Mitochondria-derived ROS bursts disturb Ca²⁺ cycling and induce abnormal automaticity in guinea pig cardiomyocytes: a theoretical study

Abstract

Mitochondria are in close proximity to the redox-sensitive sarcoplasmic reticulum (SR) Ca²⁺ release [ryanodine receptors (RyRs)] and uptake [Ca²⁺-ATPase (SERCA)] channels. Thus mitochondria-derived reactive oxygen species (mdROS) could play a crucial role in modulating Ca²⁺ cycling in the cardiomyocytes. However, whether mdROS-mediated Ca²⁺ dysregulation translates to abnormal electrical activities under pathological conditions, and if yes what are the underlying ionic mechanisms, have not been fully elucidated. We hypothesize that pathological mdROS induce Ca²⁺ elevation by modulating SR Ca²⁺ handling, which activates other Ca²⁺ channels and further exacerbates Ca²⁺ dysregulation, leading to abnormal action potential (AP). We also propose that the morphologies of elicited AP abnormality rely on the time of mdROS induction, interaction between mitochondria and SR, and intensity of mitochondrial oxidative stress. To test the hypotheses, we developed a multiscale guinea pig cardiomyocyte model that incorporates excitation-contraction coupling, local Ca²⁺ control, mitochondrial energetics, and ROS-induced ROS release. This model, for the first time, includes mitochondria-SR microdomain and modulations of mdROS on RyR and SERCA activities. Simulations show that mdROS bursts increase cytosolic Ca²⁺ by stimulating RyRs and inhibiting SERCA, which activates the Na⁺/Ca²⁺ exchanger, Ca²⁺-sensitive nonspecific cationic channels, and Ca²⁺-induced Ca²⁺ release, eliciting abnormal AP. The morphologies of AP abnormality are largely influenced by the time interval among mdROS burst induction and AP firing, dosage and diffusion of mdROS, and SR-mitochondria distance. This study defines the role of mdROS in Ca²⁺ overload-mediated cardiac arrhythmogenesis and underscores the importance of considering mitochondrial targets in designing new antiarrhythmic therapies.

cardiovascular diseases kill nearly 400,000 Americans each year (41), and many of these deaths are the result of ventricular arrhythmias leading to sudden cardiac arrest. Although arrhythmias are multifactorial and involve a sophisticated interplay among various subcellular systems such as electrophysiology, force generation, and energy metabolism, a defective intracellular Ca²⁺ cycling has been implicated to play a crucial role in arrhythmogenesis (10, 14, 51). However, in spite of extensive efforts, the detailed molecular mechanisms underlying Ca²⁺ dysregulation and Ca²⁺-mediated arrhythmias are not completely understood, hindering the development of effective therapeutic strategies.

As an important universal signaling ion, cardiac Ca²⁺ is precisely regulated by a complicated system comprising of multiple sarcolemmal and intracellular ion channels and transporters that activate and inactivate in a coordinated fashion during the cardiac cycle (6). Of those channels, the sarcoplasmic reticulum (SR) Ca²⁺ release channels, also known as ryanodine receptors (RyRs), are the primary Ca²⁺ release sites. Structural studies have shown that tetrameric RyRs have 89 cysteine residues, of which ∼21 are susceptible to oxidation by free radicals (61). Oxidation of RyR cysteine thiol residues forms disulfide bonds, leading to reversible activation of channel activity. Gen et al. (25) have shown that extracellular H₂O₂ activates SR Ca²⁺ release and causes Ca²⁺ overload in rat ventricular myocytes. Similarly, exogenous H₂O₂ can directly modify the gating of sheep cardiac RyRs and increase the channel open probability (21). In addition to RyRs, reactive oxygen species (ROS) also affect SR Ca²⁺ uptake proteins such as SR Ca²⁺-ATPase (SERCA). It has been shown that H₂O₂ and superoxide (a.k.a. O₂^·−) inhibit SERCA activity by directly oxidizing its thiol groups (42, 68) or indirectly interfering with its ATP binding sites (60). Although the effects of exogenous ROS on SR Ca²⁺ handling proteins are evident (4, 29, 59), study of how endogenous ROS influence Ca²⁺ homeostasis has begun only recently (49, 62, 63, 66), partially due to the difficulty of inducing controllable endogenous ROS production in the cell in the experimental setting.

There is convincing evidence that mitochondria, the major sites of intracellular ROS production, and SR are in close proximity and physically tethered via mitofusin proteins (12, 19). This unique mitochondria-SR tethering forms a close coupling of Ca²⁺ signaling between SR release sites and nearby mitochondria, facilitating rapid mitochondrial Ca²⁺ uptake that stimulates oxidative phosphorylation (11). This coupling is essential for matching energy supply with demand in response to increased workload (39, 52). We speculate that the mitochondria-SR tethering is also involved in the interorganellar redox signaling, allowing dynamic modulation of SR Ca²⁺ handling by mitochondria-derived ROS (mdROS). In supporting this notion, recent studies have shown that mitochondrial depolarization and associated ROS-induced ROS release (RIRR) are closely correlated to enhanced Ca²⁺ sparks in resting guinea pig cardiomyocytes (62, 66). In addition, others have reported that mdROS cause significant changes in global Ca²⁺ transients and regional Ca²⁺ sparks in smooth muscle cells (13), skeletal muscles (46), and cerebral arteries (58). Given these advancements, how mdROS may influence Ca²⁺ cycling and electrical activity in paced cardiomyocytes has not been fully elucidated. We hypothesize that the pathological mdROS can diffuse to the proximal SR and dynamically alter SR Ca²⁺ handling (e.g., RyRs and SERCA), disrupting Ca²⁺ cycling and causing erratic action potential (AP) generation.

To test the hypothesis, we developed a novel guinea pig cardiomyocyte model that integrates exciation-contraction coupling, Ca²⁺ handling, and mitochondrial energetics, as well as the modulations of RyRs/SERCA by the surrounding mdROS. The simulations show that mdROS bursts rapidly induce cytosolic Ca²⁺ ([Ca²⁺]_i) overload by stimulating RyRs and inhibiting SERCA. The elevating [Ca²⁺]_i activates the Na⁺/Ca²⁺ exchanger (NCX) and Ca²⁺-sensitive nonspecific cationic channels (nsCa) that underlie transient inward currents (I_ti), further exacerbating Ca²⁺ dysregulation and leading to APs triggered by the abnormal automaticity.

