Hydroxychloroquine (HCQ) is an antimalarial drug commonly used to treat certain types of autoimmune diseases, such as systemic lupus erythematosus (SLE), rheumatoid arthritis, and Sjogren’s syndrome [1, 2]. The drug has also been tested for the treatment and prophylaxis of COVID-19 but did not show significant efficacy in treating or preventing COVID-19 [3,4,5]. Various mechanisms of action of HCQ have been discovered [2]: HCQ interferes with Toll-like receptor (TLR) signaling pathways and inhibits lysosomal activity and autophagy. In the immune system, HCQ reduces the production of tumor necrosis factor (TNF), interleukin (IL)-6, and interferon (IFN)α in plasmacytoid dendritic cells, thereby inhibiting the type‑I IFN signaling in SLE [6,7,8]. Based on multiple immunomodulatory actions, SLE is primarily managed with HCQ to reduce disease activity, flare risk, and organ damage [9]. However, long-term and/or high-dose HCQ treatments can lead to several adverse effects, including QT prolongation, cardiomyopathy, and retinopathy [10,11,12]. Specifically, HCQ-induced QT prolongation is directly linked to life-threatening ventricular arrhythmias, such as torsade de pointes (TdP; [13, 14]).

Hydroxychloroquine inhibits several human cardiac ion channels, including rapidly activating delayed rectifier potassium current (IKr), inward rectifier potassium current (IK1), fast sodium current (INa), and L‑type calcium current (ICaL) channels [15, 16]. The net effects of HCQ on cardiac ion channels can prolong action potential duration (APD) in ventricular cells [15,16,17], eventually leading to QT interval prolongation. To examine the potential proarrhythmogenic risk of non-cardiovascular drugs such as HCQ, computational modeling-based drug toxicity screening frameworks (e.g., comprehensive in vitro proarrhythmia assay [CiPA]; [18, 19]) have been proposed and tested [15, 16]. These computational electrophysiology models have been developed based on quantitative patch-clamp data of human cardiomyocytes [20, 21] and recently utilized to perform computational modeling-guided ablation of atrial fibrillation (AF; [22, 23]) or in in silico antiarrhythmic drug trials [24, 25]. However, most of the recent preclinical studies have focused on the HCQ-associated TdP risk stratification in ventricular cells. It is still unclear how HCQ affects atrial electrophysiology and the risk of AF.

In a retrospective cohort of 1647 adult patients with SLE, Gupta et al. [26] discovered that HCQ use was associated with an 88% decreased risk of new-onset AF. This AF-protective effect of HCQ remained consistent in the propensity score-matched analysis. These results suggest that HCQ may play an antiarrhythmic role in AF in patients with SLE. However, potential antiarrhythmic mechanisms of HCQ have remained unknown, and further prospective clinical studies are needed in a general population.

In the present computational study, we systematically examined the potential effects of HCQ on atrial electrophysiology using computational models of human atrial cardiomyocytes. We quantitatively analyzed HCQ-induced changes in APD, calcium transient (CaT), and APD/CaT alternans thresholds by varying HCQ doses in normal sinus rhythm (NSR) and AF conditions. Since the paired-like homeodomain transcription factor 2 (PITX2) plays a key role in AF electrical/structural remodeling and in since genetic variants located on 4q25, an intergenic region near the PITX2 gene, are closely associated with AF [27,28,29,30], we also examined the effects of HCQ in an in silico PITX2+/- AF model.

Methods

Computational modeling of human atrial cardiomyocytes

We used a state-of-the-art mathematical model of the human atrial cardiomyocyte developed by Grandi et al. [20]. The Grandi atrial cell model incorporates comprehensive Ca2+ handling properties in both normal and AF conditions. Briefly, the cell model is described according to the following differential equation:

$$\frac{dV}{dt}=-\frac{I_{\mathrm{ion}}+I_{\text{stim}}}{C_{m}}$$
(1)

where V is the transmembrane potential, Iion is the total ionic current, Istim is the stimulus current, and Cm is the total membrane capacitance. Each ionic current is a variation of the Hodgkin–Huxley-type model, that is, a nonlinear function of the action potential, ion concentration, and ion channel gating state variables. The biophysical details are described in the study by Grandi et al. [20]. Numerical simulations were performed with MATLAB 2021b (MathWorks Inc.), and ordinary differential equations were numerically solved using a variable-step, variable-order method (ode15s function in MATLAB).

To reflect the electrical remodeling in AF, we decreased INa, ICaL, Ito,fast, IKur, ISERCA, increased IK1, IKs, INCX, IRyR, ISR,leak, and added an INaL component as described by Grandi et al. [20]. The PITX2+/- AF condition was modeled by reducing IK1 by 25% and doubling IKr from the wild-type AF baseline condition [27].

