Increased atrial effectiveness of flecainide conferred by altered biophysical properties of sodium channels

Increased atrial effectiveness of flecainide conferred by altered biophysical properties of sodium channels Journal Item How to cite: O’ Brien, Sian; Holmes, Andrew P.; Johnson, Daniel M.; Kabir, S. Nashitha; O’ Shea, Christopher; O’ Reilly, Molly; Avezzu, Adelisa; Reyat, Jasmeet S.; Hall, Amelia W.; Apicella, Clara; Ellinor, Patrick T.; Niederer, Steven; Tucker, Nathan R.; Fabritz, Larissa; Kirchhof, Paulus and Pavlovic, Davor (2022). Increased atrial effectiveness of flecainide conferred by altered biophysical properties of sodium channels. Journal of Molecular and Cellular Cardiology(166) pp. 23–35.


Abstract
Atrial fibrillation (AF) affects over 1% of the population and is a leading cause of stroke and heart failure in the elderly. A feared side effect of sodium channel blocker therapy, ventricular pro-arrhythmia, appears to be relatively rare in patients with AF. The biophysical reasons for this relative safety of sodium blockers are not known.
Our data demonstrates intrinsic differences between atrial and ventricular cardiac voltage-gated sodium currents (INa), leading to reduced maximum upstroke velocity of action potential and slower conduction, in left atria compared to ventricle. Reduced atrial INa is only detected at physiological membrane potentials and is driven by alterations in sodium channel biophysical properties and not by NaV1.5 protein expression. Flecainide displayed greater inhibition of atrial INa, greater reduction of maximum upstroke velocity of action potential, and slowed conduction in atrial cells and tissue.
Our work highlights differences in biophysical properties of sodium channels in left atria and ventricles and their response to flecainide. These differences can explain the relative safety of sodium channel blocker therapy in patients with atrial fibrillation.

Introduction
Atrial fibrillation (AF) is a common cardiac arrhythmia and a major driver of stroke, heart failure, and cardiovascular death. Incidence and prevalence of AF is expected to increase further, with projected estimates of 6 million AF patients in US and 17.9 million in Europe, over the next 30-40 years 1,2 .
Sodium channel blockers such as flecainide are commonly used to restore rhythm in patients with AF.
These were initially developed to suppress ventricular arrhythmias. Following the increased mortality found in the CAST trial comparing flecainide and encainide to placebo 3 , flecainide and similar agents are now primarily used for rhythm control therapy of patients with atrial fibrillation (AF) with normal ventricular function and without ischemic heart disease [4][5][6][7][8] . Recent controlled trials (Flec-SL and EAST-AFNET 4) demonstrate a low rate of ventricular pro-arrhythmia when flecainide is used after cardioversion and during long-term rhythm control therapy as part of an early rhythm control strategy, including in patients with heart failure with preserved ejection fraction 4,9,10 . This contrasts with the ventricular pro-arrhythmia observed when flecainide was used to suppress ventricular arrhythmias in patients with myocardial ischemia. 3 The reasons for this discrepancy are unclear and with the predicted increase in use of early rhythm therapy, as a consequence of the EAST-AFNET 4 trial, better understanding of the effects of flecainide on atrial and ventricular electrical function is waranted.
Although flecainide has been shown to inhibit the rapid delayed rectifier current at clinically relevant doses, which may account for some proarrhythmic outcomes, flecainide primarily inhibits myocardial Na + channels, of which NaV1.5 is the primary pore-forming alpha subunit found in mammalian heart 11-15 . NaV1.5 is a large transmembrane protein primarily mediating the cardiac Na + current (INa) in cardiac cells.
Localization and plasma membrane expression of the NaV1.5 is reliant on specific beta subunits 16 and NaVβ2 and NaVβ4 subunits can modulate the kinetic and voltage dependence properties of the INa 17 .
Flecainide is known to block NaV1.5 in its open, activated state, where it can bind to the pore and prevent Na + from traversing the membrane 18,19 . This leads to enhanced refractoriness especially at rapid atrial rates, as seen in AF 20 . However, inhibition of Na + channels also slows conduction velocity, which can both terminate multiple reentry and increase the predisposition to functional conduction block and macroreentrant arrhythmias 21,22 . Functional conduction block and macro-reentry are believed to contribute to the ventricular pro-arrhythmia found in the CAST trial studying flecainide, encanide, and moricizine in survivors of a myocardial infarction with reduced left ventricular function.
Several crucial questions remain unanswered, regarding flecainide and atrial and ventricular electrical function. It is not known whether flecainide exhibits differential sodium channel-blocking effects in atria and ventricles. Characterization of the atrial and ventricular conduction differences are yet to be studied in the same hearts, or indeed the extent of conduction slowing caused by flecainide. Chamber differences in peak INa when measured over a range of different holding potentials remains to be determined, as well as the sensitivity to flecainide. Whilst increased peak INa is reported in cells isolated from left atrium (LA) compared to those isolated from left ventricle (LV) [23][24][25][26][27] , it is not clear if this is conserved at more physiological resting membrane potentials (RMPs), which are likely to differentially impact on sodium channel availability, gating and flecainide efficacy. Furthermore, it is not known if chamber differences in expression of SCN5a/Nav1.5, SCN2B/NaVβ2 and SCN4B/NaVβ4 exist and if they are consistent in healthy murine and human heart tissue.
The aims of the present study were to robustly interrogate differences in biophysical properties of ventricular and atrial INa (inclusive of measurements at physiological membrane potentials), their effects on conduction, and examine chamber specific responses to flecainide. Furthermore, we investigated expression profiles of sodium channel subunits in atrial and ventricular chambers in non-failing murine and human tissue.