METHODS

Model Development

In this study, a multiscale guinea pig cardiomyocyte model was developed to examine the effect of mitochondrial oxidative stress on Ca²⁺ regulation and cellular electrophysiological behaviors. This model was based on our recently published exciation-contraction coupling, mitochondrial energetics and ROS-induced ROS release (ECME-RIRR) model (37, 67) and incorporated new model components including local control Ca²⁺-induced Ca²⁺-release (CICR), mitochondria-SR microdomain (MSM), and mdROS modulations of RyRs and SERCA. In a recent study we showed that the mdROS-mediated Ca²⁺ sparks can be suppressed by TMPyP (a O₂^·− dismutase mimetic) or 4′-Cl diazepam (the peripheral benzodiazepine receptor ligand that blocks mitochondrial O₂^·− release) (66), suggesting that mitochondria-derived O₂^·− (mdO₂^·−) plays a major role in the dynamic modulation of SR Ca²⁺ release during RIRR. Thus only the direct effect of mdO₂^·− on SR Ca²⁺ handling proteins was considered in the present model development. For simplification, we assumed that all mitochondria in the cardiomyocyte depolarize/oscillate synchronously (1). It is worth noting that such a synchronization may only occur in pathological, rather than physiological, conditions (2). Consequently, the mitochondria were lumped as a giant mitochondrion and mitochondria-SR subspaces were modeled as a single compartment. The schemes of the new whole cell model and the MSM model are shown in Supporting Information Fig. S1 and Fig. S2, respectively (see endnote).

mdO₂^·− diffusion.

Studies have shown that under pathophysiological conditions, mitochondrial O₂^·− production increases significantly and upon reaching a threshold triggers the opening of the inner membrane anionic channels (1, 3, 65) and/or mitochondrial permeability transition pore, leading to RIRR (69). We hypothesize that the pathological mdO₂^·− could diffuse from the perimitochondrial space to the proximal SR, dynamically modulating the redox-sensitive Ca²⁺ channels. Assuming that MSM is homogeneous, the concentration profiles of mdO₂^·− in the microdomain ([O₂^·−]_MSM) can be described by the Fick's second law (3):

$$\frac{\partial {\left[{\text{O}}_2^{·-}\right]}_{\text{MSM}}\left(x,t\right)}{\partial t}=D_{{\text{O}}_2^{·-}}\cdot \frac{{\partial }^2{\left[{\text{O}}_2^{·-}\right]}_{\text{MSM}}\left(x,t\right)}{\partial x^2}+v_{\text{cyto\_MSM}}\cdot \text{f}\left({\left[{\text{O}}_2^{·-}\right]}_{\text{MSM}}\left(x,t\right)\right)$$ (1)

where D_O2^·− is O₂^·− diffusion coefficient, x is the distance from the mitochondrion, v_{cyto_MSM} is cytosol and MSM effective volume ratio, and f([O₂^·−]_MSM(x, t)) is O₂^·− scavenging rate.

The numerical simulation of Eq. 1 was performed with a finite difference method, whereby the spatial component at x was approximated by the following expression (Supporting Information Fig. S3A):

$$D_{{\text{O}}_2^{·-}i}\frac{{\partial }^2{\left[{\text{O}}_2^{·-}\right]}_{\text{MSM}}\left(x,t\right)}{\partial x^2}=\frac{D_{{\text{O}}_2^{·-}i}}{\Delta x^2}\left[\left({\left[{\text{O}}_2^{·-}\right]}_{\text{MSM}}\left(x+\Delta x,t\right)-{\left[{\text{O}}_2^{·-}\right]}_{\text{MSM}}\left(x,t\right)\right)-\left({\left[{\text{O}}_2^{·-}\right]}_{\text{MSM}}\left(x,t\right)-{\left[{\text{O}}_2^{·-}\right]}_{\text{MSM}}\left(x-\Delta x,t\right)\right)\right]$$ (2)

where Δx is the spatial step size. To reduce computation time, the MSM was discretized as two subcompartments, with one (MSM_SR) adjacent to the peri-SR space and another (MSM_mito) adjacent to the perimitochondrial space (Supporting Information Fig. S3B). Assuming nonflux boundary conditions (i.e., $\frac{\partial {\left[{\text{O}}_2^{-}\right]}_{\text{MSM}}\left(0,t\right)}{\partial x}$ = 0 and $\frac{\partial {\left[{\text{O}}_2^{-}\right]}_{\text{MSM}}\left(X,t\right)}{\partial x}$ = 0, where X is SR-mitochondrion distance), the concentration of mdO₂^·− in the peri-SR space ([O₂^·−]_SR), and perimitochondrial space ([O₂^·−]_{p_mito}) can be approximated to equal [O₂^·−]_{MSM_SR} and [O₂^·−]_{MSM_mito}, respectively. Consequently, the concentration profile of [O₂^·−]_SR can be described by the following equation (the detailed derivation of the equation can be found in Supporting Information S1):

$$\frac{\text{d}{\left[{\text{O}}_2^{·-}\right]}_{\text{SR}}\left(t\right)}{\text{d}t}=D_{{\text{O}}_{\text{2}}^{·-}}\cdot \frac{{\left[{\text{O}}_{\text{2}}^{·-}\right]}_{\text{p}\_\text{mito}}\left(t\right)-{\left[{\text{O}}_{\text{2}}^{·-}\right]}_{\text{SR}}\left(t\right)}{X^2}+v_{\text{cyto}\_\text{MSM}}\cdot f\left({\left[{\text{O}}_{\text{2}}^{·-}\right]}_{\text{SR}}\left(t\right)\right)$$ (3)

The values of newly added model parameters (e.g., D_O2^·− and X) were taken from the literature and listed in the Supporting Information (Table S7.16). The effect of diffusion coefficient on mdO₂^·− concentration in the peri-SR space is plotted in Fig. 1A, which shows that the larger the diffusion coefficient, the higher the SR mdO₂^·− concentration.

Fig. 1.