Measurement of atrial electrophysiological markers

At a pacing cycling length (PCL) of 1000 ms (=1 Hz), we measured the APD at 90% of repolarization (APD90), APD at 50% of repolarization (APD50), resting membrane potential (RMP), maximum upstroke velocity (Vmax), and calcium transient amplitude (∆CaT; [31, 32]). A steady state was achieved by stimulating the cells for 60 s. The ∆CaT was defined as the difference between the maximum and minimum intracellular Ca2+ concentrations [31,32,33]. We also estimated the refractory period and APD (or ∆CaT) alternans threshold cycle lengths using a dynamic ramp pacing protocol as previously described [32]: PCLs of 1000, 600, and 500 ms were used, and the PCL was decreased to 300 ms in steps of 50 ms and further decreased in steps of 10 ms until failure of 1:1 capture. For each PCL, the cell was stimulated for 16 beats. The APD (or ∆CaT) alternans were evaluated by determining the maximum and minimum APD (or ∆CaT) values in the last three beats that differed by more than 2% [32]. We considered the beat-to-beat APD variation > 10% as prominent APD alternans.

Actions of hydroxychloroquine on cardiac ion channels

A drug block was modeled by reducing ionic current conductance values [34]. The ionic current conductance was scaled using the Hill equation as follows:

$$\frac{1}{1+\left(\frac{\left[\mathrm{HCQ}\right]}{\mathrm{I}C_{50}}\right)^{n}}$$
(2)

where [HCQ] is the HCQ concentration, IC50 is the half-maximal inhibitory concentration, and n is the Hill coefficient. As shown in Table 1, we used HCQ ion channel binding profiles in the study by Thomet et al. [15] and assumed the Hill coefficient of 1 [34]. The maximum free therapeutic plasma concentration (Cmax) of HCQ was 0.495 μM (215 ng/mL; [35]). Since HCQ has a long half-life of approximately 40 days, patients with SLE taking daily HCQ doses have blood HCQ concentrations of 1000–2000 ng/mL [36]. Thus, we systemically changed the HCQ concentrations from 0.5 × Cmax to 16 × Cmax in a log2 scale (i.e., 0, 0.5, 1, 2, 4, 8, 16 × Cmax).

Table 1 The maximum free therapeutic plasma concentration (Cmax) and half-maximal inhibitory concentration (IC50) values of hydroxychloroquine for human cardiac ion channels based on Thomet et al. [15]
Full size table

Statistical analysis

All the numerical simulations and computational analyses were performed with MATLAB 2021b (MathWorks Inc.). The ordinary differential equations used in our simulation are deterministic (not stochastic); thus, we did not perform repetitions. The HCQ concentration–electrophysiological marker relationships were evaluated using Spearman’s correlation coefficients and visualized in symmetrical log scales using Python 3.8 (https://www.python.org/). Values of p < 0.05 was considered statistically significant.

Results

Hydroxychloroquine-induced changes in atrial action potential and calcium transient

We systematically changed the HCQ concentrations from 0.5 × Cmax to 16 × Cmax and measured action potentials and CaTs in in silico human atrial cardiomyocytes. As shown in Fig. 1, HCQ increased the APD and decreased ∆CaT at a pacing frequency of 1 Hz in a dose-dependent manner. The APD prolongation reflects the outward K+ current inhibitions by HCQ, reducing phase 3 of the action potentials. The reduction in ∆CaT may be attributed to the inward Ca2+ inhibition by HCQ. The results were consistent in the NSR, wild-type AF, and PITX2+/- AF conditions. We also measured action potentials and CaTs at a pacing frequency of 2 Hz (Figure S1). The dose-dependent HCQ effects on the APD and ∆CaT were similar to those acquired at a pacing frequency of 1 Hz.

Fig. 1
figure 1

Action potentials (APs) and calcium transients (CaTs) for hydroxychloroquine concentrations of 0–16 × Cmax. Simulated steady-state AP and CaT traces are shown at the pacing frequency of 1 Hz in normal sinus rhythm (a, b), wild-type AF (c, d), and PITX2+/- AF (e, f). HCQ hydroxychloroquine, Cmax maximum free therapeutic plasma concentration, CaT calcium transient, NSR normal sinus rhythm, AF atrial fibrillation, PITX2 paired-like homeodomain transcription factor 2

