Ranolazine for Congenital and Acquired Late I_Na Linked Arrhythmias: In Silico Pharmacologic Screening

Abstract

Rationale

The antianginal ranolazine blocks the hERG-based current I_Kr at therapeutic concentrations and causes QT interval prolongation. Thus, ranolazine is contraindicated for patients with preexisting long QT and those with repolarization abnormalities. However, with its preferential targeting of late I_Na (I_NaL), patients with disease resulting from increased I_NaL from inherited defects (e.g. Long QT syndrome type-3, LQT3), or disease induced electrical remodeling (e.g. ischemic heart failure), might be exactly the ones to most benefit from the presumed antiarrhythmic properties of ranolazine.

Objective

We developed a computational model to predict if therapeutic effects of pharmacological targeting of late I_Na by ranolazine prevailed over the off-target block of hERG in the setting of inherited LQT3 and heart failure.

Methods and Results

We developed computational models describing the kinetics the interaction of ranolazine with cardiac Na⁺ channels in the setting of normal physiology, LQT3 linked ΔKPQ mutation, and heart failure. We then simulated clinically relevant concentrations of ranolazine and predicted the combined effects of Na⁺ channel and hERG blockade by both the parent compound ranolazine and its active metabolites, which have shown potent blocking effects in the therapeutically relevant range. Our simulations suggest that ranolazine is effective at normalizing arrhythmia triggers in bradycardia-dependent arrhythmias in LQT3 as well tachyarrhythmogenic triggers arising from heart failure-induced remodeling.

Conclusions

Our model predictions suggest that acute targeting of late I_Na with ranolazine may be an effective therapeutic strategy in diverse arrhythmia provoking situations that arise from a common pathway of increased pathologic late I_Na.

INTRODUCTION

The CAST and SWORD trials showed that common antiarrhythmics increased mortality and risk of sudden cardiac death in post-infarction patients ^1-3. Almost thirty years after CAST there is still no way to differentiate potentially useful and potentially harmful drugs ⁴.

Classification of antiarrhythmic drugs is based on the drug's primary effect, known as the Singh-Vaughn Williams classification ⁵. While this method is straightforward, it fails to describe the complex kinetics of the drug-channel interaction, contributions from charged and uncharged species, and for the effects of non-specific drugs on multiple ion channels. For example, while lidocaine is specific for Na⁺ channels ⁵, flecainide blocks K⁺ channels in some species ^{6, 7}, the ryanodine receptor ⁸, and the L-type calcium current ⁹ in others. Drug-channel interactions are also modified by cellular action potential properties including morphology, duration and frequency. Strong bi-directional feedback exists because drugs alter the action potential waveform, which in turn affects the potency of drugs. Electrotonic coupling in tissue leads to even more complex responses to drug application that may not be apparent in cellular level studies.

Off-target effects are of special interest for the novel antianginal agent ranolazine. While ranolazine preferentially blocks the late component of the Na⁺ current l_NaL (a depolarizing current), ranolazine also interacts with and reduces the repolarizing hERG current I_Kr with therapeutic concentrations ¹⁰. The result is a mild concentration dependent QTc prolongation ¹¹. Ranolazine is thus contraindicated for patients on other QT prolonging drugs, those with preexisting QT prolongation ¹², or ostensibly those with any form of decreased repolarization reserve. However, patients with QT prolongation from increased I_NaL from either inherited defects or disease induced electrical remodeling might be exactly the ones to most benefit from selective targeting of l_NaL.

The wealth of genetic information in recent years has led to an increased understanding of how genotype underlies clinical phenotype. For example, the long QT variant 3 (LQT3), a subset of the congenital long QT syndrome, is a group of inherited Na⁺ channel mutations that are characterized by a delay in cardiac cellular repolarization, manifesting as a prolongation of the QT interval on the ECG, resultant cardiac arrhythmias and sudden death ¹³. LQT3 mutations manifest clinically similar (a prolongation on the ECG), but are heterogeneous in mechanisms. Thus, it is not surprising that they also exhibit varied responses to antiarrhythmic drugs ^14-17.

Acquired arrhythmia syndromes such as ischemic heart failure present their own specific challenge: namely the vast heterogeneity of disease phenotypes and the continuum of severity existing throughout the natural course of disease. As opposed to genetically linked ion channel mutations, which have a fairly defined mechanistic basis, the emergent effects of heart failure associated arrhythmias are the result of multiple intersecting, deranged, and physiologic compensatory processes ¹⁸.

Another confounding factor in accurate interpretation of antiarrhythmic drug effects results from drug metabolism. For example, ranolazine is extensively biotransformed into active metabolites that exhibit strikingly different affinities to cardiac ion channels than the parent compound ^{19, 20}. Thus, ex vivo channel expression and cell studies done outside the physiologic milieu where drug metabolism is absent must be interpreted with caution.