Dynamics of mitochondria-derived O₂^·− (mdO₂^·−) and its effect on sarcoplasmic reticulum (SR) Ca²⁺ handling channels. A: model simulated concentration of mdO₂^·− in the per-SR space ([O₂^·−]_SR) with different diffusion coefficient, D_O2^·−. B: stimulatory effect of O₂^·− on ryanodine receptor (RyR) open probability (P_{O_ryr}). Line is the least square curve fit and dots are experimental data [circles from Boraso and Williams (8); triangles and squares from Oba et al. (44, 45)]. C: inhibitory effect of O₂^·− on SR Ca²⁺-ATPase (SERCA) Ca²⁺ uptake (J_up). Line is the least square curve fit and square dots are experimental data taken from Ref. 60. D: model simulated mitochondrial depolarization and associated O₂^·− burst triggered by increasing the shunt (from 2% to 10%), which shows that [O₂^·−]_{p_mito} and [O₂^·−]_SR bursts occurred concurrently with mitochondrial depolarization. RSS, residual sum of squares; ΔΨ_m, membrane potential.

Local Ca²⁺ control.

To account for the local interaction of L-type Ca²⁺ channels (LCCs) and RyRs in the dyadic microdomain that controls CICR, we incorporated the coupled 40-state LCC-RyR model developed by Gauthier et al. (24) into the ECME-RIRR model (67) (see Supporting Information S2 for details). Since in the ECME-RIRR model the SR is separated into two compartments (i.e., NSR and JSR) that is different from the Gauthier et al. model, several model equations and parameters were modified (see Supporting Information S3 for details) to better fit the simulated current-voltage (I– V) relationship and L-type Ca²⁺ current (I_CaL) trace during steady-state AP with experimental data (Supporting Information Fig. S4). It is worthy to note that the values of the peak current of I_CaL were different among experiments, which might be caused by the variance of channel density (24). In our model, the number of LCC [i.e., the number of Ca²⁺ release unit (N_CaRU)] was set to 300,000, which is within the range of experimental measurements [from ∼276,000 (7) to ∼500,000 (27)]. Other model parameters were the same as those used in the previous model (24). The modified ECME-RIRR model showed a ∼37% decrease of SR Ca²⁺ content during a normal AP cycle, which is comparable to that reported in the previous studies (∼35%) (5, 24).

ROS and RyR Ca²⁺ release.

Experimental studies have shown that ROS exponentially increase RyR open probability (P_{O_ryr}), with the enhancement effect determined by both the ROS concentration and the state of the channel: when P_{O_ryr} is low, the enhancing effect of ROS is dramatic, but when P_{O_ryr} is already high, the stimulatory effect of ROS is much less significant. Consequently, the RyR open probability in the presence of ROS (P_{O_ryr}__ROS) can be described by an exponential equation_, which is a function of [O₂^·−]_SR and P_{O_ryr}:

$$P_{\text{O}\_\text{ryr}\_\text{ROS}}=c_{\text{ryr}}\cdot \left(1-\exp \left(-k_{\text{ryr}}\left({\left[{\text{O}}_2^{·-}\right]}_{\text{SR}}-\frac{1}{k_{\text{ryr}}}\ln \left(1-\frac{P_{\text{O}\_\text{ryr}}}{c_{\text{ryr}}}\right)\right)\right)\right)$$ (4)

where c_ryr is ROS enhancement coefficient and k_ryr is effective factor. The values of these parameters (0.20 and 19.55 mM⁻¹, respectively) were obtained by the least-square curve fitting using experimental data from the literature (8, 44, 45) (Fig. 1B). The enhancement of ROS on P_{O_ryr} was incorporated into the RyR Ca²⁺ release formula (Supporting Information S4).

ROS and SERCA Ca²⁺ uptake.

Unlike RyRs, SERCA Ca²⁺ uptake (J_up; Supporting Information Eq. S6.E47) is exponentially inhibited by ROS and the effect can be described by:

$$J_{\text{up}\_\text{ROS}}=J_{\text{up}}\cdot \left[c_{\text{SERCA}}\cdot \exp \left(-k_{\text{SERCA}}\cdot {\left[{\text{O}}_2^{·-}\right]}_{\text{SR}}\right)\right]$$ (5)

where c_SERCA (= 1.02) is the inhibition coefficient and k_SERCA (= 43.67 mM⁻¹) is the effective factor of ROS inhibition. These values were obtained using the least-square curve fitting method. The experimental data used for parameter optimization is from Refs. 36, 60, and the fitting result is shown in Fig. 1C. Equation 5 (J_{up_ROS}) was then added back to ECME-RIRR model to replace J_up in Eq. S6.E47 (Supporting Information).

Simulation Protocol

The model formulations and parameters of other processes (e.g., ion currents and metabolic reactions) were the same as those in the ECME-RIRR model (67) and CICR model (24) unless otherwise indicated (see Supporting Information S6 and S7). The whole cell model was coded in C⁺⁺ (Visual Studio; Microsoft, Redmond, WA). The nonlinear ordinary differential equations were integrated numerically with CVODE as described previously (67).

Before examining the effect of mdO₂^·− on SR Ca²⁺ handling and cellular electrophysiology, the behavior of a paced (0.5 Hz) cardiomyocyte was simulated under physiological conditions (i.e., mdO₂^·− production was at physiological level). The obtained steady-state values were then fed into the model as initial conditions (Supporting Information Table S7.17) for all runs in the subsequent simulations. Mitochondrial depolarization (and associated mdO₂^·− burst) was induced by increasing the fraction of O₂^·− production from the electron transport chain (a.k.a. shunt) from 2 to 10% as described previously (67). To systematically examine the effect of mdO₂^·− on Ca²⁺ handling and AP generation, the mdO₂^·− burst was induced at different time during the AP cycle (e.g., phase 2 or 4). In some subset simulations, the extent of mdO₂^·− production or the distance between mitochondria and SR was altered to examine its effect on the mdROS-mediated Ca²⁺ transient and AP alterations. APD₉₀ was defined as the interval between the time of the maximum upstroke velocity of the AP, [dV/dt]_max, and 90% repolarization.

