 1.8 The drug-binding pocket of Kv11.1 ...... 34 1.9 Theory of ﬂuorescence-based ion channel assays ...... 38

2.1 Voltage protocol used to record drug-block of WT-Kv11.1 and mutant-Kv11.1 channels ...... 42

3.1 Schematic diagram of the open and inactivated states of Kv11.1 ...... 45 3.2 Two-pulse and three-pulse voltage protocols to investigate the voltage- dependence of steady-state inactivation of Kv11.1 ...... 47 3.3 Raw data traces from two-pulse and three-pulse voltage protocols explor- ing the voltage-dependence of steady-state inactivation ...... 49 3.4 Deactivation-corrected three-pulse current recording ...... 50 3.5 Voltage-dependence of inactivation and recovery from inactivation time constants of Kv11.1 ...... 51 3.6 Boltzmann ﬁts to steady-state inactivation data from the two-pulse and three-pulse voltage protocols ...... 53 3.7 Boltzmann curves of voltage-dependence of steady-state inactivation in high and low K+ concentrations ...... 54 3.8 Raw current traces of two-pulse voltage protocols in 5 and 100 mmol/L K+ concentrations ...... 55 3.9 Hill plots for drug-block of Kv11.1 in normal and high K+ concentrations . . 58

5 4.1 Schematic diagram of the S5, S6, pore helix, and S5P linker regions of two Kv11.1 subunits ...... 61 4.2 Conductance voltage curves for WT-Kv11.1, N588E-Kv11.1, and N588K- Kv11.1 ...... 62 4.3 Typical current traces recorded from WT-Kv11.1, N588E-Kv11.1, and N588K- Kv11.1 in the presence of 30 nM cisapride ...... 63 4.4 Hill plots of high affinity Kv11.1 blockers ...... 64 4.5 Typical current traces recorded from WT-Kv11.1, N588E-Kv11.1, and N588K- Kv11.1 in the presence of 3μM quinidine...... 66 4.6 Hill plots of low affinity Kv11.1 blockers ...... 68 4.7 Conductance voltage curves for S631A-Kv11.1 and S620T-Kv11.1 and the affinity of dofetilide for each...... 69 4.8 Markov model of drug-binding to Kv11.1 ...... 70 4.9 Dofetilide Hill plots with data points for N588K-Kv11.1 ...... 71 4.10 Correspondence between experimental and modelled Hill plots for state- dependent drugs ...... 72 4.11 Hill plots for drug-binding to S620T-Kv11.1 ...... 73

5.1 Model ﬁtting to experimental data ...... 80 5.2 Markov state models for cisapride and dofetilide block of Kv11.1 ...... 81 5.3 Voltage protocol used to record recovery from inactivation of L-type calcium channels ...... 83 5.4 Voltage protocols for use-, frequency-, and voltage-dependence ...... 83

5.5 Comparison of time constant of recovery from inactivation for ICa−L in TTP06 and mTTP06 ...... 85

5.6 Accumulation of IKs current in a modiﬁed ten Tusscher-Panﬁlov model . . . 86 5.7 Markov-state models of cisapride and dofetilide binding to Kv11.1 reproduce voltage patch-clamp experimental data ...... 89 5.8 Triangulation of cardiac action potential due to cisapride and dofetilide . . . 90

5.9 Increase in APD90 with reduction in Kv11.1 conductance ...... 91

5.10 APD90 prolongation dose-response curves for dofetilide and cisapride . . . 92 5.11 The effect of pacing cycle length on magnitude of action potential prolongation ...... 93 5.12 Poincare´ plots of action potential duration instability in simulated M cells . . 94 5.13 Dispersion in repolarisation predicted from mathematical modelling ..... 96 5.14 Early afterdepolarisations in M cells ...... 97 5.15 Simulations of transmural conduction across the ventricular wall ...... 98

6 6.1 Raw tracing of a current-clamp experiment ...... 103 6.2 Voltage change correlated to conductance of Kv11.1 ...... 106 6.3 Current-clamp, and voltage-clamp, Hill plots for quinidine, haloperidol, dofetilide and cisapride...... 107 6.4 Modelled current-clamp simulations mimicking the readout from ﬂuorescence- based high-throughput screening assays ...... 108 6.5 Incomplete recovery of Kv11.1 from haloperidol block at negative membrane voltages...... 109 6.6 Kv11.1 mutants that left-shift activation and inactivation increase the range and sensitivity of voltage responses to drugs that block Kv11.1 in CHO cells.111

7 List of Tables

1.1 IUPHAR numbering system for mammalian voltage-gated potassium channels ...... 15 1.2 Reversal voltages of sodium, potassium, and calcium ...... 18 1.3 Genes and proteins that cause the congenital form of the long QT syndrome 24

4.1 IC50 values from Hill plots of low and high afﬁnity Kv11.1-blockers ...... 67 4.2 Rate constants for drug-binding to the open and inactivated states of Kv11.1 70

4.3 Drug afﬁnities for S620T (IC50) ...... 71

6.1 Drugs withdrawn from the US market due to QT prolongation and risk of torsades de pointes ...... 102

6.2 IC50 values in nM for voltage-clamp and current-clamp experiments .....109

8 Acknowledgements

First and foremost I offer my sincerest gratitude to my supervisors Associate Professor Jamie Vandenberg and Professor Terence Campbell. These three years have been the best of my life. Your support, guidance, and considerable patience, have provided me with a wonderful introduction to the world of science and the mysteries of this peculiar ‘dancing’ channel. My thanks to: All the members of the Mark Cowley Lidwill Laboratory. My wonderful wife, Sarah and my daughter Grace. We have travelled this road together. What I would have done without you I do not know. You have been the source of much joy and happiness that has sustained me in the difﬁcult periods of this work. My father and mother, Ken and Mavis. Two better parents I can’t imagine. All my life you have encouraged and supported me, loved me at my lowest, and celebrated with me at my best. I thank God for you every day.

9 Publications and Patents Arising From This Thesis

Peer-reviewed publications

Perrin MJ, Kuchel PW, Campbell TJ and Vandenberg JI (2008). Drug-binding to the inactivated state is necessary but not sufﬁcient for high-afﬁnity binding to human ether-a-` go-go-related gene channels. Mol Pharmacol 74(5):1443–1452. (Chapter 4)

Perrin MJ, Subbiah RN, Vandenberg JI and Hill AP (2008). Human ether-a-go-go related gene (hERG) K+ channels: function and dysfunction. Prog Biophys Mol Biol 98(2- 3):137–48. (Chapters 1 and 4)

Patents hERG mutants and uses thereof Australian Patent Ofﬁce – 20 May 2008 International PCT application – 19 May 2009

10 Chapter 1

Introduction

1.1 Preface

“Scientiﬁc work proceeds at many levels of complexity. Scientists assume that all observable phenomena can ultimately be accounted for by a small number of unifying physical laws. Science, then, is the attempt to ﬁnd ever more fundamental laws and to reconstruct the long chains of causes from these foundations up to the full range of natural events.”1

An interesting figure is found in a seminal Hodgkin and Huxley electrophysiology paper of 1952 (Hodgkin and Huxley, 1952) — a calculated membrane action potential hanging in mid-air (Figure 1.1). It was not completed due, principally, to the ‘labour involved’2. When it is known that the manual calculation of a single action potential took many months, their explanation appears more than reasonable. In previous experiments they had elegantly demonstrated that current could pass through a cell membrane by charging the membrane capacity, or by ion passage through ‘the resistances parallel with the capacity’. Ionic conduction was further decomposed to a potassium, sodium, and leak current. Having established the ‘physical laws’ of membrane electrical activity, they were described mathematically to test their validity. Could they account for observed conduction and excitation in excitable cells? In Figure 1.1 the answer is clear. 1Hille B, Ion Channels of Excitable Membranes 3rd ed. p25 2In Hodgkin and Huxley’s words: “Three calculated membrane action potentials, with different strengths of stimulus, are shown in the upper part of the figure. Only one, in which the initial displacement of membrane potential was 15 mV, is complete; in the other two the calculation was not carried beyond the middle of the falling phase because of the labour involved and because the solution had become almost identical with the 15 mV action potential, apart from the displacement in time. One solution for a stimulus just below threshold is also shown (Hodgkin and Huxley, 1952)”

11 Figure 1.1: Hodgkin and Huxley’s original incomplete calculated membrane action potentials. Upper panel: numerical solutions for initial depolarisations of 90, 15, 7 and 6 mV are shown. Lower panel: tracings of membrane action potentials recorded from a giant squid axon (Hodgkin and Huxley, 1952).

The process of seeking to understand experimental observations as the cooperative interaction of a number of fundamental laws, to describe those laws, re-form them mathematically, and validate them by recapitulating independent experimental data, remains with us today and is the general form of the work I present. Happily, the months of manual calculation have been reduced to the work of a few seconds with the advent of the silicon chip. However, the need for careful experimentation and mathematical formulation remains.

12 1.2 Foundation of Cardiac Electrical Activity

1.2.1 Ion channels

1.2.1.1 Background

Erwin Neher and Bert Sakmann, inventors of the voltage patch-clamp technique, received the Nobel prize in Physiology or Medicine in 1991 for having, in the words of the commit- tee, “conclusively established the existence of ion channels”. This marked the ‘end of the beginning’ for ion channel investigation. In 1912, Julius Bernstein postulated a ‘membrane hypothesis’ to explain cell electrical activity (Bernstein, 1912). He proposed that the membrane was selectively permeable to K+ at rest, and during excitation ‘broke down’ to allow the free movement of ions down their electrochemical gradients. Subsequently, Alan Hodgkin and Andrew Huxley recognised the existence of ion- specific and voltage-dependent currents that functioned at rest and during the period of cell excitation. They formulated a kinetic model of ion currents that adequately explained recorded membrane action potentials — now eponymously, the Hodgkin Huxley (HH) model (Hodgkin and Huxley, 1952). In humility, they accepted that this was a model,able to reproduce currents qualitatively, but perhaps not reflective of the underlying molecular mechanisms. Yet it is remarkable that, 50 years later, their model is still widely employed in ion channel simulations. Perhaps more remarkable: their suggestion that underlying the measured potassium current were four independent charged particles existing in conductive and non-conductive states — analogous to the four subunits of the first structurally determined ion channel, KcsA (Doyle et al., 1998). In 1971, Bertil Hille, through the measurement of differential mobility of multiple ions across excitable membranes suggested that the pore “filter” controlled ion flow and was around 3–5 A˚ in size (Hille, 1971). Therefore, prior to direct confirmation of the existence of the ion channel by voltage patch-clamping, a theory of cell excitability prevailed according to which ions, in the presence of a suitably sized and selective conductor, flowed down their electrochemical gradient to change membrane voltage.

1.2.1.2 Naming conventions

Before detailed knowledge of the human genome, classiﬁcation of ion channels was performed on the basis of their unique electrophysiological characteristics. With the advent of molecular genetics it became clear that far more channel subtypes were coded by

Figure 1.2: Ribbon representation of KcsA. (A) – the extracellular side, revealing the central ion- conducting pore. (B) – a perpendicular aspect, demonstrating the ‘inverted teepee structure’ of the tetramer (Doyle et al., 1998). the genome than were otherwise distinguishable by electrophysiological characterisation. Further, a single gene could encode multiple different forms (isoforms) of a channel through alterations in transcription (Lees-Miller et al., 1997). The naming of individual genes follows their line of discovery. KCNH23 (commonly known as hERG – human ether-a-go-go` related gene) is the proper name for the gene that codes the channel which conducts the rapid delayed rectiﬁer potassium current in the heart. Guidelines for the appropriate naming of genes are published (Wain et al., 2002). The International Union of Pharmacology (IUPHAR) now provides a consensus document based upon structural and evolutionary characteristics that guides the naming of new receptors (Harmar et al., 2009). An example of this nomenclature is presented for the voltage-gated potassium channels in Table 1.1. In this thesis I will refer to the gene encoding the α-subunit of the rapid delayed rectiﬁer current as KCNH2, and its protein product as Kv11.1.

1.2.1.3 Properties

When the ﬁrst structure of an ion channel was determined – KscA (Figure 1.2) – (Doyle et al., 1998), it did not render obsolete the difﬁcult biophysical work of 50 years. Instead, the structure was examined, and understood, in its complementary light. The general topology of a voltage-gated potassium channel is displayed in Figure 1.3.

3KCNH2: K – potassium, CN – channel, H – subfamily H (EAG like), member – 2 (Wain et al., 2002)

14 Table 1.1: IUPHAR Numbering System for Mammalian Voltage-Gated Potassium Channels Family Ion Ofﬁcial IUPHAR name Human gene name HGNC ID Human genetic localisation Protein VGPC K+ Kv1.1 KCNA1 6218 12p13.32 VGPC K+ Kv1.2 KCNA2 6220 1p13 VGPC K+ Kv1.3 KCNA3 6221 1p13.3 VGPC K+ Kv1.4 KCNA4 6222 11p14.3-15.2 + VGPC K Kv1.5 KCNA5 6224 12p13.3 α subunit of IKur VGPC K+ Kv1.6 KCNA6 6225 12p13.3 VGPC K+ Kv1.7 KCNA7 6226 19q13.3 VGPC K+ Kv1.8 KCNA10 6219 1p13.1 VGPC K+ Kv2.1 KCNB1 6231 20q13.2 VGPC K+ Kv2.2 KCNB2 6232 8q13.2 VGPC K+ Kv3.1 KCNC1 6233 11p15 VGPC K+ Kv3.2 KCNC2 6234 12q14.1 VGPC K+ Kv3.3 KCNC3 6235 19q13.3-13.4 VGPC K+ Kv3.4 KCNC4 6236 1p21 VGPC K+ Kv4.1 KCND1 6237 Xp11.23 VGPC K+ Kv4.2 KCND2 6238 7q31 + VGPC K Kv4.3 KCND3 6239 1p13.3 α subunit of Ito,f VGPC K+ Kv5.1 KCNF1 6246 2p25 15 VGPC K+ Kv6.1 KCNG1 6248 20q13 VGPC K+ Kv6.2 KCNG2 6249 18q22-q23 VGPC K+ Kv6.3 KCNG3 18306 2p21 VGPC K+ Kv6.4 KCNG4 19697 16q24.1 + VGPC K Kv7.1 KCNQ1 6294 11p15.5 α subunit of IKs VGPC K+ Kv7.2 KCNQ2 6296 20q13.3 VGPC K+ Kv7.3 KCNQ3 6297 8q24 VGPC K+ Kv7.4 KCNQ4 6298 1p34 VGPC K+ Kv7.5 KCNQ5 6299 6q14 VGPC K+ Kv8.1 KCNV1 18861 8q22.3-24.1 VGPC K+ Kv8.2 KCNV2 19698 9p24.2 VGPC K+ Kv9.1 KCNS1 6300 20q12 VGPC K+ Kv9.2 KCNS2 6301 8p22 VGPC K+ Kv9.3 KCNS3 6302 2p24 VGPC K+ Kv10.1 KCNH1 6250 1q32-q41 VGPC K+ Kv10.2 KCNH5 6254 14q23.1 + VGPC K Kv11.1 KCNH2 6251 7q35-q36 α subunit of IKr VGPC K+ Kv11.2 KCNH6 18862 17q23.3 VGPC K+ Kv11.3 KCNH7 18863 2q24.2 VGPC K+ Kv12.1 KCNH8 18864 3p24.3 VGPC K+ Kv12.2 KCNH3 6252 12q13 VGPC K+ Kv12.3 KCNH4 6253 17q21.2 Figure 1.3: General topology of a voltage-gated potassium channel demonstrating the most important functional domains. (A) – viewed perpendicular to the membrane surface, and (B)–en face through the ion conducting pore

Permeability Ion channels are transmembrane macromolecules that provide an aque- ous conduit through a lipid bilayer. The high unitary conductance of single ion channels provided the strongest evidence for the ‘pore hypothesis’ of ion channel behaviour before crystal structures were available. Ion channel conductance may reach 300-400 pS4 –far in excess of enzyme- or carrier-like mechanisms of ion transport. This corresponds to the movement of millions of ions each second (Hille, 2001), and is the motor of rapid voltage change across the membrane.

Selectivity Biophysical characterisation suggests that ion channels act in part as molecular sieves for ions of different size. Each channel may conduct ions up to a certain radius. In voltage-gated potassium channels, Rb+ (1.5 A˚ radius) is the largest ion to pass. How- ever, ions of smaller size do not necessarily conduct. Voltage-gated potassium channels are highly selective for the larger K+ (1.3 A˚ ) over the smaller Na+ (1 A˚ ) (1000:8 in Kv11.1 (Zhou et al., 1998)). The crystal structure of KcsA provides significant insight: K+ within the selectivity filter are dehydrated, an energetically unfavourable process which is compensated by carbonyl oxygen atoms that face into the pore. The large sized filter is able to compensate for K+; but the smaller Na+ is relatively distant from the carbonyl oxygens, and therefore inadequately compensated. In the solved structure, a maximum of2K+ occupy the selectivity filter at any one time, and at distant ends. It is likely that

4Values for voltage-gated potassium channels (Kv) are typically in the range of 2 - 50 pS.

16 electrostatic repulsion, with the entrance of one ion promoting the exit of another, allows for rapid ion transit through the ﬁlter region (Doyle et al., 1998).

Gating It is not enough that ion channels conduct specific ions, there must also exist a mechanism to alternately quiet and excite the flow of charge. In voltage-gated ion channels, transitions between the open and closed stated are regulated by the first four transmembrane helices — the voltage sensor domain (VSD)(Figure 1.3). Structural rear- rangement of the VSD in response to a changing electric field is ‘transferred’ to the pore domain via a physical interaction between the S4-S5 linkers and the cytoplasmic termi- nals of the S6 helices (Long et al., 2005). Inactivation, when it occurs, may result from two mechanisms: N-type inactivation (ball and chain) or C-type inactivation (Baukrowitz and Yellen, 1995). The latter is operative in Kv11.1 (see Subsection 1.4.2).

1.2.2 Cardiac action potential

1.2.2.1 Phases

The cardiac action potential results from the complex interplay of depolarising inward, and repolarising outward currents. The principal ions of depolarisation are Na+ and Ca2+; the principal ion of repolarisation is K+.

Phase 0 The rapid upstroke (depolarisation) of the cardiac action potential is termed dv Phase 0. Vmax, the maximum dt of the upstroke reaches 400 V/s. Such rapidity is effected by the diffusion of > 107 Na+/s into the cell through Nav1.5 – the cardiac sodium channel encoded by SCN5A on chromosome 3. Nav1.5 is activated by the depolarisation of nearby cells.

Phase 1 Initial repolarisation of the cardiac action potential is termed Phase I. The transient outward current (Ito), conducted by Kv4.3 (fast component – Ito,f ) and Kv1.4 5 (slow component – Ito,s), is primarily responsible . Ito demonstrates very rapid activation (τ ∼2-10 ms) at the positive voltages reached in Phase 0, followed by a slower but still rapid inactivation (τ ∼25-80 ms).

Phase 2 Two forms of calcium channel predominate in cardiac myocytes – L- and T- type. L-type is mostly Cav1.2, and present in all cardiac myocytes. T-type: Cav3.1

5Kv4.3 and Kv1.4 in the ventricles; Kv4.3 alone in the atria

Figure 1.4: The left ventricular cardiac action potential and the principal ionic currents that form it. The APD90 (red dotted line), is the time point at 90% of maximum repolarisation, and is the most commonly quoted measure of action potential duration.

Ion Extracellular Intracellular Reversal concentration (mM) concentration (mM) voltage (mV) Na+ 145 12 +67 K+ 4 155 -98 Ca2+ 1.5 100 nM +129

Table 1.2: Reversal voltages of sodium, potassium, and calcium (Hille, 2001).

and Cav3.2, is expressed in atrial, pacemaking, and Purkinje cells. During Phase 2 – the ‘plateau’ phase – cell membrane voltage stabilises in the range ∼0-20 mV. Calcium channels activate rapidly at positive membrane voltages; however, as peak action po- 2+ tential voltage (∼+40 mV) is close to the reversal voltage of Ca , initial ICa−L is small. Repolarisation due to Ito increases the driving force for calcium conduction which quickly reaches a peak, commencing the process of excitation-contraction coupling, before de- clining to a new steady-state.

A succession of repolarising currents balance calcium inﬂux. Ito is followed by the delayed rectiﬁer potassium currents – IKr and IKs – conducted by Kv11.1 and Kv7.1/KCNE1 respectively. The gradual increase in IKr and IKs eventually overwhelms the inactivating calcium current, and the cell begins to repolarise.

Phase 3 As the plateau ends and membrane voltage becomes negative, Kv11.1 recovers from inactivation and hastens the rate of repolarisation. Spermine and Mg2+ block

18 of IK1 (Kir2.1 and Kir2.4) – responsible for its strong inward rectiﬁcation properties – is relieved by membrane repolarisation and results in a surge of repolarising current that returns the membrane voltage to its resting value.

Phase 4 Nav1.5 (INa), Kv4.3 (Ito), Cav1.2 (ICa), Kv11.1 (IKr), and Kv7.1 (IKs) are de- activated at negative membrane voltages. The resting voltage, then, is determined by a series of leak currents and Kir2.x. Interestingly, though it lacks a true VSD, Kir2.x is the predominant regulator of resting membrane voltage. Voltage ‘regulation’ of Kir2.x results from Mg2+ and linear polyamine (61) block at positive voltages. At negative membrane voltages the obstruction is relieved, allowing significant current flow. IK1 acts as a ‘buffer’ to membrane voltage change – currents tending to depolarise the membrane increase the driving force for outward K+ flow thus opposing voltage change; repolarising currents are opposed by a movement toward the equilibrium voltage of K+ which reduces the electrical driving force and, thus, outward K+ flow.

1.2.2.2 Action potential morphologies in different regions of the heart

Action potential morphologies recorded from pacemaker tissue, atria, and ventricles, differ as a result of differential ion channel expression (Figure 1.5)(Schram et al., 2002).

Sinus node The specialised pacemaker cells of the sinus node have a maximum dias- tolic potential of -50 mV, and a slow depolarising upstroke (< 2 V/s). An absence of IK1 allows the small inward current of the non-selective cation channel (If ) to slowly depolarise the cell during diastole. L-type calcium channels conduct the depolarising current of phase 0, while Kv11.1 and Kv7.1 promote repolarisation.

Atrium Atrial action potentials regain the sharp upstroke (150-300 V/s) of Nav1.5 expression, but exhibit a short plateau phase and a prolonged repolarisation phase. Re- gional differences in action potential duration (APD) occur throughout the atria as a result + 2+ of differential expression of K and Ca channels. Reduced IK1 (1/8 of the ventricle), produces a slightly more positive Phase 4 potential (-80 mV) and slowed Phase 3 repolarisation.

Atrioventricular node The atrioventricular node (AVN) delays AV conduction providing an optimal contraction sequence of atria and ventricles. Similar to the SA node, it lacks signiﬁcant INa – Phase 0 is effected by ICa. There is no signiﬁcant plateau period; instead,

Figure 1.5: Cardiac action potential shapes in different regions of the human heart. Each of the voltage waveforms are specially suited to their regional function (see text). a rapid repolarisation phase dominates conducted mostly by IKr. Large If and absent IK1 produce slow depolarisation during Phase 4.

Purkinje Purkinje cells are starkly different to AV node cells. Delay in atrial to ventricular signalling is now replaced by the need for synchronous contraction of the ventricles. The velocity of Phase 0 reaches 800 V/s (cf. 150-300 V/s in the ventricle); abundant connexin 40 expression aids rapid conduction. Slow depolarisation during Phase 4 may provide intrinsic pacemaking in the ventricle in the absence of AV nodal signals.

Ventricle The general case of the ventricular action potential is detailed in Subsec- tion 1.2.2.1. Regional differences, however, are present. Epicardial action potentials have a shorter APD, and more prominent Phase 1 notch compared with endocardial action potentials. In other respects they are alike. Mathemat-

20 ical simulations of human ventricular tissue model such changes by a 4-fold reduction in Ito conductance in endocardial cells (ten Tusscher et al., 2004; Nabauer¨ et al., 1996). In similar fashion, the M cells of the mid-myocardial layer are modelled by a 4-fold reduction in IKs compared with endocardial and epicardial cells (ten Tusscher et al., 2004; Drouin et al., 1995); this reproduces the long APD, and electrophysiological characteristics of M cells seen in ventricular-wedge preparations (Antzelevitch, 2005); however, such modiﬁ- cations are almost certainly a simpliﬁcation.

21 1.2.2.3 M cell

“One injects an animal with a foreign substance, takes out its heart, cuts it to pieces, puts it in salt water, stabs it with needles, shocks it with electricity and calls it normal.” Moe G6

The functional significance of the M cell in vivo is controversial; its existence, less so. First identified in canine ventricular tissue preparations (Sicouri and Antzelevitch, 1991), it has a long action potential duration (APD) that prolongs disproportionately, when compared to epicardial or endocardial cells, at slow cycle lengths or with block of Kv11.1 (Drouin et al., 1995; Antzelevitch, 2007). These properties correspond to reduced expression of Kv7.1, and a small increase in late sodium current (Schram et al., 2002). Heterogeneity of APD across the ventricular wall from the presence of M cells results in significant dispersion of repolarisation in ventricular wedge preparations (Antzelevitch, 2007). However, such ex vivo findings do not neatly translate to in vivo studies which universally fail to demonstrate any significant transmural difference in repolarisation time (Rodriguez-Sinovas et al., 1997; Opthof et al., 2007b). This absence may be attributed to an averaging effect of electrotonic currents flowing according to local (cell to cell) voltage gradients — the ‘longer prolonging the shorter; the shorter abbreviating the longer’. Thus, the more isolated a cell from its in vivo environment, the more individual its expression of intrinsic ionic channel concentrations: longest to shortest difference of APD at 1 Hz pacing rates — isolated cells, 109 ms (Drouin et al., 1995); ventricular tissue slices, ∼100 ms (Sicouri and Antzelevitch, 1991); ventricular wedge preparations, 30-40 ms (Antzelevitch et al., 1996), in vivo, ∼0 (Rodriguez-Sinovas et al., 1997; Opthof et al., 2007b). ‘Masked’ differential ionic expression may assume importance in the setting of reduced cellular coupling or as the result of an incompletely compensated reduction in repolarisation current, whether congenital or acquired.

6(Opthof et al., 2007a)

22 1.3 Congenital Disorders of Cardiac Action Potential Repolar- isation

1.3.1 Congenital long QT syndrome

History In 1856 Friedrich Meissner described the sudden death of a young girl born deaf (Meissner, 1856). A small act of theft brought her before the director of the asylum where she lived. It is said that when she saw his disappointment she fell down and died. The young girl, Steinin, had two brothers also born deaf who died at times of acute emotion. More than 100 years later, Jervell and Lange-Nielsen reported on four young boys among a family of 6 who suffered congenital deafness in association with prolongation of the QT interval and syncope (Jervell and Lange-Nielsen, 1957). The appellation Jervell Lange-Nielsen (JLN) was subsequently attached to this evidently autosomal recessive disorder. Subsequently, Romano and Ward published independent reports of prolongation of the QT interval associated with syncope and sudden death in the absence of a hearing deﬁcit and inherited in a manner unlike JLN (Romano et al., 1963; Ward, 1964). It is now known that this autosomal dominant form of the condition – Romano-Ward Syn- drome – is far more common than JLN. In these important case reports the association of QT prolongation and syncope was established, but the syncopal rhythm was not observed. The wonderfully named torsade de pointes7 (TdP) (“a twisting of the points”) had been recorded and published (Dupler, 1953; Schwartz and Hallinger, 1954) but not fully characterised when Dessertenne wrote his seminal paper (Dessertenne, 1966). He described the case of an elderly woman who presented with complete atrioventricular block, prolongation of the QT interval, and a form of polymorphic ventricular tachycardia that twisted about the isoelectric line. His complete and careful study ensured the survival of the descriptive term. Four years later, an association between congenital long QT syndrome (LQTS) and TdP was made (Motte et al., 1970).