Biophysical properties of INa differ between the left ventricle and left atrium
We first set out to quantitatively compare INa in adult murine cardiomyocytes isolated from left atria and left ventricles. To measure peak INa, we initially employed a standard protocol, where INa was elicited from a holding potential of -100mV. Example raw INa traces are demonstrated in Figure 1A. When normalized to capacitance (Supplemental Figure 1) and measured from a holding potential of -100mV no differences in INa density were seen between the left atria and left ventricles at any test potential (Figure 1B,C).
However, atrial cardiomyocytes exhibited an approximately 4mV more negative V50,act, compared to ventricular cardiomyocytes ( Figure 1D). We also examined whether differences in the INa inactivation kinetics exist in datasets in Figure 1. Time to peak and time to 95% decay were increased in left ventricular cardiomyocytes compared to atrial (time to peak -LA 0.86±0.03 ms, LV 1.16±0.06 ms; ****p < 0.0001 unpaired t test, n=30 cells for LA and n=29 cells for LV), (time to 95% decay -LA 4.25±0.23 ms, LV 6.47±0.24 ms; ****p < 0.0001 unpaired t test, n=30 cells for LA and n=29 cells for LV). These differences however are likely driven by larger ventricular currents.
Under normal physiological conditions, cardiomyocytes are not at a resting membrane voltage of -100mV or -120mV (conditions at which INa is usually measured), however the resting membrane potential lies in the range of -90 to -65mV 28,29 . Thus, the magnitude of INa over this physiological range will depend not only on the population size of functional NaV1.5 channels but also on the number of channels that are available to conduct ions at a given resting membrane voltage. Therefore, we went on to determine how resting membrane potential could alter the peak INa current (Figure 2A). The holding potential had a marked and differential effect on the left atrial and left ventricular INa. At a holding potential of -120mV, atrial peak INa was larger (-35.7±1.3 pA/pF, n=40/13 cells/mice) when compared to ventricular cardiomyocytes (-30.9±1.7, p=0.027, Figure 2B). However, this was reversed when sodium channels were activated from a more physiologically relevant holding potential of -75mV; peak current density was lower in left atrial cardiomyocytes (6.7±0.8 pA/pF, n=40/13 cells/mice) when compared to the ventricular cardiomyocytes (9.7±1.1pA/pF, n=28/13 cells/mice, p=0.026), Figure 2C. Thus, although atrial cardiomyocytes have a greater maximal INa density when activated from -120 mV, the physiologically relevant INa size is actually significantly lower in atrial than ventricular cardiomyocytes.