RESULTS

Model Validation

After parameters were optimized, the new/modified model modules were incorporated into the ECME-RIRR model. We first simulated the dynamics of mitochondrial energetics under stress conditions. Increasing shunt from 2 to 10% triggered sustained mitochondrial oscillations including membrane potential (ΔΨ_m) and mdO₂^·− production (Supporting Information Fig. S5), as well as NADH oxidation and reduced glutathione depletion (data not shown) in a paced cardiomyocyte (0.5 Hz). These simulations were consistent with our previous experimental and computational studies (1, 17, 65, 67), suggesting that the addition of new model components (e.g., mdO₂^·− modulation of RyRs and local Ca²⁺ control) did not influence the dynamics of the existing model subsystems. Simulations also showed that [O₂^·−]_{p_mito} and [O₂^·−]_SR bursts occurred concurrently with ΔΨ_m depolarization during each oscillation cycle (see Fig. 1D for enlargement). Thus, for better visualization, ΔΨ_m depolarization was plotted to represent the O₂^·− burst in some figures.

The model was further validated by simulating mdROS-mediated abnormal APs and comparing the results with experimental data. Due to the lack of experimental studies on the direct effect of mdROS on APs in guinea pig cardiomyocytes, the comparisons were made between model simulations and data obtained from isolated rabbit cardiomyocytes exposed to external oxidative stress. A mdO₂^·− burst induced before AP firing led to an early afterdepolarization (EAD; Fig. 2A) similar to that observed in a cardiomyocyte subject to H₂O₂ perfusion (Fig. 2A, inset) (59), including the ∼2.5-fold APD₉₀ prolongation (Fig. 2B). The mdO₂^·− burst also caused significant increase in Ca²⁺ transient (1.63-fold higher than that of normal AP cycles) and reduction in SR Ca²⁺ storage (∼69%; Fig. 2C). The mdO₂^·− burst-induced [Ca²⁺]_i elevation was comparable to that reported in a recent experimental study (57). In addition, a mdO₂^·− burst induced during phase 4 of the AP cycle elicited a delayed afterdepolarization (DADs) that was also observed in the H₂O₂ perfusion experiments (Fig. 3A, inset) (59). These validations, although semiquantitatively, demonstrated the utility and capability of our model in examining the interaction between mitochondrial energetics and cellular electrophysiology.

Fig. 2.

Model-simulated mdO₂^·− burst-induced early afterdepolarization (EAD) in a “beating” (0.5 Hz) cardiomyocyte. A: mitochondria-derived reactive oxygen species (mdROS) burst-elicited EAD (inset: EAD observed in a rabbit cardiomyocyte subjected to ROS perfusion; Ref. 59). B: comparison of APD₉₀ between model simulation (guinea pig) and experimental data (rabbit) under control and oxidative stress conditions {where APD₉₀ is the interval between the time of the maximum upstroke velocity of the action potential (AP), [dV/dt]_max, and 90% repolarization}. C– F: model simulated intracellular Ca²⁺ concentration ([Ca²⁺]_i) and SR Ca²⁺ concentration ([Ca²⁺]_SR) (C), RyR Ca²⁺ release (J_rel), and SERCA Ca²⁺ uptake (J_up) (D); Na⁺/Ca²⁺ exchanger current (I_NaCa) and Ca²⁺-sensitive nonspecific cationic current (I_nsCa) (E); and I_CaL (F) before and after a mdROS burst. The solid arrow in A represents the AP reversal; in C, D, and E, the [Ca²⁺]_i elevation, SR Ca²⁺ release, and activations of I_NaCa and I_nsCa caused by the mdO₂^·− burst, respectively; in F, the L-type Ca²⁺ current (I_CaL) reactivation. The dashed arrow in C, D, and E represents the second [Ca²⁺]_i elevation, SR Ca²⁺ release, and activations of I_NaCa and I_nsCa caused by I_CaL reactivation, respectively. The dashed lines represent ΔΨ_m and its depolarization represents the acute mdO₂^·− burst.

Fig. 3.

Model-simulated mdO₂^·− burst-induced delayed afterdepolarizations (DAD) in a “beating” (0.5 Hz) cardiomyocyte. A: mdROS burst-induced DAD (inset: DAD observed in a cardiomyocyte perfused by external H₂O₂; Ref. 59). B– F: model simulated SR Ca²⁺ release (J_rel) (B), [Ca²⁺]_i and [Ca²⁺]_SR (C), I_NaCa (D), I_nsCa (E), and I_CaL (F) before and after the induction of a mdROS burst. The dash lines represent ΔΨ_m and its depolarization represents the acute mdO₂^·− burst. The arrow in C indicates the mdO₂^·− burst-induced SR Ca²⁺ depletion and in D and E represents activation of I_NaCa and I_nsCa.

Effect of mdO₂^·− on Ca²⁺ Handling Channels

We next examined the ionic mechanisms underlying the mdO₂^·−-induced EAD or DAD described above. When a mdO₂^·− burst was induced before AP firing, the activated J_rel (Fig. 2D, black line, solid arrow) and inhibited J_up (Fig. 2D, gray line) caused not only [Ca²⁺]_i elevation but also [Ca²⁺]_SR reduction (Fig. 2C, solid arrow). The accumulated cytosolic Ca²⁺ subsequently enhanced the Na⁺/Ca²⁺ exchanger current (I_NaCa) in the forward mode (Fig. 2E, black curve, solid arrow) and the Ca²⁺-sensitive nonspecific cationic current (I_nsCa) (Fig. 2E, gray curve), eliciting I_ti that caused the reduction of AP repolarization reserve (Fig. 2A, arrow). Intriguingly, the transient reversal of AP repolarization reactivated the I_CaL (Fig. 2F, arrow), which triggered a second CICR (Fig. 2D, dashed arrow), causing a further [Ca²⁺]_i elevation and [Ca²⁺]_SR depletion (Fig. 2C, dashed arrow). The resultant Ca²⁺ overload caused further I_NCX and I_nsCa activation (Fig. 2E, dashed arrow), which eventually elicited an EAD (Fig. 2A).

In the case of DAD (Fig. 3A, inset), the mdO₂^·−-activated J_rel (Fig. 3B) evoked an extra Ca²⁺ transient, which was slightly smaller than that of normal AP cycles (Fig. 3C). The enhanced RyR Ca²⁺ release also caused a significant SR Ca²⁺ depletion (Fig. 3C, arrow). Consequently, both I_NaCa and I_nsCa were enhanced (Fig. 3, D and E, arrow), which induced I_ti that underlay the DAD. However, I_CaL was not reactivated since the membrane potential was low (Fig. 3F).