Molecular basis While a mechanism was not advanced in the initial descriptions of LQTS, the close association of syncope with times of sudden emotion or exercise suggested a role for the autonomic nervous system. Modulation of the QT interval through interruption or stimulation of the sympathetic nervous system (Yanowitz et al., 1966) and the initiation of T wave alternans by stellate ganglion stimulation (Schwartz and Malliani, 1975) supported a theory of sympathetic imbalance. The strong protection afforded by beta-blockers led to widespread acceptance of the sympathetic hypothesis (Schwartz

7Throughout, a single twisting will be termed torsade de pointes,arunof‘twistings’torsades de pointes following Dessertenne (Moise, 1999)

23 et al., 1975) and was the common belief for two decades, notwithstanding suggestions that a hitherto undeﬁned intra-cardiac abnormality may be present (Attwell and Lee, 1988; Jackman et al., 1988).

Gene Locus Syndrome Protein Function Occurs in KCNQ1 11p15.5 LQTS1,SIDS Kv7.1 αIKs ↓ 30-35% KCNH2 7q35 LQTS2, SIDS Kv11.1 αIKr ↓ 25-30% SCN5A 3p21 LQTS3, SIDS Nav1.5 αINa ↑ 5-10% ANK2 4q25 LQTS4, ABS Ankyrin-B INa,K ↓ INCX ↓ 1-2% KCNE1 21q22.1 LQTS5 mink βIKs ↓ 1% KCNE2 21q22.1 LQTS6, SIDS MIRP1 βIKr ↓ rare KCNJ2 17q23 LQTS7, ATS Kir2.1 αIK1 ↓ rare CACNA1C 12p13.3 LQTS8, TS Cav1.2 α1c ICa,L ↑ rare CAV3 3p25 LQTS9, SIDS Caveolin-3 INa ↑ rare SCN4B 11q23 LQTS10 Nav1.5 β4 INa ↑ rare AKAP9 7q21-q22 LQTS11 AKAP9 IKs ↓ rare SNTA1 20q11.2 LQTS12 α1-syntrophin INa ↑ rare KCNQ1 11p15.5 JLNS1 Kv7.1αIKs ↓↓ rare KCNE1 21q22.1 JLNS2 mink βIKs ↓↓ rare

Table 1.3: Genes and proteins that cause the congenital form of the long QT syndrome (LQTS). Anderson Tawii syndrome (ATS): ∼50-60% are KCNJ2 mutation carriers and suffer periodic mus- cle paralysis and developmental abnormalities. Ankyrin-B syndrome (ABS): sinus bradycardia, paroxysmal AF, VF, and polyphasic T waves. JLNS1 (Jervell Lange-Nielsen Syndrome Type 1): congenital deafness, QTc prolongation, ventricular arrhythmias. AKAP9: A-kinase anchoring protein 9. JLNS2: compound heterozygous mutations, and asymmetric T waves with rapid terminal conﬁguration. LQTS accounts for ∼10-15% of cases of SIDS. (Adapted from Lehnart) (Lehnart et al., 2007)

The genetic basis of LQTS was conﬁrmed in linkage studies during the early 1990s (Keating et al., 1991; Jiang et al., 1994). The three genes responsible for the vast majority of cases of LQTS were identiﬁed in 1995 and 1996: KCNH2, the cause of LQTS2 (Curran et al., 1995); SCN5A, the cause of LQTS3 (Wang et al., 1995); and KCNQ1, the cause of LQTS1 (Wang et al., 1996).

Diagnosis A diagnosis is made by the presence of classic symptoms – ‘seizures’, syncope, sudden cardiac death – in the setting of a prolonged QT interval. The QT interval is measured from the beginning of the QRS complex to the end of the T wave. It should be measured when the resting HR is close to 60 BPM, averaged over 3-5 beats, and be the longest interval present in lead II, V5 or V6. When the end of the T wave is difficult to identify, a tangent line may be drawn from the steepest portion of the downstroke of the T wave to the isoelectric line – the point of intersection corresponding to the end of the T wave. The abnormal repolarisation of LQTS may produce unusual T waves with biphasic or bifid morphologies; these are often mistaken for the U wave. As a general rule, deflections in the latter half of the T wave that are of smaller magnitude than the

24 Figure 1.6: ECG ﬁndings in a patient with long QT syndrome type 2. (A) – ECG of an 18 year- old woman with recurrent syncope. The QT interval is prolonged at ∼550 ms in V6. In this case, the biphasic T waves in V2 and Lead III were consistent with a Kv11.1 mutation. (B) – cardiac telemetry recording showed the onset of TdP, with an ‘R on T’ initiation, consistent with an early afterdepolarisation (EAD). (C) – subsequent ECG, during syncope, illustrates the characteristic appearance of torsades de pointes (TdP) (Adapted from Perrin et al. (2008b).

25 peak, should be ignored and the end of the T wave established by the tangent method as detailed above. When the two peaks are equal in size, the end of the T wave is identified as the end of the second peak (Goldenberg et al., 2006). Careful studies have identified T wave morphologies that correspond, though with some overlap, to the underlying genotype. LQTS1 typically displays a low amplitude, broad-based, T wave. LQTS2 T waves are biphasic or bifid (Figure 1.6). The T wave of LQTS3 may be peaked, and is usually of late onset (Zhang et al., 2000). The specificity of each change is not sufficient to confirm genotype from morphology, however it may provide a useful starting point for the identification of an underlying mutation in an affected individual. The diagnosis of congenital LQTS is made more difficult in the presence of a normal or borderline abnormal QT interval at rest. Provocative tests can then be helpful to identify latent QT prolongation, and predict the likely genotype. With exercise, LQTS patients may demonstrate paradoxical increases in the QT interval, or else an inappropriately shortened QTc (corrected QT interval) (Swan et al., 1999). This abnormal response is most marked in patients with loss of function mutations in Kv7.1 — LQTS1. Adrenaline infusion may unmask latent-LQTS1 in a similar manner. LQTS2 patients also exhibit a paradoxical increase in QT interval early during exercise or adrenaline infusion, but an accumulation of IKs, due to its slow deactivation kinetics, reduces the final QT interval to normal levels. An increment in the second peak of the bifid T wave to a level greater than the first peak is a specific but insensitive finding in LQTS2 patients (Vyas and Ackerman, 2006). Last, a molecular diagnosis may be made through primary screening, or after the molecular diagnosis of a close relative. However, up to 40% of clinically confirmed patients with congenital LQTS will be genotype negative (Splawski et al., 2000; Bai et al., 2009).

Triggers Perhaps the most interesting aspect of congenital LQTS is the strong association between genotype and and individual triggers for syncope and sudden cardiac death. In a landmark international study, Schwartz and colleagues examined 670 patients with genotype confirmed LQTS and cardiac symptoms for genotype-specific triggers of arrhythmia. A significant association for exercise – LQTS1, emotion – LQTS2, and sleep – LQTS3, was established. Most marked was the association between these triggers and the cause of sudden cardiac death for each genotype (Schwartz et al., 2001).

Predicting individual risk Risk assessment in congenital LQTS has been said to be “the Achilles heel of appropriate treatment” (Vincent, 2005). However, some general

26 predictions may be made: an early age of onset, and a longer QT interval, mark a greater risk of sudden cardiac death (Priori et al., 2003); mutations affecting SCN5A (LQTS3) are, in general, more malignant than those causing LQTS1 or LQTS2 (Zareba et al., 1998); and an individual’s risk of sudden cardiac death is determined by their clinical expression of disease – e.g. personal history of syncope, absolute length of QT interval – rather than the clinical expression of a particular genotype in an affected sibling (Kaufman et al., 2008). Yet, at the genetic level there is a low penetrance of disease (Schwartz et al., 2003), the QT interval is variable over time, and the risk for the individual, even now, may only be described in general terms.

Treatment The efficacy of beta-blocker therapy in LQTS1 is well established (Vincent et al., 2009). It appears that protection against sudden cardiac death is near to 100% in patients who are compliant with therapy; therapy ‘failures’ are usually attributable to non-compliance or the addition of a QT-prolonging drug (Vincent et al., 2009). Such high efficacy is predictable for LQTS1 — clinical events are more common during exercise. Some evidence suggests that beta-blocking therapy may be useful in LQTS2, perhaps as a result of protection against the sympathetic burst of an acute emotional event (Schwartz et al., 2001). No evidence exists for the effectiveness of beta-blockade in LQT3; most clinical events in this form occur during the slow heart rates and low sympathetic tones of sleep; further slowing of the heart rate may possibly be harmful. The provision of an implantable cardiac defibrillator to all patients surviving cardiac arrest is mandatory. The decision to implant in a patient as primary prevention is difficult, taking into account personal preference, length of the QT interval, and suspected genotype.

1.3.2 Congenital short QT syndrome

As gain of function mutations in depolarising currents, and loss of function mutations in repolarising currents may result in congenital LQTS, it might be conjectured that the converse would provoke a shortened QT. This was confirmed when, in 2000, a new syndrome was characterised by short QT intervals on the ECG, and early-onset atrial fibril- lation; sudden cardiac death was later described as a common manifestation. The first causative gene was KCNH2 – SQTS1 – through a gain of function mutation; other genes have now been identified: KCNQ1 (Bellocq et al., 2004), KCNJ2 (Priori et al., 2005), and CaCNA1C and CACNB2b (Antzelevitch et al., 2007), (SQTS 2-5 respectively). Short QT syndrome type 1 is characterised by persistently short QT interval, symmet-

27 ric T waves, and a propensity to atrial ﬁbrillation and early sudden cardiac death. The genetic defect is a missense mutation leading to substitution of asparagine for lysine at position 588 (N588K), in the long SSP linker region of Kv11.1. Serendipitously, this mutation, due to its peculiar effect on channel kinetics, is an ideal construct to investigate drug-binding to Kv11.1 (Clarke et al., 2006; Perrin et al., 2008).

28 1.4 Acquired Disorders of Cardiac Action Potential Repolari- sation

1.4.1 Acquired long QT syndrome

Background Notwithstanding the importance of congenital LQTS, the acquired form of the condition is a more common cause of TdP. The terms drug-induced LQTS and acquired LQTS are often interchanged but are not synonymous: the first ECG record of TdP is attributed to severe hypomagnesaemia (Zwillinger, 1935), and the first thorough description of TdP involved a case of complete atrioventricular block (Dessertenne, 1966). Numerous other causes of acquired LQTS have been described including: acute myocardial infarction (Halkin et al., 2001), transient apical ballooning of the left ventricle (Denney et al., 2005), starvation (Petrov, 2003), subarachnoid haemorrhage (Di Pasquale et al., 1984) and electrolyte abnormalities. However, drugs are the most frequent cause and this section will focus on this condition. Drug-induced prolongation of the QT interval may occur as a predictable response to a compound designed to block cardiac repolarising currents, e.g. dofetilide, sotalol and ibutilide. Unfortunately, QT prolongation and TdP may also arise as an unwanted side-effect of a compound designed to act at non-cardiac targets. The most common cause of drug-withdrawal or restriction in already marketed drugs is prolongation of the QT interval and TdP (Roden, 2004). In theory, drug interaction with any ion channel that contributes to cardiac repolarisation could cause the acquired syndrome; but clinically relevant drug-induced QT prolongation is almost invariably the result of drug-block of Kv11.1 (Roden et al., 1996; Cavero et al., 2000), or interruption of Kv11.1 trafficking to the cell membrane (Dennis et al., 2007). There are, however, some rare instances of drugs that result in increased QT interval without affecting Kv11.1, including, e.g. alfuzosin (enhances sodium current; (Lacerda et al., 2008)).

Predicting risk Post-marketing reports of TdP and sudden death have led to the withdrawal of multiple therapeutic compounds (Roden, 2004). Consequently, the Food and Drug Agency (FDA) in the United States and other regulatory authorities have mandated that no new drug can be released to market without an assessment of its afﬁnity for Kv11.1, and a prediction of its propensity to prolong the QT interval (Guth, 2007). Such regulations presuppose these markers as a surrogate for the risk of TdP and sudden death; but the precise molecular and physiological interrelationships between Kv11.1 blockade, QT prolongation, and TdP are uncertain. Cardiac myocyte depolarisation and repolarisation is complex, the result of multiple competing ionic currents. High afﬁnity

29 blockers of Kv11.1 may not prolong the QT interval in vivo if they also perturb other ion channels. Verapamil, for instance, is a potent blocker of Kv11.1 repolarising current, but does not prolong the QT interval or cause TdP, due to compensatory block of calcium channel depolarising current (Fauchier et al., 1997). Amiodarone blocks Kv11.1 and prolongs the QT interval, but is a rare cause of TdP (Yang et al., 2001). Therefore while almost all drugs that cause TdP block Kv11.1, the converse is not true. Nor is it true that all QT prolongation is proarrhythmic. To improve prediction models many electrocardiographic markers of TdP risk have been investigated. Among these, the measurement T-peak to T-end – that is from the peak of the T wave to its end – has shown the most promise. It correlates with an increased risk of TdP in acquired LQTS (Yamaguchi et al., 2003) and a range of other conditions (Shimizu et al., 2002; Lubinski et al., 1998). Drugs that prolong the QT interval but do not increase T-peak to T-end across an arterially perfused left ventricular wedge, rarely cause TdP in humans (Antzelevitch, 2005). Experiments in this preparation suggest that T-peak to T-end represents the time from the end of the earliest repolarisation (epicardial) to the end of the latest repolarisation (M or mid-myocardial cells), and is therefore a measure of the transmural dispersion of repolarisation (Antzelevitch, 2007). However, intact heart experiments suggest T-peak to T-end is a measure of the total dispersion of repolarisation across the ventricle (Xia et al., 2005; Opthof et al., 2007b). This does not invalidate the utility of the parameter, but questions its origin. Further study is required into the predictive value, and genesis, of T-peak to T-end but preliminary data suggests it may be an improvement over the traditional measurement of QT interval.

Individual variations in clinical effects of drug blockade The considerations above may be summarised as the intrinsic propensity of a drug to cause TdP. It is now well known that for similar doses of drug the extent of QT prolongation – or dispersion of repolarisation – varies markedly between individuals. Equal doses of drug may have different effects on QT interval due to altered metabolism increasing plasma levels of the drug; for example, consumption of grape fruit juice, which inhibits cytochrome P450, results in increased plasma levels of terfenadine and an increased QT interval (Benton et al., 1996). However, in many instances the prediction of which individuals in particular will develop TdP after the administration of a drug is not easy. To explain the apparently idiosyncratic response of some patients to standard drug doses, the concept of “repolarisation reserve” has been proposed (Roden, 1998). In this framework carefully balanced ionic currents act to maintain homogenous repolarisation across the ventricles such that early afterdepolarisations are rare, and no substrate for re-entry exists. A variety of stres- sors may reduce the reserve including genetic mutations, common polymorphisms (Sesti et al., 2000), female gender, electrolyte abnormalities and concomitant disease such that

30 the addition of Kv11.1 blockade markedly increases dispersion across the ventricle and provides a substrate for TdP. It is reported that between 34% and 52% of the variation in QT interval between individuals is heritable (Hong et al., 2001; Busjahn et al., 1999; Newton-Cheh et al., 2005). In addition, individual susceptibility to drug-induced LQTS may be determined by a host of genetic variations (mutations and polymorphisms) that do not prolong the QT interval at baseline but reduce overall repolarisation reserve. Multiple clinical mutations have been identified in patients with drug-induced LQTS (Donger et al., 1997; Napolitano et al., 2000; Yang et al., 2002; Saenen et al., 2007). The A555C mutation in KVLQT1 identified by Donger et al. (1997), is illustrative of a mutation that results in no or minimal QT prolongation, but is exposed in the context of Kv11.1 blocking drugs. Single nucleotide polymorphisms may also influence risk in drug-acquired LQTS. The Y1102 allele in SCN5A, common in African-Americans, results in a modest alteration in gating characteristics, and a small lifetime risk of a fatal arrhythmia. It was, however, vastly overrepresented in an examination of cases of acquired arrhythmia where 56.5% carried at least one allele compared to 13% of controls (Splawski et al., 2002). Other polymorphisms may influence risk by sensitising Kv11.1 to drug-block, e.g. T8A in the regulatory subunit of Kv11.1, MiRP1. A recent report illustrates this interplay of congenital and acquired factors in another cause of acquired LQTS; in a group of patients presenting with a prolonged QT interval (>600 ms in their study) complicating complete atrioventricular block, 17% had mutations in KCNH2 (Chevalier et al., 2007). The shift in current thinking is therefore towards a continuum of arrhythmic risk rather than the binary categorisation of patients into congenital or acquired LQTS.

1.4.2 Kv11.1 (hERG) channel

History Kv11.1 was discovered in 1994 (Warmke and Ganetzky, 1994) in human hip- pocampal cDNA libraries as a homolog of Drosophila EAG protein. Mutations in Kv11.1 were found to cause congenital LQTS in 1995 (Curran et al., 1995). Kv11.1 was subsequently shown to conduct the rapid delayed rectiﬁer current (IKr) (Sanguinetti et al., 1995), and be blocked by methanesulfonanilides (Trudeau et al., 1995), providing a mechanistic link between congenital and acquired LQTS.

Structure Kv11.1 is a homotetramer, each major subunit containing six transmembrane domains (Figure 1.3). S5, the pore helix, and S6 form the pore domain which surrounds the ion conduction pathway. S1-S4 in each subunit form the voltage sensing domains (VSD), that move through the membrane in response to changes in the trans-membrane

31 electric field. An unusually long extracellular domain between S5 and the P-domain (S5P linker) of 40 amino acids is unique to the EAG family of proteins (Torres et al., 2003). The N-terminal intracellular region contains a Per-Arnt-Sim (PAS) domain found only in EAG channels in mammals. In Kv11.1, PAS plays a significant role in channel gating: its deletion accelerates deactivation (Schonherr¨ and Heinemann, 1996; London et al., 1997). A cyclic nucleotide binding domain (cNBD) is present at the C-terminal. Its role in channel function is not clear, though mutations in this area result in trafficking defects (Akhavan et al., 2005).

Gating Kv11.1 has three distinct gating conformations: closed (non-conducting), open (conducting), and inactivated (non-conducting) (Figure 1.7). Activation is effected by changes in transmembrane voltage that provoke a structural alteration in S1-S4 (that is, the VSD); the resultant mechanical force is transmitted via the S4-S5 linker to the cytoplasmic domains of S6 to either open or close the bundle- like activation gate. The exact conformational changes that occur in the VSD are, at this point, uncertain (Jiang et al., 2003; Bezanilla, 2008). C-type inactivation is the product of structural changes in the selectivity ﬁlter which produce ‘constriction’ of the conductance pathway (Spector et al., 1996). While analogous to C-type inactivation in Shaker, Kv11.1 inactivation is considerably faster, and displays a marked voltage-dependence. Inactivation may be largely abolished by mutations at Ser620 in the P-domain or demonstrate altered voltage-dependence with mutation of Ser631 (Schonherr¨ and Heinemann, 1996). Further, charge substitution at Asn588 – distant to the pore on the long S5-P linker – titrates the voltage-dependence of inactivation with little effect on activation (Clarke et al., 2006). This suggests a separation of activation and inactivation gating in Kv11.1.

Function Kv11.1 is slowly activated during phase 0 (fast depolarisation) of the cardiac action potential, and then transitions rapidly into an inactivated state during the action potential plateau. This results in a small spike of current at onset, followed by a rapid reduction in current to low levels during the plateau (Figure 1.7B). Therefore, early depolarisation is largely unopposed allowing optimal excitation-contraction coupling. As the cell repolarises, Kv11.1 recovers rapidly from inactivation into the open state and then slowly deactivates. The long open state dwell-time allows passage of a large outward current at negative membrane voltages. In the event of a sudden depolarising current during phase 3 (that is, an early afterdepolarisation), Kv11.1 is able to respond with a large repolarising current (Figure 1.7B),

32 Figure 1.7: The gating transitions of Kv11.1. (A) – activation and deactivation are slow compared to the rapid C-type inactivation conferred by selectivity filter constriction. Consequently, after channel activation, current flow is small due to fast inactivation. Repolarisation of the membrane voltage favours channel recovery from inactivation, which is then delayed in the open-state by slow deactivation producing large current flow – (B). The unique kinetic properties of Kv11.1 protect the myocyte from depolarising forces during the terminal stages of the cardiac action potential; in response to an early afterdepolarisation, a large flow of current passes through open Kv11.1 channels (B) thus opposing the propagation of a premature beat.

33 thus opposing early reactivation of the cell (Lu et al., 2001)

1.4.3 Drug-binding to Kv11.1

Molecular basis of drug-binding to Kv11.1 Kv11.1 binds a wide variety of drugs of different class and structure. This promiscuous nature is conferred by the presence of two aromatic amino acid residues located on each of the four subunits that line the central cavity of the channel pore (Figure 1.8). A tyrosine at position 652 and a phenylalanine at 656 are predicted by homology studies to face the inner cavity, consistent with experiments demonstrating the necessity of channel activation for drug-block. Alanine scanning mutagenesis has demonstrated that these two residues dictate the high afﬁnity binding of most drugs that bind to Kv11.1 (Mitcheson et al., 2000).

Figure 1.8: Schematic diagram of the drug-binding pocket of Kv11.1 and key binding residues. 2 of the 4 subunits are displayed. Note that G648 is probably not exposed to the pore cavity, and is thought to inﬂuence drug-block by changing the conformation of the drug-binding pocket. The aromatic residues – Y652 and F656 – interact with most drugs binding to the channel.

Unfortunately, while the residue determinants of binding have been identiﬁed and are shared by different drugs, it is likely that the precise docking arrangement is not the same for all drugs. Computer modelling suggests that most drugs do not interact with a static arrangement of residues, but instead can adopt multiple binding arrangements (Masetti et al., 2008; Farid et al., 2006). Drug molecules bind to different combinations of residues within and between different subunits (Myokai et al., 2008). Further, the spatial relationships of these residues is likely to change with channel gating; some evidence suggests that the inactivated state is preferred by high afﬁnity binding drugs (Ficker et al., 2001; Herzberg et al., 1998). The complexity of the drug-Kv11.1 interaction is a principal rea-

34 son for the difﬁculty in providing an accurate assessment of Kv11.1 afﬁnity from chemical structure (Recanatini et al., 2008).

State-dependent drug-binding to Kv11.1 It is well established that drugs bind in the central cavity of the pore region of Kv11.1 (Mitcheson et al., 2000), and that the channel must activate (open) before this can occur (Kiehn et al., 1996). At depolarised voltages, Kv11.1 may occupy either the open or inactivated state (Sanguinetti et al., 1995; Smith et al., 1996), but it is not established whether the open or inactivated state is preferred for drug-binding. Evidence in support of preferential binding to the inactivated state comes primarily from studies demonstrating reduced affinity for mutant channels that either abolish (S620T; G628C:S631C) or reduce inactivation (S631A) (Ficker et al., 1998; Weera- pura et al., 2002). These mutations, however, lie proximate to the selectivity filter and putative drug-binding pocket and so may affect drug-block by gating-independent means through local changes in the drug-binding pocket. Mutant-Kv11.1 channels G648A and T623A promote inactivation but reduce methanesulfonanilide block (Mitcheson et al., 2000); and voltage protocols designed to favour occupancy of the inactivated state (i.e. strong depolarisations) relieve drug-block (Kiehn et al., 1996; Numaguchi et al., 2000). Therefore, the evidence is conflicted, and no definitive answer can be given regarding the existence of state-dependent drug-binding to Kv11.1. More certain: drug-unbinding from Kv11.1 is state-dependent. Elegant work by Stork and colleagues has demonstrated the marked frequency-dependence displayed by some drugs when binding to Kv11.1 (Stork et al., 2007). This implies that some compounds are able to dissociate from the inner channel cavity at hyperpolarised (negative) membrane voltages, while others are seemingly trapped in the cavity. Drug escape may be interpreted in two ways: drugs are able to exit the the true closed conformation of the channel, or else channel closure is delayed until drug escape occurs.

1.4.4 Safety pharmacology

Background The correlation between Kv11.1 block and sudden cardiac death is of sufficient strength that testing of all new compounds for their affinity to Kv11.1 is now mandated. The International Conference on Harmonisation of Technical Requirements for Registration of Pharmaceuticals for Human Use (ICH) Document ICH S7B provides guidelines for an assessment of the QT liability of a drug, and was endorsed by the US Food and Drug Administration in 20058. In brief, such recommendations are limited to an in vitro measurement of drug-Kv11.1 affinity and in vivo QT prolongation studies in

8http://www.fda.gov/RegulatoryInformation/Guidances/ucm129121.htm

35 animals. Neglecting to emphasise in vitro action potential duration (APD) assays in the front-line of testing, these guidelines have received some criticism (Chen et al., 2009). In regard to the in vitro measurement of drug-Kv11.1 afﬁnity, the available testing regimes may be classiﬁed separately according to their throughput.

1.4.4.1 Low-throughput drug-binding assays

Voltage patch-clamping Without dispute, the gold-standard measurement of a drug’s affinity for Kv11.1 is the voltage patch-clamp technique. In this method, a micro-pipette is advanced to the cell membrane and with the aid of gentle suction, forms a tight seal (a gigaseal) with its surface. Stronger suction disrupts the underlying membrane providing electrical access to the interior of the cell. Consequently, voltage across the cell membrane may be tightly controlled – clamped – and the flow of current across the lipid bilayer measured (Hamill et al., 1981). Recordings obtained in cells expressing Kv11.1 may be used to determine the affinity of a drug for the channel by measuring the membrane current in control conditions and in the presence of increasing drug concentrations. Fur- ther, with careful attention to technique, kinetic measurements – on- and off-rates – may be calculated or inferred from computer modelling simulations reproducing the experimental data. Unfortunately, the technique of voltage patch-clamping is labour intensive, and requires significant training. A competent practitioner may be expected to produce a maximum of one dose-response curve per day; kinetic measurements require more time.

1.4.4.2 Medium-throughput drug-binding assays

Automated patch-clamping Several automated versions of the ‘whole-cell’ voltage patch-clamp technique have been marketed including QPatch (Sophion Bioscience A/S), Patchliner (Nanion Technologies GmbH), CytoPatch (Cytocentrics GmbH) and PatchX- press (Molecular Devices) (Chen et al., 2009). Such automation reproduces the voltage recording profile and the fast on and off perfusion of the manual technique. However, many of the systems are prohibitively expensive, and are at best low to medium throughput. Further, some lipophilic compounds may have measured affinities 10-100s of times greater than obtained from the manual method (Dubin et al., 2005). A different approach is employed by the IonWorks (Molecular Devices) automated ‘planar array’ electrophysiology platform (Schroeder et al., 2003). In this system, cells settle in the base of a 384-well plate and form an electrical seal over a small hole of 1-2 μm diameter. Amphotericin is used to permeabilise the membrane of the cell which is then in continuity with a ground electrode. Voltage-current profiles of Kv11.1 in the pres-

36 ence of hundreds of drugs may be assessed in a short time. Seal resistances, however, may be as low as 50 Ω, averaging 100-200 Ω, thus preventing precision voltage-clamping; further, series resistance and capacitative artefacts cannot be compensated. Nonethe- less, there is a reasonable correlation between recordings from IonWorks and the manual patch-clamp technique. A 64-well format of the technique has been developed that by averaging the recordings over all wells – for a single compound – markedly improves reproducibility and accuracy, though at the expense of throughput (Chen et al., 2009).

Competitive binding assays Competitive binding assays compare the affinity of a test compound to that of a radiolabelled, and potent, blocker of the Kv11.1 channel e.g. [3H]- dofetilide or [35S]-MK-499. The displacement of reference compound in the presence of test drug is determined by the relative affinities of each for the drug-binding site. As drug-binding to Kv11.1 is complex, and likely to involve more than one discrete binding site, the affinity of compounds which bind to sites different from the binding-site of the reference compound may be underestimated. Cisapride, when tested with a [35S]-MK-499 binding assay, has an IC50 roughly 3x reduced compared to the voltage-clamp measurement. This results from an alternative binding-site to the methanesulfonanilides (Mitche- son et al., 2000). For this reason, competitive binding assays are not recommended by the ICH S7A document.

