A Molecularly Detailed NaV1.5 Model Reveals a New Class I Antiarrhythmic Target

Visual Abstract

These results provide a roadmap for the design of novel molecular-based therapy to treat myriad arrhythmia syndromes, including ventricular tachycardia, heart failure arrhythmias, and inherited arrhythmia syndromes.
In summary, computational modeling approaches to drug discovery represent a novel tool to design and test precision-targeted therapeutic agents. By exploiting nontraditional ion channel drug targets, an entirely new dimension can be added to the wide parameter space of traditional antiarrhythmic drugs to develop more precision-targeted and potent Class I therapeutic agents.

SUMMARY
Antiarrhythmic treatment strategies remain suboptimal due to our inability to predict how drug interactions with ion channels will affect the ability of the tissues to initiate and sustain an arrhythmia. We built a multiscale molecular model of the Na þ channel domain III (domain III voltage-sensing domain) to highlight the molecular underpinnings responsible for mexiletine drug efficacy. This model predicts that a hyperpolarizing shift in the domain III voltage-sensing domain is critical for drug efficacy and may be leveraged to design more potent Class I molecules. The model was therefore used to design, in silico, a theoretical mexiletine booster that can dramatically rescue a mutant resistant to the potent antiarrhythmic effects of mexiletine. Our framework provides a strategy for in silico design of precision-targeted therapeutic agents that simultaneously assesses antiarrhythmic markers of success and failure at multiple spatial and time scales. This approach provides a roadmap for the design of novel molecular-based therapy to treat myriad arrhythmia syndromes, including ventricular tachycardia, heart failure arrhythmias, and inherited arrhythmia syndromes.  This outcome is due, in part, to the complex kinetics of the drug channel interaction that includes strong bidirectional feedback between how drugs alter the action potential waveform affecting voltagedependent potency, as well as electrotonic coupling in tissue, which we (3) and others (4) have shown can lead to an even more complex response to drugs that may not be appreciated in single-cell studies. For both acquired and inherited arrhythmia syndromes, this scenario leads to a dangerous trial-and-error approach to choosing appropriate pharmacotherapy for patients. Furthermore, despite an enormous amount of research into antiarrhythmic drug therapy that has produced molecules with a wide range of pharmacokinetic properties, they all block a single target: the channel pore. Thus, targets of ion channels other than the channel pore may add an entirely new dimension to antiarrhythmic drug therapy.
Structurally, the alpha-subunit of Na þ channels is formed by a monomer with 4 homologous domains (DI to DIV), each with 6 transmembrane subunits (S1 to S6). The S1-S4 of each domain forms the voltagesensing domain (VSD), and S5-S6 forms the pore  have shown that VSD movement is modulated by binding of local anesthetics, and that when lidocaine binds to the Na þ channel, it stabilizes the domain III VSD (DIII-VSD) in an activated conformation (7). Our recent experimental results (8) show that DIII-VSD dynamics significantly regulate mexiletine blockade of Na V 1.5, and the differential response of long QT syndrome type 3 (LQT3) carriers to mexiletine is due, in large part, to mutation-specific VSD dynamics.
To date, the myriad parameters within Na þ channel kinetic models have been shown to have a significant impact on the ability of the heart to initiate and sustain an arrhythmia, and exploiting these parameters has been useful for understanding Na þ channel pharmacology (9). However, it has not yet been possible to design targeted interventions that alter these parameters because they are not specifically connected to the channel structure.
The current study used our experimental results (8) to develop a computational model that tracks molecular DIII-VSD movement, Na þ channel electrophysiology, and the response of both to mexiletine drug blockade for 2 LQT3 mutations: R1626P, shown to be mexiletine sensitive, and M1652R, shown to be mexiletine resistant. The fidelity of the model for these 2 mutants provides confidence that the relation between the DIII-VSD and channel gating is well represented by the model and allows us to predict novel therapeutic approaches based on this relation. We focused on the R1626P (RP) and the M1652R (MR) mutations because of their marked differential responses to mexiletine. Both mutations lie within the DIV S4 segment and produce an increased late Na þ current w0.7% to 1.0% of the peak current. Aside from a w15 mV depolarizing shift in steady-state availability (SSA) for the MR mutation, and a w8 mV hyperpolarization of the RP mutation, the electrophysiology of both mutations is similar (15). At resting membrane potential, however, the DIII-VSD of RP is nearly w90% activated ("up" position), whereas the MR DIII-VSD is only w50% activated (8).