Voltage dependent steady state inactivation of INa is critical to the function of the cardiomyocyte as it determines the number of available NaV1.5 channels, and thus can have major effects on peak depolarizing current, conduction and interactions with pharmacological agents. Left atrial INa inactivated at significantly more negative voltages than in left ventricular cardiomyocytes, with significant differences observed over a wide range of voltages (Supplemental Figure 2A). The mean voltage of half-maximal steady-state inactivation (V50, inact) of the atrial cardiomyocytes was 6.5±1.5 mV more negative than that of ventricular cardiomyocytes (Supplemental Figure 2B). This has important physiological consequences; at -120 mV there was no difference in steady-state inactivation %, as nearly all channels are available. However, at less negative voltages, inactivation % is greater in the LA when compared to the LV. For example, at -90 mV, 47.7±3.3 % of channels were no longer available for activation in the LA, compared to only 24.8±2.6 % in the LV (p<0.0001). At a more physiological potential of -75 mV, the difference in inactivated channels between the left atria (82±2.2%) and left ventricles (71.8±2.8%, P<0.0049, Supplemental Figure 2C) is significant. Voltage inactivation slope constants (k) were consistent between LA (7±0.3, N=12) and LV (6.9±0.1, N=19, P>0.05).
Recovery from inactivation was examined using a double pulse, P1 and P2 protocol that delivered two identical depolarizing pulses to -30 mV of 25 ms duration. Atrial INa was also slower to recover from inactivation when comparing the ventricle P50 of recovery in LA (22.6±1.7ms) and LV (15.5 ±1.5 ms) (Supplemental Figure 3). Recovery differences shown here are therefore likely to contribute to the reduced INa at physiological holding potentials, action potential upstroke and conduction in the atria observed in this study.
In order to examine whether these differing INa characteristics can lead to alterations in action potential morphology, specifically action potential (AP) upstroke, we went on to include these biophysical data in a mathematical model of the ventricular cardiomyocyte 30 . As shown in Figure 3, the modelling data fitted the experimental data well in terms of current activation and inactivation (Figure 3A and 3B).
Incorporating the kinetics of the sodium channel from the left atrial and left ventricular cardiomyocytes resulted in a reduced current density in the 'atrial' cell as well as slower current decay during an AP when compared to the ventricular cell ( Figure 3C). There was also a slower peak maximum upstroke velocity (dV/dt) of the modelled 'atrial' cell compared to the ventricular cell (121.59 mV/ms for the LA when compared to 130.32 mV/ms for LV) and a smaller AP amplitude, when measured at 1Hz pacing cycle length and an RMP of -80mV, Figure 3D. We followed these modelling experiments with current-clamp experiments on single isolated cardiac myocytes (Figure 4) to determine the applicability of the modelling data. As shown in Figure 4B, when paced at 1Hz, left atrial myocytes had an RMP of 70.8±1.4 mV, whereas left ventricular myocytes were generally less depolarized with an RMP of -74.4±1.0 mV (p=0.0002, n=23-40/5 cells/mice). Furthermore, maximal upstroke velocity of the AP was significantly higher in single ventricular cells when compared to atrial cells (at 1Hz, LA 238.6±27.7 mV/ms; LV 304.2±27.9 mV/ms *p=0.0018, n=23-40/5 cells/mice) ( Figure 4C).