1.4.4.3 High-throughput drug-binding assays

Rubidium flux assays Rubidium ions pass through Kv11.1 channels with similar flux to K+. In the rubidium assay, cells expressing Kv11.1 are incubated with Rb+, displacing K+ from the system. A test compound is then applied. Depolarisation of the cell membrane activates Kv11.1 allowing Rb+ efflux into the supernatant. The degree of efflux, and therefore the conductance of Kv11.1 channels, is determined by comparing the supernatant [Rb+] to that of the cell [Rb+]. In the presence of a drug blocking Kv11.1, the [Rb+] in the supernatant will be low. A reasonable correlation exists between such measures and those obtained from voltage patch-clamping (Tang et al., 2001). Rb+, however, shifts the V0.5 of voltage-dependence of steady-state inactivation to the right, thus reducing the proportion of channels in the inactivated state at the voltage levels of the assay. It may, therefore, give a reduced estimate of drug affinity for the channel when the drug preferentially binds the inactivated state (Rezazadeh et al., 2004). Further, the assay operates at membrane voltages in the range of heterologously expressing Kv11.1 cells, roughly -70mV to -25 mV, and therefore underestimates affinities drugs that bind in a voltage-dependent manner. A previous disadvantage of this system was the formation of radioactive waste, with all of its attendant dangers and difficulties. This has largely

37 been overcome by the use of atomic-absorption spectroscopy, though at the expense of a lower throughput (Netzer et al., 2003).

Fluorescence-based ion channel assays Fluorescence-based ion channel assays utilise voltage-sensitive dyes to provide a ‘read-out’ of the membrane voltage of cells expressing Kv11.1 in the presence and absence of drug. As may be predicted from theory, the addition of a potassium channel to a cell will shift the resting membrane voltage toward the reversal potential of potassium (-85 mV with standard solutions); -60 mV is typical. With increasing channel block, the membrane voltage will move toward the resting membrane voltage of a ‘blank’ cell (typically, -30 mV) (Figure 1.9). The degree of voltage change provides a rank order of compounds according to their Kv11.1 affinity; calibration of each recording against the known affinity of a single compound may then allow the IC50 for each drug to be estimated. Automation of this system allows the affinity of thousands of compounds to be determined in a short period of time. After the start-up costs, which may be significant, the day to day running costs are small.

- - I - I F F & F & B B 3 B 3 L L ( L (

Figure 1.9: The theory of ﬂuorescence-based ion channel assays. In our test system the resting membrane voltage of a blank Chinese hamster ovary (CHO) cell was ∼-30 mV – (A). (B) – CHO cells stably transfected with Kv11.1 have a resting membrane voltage ∼-60 mV. Drug block of Kv11.1 – (C) – shifts membrane voltage in a positive direction; in the presence of a voltage- sensitive dye, the magnitude of voltage shift may be recorded and assumed to correlate with the degree of channel block.

Unfortunately, as fast, cheap, and safe as it is to run, it is also invariably inaccurate. This was initially explained as a result of the slow kinetics of the dye utilised (Tang et al., 2001); however, improvement in the ﬂuorescent dyes has only partially solved the problem. Some of the inaccuracy may relate to quenching artefacts and other technology

38 related issues. But it is our hypothesis that much of the inaccuracy arises from the effect of atypical channel state distributions within the membrane voltage range assayed in the procedure (i.e. -60 mV to -30 mV). If it is remembered that the cardiac action potential spends a considerable portion of its time at positive membrane voltages, and therefore Kv11.1 is predominantly inactivated during its course, it is self-evident that drug-binding assays should predominantly assay the inactivated form of the channel – presuming, of course, that a drug has measurable differences in afﬁnity for different states of the channel.

39 1.5 Aims

In summary of this thesis introduction, drug-binding to Kv11.1 is a significant clinical and drug-development problem. It is almost the sole cause of the drug-acquired LQTS and may inflict the pain of sudden cardiac death upon a family; it is a leading cause of drug- withdrawal from the marketplace. Programs are in place to prevent ultimate clinical harm. Early in drug-development, a prediction may be made as to the proarrhythmic propensity of a drug. The nature of the development process demands that the tools of prediction be both accurate, and of high-throughput. Yet the leading method – the fluorescence-based ion channel assay, only fulfils the latter. In this work, I seek to investigate the kinetic interactions of drug-binding to Kv11.1 in an effort to reveal the ‘chain of causes’ that lie between a drug’s influence on ion channel current and cardiac arrhythmia. It is hoped that the completion of this investigation will reveal the cause of fluorescence-based ion channel assay inaccuracy, and provide a means to fix it.

Speciﬁcally, my aims are:

1. To determine: the most accurate method to measure Kv11.1 inactivation; the effect of potassium on inactivation; and the effect of potassium on drug-binding to the channel.

2. To develop an assay that permits differentiation of drug-binding to the open versus the inactivated state of Kv11.1.

3. To explore the kinetics of drug-binding and unbinding to Kv11.1 using a mathematical model of Kv11.1 drug-block, and to investigate the cause of reduced drug afﬁnity at negative membrane voltages.

4. To describe the effect of different Kv11.1 drug-binding kinetics on the preclinical markers of arrhythmic risk.

5. To reproduce the membrane voltage readouts of the ﬂuorescence based ion channel assays in current-clamp experiments.

6. To produce a mathematical model of Kv11.1 drug-block in a standard cell expression system, and explore the effect of drug-block on mutant-Kv11.1 channels on cell membrane voltage.

40 Chapter 2

General Methods

2.1 Molecular Biology

Experiments on wild-type human ether-a-go-go` related gene (WT-Kv11.1) channels were performed using a Chinese hamster ovary (CHO) cell line stably expressing the Kv11.1 potassium channel (Kv11.1 cDNA kindly donated by Dr. Gail Robertson), constructed as previously described (Walker et al., 1999c). CHO cells were cultured in D-MEM/F12 with ◦ 10% FBS and maintained at 37 Cin5%CO2. Cells were studied at least 24 h after being plated on microscope coverslips. The mutant-Kv11.1 constructs (N588E, N588K, S631A, and S620T) were generated using the megaprimer PCR method (Clarke et al., 2006), sequenced to ensure successful generation, and then subcloned into the pIRES2-eGFP vector. Blank CHO cells were plated onto sterilized glass coverslips. After 24 h, the cells were transfected with the mutant-Kv11.1 construct using PolyFect Transfection Reagent (Qiagen). Successfully transfected cells, identiﬁed by the expression of eGFP, were studied using the whole-cell conﬁguration of the patch-clamp technique, in voltage- or current-clamp mode (Hamill et al., 1981).

2.2 Electrophysiology

Cells were superfused with a Tyrode solution containing (in mM): 130 NaCl, 5 KCl, 1 MgCl2, 1 CaCl2, 12.5 Glucose, and 10 Hepes. dl-Sotalol was purchased from Bristol- Myers Squibb (Victoria, Australia); all other drugs were purchased from Sigma Aldrich (NSW, Australia). Cisapride, astemizole, terfenadine, erythromycin, dofetilide, haloperidol, and quinidine were prepared as stock solutions in dimethyl sulfoxide (DMSO) and

41 subsequently diluted as required with superfusate (maximum final DMSO concentration = 0.1% vv−1). dl-Sotalol was prepared as a stock solution in Tyrode’s solution and perhexiline as a stock solution in methanol. Our laboratory has previously reported that DMSO at 0.1% vv−1 has no effect on the parameters under study (Walker et al., 1999c). Borosilicate glass tubing patch pipettes, with resistances of1-4MΩ when filled with internal solution, were made using a vertical 2-stage puller (PP-830; Narishige, Japan). The internal solution contained (in mM) 120 potassium gluconate, 20 KCl, 1.5 MgATP, 5 EGTA and 10 HEPES (pH 7.3 with KOH). Membrane potentials were adjusted by -15 mV to correct for the junction potential (Barry, 1994) between low Cl− pipette solution and external bath solution in voltage-clamp experiments. Currents were amplified (Axopatch 200B amplifier; Molecular Devices, Sunnyvale, Ca) and digitised (DigiData 1200; Molec- ular Devices) before storage on a personal computer. In voltage-clamp experiments, capacitance current transients were electronically subtracted, and series resistance com- pensation was 80-90%. Current signals were digitised at 5 kHz and low-pass filtered at 2 kHz. Some current traces were leak-subtracted off-line. Acquisition was performed using pClamp 9 and 10 software.

2.2.1 Drug-block protocol for high and low potassium, and mutant- and WT-Kv11.1

In voltage-clamp drug-block experiments, currents were measured with a two-step voltage protocol: a ﬁrst step to +20 mV for 3 s to fully activate the 20 time (s) channels and a second step to a negative mem- // 0.1 Hz 1 2 3 4 5 brane potential, -110 mV. This second step was 20 40 termed ‘revelatory’, as it allowed us to estimate the 60 total conductance of activated channels (open and (mV)voltage 80 inactivated) after the 3 s period of depolarization. Drug block was calculated as I[drug]/I[control], with Figure 2.1: Voltage protocol used to all currents measured at the end of the activating record drug-block of WT-Kv11.1 and step. For WT-Kv11.1 a single exponential was ﬁtted mutant-Kv11.1 channels. to the deactivation phase of the revelatory step and extrapolated back to the beginning of the revelatory step. Voltage-protocols were run at 0.1 Hz. Control currents were recorded2-5minaf- ter patch rupture, i.e. the time taken for current levels to stabilise at a steady-state level. Between 2 and 4 concentrations of drug were applied to each cell and all experiments completed within 20 min.

42 2.3 Statistics

Initial data analysis was performed using the Clampfit module of the pClamp 9.0 and 10.0 software. Subsequent data analysis and preparation of data for figures were performed with Mathematica 6.01 and 7.01 (Wolfram Technologies). All data are expressed as mean ± S.E.M (n) and statistical significance (P<0.05) was determined using paired t-tests.

43 Chapter 3

Potassium and Inactivation of Kv11.1

3.1 Background

Kv11.1 is a voltage-gated potassium channel with a kinetic pairing of slow activation and rapid inactivation that confers functional, if not true, inward rectification (Smith et al., 1996). Such ‘inward rectification’ is critical to its role in the cardiac action potential. Kv11.1 conducts little current during the initial upstroke of depolarisation; but substantial current passes during phase 3 promoting timely repolarisation (Subsection 1.2.2.1). Further, in response to a premature ventricular stimulus a very large current will flow to oppose depolarisation (Smith et al., 1996; Lu et al., 2001). Inactivation of Kv11.1 resembles C-type inactivation in Shaker. It is sensitive to external TEA as well as to the external [K+] (Smith et al., 1996) — channel inactivation is promoted by low external [K+]. Unlike Shaker, however, Kv11.1 inactivation is rapid and voltage-dependent (Smith et al., 1996). External [K+] also likely modulates drug-block of Kv11.1. The affinity of dofetilide and quinidine for Kv11.1 has been shown to increase in the setting of hypokalaemia (Yang and Roden, 1996); similarly – E-4031 (Wang et al., 1997b), cisapride (Walker et al., 1999b), and nicotine (Wang et al., 1999). As low external [K+] favours channel inactivation, these findings support a theory of preferential binding to the inactivated state of the channel (Figure 3.1). However, another possibility is that a high [K+] effectively displaces drugs from their binding pocket in the channel pore – i.e. exerts a direct permeant ion effect on drug-binding.

44 Figure 3.1: Schematic diagram of the open and inactivated states of Kv11.1. Opening of the activation gate is required for drug entry into the channel pore. High external [K+] is thought to reduce drug afﬁnity either by displacing drug from its binding pocket, or by reducing inactivation driven stabilisation of drug-binding. Note that inactivation is effected by ‘collapse’ or constriction of the selectivity ﬁlter thus rendering the channel non-conducting; the activation gate, however, remains open allowing drug-entry to the channel.

45 In evidence of the latter theory, E4031 has a reduced affinity for the inactivation- deficient Kv11.1[GS:CC] channel compared to WT-Kv11.1, yet the intervention of raising external [K+] in either channel reduces drug-binding to the same extent (Wang et al., 1997b). However, Lin and colleagues report that the S620T mutant-Kv11.1 channel both abolishes inactivation and the external [K+] dependence of cisapride block (Lin et al., 2005). Confusingly, several studies question whether low external [K+] promotes drug- block of Kv11.1 at all. Taking dofetilide as an example, equal numbers of studies support the idea that external [K+] modifies drug-binding to Kv11.1 (Yang and Roden, 1996; Yang et al., 2004; Lin et al., 2007), as refute it (Limberis et al., 2006; Kiehn et al., 1996; Weer- apura et al., 2002). No doubt, differences in voltage-clamp protocols, cell and drug preparations, have produced these discordant findings. In regard to the present thesis, the degree to which potassium is likely to affect drug-binding is of importance as two of the principal preclinical screening methods for a drug’s arrhythmic potential – the rubidium flux assay, and fluorescent screening assay – employ high potassium to activate Kv11.1 channels (Subsection 1.4.4.3 ). Carefully conducted experiments examining the mechanism and magnitude of external [K+] modification of Kv11.1 drug-block are essential if the inaccuracy of high-throughput screening (HTS) is to be remedied. In this chapter, therefore, I seek to: first, investigate different measurement methods of inactivation measurement; second, determine the effect of different external [K+]on channel inactivation; and last, establish the magnitude of effect of high external [K+]on reducing, if at all, the affinity for several drugs to Kv11.1.

46 3.2 The Measurement of Inactivation in Kv11.1

3.2.1 Introduction

Two-pulse peak-antipeak In 1995, Sanguinetti and colleagues employed a two-pulse voltage protocol to investigate the voltage-dependence of steady-state inactivation(Sanguinetti et al., 1995). In brief, cells were depolarised (+40 mV) to activate and inactivate the channels, before peak or anti-peak tail currents were measured at test-voltages ranging from +20 mV to -120 mV (Figure 3.2). A rectiﬁcation factor (R) was calculated based on tail current amplitude (Equation 3.1) normalised to the electrical driving force (V - Ek) and the maximal conductance of the channel. R was then plotted against voltage and ﬁt to a Boltzmann equation (Equation 3.2). This method will be termed the ‘two-pulse peak- antipeak’ method of steady-state inactivation measurement (Figure 3.2A).

+40 mV A B +20 mV 3s 2s 30ms

-80 mV -80 mV -80 mV

-140 mV - 140 mV

Figure 3.2: (A)–two-pulse peak-antipeak, and (B)–three-pulse, voltage protocols employed to investigate the voltage-dependence of steady-state inactivation. Both employ a long depolarising step in order to activate all Kv11.1 channels. At +20-40 mV almost all channels will enter the inactivated state; a hyperpolarised voltage – the second pulse of (A) and (B) – favours channel recovery from inactivation. (Figure 3.6)

Three-pulse In 1996, Smith and colleagues (Smith et al., 1996) investigated the voltage- dependence of Kv11.1 inactivation and its mechanism. They employed a ‘three-pulse’ voltage protocol. Similar to the two-pulse peak-antipeak, a depolarising voltage (+20 mV) was used to completely open and inactivate the channels. The second step was short – 30 ms – with test-voltages employed in 10 mV decrements with each new sweep to a minimum of -130 mV. During this brief step, channels recovered from inactivation and relaxed to steady-state open-inactivated distributions. A ﬁnal voltage step to +20 mV allowed measurement of the proportion of channels that had recovered from inactivation during the test-voltage (Figure 3.2B).

47 Two-pulse extrapolated A third variation is the two-pulse extrapolated method (Torres et al., 2003). Similar to the method of Sanguinetti and colleagues, a two-pulse voltage protocol is employed. Unlike their method, the proportion of channels recovering from inactivation during the second pulse is determined by ﬁtting a single or double exponential to the deactivation phase of the current trace, and recording the peak or antipeak current as the intersection of this exponential line with the beginning of the second pulse (Figure 3.3). In this section I use the two-pulse peak-antipeak, three-pulse, and two-pulse extrapolated methods to investigate the voltage-dependence of steady-state inactivation of Kv11.1. The strengths and weaknesses of each method are discussed. The two- pulse extrapolated method is proposed as the most accurate measure of the voltage- dependence of steady-state inactivation of Kv11.1.

3.2.2 Experimental protocols

CHO cells stably expressing Kv11.1 were voltage patch-clamped as previously described (Section 2.2). The two-pulse peak-antipeak voltage protocol consisted of a depolarising step to +40 mV for 3 s from a holding potential of -80 mV, followed by a series of test potentials from +40 mV to -130 mV (Figure 3.2A) Following the analytical method of Sanguinetti and colleagues (Sanguinetti et al., 1995), a rectiﬁcation factor (R) was calculated at each potential using the following equation:

G.n.(V − E ) R = t k (3.1) IKv11.1

where G is the maximum conductance, n is the activation factor at +40 mV i.e. 1, Vt is the test voltage, Ek is the reversal voltage of potassium, and IKv11.1 is the peak or antipeak of the tail current at each voltage. R was plotted against test voltage, and the ﬁnal data ﬁt to a Boltzmann equation:

1 G = (3.2) V0.5−Vt (1 + e k )

where V0.5 is the voltage at which equal numbers of channels are in the open and inactivated states; Vt is the test voltage; and k is the slope of the Boltzmann curve. The three-pulse protocol followed Smith and colleagues (Smith et al., 1996). From a holding potential of -80 mV, the cell was depolarised to +20 mV for 2 s, followed by

48 A D

B E

C F

Figure 3.3: Raw data traces from two-pulse (A) and three-pulse (D) voltage protocols employed to measure voltage-dependence of steady-state inactivation. (B) – typical experimental recording from the two-pulse voltage protocol displaying‘’ current traces recorded from the 2 s hyperpolarising step. (C) – magnified version of (B) displaying the method of current measurement. In the two-pulse peak-antipeak method (Subsection 3.2.2) the peak or antipeak of the current recorded from the hyperpolarising step was measured to determine the proportion of channels that had recovered from inactivation during the test voltage. In the two-pulse extrapolated method, a single exponential curve was fit to the deactivation phase of the hyperpolarising step, and extrapolated back to the start of the voltage step — the intersection of this curve and the start of the step was substituted in place of the ‘antipeak’ current. (E) – typical experimental recording of the three-pulse voltage protocol showing only the brief 30 ms hyperpolarising voltage step and the beginning of the third step. (F) – significant deactivation is noted at negative voltages (arrow – -130 mV), as well as incomplete recovery from inactivation at more positive voltages (arrow – 0 mV).

49 test voltages ranging from +20 mV to -130 mV in 10 mV decrements. Upon return to a membrane voltage of +20 mV, the number of activated channels was measured. As the 30 ms test potential resulted in signiﬁcant deactivation of recovering channels at negative voltages (Figure 3.3B), these current traces were corrected for the magnitude of deactivation, using time constants of deactivation obtained from the two-pulse voltage protocol described above. The two-pulse extrapolated method was recorded as per the two-pulse peak-antipeak and analysed according to the method presented in Figure 3.3.

3.2.3 Results

Voltage protocols and raw traces are displayed in Figure 3.3. Only cells where both the 2 pulse and 3 pulse protocols could be recorded were chosen. An n of 7 was obtained for the two-pulse protocol, and an n of 6 for the three-pulse — in one cell, the three pulse protocol was successfully recorded but no line of best ﬁt could be found. The V0.5 of inactivation for WT-Kv11.1 expressed in CHO cells was -75.2 ± 4.7

(μA) current mV with a slope of -30.0 (n = 7) for the two-pulse extrapolated method; -50.9 ± 3.9 mV with a slope of -21.7 (n = 7) for the two-pulse peak-antipeak method and -77.7 ± 3.8 mV with a slope of -21.5 (n voltage (mV) = 6) for the three-pulse extrapolated method. The Figure 3.4: Typical currents recorded Boltzmann curves are displayed in Figure 3.6 from the third pulse of the three-pulse The Erev was -81.7 mV for the two-pulse ex- voltage protocol. – uncorrected for deactivation. – corrected for deacti- trapolated method, and -81.2 mV for the two-pulse vation. peak-antipeak method.

3.2.4 Discussion

The voltage protocol used to examine the voltage-dependence of steady-state inactivation, and the method of analysis determine the ﬁnal result. In general, an ideal method involves simple recordings and analysis, and isolates inactivation as far as possible — that is, changes to gating transitions apart from inactivation (channel activation and deactivation) should not inﬂuence the measurement.

(ms) time constant voltage (mv)

Figure 3.5: Voltage-dependence of inactivation and recovery from inactivation time constants in Kv11.1 recorded from a two-pulse voltage protocol (see Figure 3.2A). Blue curve – exponential curve ﬁt to time constants (τf ) obtained from -100 to -150 mV and describing the voltage- dependence of recovery from inactivation; red curve – exponential curve ﬁt to time constants (τb) obtained from 0 to +20 mV and describing the voltage-dependence of inactivation. Black curve – 1 1 + 1 τf τb

Problems with the three-pulse method The three-pulse method relies on a brief hyperpolarising step (Figure 3.3) to promote recovery from inactivation, followed by a depolarising step to measure the conductance resulting from recovered channels. Three problems present. First, though the second voltage step is brief, signiﬁcant deactivation of Kv11.1 occurs at negative potentials (Figure 3.3) and must be corrected. In practice this necessitates the recording of a second voltage protocol to measure the rate of deactivation at each voltage, and a subsequent mathematical correction of the affected current data. Second, the brevity of the second voltage step does not allow full recovery of the channel from inactivation at more positive voltages. To examine this phenomenon, time constants (τ) for rate of recovery from inactivation were extracted from two-pulse inactivation protocols — results in Figure 3.5. As an example of error, the τ of recovery from inactivation at -40 mV is 12.3 ms. This implies that ∼10% of channels will fail to recover from inactivation after a 30 ms pulse. As this error will only affect data points at more positive voltages, the overall effect on steady-state inactivation measurement will be to steepen the Boltzmann curve with only a small shift in the calculated V0.5 of inactivation (note annotations on Figure 3.6B). A longer hyperpolarising pulse (e.g. 50-60 ms) will reduce this error, but increase the degree of deactivation.

51 Last, raw data traces obtained with this method do not often demonstrate a typical Boltzmann distribution increasing the difﬁculty of curve ﬁtting. This probably occurs due to the contaminating effects of deactivation, and incomplete recovery from inactivation described above.

Comparing the two-pulse methods The principal difficulties associated with the three- pulse protocol do not arise in the two-pulse. Here, long depolarisation and hyperpolar- isation steps allow complete relaxation to steady-state channel distributions. However, deactivation during the second pulse may be rapid at negative voltages. Two methods have been used to overcome this measurement difficulty. First, the peak or antipeak of current1 which occurs soon after the onset of the second pulse may be measured. As recovery from inactivation is rapid, and deactivation slow, it is assumed that significant deactivation will not have occurred at measurement of the peak or antipeak of current. However, it will be seen from Figure 3.3 that at more negative voltages, significant deactivation has occurred — note the difference between antipeak and extrapolated current. This is a critical problem as the data points obtained at these voltages strongly determine the V0.5 of steady-state inactivation. Using a method of curve extrapolation, full recovery from inactivation is found to occur at more negative voltages than suggested by the peak-antipeak method (Figure 3.3C). It may be argued that as long as one protocol and method is consistently used in a study, whether the figure obtained is ‘true’ or not is of little consequence; the relative effect of a channel mutation, or change in ionic concentration will be the same. This is true to a point. Yet, it is possible that a perturbation in one gating transition may contaminate the method used to examine another transition. For example, imagine a mutation that speeds deactivation alone without affecting inactivation. Use of the two- pulse extrapolated method reveals no significant change in the V0.5 of inactivation — more rapid deactivation is compensated by a ‘steeper’ extrapolated curve. Use of the two-pulse peak-antipeak method, however, will give a greater error in the measured current of the rapidly deactivating mutant and will, falsely, right-shift the steady-state inactivation curve. For the reasons outlined above, the two-pulse extrapolated method is employed in this thesis for the measurement of the voltage-dependence of steady-state inactivation: it provides an accurate measurement of inactivation and isolates this process from other gating transitions; further, it requires only one recording, from which kinetic information may be obtained on the rate of deactivation and rate of recovery from inactivation, as well as the reversal potential of the experiment.

1The peak positive to the reversal potential, the antipeak negative to the reversal potential

52 A BC

conductance conductance conductance

53 voltage (mV) voltage (mV) voltage (mV)

Figure 3.6: Boltzmann ﬁts to steady-state inactivation data from the two-pulse and three-pulse voltage protocols. (A) – comparison of two-pulse extrapolated () and two-pulse peak-antipeak () methods to investigate the voltage-dependence of inactivation; failure to account for signiﬁcant deactivation at negative membrane voltages results in apparent complete recovery of channels at voltages around -80 mV (arrow), and a right

shifted voltage-dependence of inactivation curve. (B) – the V0.5 of inactivation is only marginally left shifted for the three-pulse (dotted-)vs.two pulse extrapolated () method; incomplete recovery from inactivation during the hyperpolarising second pulse (Figure 3.2B) results in a steeper slope of the Boltzmann ﬁtted curve. (C) – modelled family of curves showing the effect of shorter hyperpolarising second pulses, and their effect

to steepen the Boltzmann ﬁtted curve (dotted curves and arrows). The V0.5 of inactivation for WT-Kv11.1 expressed in CHO cells was -75.2 ± 4.7 mV with a slope of -30.0 (n = 7) for the two-pulse extrapolated method; -50.9 ± 3.9 mV with a slope of -21.7 (n = 7) for the two-pulse peak-antipeak method; and -77.7 ± 3.8 mV with a slope of -21.5 (n = 6) for the three-pulse method. 3.3 External K+ Concentration and Inactivation of Kv11.1

3.3.1 Introduction

Kv11.1 current magnitude increases as a linear function of external [K+] (Sanguinetti et al., 1995). The reversal voltage of Kv11.1 has a Nernstian relationship to external [K+] suggesting high selectivity for K+ (Trudeau et al., 1995) estimated at > 100:1, K+ to Na+. External [K+] also shifts the voltage-dependence of steady-state inactivation for expressed Kv11.1 (Wang et al., 1997a), and IKr (Yang et al., 1997) presumably through an interaction with an external channel site involved with inactivation. High [K+] is sometimes employed in HTS assays; it may cause assay inaccuracy through changes to inactivation gating, or else by directly displacing drug from its binding site; here, therefore, we examine the effect of different [K+] on the voltage-dependence of steady-state inactivation.

3.3.2 Results

Steady-state voltage-dependence of inactivation curves were obtained using the two- pulse extrapolated method (Subsection 3.2.2), in identical Tyrode solutions (Section 2.2) apart from external [K+]. [K+] included: 0, 5, 25 and 100 mmol/L. High [K+] was matched by lower [Na+] to maintain constant osmolarity.

conductance conductance

voltage (mV) voltage (mV)

Figure 3.7: Boltzmann curves of voltage-dependence of steady-state inactivation in high and low [K+]. Recordings with external [K+] of 0 mmol/L (green), 5 mmol/L (black), 25 mmol/L (blue), and 100 mmol/L (red) are displayed. Dotted lines represent the V0.5 of inactivation. (A)–V0.5 of inactivation: 25 mmol/L [K+], -87.0 mV; 5 mmol/L [K+], -78.2 mV; 100 mmol/L [K+], -61.0 mV. + (B) – gradual left-shifting of V0.5 of inactivation with 0, 5 and 25 mmol/L [K ] (-75.1 mV, -78.2 mV, -87.0 mV respectively).

54 + The V0.5 of steady-state inactivation was: for 5K , -78.2 ± 2.0 mV with a slope of -26.7 (n = 11); for 25K+, -87.0 ± 2.2 mV with a slope of -27.9 (n = 8); for 100K+, -61.0 ± 3.5 mV with a slope of -27.0 (n = 11) — see Figure 3.7. All differences were statistically signiﬁcant (p<0.05). The respective Erev were: -86.5 mV, -48.1 mV, and -9.4 mV. The differences in the rates of recovery from inactivation at 5 mmol/L [K+] and 100 mmol/L [K+] may be noted in Figure 3.8B. + As the V0.5 of steady-state inactivation result was unusual for 25 mmol/L [K ], experiments were performed in cells where both 5 mmol/L [K+] and 25 mmol/L [K+] inactivation protocols could be recorded. The left-shift of steady-state inactivation was maintained in this analysis: for 5K+, -80.5 ± 2.4 mV (n = 4); for 25K+, -85.6 ± 3.8 mV (n = 4) with a trend to statistical signiﬁcance (p = 0.09).