In a recent study ²¹, we developed a computational modeling approach, informed and validated by experimental data, that simulated the interaction kinetics of the antiarrhythmic drugs flecainide and lidocaine with cardiac Na⁺ channels. We then used the model to predict the drug effects on human ventricular cellular and tissue electrical activity and in the setting of one common arrhythmia trigger, spontaneous ventricular ectopy. The model predicted when clinically relevant concentrations of the antiarrhythmic drugs flecainide and lidocaine would exacerbate, rather than ameliorate, arrhythmia. Here we expand this computational framework to predict the effects of promising genotype-specific therapeutic candidates for inherited LQT3-linked arrhythmias as well as acquired arrhythmia syndromes (e.g. heart failure) on emergent electrical activity in virtual cells and tissue. Computational analyses of disease-specific alterations and pharmacology present an opportunity to screen drugs for improved phenotype for a given disease process, and when a drug should be excluded if it exacerbates arrhythmogenic potential.

METHODS

Full methods are contained in the online Supplementary material. Source code is available upon request.

RESULTS

To compare the potential usefulness of ranolazine as an antiarrhythmic in the setting of LQT3, we expanded an existing theoretical model of Na⁺ channel gating to include drug interactions (Supplementary material), which takes into account channel conformation dependence of drug accessibility and binding affinity, and channel kinetics after drug binding ^{22, 23}.

From experimentally obtained data, we first developed a model of ranolazine interaction with the wild-type cardiac Na⁺ channel as described in Supplementary material and in ²¹. A depiction of the model is shown in Figure 1A. The model contains eight discrete background states to represent the drug free channel conformations (black) and eight additional states (green) that represent drug bound channel states. We have also included 4 additional states (not shown for clarity) to represent channel bursting ²⁴: a small population of channels that transiently fail to inactivate, producing a persistent Na⁺ current that represents 0.1% of the peak Na⁺ current as described for WT Na⁺ channels ²⁵. The drug channel representation is based on assumptions derived from the modulated receptor hypothesis, which suggests that any discrete conformational state of the channel can exist in a drug free or drug bound form ^{21, 22}.

Figure 1. Schematic of the drug-channel interaction and simulated (lines) and experimental (symbols) ranolazine – WT Na⁺ channel interactions.

Panel (A) shows a schematic indicating drug free and drug bound model states. (B) Dose dependence of tonic block (TB) for peak (solid line) and late (dashed line) current. One depolarizing pulse from −100 mV to −10 mV was elicited. Block is peak current fraction normalized to drug free conditions. (C) Steady-state channel availability. Currents measured at −10 mV in drug free conditions (dotted), or with 10 μM ranolazine (solid) pulsed from −120 to −40 in 5 mV increments (normalized to tonic block at −120 mV). (D) Dose-dependence of use-dependent block (UDB) from 300 pulses to −10 mV for 25 ms from −100 mV at 5 Hz with indicated drug dose. Block is peak current at last pulse normalized to drug free conditions. (E) Frequency dependence of UDB. Protocol as in (D) at indicated pacing frequencies with 100 μM ranolazine. (F) Recovery from UDB induced by trains of pulses (−10 mV for 25 ms at 25 Hz) from −100 mV for drug free (dashed) and with ranolazine (solid). Test pulses (−10 mV) were after variable recovery intervals at −100 mV. Currents were normalized to tonic block. In B – F, the points are experiment, and lines are simulation. Yellow boxes in (B) and (D) indicate therapeutic ranolazine.

Ranolazine binds to closed cardiac Na⁺ channels (IC₅₀ = 165 μM), indicated by resting tonic block at hyperpolarized membrane potentials that favor the closed channel conformation as in Figure 1 panel B – dose dependent tonic block of peak current in solid line. Ranolazine also tonically blocks late Na⁺ current with higher affinity - IC₅₀ = 5-21 μM ^{10, 26}. Simulated tonic block of late current is shown in Figure 1B (dashed line). The clinically relevant concentration range of drug is shown in yellow. Unlike some Na⁺ channel blocking drugs ^{15, 27}, ranolazine does not bind to inactivated Na⁺ channels, since no shift in steady-state inactivation is observed upon drug application (panel C). In response to repetitive depolarization, ranolazine exhibits potent open state use-dependent block (UDB) (IC₅₀ = 100.5 μM), resulting from cumulative build-up of drug bound channels and incomplete recovery during the interstimulus interval (panel D) ²⁸. Use-dependent block by ranolazine is frequency dependent, with marked increases in UDB observed at faster pacing frequencies (panel E). Ranolazine also dramatically slows Na⁺ channel recovery from use-dependent block following a rapid series of depolarizing pulses (panel F).

We then optimized a model of the Long QT3 linked Na⁺ channel mutation ΔKPQ to fit experimental data obtained from drug-free ΔKPQ mutant channels (Online Figure I) using the methods as described for WT and as previously ²¹. Notably, aside from a bursting-induced persistent Na⁺ current of ~0.5% of the peak Na⁺ current, the ΔKPQ channel recovers slightly faster from inactivation, but has similar mean open time ²⁹, peak current density ³⁰, steady state availability ³¹, and activation ²⁹, making ΔKPQ a particularly well suited mutation to compare to wild-type for efficacy of mutation specific persistent Na⁺ current blockade.