Atria display slower conduction velocity and are more sensitive to flecainide
In order to confirm the earlier cellular and modelling findings, we set out to investigate whether conduction of electrical signals is different across matched murine left ventricle and left atria, using optical mapping. Figure 6 illustrates that LA conduction velocities are significantly reduced compared to the LV at 10Hz pacing cycle length (24.4±2.7 vs 36.1±6.2 cm/s, p=0.03. N=5). As expected, action potential duration was prolonged in the LV when compared to the LA (APD50 27.1±2.4 ms in the LV compared to 12.5±0.8 ms in the LA, p=0.0011, N=5).
In a separate set of experiments, we investigated if flecainide exerted differential effects on LA and LV CVs in the intact heart. In agreement with the cellular data, flecainide significantly decreased conduction in the left atrium (59.6± 7.2% of baseline) ( Figure 6F) whilst conduction was not significantly affected in the left ventricle (73.7± 4.7% of baseline) ( Figure 6G). Furthermore, no differences were seen in action potential duration between the two chambers after exposure to flecainide (Figure 6H,I).

NaV1.5, ß2 and ß4 subunits are differentially expressed in murine left atrium and left ventricle
We then determined whether potential differences in protein expression could explain the differing biophysical and pharmacological properties between the left atrium and the left ventricle. Specifically, we examined the expression of NaV1.5, NaVβ2 and NaVβ4 in murine atrial and ventricular tissue, using Western blotting. NaVβ2 and NaVβ4 were previously demonstrated to affect activation, inactivation and recovery of INa 25 . As can be seen in Figure 7A, NaV1.5 showed a higher relative expression in the left atria when compared to the left ventricle in the mouse heart (LA=3.18 ± 0.40 AU; LV=1.00 ± 0.16AU; n=13, p<0.0001). A number of additional bands of lower molecular weight were also noted on the Western blot, likely due to the polyclonal nature of the NaV1.5 antibody (Supplemental Figure 5)). In contrast to this, both NaVβ2 (LA=0.22 ± 0.03AU; LV=0.41 ± 0.05AU; n=6, p=0.0123) and NaVβ4 (LA=0.09 ± 0.003 AU; LV=0.21 ± 0.01 AU; n=6, p<0.0001) were shown to be expressed at lower levels in the left atria when compared to the left ventricles ( Figure 7B,C).

SCN4B transcripts are decreased, whilst SCN5A are increased in left atrium compared to left ventricle in non-failing human heart tissue
Finally, we sought to determine if these findings in murine tissue are replicated in human tissue. To do so we conducted differential expression analysis on RNAseq from matched human left atrial and left ventricular tissue derived from 79 non-failing hearts. As seen in Figure 7D,E, transcripts for SCN5A, the gene that encodes for NaV1.5, was found at significantly higher levels in human left atria, compared to human left ventricles (1.66-fold change, Padjusted = 7.71*10 -23 ), consistent with the murine expression data ( Figure 7A). Furthermore, transcript levels of SCN4B, which codes for NaVβ4, were significantly lower in human left atria when compared to the left ventricles (1.24-fold change, padjusted = 0.0057). SCN2B (NaVβ2) expression was modestly reduced in the left atria but this was not significant (1.09-fold change, Padjusted = 0.407).

Discussion
The present findings demonstrate that flecainide exhibits some 'atrial-selective' inhibition of INa in the normal adult heart. These effects can explain the relatively low risk of ventricular pro-arrhythmia when flecainide is used in patients with atrial fibrillation. In more detail, the data demonstrate that: 1) left atrial tissue has slower conduction velocities then the left ventricle and is more sensitive to flecainide, when measured in the same heart; 2) Despite the left atrium expressing more SCN5A/NaV1.5 protein then the left ventricle, at physiological membrane potentials, peak INa density is reduced in the atria; 3) The reduction in INa at physiological resting membrane potentials is sufficient to account for the decreased action potential upstroke velocity in the left atrium compared to the left ventricle; 4) Flecainide is more selective for left atrial INa compared to left ventricular cardiomyocytes; 5) Atrial peak INa density and sensitivity to flecainide is dramatically more influenced by resting membrane potential than the ventricular peak INa density. 6) Flecainide reduces the peak AP upstroke velocity more in atrial cells compared to ventricular cells.