Effect of Timing of mdROS Burst Induction on AP Morphology

Simulations shown in Figs. 2 and 3 suggest that the pattern of mdO₂^·−-induced AP abnormality is dependent on the timing of mdO₂^·− burst induction during a AP cycle or the time interval between the firing of AP and the burst of mdO₂^·− (t_AP-mdROS). To better understand the ionic mechanisms underlying the mdO₂^·−-induced AP abnormality, we analyzed the correlation between t_AP-mdROS and the morphology of mdO₂^·−-mediated AP.

Figure 4Ai shows the control AP (i.e., mdO₂^·− production was at physiological level). The mdO₂^·− burst induced before AP firing (e.g., t_AP-mdROS = −160 ms) caused APD₉₀ prolongation (from 167 to 274 ms) (Fig. 4Aii). A mdO₂^·− burst began to elicit EADs when t_AP-mdROS gradually increased. The mdROS burst induced at t_AP-mdROS = −130 ms caused a single EAD (Fig. 4Aiii), and a delayed mdO₂^·− burst (e.g., t_AP-mdROS = 117 ms) elicited multiple EADs (Fig. 4Aiv). The multiple EADs degraded into a single EAD (Fig. 4Av) and then APD prolongation (data not shown) when the occurrence of mdO₂^·− burst was further postponed to phase 3 of AP. Finally, when mdO₂^·− burst was induced after the AP was fully repolarized (e.g., t_AP-mdROS = 355 ms), a DAD (Fig. 4Avi) ensued.

Fig. 4.

Effect of t_AP-mdROS (the time interval between mdO₂^·− burst induction and AP firing) on the mdO₂^·−-induced [Ca²⁺]_i and AP abnormality. i: control; ii: t_AP-mdROS = −160 ms; iii: t_AP-mdROS = −130 ms; iv: t_AP-mdROS = 117 ms; v: t_AP-mdROS = 136 ms; and 6) t_AP-mdROS = 355 ms. A– E: AP, [Ca²⁺]_i, I_NaCa, I_nsCa, and I_CaL, respectively. The arrows in Eiii-Ev represent I_CaL reactivation(s). F: summary of the effect of t_AP-mdROS on mdO₂^·− burst-induced AP abnormality: a mdO₂^·− burst induced in zone 1 (−135 to −37 and 136 to 140 ms) elicited a single EAD, in zone 2 (−36 to 135 ms) elicited multiple EADs, in zone 3 (147 to 179 ms and −680 to −136 ms) elicited APD prolongation, and in zone 4 (the remaining time window) elicited a DAD. The dash lines represent ΔΨ_m and its depolarization represents the acute mdO₂^·− burst.

The dynamics of [Ca²⁺]_i and Ca²⁺ currents were then analyzed to explore the ionic mechanism accounting for the variance of the mdO₂^·− burst-induced APs. When t_AP-mdROS = −160 ms (Fig. 4Aii), a mdO₂^·− burst caused a moderate [Ca²⁺]_i increase (Fig. 4Bii), which activated I_NaCa (Fig. 4Cii) in the forward mode and enhanced I_nsCa (Fig. 4Dii). The resultant I_ti caused a small AP depolarization and prolongation of APD₉₀ (Fig. 4Aii) and I_CaL (Fig. 4Eii) but could not reverse AP repolarization.

When t_AP-mdROS = −130 ms, the mdO₂^·− burst gradually raised [Ca²⁺]_i (Fig. 4Biii), which augmented I_NaCa and I_nsCa (Fig. 4, Ciii and Diii) and evoked I_ti, which reduced AP repolarization reserve, reactivated I_CaL (Fig. 4Eiii), and elicited an EAD (Fig. 4Aiii). The reactivated I_CaL subsequently triggered CICR and led to further increases of [Ca²⁺]_i and augmentation of I_ti. Further postponing the induction of mdROS (e.g., t_AP-mdROS = 117 ms; Fig. 4Aiv) led to a larger Ca²⁺ transient, which enhanced I_NaCa and I_nsCa (Fig. 4, Biv-Div). The resultant large I_ti activated I_CaL repetitively (Fig. 4Eiv), generating an EAD with multiple phases (Fig. 4Aiv).

When the mdROS was induced in phase 3 (e.g., t_AP-mdROS = 136 ms) of the AP cycle, the mdO₂^·− burst-induced Ca²⁺ elevation (Fig. 4Bv) and I_NaCa and I_nsCa enhancements (Fig. 4, Cv and Dv) could only trigger a single EAD (Fig. 4Av). As the cell membrane voltage was low when I_CaL was reactivated, the amplitude of reactivated I_CaL was larger (Fig. 4Ev, arrow) than that evoked in the previous cases (Fig. 4, Eiii and Eiv). This is consistent with the experimental result showing that a more repolarized AP produced larger EAD amplitude (59). Further delay of the mdROS burst in phase 3 diminished mdO₂^·− burst-induced Ca²⁺ elevation and induced APD prolongation (data not shown).

When the mdO₂^·− burst was induced in phase 4 of the AP cycle (Fig. 4Avi), the mdO₂^·− burst-induced SR Ca²⁺ release generated a second Ca²⁺ transient (Fig. 4Bvi). In this case, the gradual AP depolarization induced by I_NaCa and I_nsCa activations (Fig. 4, Cvi and Dvi) slightly depolarized the membrane potential and evoked a DAD (Fig. 4Avi). However, I_CaL was not activated (Fig. 4Evi). The relationship between t_AP-mdROS and mdO₂^·− burst-induced AP abnormality is summarized in Fig. 4F, with zones 1–4 representing single EAD, multiple EADs, APD prolongation, and DADs, respectively.

Effect of mdO₂^·−-Mediated RyR Activation or SERCA Inhibition Alone on Ca²⁺ Cycling and AP

We next examined how mdO₂^·− could affect Ca²⁺ transients and AP via modulating solely the activity of RyRs or SERCA. A mdO₂^·− burst induced during AP firing, when modeled to activate RyR Ca²⁺ release solely, caused a substantial [Ca²⁺]_SR reduction (∼65%) (Fig. 5A) and a significant [Ca²⁺]_i elevation (∼1.7-fold in amplitude and ∼2.3-fold in duration) (Fig. 5B). The elevated Ca²⁺, as we expected, caused an EAD (Fig. 5B) similarly to that shown in Fig. 2A in which the mdO₂^·− burst affected both RyRs and SERCA. The mdO₂^·− burst-induced RyR activation itself, however, could not elicit multiple EADs regardless the timing of mdROS burst induction during the AP cycle (data not shown).