Next, we modeled ranolazine effects on ΔKPQ mutant channels. Figure 2 shows the model fits (lines) to data (symbols) following parameter optimization for ranolazine interaction with ΔKPQ channels. The clinical range of drug is shown in yellow. Panel A shows the differential sensitivity of ranolazine to peak (IC₅₀ = 120.8 μM) and late (IC₅₀ = 12.7 μM) ΔKPQ current elicited by a single depolarizing pulse. The data summary (panel B) reveals that peak current from ΔKPQ channels is slightly more sensitive to blockade by ranolazine than WT (IC₅₀ = 120.8 μM vs. 165.2 μM). Ranolazine blocks late versus peak Na⁺ current preferentially by a factor of 9.5, consistent with previously published reports ^{32, 33}. Importantly, experiments indicate that even though ranolazine preferentially blocks late current, ΔKPQ is two-fold less sensitive to blockade of the late component of the Na⁺ current when compared to WT (12 μM vs. 6 μM, respectively). Ranolazine minimally shifted the ΔKPQ steady state inactivation curve (<2 mV), indicating low inactivated state affinity (panel C). At 5 Hz pacing, ΔKPQ channels exhibit a similar extent of use-dependent block by ranolazine as WT channels (IC₅₀ for peak Na⁺ current blockade, 83 μM and 100.5 μM, respectively) (panel D). However, the rate-dependence of UDB for ΔKPQ channels was markedly blunted compared to WT channels (compare panel E in Figure 1). Panel F shows that ΔKPQ channels also recover from UDB induced by 10 μM ranolazine faster than WT channels.

Figure 2. Simulated (lines) and experimental (symbols) ranolazine – ΔKPQ Mutant Na⁺ channel interactions.

(A) Dose dependence of tonic block (TB) for peak (solid) and late (dashed) current. One depolarizing pulse from −100 mV to -10 mV was elicited. Block is peak current normalized to drug free conditions. Late current is measured after a 200 ms depolarizing pulse. (B) Comparison of peak and late currents in WT and ΔKPQ mutant Na⁺ channels. (C) Steady-state channel availability. Currents measured at -10 mV in drug free conditions (dotted), or with 10 μM ranolazine (solid) pulsed from -120 mV to -40 mV in 5 mV increments (normalized to tonic block at −120 mV). (D) Dose-dependence of use-dependent block (UDB) from 300 pulses to −10 mV for 25 ms from −100 mV at 5 Hz with indicated drug dose. Block is peak current at last pulse normalized to drug free conditions. (E) Frequency dependence of UDB. Protocol is as in (D) at indicated pacing frequencies with 10 μM ranolazine. (F) Recovery from UDB induced by 100 pulses (−10 mV for 25 ms at 25 Hz) from a −100 mV in drug free conditions (dotted), 10 μM ranolazine (solid top), and 100 μM ranolazine (solid bottom). Test pulses (−10 mV) were after variable recovery intervals at -100 mV. Currents were normalized to tonic block. Yellow boxes in (A) and (D) indicate therapeutic ranolazine.

Ranolazine binds to the promiscuous drug target hERG

Ranolazine is a potentially promising therapeutic for LQT3 patients due to its targeting of late I_Na, but is contraindicated for LQTS patients due to off-target interactions with the promiscuous drug target hERG, which underlies the key human repolarizing current I_Kr. The rapid kinetic interaction of ranolazine with hERG yields frequency-independent block, thus allowing for a much simplified model representation compared to that required for the Na⁺ channel ³⁴. Thus, to account for the off-target interactions of ranolazine with hERG, we incorporated a concentration dependent block of I_Kr peak using a concentration response relationship with a Hill coefficient of 1 (n = 1). Multiple studies concur that ranolazine blocks hERG with an IC₅₀ of I_Kr = 12 μM ^{10, 34}. Clinical studies also suggest hERG blockade: administration of 2 – 6 μM ranolazine yields a proportional increase in QTc of 2 – 6 ms ^{26, 35}. We carried out simulations in a one-dimensional transmural tissue informed by human data (see Supplementary material) comprising 165 cardiac cells and simulated the effect of 6 μM ranolazine on the computed electrogram generated by the model. In the model, we observed a marked QTc prolongation of 40 ms, a prediction that was not consistent with the clinical data.

A survey of the literature revealed a plausible and testable explanation for the discrepant model predictions and clinical findings. Pharmacokinetic studies of ranolazine suggest extensive metabolism via CYP3A mediated pathways of biotransformation, with less than 5% of the parent compound unmetabolized ²⁰. Four predominant metabolites were identified in healthy volunteers at plasma concentrations 30 – 40% of the parent compound, all of which produce a substantially weaker inhibition of I_Kr (40 – 50% inhibition at 50 μM). IC₅₀ values for an additional 7 metabolites tested were all >50 μM ²⁶. Importantly in contrast, all 11 metabolites potently inhibited I_NaL by 12 – 57% at 10 μM ²⁶.

In light of this pharmacokinetic data, we used the model to make a prediction about the role of weaker ranolazine metabolite inhibition of I_Kr to explain the clinically observed changes in QTc upon ranolazine administration. Shown in Figure 3A (and Online Figure II) are computed electrograms from transmural tissues spanning the range of measured affinities (50 μM to 12 μM) for the parent compound ranolazine and its metabolites on I_Kr. Notably, an intermediate value that best reflects the physiological situation encompassing a weighted average of I_Kr inhibition from high affinity block by ranolazine and low affinity block by metabolites, produced 8 ms prolongation of computed QTc at 6 μM ranolazine, fully consistent with clinical data ²⁶. In Figure 3B, the predicted concentration dependent increase in QTc with increasing doses of ranolazine is shown. Low dose ranolazine (2 μM) increased QTc by 2.5 ms, while high dose (10 μM) increased the QTc by 14 ms. The simulated QTc prolonging effects are approximately linear with a slope of ~3 ms per 1000 ng/mL, consistent with the clinically observed change of 2.4 ms per 1000 ng/mL ²⁶.