Figure 3.8: Raw current traces during the second pulse of the two-pulse voltage protocol. (A) – 5mmol/L [K+] and (B) – 100mmol/L [K+] The slower recovery from inactivation in the higher [K+] is clearly seen (arrow). Deactivation in 100 mmol/L [K+] is markedly slower. Time scales are identical between traces. .

3.3.3 Discussion

The absolute shift in the V0.5 of steady-state inactivation between normal (5 mmol/L) and high (100 mmol/L) [K+]of∼17 mV is smaller than that obtained by Wang and colleagues (Wang et al., 1997a), though their results were measured at 2 and 98 mmol/L [K+]in oocytes perhaps explaining the difference. + The left shift in V0.5 of steady-state inactivation between 5 mmol/L and 25 mmol/L [K ] was unexpected. As detailed above this result was conﬁrmed by only examining traces where both inactivation curves could be obtained from the same cell, removing the play of chance in determining the result. Thus we conﬁrmed that high external concentrations of [K+] – at least at 100 mmol/L

55 – produce a right shift in the V0.5 of steady-state inactivation, in our experiments ∼17 mV. The left-shift at 25mmol/L [K+] suggests that the effect of [K+] on inactivation may be biphasic, at lower concentrations shifting inactivation to the left (Figure 3.7B); at higher concentrations, shifting to the right (Figure 3.7A).

56 3.4 External K+ Concentration and Drug Block of Kv11.1

3.4.1 Introduction

As previously discussed, the effect of external [K+] on Kv11.1 is contentious (Section 3.1). However, the weight of evidence would suggest some reduction of drug affinity at high [K+]. The cause is unknown. Of relevance to the present thesis is the effect of external [K+] on drug-block in our mammalian cell expression system. This knowledge will influence the design of our high-throughput assay. Here, therefore, we examine the affinity of 6 drugs, previously reported to block Kv11.1, for Kv11.1 in the conditions of normal (5 mmol/L) and high (100 mmol/L) external [K+] using the Kv11.1 protocol described in Subsection 2.2.1.

3.4.2 Results

Between 4 and 8 cells were recorded for each external [K+]. As expected, maximum + conductance was greater at high [K ] (Sanguinetti et al., 1995). The IC50 of Kv11.1 for Kv11.1 in 5 and 100 mmol/L [K+] was: for cisapride, 20.5 nM and 17.9nM; for terfenadine 61.3 nM and 97.8 nM; for astemizole, 5.1 nM and 5.1 nM; for dofetilide,51nM and 57 nM, for quinidine,3.1μM and 8.7 μM; and for perhexiline, 5.9 μM and 5.0 μM (Figure 3.9).

3.4.3 Discussion

As may be appreciated from Figure 3.9 there was very little difference in drug affinity for Kv11.1 in normal and high external [K+]. The result for quinidine was most marked, a ∼2.5-fold reduction in affinity between the two concentrations; there was no appreciable difference for most of the drugs tested. This suggests that external [K+] has little effect on drug-block; however, this conclusion would be premature. Our results demonstrate that there is little or no effect of external [K+] on drug-Kv11.1 affinity at +20 mV – generalising to other voltages is not possible. Two principal theories prevail for the effect of external [K+] on drug-block: a permeant ion effect – whereby K+ effectively displaces drug from its binding pocket; or an indirect effect of K+ to shift inactivation (Figure 3.7), thereby reducing the affinity of drugs which bind preferentially to the inactivated state of the channel. To examine the latter theory a + significant difference in the V0.5 of inactivation must exist between the two [K ] at the test voltage (+20 mV). As may be seen readily from Figure 3.7, no such difference exists. This implies that channel state distributions are roughly equal between the two potassium con-

57 cisapride terfenadine

conductance conductance

[drug] (nM) [drug] (nM)

astemizole dofetilide

conductance conductance

[drug] (nM) [drug] (nM)

quinidine perhexiline

conductance conductance

[drug] (μM) [drug] (μM)

Figure 3.9: Hill plots for Kv11.1 of Kv11.1 in normal (5mmol/L – black) and high (100 mmol/L – red) external [K+]. Only terfenadine and quinidine demonstrated a signiﬁcant difference in afﬁnity for the channel between the two concentrations, and the overall effect was small. The IC50 of Kv11.1 for Kv11.1 in 5 and 100 mmol/L [K+] respectively: for cisapride, 20.5 nM and 17.9nM; for terfenadine 61.3 nM and 97.8 nM; for astemizole, 5.1 nM and 5.1 nM; for dofetilide,51nMand 57 nM, for quinidine, 3.1 μM and 8.7 μM; and for perhexiline, 5.9 μM and 5.0 μM.

58 centrations at this voltage. Therefore, no conclusions on state-dependent drug-block may be reached. Importantly, however, K+ displacement of drug from its drug-binding pocket (the former theory) would be expected be operative at all test voltages. Consequently, the relative absence of reduced drug afﬁnity at high external [K+] in our experiments is evidence against this theory.

The greatest difference in the V0.5 of steady-state inactivation between different external [K+] occurs between -50mV and -100mV (Figure 3.7). It is impractical to perform drug-binding experiments at these voltages due to the small amount of activated Kv11.1 in this voltage range. Another method is required to investigate the effect of gating transitions on drug-block of Kv11.1 and forms the substance of the chapter following.

59 Chapter 4

State-Dependent Drug-Binding to Kv11.1

4.1 Background

It is well established that drugs bind in the central cavity of the pore region of Kv11.1 (Mitcheson et al., 2000) and that channels need to open before this can occur (Kiehn et al., 1996). At depolarising potentials, Kv11.1 channels occupy either an open or an inactivated state (Sanguinetti et al., 1995; Smith et al., 1996), yet it is not established whether the open or the inactivated state is preferred for drug-binding. Evidence in support of preferential binding to the inactivated state comes primarily from studies showing reduced affinity for mutant channels that either abolish (S620T; G628C+S631C) or reduce, inactivation (S631A) (Ficker et al., 1998; Weerapura et al., 2002). These mutations, however, lie proximate to the selectivity filter and putative drug-binding pocket (Figure 4.1) and so may affect drug-block by gating-independent means – e.g. local structural changes in the drug-binding pocket. In Chapter 3, despite a significant right shift in the voltage-dependence of inactivation at 100 mmol/L [K+], most drugs showed no or little reduction in affinity for Kv11.1. It was clear that despite altered inactivation the absolute change in channel-state distribution at +20 mV was minimal (see Figure 3.7); therefore, no firm conclusions regarding preference of drugs for one state over another were reached. Here we examine state-dependent drug-binding to Kv11.1 by a different method. Mu- tations to residues that are remote from the central pore (putative drug-binding pocket) but affect inactivation, are used to measure the affinities of different drugs for the open and inactivated states of Kv11.1. Specifically, we mutate residue N588, located on the α-helix of the S5P linker (Torres et al., 2003) and thought to be distant from the drug-

60 binding pocket (Yi et al., 2007), to either glutamate (N588E) or lysine (N588K) (Clarke et al., 2006).

Figure 4.1: Schematic diagram of the S5, S6, pore helix, and S5P linker regions of two Kv11.1 subunits. The most important molecular determinants of drug-block are denoted with grey circles. Commonly employed mutants that disrupt inactivation are displayed with white stars. Note that G648 is probably not exposed to the pore cavity, and is thought to inﬂuence drug-block by determining the conformation of the drug-binding pocket; V659 may inﬂuence drug-block by altering the gating characteristics of the channel (Mitcheson et al., 2000; Mitcheson, 2008). The S5P linker is dashed, as its exact position relative to the pore is unknown.

4.2 Results

4.2.1 V0.5 of steady-state inactivation in WT-Kv11.1, N588E-Kv11.1 and N588K- Kv11.1 expressed in CHO cells.

To investigate the link between drug-binding and state-dependence we chose to use mutants of residue N588 located in the α-helix of the S5P linker of Kv11.1 (Figure 4.1). This residue has two important features: ﬁrst, it is though to be located distant to the drug-binding pocket (Yi et al., 2007); and second, it is possible to titrate the voltage- dependence of inactivation of the channels by introducing different charges at this residue (Clarke et al., 2006). Previous studies of N588 mutant channels in our laboratory were carried out using the Xenopus laevis oocyte expression system (Clarke et al., 2006); however, mammalian cells are the preferred heterologous expression system to use for drug-binding studies

61 Figure 4.2: Conductance voltage curves for N588E-Kv11.1 (), WT-Kv11.1 (), and N588K- Kv11.1 (). Data points are mean ± S.E.M and the curve ﬁtted to the data is a Boltzmann function with a V0.5 of inactivation of -114 ± 3 mV and slope factor of -19 mV (n = 13) for N588E- Kv11.1, V0.5 of -78 ± 2 mV and slope factor of -26 mV (n = 11) for WT-Kv11.1, and a V0.5 of 45 ± 4 mV and slope factor of -14 mV (n = 6) for N588K-Kv11.1.

(Cavalli et al., 2002). Therefore, we ﬁrst wanted to conﬁrm that the properties of N588 mutant channels were similar in mammalian cells and Xenopus laevis oocytes. When expressed in CHO cells, the V0.5 of inactivation for N588E-Kv11.1 was -114 ± 3mVwith a slope factor of -19 mV ± 1 mV (n = 13); the V0.5 of inactivation for WT-Kv11.1 was -78 ± 2mVwith a slope factor of -27 mV ± 2 mV (n = 11); and the V0.5 of inactivation for N588K-Kv11.1 was 45 ± 4mVwith a slope factor of -14 mV ± 1 mV (n = 6) (Figure 4.2). All values are similar to those reported for Xenopus oocytes (Clarke et al., 2006). The reversal potential for N588K-Kv11.1, -83.4 ± 2.0 mV (n = 5), was slightly shifted compared to that for WT-Kv11.1, -88.5 ± 2.2 mV (n = 10). This is very similar to previous reports (Cordeiro et al., 2005). The N588K-Kv11.1 channels are nevertheless still highly selective for K+ over Na+. The reversal potential in the N588E-Kv11.1 construct could not be determined due to the small magnitude of its activating currents. However, when expressed in oocytes with much larger currents it was found to be similar to WT-Kv11.1 (Clarke et al., 2006). These properties together make N588 mutant constructs ideal for the investigation of inactivation mediated drug-binding to Kv11.1.

4.2.2 High afﬁnity drug-binding is modulated by N588 charge mutants

We initially investigated the afﬁnity of four drugs, previously established to block Kv11.1 in the low nanomolar range — astemizole (Zhou et al., 1999), cisapride (Mohammad et al., 1997), dofetilide (Kiehn et al., 1996), and terfenadine (Yang et al., 2004) – for N588E-

62 Figure 4.3: Typical examples of current traces recorded from the drug-block voltage protocol illustrated at top. (A) – WT-Kv11.1, (B) – N588E-Kv11.1, and (C) – N588K-Kv11.1, in the presence of 30 nM cisapride. Dashed lines represent zero current in each case. Note the different scales on each recording; (A) and (B) are currents recorded during the revelatory voltage step, while (C) displays the non-rectifying current of the 3 s depolarising step in N588K-Kv11.1. The percentage of drug-block for each recording was determined by dividing the current measured at the end of the 3 s depolarising step to +20 mV after application of the drug (Idrug) by the current measured at the beginning of the recording (Icontrol). In the case of WT-Kv11.1 and N588E-Kv11.1, the current at this point was determined by ﬁtting an exponential curve to the ﬁrst part of the deactivating current during the revelatory voltage step and extrapolating this back to the end of the depolarising step, as illustrated in (A) and (B).

Figure 4.4: Hill Plots of high afﬁnity Kv11.1 blockers. Drug-binding to WT-Kv11.1 (), N588E- Kv11.1 (), and N588K-Kv11.1 () for (A) – astemizole; (B) – cisapride; (C) – dofetilide; and (D) – terfenadine. Each data point represents Idrug / Icontrol ± S.E.M for n = 4-9 cells. (E) – summary data showing the mean IC50 ± S.E.M for each drug. Signiﬁcant differences between WT-Kv11.1 and N588K-Kv11.1 are denoted by *. All values are printed in Table 4.2.3

64 Kv11.1, WT-Kv11.1 and N588K-Kv11.1 expressed in CHO cells. Figure 4.3 shows typical examples of WT-, N588E- and N588K-Kv11.1 traces under control conditions and after 5 min equilibration with 30 nM cisapride. Percentage of drug-block was measured at the end of the 3 s activating step to +20 mV in all cells. This was performed directly in N588K (Figure 4.3C), or else by fitting a single exponential curve to the first part of the current trace during the revelatory step in N588E-Kv11.1 (Figure 4.3B) and WT-Kv11.1 (Figure 4.3A), and extrapolating this back to the end of the activating step. The data in Figure 4.3 indicate that 30 nM cisapride caused less block of N588K-Kv11.1 channels compared to N588E- or WT-Kv11.1. This is more clearly seen from the summary Hill plots shown in Figure 4.4; cisapride affinity for WT-Kv11.1: 20.5 ± 2.2 nM (n = 4) was similar to that for N588E-Kv11.1: 13.1 ± 4.9 nM (n = 4) but significantly reduced for N588K-Kv11.1: 55.9 ± 4.2 nM (n = 4). All four high affinity blockers showed a similar pattern of reduced affinity for N588K- compared to WT- and N588E-Kv11.1 (see Figure 4.4). The affinities of all drugs for WT and N588 mutant constructs are summarised in Table 4.2.3.

4.2.3 Low afﬁnity drug-binding to N588 charge mutants

We next investigated the affinity of 4 drugs previously established to block Kv11.1 in the high nanomolar or micromolar range -– quinidine (Lees-Miller et al., 2000), perhexiline (Walker et al., 1999c), erythromycin (Volberg et al., 2002), and dl-sotalol (Kirsch et al., 2004) – for N588E-Kv11.1, WT-Kv11.1 and N588K-Kv11.1 constructs expressed in CHO cells. Typical examples of current traces recorded from WT-, N588K- and N588E-Kv11.1 in the presence and absence of 3 μM quinidine are illustrated in Figure 4.5. Quinidine caused a similar degree of block of WT-Kv11.1 and the two N588 charge mutants. This is also seen from the summary Hill plots (Figure 4.6A). Perhexiline and erythromycin showed the same pattern as that observed for quinidine, i.e. similar affinities for all Kv11.1 constructs (Figure 4.6A and C). In contrast, dl-sotalol showed a significantly higher affinity for N588E-Kv11.1 and WT-Kv11.1 compared to N588K-Kv11.1 (Figure 4.6D, Table 4.2.3).

4.2.4 Does reduced afﬁnity for N588K-Kv11.1 reﬂect state-dependent binding?

The data from Figures 4.3 and 4.4 clearly demonstrate that the four high affinity drugs used in this study had reduced affinity for the inactivation-deficient N588K-Kv11.1 channels. To determine whether this reduced affinity for N588K-Kv11.1 reflected state-dependence of drug-binding we investigated whether there was a similarly reduced affinity for a range of inactivation-deficient mutants. Specifically, we investigated binding of dofetilide to S631A-Kv11.1 and S620T-Kv11.1. S631A-Kv11.1 has a markedly right-shifted V0.5 of

Figure 4.5: Typical examples of current traces recorded from (A) WT-Kv11.1, (B) N588E-Kv11.1, and (C) N588K-Kv11.1 in the presence and absence of 3 μM quinidine. The drug-block voltage protocol is shown at top. Dashed lines represent zero current in each case. Note the 50% block of N588K-Kv11.1 by 3 μM quinidine which is similar to the percentage block in WT-Kv11.1 and N588E-Kv11.1. Rates of deactivation in N588E-Kv11.1 are faster than WT-Kv11.1 due to the use of a -120 mV ‘revelatory’ voltage step, in order to produce adequate opening of channels to measure current. The rate of deactivation appears to slow in (A) and (B) with application of drug; this is probably due to some drug unblocking during this step, rather than a true effect on the kinetics of deactivation.

66 Table 4.1: Collated IC50 values calculated from the Hill plots in Figure 4.4 and Figure 4.6. * denotes a significant difference from N588K-Kv11.1 to N588E-Kv11.1; † denotes a significant difference between N588K- and WT-Kv11.1. h is the Hill coefficient calculated from the curve of best fit to the data (see Figure 4.4 and 4.6). All IC50 values are mean ± S.E.M (n cells).

steady-state inactivation compared to WT-Kv11.1 that is very similar to that observed for N588K (Figure 4.7A) whereas S620T-Kv11.1 does not inactivate at measurable voltages (Ficker et al., 1998). Accordingly, at +20 mV, the proportion of channels in the open:inactivated states is ∼85:15 for N588K and S631A, compared to 100:0 for S620T, but 2:98 for WT-Kv11.1 (Figure 4.7A). The affinity of dofetilide for S631A-Kv11.1 was not statistically different to that for N588K-Kv11.1 (404 ± 95 nM (n = 4), and 377 ± 69 nM (n = 4) respectively), an 8-fold reduction compared to WT-Kv11.1. That these two mutants, with very similar effects on inactivation but apparently not located near each other, have very similar effects on drug- binding suggests that the reduced affinity for drug-binding is mediated by reduced inactivation of the channel. However, the affinity of dofetilide for S620T, was reduced a further 10 fold (3494 nM ± 374nM,n=4)compared to its affinity for S631A or N588K. Given that there is relatively little difference in the extent to which S631A and N588K channels occupy the open state at +20 mV (the voltage at which we assayed drug-binding) compared to S620T channels (i.e. ∼85% compared to 100%), a marked reduction in drug-affinity for S620T-Kv11.1 suggests a gating-independent effect on drug-binding by this mutant. An alternative hypothesis is that, despite the relative infrequency with which S631A and N588K channels occupy the inactivated state at +20 mV, the kinetics of drug-binding and unbinding are such that whenever the channel enters the inactivated state it binds drug that, with a very slow off rate, remains bound for a long period. According to this hypothesis, binding of drug to the S620T mutant would only encounter the open state and so

Figure 4.6: Hill plots of low afﬁnity Kv11.1 blockers. Drug-binding to WT-Kv11.1 (), N588E- Kv11.1 (), and N588K-Kv11.1 () for (A) – quinidine; (B) – perhexiline; (C) – erythromycin; and (D) – dl-sotalol. Each data point represents Idrug / Icontrol ± S.E.M for n = 4-6. (E) – summary data showing the mean IC50 ± S.E.M for each drug. Signiﬁcant differences between WT-Kv11.1 and N588K-Kv11.1 are denoted by *. All values are printed in Table 4.2.3

68 reflect the affinity for the open state, whereas binding to WT or N588K channels would reflect a weighted average of the affinity for the inactivated and open states dependent on the relative rates of transitions between the two states and drug-binding and unbinding rates. To test this hypothesis we set up a computer model of drug-binding to Kv11.1 channels as depicted in Figure 4.8.

Figure 4.7: (A) – conductance voltage curves for S631A-Kv11.1(), and S620T-Kv11.1(2). Data points are mean ± S.E.M and the curve fitted to the data points is a Boltzmann function. The V0.5 of steady-state inactivation for the S631A-Kv11.1 mutant is 43 ± 2 mV with a slope factor of - 16.5 mV (n = 6). S620T-Kv11.1 does not inactivate in the voltage range shown data from (Ficker et al., 1998). Dotted, plain, and dashed lines show data for N588E-Kv11.1, WT-Kv11.1, and N588K-Kv11.1, reproduced from Figure 4.2. (B) – affinity of dofetilide for S631A-Kv11.1 (), and S620T-Kv11.1 (2). Dotted, plain, and dashed lines show data for N588E-Kv11.1, WT-Kv11.1, and N588K-Kv11.1 respectively, reproduced from Figure 4.4C. Note the similar drug affinity for the N588K and S631A cell constructs, and the marked right-shift (reduced affinity) of dofetilide for the S620T construct.

4.2.5 Modelling kinetics of drug-binding to open and inactivated states

The Markov chain model for Kv11.1 kinetics is based on that developed by Lu et al. (Lu et al., 2001) with the addition of two states: drug-bound open state and drug-bound inactivated state (Figure 4.8). The rate constants from Lu et al., (2001), were scaled to 22 ◦C. Drug-binding rate constants are presented in Table 4.2. To model the S620T-

69 Figure 4.8: Markov model of drug-binding to Kv11.1. Cx = closed states. O = open state. I = inactivated state. OD = drug bound to open state. ID = drug bound to inactivated state. Greyed out portions of the model were not altered during modelling simulations.

Kv11.1 mutant the rate constants for transition steps to the inactivated state were set to 0. To model the N588K mutant, the rate constants for the transition between the open and inactivated states were adjusted to reproduce the experimentally observed changes in these rate constants and steady-state inactivation. For simplicity, we assumed that the kinetics of activation were the same for WT-, S620T- and N588K-Kv11.1 channels. To calculate the rate constants for open and inactivated drug blocked state we assumed that drug affinity for S620T represented the affinity for the open state. The rate constants for open state drug-block, kf6 and kb6, were constrained to fit the data for drug-binding to the S620T-Kv11.1 channels. The values for kf6 and kb6 were then held constant and the values for drug-binding to the inactivated state, kf7 and kb7, constrained to fit the time course of drug-block for WT channels. In the case of dofetilide binding, the calculated affinity for the open state (kf6/kb6) was 3.5 μM, i.e. the value for S620T, and the calculated affinity for the inactivated state (kf7/kb7) was 47.8 nM. The measured affinity for dofetilide binding to WT-Kv11.1 channels, 50.1 nM, is much closer to the calculated value for the affinity to the inactivated state reflecting both the greater proportion of time WT-Kv11.1 channels spend in the inactivated state and the slower dissociation of drug from the inactivated state. If the original assumption was correct, i.e. drug affinity for S620T-Kv11.1 is the true affinity of the drug for the open state, then substitution of the values for kf6, kb6, kf7 and kb7 into our model for N588K should reproduce the experimentally determined IC50 for dofetilide binding to N588K. As can be seen in Figure 4.9, the model predicted values are very close to the experimental data. The same procedure

astemizole cisapride dofetilide dl-sotalol terfenadine kf kb kf kb kf kb kf kb kf kb 6 6 6 6 6 O-OD 5x10 0.099 5 x 10 0.7884 5 x 10 17.55 5 x 10 11138 5 x 10 1.363 6 6 5 6 6 I-ID 5x10 0.02425 5 x 10 0.09697 1 x 10 0.00478 5 x 10 2450 5 x 10 0.292 I:O 4.1 8.1 73.4 4.5 4.7

Table 4.2: Rate constants for drug-binding to the open and inactivated states of Kv11.1

70 drug concentration (nM) astemizole 19.8 ± 3.7 (n = 10) cisapride 157.1 ± 20.2 (n = 6) dofetilide 3494 ± 425.0 (n = 4) dl-sotalol 2218 ± 380.2 (n = 5) terfenadine 271.8 ± 120.3 (n = 4)

Table 4.3: Drug afﬁnities for S620T (IC50) was repeated for each of the state-dependent drugs formerly assessed (Figure 4.10A-D) with similarly good approximations to the experimental data.

Figure 4.9: Dofetilide Hill plots with data points for N588K-Kv11.1 and the modelling curve superimposed. Thin plain line, dotted line, and dashed line show WT-, N588K-, and S620T-Kv11.1 respectively. The modelled line is thick black. Experimental IC50 values for dofetilide block of N588K, and the modelled values respectively: 377 ± 57.8 vs 302 nM.

71 AB

C D

Figure 4.10: Hill plots for (A) – astemizole, (B) – cisapride, (C) – terfenadine, and (D) – dl- sotalol with data points for N588K-Kv11.1 and the modelling curves superimposed. Thin plain line, dotted line, and dashed line show WT-, N588K-, and S620T-Kv11.1 respectively. Modelled lines are thick black. Experimental IC50 values for drug-block of N588K-Kv11.1 and the modelled values respectively: astemizole – 14.8 ± 1.1 vs 13.6 nM; cisapride – 55.9 ± 4.2 vs 77.3 nM; terfenadine – 170 ± 32.3 ± 177 nM; dl-sotalol – 1392 ± 266.5 vs 1461 μM. Drug IC50 values for S620T-Kv11.1 are presented in Table 4.3; Hill plots for S620T-Kv11.1 are presented in Figure 4.11

Figure 4.11: Hill plots for drug-binding to S620T-Kv11.1. (A) – astemizole, (B) – cisapride, (C)– dl-sotalol, (D) – terfenadine. Summary IC50 values are in Table 4.3.

73 4.3 Discussion

4.3.1 State-dependence of drug-binding to Kv11.1

Most drugs that block Kv11.1 require channel opening (Kiehn et al., 1996). Some evidence suggests that, once activated, inactivation increases drug affinity for the channel: first, mutant-Kv11.1 channels with disrupted inactivation (G628C/S631C, S620T, and S631A) reduce drug-block by multiple agents (Ficker et al., 1998; Lees-Miller et al., 2000; Numaguchi et al., 2000; Yang et al., 2004); second, mutations introduced into Bovine- EAG (bEAG) and hEAG1 that enable inactivation also confer sensitivity to dofetilide block (Ficker et al., 2001; Gomez-Varela´ et al., 2006). However, the inactivation disrupting mutations could affect drug-block through gating independent means. S631 and S620 lie proximate to the drug-binding space, and mutations at these positions may produce conformational changes at the base of the pore helix – an important molecular determinant of drug-binding (Mitcheson et al., 2000). G628C/S631C markedly reduces the K+ selectivity of Kv11.1 (Smith et al., 1996), suggesting a conformational change in the vicin- ity of the selectivity filter. Further, mutant-Kv11.1 channels G648A and T623A promote inactivation, but reduce methanesulfonanilide block of Kv11.1 (Mitcheson et al., 2000). In addition, voltage protocols designed to favour occupancy of the inactivated state during drug-binding (i.e. large depolarizations) also relieve drug-block (Kiehn et al., 1996; Numaguchi et al., 2000). The data reported in this study provides strong evidence that drug-binding to Kv11.1 K+ channels is influenced by whether the channels are in the inactivated or the open state. In addition, we have demonstrated that this phenomenon is drug-dependent, with the ratio of affinities for the open to inactivated state varying from 1:1 (i.e. not state- dependent: quinidine, erythromycin, perhexiline) to 1:70 (dofetilide). First, we have shown that a mutation, N588K, introduced into a region thought to be remote from the drug-binding pocket (Yi et al., 2007) that shifts the voltage-dependence of inactivation (Brugada et al., 2004; Clarke et al., 2006) affects drug-binding. A combination of structural studies (Torres et al., 2003), toxin binding studies (Pardo-Lopez et al., 2002) and molecular dynamics modelling studies (Yi et al., 2007) suggest that N588 on the hydrophilic surface of the S5P α-helix faces the extracellular space and is therefore remote from the drug-binding pocket. Second, we have shown that two distinct mutants that have very similar effects on inactivation (N588K and S631A) have similar effects on drug-binding. The simplest explanation is that it is the effect these mutations have on inactivation that accounts for the altered drug-binding. Third, our kinetic modelling of drug- binding to WT, S620T and N588K mutant channels demonstrates that all the differences in drug-binding between these mutants can be explained on the basis of differences in

74 occupancy of open and inactivated states and the kinetics of drug-binding. Our kinetic modelling studies have also enabled, for the first time, estimation of the binding affinities for both the open and inactivated states for a range of drugs. Stork et al. (2008) have recently published data extending the concept of state dependence to the dissociation of drugs from the Kv11.1 channel. As they have elegantly shown, some drugs require opening of the activation gate to dissociate from the inner cavity of the channel. Our data is complementary to their own, demonstrating that the rate of drug dissociation will depend not only on the relative proportions of activated and closed channels at a given voltage, but also the proportion of activated channels in the inactivated or open state, itself a function of voltage. In contrast with our data, there is one report in the literature that suggests mutations to S620 could have a gating-independent effect on drug-binding (Guo et al., 2006). Guo and colleagues showed that S620T and S620C both abolished inactivation gating but had different affinities for E-4031 – a methanesulfonanilide similar to dofetilide. This data, however, is not necessarily incompatible with our results. It is possible that the abolition of inactivation alters drug affinity in both the S620T and S620C. However, whereas the S620T represents the true affinity for the open state, the cysteine side chain in S620C, is able to bind to the drug and thereby increase the affinity relative to that for S620T.