Fig. 5.

Model simulated effect of mdO₂^·− burst-mediated RyR activation or SERCA inhibition alone on SR Ca²⁺ content (A and C) and AP/[Ca²⁺]_i morphology (B and D). In simulations shown in A and B, mdO₂^·− burst had no effect on SERCA activity, and in C and D mdO₂^·− burst had no effect on RyR activity. The dash line is ΔΨ_m and its depolarization represents the acute mdO₂^·− burst.

The effects of mdO₂^·− burst on Ca²⁺ transients and AP were much less evident when it was modeled to modulate SERCA only. Particularly, the mdO₂^·− burst-inhibited SERCA Ca²⁺ uptake (J_up, from ∼0.144 to ∼0.097 mM/s) led to a small [Ca²⁺]_SR reduction (Fig. 5C) and a modest [Ca²⁺]_i increase (Fig. 5D). In this case, the mdO₂^·− burst could only cause APD prolongation (from 167 to 207 ms in Fig. 5D), implying that the mdO₂^·−-mediated SERCA inhibition had a smaller effect on AP than mdO₂^·−-mediated RyR activation in our model.

Effect of Shunt or Mitochondria-SR Distance on AP Morphology

In addition to mdO₂^·− induction timing, the pattern of mdROS-mediated abnormal AP was also affected by the concentration of [O₂^·−]_SR, which was determined by mdO₂^·− production rate (shunt) and mitochondria-SR distance (X) (or diffusion coefficient). Specifically, increasing shunt enhanced mdO₂^·− burst-induced Ca²⁺ elevation (Fig. 6A), promoting earlier occurrence of EAD, and potentially shifting APD prolongation to EAD and single EAD to multiple EADs (Fig. 6B). Reducing X had the similar effect: the shorter the distance, the stronger the effects of mdO₂^·− on Ca²⁺ and AP (Fig. 6, C and D). It is worth noting that the Ca²⁺ transient became “nonphysiologically” large when O₂^·− production was very high (e.g., shunt = 20%) or mitochondria-SR distance was very short (e.g., X = 25 nm) (Fig. 6, A and C). Under those extreme conditions, mdO₂^·− caused dramatic SR Ca²⁺ depletions compared with those under normal conditions (76–85 vs. ∼37%). In the case when mdO₂^·− burst was induced in phase 4 of the AP cycle, increasing shunt (from 10 to 30%) or reducing X (from 75 to 25 nm) significantly exaggerated Ca²⁺ elevation (Fig. 6E). The resultant large I_ti depolarized the membrane potential to the threshold for I_Na activation, leading to triggered activities (Fig. 6F). Increasing mdO₂^·− diffusion coefficient (D_O2^·−) similarly enhanced the effects of mdO₂^·− burst on Ca²⁺ transients and APs (Supporting Information Fig. S6).

Fig. 6.

Model simulated effect of mdO₂^·− production rate (shunt) or mitochondrion-SR distance (X) on the mdO₂^·−-mediated Ca²⁺ transient and AP alteration. A and B: increasing shunt caused higher Ca²⁺ elevation and converted normal AP to APD prolongation, single EAD and eventually multiple EADs. In these simulations, X = 75 nm. C and D: decreasing X had a similar effect on Ca²⁺ elevation and AP as increasing shunt. In these simulations, shunt = 0.10. E and F: effects of shunt and/or X on the phase 4 mdO₂^·− burst-induced Ca²⁺ elevation and AP abnormality. Increasing shunt from 10 to 30% or reducing X from 75 to 25 nm caused extraordinary increase of [Ca²⁺]_i and translated DAD to triggered activity. The dash gray lines represent ΔΨ_m and its depolarization represents the acute mdO₂^·− burst.

Effect of Blocking Ca²⁺ Handling Channel on mdROS-Induced EADs

Finally, we examined the effect of blocking individual Ca²⁺ handling channels on the mdROS-induced AP abnormality. Under normal mdO₂^·− production conditions, blocking I_NaCa had minimal effects on both AP and [Ca²⁺]_i (Fig. 7, Ai and Aii). In the presence of pathological mdO₂^·− bursts, the blockade of I_NaCa (Fig. 7Aiii) inhibited I_ti in spite of enhanced I_nsCa activation (Fig. 7Aiv), which suppressed I_CaL reactivation (Fig. 7Av). The lack of the subsequent CICR lowered Ca²⁺ elevation (Fig. 7Aii) and abolished the EAD (Fig. 7Ai). However, since the Ca²⁺ extrusion via I_NaCa was blocked, the duration of Ca²⁺ transient was prolonged (Fig. 7Aii).

Fig. 7.

Model simulated effects of blocking I_NaCa (A), I_nsCa (B), or I_CaL (C) on AP (i), Ca²⁺ transient (ii), I_NaCa (iii), I_nsCa (iv), and I_CaL (v) under control and mdO₂^·− burst conditions. In the legend, I_x represents I_NaCa (Ai-Av), I_nsCa (Bi-Bv), or I_CaL (Ci-Cv), respectively. The dash gray lines represent ΔΨ_m and its depolarization represents the acute mdO₂^·− burst.

Blocking I_nsCa did not significantly affect AP (Fig. 7Bi) and Ca²⁺ transient (Fig. 7Bii) no matter whether there was a pathological mdO₂^·− burst. Particularly, blocking I_nsCa slightly shortened the APD (Fig. 7Bi) and shifted the second Ca²⁺ transient to the left by ∼10 ms (Fig. 7Bii). However, blocking I_nsCa slightly reduced I_ti (Fig. 7Biv), shifted the I_CaL reactivation to a more polarized AP, and therefore increased the amplitude of the reactivated I_CaL (Fig. 7Bv).