Figure 3. Consideration of ranolazine metabolites for I_Kr inhibition predicts clinical QTc prolongation.

Shown in (A) are computed ECGs of a 165-cell cardiac fiber with 6 μM ranolazine (therapeutic concentration), and varying values for IC₅₀ of I_Kr inhibition (see Supplementary material). I_Kr inhibition at IC₅₀ = 50 μM (green) produces a 3 ms QTc prolongation; I_Kr inhibition at IC₅₀ = 12 μM (blue) prolongs QTc by 40 ms; I_Kr inhibition at IC₅₀ = 35 μM (red) prolongs QTc by 8 ms. (B) Concentration-dependent ΔQTc is approximately linear over the therapeutic range of ranolazine. See text for details.

Potential for ranolazine to normalize ΔKPQ arrhythmia triggers

In order to explore the potential for ranolazine to improve abnormal cellular electrical phenotypes arising from the ΔKPQ mutation, we incorporated the channel model with and without drug in the O'Hara and Rudy ³⁶ (Figure 4, left), and Grand-Bers ³⁷ (Figure 4, right) human ventricular myocyte models. The ten-Tusscher ^{38, 39} model is shown in Online Figure III. We conducted simulations in the full complement of existing human ventricular action potential models in order to ensure model independence of our predictions. Consistent with experimental data ^{30, 32, 40} and previous computationally based studies, the ΔKPQ mutation led to dramatic APD prolongation that worsened with slowing of pacing frequency. As shown in Figure 4 for each model, following 500 stimuli at bradycardic pacing intervals, the ΔKPQ mutation resulted in persistent late Na⁺ current (Figure 4, row 2) and continued arrhythmogenic early afterdepolarizations (EADs) that arose from an extended phase 2 plateau (Figure 4, row 1), which allowed for reactivation of the L-type Ca²⁺ channel (Figure 4, row 3). For rows 2 and 3, peak currents of both Na⁺ and L-type Ca²⁺ currents are off scale. Within the therapeutically relevant range, both high dose (10 μM – teal lines) and low dose (5 μM – red lines) ranolazine normalized the ΔKPQ action potential morphology, an effect that was model-independent.

Figure 4. Effects of ranolazine on ΔKPQ mutant action potentials.

The effects of the ΔKPQ mutation in two human ventricular myocyte models yield qualitatively similar results. Shown in column 1 is the O'Hara Rudy ³⁶ model at BCL 1000, and in column 2 is the Grandi-Bers ³⁷ model at BCL 4000. For each, row (A) depicts cellular APs, row (B) depicts Na⁺ currents (peak off scale), and row (C) depicts the L-type Ca²⁺ currents (peak off scale). In both models, low (5 μM), and high (10 μM) ranolazine progressively shortens the APD but fails to fully normalize to WT (blue line). Row (D) depicts concentration dependent effects of ranolazine on action potential duration (APD) and upstroke velocity (UV) at BCL 1000. Note that while APD is normalized, UV remains robust with high therapeutic concentrations of drug.

The fourth row (D) of Figure 4 shows a summary of the effects of clinically relevant concentrations of ranolazine on ΔKPQ action potential duration and cellular excitability (upstroke velocity (UV) of the action potential (AP)) for simulated epicardial cells at nominal pacing (BCL 1000). Over the clinically relevant dosing regime (1 – 10 μM), ranolazine effectively normalizes APD without compromising cellular excitability – a potentially confounding occurrence and cellular level marker that was previously shown to be strongly proarrhythmic in coupled tissue ²¹. Because there was minimal UV depression, we further tested supratherapeutic ranolazine (15 and 20 μM) and found similar results.

Efficacy of ranolazine to normalize pause-induced EADs

It has been widely documented that LQT3-linked arrhythmias are typically preceded by sinus pauses and short-long-short sequences ^41-45. The presumed mechanisms have been shown experimentally and predicted computationally, and results from the emergence of early afterdepolarizations on action potentials triggered after a pause. Thus, ideal drug therapy for LQT3 patients must normalize arrhythmia triggers occurring subsequent to long diastolic intervals. We used computational one-dimensional transmural tissue models to test the potential for ranolazine to normalize action potentials following long rest intervals in coupled tissue.

Shown in Figure 5A is a space-time-membrane voltage plot showing the last 3 S1 beats (stars) at BCL 750 (after steady state pacing - 500 beats), followed by an S2 (arrow) stimulus applied following a 1.05 second pause. Underneath each voltage in time plot is a computed electrogram from the tissue. The electrogram in A shows an early downward deflection due to EAD generation that occurs first in endocardial cells. Flattening of the electrogram then occurs and finally a positive t-wave deflection as epicardial cells repolarize prior to endocardial cells. Panels B and C show the effect of pretreatment with moderate (5 μM) and high (10 μM) clinical doses of ranolazine. The model predicts that 5 μM ranolazine improves the cellular phenotype following the pause, but is unable to fully normalize the arrhythmogenic trigger following the pause (B). High-dose ranolazine application (C), on the other hand, completely prevents EAD formation following a long pause.