Full size image

Effects of hydroxychloroquine on atrial electrophysiological markers

To quantitatively evaluate the effects of HCQ on atrial electrophysiology, we measured various electrophysiological markers in the NSR (Fig. 2), wild-type AF (Fig. 3a), and PITX2+/- AF (Fig. 3b) conditions. The HCQ concentrations were positively correlated with the APD90, APD50, RMP, refractory period, APD alternans threshold, and ∆CaT alternans threshold (all p < 0.05). By contrast, HCQ concentrations were inversely correlated with Vmax and ∆CaT (all p < 0.05). When a therapeutic concentration of HCQ (Cmax) was applied, HCQ increased the APD90 by 1.4% in NSR (409.6 to 415.3 ms), 1.8% in wild-type AF (242.8 to 247.1 ms), and 2.6% in PITX2+/- AF (266.8 to 273.8 ms). By contrast, HCQ reduced ∆CaT by 0.5% in NSR (0.258 to 0.257 μM), 0.5% in wild-type AF (0.0699 to 0.0696 μM), and 0.1% in PITX2+/- AF (0.0714 to 0.0714 μM). At Cmax, HCQ did not increase the APD and ∆CaT alternans thresholds in the NSR, wild-type AF, and PITX2+/- AF conditions. Overall, the HCQ-induced changes in atrial electrophysiological markers suggest no significant atrial proarrhythmogenic effects of HCQ at Cmax (Figs. 2 and 3).

Fig. 2
figure 2

Electrophysiological markers in the normal sinus rhythm (NSR) condition. The action potential duration at 90% of repolarization (APD90), APD at 50% of repolarization (APD50), resting membrane potential (RMP), maximum upstroke velocity (Vmax), calcium transient amplitude (∆CaT), refractory period, APD alternans threshold cycle length (CL), and CaT alternans threshold CL are shown, depending on hydroxychloroquine concentrations of 0–16 × Cmax (see Methods). Red lines indicate Cmax. Spearman’s correlation coefficients and p values are presented. HCQ hydroxychloroquine, Cmax maximum free therapeutic plasma concentration, NSR normal sinus rhythm

Full size image
Fig. 3
figure 3

Electrophysiological markers in the wild-type AF (a) and PITX2+/- AF (b) conditions. The action potential duration at 90% of repolarization (APD90), APD at 50% of repolarization (APD50), resting membrane potential (RMP), maximum upstroke velocity (Vmax), calcium transient amplitude (∆CaT), refractory period, APD alternans threshold cycle length (CL), and CaT alternans threshold CL are shown, depending on hydroxychloroquine concentrations of 0–16 × Cmax (see Methods). Red lines indicate Cmax. Spearman’s correlation coefficients and p values are presented. HCQ hydroxychloroquine, Cmax maximum free therapeutic plasma concentration, AF atrial fibrillation, PITX2 paired-like homeodomain transcription factor 2

Full size image

However, HCQ at 2 × Cmax began to increase the APD alternans threshold in the PITX2+/- AF model (Fig. 3b). To examine the effects of HCQ on alternans, we paced in silico atrial cells at 4 Hz. As highlighted in Fig. 4, HCQ at 4 × Cmax and 16 × Cmax caused prominent alternating long and short action potentials in the PITX2+/- AF condition (APD variation > 10%, Fig. 4c), whereas the NSR and wild-type AF models did not show prominent APD alternans (APD variation < 10%).

Fig. 4
figure 4

Simulated action potentials (APs) at the pacing frequency of 4 Hz in normal sinus rhythm (a), wild-type AF (b), and PITX2+/- AF (c). With the presence of hydroxychloroquine, alternating long and short APs are prominent in PITX2± AF (beat-to-beat APD variation > 10%; c). HCQ hydroxychloroquine, Cmax maximum free therapeutic plasma concentration, NSR normal sinus rhythm, AF atrial fibrillation, PITX2 paired-like homeodomain transcription factor 2

Full size image

Discussion

We systematically measured atrial electrophysiological markers after applying HCQ to in silico human atrial cardiomyocytes. We found that HCQ increased the APD and decreased ∆CaT but did not change the alternans thresholds at Cmax. Additionally, HCQ reduced the maximum upstroke velocity, potentially slowing the myocardial conduction velocity. Since AF is characterized by shortening of the APD and rate-dependent APD alternans occurrence [33, 37, 38], our computational results suggest no significant atrial arrhythmogenic effects of HCQ at Cmax, instead indicating a potential antiarrhythmic role of low-dose HCQ in AF. However, with high doses of HCQ and high heart rates, APD alternans were prominent, particularly in the PITX2+/- AF condition. This potential proarrhythmogenic role of high-dose HCQ in PITX2+/- AF should be further investigated experimentally.