4.3.2 What explains the strong preference for inactivated state binding?

The molecular determinants of drug-binding to Kv11.1 include two aromatic residues in the S6 helix, F656 and Y652; and to a variable extent: three residues close to the selectivity filter: T623, S624, V625, and two residues in S6: G648, and V659 (Figure 4.1). In the absence of a high-resolution structure of Kv11.1, the exact conformation of these residues in relation to each other cannot be determined. It has been suggested that inactivation of Kv11.1 channels involves changes in the structure of the pore region (Chen et al., 2002),and so the spatial relationships of the different components of the drug- binding pocket are likely to vary between the open and inactivated state. One can imagine two scenarios that could arise. First, a conformational change around the selectivity filter and base of the pore-helix secondary to inactivation could result in a reorientation of the drug-binding residues at the base of the selectivity filter relative to the drug-binding residues on S6. Second, inactivation could involve reorientation of the S6 helixes, relative to their positions in the open state. In this hypothesis not only would the relationship between the drug-binding residues on S6, Y652 and F656, alter with respect to drug- binding residues at the base of the pore-helixes (Chen et al., 2002; Lees-Miller et al., 2000), but also the inter-subunit relationships between Y652 and F656 would change (Myokai et al., 2008). Evidence in support of a primary role for reorientation of S6 in

75 favouring inactivated state drug-binding comes from a study investigating the effect of mutating F656 to methionine on the binding of droperidol (Luo et al., 2008). The affinity of droperidol for S631A-Kv11.1 was reduced compared to WT-Kv11.1, but introduction of S631A into the background of the F656M mutant did not reduce the affinity compared to F656M-Kv11.1 alone. The reduction in affinity of dofetilide for S620T-Kv11.1 compared to WT-Kv11.1 (70- fold) is comparable to the reduction in drug-affinities seen when residues Y652 or F656 are mutated (Mitcheson et al., 2000; Lees-Miller et al., 2000) This suggests that abolition of inactivation results in the complete elimination of one of the interactions between dofetilide and the channel. However, for astemizole, cisapride, sotalol and terfenadine the reduction in affinity is more modest suggesting an interaction has been reduced but not eliminated.

4.3.3 Inactivated state binding is necessary but not sufﬁcient for high afﬁn- ity binding

All the high affinity blockers tested in this work showed a marked preference for the inactivated state. At depolarised potentials, WT-Kv11.1 channels predominantly reside in the inactivated state therefore it is unsurprising that the drugs with the highest affinity for the inactivated state show the highest affinity for WT-Kv11.1. However, our data for dl-sotalol clearly indicates that state-dependence of binding is not sufficient to produce high affinity binding. One plausible explanation for the low affinity yet state-dependence of dl-sotalol binding is that it binds to the residues most critical for state-dependent binding but does not bind to any other residues whereas the higher affinity state-dependent blockers bind both to the critical state-dependent residues as well as others. If this hypothesis is correct, then determining the molecular basis of sotalol binding to Kv11.1 would be a useful probe for determining the minimum requirements for state-dependent binding to Kv11.1.

4.3.4 Relevance for drug-binding in SQTS

The most common form of short QT syndrome (SQTS) results from the N588K mutation in Kv11.1 (Brugada et al., 2004). The wide spectrum of drugs known to block Kv11.1 provides multiple candidates for treatment. However, initial testing demonstrated that d- sotalol failed to prolong the QT interval (Brugada et al., 2004). Of tested candidates, only quinidine (Gaita et al., 2004), disopyramide (McPate et al., 2006), and doxepin (Duncan et al., 2007) have been shown to block N588K-Kv11.1 at affinities similar to WT-Kv11.1. Of significance, all block WT-Kv11.1 in the micromolar (low affinity) range. Of note, al- though the binding affinity of astemizole for N588K-Kv11.1 is reduced compared to WT-

76 Kv11.1, its afﬁnity for N588K-Kv11.1 is 250 fold greater than quinidine. Combined with a benign side-effect proﬁle, it is a good candidate for evaluation as a treatment for SQTS type 1.

4.3.5 Relevance for high-throughput assays

Given the mandated need to screen all drugs for Kv11.1 binding, there has been considerable effort put into developing high-throughput screens for assaying drug-binding to Kv11.1 (Dorn et al., 2005; Tang et al., 2001). In general, however, the results of these screens have been poor and we suggest that this may be because they predominantly assay binding to the open state and therefore underestimate the afﬁnity of drugs that preferentially bind the inactivated state. Given that the difference in afﬁnity between the open and inactivated states can be 70-fold, it is important that any high-throughput screening system must assay binding to the inactivated state.

77 Chapter 5

Drug-Kv11.1 Kinetics and Risk of Acquired Long QT Syndrome

5.1 Background

Measured Kv11.1-drug affinity is a core component of regulatory guidelines for the preclinical investigation of a drug’s propensity to prolong the QT interval (ICH S7A and S7B, http://www.ich.org). An integrated approach to pre-clinical testing identifies most compounds with proarrhythmic potential (Wallis, 2009). The failings of this strategy, when they occur, are largely due to the use of surrogate markers of arrhythmic risk with less than perfect predictive value. For example, a drug may cause significant QT interval prolongation yet have a low risk for arrhythmia, such as amiodarone (Yang et al., 2001). The potential therefore exists for a small but significant number of useful, and ultimately safe, medications to be withdrawn from development for failing pre-clinical QT studies. To avoid this possibility, complementary measures of arrhythmia risk have been proposed, foremost among them is TRIaD (Triangulation, Reverse use-dependence, Instability, and Dispersion)(Hondeghem et al., 2003; Hondeghem, 2007). TRIaD involves examination of what phase of the cardiac action potential is responsible for QT prolongation (Triangulation); under what conditions i.e. high or low heart rates (HR), the QT will prolong (Reverse use-dependence); whether the drug provokes beat to beat changes in action potential duration (Instability); and whether the drug increases differences in repolarization times across the heart (Dispersion). These markers, when present, predict with accuracy the likelihood that a drug will be associated with a significant risk of arrhythmia (Hondeghem et al., 2003). The origin, however, of these repolarization abnormalities is largely unexplored. Are they simply in vivo manifestations of

78 Kv11.1 drug-block, or the product of hitherto unexplored drug effects? It is now established that drug-block of Kv11.1 is time-, voltage-(Kamiya et al., 2008) and state-dependent (Perrin et al., 2008) (Chapter 4). Further, these properties may vary substantially from drug to drug (Perrin et al., 2008) (Chapter 4). In this chapter, we examine the hypothesis that the differences in drug effects on the principal pre-clinical markers of arrhythmic risk as represented by TRIaD can be explained by differences in the time-, voltage-, and state-dependence of drug-binding to Kv11.1. Here, we present Markov-state models for cisapride and dofetilide block of Kv11.1 that recapitulate their principal Kv11.1-binding characteristics. The Markov descriptions are placed within a modified ten Tusscher-Panfilov (mTTP06) model of human ventricular tissue (ten Tusscher and Panfilov, 2006) in order to investigate the effect of different drug-binding kinetics on cardiac action potential repolarization. The complete model is then used to examine each of the individual components of TRIaD, revealing important and fundamental differences between the two drugs. Lastly, our modified TTP06 model is used in cable simulations of the left ventricular wall and ‘pseudo-electrocardiograms’ recorded demonstrating that a fundamental knowledge of the time-, voltage- and state- dependence of drug-binding to Kv11.1 enables the full range of arrhythmic risk markers to be reproduced in silico.

5.2 Methods

5.2.1 Markov state models of cisapride- and dofetilide-binding to Kv11.1

The ﬁve-state Markov state model of Kv11.1 (Mazhari et al., 2001) formed the backbone of our drug-binding models. As previously described in Chapter 4, cisapride block of the channel was modelled by the addition of two states: an open state bound to drug (OD), and an inactivated state bound to drug (InD) (Section 4.2). Evidence suggests that the methanesulfonanilides are trapped by the activation gate of Kv11.1 at negative membrane potentials (Mitcheson et al., 2000; Stork et al., 2007). In an effort to improve the accuracy of our modelling simulations for this more intricate work, dofetilide, therefore, was modelled by the addition of three closed-drug states (C1D, C2D, and C3D) with free movement of the channel through all states while bound to drug (Figure 5.1) – compare to model presented in Section 4.2. The determination of rate and drug-binding parameters is illustrated in Figure 5.1. In brief, for cisapride: the ﬁve-state Markov model was constrained to current data from a voltage protocol that interrogated channel activation, inactivation, recovery from inactivation, and deactivation of Kv11.1 with non-linear least squared optimisation performed

79 Markov-state models voltage protocols experimental data

+ 40 mV 3 s C1 C2 C3 O 1 Ͳ Ͳ Ͳ

InͲ Ͳ 4 s - 65 mV - 125 mV 0.5 nA 0.5 s

control S620T-hERG + 20 mV C C C O * 2 1 Ͳ 2 Ͳ 3 Ͳ ͲOD 3 s // 0.1 Hz - 80 mV - 80 mV 1 nA - 110 mV 500 ms

WT-hERG + 20 mV 3 C C C O OD * 1 Ͳ 2 Ͳ 3 Ͳ Ͳ 3 s // 0.1 Hz

Ͳ Ͳ InD * In Ͳ - 80 mV 0.5 nA * - 80 mV - 110 mV control 100 ms

A cisapride model S620T-cisapride simulation cisparide-WT simulation

control

C1 Ͳ C2 Ͳ C3 Ͳ O Ͳ OD Ͳ * * InͲͲ InD 500 ms * 100 ms * control

B dofetilide model dofetilide-WT raw tracing dofetilide-WT simulation

C1 C1D * * Ͳ

C2 C2DͲ

Ͳ Ͳ

C3 Ͳ O Ͳ OD ͲͲ C3 100 ms 100 ms Ͳ Ͳ In Ͳ InDͲ control control

Figure 5.1: Model fitting to experimental data. Intermediate and final Markov-state models employed in the fitting scheme for cisapride (left column). Voltage protocols used to obtain experimental current data from S620T-Kv11.1 and WT-Kv11.1 (middle column). Raw experimental data from voltage-clamp experiments (right column). The portions of the voltage protocol outlined in red are responsible for the displayed current data. Step 1: the rate parameters (unknown transitions in red) of a five-state model of Kv11.13 (Mazhari et al., 2001) were constrained to current data resulting from a voltage protocol that interrogated Kv11.1 activation, deactivation, inactivation, and recovery from inactivation. Step 2: the inactivated state was then removed from the Markov-state model, and a new open-drug (OD) state appended; the rate parameters describing channel activation and deactivation were retained, and the new rates for drug-binding and unbinding from the open state were fit to current data from drug-binding experiments in the non-inactivating S620T- Kv11.1 mutant. Step 3: the inactivated state was returned to the model, including relevant rate parameters from step 1, and the inactivated-drug state appended (InD); the final unknown rate parameters were fit to current traces of cisapride block of WT-Kv11.1. The final model – (A)– reproduced drug-binding experiments to S620T-Kv11.1 (100 nM cisapride) and WT-Kv11.1 (3, 10 and 30 nM cisapride). Experimental evidence suggests that methanesulfonanilides may be trapped in the channel pore by the activation gate (Mitcheson et al., 2000; Stork et al., 2007). To model this process, three closed states bound to drug (C3D, C2D, and C1D) were added to the model – (B). The channel is able to transit through all drug-bound states. The rate parameters for each drug-bound transition mirrors that of the corresponding transition in the absence of drug i.e. rate constants for C1 C2 are equal to those for C1D C2D. (B) (middle and right columns) – simulated 1 μM dofetilide-binding to Kv11.1. Rate constants for both models are given in Figure 5.2 80 cisapride kb

123o

kfo kb i kfi dofetilide

1 2 3 kfo 3 2 1

kb i kfi A

transition Gα zα Gβ zβ

C1 - C2 69523 2.69 76138 2.48 (C1D-C2D) C2 - C3 56175 7.6 x10-9 67718 1.3 x 10-9 (C2D-C3D) C3 - O 68446 0.01 72630 0.75 (C3D-OD) C3 - In 69774 0.44 83506 0.17 (C3D-InD) O - In 58962 0.67 68246 0.78 (OD-InD) B

kfo kbo kfi kbi

cisapride 1 x 106 s-1 0.048 s-1 2.5 x 106 s-1 0.016 s-1

dofetilide 0.025 x 106 s-1 0.028 s-1 0.1 x 106 s-1 0.003 s-1

Figure 5.2: Markov-state models for cisapride and dofetilide block of Kv11.1. Rate constants for each gating transition are calculated from Equation 5.1 and 5.2, according to the relevant values of Table A. The rate constants for drug-binding to the open and inactivated states are displayed in Table B, where kfo is the rate constant for drug-binding to the open state, kbo – rate constant for drug unbinding from the open state, kfi – rate constant for drug-binding to the inactivated state, and kbi – rate constant for drug unbinding from the inactivated state.

81 using a curvilinear search method (Dokos and Lovell, 2004). Next, the inactivated states were removed from the Markov model, and a new open-drug (OD) state appended; the rate parameters describing channel activation and deactivation were retained, and the new rates for drug-binding to and unbinding from the open state were fit to current data from cisapride block of the non-inactivating S620T-Kv11.1 mutant. In a final step, the inactivated state was returned to the model, including relevant rate parameters from the first step, and the inactivated-drug state appended (InD). The final unknown rate parameters (binding to, and dissociating from, the open state) were fit to current traces of cisapride block of wild-type Kv11.1. The fitting routine for dofetilide varied only by inclusion of drug trapping in the three closed states (Figure 5.1B); rate transitions for transitions through drug bound states mirrored those determined for non-drug bound states. Each forward rate constant was defined thus:

k T VFzα−Gα k = B e RT (5.1) f h

and each backward rate constant:

k T VFzβ −Gβ k = B e RT (5.2) b h

where kB is Boltzmann’s constant, T is the absolute temperature, h is Planck’s constant, R is the gas constant, V is the membrane voltage, F is Faraday’s constant, and zα/β and Gα/β are the voltage-dependent and independent components respectively (Fig- ure 5.2, Table A). In action potential experiments, all gating rate constants were scaled by aQ10 value of 2.8 to account for the difference between room temperature experiments, and simulated recordings at 37 ◦C.

5.2.2 Voltage protocols employed in modelling experiments

5.2.2.1 Recovery from inactivation of calcium Channels

A double-pulse voltage protocol was employed to measure recovery from inactivation of calcium channels following Zygmunt and colleagues (Zygmunt et al., 1997) – Figure 5.3.

5.2.2.2 Use-, frequency-, and voltage-dependence

Use-, frequency, and voltage-dependence were measured using voltage protocols adopted from Walker et al. (1999b) and Stork et al. (2008) – Figure 5.4.

82 20 time (s) 1. 1.2 1.4 1.6 1.8 20 40

voltage (mV)voltage 60 80

Figure 5.3: Voltage protocol used to record recovery from inactivation of L-type calcium channels. From a holding potential of -80 mV the membrane voltage was stepped to -45 mV for 25 ms, followed by +25 mV for 150 ms (P1) to promote inactivation of the channel. P2 (-40 mV), was 20 ms duration in the 1st sweep, and incremented by 20 ms on each subsequent sweep. The anti-peak currents of a ﬁnal step to +25 mV were plotted against the duration of P2 and ﬁt with a single exponential curve to give the time constant of recovery from inactivation (Zygmunt et al., 1997).

A +35 mV 300 ms -55 mV 5 min 300 ms -80 mV

BC +35 mV 3 s 300 ms 0.1 -55 mV 5 s 0.3 Hz 300 ms -80 mV -80 mV 0.6 1

Figure 5.4: Voltage protocols for use- (A), frequency- (B), and voltage-dependence (C) of Markov state models of Kv11.1. To determine use-dependence, – (A) – tail currents were recorded at -55 mV after a 300 ms depolarising pulse to + 35 mV; drug was then perfused while the membrane voltage was held at -80 mV; after 5 min at -80 mV, the tail currents were again recorded and compared to the control values. To assess frequency-dependence of channel block, – (B)–atwo- pulse protocol was delivered at frequencies of 0.1, 0.3, 0.6 and 1 Hz; percentage drug-block was measured at steady-state from the tail-current (-55 mV) (Stork et al., 2007). Voltage-dependent drug-block – (C) – was measured at steady-state from the tail-current resulting from a -55 mV voltage step (P2); each experiment employed, in turn, a different test voltage for P1 ranging from +35 mV to -35 mV (Walker et al., 1999b).

83 5.2.3 Ventricular cell modelling

The ten Tusscher-Panﬁlov ventricular cell model of 2006 (TTP06) is the most detailed and complete description of human ventricular electrical activity available (ten Tusscher and Panﬁlov, 2006). We made three changes to this model to provide more complete descriptions of those aspects of ventricular cell electrical activity that pertain particularly to the proposed mechanisms of arrhythmogenesis in the long QT syndrome.

First, IKr is described in TTP06 by a Hodgkin Huxley (HH) formulation of channel gating that cannot replicate the complex kinetic transitions of Kv11.1-drug-binding experiments. We therefore replaced IKr with Markov models of Kv11.1 drug-block that allowed full expression of channel kinetics. Second, in 2008, Guo and colleagues showed that the ratio of the time constant for recovery from inactivation (τrec) for the L-type calcium current (ICa−L) and the duration of the action potential (APD) at 90% repolarization was ∼0.5 across species (Guo et al., 2008). Delayed recovery of ICa−L prevents reactivation of ICa−L in the setting of delayed repolarization, thus preventing early after depolarizations (EADs) — an important factor in drug-induced arrhythmia (Antzelevitch and Sicouri, 1994). Measurement of ICa−L recovery in TTP06 using the double-pulse protocol of Guo and colleagues, gave a τrec of 452 ms and ratio of τrec:APD90 of 1.5, far in excess of the normal range. We therefore modiﬁed the ratio by simply scaling the time constants describing recovery from ICa−L inactivation by a constant factor (3.6) at voltages ≤ 0 mV to produce a time constant of recovery from inactivation of ∼150 ms in the double-pulse protocol, and thus a ﬁnal ratio of 0.5 in our model.

Last, we incorporated a new description of the slow-delayed rectifier current (IKs). Experimental evidence suggests that IKs activation is best fit by a triexponential curve (Terrenoire et al., 2005). Further, evidence from the same work supports the hypothesis that the accumulation of IKs channels in the open state is a significant contributor to shortening of the action potential duration at high heart rates (Jurkiewicz and Sanguinetti, 1993). As the interplay between IKs and IKr is likely to be important in drug-induced arrhythmia, the IKs model of Terrenoire and colleagues was substituted for that in the original TTP06. It may be noted in Figure 5.6 that IKs current accumulates (increases) at high heart rates with the new model, and declines slightly in the TTP06 model. In summary, the modified model includes three significant changes to TTP06: a Markov model of state-dependent drug-binding to Kv11.1 has replaced the Hodgkin- Huxley (HH) IKr formulation; the ratio of ICaL τrec to APD90 has been scaled to 0.5 and the HH formulation of IKs has been replaced with a HH model incorporating a third activation gate and open channel accumulation at high heart rates.

84 A B time (ms) 1. 3400 3800 4200 4600 0.8 10 I/I max 0.6 20 0.4 30 0.2

current (pA) current τrec = 452 ms 40 0 250 500 750 1000 time (ms) CD time (ms) 1. 3400 3800 4200 4600 10 0.8 I/I max 0.6 20 0.4 30

current (pA) current 0.2 τ = 154 ms 40 rec 0 250 500 750 1000 time (ms)

Figure 5.5: Time constant of recovery (τrec) from inactivation for ICa−L for the original ten Tusscher-Panfilov (TTP06) (A – B), and our modified model (C – D). A double pulse protocol, see Guo et al., (2008) and Subsection 5.2.2.1, was used to record current traces in (A) and (C) (Terrenoire et al., 2005). A single exponential was fit to the envelope of peak current values to determine τrec. The time constants describing ICa−L recovery in TTP06 (tauf and tauf2) were scaled by a factor of 3.6 to give a τrec:APD90 (epicardial) ratio of 0.5

85 I I m Ks TTP06 Ks TTP06 1000 ms current (pA/pF) current (pA/pF) current time (s) time (s)

600 ms current (pA/pF) current (pA/pF) current time (s) time (s)

400 ms current (pA/pF) current (pA/pF) current time (s) time (s)

Figure 5.6: Accumulation of IKs current in the modiﬁed ten Tusscher-Panﬁlov model (mTTP06). At short cycle lengths (CL), there is no accumulation of IKs current in the original TTP06 model due to rapid deactivation (left column). Slow deactivation in the model of Terrenoire and colleagues (Terrenoire et al., 2005) results in an accumulation of IKs current at short CL (right column).

86 After scaling, IKs and IKr conductance to an epicardial APD of 304 ms, the model reproduced the general shape of action potentials from different areas of the heart, the relative APD90 durations between different cell types, and displayed similar APD restitu- tion properties to that described in the original paper (slope 1.1). Thus we increased the ‘resolution’ of the model in regard to proposed mechanisms of drug-induced arrhythmia, but maintained its fundamental properties.

5.2.4 Cable simulations

A 1D transmural fibre consisting of endocardial, mid-myocardial, and epicardial ‘cells’ in a 30:35:30 ratio (Gima and Rudy, 2002) was developed with membrane dynamics described by our modified ten Tusscher-Panfilov model of human ventricular tissue (ten Tusscher and Panfilov, 2006). The fibre was considered to be within an extracellular medium of very high conductance (Plonsey and Barr, 2007), and was described thus:

∂Vm(x, t) 1 = (Imem − Iion) (5.3) ∂t Cm

where

∂V (x, t) 1 I = m × (5.4) mem ∂x2 ρS

−2 and where ρ (cellular resistivity) = 450 Ωcm, Cm (cell capacitance) = 2 μFcm , and S (surface to volume ratio) = 0.2 μm−1 to give a “diffusion” co-efﬁcient (D):

1 D = (5.5) ρSCm of 0.556 × 10−3 cm2 ms−1. The conduction velocity was 0.45 ms−1, which is similar to experimental recordings across transmural wedge preparations (Yan et al., 1998a). The pseudo-ECG was recorded from an electrode 2 cm from the epicardial edge, excluding 1% of edge cells due to boundary effects, and calculated using the equation: x 1 φe = k −∇Vm.∇ dx (5.6) 0 r where ∇Vm is the spatial gradient of membrane voltage, r is the Euclidean distance from a source point to the electrode point, and k is a scalar approximating:

a2σ k = i 4σe

87 where σi is the intra-cellular conductivity, σe is the extracellular conductivity, and a is the radius of the ﬁbre. The ﬁbre was paced from the endocardial side.

5.3 Results

5.3.1 Markov-state models of Kv11.1 drug-block reproduce voltage-clamp experimental data

Cisapride binding to Kv11.1 exhibits use-1, frequency-, and marked voltage-dependence (Walker et al., 1999b). Frequency-dependence implies that a drug can dissociate from the channel at negative membrane voltages. This property has been attributed to dissociation from the closed state (Stork et al., 2007). We trialled Markov-state models incorporating dissociation from a closed state but no set of rate parameters could be found which reproduced all the experimental data. However, a simpler model (Figure 5.1A), that required drug dissociation from the open channel before channel closure, and included unequal affinities for the open and inactivated states (Perrin et al., 2008) was able to reproduce all the experimental data including frequency-dependence and voltage- dependence of channel block (Figure 5.7). It will be noted that no specific function describing voltage-dependence was incorporated within the model (Figure 5.2). Instead, voltage-dependence arose indirectly from the different distribution of channel states at each voltage. Dofetilide binding to Kv11.1 is use-dependent, frequency-independent, and minimally voltage-dependent (Tsujimae et al., 2004; Carmeliet, 1992). These properties were all reproduced by a Markov-state model that incorporated unequal affinities for the open and inactivated states (∼1:40, open:inactivated) and drug-trapping in the closed states (Figure 5.1B). Dofetilide has been reported as voltage-dependent (Carmeliet, 1992) and -independent (Tsujimae et al., 2004). However, in the former, the voltage-dependence is small (27% change from -20 mV to +10 mV) and in the latter a trend to reduced binding at negative voltages may be appreciated (see Figure 4 of Tsujimae et al. (2004)); it is therefore likely that dofetilide demonstrates a small amount of voltage-dependence as

1There is some confusion in the literature between the terms describing the characteristics of drug-block in voltage-clamp studies and those in action potential experiments. Frequency-dependence in voltage-clamp studies describes a drug property of increased channel block at higher frequency voltage pulsing. Use- dependence in voltage-clamp studies suggests that drug-binding may only occur during channel activation (opening). In action potential studies, reverse use-dependence (RUD) (also called reverse rate-dependence (RRD), or reverse frequency-dependence) describes the phenomenon of increasing action potential prolongation at low heart rates compared to high, at a constant drug concentration, and is therefore essentially unrelated to the use-dependence of voltage-clamp studies though using similar terms. Dofetilide, for example, is said to be use- but not frequency-dependent in voltage clamp studies, but display reverse use-dependence in action potential experiments.

88 use-dependence frequency-dependence voltage-dependence A block block 5 min 1 at -80 mV 1 current 0.8 0.1 Hz 0.8 0.6 0.6 0.4 0.4 1 Hz +35 mV 300 ms 0.2 0.2 voltage -55 mV 300 ms 0 0 30 60 90 35 15 5 25 -80 mV pulse number voltage (mV) B 5 min block block at -80 mV current 1 1 0.8 0.8 0.6 0.6 0.1 Hz 0.4 0.4 +35 mV 300 ms 0.2 1 Hz 0.2 voltage -55 mV 300 ms 0 0 50 100 150 35 15 5 25 -80 mV pulse number voltage (mV)

Figure 5.7: Markov-state models of cisapride and dofetilide binding to Kv11.1 reproduce voltage patch-clamp experimental data. The results are from simulated experiments. Use-, frequency-, and voltage-dependence of cisapride block (upper panels) and dofetilide block (lower panels) of Kv11.1. 5 min after drug application with membrane voltage held at -80 mV, channel block was near 0% in both (A) – cisapride, and (B) – dofetilide simulations (left column), modelling the re- quirement for activation gate opening before channel block. Frequency-dependence of channel block was tested at 0.1 (black), 0.3 (purple), 0.6 (blue) and 1 Hz (red) (centre column). Cis- apride block of Kv11.1 was markedly frequency-dependent, and dofetilide block of Kv11.1 was frequency-independent. Voltage-dependence of channel block was tested with a two-pulse voltage protocol, with an initial depolarising pulse between -45 and +35 mV at a concentration predicted to block 60% of current at maximum effect (right column). Cisapride block was markedly voltage-dependent with 16% block at -35 mV and 60% block at +35 mV. Dofetilide displayed a marginal voltage-dependence of Kv11.1 block with 43% block at -35 mV and 60% block at +35 mV.

89 reproduced by our model ( 17% from -35 to +35 mV at a test dose of 100 nM) (Figure 5.7). Similar to cisapride, this voltage-dependence is a measure of an altered distribution of open and inactivated states at negative voltages, and unequal afﬁnities to each state. It is small in comparison to cisapride as dofetilide is trapped in the closed states at negative membrane voltages.

20 1. APD30 0.8 0 0.6 0.4 20 0.2 0. 40 (pA/pF) current

voltage (mV)voltage 100 ms 60 APD90 80 100 ms B

20 1. 0.8 APD30 0 0.6 0.4 20 0.2 0. 40 (pA/pF) current 100 ms voltage (mV)voltage 60 APD90 80 100 ms

Figure 5.8: Triangulation – measured by an increase in time from APD30 to APD90 –ofthe epicardial cardiac action potential. (A) – cisapride and (B) – dofetilide. Prolongation of the cardiac action potential was most prominent in phase III. Control APs shown as dotted lines, Kv11.1 drug- blocked APs shown in solid lines. A concentration of drug predicted to block 90% of IKr (inset: IKr currents for control, dashed line, and in presence of drug, solid line), resulted in a 15 ms prolongation of the APD30, with a 41 ms prolongation of APD90, producing a ‘triangulation’ of the action potential shape.