Mechanisms promoting atrial susceptibility to arrhythmias
We robustly interrogated differential conduction velocities in the atria and ventricles by multi-vector, single-vector and activation time methodologies. 31 . Crucially, our optical mapping data compares atrial and ventricular recordings in the same heart. Conduction velocities recorded in our study are in the expected range, as demonstrated by others 32 . We demonstrate for the first time that mouse atrial tissue display circa 30% reduced conduction velocity compared to the ventricle, regardless of the methodology (multi vs single vector) utilized (Figure 6; Supplemental Figure 6). These data shed some light on the potential mechanisms contributing to enhanced atrial susceptibility for maintenance of re-entrant arrhythmias. In agreement with our data, van Veen et al. whilst employing electrogram array to record conduction in 3-4 months old mice, showed slightly lower conduction velocities in left atria (circa 30 cm/s) than the ventricles (circa 36 cm/s), though these were not directly compared 33 . Additionally, Thomas et al. detected a small but non-significant increase in conduction in mouse ventricles compared to the atria, using a liner epicardial electrode array 34 .
Remarkably, reduced conduction in atria was observed despite a clear increase in NaV1.5 protein expression in left atrial tissue, compared to left ventricle. Increased murine atrial NaV1.5 expression was consistent with our data showing significantly increased expression of SCN5A in non-failing human left atrium, compared to left ventricle (Figure 7). Consistent with these molecular biology findings, left atrial cardiomyocytes show increased peak INa compared to ventricular, when all sodium channels are activated, from a holding potential of -120 mV (Figure 2A,B). Our patch clamp findings are in line with reports of others that use the holding potentials of -120 mV. Li et al. 24 were the first to describe marked increase in INa density in isolated atrial guinea pig cardiomyocytes when compared to ventricular epicardial cells.
Similar findings have also been reported in rat, rabbit, and canine cardiomyocytes 23,[25][26][27] . These studies have looked at the current density from a holding potential of -120 mV, when all channels are available, which is not the case under physiological conditions. Healthy ventricular cells have a resting membrane potential of between -65 and -90 mV in situ whilst atrial cells are slightly more depolarized, lying between -65 and -80 mV 28,29 . In the present study we show that atrial peak INa density is differentially influenced by resting membrane potential, more so than observed in the ventricular cardiomyocyte ( Figure 2). Thus, at physiological resting membrane potentials, atria actually display a smaller peak INa density when compared to the ventricles. Consistent with the observed reduced peak INa in the left atrium, the action potential amplitude and Vmax of atrial APs have been experimentally described as being between ≅150-300 V/s 35,36 , compared with higher values of 250-450 V/s for human ventricular cells [37][38][39] . Similar differences appear to be conserved in smaller species 28,40,41 . Our modelling and cellular data consistently show a faster upstroke in the ventricle compared to the atria (Figures 3 and 4), in agreement with the cardiac optical mapping studies (Figure 6). With regards to the modelling data, we likely underestimate the impact of the reduced atrial INa on the upstroke, as we utilize a ventricular cell model and do not account for many other differences that exist between the atria and ventricle, such as the altered resting membrane potential. Since the magnitude of depolarizing INa is the key factor that determines conduction, a reduced INa at physiological membrane potentials is the most likely explanation for the slower conduction that we observed in the left atrium compared to the left ventricle ( Figure 6).
So, what determines the lower atrial INa at physiological membrane potentials, despite expression of a greater number of NaV1.5 channels? Our patch clamp data demonstrate that left atrial cardiomyocytes display a greater negative shift in the voltage dependence of sodium channel inactivation (Supplemental Figure 2). This is expected to result in a markedly reduced number of available Nav1.5 channels, and a lower peak INa, when initiated from physiological holding potentials. We also detected a negative shift in atrial V50act, meaning that the voltage at which half of the sodium channels are activated is more negative in the left atrial cardiomyocytes, when compared to the left ventricular cells (Figure 1). This is in line with previous findings in rabbit cardiomyocytes 23 Figure 4A. This may well account for the increased incidence of ectopy and re-entrant arrhythmia in the atria compared to ventricle although more work is required to validate this.