Figure 5. Effects of a pause on incident EADs generated in the subsequent beat.

(A) Space-time plot of 3 S1 beats at basic cycle length (BCL) 750 (after steady state pacing – 500 beats, asterisks) in the absence of drug. A 1050 ms pause, followed by an S2 stimulus (arrows) elicits an EAD throughout the 165-cell cardiac 1-dimensional tissue. (B) Pretreatment with 5 μM ranolazine is insufficient to abolish the EAD throughout the tissue, but with pretreatment of 10 μM ranolazine (C), the S2 stimulus fails to elicit an EAD and monotonic repolarization is restored. For A – C, x-axis is time, y-axis is cell number, z-axis is voltage. A computed ECG is underneath the space-time plot. (D) The pause necessary to elicit an EAD depends increases with drug concentration after steady state pacing (500 beats) at BCL 750. The O'Hara-Rudy model ³⁶ was used for this simulation.

We next quantified the effect of high and low concentrations of ranolazine to prevent pause-induced arrhythmia triggers over a physiologically relevant pause interval range. Shown in Figure 5D is the increase in pause length threshold for EAD normalization after pretreatment with drug at 5 μM and 10 μM ranolazine following pacing to steady-state at BCL 750. The simulations suggest that high dose ranolazine can normalize EADs when the incident pause is less than 1150 ms. Supratherapeutic levels of ranolazine, which we show in Figure 4 maintains UV in single cells, substantially increases the pause duration safety window before arrhythmogenic EADs are noticed (2150 ms at 20 μM).

Thus far, our model simulations have suggested that within the clinically relevant dosages, ranolazine resolves arrhythmia triggers that result from persistent LQT3-linked Na⁺ current. We next wanted to test whether ranolazine had the potential to normalize arrhythmia triggers stemming from acquired dysfunction such as human heart failure, which has been linked to a pathologic increase in late I_Na, and suggested as a potential therapeutic target ^{46, 47}.

Formulation of a human heart failure model

The range of heart failure phenotypes is complex, and there currently exists no adequate computational model that incorporates the myriad ionic and hormonal dysregulation found in end-stage ischemic heart disease. We thus turned to the literature ^48-50 to find the most up-to-date and reproducible human heart failure data and incorporated the deranged ionic fluxes into the Grandi-Bers human ventricular model ³⁷. We chose the model of Grandi-Bers because it incorporates intricacies of Ca²⁺ handling that are known to play a key role in Ca²⁺-induced arrhythmia triggers. We married the Grandi-Bers model of the action potential to the Soltis-Saucerman model formulation of the β-adrenergic pathway. This includes CaMKII and PKA signaling ⁵¹, important regulatory pathways shown to be upregulated in human heart failure ⁵². Complete details of our human heart failure model formulation can be found in Table VI in Supplementary Material.

In Figure 6, we show cellular simulations generated by the human heart failure model. Figure 6A shows a prolonged APD under the influence of 1 μM isoproterenol, consistent with experiments ^{49, 53}. Other important ionic fluxes include an outward shift in NCX current (panel C), reduced intracellular Ca²⁺ transient (panel D), a delayed recovery of SR Ca²⁺ load (panel E), an increased late I_Na optimized to yield ~1% late current (panel F) ⁵⁴, and an increased intracellular Na⁺ concentration (panel G) ^{48, 50} also due to increased I_Na,L and I_Na,leak, and decreased NKA. Summary data comparing experiment to simulation is shown in panel H.

Figure 6. Cellular AP and ionic currents for human heart failure under the influence of β-adrenergic stimulation.

Shown in (A) are cellular APs in the control (black) and human heart failure (HF - red) models after steady state pacing at BCL1000 (500 beats) and under the influence of 1μM isoproterenol, 8-fold increase in I_Na,Leak, and 10% decrease in NKA (our “nominal” HF condition). The model incorporates the Soltis-Saucerman ⁵¹ model of β-adrenergic stimulation into the human ventricular myocyte model of Grandi et al. ³⁷ as described in the Supplementary material. The APs in (A) are reproduced in both columns for ease of comparison. (B) I_CaLcurrent; (C) Na⁺/Ca²⁺ exchange current (I_NCX); (D) intracellular Ca²⁺ transient; (E) sarcoplasmic reticulum Ca²⁺ load; (F) peak (I_Na) current (note peak off scale); and (G) intracellular Na⁺ concentration. (H) Comparison between experiment and simulation for APD ^{49, 53, 75-77}, and [Na]_i ^{48, 50}. Note, both experiments and simulations were at 1 Hz (BCL 1000 ms).

Efficacy of ranolazine to ameliorate heart failure induced arrhythmia triggers

A hallmark arrhythmia trigger in human heart failure is the occurrence of Ca²⁺-induced delayed afterdepolarizations (DADs). While the complete pathway is not fully elucidated, Ca²⁺ modulation of the Na⁺ channel has been demonstrated ^{55, 56} via CaMKII ⁵⁷, and multiple upstream pathways converge on pathologic late I_Na (I_NaL) (e.g. increased mitochondrial oxidative phosphorylation ⁵⁸, increased ROS ^{58, 59} , increased Na⁺/H⁺ exchange ^60-62, decreased Na⁺/K⁺-ATPase [NKA] ⁶³). Increased I_NaL and elevated intracellular Na⁺ then leads to increased [Ca]_i, via a in NCX and ultimately mechanical and electrical instabilities (e.g. DADs, beat-to-beat variability in APD) leading to further ischemia and ventricular arrhythmias ¹⁸.