5.3.2 IC50 from voltage-clamp experiments and action potential duration experiments

The correspondence between an IC50 value in voltage patch-clamp studies and the IC50 of APD prolongation (IC50-APD) remains unclear (Zhou et al., 2009). Simulations in the mTTP06 model demonstrated a near-linear dependence between GKr (conductance of the Kv11.1 channel) and APD (Figure 5.9). This allowed us to determine dose-response curves for cisapride and dofetilide effects on APD90 (Figure 5.9). At a pacing cycle length

90 (CL) of 1000 ms the IC50-APD was 27 nM for cisapride and 85 nM for dofetilide-a40% and 80% increase, respectively, compared to the IC50 determined from stepped voltage patch-clamp experiments (Kamiya et al., 2008). By integration, the relative proportion of time spent in the open:inactivated state is 1:9 during the epicardial action potential vs. 1:24 in the voltage-step protocol. Therefore the measured IC50-APD is more highly influenced by affinity for the open state than the IC50 of the voltage-step protocol. The calculated open:inactivated affinity for cisapride is ∼1:8 compared to ∼1:40 for dofetilide. This can explain the greater decrease in apparent dofetilide affinity for IKr under AP conditions compared to Kv11.1 under voltage-step conditions in our studies (Perrin et al., 2008).

340

330 APD90 (ms) 320

310

0.2 0.4 0.6 0.8 1. G G Kr/ Kr(control)

Figure 5.9: Increase in APD90 with reduction in Kv11.1 conductance (GKr). In simulated recordings of epicardial action potentials in the modiﬁed ten Tusscher-Panﬁlov model (mTTP06), APD90 increased in an almost linear fashion with reducing GKr.

5.3.3 Cisapride and dofetilide cause triangulation of the cardiac action potential

Prolonged phase III repolarization results in ‘triangulation’ of the cardiac action potential shape. The degree of triangulation was measured by the increment in time from 30% of repolarization (APD30) to 90% of repolarization (APD90) (Hondeghem et al., 2003). Sim- ulations were carried out in epicardial cells. The effect of IC90-APD concentrations of cisapride and dofetilide on the action potential shape (230 nM and 700 nM respectively, supplementary Figure 5.10) were compared to control simulations (Figure 5.8). Both drugs markedly prolonged the APD30 to APD90 repolarization time — 15 ms in control, 41 ms in the presence of drug. This was consistent with experimental studies demonstrating triangulation of the action potential in the presence of cisapride and dofetilide

91 (Hondeghem et al., 2003). Importantly, there was no signiﬁcant difference between the two drugs when comparing equivalent APD prolonging doses (see above).

AP D [drug] −APD[control] AP D −APD 1− 1− [drug] [control] AP D −APD [G Kr =0] [control] AP D [G Kr =0] −APD[control] 1 1

0.5 0.5

10 102 103 104 1 10 102 103 104 [drug] (nM) dofetilide [drug] (nM) cisapride

Figure 5.10: APD90 prolongation dose-response curves for dofetilide and cisapride Markov-state models in the modified ten Tusscher-Panfilov model (mTTP06). Simulations were carried out at a pacing cycle length of 1000 ms in epicardial cells. APD (Bany´ asz´ et al., 2009) — APD90 at a particular drug concentration; APD (Barrett et al., 2001) — APD90 at baseline; APDGKr=0 — APD90 when conductance of IKr was set to 0. The APD90-IC50 for dofetilide was 85.4 nM, and 27.4 nM for cisapride, with a Hill coefficient of ∼1 for both. The dashed line indicates the IC50.

5.3.4 Dofetilide is reverse use-dependent; cisapride is use-dependent

Reverse use-dependence (RUD) was determined by calculating APD90 in the presence of drug minus APD90 in control simulations at CLs ranging from 400 ms to 3000 ms. Both M and epicardial cells were simulated. The CL of 400 ms could not be tested in M cells as the stimulation fell within the absolute refractory period of the action potential. Dofetilide produced a greater relative increase in the APD90 in both M and epicardial cells at longer CLs (Figure 5.11). M cell APD90 increased by 48 ms at 600 ms CL, and by 76 ms at 3000 ms CL. This is consistent with previous studies that have demonstrated RUD for dofetilide and other methanesulfonanilides (Hondeghem et al., 2003; Bany´ asz´ et al., 2009; Strauss et al., 1970). In contrast, cisapride demonstrated marked use-dependence over the range of CLs tested. The cisapride-induced increase in M cell APD90 fell from 66 ms at 600 ms CL to 37 ms at 3000 ms CL (Figure 5.11). Over the range of CLs of 600 to 1000 ms, the marginal use-dependence in epicardial cells, and neither use- or reverse use-dependence in M cells in our model is consistent with the data of Hondeghem and colleagues(Hondeghem et al., 2003).

92 AB

80 80 60 60 ΔAPD ΔAPD (ms) (ms) 40 40 20 20

0 1000 2000 3000 0 1000 2000 3000 time (ms) time (ms)

Figure 5.11: The effect of pacing cycle length on magnitude of action potential prolongation in M () and epicardial () cells for (A) – dofetilide and (B) – cisapride. The IC50 of APD prolongation was used in simulations. Signiﬁcant RUD was demonstrated for dofetilide, most marked in M cells (absolute increase of 29 ms between CL 600 ms and CL 3000 ms). Cisapride, in contrast, displayed marked use-dependence, especially at CLs > 1000 ms, and again most marked in M cells (absolute reduction of 29 ms between CL 3000 ms and CL 600 ms).

5.3.5 Cisapride and dofetilide cause instability of the action potential duration

After constant pacing at stable CLs (600, 1000, and 2000 ms) for 3 min, we introduced a degree of instability by varying the beat-to-beat intervals from the base pacing rates by random small amounts in both directions. The simulations were continued for a further 3 min. This artiﬁcial instability was calibrated to produce a variance from the mean of APD90 of ∼10 ms in control simulations. Poincare´ plots were produced for simulations at each CL in the presence and absence of drug and are displayed in Figure 5.12. Both drugs produced signiﬁcant increases in APD instability at all CLs. The degree of instability increased markedly at longer CLs for dofetilide (Figure 5.12), consistent with its RUD. However, for cisapride, the degree of instability was similar for all CLs tested, increasing marginally at shorter durations (Figure 5.12), i.e. marginal use-dependence.

5.3.6 Cisapride and dofetilide increase dispersion of repolarization

In control simulations the difference between epicardial APD90 and M cell APD90 was 107 ms. In the presence of IC90-APD concentrations of drug the difference increased to

93 A B IC IC50 IC75 50 IC75 460 460 460 460

420 420 420 420 600 380 380 380 380 ms APD (ms) APD (ms) APD (ms) APD (ms) 340 340 340 340

340 380 420 460 340 380 420 460 340 380 420 460 340 380 420 460 APD (ms) APD (ms) APD (ms) APD (ms) 600 600 600 600 2000 ms 520 520 520 520 APD (ms) APD (ms) APD (ms) APD (ms) 440 440 440 440

440 520 600 440 520 600 440 520 600 440 520 600 APD (ms) APD (ms) APD (ms) APD (ms) 160 160

30 30 120 drug 120 drug 80 20 80 20

variance (ms) variance (ms) variance (ms) variance (ms) variance 40 40 control control 10 10 600 1000 2000 600 1000 2000 600 1000 2000 600 1000 2000 cycle length (ms) cycle length (ms) cycle length (ms) cycle length (ms)

Figure 5.12: Poincare´ plots of action potential duration instability in simulated M cells. Each point represents the duration of an AP plotted against the duration of the immediately prior action potential i.e. (x,y→APDn, APDn−1). Graphically, the mean of APD prolongation is approximated by the centre of each ‘knot’, and beat to beat variation from that mean by the degree of scatter. Row 1: Poincare´ plots at 600 ms CL at the IC50- and IC75-APD of (A) – dofetilide, and (B)– cisapride. Row 2: Poincare´ plots at 2000 ms CL at the IC50- and IC75-APD of dofetilide and cisapride. Row 3: Plots of variance from mean (a measure of instability) for each drug at IC50- and IC75-APD, and at CLs 600, 1000 and 2000 ms; drug variance (), control variance (). A marked increase in instability at longer CLs may be noted for dofetilide, with an opposite, though less marked effect for cisapride.

94 186 ms for dofetilide and 182 ms for cisapride (Figure 5.13). The increase was due to a relatively larger increase in APD90 in M cells, +116 ms for cisapride and +120 ms for dofetilide compared to +41 ms in epicardial cells for both cisapride and dofetilide. The slightly greater increase in M cell APD90 in the presence of dofetilide, compared to cisapride, may be attributed to a small increase in Kv11.1 inactivation during the prolonged plateau of the M cell action potential.

5.3.7 Cisapride and dofetilide induce early afterdepolarisations in M cells

We tested the effect of 3 μM concentrations of cisapride and dofetilide in M cells, to determine if early afterdepolarisations (EADs) could be induced by Kv11.1 block alone. Both cisapride and dofetilide at high concentration resulted in EADs and triggered activity. However, such effects were only seen at slow pacing rates (2000 ms CL). It is common knowledge that some drugs that block Kv11.1, do not commonly cause TdP (Yang et al., 2001; Fauchier et al., 1999). Such an effect is thought to result from balanced reduction in opposing currents. To test this hypothesis in our model system, systematic reductions of the fast inward sodium current (INa) and ICa−L currents were performed to examine the effect on EADs. Using dofetilide at a concentration that produced EADs in the previous experiments, a gradual reduction of INa was performed. At no level of INa current (down to 10% of Gmax) were EADs prevented in the simulation. Reducing ICa−L conductance, however, had a marked effect. At 90% of baseline ICa−L current, all triggered activity was abolished. At 80% of ICa−L current, all EADs had dis- appeared (Figure 5.14).

5.3.8 Cisapride and dofetilide delay repolarization in cable models of the ventricular wall

A 1D transmural fibre consisting of endocardial, mid-myocardial, and epicardial ‘cells’ in a 30:35:30 ratio was simulated in the presence and absence of drug. Both drugs caused significant delay in transmural repolarization; results for dofetilide shown in Figure 5.15. Importantly, the T wave was also noted to broaden and reduce in amplitude in the pseudo- ECG (Figure 5.15B). As predicted by in vivo studies, the degree of dispersion (longest to shortest APD90) measured in control simulations of the transmural fibre was markedly reduced compared to values suggested by single cell experiments (Opthof et al., 2007b), from ∼100 ms to 30 ms. This demonstrates the effect of the synchronisation of repolarization resulting from electrotonic interactions that occur in vivo.

95 A control

20 time 0 100 200 300 400 500 (ms) 20 40 voltage (mv) voltage 60 80

B dofetilide

20 time 0 100 200 300 400 500 (ms) 20 40 voltage (mv) voltage 60 80

C cisapride

20 time 0 100 200 300 400 500 (ms) 20 40 voltage (mv) voltage 60 80

Figure 5.13: Dispersion (difference in APD, shortest to longest) between epicardial (black) APD90 and M cell (red) APD90 was 107 ms and increased to 186 ms for (B) – dofetilide and 182 ms for (C) – cisapride at their respective IC90-APDs. The increase was driven by a disproportionate increase in M cell – compared to epicardial and endocardial (green) – APD in the presence of drug, and demonstrates the greater reliance of M cells on IKr repolarising current.

96 1 2

A IC90-APD 3 μM 40 40 20 20 0 0 20 20 40 40 voltage (mv) voltage 60 60 80 80 1 S 1 S B 40 0 20 2 *** 0 20 4 40

voltage (mv) voltage 6

60 current(pA/pF) 8 80 G 100% C Ca-L 40 0 20 2 ** 0 20 4 40 voltage (mv) voltage 6 60 current(pA/pF) 80 G 8 Ca-L90% D 40 0 20 2 0 20 4 40

voltage (mv)voltage 6 (mv) voltage 60 current(pA/pF) 80 G 8 Ca-L80%

Figure 5.14: A(1) – early afterdepolarisations (EADs) elicited from M cells when pacing at long CLs (2000 ms) in the presence of dofetilide at IC90-APD. (2)–3μM dofetilide elicited constant triggered activity. (B) – action potentials and ICa−L in the presence of 3 μM dofetilide; asterisk denotes reactivation of calcium channels that coincides with triggered activity. (C) – 10% reduction in the conductance of L-type calcium-channel conductance (GCa−L) eliminated triggered activity. (D) 20% reduction in GCa−L also eliminated EADs. The pattern of EAD induction and elimination was similar for both cisapride and dofetilide; only simulations of dofetilide-Kv11.1 are shown.

97 A control drug endo. endo.

(mm) (mm)

epi. epi. voltage (mV)voltage (mV)voltage time (ms) time (ms) B voltage voltage

time 100 200 300 400 500 600 (ms)

Figure 5.15: (A) – simulated transmural conduction across the left ventricular wall progressing from the endocardium (top) to epicardium (bottom) in each panel. Internal resistivity (Ri)was calibrated to give a conduction velocity across the wall of ∼450 mms−1 in agreement with experimental data from left ventricular wedge preparations (Yan et al., 1998b). The effect of the synchronisation of repolarization (Conrath and Opthof, 2006) may be seen in the control experiment (left) with a dispersion of repolarization of ∼30 ms, significantly reduced compared to that of isolated cells of ∼105 ms (see Figure 5.14). Consequently, fibre simulations at high concentration of drug (3 μM) did not induce EADs (right) producing an increase in APD of only 50 ms in M cells. (B) – pseudo-electrogram demonstrating that 3 μM dose dofetilide (red tracing) broadened and reduced the amplitude of the T wave, similar to in vivo findings (Vaglio et al., 2008). This is the electrical correlate of action potential triangulation (Shah and Hondeghem, 2005).

98 5.4 Discussion

In this study we demonstrate that the principal pre-clinical markers of drug-acquired arrhythmia (TRIaD) may be reproduced by Kv11.1 block alone. Our method was to completely describe each drug’s interaction with Kv11.1 in Markov chain form, and from this foundation ‘reconstruct the long chains of causes’ (Hille) to the observable phenomena of arrhythmic risk. Furthermore, differences in drug speciﬁc effects on the various TRIaD markers could be explained by the differences in the kinetics and state-dependence of drug-binding.

5.4.1 Mechanistic insight from Markov models

Markov-state models provide a versatile structure for modelling ion channel kinetics and state-dependent drug-binding (Fink and Noble, 2009). The validity of our models was demonstrated by the fact they could reproduce complex experimental data that was not used to constrain them (Figure 5.7). Our cisapride-Kv11.1 model was noteworthy in this respect. While care must be taken in inferring molecular structural changes from Markov- state models (Fink and Noble, 2009), we would reason that the inability of Markov-state schemes incorporating closed-state dissociation to reproduce even closely our array of experimental data for cisapride-Kv11.1 block, argues against pure dissociation from a closed-state (activation gate closed) (Stork et al., 2007). Conversely, the relatively simple Markov-state model presented in Figure 5.1A is able to reproduce the marked voltage- dependence of cisapride-Kv11.1 block (Walker et al., 1999b) (Figure 5.7) even though it contains no direct description of voltage-dependence. Instead, voltage-dependence arose indirectly as a function of cisapride’s unequal afﬁnities for the open and inactivated states (Perrin et al., 2008). Furthermore, it had a tendency to dissociate from these states at negative voltages (Stork et al., 2007), in concert with different channel-state distributions at varying membrane voltage i.e. by way of a kinetic, not a charge-related mechanism. This suggests that other drugs which block the channel in a voltage-dependent manner, such as quinidine (Tsujimae et al., 2004), may also do so through non-charge- related mechanisms.

5.4.2 Markov models reproduce TRIaD markers of proarrhythmic risk

Placing each of our models of drug-block into a modiﬁed TTP06 ‘myocyte’ allowed us to examine the particular inﬂuence of Kv11.1-drug kinetic interactions on TRIaD markers of arrhythmic risk. Our changes to the TTP06 model were conservative, and did not alter its fundamental characteristics, but did improve the ability of the model to simulate complex

99 alterations in repolarization as demanded by TRIaD. Triangulation of the action potential, the T of TRIaD, is proarrhythmic principally because it promotes the reactivation of recovered L-type calcium channels during phase III repolarization (Shah and Hondeghem, 2005). Unsurprisingly, as triangulation is a general outcome of IKr reduction (Romero et al., 2009), both Markov-state models produced this effect on the cardiac action potential. Reactivation of L-type calcium channels, however, only occurred at long CLs (2000 ms) and at ≥ 90% reduction of Kv11.1 current (Figure 5.14). It has been noted previously that the electrocardiographic correlate of triangulation is reduced amplitude and broadening of the T wave (Shah and Hondeghem, 2005); our electrocardiogram of left ventricular transmural conduction reproduced these findings (Figure 5.15). Reverse use-dependence describes the tendency of a drug to prolong the APD to a greater extent at low compared to high heart rates. It is predicted to be arrhythmogenic through a prolongation of the already lengthened APDs present at low heart rates to a point where L-type calcium reactivation (the initiator of EADs) may occur. Importantly, reverse use-dependence is now recognised as an intrinsic property of the cardiac myocyte (Bany´ asz´ et al., 2009) – i.e. any agent that prolongs the APD is predicted to display RUD. Our simulations of dofetilide-Kv11.1 block in myocytes confirmed this. Despite the frequency independence of dofetilide block of Kv11.1 (Figure 5.7), prolongation of the APD is markedly greater at long CLs compared to short CLs (Figure 5.11). The opposite effect, however, is noted with cisapride. Reduced block of the Kv11.1 channel at low heart rates (Figure 5.7), due to drug dissociation at negative membrane potentials, effectively opposes the intrinsic RUD of the myocyte model. The difference in APD prolongation is minimal between CLs of 600 and 1000 ms in agreement with experimental data (Shah and Hondeghem, 2005). The important implications of these results are visualised in simulations of APD instability. Dofetilide produced significant instability at long CLs, but the increase in instability is much less for cisapride. Consistent with its lack of RUD, the instability evoked by cisapride increased at short CLs. This also matches the experimental observation that induction of cisapride instability requires the use of short-cycle length protocols (Fossa et al., 2002) while instability for dofetilide is more easily inducible. Instability is predicted to increase chaotic behaviour and be proarrhythmic (Thomsen et al., 2004; Weiss et al., 1999). Finally, the degree of dispersion indicated by single cell experiments is roughly equal between the two drugs at a constant CL of 1000 ms. However, in line with RUD for dofetilide but not cisapride, dispersion increased at long CLs for dofetilide with an opposite pattern for cisapride. The effect of dispersion is ameliorated in vivo by electronic coupling of cells, as demonstrated in our transmural fibre experiments (Figure 5.15), where

100 the degree of dispersion decreased from ∼100 ms in single cells, to ∼30 ms in the cable simulations. Absolute dispersion across the ventricular wall would likely be even smaller in the intact heart where M cell characteristics are not seen (Opthof et al., 2007b). It is important to note that our simulations reveal repolarization changes that are most likely evident only transiently in vivo. Arrhythmia in the congenital and aLQTS is rarely inces- sant but typically occurs episodically as the result of a ‘second hit’ – whether that be exposure to a Kv11.1-blocking drug on a genetic background of LQTS, hypokalaemia, cellular uncoupling associated with myocardial ischaemia, or a large sympathetic outﬂow caused by a loud noise or fright. Our simulations revealed changes to repolarization that will become manifest when the mechanisms that smooth and synchronise repolarization in vivo, both anatomical and electrical, are compromised.

5.4.3 Predicting arrhythmogenicity

An intriguing question relates to the accuracy of the predicted arrhythmogenicity of cisapride and dofetilide based on their different effects on TRIaD. As cisapride triangulates the action potential and increases the dispersion of repolarization, a degree of proarrhythmic potential is likely. However, cisapride’s lack of RUD, and reduced instability at long CLs, may oppose arrhythmogenesis, especially in the setting of bradycardia. The clinical event rate for cisapride is estimated at 1/120,000 (Wysowski and Bacsanyi, 1996), with one study finding no significant increase in risk (Walker et al., 1999a). Conversely, dofetilide which exhibits all four features of TRIaD has a significantly higher risk of TdP at 0.3-10% (Lenz and Hilleman, 2000). Much of this difference may relate to higher free drug concentrations in plasma for dofetilide, but we propose that the peculiar kinetic qualities of cisapride also contribute to its relative safety.

5.5 Conclusion

Beginning with first-principle models describing the interaction of two different compounds with the Kv11.1 K+ channel, the full range of pre-clinical risk markers as represented by TRIaD, and drug-specific differences in these effects, may be reproduced. We conclude that Kv11.1 block by itself accounts for TRIaD. The absence of TRIaD in a known Kv11.1 blocker suggests that it must exert a significant effect on other ion channels influencing repolarization. Further, we suggest that an agent that is able to dissociate from the channel at negative membrane voltages, and therefore oppose the intrinsic RUD of the cardiac myocyte, thus exerting its maximum effect at high HR (e.g. tachyarrhythmia), would provide a significant improvement on the current generation of antiarrhythmic medications.

101 Chapter 6

Improving High-Throughput Screening Assays

6.1 Background

The most feared side-effect of any drug is sudden cardiac death (SCD). This tragic event is a frequent reason for the withdrawal of compounds post-marketing (Table 6.1). Drug block of Kv11.1 is the common mechanism (Roden, 2004). It is unsurprising, then, that a measurement of drug-Kv11.1 affinity is now mandated (ICH S7A and S7B, http://www.ich.org) in the preclinical assessment of a drug’s propensity to prolong the QT. Though the TRIaD markers are noted for their accuracy in identifying the proarrhythmic substrate (Hoffmann and Warner, 2006) (Chapter 5), they are not in routine use; instead, an integrated approach is typical, comprising: measurement of Kv11.1 inhibition in mammalian cells and APD prolongation in canine Purkinje fibres; and QT monitoring in telemetered dog studies (Wallis, 2009). While this approach identifies the majority of drugs likely to prolong the QT in vivo, it is costly and time-consuming. Ideally, only a small number of compounds will require

Drug Class Target Withdrawn terodiline antimuscarinic muscarinic + calcium channels 1993 terfenadine antihistamine H1-receptor 1997 astemizole antihistamine H1-receptor 1999 grepaﬂoxacin ﬂuoroquinolone antibiotic bacterial DNA gyrase 1999 cisapride prokinetic agent 5-HT4 2000

Table 6.1: Drugs withdrawn from the US market due to QT prolongation and risk of torsades de pointes (Whellan et al., 2009).

102 thorough testing having shown themselves unlikely to prolong the QT at concentrations that favourably modify the intended target. For this purpose, high-throughput screening (HTS) methods are employed early in drug development. Lead compounds, of which there may be many, are assayed and a ratio calculated of affinity-for-target compared to affinity-for-Kv11.1 – the ‘safety factor’. The large amount of data obtained in this process becomes useful at a later stage. Then, computational methods may be employed to analyse the dataset to determine those compounds having similar structure, but markedly dissimilar affinity for Kv11.1. The small structural difference is assumed responsible for channel binding and may be modified by medicinal chemists to reduce the Kv11.1-affinity of interesting compounds (Bell and Bilodeau, 2008). It is of great importance, then, that the HTS method employed be accurate. Unfor- tunately, this is not the case; both radio-ligand binding studies (Netzer et al., 2003), and fluorescence-based screening assays (Tang et al., 2001), provide drug-Kv11.1 affinity measurements which differ from voltage-clamp values by significant amounts. Our work suggests that the complex kinetic interactions of a drug with Kv11.1, and the effect of state-dependent binding, may significantly reduce the measured affinity when testing conditions are altered (Chapter 3-5). In this chapter, we seek to further investigate these effects. Current-clamping experiments are performed to mimic the voltage readout from fluorescence-based screening assays and allow an investigation into the theoretical Figure 6.1: Raw tracing of a current- foundation of such testing. Further, mathemat- clamp experiment. Quinidine at a ical models of membrane-voltage change with concentration of 3 μM produced a voltage drug-block provide insight into the cause of as- change near 5 mV (-53 to -48 mV) in this say inaccuracy and, more importantly, enable us recording. Starting membrane voltages -53 mV were accepted for dose-response to predict favourable changes to Kv11.1 channel curves; most membrane voltages were gating – through mutation – that are likely to sub- -60 mV. stantially increase the sensitivity and accuracy of fluorescence-based screening methods.

103 6.2 Mathematical Model of a High-Throughput Assay

A number of ﬂuorescence-based high-throughput screening assays have been described e.g. (Tang et al., 2001; Dorn et al., 2005). The theoretical foundation of such testing is described in Subsection 1.4.4. We employed the method of Dorn and colleagues (Dorn et al., 2005) as the template for our mathematical model due to its simplicity, and use of Chinese hamster ovary (CHO) cells expressing Kv11.1. Their method in brief: CHO cells stably transfected with Kv11.1, in combination with a membrane voltage sensitive dye, are exposed to drug; the change in ﬂuorescence signal – corresponding to change in membrane voltage – correlates with the degree of Kv11.1 channel block (Subsection 1.4.4).

6.2.1 Formation of the mathematical model

Experiments previously performed in our laboratory on non-transfected CHO cells, have demonstrated the absence of signiﬁcant endogenous current, and a resting membrane voltage of -31 mV (n = 30) (Walker et al., 1999b). The ionic permeability responsible for the resting membrane voltage was therefore modelled as a linear ‘leak’ current with a reversal potential of -31 mV. Our current-clamp experiments revealed an average resting membrane voltage of -60 ± 1 mV (n = 18) for CHO cells stably expressing Kv11.1.

By deﬁnition, at the resting membrane voltage – Vrm –, the sum of all currents is equal to 0:

IVrm = GK (Vrm − EK )+GL(Vrm − EL) (6.1)

where IVrm is the sum of current at the resting membrane voltage and equal to 0, Vrm is the resting membrane voltage in CHO-Kv11.1 cells: -60 mV, GK is the Kv11.1 channel conductance, EK is the reversal potential of potassium (-81 mV with our solutions), GL is the conductance of the leak channel, and EL is the reversal potential of the leak current: -31 mV. 1 GK was set to a realistic but arbitrary value of 30 nS at -60 mV , and Equation 6.1 solved, giving a GL of 0.22 nS. As the leak current was assumed linear, GL is a constant. GK, however, is a function of the proportion of Kv11.1 channels in the open (O) channel state: 1The ‘true’ ﬁgure is unimportant in modelling terms as the membrane voltage is a function of the relative conductances, and not the absolute values

104 −9 [O] GK =30× 10 × (6.2) [O]Vrm

where O is the proportion of channels in the open – conducting – state at each mo- ment in time, and [O]Vrm is the proportion of channels in the open state at the resting membrane voltage of the cell. dV Change in membrane voltage with time – dt – was described by a Goldman-Hodgkin- Katz equation in this form:

dV 1 = − × (GK (V − EK )+GL(V − EL)) (6.3) dt Cm

where Cm is the capacitance of the CHO cell. Equation 6.3 was solved numerically with the differential equations of the various Markov state models described previously (Chapter 5). A simpliﬁed model was also produced using Hodgkin-Huxley descriptions for Kv11.1 gating in order to rapidly correlate reduced conductance of Kv11.1 with observed membrane voltage change. Initial simulations revealed that membrane voltage was not a linear function of Kv11.1 conductance – a 50% reduction in GK produced only a 3.5 mV shift in membrane voltage (Figure 6.2). Before predictive modelling simulations were performed, experiments examining membrane voltage change with drug-block were recorded in order to validate the model.