Experimental studies have demonstrated that blocking LCC eliminates the oxidative stress-induced EADs and causes remarkable APD prolongation (59). Our simulations showed the similar results: the mdO₂^·−-induced EAD was suppressed but APD remained prolonged when I_CaL was blocked (Fig. 7Ci, dashed green curve). The prolonged APD was mainly caused by the mdO₂^·−-induced RyR activation and SERCA inhibition, which resulted in a relatively small but prolonged [Ca²⁺]_i elevation (Fig. 7Cii). However, blockade of I_CaL (Fig. 7Cv) hindered the subsequent CICR and prevented the further Ca²⁺ elevation. Consequently, I_NaCa and I_nsCa enhancements were suppressed (Fig. 7, Ciii and Civ) and the I_ti was too small to elicit an EAD. The outcomes of I_CaL blockade in the absence of mdROS bursts were shortened APD and abolished AP plateau (Fig. 7Ci) as well as diminished Ca²⁺ transient (Fig. 7Cii), which were consistent with previous studies (16, 23).

DISCUSSION

In this study, a novel multiscale guinea pig cardiomyocyte model was developed, which incorporates mitochondrial energetics, excitation-contraction coupling, local Ca²⁺ control, and RIRR, as well as mdROS-mediated modulations of RyR and SERCA activities. This model, for the first time, provides a computational framework to quantitatively examine how mdROS may influence cardiomyocyte Ca²⁺ regulation and electrophysiological behaviors under oxidative stress. Our main findings are 1) mdO₂^·−-induced AP abnormality involves alterations of multiple Ca²⁺ handling channels in a coordinated fashion, including RyRs, SERCA, I_nsCa, I_NaCa, and I_CaL; 2) the variance of mdO₂^·− burst-induced AP abnormality largely depends on the time interval between its induction and AP firing; and 3) the mdO₂^·−-induced AP abnormality is also influenced by mdO₂^·− dosage and/or mitochondria-SR distance.

It has been shown that the excessive O₂^·− produced by a stressed mitochondrion can diffuse to its immediate neighbors and activate their inner membrane anion channels, triggering RIRR in a self-amplifying mode (70). The present work suggests that if mdO₂^·− could travel to the proximal SR, they may activate the redox-sensitive RyRs, triggering ROS-induced Ca²⁺ release (RICR). This RICR alone, or together with inhibition of SERCA-mediated SR Ca²⁺ uptake, evoked erratic APs similar to those observed in cardiomyocytes exposed to H₂O₂ perfusion (59). Importantly, we showed that the mdO₂^·−-mediated SERCA inhibition had smaller effects on Ca²⁺ cycling and AP generation than the mdO₂^·−-mediated RyR activation, implying that enhanced RyR Ca²⁺ release might play a major role in the mdROS-induced cardiac arrhythmias. It is worth noting that in our simulations the inhibitory effect of mdO₂^·− on SERCA is transient. Sustained SERCA inhibition, such as that induced by continuous H₂O₂ perfusion, would have a much more prominent effect on Ca²⁺ handling. Intriguingly, while the proarrhythmic effect of mdO₂^·− burst-induced SERCA inhibition alone was minor, it significantly exacerbated the effect of RyR activation on AP. Particularly, concurrent SERCA inhibition shifted the RyR activation-induced APD prolongation to a single EAD and the single EAD to multiple EADs. These model simulations endorse the antiarrhythmic effect of increasing SERCA2a gene expression in heart failure treatment (18, 20).

Another important finding is that mdO₂^·− burst can elicit various patterns of AP abnormality such as APD prolongation, single EAD, and multiple EADs and DAD, similar to those observed in experimental studies (59). Our analysis suggested that mdO₂^·− bursts elicit erratic APs by triggering the following events in a coordinated way: 1) the mdO₂^·− burst activates RyR Ca²⁺ release and inhibits SERCA Ca²⁺ uptake, resulting in a [Ca²⁺]_i increase; 2) Ca²⁺ accumulation enhances I_NaCa in the forward mode and activates I_nsCa, evoking I_ti; 3) I_ti reduces AP repolarization reserve, reactivating I_CaL that triggers CICR and causes further [Ca²⁺]_i elevation; and 4) the resulting large [Ca²⁺]_i increment augments I_NaCa and I_nsCa, eliciting larger I_ti, which may repetitively reactivate I_CaL. Depending on the timing (t_AP-mdROS) of mdROS bursting, mdROS dosage, and mitochondria-SR distance, one or more of these events would not be activated, or be activated at different levels, resulting in various Ca²⁺ transient and AP morphologies (Fig. 4F). This paradigm is strengthened by the Ca²⁺ current inhibition studies, which showed that blocking I_NCX or I_CaL significantly altered the pattern of mdO₂^·− burst-induced AP abnormality. These finding are highly significant, as under pathological conditions such as ischemia-reperfusion mitochondrial depolarization and associated ROS bursts may occur asynchronously in various cells, resulting in regional electrical heterogeneity that increases the propensity for arrhythmogenesis.

Some studies suggested that direct ROS activation of I_CaL is a primary cause of oxidative stress-induced EADs (59), while others argued that the effects of ROS on I_CaL are controversial (9, 30, 50, 55). Notably, we showed that I_CaL were indirectly activated during mdROS bursts such as in the cases of single and multiple EADs (Fig. 4, Eiii and Eiv). Further analyses revealed that whether or not I_CaL can be reactivated was largely determined by two factors: 1) the amplitude of mdO₂^·− burst-induced SR Ca²⁺ release, which must be large enough to enhance I_NaCa and I_nsCa so that the resultant I_ti can reverse AP repolarization, and 2) the membrane potential when the reversal occurred, which needs to be in the range where I_CaL can be activated. The latter factor also determined the amplitude of the reactivated I_CaL, which was in agreement with previous experimental studies (59). Our simulations also suggested that when mdROS amplitude and duration are sufficiently large, I_CaL can be reactivated repeatedly, producing multiple EADs. These types of substantial Ca²⁺ transients (and EADs) may be seen in real life in more extreme situations such as apoptotic conditions, when multiple aspects of the cellular homeostasis are changed significantly and irreversibly.