In Figure 7 (and expanded analysis in Online Figures V - VII), we tested the effects of ranolazine to inhibit DAD generation under conditions of heart failure in the presence of beta-adrenergic stimulation. After conditions of tachycardic pacing (BCL 500), panel B, left, depicts a nonstimulated DAD beat (red arrowhead) not present in control conditions (panel A and Online Figure V). Moderate dose ranolazine (5 μM) is sufficient to inhibit its occurrence (panel B, right, and Online Figure VI). Online Figure VII depicts high dose ranolazine (10 μM). Panel C depicts a more “severe” derangement (20% decrease in NKA, 6-fold increase in Na⁺ leak current) eliciting four spontaneous beats (red arrowheads panel C, left), which is again ameliorated by 5 μM ranolazine (panel C, right). An expanded analysis of ionic current changes for this severe condition (Fig. 7C) is in Online Figure XI. Of note, no data exists for the affinity of ranolazine to the Na⁺ leak current (I_Na,leak); for these simulations, we assumed that the I_Na,leak had similar affinity to WT I_NaL (6 μM). For completeness, we also tested differing ranolazine affinities to I_Na,leak in Online Figures IX and X. Panel D summarizes the results of simulations spanning physiologically reasonable range of combinations of increased Na⁺ leak and decreased NKA. When Na⁺ leak is increased ten-fold, full repolarization failure occurs over all conditions tested (blue “stop signs” in top row of panel D, left). Panel D, right, shows ranolazine is effective at restoring repolarization and inhibiting DAD generation in the majority of physiologically plausible conditions tested (compare blue “stop signs” and red arrowheads, left, to filled circles, right).

Figure 7. Effects of ranolazine on DAD generation under heart failure (HF) conditions.

Row (A) depicts control; (B) HF condition of 10% decrease in Na⁺/K⁺ ATPase (NKA), 8-fold increase in Na⁺ leak; (C) HF condition of 20% decrease in NKA, 6-fold increase in Na⁺ leak. For A – C, column 1 is drug free, and column 2 is with 5 μM ranolazine. Red arrowheads over APs indicate non-paced beats (DADs) that were observed after stimulus was removed. (D) Summary data for parameter space spanning plausible ranges of decreased NKA and increase Na⁺ leak. Filled black circles (□) indicate absence of DADs, upside down red triangles indicate presence of DADs, blue stop signs (□) indicate repolarization failure.

Finally, in Figure 8 and Online Figure XIII, we probe the model components to reveal the ionic mechanism for ranolazine efficacy for one of the conditions tested in Figure 7, namely an 8-fold increase in Na⁺ leak, 10% decrease in NKA (condition in Figure 7B). As compared to control (Figure 8, column 1), the heart failure condition exhibits an approximate 20% increase in [Na⁺]_i (row B, column 2 vs. column 1), which slows inward NCX (Ca²⁺ extrusion) and enhances outward NCX (Ca²⁺ entry, at the beginning of the AP). This, coupled with AP prolongation due to extensive ionic remodeling in HF allows increased Ca²⁺ entry and maintains adequate SR Ca²⁺ load and Ca²⁺ transient despite reduced SERCA function. On the other hand, [Na⁺]_i-induced Ca²⁺ enhancement in combination with hypersensitive RyRs causes diastolic SR Ca²⁺ release and the occurrence of a spontaneous Ca²⁺ transient (row C, column 2 vs. column 1 – red arrowhead). This Ca²⁺ is extruded by NCX, which generates an inward current (row D column 2 – red arrowhead) that depolarizes the membrane potential leading to I_Na activation and triggered AP (row A, column 2 vs. column 1 – red arrowhead). Note that the more depolarized resting membrane potential in HF (due to decreased I_K1 and increased I_Na,leak) is likely to favor AP triggering. Application of 5 μM ranolazine partly normalizes [Na⁺]_i (row B, column 3 vs. column 1) and inward NCX (row D, column 3), and abolishes the spontaneous Ca²⁺ transient (row C, column 3) and triggered AP (row A, column 3). Notably, ranolazine also hyperpolarizes the resting membrane potential thus elevating the threshold for triggered diastolic events. These simulations are fully consistent with recent experimental data for a hypertrophic cardiomyopathy experimental model with ranolazine ⁶⁴, and suggest even moderate dose ranolazine may be an appropriate antiarrhythmic therapeutic for the prevention of arrhythmia triggers driven by spontaneous SR Ca²⁺ release.

Figure 8. Ionic mechanisms of ranolazine efficacy in the heart failure (HF) model.

Shown in row (A) are cellular APs, (B) intracellular Na⁺ concentration, (C) intracellular Ca²⁺ concentration, and (D) Na⁺/Ca²⁺ exchange current. Column 1 is drug-free control, column 2 depicts 10% decrease NKA, 8-fold increased Na⁺ leak (case B from Figure 7) and column 3 is the same condition with 5 μM ranolazine. The red arrowheads indicate spontaneous DAD-triggered beats.