Do NaVβ-subunit expression differences mediate INa differences between the atria and ventricles?
Here we show that atrial expression of NaVβ2 and NaVβ4 is dramatically reduced in mouse hearts. In addition, RNAseq data from non-failing human hearts showed a similar pattern, reduction in transcript levels of SCN2B and SCN4B in the left atrium (Figure 7). Chen et al., showed that kinetic properties of sodium channels co-expressing NaV1.5 with NaVβ2 and NaVβ4 subunits were similar to those of the ventricular sodium channels, with more positive activation potential, more positive inactivation and faster recovery of the sodium channels. Equally, kinetic properties of sodium channels expressing NaV1.5 alone were similar to those of the atrial sodium channels 25 . However, as experiments were performed in HEK293 cells co-expressing NaVβ2 and NaVβ4 subunits together, it is unclear whether the effects on channel kinetics required expression of both NaVβ2 and NaVβ4 or just one of the subunits. Indeed, Malhotra et al. showed that β2 had no detectable effects on channel kinetics using a heterologous expression system, suggesting that the effects of β2 may involve cell adhesion and cytoskeletal communication as opposed to channel gating 42 . Meanwhile, co-expression of the β4 subunit with NaV1.5 resulted in a negative shift in the voltage dependence of inactivation when compared to NaV1.5 alone 43 .
Furthermore, mutations in both of these subunits have been associated with arrhythmogenic consequences 43,44 . In addition to β2 and β4 subunits Nav1.5 may also interact with β1 and β3 subunits, which have also been shown to modulate sodium channel activity 45,46 . Interestingly, recent work has illustrated that the loss of Scn1b (NaVβ1) in murine myocytes differentially affected the potencies of lidocaine and ranolazine, whilst also showing a distinct difference in transcriptional expression between human atria and ventricles 47 In the present study we did not investigate the effects of NaVβ1 and NaVβ3 expression on sodium channel function and localization, however, this should be an area of further study.
In addition to direct effects of β subunits on biophysical properties of INa, β subunits can also affect sodium channel localization. For AP conduction, location of NaV1.5 on the cell membrane is particularly important, as are the number of channels in a particular cluster of channels as recently shown by Hichri et al. 48 Furthermore, work has shown that there are a number of different sub-pools of channels at the lateral membrane 49 . Whether β-subunits are involved in localization of channels to a particular microdomain remains to be investigated. It is well-known that the atrial resting membrane potential is more positivedepolarizedthan the ventricular resting membrane potential and this has also been shown in the current data set 29,40 . Our data now add novel insights into the differences between atrial and ventricular sodium channel properties and expression profiles and suggest that these physiological differences lead to further inherent atrial selectivity of flecainide (Figures 4, 5 and 6). We have previously shown that the extent of flecainide's INa inhibition is highly sensitive to small changes in atrial resting membrane potential, whilst our more recent work has shown that atrial RMP modifies the effectiveness of several clinically used AADs 28,51 . This sensitivity clearly contributes to the observed lower atrial INa at physiological membrane potentials. In addition, other factors affecting the differential effects of flecainide on Nav1.5 channels in the atria and ventricles clearly exist. Whether the alterations in β-subunit chamber specific expression seen in this study contribute to differences in flecainide effectiveness remains unanswered. Further studies determining the supra-molecular clustering of sodium channels including their subunit on cardiomyocyte membranes are warranted. In murine Scn3b −/− hearts, where the β3 subunit is not expressed, flecainide produced reduced arrhythmic incidences combined with prolonged refractory periods and shortened APDs, despite the fact that these cardiomyocytes showed a reduction in INa and a negative shift in NaV1.5 channel inactivation 52 In light of our findings, atrial selectivity of flecainide in healthy myocardium is apparent and thus our data explain the lack of ventricular arrhythmias observed in Flec-SL and EAST-AFNET4 trials. The dosing regimens in these trials is not dissimilar to that of the CAST and CAST II trials so perhaps depolarization of the ventricular resting membrane potential during ischemia and in the border zone of myocardial infarctions enhanced flecainide effectiveness in "dangerous" areas in the left ventricle with scars (which were present in all patients in CAST and CAST II).