DRUG-FREE MODEL DEVELOPMENT. We began by developing drug-free models of both mutations as well as wild type, which were easily fit with a well- peak current) for the WT, M1652R, and R1626P constructs, respectively (15). We then focused on expanding the kinetic model to account for DIII-VSD movement, shown to be stabilized in an active conformation by mexiletine (8). Using voltage-clamp fluorometry methods previously described (16), the kinetics of DIII-VSD were simulated in response to channel activation. Briefly, by attaching a fluorescent tag to the DIII-VSD and expressing Na V 1.5 in a Xenopus oocyte cell expression system, we could simultaneously record current kinetics and DIII-VSD movement (16,17).
As can be seen in Figure 1E Figure 1B.  One benefit of using a computational approach is that data from different expression systems can be reconciled by altering the appropriate parameters.
For example, in simulating MR SSA, we chose to incorporate a 15 mV depolarizing shift of MR compared with WT to more closely simulate the results obtained by us and others (15) in the HEK expression system. Thus, in Figure 1C Figure 1E is labeled as "DIII-VSD activation" (which is equivalent to the DIII-VSD position in the membrane). This is congruent with Figure 1F labeling: at -160 mV, all constructs fluoresce maximally, shown as the upward deflection of the fluorescence curve. Upon depolarization, the fluorescence is quenched to 0%, as DIII-VSD is in an "up" position.

MEXILETINE DRUG-BOUND MODEL DEVELOPMENT.
We next expanded the model to account for mexiletine drug binding. Our previous results suggested that the voltage dependence of DIII-VSD activation strongly correlates with tonic block by mexiletine ( Figure 4 of Zhu et al. [8]). Briefly, tonic block, a measure of first-pulse block, is assessed at holding potentials before much closed-state inactivation occurs; thus, the apparent differences in tonic block seen for 15 different mutants were primarily accounted for by the difference in the DIII-VSD. For the RP mutation, the V 1/2 of DIII-VSD activation is -143 mV and has a 10-fold lower half-maximal inhibitory concentration (IC 50 ) for tonic block (69.5 mM), compared with MR, which has a V 1/2 of -100 mV and a tonic block IC 50 of 624 mM. Our results, as well as others (15), suggest that the intrinsic affinity of mexiletine to the local anesthetic receptor for the mutants is likely the same, but the measured tonic block differences represent both contamination by the inactivated state and differences in DIII-VSD guarding the receptor.   (14), with substitution of our Na þ Given mexiletine's safety and widespread use in the clinic, we hypothesized that combination therapy with mexiletine and a "booster" drug might synergize for more potent antiarrhythmic effects. Because mexiletine sensitivity between RP and MR seems to be driven by the relative position of the DIII-VSD, we hypothesized that by holding the DIII-VSD in an "up" and activated position, we could enhance the efficacy of mexiletine for patients found to be mexiletine resistant (e.g., MR mutants). We thus turned to the computational model to design (i.e., in silico) a mexiletine booster and then tested its efficacy in combination with mexiletine.
This concept was first tested experimentally ( Figure 4). We used MTSEA-biotin (biotin) to alter the conformation of the DIII-VSD. It was previously shown that extracellular application of biotin can modulate the cysteine residue at the 1,306 location (7,19). It stabilizes the DIII-VSD in an activated position in R1306C channels. We engineered the R1306C mutation as a biotin target, as well as an R1306C M1652R double mutation into Na V 1.5, and expressed the mutant channels in the HEK cell expression system. As shown in these previous studies, application of biotin to R1306C channel decreases peak Na þ current, which stabilized w20 min after biotin perfusion.
The decrease in Na þ current amplitude suggests that MTSEA-biotin binds to the cysteine residue and modulates the DIII-VSD conformation. We also observed a reduction in peak Na þ current in the double-mutant R1306C M1652R 20 min after biotin application ( Figure 4B). Biotin caused a small depolarizing shift in the activation (conductance-voltage [GV]) curve, but no effect on the SSA curve, for the R1306C channel ( Figure 4C).  When application of boosted mexiletine was simulated in single cells, a dramatic response in the M1652R mutation was observed. Figure 5A shows the movement of the DIII-VSD during the action potential for the 3 constructs. For WT, the DIII-VSD transits between 92% and 100% "up" during an action potential. For RP, the DIII-VSD remains nearly 100% up throughout the entire duration of the action potential, underlying the sensitivity of RP to mexiletine. In contrast, the DIII-VSD of MR, even with mexiletine 10 mM, transits between 75% and 100% activated, with most of the cardiac cycle at 75% (during diastole). Application of boosted mexiletine holds "up" the DIII-VSD to w90% (similar to WT) and allows increased mexiletine access. In Figure 5B, the action potentials Na þ current and APD 90 are plotted in response to boosted mexiletine in a fashion similar to that shown in Figure 3. Combination therapy dramatically shortened APD and late Na þ current and rescued the phenotype to resemble WT.