105 6.3 Current-Clamp Experiments

Current-clamp experiments were performed according to this method: after gaining electrical access to the cell, the recording mode was switched from voltage-clamp to current- clamp, and the injected current set to 0 pA — effectively a ‘voltage-following’ mode. Upon reaching the steady-state membrane-voltage, up to three concentrations of drug were applied, and the magnitude of voltage-change with each concentration measured. Cells with a resting membrane voltage of ≤ -53 mV, and a ﬁnal membrane voltage after the wash-on of drug of ≤ -30 mV were accepted for analysis. The conductance of Kv11.1 for each drug concentration was ‘read’ from the simpli- ﬁed Hodgkin-Huxley Kv11.1-CHO cell model (Section 6.2). An example experimental recording is presented in Figure 6.1.

6.3.1 Results

Dose-response curves were obtained for 4 drugs: cisapride, haloperidol, dofetilide, and quinidine. The estimated IC50 for quinidine was similar to that obtained by the voltage- clamp method (Table 6.2 and Figure 6.3), but each of haloperidol, dofetilide, and cisapride, had a right shifted drug-afﬁnity (i.e. reduced) compared to their respective voltage- clamp experiments. A reduction in apparent afﬁnity was most marked for cisapride — 8 fold.

Having demonstrated a significant difference in drug-Kv11.1 affinity measurements between current- and voltage-clamp experiments, we sought to determine if our mathematical model of membrane voltage change with drug-block (Section 6.2) reproduced these results. Mod- elling simulations were performed by simply placing the relevant rate constants for each drug (Figure 5.2) into the mathematical description of Figure 6.2: Membrane voltage change the model cell. correlated to Kv11.1 conductance in a simplified mathematical model of a sta- Haloperidol was modelled as a drug that was bly Kv11.1-transfected CHO cell. Voltage trapped by the activation gate at negative mem- change is demonstrably not a linear func- brane voltages (Figure 6.5); quinidine, was mod- tion of Kv11.1 conductance. 50% block correlates to a 3.5 mV shift in membrane elled in a similar manner to cisapride (Subsec- voltage. tion 5.2.1) as experimental evidence has sug-

106

Figure 6.3: Current-clamp, and voltage-clamp, derived Hill plots for (A) – quinidine; (B)– haloperidol; (C) – dofetilide; and (D) – cisapride. Voltage-clamp Hill plots in red; current-clamp Hill plots in black. A significant right-shift (reduction) in drug-affinity is noted for each compound except quinidine. Comparative IC50 values are presented in Table 6.2. gested that it is able to dissociate from the channel at negative membrane voltages (Tsu- jimae et al., 2004). The Hill plots obtained from these simulations are displayed together with the experimental data in Figure 6.4. As may be seen, a reasonable correlation between current- clamp experiments and modelled simulations was present. The closest agreement ex- isted for quinidine and haloperidol. Importantly, the rank order of reduction in affinity for current-clamp IC50 obtained from experimental and modelled data, compared to voltage- clamp experiments, was maintained, i.e. quinidine < haloperidol < dofetilide < cisapride.

6.3.2 Discussion

Our experiments demonstrate a discordance between measurements of Kv11.1-afﬁnity obtained by current-clamp experiments, and those determined by voltage-clamp experiments. While there may be multiple reasons for this discrepancy e.g. dye quenching

107

Figure 6.4: Modelled current-clamp compared to experimental current- and voltage-clamp experiments. The modelled simulations (red) are presented with current-clamp experiments (black) and voltage-clamp experiments (dotted) for (A) – quinidine; (B) – haloperidol; (C) – dofetilide; and (D) – cisapride. The modelled Hill plots approximated the experimental data for quinidine and haloperidol, and were right-shifted compared to the experimental data for dofetilide and cisapride probably as a result of the slow onset of block for these drugs with consequent increase in patch-leak. IC50 values for each of the experiments/simulations are presented in Table 6.2. effects, a more fundamental flaw lies at the heart of the method: viz. voltage change with drug-block is not a surrogate of drug-Kv11.1 affinity measured under voltage-clamp conditions. Of the four drugs tested, the measured affinity of quinidine in all assays was the most similar; then, in order of reduced concordance: haloperidol, dofetilide, and cisapride. Drug-binding to Kv11.1, as described in Chapters 4 and 5, is time (frequency-dependent or -independent), voltage-, use-, and state-dependent. The typical resting membrane voltage of CHO cells stably transfected with Kv11.1 is -60 mV. At this voltage, the channel is principally in the closed state (>99%), and of the small amount of ‘activated’ (open and inactivated) channels, the majority will occupy the open state (see steady-state activation curve of Figure 6.6A). Though a small percentage of the whole, open-channel state occupancy is, by definition, responsible for the resting membrane voltage. While drug- block of the open state will shift the membrane voltage, the affinity for the open state may

108 Drug Voltage-Clamp Current-Clamp Modelled Current-Clamp quinidine 3090 3074 5691 haloperidol 90 211 258 dofetilide 51 137 344 cisapride 21 175 348

Table 6.2: IC50 values in nM for voltage-clamp and current-clamp experiments.

be reduced 50 or more times compared to its affinity for the inactivated state (Chapter 4). Further, at negative membrane voltages, some drugs will dissociate from the channel (Chapter 5). The combined effect renders the fluorescence-method an assay of drug- binding to the open-state modified by deactivation-driven drug-unbinding, rather than an assay of drug-Kv11.1 affinity per se. Though quinidine may dissociate from the channel at negative voltages, its very fast on- rate (Tsujimae et al., 2004), and lack of state-

dependence (Perrin et al., 2008), effectively opposes this effect, and results in a reasonable correlation between measured IC50 in current-

and voltage-clamp experiments. On the other hand cisapride is markedly state-dependent (Perrin et al., 2008), and dis- sociates rapidly from the channel at negative voltages (Stork et al., 2007). Thus its apparent Figure 6.5: Incomplete recovery of Kv11.1 from haloperidol block at negative afﬁnity for Kv11.1 at -60 mV, when the closed membrane voltages, performed according state predominates, is far from its voltage- to method of Stork et al. (2007) – test- clamp measurement (Figure 5.7). Haloperi- pulses (Figure 5.4) were applied at timed dol and dofetilide, though trapped by the ac- intervals; interval membrane voltage: -80 mV. The ﬁnal recovery current was 10% of tivation gate at negative voltages and there- control. fore unable to unbind, are both state-dependent – dofetilide (∼50:1, inactivated:open) more so than haloperidol (∼14:1) (Chapter 5). Therefore, the current-clamp IC50 for haloperidol will more closely approximate the voltage-clamp derived value than dofetilide.

109 6.4 Modelling an Improved High-Throughput Assay

In the final part of this chapter we produce models of voltage change with drug-block based upon the approximated gating properties of several mutant-Kv11.1 channels. We have shown that mutants may reduce the affinity of drugs for Kv11.1 by shifting channel distributions (Chapter 4); here, we seek to determine if mutant channels can effectively rescue ‘Kv11.1 affinity’ by opposing the forces that reduce it at negative membrane voltages. Two problems present when measuring drug-Kv11.1 affinity at -60 mV: first, the open- inactivated distribution is left-shifted compared to +20 mV – the testing voltage of voltage- clamp experiments, and an approximation to the plateau voltage of the cardiac action potential; second, occupancy of the closed-states is predominant – the drive into these states promoting drug-unbinding from the open and inactivated states (Chapter 5). There- fore, we hypothesised that mutant channels with left-shifted inactivation and activation would improve the accuracy of drug-testing at negative membrane voltages. The results for our modelling simulations are presented in Figure 6.6, and demonstrate that Kv11.1 mutants that left-shift activation and inactivation increase the range and sensitivity of voltage responses to drugs that block Kv11.1 in CHO cells.

110

"!%$! "!%$! "!%$! "!%$!

&"$ &"$ &"$ &"$

111 &"$ &"$ &"$ &"$ $ # $ # $ # $ #

Figure 6.6: Upper panel: black lines indicate the voltage-dependence of steady-state activation (closed and open states); red lines, the voltage- dependence of steady-state inactivation (open and inactivated states). Positive voltages favour movement out of the closed states and into the inactivated states. The shaded area, overlap of the two curves, represents open-state occupancy at steady-state conditions. The relative occupancy of the open state between different channels is displayed for: (A)–WT;(B) – V535A (left-shifted activation); (C) – N588E (left-shifted inactivation); and (D) – V535A + N588E (left-shifted activation and inactivation). The double mutant – D – produces a channel with a large voltage range, though channel-occupancy still favours the inactivated state. Lower panel: modelled membrane voltage change with channel block in the

various mutant channels when perfused with 20 nM cisapride i.e. the IC50 under voltage-clamp conditions. The change in membrane voltage is 0.5 mV for WT-Kv11.1, 3.2 mV for V535A, 1.2 mV for N588E, and 8.6 mV for V535A + N588E. Mutant channels were modelled by altering relevant rate constants describing activation and inactivation according to unpublished laboratory data. Chapter 7

Conclusion

This thesis is presented as an attempt to ‘understand experimental observations as the co-operative interaction of fundamental laws, to describe those laws, re-form them mathematically, and validate them by recapitulating independent experimental data’. First, we gained mechanistic insight into the effect of high external [K+] to reduce drug-block of KV11.1. Two mechanisms have been proposed for this observation: drug displacement from the channel pore in the presence of high potassium; and potassium- induced shift of the voltage-dependence of inactivation with a change in channel state distributions favouring the open – and theorised lower affinity – state. Our results are evidence against the former proposal. At +20 mV, block of Kv11.1 for a range of different drugs was not substantially changed between normal and high potassium concentrations. As channel-state distributions between the two potassium concentrations are similar at +20 mV, absence of reduced affinity suggests absence of a direct effect of potassium to displace drug. Second, using mutants that shift the voltage-dependence of steady-state inactivation without altering the fundamental properties of Kv11.1, we demonstrated that high affinity blockers of Kv11.1 bind preferentially to the inactivated state of the channel. Conversely, most drugs that bind at low affinity to Kv11.1 show no preference for the open or inactivated conformations. Using a process of simple mathematical modelling we calculated relative affinities for the open and inactivated states for a range of drugs. These results have an important implication for the screening of drugs that block Kv11.1: assays should predominantly measure affinity for the inactivated state. Further, we discovered a high affinity blocker of the N588K-Kv11.1 channel, providing a new therapeutic candidate for the short QT syndrome type I. Third, knowledge of the kinetics and state-dependence of drug-binding to Kv11.1 allowed us to produce Markov-state models of drug-block that were able to recapitulate

112 drug-specific characteristics of Kv11.1 block. This involved a process of re-forming mathematically the ‘fundamental laws’ of chapter 3 and 4 and validating them by reproducing independent experimental data from the literature. Placing the Markov models into a mathematical description of human ventricular tissue allowed us to investigate the genesis of important markers of preclinical risk, viz. TRIaD. TRIaD was found to be entirely predictable from a knowledge of the drug-specific kinetics and state-dependence of channel block. We were therefore able, for the first time, to suggest the underlying mechanism of TRIaD. Our results also suggested that drug-unbinding from Kv11.1 at negative membrane voltages was protective against drug-induced arrhythmia. Thus a new avenue for safer drug-development was provided from this work. Last, we investigated high-throughput screening assays used for measuring drug- Kv11.1 affinity. The inaccuracy of these methods is well known. By current-clamp experiments, and a mathematical model of a stably-transfected Kv11.1 CHO cell, we established that drug-specific frequency-dependence, and state-dependence, altered the apparent affinity of the drug for the channel when measured at the negative membrane voltages of the high-throughput assay. Our modelling results indicate that the accuracy of such screening may be improved by the use of mutant-Kv11.1 channels that shift the voltage-dependence of activation and inactivation. All that remains is the experimental validation of this final proposal. Such experiments continue; if successful, a substantial improvement over currently employed methods is hoped with consequent benefit to pharmaceutical development, and ultimately the millions who will require prescription medications.

113 Bibliography

Akhavan A, Atanasiu R, Noguchi T, Han W, Holder N and Shrier A (2005). Identiﬁcation of the cyclic-nucleotide-binding domain as a conserved determinant of ion-channel cell- surface localization. J Cell Sci 118(Pt 13):2803–12.

Antzelevitch C (2005). Cardiac repolarization. The long and short of it. Europace 7 Suppl 2:3–9.

Antzelevitch C (2007). Ionic, molecular, and cellular bases of QT-interval prolongation and torsade de pointes. Europace 9 Suppl 4:iv4–15.

Antzelevitch C, Pollevick GD, Cordeiro JM, Casis O, Sanguinetti MC, Aizawa Y, Guer- chicoff A, Pfeiffer R, Oliva A, Wollnik B, Gelber P, Bonaros EP, Burashnikov E, Wu Y, Sargent JD, Schickel S, Oberheiden R, Bhatia A, Hsu LF, Ha¨ıssaguerre M, Schimpf R, Borggrefe M and Wolpert C (2007). Loss-of-function mutations in the cardiac calcium channel underlie a new clinical entity characterized by ST-segment elevation, short QT intervals, and sudden cardiac death. Circulation 115(4):442–9.

Antzelevitch C and Sicouri S (1994). Clinical relevance of cardiac arrhythmias generated by afterdepolarizations. Role of M cells in the generation of U waves, triggered activity and torsade de pointes. J Am Coll Cardiol 23(1):259–77.

Antzelevitch C, Sun ZQ, Zhang ZQ and Yan GX (1996). Cellular and ionic mechanisms underlying erythromycin-induced long QT intervals and torsade de pointes. J Am Coll Cardiol 28(7):1836–1848.

Attwell D and Lee JA (1988). A cellular basis for the primary long Q-T syndromes. Lancet 1(8595):1136–1139.

Bai R, Napolitano C, Bloise R, Monteforte N and Priori SG (2009). Yield of genetic screening in inherited cardiac channelopathies: how to prioritize access to genetic testing. Circ Arrhythm Electrophysiol 2(1):6–15.

114 Bany´ asz´ T, Horvath´ B, Virag´ L, Bar´ andi´ L, Szentandrassy´ N, Harmati G, Magyar J, Marangoni S, Zaza A, Varro´ A and Nan´ asi´ PP (2009). Reverse rate dependency is an intrinsic property of canine cardiac preparations. Cardiovasc Res 84(2):237–44.

Barrett TD, Hennan JK, Fischbach PS, O’Neill BP, Driscoll Jr EM Jr and Lucchesi BR (2001). Tedisamil and dofetilide-induced torsades de pointes, rate and potassium dependence. Br J Pharmacol 132(7):1493–500.

Barry PH (1994). JPCalc, a software package for calculating liquid junction potential corrections in patch-clamp, intracellular, epithelial and bilayer measurements and for correcting junction potential measurements. J Neurosci Methods 51(1):107–16.

Baukrowitz T and Yellen G (1995). Modulation of K+ current by frequency and external [K+]: a tale of two inactivation mechanisms. Neuron 15(4):951–960.

Bell IM and Bilodeau MT (2008). The impact of I(Kr) blockade on medicinal chemistry programs. Curr Top Med Chem 8(13):1128–39.

Bellocq C, van Ginneken A, Bezzina C, Alders M, Escande D, Mannens M, Baro I and Wilde A (2004). Mutation in the KCNQ1 gene leading to the short QT-interval syndrome.

Benton RE, Honig PK, Zamani K, Cantilena LR and Woosley RL (1996). Grapefruit juice alters terfenadine pharmacokinetics, resulting in prolongation of repolarization on the electrocardiogram. Clin Pharmacol Ther 59(4):383–388.

Bernstein J (1912). Elektrobiologie, die Lehre von den elektrischen Vorgangen¨ im Organ- ismus auf moderner Grundlage dargestellt. Die Wissenschaft, 44 Hft. Braunschweig, F. Vieweg.

Bezanilla F (2008). How membrane proteins sense voltage. Nat Rev Mol Cell Biol 9(4):323–32.

Brugada R, Hong K, Dumaine R, Cordeiro J, Gaita F, Borggrefe M, Menendez TM, Bru- gada J, Pollevick GD, Wolpert C, Burashnikov E, Matsuo K, Wu YS, Guerchicoff A, Bianchi F, Giustetto C, Schimpf R, Brugada P and Antzelevitch C (2004). Sudden death associated with short-QT syndrome linked to mutations in HERG. Circulation 109(1):30–5.

Busjahn A, Knoblauch H, Faulhaber HD, Boeckel T, Rosenthal M, Uhlmann R, Hoehe M, Schuster H and Luft FC (1999). QT interval is linked to 2 long-QT syndrome loci in normal subjects. Circulation 99(24):3161–3164.

115 Carmeliet E (1992). Voltage- and time-dependent block of the delayed K+ current in cardiac myocytes by dofetilide. J Pharmacol Exp Ther 262(2):809–17.

Cavalli A, Poluzzi E, Ponti FD and Recanatini M (2002). Toward a pharmacophore for drugs inducing the long QT syndrome: insights from a CoMFA study of HERG K(+) channel blockers. J Med Chem 45(18):3844–53.

Cavero I, Mestre M, Guillon JM and Crumb W (2000). Drugs that prolong QT interval as an unwanted effect: assessing their likelihood of inducing hazardous cardiac dysrhyth- mias. Expert Opin Pharmacother 1(5):947–973.

Chen J, Seebohm G and Sanguinetti MC (2002). Position of aromatic residues in the S6 domain, not inactivation, dictates cisapride sensitivity of HERG and eag potassium channels. Proc Natl Acad SciUSA99(19):12461–6.

Chen MX, Helliwell RM and Clare JJ (2009). In vitro proﬁling against ion channels beyond hERG as an early indicator of cardiac risk. Curr Opin Mol Ther 11(3):269–81.

Chevalier P, Bellocq C, Millat G, Piqueras E, Potet F, Schott JJ, Baro I, Lemarec H, Barhanin J, Rousson R and Rodriguez-Lafrasse C (2007). Torsades de pointes complicating atrioventricular block: evidence for a genetic predisposition. Heart Rhythm 4(2):170–174.

Clarke CE, Hill AP, Zhao J, Kondo M, Subbiah RN, Campbell TJ and Vandenberg JI (2006). Effect of S5P alpha-helix charge mutants on inactivation of hERG K+ channels. J Physiol 573(Pt 2):291–304.

Conrath CE and Opthof T (2006). Ventricular repolarization: an overview of (patho)physiology, sympathetic effects and genetic aspects. Prog Biophys Mol Biol 92(3):269–307.

Cordeiro JM, Brugada R, Wu YS, Hong K and Dumaine R (2005). Modulation of I(Kr) inactivation by mutation N588K in KCNH2: a link to arrhythmogenesis in short QT syndrome. Cardiovasc Res 67(3):498–509.

Curran ME, Splawski I, Timothy KW, Vincent GM, Green ED and Keating MT (1995). A molecular basis for cardiac arrhythmia: HERG mutations cause long QT syndrome. Cell 80(5):795–803.

Denney SD, Lakkireddy DR and Khan IA (2005). Long QT syndrome and torsade de pointes in transient left ventricular apical ballooning syndrome. Int J Cardiol 100(3):499–501.

116 Dennis A, Wang L, Wan X and Ficker E (2007). hERG channel trafﬁcking: novel targets in drug-induced long QT syndrome. Biochem Soc Trans 35(Pt 5):1060–1063.

Dessertenne F (1966). [Ventricular tachycardia with 2 variable opposing foci]. Arch Mal Coeur Vaiss 59(2):263–272.

Di Pasquale G, Lusa AM, Manini GL, Dominici P, Andreoli A, Limoni P and Pinelli G (1984). [Cardiac arrhythmias associated with subarachnoid hemorrhage. Prospective study with dynamic electrocardiography]. G Ital Cardiol 14(5):323–329.

Dokos S and Lovell NH (2004). Parameter estimation in cardiac ionic models. Prog Biophys Mol Biol 85(2-3):407–31.

Donger C, Denjoy I, Berthet M, Neyroud N, Cruaud C, Bennaceur M, Chivoret G, Schwartz K, Coumel P and Guicheney P (1997). KVLQT1 C-terminal missense mutation causes a forme fruste long-QT syndrome. Circulation 96(9):2778–2781.

Dorn A, Hermann F, Ebneth A, Bothmann H, Trube G, Christensen K and Apfel C (2005). Evaluation of a high-throughput ﬂuorescence assay method for HERG potassium channel inhibition. J Biomol Screen 10(4):339–47.

Doyle DA, Morais Cabral J, Pfuetzner RA, Kuo A, Gulbis JM, Cohen SL, Chait BT and MacKinnon R (1998). The structure of the potassium channel: molecular basis of K+ conduction and selectivity. Science 280(5360):69–77.

Drouin E, Charpentier F, Gauthier C, Laurent K and Le Marec H (1995). Electrophysio- logic characteristics of cells spanning the left ventricular wall of human heart: evidence for presence of M cells. J Am Coll Cardiol 26(1):185–192.

Dubin AE, Nasser N, Rohrbacher J, Hermans AN, Marrannes R, Grantham C, Van Rossem K, Cik M, Chaplan SR, Gallacher D, Xu J, Guia A, Byrne NG and Mathes C (2005). Identifying modulators of hERG channel activity using the PatchXpress planar patch clamp. J Biomol Screen 10(2):168–81.

Duncan RS, McPate MJ, Ridley JM, Gao Z, James AF, Leishman DJ, Leaney JL, Witchel HJ and Hancox JC (2007). Inhibition of the HERG potassium channel by the tricyclic antidepressant doxepin. Biochem Pharmacol 74(3):425–37.

Dupler DA (1953). Ventricular Arrhythmia and Stokes-Adams Syndrome. Circulation 7:585–590.

Farid R, Day T, Friesner RA and Pearlstein RA (2006). New insights about HERG blockade obtained from protein modeling, potential energy mapping, and docking studies. Bioorg Med Chem 14(9):3160–3173.

117 Fauchier L, Babuty D, Autret ML, Cosnay P and Fauchier JP (1997). Effect of verapamil on heart rate variability in subjects with normal hearts. Am J Cardiol 80(9):1234–1235.

Fauchier L, Babuty D, Poret P, Autret ML, Cosnay P and Fauchier JP (1999). Effect of verapamil on QT interval dynamicity. Am J Cardiol 83(5):807–8, A10–1.

Ficker E, Jarolimek W and Brown AM (2001). Molecular determinants of inactivation and dofetilide block in ether a-go-go (EAG) channels and EAG-related K(+) channels. Mol Pharmacol 60(6):1343–1348.

Ficker E, Jarolimek W, Kiehn J, Baumann A and Brown AM (1998). Molecular determinants of dofetilide block of HERG K+ channels. Circ Res 82(3):386–95.

Fink M and Noble D (2009). Markov models for ion channels: versatility versus identiﬁa- bility and speed. Philos Transact A Math Phys Eng Sci 367(1896):2161–79.

Fossa AA, DePasquale MJ, Raunig DL, Avery MJ and Leishman DJ (2002). The relationship of clinical QT prolongation to outcome in the conscious dog using a beat-to-beat QT-RR interval assessment. J Pharmacol Exp Ther 302(2):828–33.

Gaita F, Giustetto C, Bianchi F, Schimpf R, Haissaguerre M, Calo` L, Brugada R, Antzele- vitch C, Borggrefe M and Wolpert C (2004). Short QT syndrome: pharmacological treatment. J Am Coll Cardiol 43(8):1494–9.

Gima K and Rudy Y (2002). Ionic current basis of electrocardiographic waveforms: a model study. Circ Res 90(8):889–96.

Goldenberg I, Moss AJ and Zareba W (2006). QT interval: how to measure it and what is ”normal”. J Cardiovasc Electrophysiol 17(3):333–336.

Gomez-Varela´ D, Contreras-Jurado C, Furini S, Garc´ıa-Ferreiro R, Stuhmer¨ W and Pardo LA (2006). Different relevance of inactivation and F468 residue in the mechanisms of hEag1 channel blockage by astemizole, imipramine and dofetilide. FEBS Lett 580(21):5059–66.

Guo D, Zhou J, Zhao X, Gupta P, Kowey PR, Martin J, Wu Y, Liu T and Yan GX (2008). L-type calcium current recovery versus ventricular repolarization: preserved membrane-stabilizing mechanism for different QT intervals across species. Heart Rhythm 5(2):271–9.

Guo J, Gang H and Zhang S (2006). Molecular determinants of cocaine block of human ether-a-go-go-related´ gene potassium channels. J Pharmacol Exp Ther 317(2):865– 74.

118 Guth BD (2007). Preclinical cardiovascular risk assessment in modern drug development. Toxicol Sci 97(1):4–20.

Halkin A, Roth A, Lurie I, Fish R, Belhassen B and Viskin S (2001). Pause-dependent torsade de pointes following acute myocardial infarction: a variant of the acquired long QT syndrome. J Am Coll Cardiol 38(4):1168–1174.

Hamill OP, Marty A, Neher E, Sakmann B and Sigworth FJ (1981). Improved patch- clamp techniques for high-resolution current recording from cells and cell-free membrane patches. Pﬂugers Arch 391(2):85–100.

Harmar AJ, Hills RA, Rosser EM, Jones M, Buneman OP, Dunbar DR, Greenhill SD, Hale VA, Sharman JL, Bonner TI, Catterall WA, Davenport AP, Delagrange P, Dollery CT, Foord SM, Gutman GA, Laudet V, Neubig RR, Ohlstein EH, Olsen RW, Peters J, Pin JP, Ruffolo RR, Searls DB, Wright MW and Spedding M (2009). IUPHAR-DB: the IUPHAR database of G protein-coupled receptors and ion channels. Nucleic Acids Res 37(Database issue):D680–5.

Herzberg IM, Trudeau MC and Robertson GA (1998). Transfer of rapid inactivation and sensitivity to the class III antiarrhythmic drug E-4031 from HERG to M-eag channels. J Physiol 511(Pt1):3–14.

Hille B (1971). The permeability of the sodium channel to organic cations in myelinated nerve. J Gen Physiol 58(6):599–619.

Hille B (2001). Ion channels of excitable membranes. Sinauer, Sunderland, Mass., 3rd ed edition. ISBN 0878933212.

Hodgkin AL and Huxley AF (1952). A quantitative description of membrane current and its application to conduction and excitation in nerve. J Physiol 117(4):500–544.

Hoffmann P and Warner B (2006). Are hERG channel inhibition and QT interval prolongation all there is in drug-induced torsadogenesis? A review of emerging trends. Journal of pharmacological and toxicological methods 53(2):87–105.

Hondeghem LM (2007). Relative contributions of TRIaD and QT to proarrhythmia. J Cardiovasc Electrophysiol 18(6):655–7.

Hondeghem LM, Lu HR, van Rossem K and De Clerck F (2003). Detection of proarrhythmia in the female rabbit heart: blinded validation. J Cardiovasc Electrophysiol 14(3):287–94.

Hong Y, Rautaharju PM, Hopkins PN, Arnett DK, Djousse L, Pankow JS, Sholinsky P,Rao DC and Province MA (2001). Familial aggregation of QT-interval variability in a general population: results from the NHLBI Family Heart Study. Clin Genet 59(3):171–177.

119 Jackman WM, Friday KJ, Anderson JL, Aliot EM, Clark M and Lazzara R (1988). The long QT syndromes: a critical review, new clinical observations and a unifying hypothesis. Prog Cardiovasc Dis 31(2):115–172.

Jervell A and Lange-Nielsen F (1957). Congenital deaf-mutism, functional heart disease with prolongation of the Q-T interval and sudden death. Am Heart J 54(1):59–68.

Jiang C, Atkinson D, Towbin JA, Splawski I, Lehmann MH, Li H, Timothy K, Taggart RT, Schwartz PJ and Vincent GM (1994). Two long QT syndrome loci map to chromosomes 3 and 7 with evidence for further heterogeneity. Nat Genet 8(2):141–147.

Jiang Y, Lee A, Chen J, Ruta V, Cadene M, Chait BT and MacKinnon R (2003). X-ray structure of a voltage-dependent K+ channel. Nature 423(6935):33–41.

Jurkiewicz NK and Sanguinetti MC (1993). Rate-dependent prolongation of cardiac action potentials by a methanesulfonanilide class III antiarrhythmic agent. Speciﬁc block of rapidly activating delayed rectiﬁer K+ current by dofetilide. Circ Res 72(1):75–83.

Kamiya K, Niwa R, Morishima M, Honjo H and Sanguinetti MC (2008). Molecular determinants of hERG channel block by terfenadine and cisapride. J Pharmacol Sci 108(3):301–7.