DISCUSSION

Recently, there has been interest in the antiarrhythmic potential of the novel antianginal agent, ranolazine, the first FDA approved drug that specifically blocks the late component of the Na⁺ current. Like most antiarrhythmics that target cardiac ion channels (e.g. flecainide and amiodarone), ranolazine blocks multiple channels, including the repolarizing hERG current I_Kr with therapeutic concentrations. The result is a mild concentration dependent QTc prolongation seen in patients with chronic stable angina ¹¹. Because of this, ranolazine is “contraindicated” for patients on other QT prolonging drugs, those with preexisting QT prolongation ¹², and those with repolarization abnormalities.

In this study, we sought to use a computationally based approach to determine whether ranolazine's unintended pathological block of promiscuous K⁺ channels would prevail over therapeutic drug effects in two specific patient populations: LQT3-ΔKPQ carriers, and those with acquired arrhythmias arising from heart failure.

With regard to congenital LQT3, although many in vitro studies ^65-68 have suggested ranolazine as an ideal therapeutic, to date, only one clinical study has been carried out on a small number (5 carriers) of the ΔKPQ mutation-afflicted patients ⁶⁹. Moss et al. showed an unequivocal decrease in QTc with ranolazine treatment in these patients (4% decrease in QTc at ~ 5 μM), but due to its small size and limited endpoints, it is unclear if ranolazine would be effective at preventing bradyarrhythmias rather than just impacting surrogate markers of arrhythmia (e.g. the corrected QT interval).

When we tested ranolazine in cellular simulations, we found another potential mechanism of safety, namely that unlike other Na⁺ channel blockers (e.g. flecainide and lidocaine), ranolazine does not cause depression of cellular excitability and tissue conduction velocity ²¹. Another drug that had similar promise was the lidocaine oral analog mexiletine. While mexiletine has proven useful in a small clinical trial of LQT3 patients ⁷⁰, it carries proarrhythmic side effects like many other antiarrhythmic drugs. Sustained ventricular tachycardia has been reported ^{71, 72}, as well as exacerbation of arrhythmia in 10 – 15% of patients ⁷³. More importantly for the LQT3 patient population, where the characteristic phenotype is a bradyarrhythmia, mexiletine is associated with sinus node depression, resulting in sinus bradycardia and prolonged sinus node recovery time ^{71, 73}, potentially exacerbating the arrhythmia phenotype.

Clinical studies have shown that administration of ranolazine in the clinic (2 – 6 μM) causes proportional increases in QTc of 2 – 6 ms ^{26, 35}, presumably arising from hERG block. We would have therefore expected to see an approximate 6 ms increase in the QTc on our computed electrograms when model tissues were pretreated with 6 μM ranolazine with incorporation of the rapid blockade of hERG ³⁴. Instead, we observed a dramatic QTc prolongation (40 ms), a prediction that was not consistent with the clinical data (Figure 3A).

However, when we additionally considered the effects of active metabolites of ranolazine, the simulation confirmed clinical findings. Like most drugs, ranolazine undergoes extensive metabolism, primarily via the CYP3A system, with less than 5% of the parent compound excreted in the urine unchanged ²⁰. All 11 active metabolites potently inhibit I_NaL by 12 – 57% at 10 μM, similar to the parent compound ²⁶. In contrast, the four predominant metabolites comprising 30 – 40% of the parent compound produce substantially weaker inhibition of I_Kr (40 – 50% inhibition at 50 μM). Inhibitory concentration (IC₅₀) values for an additional 7 metabolites tested were all >50 μM ²⁶. Our model simulations suggested that a weighted average of the affinities of parent compound and active metabolites led to an apparent affinity for ranolazine and metabolites for I_Kr in the range of 35 μM, causing the clinically observed moderate changes in QTc. Thus, there exists a large margin of safety for drug administration of ranolazine that reflects the difference in ranolazine affinity, and thus targeting, for I_Kr and I_NaL.

We validated our model predictions against surrogate markers of arrhythmia risk (e.g. normalization of the QT interval), and then sought to determine if ranolazine could prevent pause-induced EADs, a clinically significant precedent event to torsades de pointes ^{45, 74}. We found that at a pacing interval of 750 ms, a minimal pause (~100 ms) that extended the diastolic interval beyond normal (S2 = 850 ms), induced action potential prolongation and EAD triggers in a transmural tissue model. Pretreatment of the cardiac fiber with high therapeutic ranolazine (10 μM) delayed the onset of EAD generation by 35% (S2 = 1150 ms, compared to 850 ms in drug free conditions).

In a comprehensive review of the incidence of pause-dependent torsadogenic arrhythmias in congenital LQTS, Viskin et al. ⁴⁵ found an average precedent pause was ~1000 + 300 ms, similar to the threshold range we predicted with ranolazine treatment. We then tested supratherapeutic ranolazine (15 – 20 μM) because of the promising results of robust UV in single cell, and we found a dramatic increase in the safety window of a pause necessary to elicit an EAD (2150 ms). This suggests that high dose ranolazine may reduce the need for cardiac pacing, which itself perpetuates the short-long sequence of torsades de pointes ⁴⁵, but clinical studies of ranolazine with Holter ECG recording will be needed to confirm this clinically relevant prediction.