Limitations.
While we verified key molecular changes in human tissue, this study was mainly performed in healthy mouse hearts. Validation in large mammals and human cardiomyocytes is warranted. We must also consider that our findings are relevant for hearts without structural heart disease, the main group of patients in whom sodium channel blockers are currently used, this study did not directly investigate the proarrhythmic effects of flecainide which are mainly found in infarcted hearts with structural heart disease or during ischemia. In addition to direct effects of β subunits on biophysical properties of INa, β subunits can also affect sodium channel clustering, a novel modulator of Nav1.5 function 48 It is conceivable that altered clustering contribute to the physiological differences between atrial and ventricular INa found. In addition it has been shown that other alpha-subunits, such as Nav1.8, may contribute to cardiac conduction and mRNA for such proteins have been detected in both atrial and ventricular tissue 53 . How these may contribute to the chamber differences observed here is unclear.
Furthermore, work has shown that there are a number of different sub-pools of channels at the lateral membrane 49 . Whether β-subunits are involved in localization of channels to a particular microdomain remains to be investigated. Finally we also need to consider that it is not only the number of available channels at physiological diastolic potentials that will determine the conduction velocity but it can also be determined by other factors such as the density of gap junctions, sodium and calcium concentrations as well as cell geometry [54][55][56] . These factors go beyond the scope of the present study but should definitely be investigated in the future.

Conclusion and clinical significance
Our data illustrate the striking differences in the electrical properties of left atrial and ventricular myocardium. Of most clinical interest is demonstration that flecainide is more effective at inhibiting atrial sodium channels. Our findings can explain the good safety profile of sodium channel blockers in patients without myocardial scars, e.g. those with atrial fibrillation and with preserved left ventricular function, while also providing a reasonable explanation for the pro-arrhythmia seen in earlier studies in survivors of a myocardial infarction with heart failure with reduced ejection fraction. Determining the molecular drivers of the atrial and ventricular resting membrane potential and identifying the molecular interaction partners of flecainide and similar substances, may help to identify new targets for safe antiarrhythmic drug therapy. This is much needed to implement systematic, early rhythm control therapy in patients with AF.

Animal Model
All experiments were conducted under the Animals (Scientific Procedures) Act 1986 and approved by the home office (PPL numbers 30/2967 and PFDAAF77F) and the institutional review board at the University of Birmingham.

Optical Mapping of Murine Atria and Ventricles
Mouse (CD-1, 29-49 g, Charles River, United Kingdom) hearts were isolated under deep isofluraneinduced inhalation anaesthesia (4% in O2, 1.5 L min -1 ). Optical mapping of mouse whole hearts was performed as previously described 57  Hearts were paced at 10Hz with 1ms pulses from the right atrium epicardial surface using silver bipolar electrodes. Pulse amplitude was 2x the diastolic threshold. Hearts were illuminated during recordings by two dual LEDs (530±25nm) and emitted light collected at >630nm. 10s recordings were acquired at 0.5kHz at 51*51 pixel resolution using a Evolve Delta EMCCD camera (Photometrics, USA).
To ensure effective recording of the signals, sequential recordings were collected. Following recording of the anterior ventricular surface, the hearts were re-orientated. The atria were moved so they were in the focal plane and removed from the ventricles to prevent signal overlap. To enable direct comparison between the left atria and ventricle, a 9x9 pixel (1.4cm2) region was analysed from both chambers ( Figure 1). Optical action potentials (APs) and conduction velocities (CV) were analysed using ElectroMap, as previously described 31 .