Kaufman ES, McNitt S, Moss AJ, Zareba W, Robinson JL, Hall WJ, Ackerman MJ, Ben- horin J, Locati ET, Napolitano C, Priori SG, Schwartz PJ, Towbin JA, Vincent GM and Zhang L (2008). Risk of death in the long QT syndrome when a sibling has died. Heart Rhythm 5(6):831–836.

Keating M, Atkinson D, Dunn C, Timothy K, Vincent GM and Leppert M (1991). Linkage of a cardiac arrhythmia, the long QT syndrome, and the Harvey ras-1 gene. Science 252(5006):704–706.

Kiehn J, Lacerda AE, Wible B and Brown AM (1996). Molecular physiology and pharmacology of HERG. Single-channel currents and block by dofetilide. Circulation 94(10):2572–9.

Kirsch GE, Trepakova ES, Brimecombe JC, Sidach SS, Erickson HD, Kochan MC, Shyjka LM, Lacerda AE and Brown AM (2004). Variability in the measurement of hERG potassium channel inhibition: effects of temperature and stimulus pattern. J Pharmacol Toxicol Methods 50(2):93–101.

Lacerda AE, Kuryshev YA, Chen Y, Renganathan M, Eng H, Danthi SJ, Kramer JW, Yang T and Brown AM (2008). Alfuzosin delays cardiac repolarization by a novel mechanism. J Pharmacol Exp Ther 324(2):427–433.

120 Lees-Miller JP, Duan Y, Teng GQ and Duff HJ (2000). Molecular determinant of high- afﬁnity dofetilide binding to HERG1 expressed in Xenopus oocytes: involvement of S6 sites. Mol Pharmacol 57(2):367–74.

Lees-Miller JP, Kondo C, Wang L and Duff HJ (1997). Electrophysiological characteriza- tion of an alternatively processed ERG K+ channel in mouse and human hearts. Circ Res 81(5):719–726.

Lehnart SE, Ackerman MJ, Benson DW Jr, Brugada R, Clancy CE, Donahue JK, George AL Jr, Grant AO, Groft SC, January CT, Lathrop DA, Lederer WJ, Makielski JC, Mohler PJ, Moss A, Nerbonne JM, Olson TM, Przywara DA, Towbin JA, Wang LH and Marks AR (2007). Inherited arrhythmias: a National Heart, Lung, and Blood Institute and Ofﬁce of Rare Diseases workshop consensus report about the diagnosis, phenotyping, molecular mechanisms, and therapeutic approaches for primary cardiomyopathies of gene mutations affecting ion channel function. Circulation 116(20):2325–45.

Lenz TL and Hilleman DE (2000). Dofetilide, a new class III antiarrhythmic agent. Phar- macotherapy 20(7):776–86.

Limberis JT, Su Z, Cox BF, Gintant GA and Martin RL (2006). Altering extracellular potassium concentration does not modulate drug block of human ether-a-go-go-related gene (hERG) channels. Clin Exp Pharmacol Physiol 33(11):1059–65.

Lin C, Ke X, Cvetanovic I, Ranade V and Somberg J (2007). The effect of high extracellular potassium on IKr inhibition by anti-arrhythmic agents. Cardiology 108(1):18–27.

Lin J, Guo J, Gang H, Wojciechowski P, Wigle JT and Zhang S (2005). Intracellular K+ is required for the inactivation-induced high-afﬁnity binding of cisapride to HERG channels. Mol Pharmacol 68(3):855–65.

London B, Trudeau MC, Newton KP, Beyer AK, Copeland NG, Gilbert DJ, Jenkins NA, Satler CA and Robertson GA (1997). Two isoforms of the mouse ether-a-go-go-related gene coassemble to form channels with properties similar to the rapidly activating component of the cardiac delayed rectiﬁer K+ current. Circ Res 81(5):870–8.

Long SB, Campbell EB and Mackinnon R (2005). Voltage sensor of Kv1.2: structural basis of electromechanical coupling. Science 309(5736):903–908.

Lu Y, Mahaut-Smith MP, Varghese A, Huang CL, Kemp PR and Vandenberg JI (2001). Effects of premature stimulation on HERG K(+) channels. J Physiol 537(Pt 3):843–51.

Lubinski A, Lewicka-Nowak E, Kempa M, Baczynska AM, Romanowska I and Swiate- cka G (1998). New insight into repolarization abnormalities in patients with congenital

121 long QT syndrome: the increased transmural dispersion of repolarization. Pacing Clin Electrophysiol 21(1 Pt 2):172–175.

Luo T, Luo A, Liu M and Liu X (2008). Inhibition of the HERG channel by droperidol depends on channel gating and involves the S6 residue F656. Anesth Analg 106(4):1161–70, table of contents.

Masetti M, Cavalli A and Recanatini M (2008). Modeling the hERG potassium channel in a phospholipid bilayer: Molecular dynamics and drug docking studies. J Comput Chem 29(5):795–808.

Mazhari R, Greenstein JL, Winslow RL, Marban´ E and Nuss HB (2001). Molecular interactions between two long-QT syndrome gene products, HERG and KCNE2, rational- ized by in vitro and in silico analysis. Circ Res 89(1):33–8.

McPate MJ, Duncan RS, Witchel HJ and Hancox JC (2006). Disopyramide is an effective inhibitor of mutant HERG K+ channels involved in variant 1 short QT syndrome. J Mol Cell Cardiol 41(3):563–6.

Meissner F (1856). Taubstummheit and Taubstummenbildung. Leipzig and Heidelberg.

Mitcheson JS (2008). hERG potassium channels and the structural basis of drug-induced arrhythmias. Chem Res Toxicol 21(5):1005–10.

Mitcheson JS, Chen J and Sanguinetti MC (2000). Trapping of a methanesulfonanilide by closure of the HERG potassium channel activation gate. J Gen Physiol 115(3):229–40.

Mohammad S, Zhou Z, Gong Q and January CT (1997). Blockage of the HERG human cardiac K+ channel by the gastrointestinal prokinetic agent cisapride. Am J Physiol 273(5 Pt 2):H2534–8.

Moise NS (1999). As Americans, we should get this right. Circulation 100(13):1462.

Motte G, Coumel P, Abitbol G, Dessertenne F and Slama R (1970). [The long QT syndrome and syncope caused by spike torsades]. Arch Mal Coeur Vaiss 63(6):831–853.

Myokai T, Ryu S, Shimizu H and Oiki S (2008). Topological mapping of the asymmetric drug-binding to the human ether-a-go-go-related gene product (HERG) potassium channel by use of tandem dimers. Mol Pharmacol 73(6):1643–1651.

Nabauer¨ M, Beuckelmann DJ, Uberfuhr P and Steinbeck G (1996). Regional differences in current density and rate-dependent properties of the transient outward current in subepicardial and subendocardial myocytes of human left ventricle. Circulation 93(1):168–77.

122 Napolitano C, Schwartz PJ, Brown AM, Ronchetti E, Bianchi L, Pinnavaia A, Acquaro G and Priori SG (2000). Evidence for a cardiac ion channel mutation underlying drug- induced QT prolongation and life-threatening arrhythmias. J Cardiovasc Electrophysiol 11(6):691–696.

Netzer R, Bischoff U and Ebneth A (2003). HTS techniques to investigate the potential effects of compounds on cardiac ion channels at early-stages of drug discovery. Curr Opin Drug Discov Devel 6(4):462–9.

Newton-Cheh C, Larson MG, Corey DC, Benjamin EJ, Herbert AG, Levy D, D’Agostino RB and O’Donnell CJ (2005). QT interval is a heritable quantitative trait with evidence of linkage to chromosome 3 in a genome-wide linkage analysis: The Framingham Heart Study. Heart Rhythm 2(3):277–284.

Numaguchi H, Mullins FM, Johnson JP Jr, Johns DC, Po SS, Yang IC, Tomaselli GF and Balser JR (2000). Probing the interaction between inactivation gating and Dd-sotalol block of HERG. Circ Res 87(11):1012–8.

Opthof T, Coronel R, Janse M and Rosen M (2007a). A wedge is not a heart. Heart Rhythm 4(8):1116–1119.

Opthof T, Coronel R, Wilms-Schopman FJG, Plotnikov AN, Shlapakova IN, Danilo PJ, Rosen MR and Janse MJ (2007b). Dispersion of repolarization in canine ventricle and the electrocardiographic T wave: Tp-e interval does not reﬂect transmural dispersion. Heart Rhythm 4(3):341–348.

Pardo-Lopez L, Zhang M, Liu J, Jiang M, Possani LD and Tseng GN (2002). Mapping the binding site of a human ether-a-go-go-related gene-speciﬁc peptide toxin (ErgTx) to the channel’s outer vestibule. J Biol Chem 277(19):16403–11.

Perrin MJ, Kuchel PW, Campbell TJ and Vandenberg JI (2008). drug-binding to the inactivated state is necessary but not sufﬁcient for high-afﬁnity binding to human ether-a- go-go-related gene channels. Mol Pharmacol 74(5):1443–1452.

Petrov DB (2003). QT interval lengthening after fasting complicated by a sudden attack of torsades de pointes. Tex Heart Inst J 30(1):86–87.

Plonsey R and Barr RC (2007). Bioelectricity: a quantitative approach. Springer, New York, NY, 3rd ed edition. ISBN 9780387488646.

Priori SG, Pandit SV, Rivolta I, Berenfeld O, Ronchetti E, Dhamoon A, Napolitano C, Anumonwo J, di Barletta MR, Gudapakkam S, Bosi G, Stramba-Badiale M and Jalife J (2005). A novel form of short QT syndrome (SQT3) is caused by a mutation in the KCNJ2 gene. Circ Res 96(7):800–7.

123 Priori SG, Schwartz PJ, Napolitano C, Bloise R, Ronchetti E, Grillo M, Vicentini A, Spaz- zolini C, Nastoli J, Bottelli G, Folli R and Cappelletti D (2003). Risk stratiﬁcation in the long-QT syndrome. N Engl J Med 348(19):1866–1874.

Recanatini M, Cavalli A and Masetti M (2008). Modeling HERG and its interactions with drugs: recent advances in light of current potassium channel simulations. ChemMed- Chem 3(4):523–535.

Rezazadeh S, Hesketh JC and Fedida D (2004). Rb+ ﬂux through hERG channels affects the potency of channel blocking drugs: correlation with data obtained using a high- throughput Rb+ efﬂux assay. J Biomol Screen 9(7):588–97.

Roden DM (1998). Taking the ”idio” out of ”idiosyncratic”: predicting torsades de pointes. Pacing Clin Electrophysiol 21(5):1029–1034.

Roden DM (2004). Drug-induced prolongation of the QT interval. N Engl J Med 350(10):1013–1022.

Roden DM, Lazzara R, Rosen M, Schwartz PJ, Towbin J and Vincent GM (1996). Multiple mechanisms in the long-QT syndrome. Current knowledge, gaps, and future directions. The SADS Foundation Task Force on LQTS. Circulation 94(8):1996–2012.

Rodriguez-Sinovas A, Cinca J, Tapias A, Armadans L, Tresanchez M and Soler-Soler J (1997). Lack of evidence of M-cells in porcine left ventricular myocardium. Cardiovasc Res 33(2):307–313.

Romano C, Gemme G and Pongiglione R (1963). [Rare cardiac arrhythmias of the paediatric age: syncopal attacks due to ventricular ﬁbrillation. (Presentation of 1st case in Italian Paediatric Literature]. Clin Pediatr (Bologna) 45:656–683.

Romero L, Pueyo E, Fink M and Rodr´ıguez B (2009). Impact of ionic current variability on human ventricular cellular electrophysiology. Am J Physiol Heart Circ Physiol 297(4):H1436–45.

Saenen JB, Paulussen ADC, Jongbloed RJ, Marcelis CL, Gilissen RAHJ, Aerssens J, Snyders DJ and Raes AL (2007). A single hERG mutation underlying a spectrum of acquired and congenital long QT syndrome phenotypes. J Mol Cell Cardiol 43(1):63– 72.

Sanguinetti MC, Jiang C, Curran ME and Keating MT (1995). A mechanistic link between an inherited and an acquired cardiac arrhythmia: HERG encodes the IKr potassium channel. Cell 81(2):299–307.

124 Schonherr¨ R and Heinemann SH (1996). Molecular determinants for activation and inactivation of HERG, a human inward rectiﬁer potassium channel. J Physiol 493 ( Pt 3):635–42.

Schram G, Pourrier M, Melnyk P and Nattel S (2002). Differential distribution of cardiac ion channel expression as a basis for regional specialization in electrical function. Circ Res 90(9):939–950.

Schroeder K, Neagle B, Trezise DJ and Worley J (2003). Ionworks HT: a new high- throughput electrophysiology measurement platform. J Biomol Screen 8(1):50–64.

Schwartz PJ and Malliani A (1975). Electrical alternation of the T-wave: clinical and experimental evidence of its relationship with the sympathetic nervous system and with the long Q-T syndrome. Am Heart J 89(1):45–50.

Schwartz PJ, Periti M and Malliani A (1975). The long Q-T syndrome. Am Heart J 89(3):378–390.

Schwartz PJ, Priori SG and Napolitano C (2003). How really rare are rare diseases?: the intriguing case of independent compound mutations in the long QT syndrome. J Cardiovasc Electrophysiol 14(10):1120–1121.

Schwartz PJ, Priori SG, Spazzolini C, Moss AJ, Vincent GM, Napolitano C, Denjoy I, Guicheney P,Breithardt G, Keating MT, Towbin JA, Beggs AH, Brink P,Wilde AA, Toivo- nen L, Zareba W, Robinson JL, Timothy KW, Corﬁeld V, Wattanasirichaigoon D, Corbett C, Haverkamp W, Schulze-Bahr E, Lehmann MH, Schwartz K, Coumel P and Bloise R (2001). Genotype-phenotype correlation in the long-QT syndrome: gene-speciﬁc triggers for life-threatening arrhythmias. Circulation 103(1):89–95.

Schwartz SP and Hallinger LN (1954). Transient ventricular ﬁbrillation. VI. Observations on the peripheral arterial pulse pressures in the course of transient ventricular ﬁbrilla- tion during established auriculoventricular dissociation. Am Heart J 48(3):390–404.

Sesti F, Abbott GW, Wei J, Murray KT, Saksena S, Schwartz PJ, Priori SG, Roden DM, George ALJ and Goldstein SA (2000). A common polymorphism associated with antibiotic-induced cardiac arrhythmia. Proc Natl Acad SciUSA97(19):10613–10618.

Shah RR and Hondeghem LM (2005). Reﬁning detection of drug-induced proarrhythmia: QT interval and TRIaD. Heart Rhythm 2(7):758–72.

Shimizu M, Ino H, Okeie K, Yamaguchi M, Nagata M, Hayashi K, Itoh H, Iwaki T, Oe K, Konno T and Mabuchi H (2002). T-peak to T-end interval may be a better predictor of high-risk patients with hypertrophic cardiomyopathy associated with a cardiac troponin I mutation than QT dispersion. Clin Cardiol 25(7):335–339.

125 Sicouri S and Antzelevitch C (1991). A subpopulation of cells with unique electrophysiological properties in the deep subepicardium of the canine ventricle. The M cell. Circ Res 68(6):1729–41.

Smith PL, Baukrowitz T and Yellen G (1996). The inward rectiﬁcation mechanism of the HERG cardiac potassium channel. Nature 379(6568):833–6.

Spector PS, Curran ME, Zou A, Keating MT and Sanguinetti MC (1996). Fast inactivation causes rectiﬁcation of the IKr channel. J Gen Physiol 107(5):611–9.

Splawski I, Shen J, Timothy KW, Lehmann MH, Priori S, Robinson JL, Moss AJ, Schwartz PJ, Towbin JA, Vincent GM and Keating MT (2000). Spectrum of mutations in long- QT syndrome genes. KVLQT1, HERG, SCN5A, KCNE1, and KCNE2. Circulation 102(10):1178–1185.

Splawski I, Timothy KW, Tateyama M, Clancy CE, Malhotra A, Beggs AH, Cappuccio FP, Sagnella GA, Kass RS and Keating MT (2002). Variant of SCN5A sodium channel implicated in risk of cardiac arrhythmia. Science 297(5585):1333–1336.

Stork D, Timin EN, Berjukow S, Huber C, Hohaus A, Auer M and Hering S (2007). State dependent dissociation of HERG channel inhibitors. Br J Pharmacol 151(8):1368–76.

Strauss HC, Bigger JT Jr and Hoffman BF (1970). Electrophysiologial and beta-receptor blocking effects of MJ 1999 on dog and rabbit cardiac tissue. Circ Res 26(6):661–78.

Swan H, Viitasalo M, Piippo K, Laitinen P, Kontula K and Toivonen L (1999). Sinus node function and ventricular repolarization during exercise stress test in long QT syndrome patients with KvLQT1 and HERG potassium channel defects. J Am Coll Cardiol 34(3):823–829.

Tang W, Kang J, Wu X, Rampe D, Wang L, Shen H, Li Z, Dunnington D and Garyantes T (2001). Development and evaluation of high throughput functional assay methods for HERG potassium channel. J Biomol Screen 6(5):325–31.

Terrenoire C, Clancy CE, Cormier JW, Sampson KJ and Kass RS (2005). Autonomic control of cardiac action potentials: role of potassium channel kinetics in response to sympathetic stimulation. Circ Res 96(5):e25–34.

Thomsen MB, Verduyn SC, Stengl M, Beekman JDM, de Pater G, van Opstal J, Volders PGA and Vos MA (2004). Increased short-term variability of repolarization predicts d-sotalol-induced torsades de pointes in dogs. Circulation 110(16):2453–9.

Torres AM, Bansal PS, Sunde M, Clarke CE, Bursill JA, Smith DJ, Bauskin A, Breit SN, Campbell TJ, Alewood PF, Kuchel PW and Vandenberg JI (2003). Structure of the

126 HERG K+ channel S5P extracellular linker: role of an amphipathic alpha-helix in C- type inactivation. J Biol Chem 278(43):42136–48.

Trudeau MC, Warmke JW, Ganetzky B and Robertson GA (1995). HERG, a human inward rectiﬁer in the voltage-gated potassium channel family. Science 269(5220):92– 5.

Tsujimae K, Suzuki S, Yamada M and Kurachi Y (2004). Comparison of kinetic properties of quinidine and dofetilide block of HERG channels. Eur J Pharmacol 493(1-3):29–40. ten Tusscher KHWJ, Noble D, Noble PJ and Panﬁlov AV (2004). A model for human ventricular tissue. Am J Physiol Heart Circ Physiol 286(4):H1573–89. ten Tusscher KHWJ and Panﬁlov AV (2006). Alternans and spiral breakup in a human ventricular tissue model. Am J Physiol Heart Circ Physiol 291(3):H1088–100.

Vaglio M, Couderc JP, McNitt S, Xia X, Moss AJ and Zareba W (2008). A quantitative assessment of T-wave morphology in LQT1, LQT2, and healthy individuals based on Holter recording technology. Heart Rhythm 5(1):11–8.

Vincent GM (2005). Risk assessment in long QT syndrome: the Achilles heel of appropriate treatment. Heart Rhythm 2(5):505–506.

Vincent GM, Schwartz PJ, Denjoy I, Swan H, Bithell C, Spazzolini C, Crotti L, Piippo K, Lupoglazoff JM, Villain E, Priori SG, Napolitano C and Zhang L (2009). High efﬁcacy of beta-blockers in long-QT syndrome type 1: contribution of noncompliance and QT- prolonging drugs to the occurrence of beta-blocker treatment ”failures”. Circulation 119(2):215–221.

Volberg WA, Koci BJ, Su W, Lin J and Zhou J (2002). Blockade of human cardiac potassium channel human ether-a-go-go-related gene (HERG) by macrolide antibiotics. J Pharmacol Exp Ther 302(1):320–7.

Vyas H and Ackerman MJ (2006). Epinephrine QT stress testing in congenital long QT syndrome. J Electrocardiol 39(4 Suppl):S107–13.

Wain HM, Bruford EA, Lovering RC, Lush MJ, Wright MW and Povey S (2002). Guidelines for human gene nomenclature. Genomics 79(4):464–470.

Walker AM, Szneke P, Weatherby LB, Dicker LW, Lanza LL, Loughlin JE, Yee CL and Dreyer NA (1999a). The risk of serious cardiac arrhythmias among cisapride users in the United Kingdom and Canada. Am J Med 107(4):356–62.

127 Walker BD, Singleton CB, Bursill JA, Wyse KR, Valenzuela SM, Qiu MR, Breit SN and Campbell TJ (1999b). Inhibition of the human ether-a-go-go-related gene (HERG) potassium channel by cisapride: afﬁnity for open and inactivated states. Br J Pharma- col 128(2):444–50.

Walker BD, Valenzuela SM, Singleton CB, Tie H, Bursill JA, Wyse KR, Qiu MR, Breit SN and Campbell TJ (1999c). Inhibition of HERG channels stably expressed in a mammalian cell line by the antianginal agent perhexiline maleate. Br J Pharmacol 127(1):243–51.

Wallis RM (2009). Integrated risk assessment and predictive value to humans of non- clinical repolarization assays. Br J Pharmacol .

Wang HZ, Shi H, Liao SJ and Wang Z (1999). Inactivation gating determines nicotine blockade of human HERG channels. Am J Physiol 277(3 Pt 2):H1081–8.

Wang Q, Curran ME, Splawski I, Burn TC, Millholland JM, VanRaay TJ, Shen J, Timothy KW, Vincent GM, de Jager T, Schwartz PJ, Toubin JA, Moss AJ, Atkinson DL, Landes GM, Connors TD and Keating MT (1996). Positional cloning of a novel potassium channel gene: KVLQT1 mutations cause cardiac arrhythmias. Nat Genet 12(1):17–23.

Wang Q, Shen J, Splawski I, Atkinson D, Li Z, Robinson JL, Moss AJ, Towbin JA and Keating MT (1995). SCN5A mutations associated with an inherited cardiac arrhythmia, long QT syndrome. Cell 80(5):805–811.

Wang S, Liu S, Morales MJ, Strauss HC and Rasmusson RL (1997a). A quantitative analysis of the activation and inactivation kinetics of HERG expressed in Xenopus oocytes. J Physiol (Lond) 502(Pt1):45–60.

Wang S, Morales MJ, Liu S, Strauss HC and Rasmusson RL (1997b). Modulation of HERG afﬁnity for E-4031 by [K+]o and C-type inactivation. FEBS Lett 417(1):43–7.

Ward OC (1964). A new familial cardiac syndrome in children. J Ir Med Assoc 54:103– 106.

Warmke JW and Ganetzky B (1994). A family of potassium channel genes related to eag in Drosophila and mammals. Proc Natl Acad Sci USA 91(8):3438–42.

Weerapura M, Hebert´ TE and Nattel S (2002). Dofetilide block involves interactions with open and inactivated states of HERG channels. Pﬂugers Arch 443(4):520–31.

Weiss JN, Garﬁnkel A, Karagueuzian HS, Qu Z and Chen PS (1999). Chaos and the transition to ventricular ﬁbrillation: a new approach to antiarrhythmic drug evaluation. Circulation 99(21):2819–26.

128 Whellan DJ, Green CL, Piccini JP and Krucoff MW (2009). QT as a safety biomarker in drug development. Clin Pharmacol Ther 86(1):101–4.

Wysowski DK and Bacsanyi J (1996). Cisapride and fatal arrhythmia. N Engl J Med 335(4):290–1.

Xia Y, Liang Y, Kongstad O, Liao Q, Holm M, Olsson B and Yuan S (2005). In vivo validation of the coincidence of the peak and end of the T wave with full repolarization of the epicardium and endocardium in swine. Heart Rhythm 2(2):162–169.

Yamaguchi M, Shimizu M, Ino H, Terai H, Uchiyama K, Oe K, Mabuchi T, Konno T, Kaneda T and Mabuchi H (2003). T wave peak-to-end interval and QT dispersion in acquired long QT syndrome: a new index for arrhythmogenicity. Clin Sci (Lond) 105(6):671–676.

Yan GX, Shimizu W and Antzelevitch C (1998a). Characteristics and distribution of M cells in arterially perfused canine left ventricular wedge preparations. Circulation 98(18):1921–7.

Yan GX, Shimizu W and Antzelevitch C (1998b). Characteristics and distribution of M cells in arterially perfused canine left ventricular wedge preparations. Circulation 98(18):1921–7.

Yang Bf, Xu Dh, Xu Cq, Li Z, Du Zm, Wang Hz and Dong Dl (2004). Inactivation gating determines drug potency: a common mechanism for drug blockade of HERG channels. Acta Pharmacol Sin 25(5):554–60.

Yang P, Kanki H, Drolet B, Yang T, Wei J, Viswanathan PC, Hohnloser SH, Shimizu W, Schwartz PJ, Stanton M, Murray KT, Norris K, George ALJ and Roden DM (2002). Allelic variants in long-QT disease genes in patients with drug-associated torsades de pointes. Circulation 105(16):1943–1948.

Yang T and Roden DM (1996). Extracellular potassium modulation of drug block of IKr. Implications for torsade de pointes and reverse use-dependence. Circulation 93(3):407–11.

Yang T, Snyders D and Roden DM (2001). Drug block of I(kr): model systems and relevance to human arrhythmias. J Cardiovasc Pharmacol 38(5):737–744.

Yang T, Snyders DJ and Roden DM (1997). Rapid inactivation determines the rectiﬁcation and [K+]o dependence of the rapid component of the delayed rectiﬁer K+ current in cardiac cells. Circ Res 80(6):782–9.

Yanowitz F, Preston JB and Abildskov JA (1966). Functional distribution of right and left stellate innervation to the ventricles. Production of neurogenic electrocardiographic changes by unilateral alteration of sympathetic tone. Circ Res 18(4):416–428.

129 Yi H, Cao Z, Yin S, Dai C, Wu Y and Li W (2007). Interaction simulation of hERG K+ channel with its speciﬁc BeKm-1 peptide: insights into the selectivity of molecular recogni- tion. J Proteome Res 6(2):611–20.

Zareba W, Moss AJ, Schwartz PJ, Vincent GM, Robinson JL, Priori SG, Benhorin J, Lo- cati EH, Towbin JA, Keating MT, Lehmann MH and Hall WJ (1998). Inﬂuence of genotype on the clinical course of the long-QT syndrome. International Long-QT Syndrome Registry Research Group. N Engl J Med 339(14):960–965.

Zhang L, Timothy KW, Vincent GM, Lehmann MH, Fox J, Giuli LC, Shen J, Splawski I, Priori SG, Compton SJ, Yanowitz F, Benhorin J, Moss AJ, Schwartz PJ, Robinson JL, Wang Q, Zareba W, Keating MT, Towbin JA, Napolitano C and Medina A (2000). Spectrum of ST-T-wave patterns and repolarization parameters in congenital long-QT syndrome: ECG ﬁndings identify genotypes. Circulation 102(23):2849–2855.

Zhou Q, Zygmunt AC, Cordeiro JM, Siso-Nadal F, Miller RE, Buzzard GT and Fox JJ (2009). Identiﬁcation of Ikr kinetics and drug-binding in native myocytes. Ann Biomed Eng 37(7):1294–309.

Zhou Z, Gong Q, Ye B, Fan Z, Makielski JC, Robertson GA and January CT (1998). Properties of HERG channels stably expressed in HEK 293 cells studied at physiological temperature. Biophys J 74(1):230–241.

Zhou Z, Vorperian VR, Gong Q, Zhang S and January CT (1999). Block of HERG potassium channels by the antihistamine astemizole and its metabolites desmethylastemi- zole and norastemizole. J Cardiovasc Electrophysiol 10(6):836–43.

Zwillinger L (1935). Uber¨ die Magnesiumwirkung auf das Herz. Journal of Molecular Medicine 14(40):1429–1433.

Zygmunt AC, Robitelle DC and Eddlestone GT (1997). Ito1 dictates behavior of ICl(Ca) during early repolarization of canine ventricle. Am J Physiol 273(3 Pt 2):H1096–106.

130