By targeting pathologic late Na⁺ current, ranolazine shows therapeutic promise for treatment of I_NaL induced arrhythmias – both congential and acquired. Mechanistically, ranolazine does this by 1) by limiting [Na⁺]_i and restoring normal NCX forward mode that limits Ca²⁺ entry via NCX and speeds up Ca²⁺ extrusion, 2) by shortening APD, thus further limiting Ca²⁺ entry, and 3) by hyperpolarizing the membrane potential which elevates the threshold of triggered activity (presumably by ranolazine's effect on I_Na,Leak, a hypothesis borne out by recent experiments ⁶⁴ and suggested by the model). We simulated a physiologically realistic transmural ventricular cardiac tissue based on data obtained from transmural wedge preparations from both normal and failing human myocardium ⁴⁹. The simulations recapitulated a modest decrease in QTc with therapeutic ranolazine (5 μM), and showed that higher dose ranolazine (10 μM) can decrease the QTc interval at static pacing (~12% decrease in QTc).

For LQT-ΔKPQ carriers, ranolazine can also ameliorate the effects of pause-induced early afterdepolarizations, a hallmark clinical precedent to torsades de pointes. In the heart failure setting, we found that even moderate dose ranolazine (5 μM) was predicted to nearly normalize many of the derangements in intracellular ionic concentrations and aberrant currents, which led to an effective abolishment of arrhythmia triggers. Of note, this therapeutic effect may be specific to LQT arising from elevated late I_Na. In the absence of substantial late I_Na, the model predicts that ranolazine will prolong APD and consequently QT interval (Online Figure XII).

Because there exists heterogeneity of heart failure phenotype severity, we attempted to survey a wide parameter space including the effects of varying the Na⁺ leak current and NKA expression. We find that phenotypes arising from a large component of I_Na,Leak are more susceptible to ranolazine blockade than those from decreased NKA expression. We also find that even moderate dose ranolazine (5 μM) shows efficacy in suppressing spontaneous depolarizations (Figure 7, right), and high dose ranolazine (10 μM) suppresses all but the two most severe phenotypes (Online Figure VII).

In summary, we have built genotype specific computational models of the LQT3-ΔKPQ mutation and a human heart failure model to specifically test a therapeutic intervention that targets the aberrant molecular mechanism (persistent late Na⁺ current) in two different pathological settings. Our multidimensional framework largely relies on experimental functional data, but is refined and validated by clinical electrophysiological data from numerous clinical trials. The results of our study suggest that the therapeutic potential of ranolazine derives largely from metabolism of the parent compound into active metabolites, which show significant selectivity between repolarizing current blockade (I_Kr) and pathologic current blockade (I_NaL). Computational modeling of the effects of metabolism is thus a vitally important for accurate and physiologically realistic electrophysiological models of drug blockade. Our studies extend the results of the clinical literature to show that ranolazine further ameliorates the effects of specific torsadogenic activation sequences common to LQT carriers, as well as selectively targeting upstream pathways which lead to mechanical and electrical instability within the ischemic heart failure setting. Both results suggest potential avenues for further clinical testing.

This study represents one step toward construction of an in silico high throughput drug testing system based on both specific genetic defects as well as a continuum of acquired syndromes. Computational models of the kind we present in this study can be used to test vast parameter spaces that include variation in severity of disease. This may allow for rapid preclinical identification of potentially proarrhythmic or antiarrhythmic agents with high fidelity – unencumbered by problems inherent to large-scale clinical trials, which are heterogeneous in patient population and disease severity. Our approach constitutes a tractable methodology to screen for which agents merit further testing with experiments and tailored clinical trials.

Novelty and Significance.

What Is Known?

• Drug therapy for long-term management of cardiac arrhythmia has had limited success, in part, because it is difficult to predict how drug therapy will alter the emergent electrical behavior of the heart.

• Ranolazine is contraindicated for patients on drugs that prolong the QT interval or those with preexisting QT prolongation because its off-target effects could worsen these conditions.

• Patients with QT prolongation due to inherited defects or disease might benefit from drugs like ranolazine that target the pathological late Na current.

What New Information Does This Article Contribute?

• We developed a computational framework to predict the effects of ranolazine in two distinct disease states marked by late Na current, inherited LQT3 and heart failure.

• The computational modeling framework is utilized to improve understanding of antiarrhythmic drug actions across multiple spatial scales of the cardiac system, from molecule, to channel, to cell, and to tissue.

• Our simulated data suggest that ranolazine may be effective in preventing multiple types of arrhythmia triggers that arise from late I_Na.

In this study we adopted an interdisciplinary approach combining laboratory experiments, computational biology, high performance computing, and clinical observation. Collectively, this led to the development of a computational approach to predict the effects of a drug in specific disease states that promote cardiac arrhythmia. We used this approach to predict if ranolazine would be useful in two distinct clinical syndromes associated with an increase in pathological Na. This framework forms a base that can be readily expanded for virtual drug screening of other agents.

Nonstandard Abbreviations and Acronyms

LQT3

Long QT syndrome type-3

EADs

early afterdepolarizations

DADs

delayed afterdepolarizations

UV

upstroke velocity

AP

action potential

APD

action potential duration

hERG

human ether-a-go-go related gene

CAST

cardiac arrhythmia suppression trial

SWORD

survival with oral d-sotalol

CYP3A

cytochrome P450, family 3, subtype A

BCL

rapidly activating component of the delayed rectifier potassium current (I_Kr), basic cycle length