This boost is further shown with the monotonic, stable APDs as a function of cycle in Figure 5B. In sum, a mexiletine booster that was designed to hold up DIII-VSD in drug-free conditions enhances mexiletine efficacy and normalizes cellular markers of arrhythmia.
To further characterize the efficacy of the mexiletine booster, an in silico dose-finding experiment was conducted in which we quantified the "equivalent dose" of mexiletine that would be needed to achieve comparable results with the combined booster þ mexiletine 10 mM; the experiment used simulations in increments of 5 mM mexiletine, starting from the maximum therapeutic dose (10 mM). As can be seen in Figure 6, achieving similar APD shortening at BCL2000 with combination therapy (com-  To create our model, 2 unique LQT3 mutations were used as "guideposts," given their varied clinical responses to mexiletine (8,15,22). These data allowed us to integrate our experimental findings into a model of the Na þ channel that is able to explicitly represent the molecular movements shown to be critical for mexiletine drug efficacy. To our knowledge, this  In the extended time course for the M1652R mutation (Supplemental Figure 4), chaotic behavior was noted, with salvos of sustained membrane depolarization preceded by increasing APD, strikingly similar to the 2:1 atrioventricular block as well as sudden cardiac death/ventricular fibrillation at the whole heart level seen clinically (15). Conversely, the R1626P mutation exhibited a 28% reduction in APD 90 from baseline with mexiletine 10 mM, similar to clinical results.
We chose to simulate slow pacing frequencies, given that LQT3 arrhythmia syndromes are bradycardia dependent, happening mostly during sleep and periods of inactivity (30). Often, they are much less pronounced during normal (and fast) heart rates, given the rate-dependent QT shortening. This rate dependence can best be seen in Supplemental Based on our modeling, we found that R1626P significantly prolonged the QT interval even more than M1652R in single-cell simulations. Furthermore, Figure 3F shows that mid-dose mexiletine (5 mM) was not particularly effective for the R1626P mutation, which displays beat-to-beat variability in APD. However, we found no simulations that suggested further liver, thyroid, and cornea deposition). Interestingly, mexiletine has also been used to treat neuropathic pain (31). Although widely used, tolerance to mexiletine is inversely proportional to the dose, with gastrointestinal distress and nausea often limiting high therapeutic concentrations.
As a necessary first step and proof-of-concept, we focused on 2 LQT3 mutations that represent the "extremes" of DIII-VSD sensitivity to mexiletine. By framing the modeling study in terms of these 2 mutations, however, we were able to delineate the putative molecular mechanism that underlies Na V 1.5 sensitivity to mexiletine, regardless of the presence or absence of a mutation. In our previous study (8), we further found a strong correlation between DIII-VSD activation and mexiletine sensitivity ( Figure 4 and Online Table II    STUDY LIMITATIONS. Although the Na þ channel model is highly complex and is coupled with a highly parameterized human ventricular myocyte computational model, it represents a simplification of the true underlying pathophysiology. The molecular movements include only the contribution from the DIII-VSD, as our experimental results (8) suggest that this domain is the most important for determining Class I antiarrhythmic drug efficacy.
Future studies may incorporate the contribution of the DIV-VSD as a next step. As noted, all current Class I drugs are designed to have different binding rates and affinities to the channel pore; a "booster" allows for the modulation of the state of the channel. However, it is possible that activation of DIII-VSD may be proarrhythmic. Based on our simulations, our in silico booster molecule seems to be extremely effective as an adjunct therapeutic strategy with mexiletine and could be a clinically useful therapy. However, such a molecule does not currently exist, and in fact, despite the existence of DIV-VSD-binding molecules, we are aware of no DIII-VSD molecule currently in clinical development.
We further note that we used a saturating concentration of biotin, and our simulations of "booster" necessarily pushed the DIII-VSD into a nearly complete "upward" position to fit the desired higher affinity tonic and use-dependent block. Theoretically, one could pursue further dose-finding strategies to "fine-tune" the desired therapeutic effects of the "booster" molecule. Although a booster does not exist currently, this study provides a rationale to pursue such a molecule.