Murine Cardiomyocyte Isolation
LA and LV murine cardiomyocytes were isolated from male and female adult mice (12-20 weeks old), bred on the 129/sv background, as previously described 28

Electrophysiology
Dissociated mouse LA and LV cardiomyocytes were transferred to an initially static bath recording chamber and allowed to adhere to laminin-coated coverslips (10mm diameter). Whole cell patch clamp recordings were obtained in voltage clamp mode using borosilicate glass pipettes (tip resistances 2. signals were sampled at 50kHz and low pass filtered at 20kHz. INa was normalised to cell capacitance and expressed as pA/pF. To assess current-voltage relationships, INa was elicited using 100ms step depolarizations at 1Hz over a test potential range of -95mV to +10mV, in 5mV increments, from a holding potential of -100mV. I/V curves were fitted using the modified Boltzmann equation as described previously 58 . Measurements of steady state inactivation of INa, were made by applying pre-pulses ranging from −120 to −40mV in 5mV increments for 500ms prior to the test potential (−30mV for 100ms). As well as being able to calculate V50 inactivation voltages, this protocol also allowed for measurements of INa at different holding voltages. This also allowed for measurements of instantaneous Following baseline recordings, cells were superfused with 1 µM flecainide for 5 minutes before repeating the recordings. Action potential characteristics were measured using modified algorithms from the ElectroMap software as previously described 51 . Following 60 seconds pacing to reach steady state, resting membrane potential (RMP) was defined as the minimum diastolic membrane potential. Maximum upstroke velocity (dV/dt) was measured as the maximum derivative of the action potential recording.

Biophysical Modelling
Full details for the modelling methodology and data can be found in the supplement. In brief, sodium channel activation and inactivation protocols were fitted to a standard Hodgkin and Huxley formulation of the INa channel: GNa is the maximum channel conductance, m is the activation gating variable, h and j are inactivation gating variables, V is the membrane potential, ENa is the sodium Nernst potential. We assume j and h are equivalent as insufficient data is available to fit distinct inactivation gates. The gating variables were fitted to activation and inactivation data for atria and ventricles.
Flecainide inhibition is dependent on the holding potential and is distinct for the atria and ventricle. To approximate flecainide effects, we introduced a linear scalar that represented sodium channel inhibition.
We fitted this scalar to data to inhibition data for atria and ventricle sodium channels. To predict the impact of different sodium ion channel kinetics on emergent action potential morphology, we introduced the fitted atrial and ventricle sodium channel models into the mouse variant of the Pandit ventricular myocyte model 30 . We used the same model for both the atria and ventricle sodium channel model.

Western Blotting
Tissue samples were homogenised in homogenisation buffer (Tris adjusted to

RNAseq
Kallisto (version 0.42.3) was used to quantify the transcripts in all experiments 59  After quantification of transcript abundance using kallisto, we used the tximport package to import quantification data into the DEseq2 package in R 60,61 . To collapse transcripts into gene level summaries, we used the "hsapiens_gene_ensembl" repository through the R package biomaRt 62 . Only genes with at least 10 counts across the row were retained to filter noise at the low end of the distribution of transcript abundance. To identify DE genes, we used the function "DESeqDataSetFromTximport" to ingest the filtered data, then the function "results" to produce lists of DE genes. Unadjusted P-values are produced using the Wald test, and P-values are adjusted for multiple testing corrections using Benjamini-Hochberg.
We used the DEseq2 function lfcShrink to transform expression values to log2 scale. All analyses were performed in R using version 3.5.1.

Statistical Analyses
Data are expressed as mean ± standard error, unless otherwise stated. Data were checked for normal distribution and statistical analysis was performed using (un)paired Student      Each dot represents an individual cell, **p < 0.01, ***p<0.001, ****p<0.0001 (Unpaired t test).