Recent studies suggest that HCQ may play a cardioprotective role in patients with autoimmune diseases [39]. Gupta et al. [26] showed that HCQ use was associated with an 88% decrease in the risk of AF in patients with SLE. We found that HCQ increased the atrial APD in a dose-dependent manner without affecting the alternans thresholds at Cmax. Our results suggest potential antiarrhythmic effects of HCQ against AF. Other studies have shown that HCQ treatments did not significantly increase the risk of cardiac arrhythmias, including ventricular arrhythmias, in patients with SLE [40, 41] and common rheumatic diseases [42]. Although some case reports showed that HCQ could lead to QT prolongation [13, 43], HCQ use was not associated with QTc length in a large cohort of patients with rheumatic diseases [44]. By contrast, HCQ may increase the risk of life-threatening ventricular arrhythmias in patients with COVID-19 [45]. In patients with COVID-19, HCQ treatment increased the risks of QT interval prolongation, TdP, and cardiac arrest [45,46,47,48,49]. For COVID-19, HCQ is typically started at high dosages of 2000 mg and reduced to 800 mg, which is higher than the typical maintenance dosage of 200 mg in patients with SLE. Therefore, adverse cardiac effects of HCQ in COVID-19 may be attributed to the use of high-dose HCQ, which could lead to prominent APD alternans according to our data (Fig. 4). Further large-scale cohort studies are needed to determine the causal effects of HCQ on AF.

Most antiarrhythmic drugs for AF increase the atrial APD. However, an excessively prolonged atrial APD can lead to atrial arrhythmias [50] and concomitant effects on ventricular APD can induce TdP [51, 52]. Our data on atrial electrophysiology and other studies suggest [15, 16] that HCQ increases both the atrial and ventricular APDs. Thus, estimating the safe therapeutic range of HCQ is essential in order to reduce the risk of AF and TdP. In our study, HCQ increased the atrial APDs by only 1.4–2.6% at Cmax. However, patients with SLE taking daily HCQ doses may have high blood HCQ concentrations due to its long half-life [36]. With high-dose HCQ, we observed prominent APD alternans, particularly in the PITX2+/- AF condition (Fig. 4). Therefore, patients with concomitant autoimmune diseases and PITX2 mutations may have an increased risk of AF, and this potential proarrhythmogenic effect should be further investigated using real-world clinical data.

Besides the direct effects of HCQ on atrial electrophysiology, HCQ may play a cardioprotective role in AF by inhibiting autophagy flux. Yuan et al. [53] discovered that autophagy induces atrial electrical remodeling through ubiquitin-dependent degradation of the L‑type calcium channel. Additionally, serum autophagy-related 5 (ATG5) was associated with left atrial enlargement in patients with paroxysmal AF [54]. Thus, interruption of autophagy flux by HCQ may prevent the electrical and structural remodeling of AF.

Cardiac computational modeling approaches can be used to quantitatively understand complex subcellular or spatiotemporal dynamics of AF [23, 25, 33] and to perform personalized AF ablation as shown in the OPTIMA [22] and PersonAL [23] studies. Recently, these computational models have also been utilized for in silico antiarrhythmic drug trials and drug toxicity screening [18, 19, 24, 25]. Although we did not perform 2D/3D tissue-level simulations, our computational approaches can be directly extended to examine how HCQ quantitatively affects the spatiotemporal wave dynamics of AF, antiarrhythmic drug efficacies, and ablation outcomes in normal, wild-type AF, and PITX2+/- AF conditions [55]. We also expect that cardiac computational modeling might be helpful for personalizing AF risk stratification and management in patients with SLE taking HCQ.

Limitations

There are several limitations to our computational study. Although we used a state-of-the-art computational model of the human atrial cardiomyocyte [20], the results need to be validated using human induced pluripotent stem cells, animal models, or clinical electrophysiology data. We assumed fixed baseline parameter conditions; however, population-based approaches would provide more statistically robust results by considering the biological uncertainty/variability in ion channel parameters [17, 25, 31, 32]. Since atrial cell models have pacing protocol-dependent memory effects [56], various pacing protocol-related sensitivity analyses should be considered to obtain more robust and reliable results [57]. To elucidate the comprehensive mechanisms of action of HCQ for drug repurposing in AF, transcriptome/proteome perturbation data or protein–protein interaction network analysis would be helpful [58, 59].

Conclusion

In the computational model of human atrial cardiomyocytes, HCQ increases the atrial APD and decreases ∆CaT but does not change the alternans thresholds at Cmax. The overall results suggest a potential antiarrhythmic role of low-dose HCQ in AF. However, concomitant PITX2 mutations and high-dose HCQ treatments may increase the risk of AF, and this potential genotype/dose-dependent effect of HCQ should be investigated experimentally.