Transient inward currents (I_ti) have been known to trigger EADs or DADs, eliciting Ca²⁺ overload-mediated arrhythmias. However, the contribution of I_nsCa to I_ti is controversial and appears to vary with different cell types (26, 28, 31, 32, 34, 47). Ca²⁺-sensitive nonspecific cationic channels (nsCa) belong to the “transient receptor potential” protein family and are expressed in both excitable and nonexcitable mammalian cells (26). In human atrial myocytes I_nsCa was shown to contribute to the genesis of arrhythmias during Ca²⁺ overload (34). However, the situation in ventricular myocytes is more controversial. Several groups have shown that I_nsCa could be activated by extracellular oxidative stress in guinea pig ventricular myocytes (31, 32); but in rabbit ventricular myocytes, Pogwizd et al. (48) showed that the contribution of I_nsCa to I_ti was insignificant in both control and failing cells. This may be due to the low expression of those cationic channels in ventricular myocardium (34). In our simulations, while blocking I_NaCa completely suppressed the mdROS burst-induced EAD (Fig. 7Ai), the blockade of I_nsCa had only minor effects on the morphologies of AP and Ca²⁺ transients. The results suggested that the contribution of I_nsCa to the mdROS burst-induced AP abnormality was smaller than that of I_NaCa, which is consistent with the observations of Pogwizd et al. (48). The differences in the contributions of these currents to I_ti under stress conditions are also in agreement with their roles in shaping AP under physiological conditions. One limitation of our study is that I_nsCa might be activated directly by oxidative stress in guinea pig ventricular myocytes (31, 32) but this effect is not considered in the present model. Nevertheless, our simulations underscore the importance of targeting the appropriate I_ti component in the treatment of oxidative stress-mediated cardiac arrhythmias.

It is well appreciated that mitochondrial dysfunction inhibits cell contraction. However, our simulations showed that the force of contraction was augmented during mitochondrial depolarization and ROS burst (Supporting Information Fig. S7). This paradox could be attributed to several factors: 1) Kohasi et al. (33) showed that enhanced ROS production could directly inhibit cardiomyocyte contraction; however, this effect is not included in the present model; and 2) it is known that ATP depletion directly inhibits cell contraction; however, the reduction of ATP during mitochondrial depolarization is very small in our simulations (∼3%) (Supporting Information Fig. S7). Under more stressed conditions such as higher pacing frequency or sustained mitochondrial dysfunction, the concentration of ATP may decline substantially, inhibiting SERCA and contraction (67).

Model Limitations and Future Directions

One of the major limitations is that the diffusion of mdROS in MSM cannot be validated by experimental study due to the lack of methods to measure ROS in such small domains. In addition, the present model does not consider the effects of mitochondrial-derived H₂O₂ or other oxidizing species. Our simplification is based on the experimental observations that the mdO₂^·− scavenger significantly suppressed enhanced SR Ca²⁺ release during RIRR. However, these results cannot completely exclude the effect of H₂O₂ as O₂^·− can be readily dismutated to generate H₂O₂. Whether it is O₂^·−, H₂O₂, or both that are responsible for the enhanced Ca²⁺ release deserves further investigations. Compared with O₂^·−, H₂O₂ is more stable and can diffuse further, thus affecting ion channels not only on the proximal SR membrane but also on the sarcolemmal membrane. In this context, various studies have shown that H₂O₂ can directly modulate I_CaL, although the effect is controversial (or dependent on H₂O₂ concentration) (54–57). Furthermore, ROS can affect I_Na either directly (38) or indirectly by activating CaMKII (22, 53, 54). The ROS-mediated CaMKII activation can also enhance SERCA activity via phospholamban phosphorylation (40), counteracting the effect of ROS-induced SERCA inhibition. Other redox-sensitive ion channels include K⁺ channels (K_ir and K_v) and the Na⁺/Ca²⁺ exchanger (NCX) (4, 15, 35, 43, 55, 68). It is expected that adding the effects of H₂O₂ on these sarcolemmal ion channels would exacerbate the influence of mdROS and lead to more complex AP morphologies as observed experimentally (59). Nonetheless, the lack of mdROS-induced I_CaL, CaMKII, and/or I_Na modulations would not impact our main conclusions about the mechanisms underlying mdROS-induced Ca²⁺ dysregulation and abnormal APs. Indeed, the lack of mdROS-mediated membrane channel remodeling allowed us to focus on the crucial role of mdROS on SR Ca²⁺ handling and genesis of EADs and DADs under oxidative stress conditions. Nevertheless, the effects of mdROS on these proteins should be incorporated into the model when more experimental data become available.

Moreover, we assumed that all mitochondria in the cardiomyocyte depolarize and release ROS simultaneously so that we can focus on the ionic mechanisms underlying mdROS burst-induced alterations in Ca²⁺ cycling and cellular electrophysiology. In a real cardiomyocyte, the rate of ROS increase might be slower due to the heterogeneity of mitochondrial energetic state and network ultrastructure. The chronic, sustained mdROS increase may deplete SR Ca²⁺ loading if it spans over several beats or inhibits RyRs and I_CaL activities if the dose is too high. Apparently, systematic experimental studies will be needed to gain a more complete understanding of how mdROS may exactly influence intracellular ion handling under certain pathological conditions. However, again these limitations are not expected to alter the mechanisms that underlie the mdROS-induced abnormal Ca²⁺ cycling and AP generation unraveled here.

Finally, a recent study showed that carbonyl cyanide p-trifluoromethoxyphenylhydrazone (FCCP)-induced mitochondrial depolarization increased the frequency and amplitude of Ca²⁺ waves and induced triggered activities, probably via the mitochondrial Ca²⁺ efflux mediated by the mitochondrial permeability pore (mPTP) (64). Since the present model did not incorporate mPTP, it cannot be used to examine this mechanism and study the interaction between the ROS-modulated and mPTP Ca²⁺ efflux-mediated cytosolic Ca²⁺ dynamics.

In summary, the developed multiscale model is capable of simulating the acute effect of mdROS on Ca²⁺ cycling, providing a novel tool to examine how alterations in mitochondrial energetics can impact cardiomyocyte electrophysiology and electrical activity. The results highlight the role of mdROS in Ca²⁺ overload-mediated cardiac arrhythmogenesis and abnormal automaticity. They also underscore the importance of considering mitochondrial targets in designing new antiarrhythmic therapies in the context of sudden cardiac death.

GRANTS

This work was supported by National Heart, Lung, and Blood Institute Grants R00-HL-095648 (to L. Zhou), R01-HL-101235, and R37-HL-054598 (to B. O'Rourke) and American Heart Association Grant GRNT8000059 (to S. M. Pogwizd).

ENDNOTE

At the request of the authors, readers are herein alerted to the fact that additional materials related to this manuscript may be found at the institutional website of one of the authors, which at the time of publication they indicate is http://www.uab.edu/medicine/cscb/images/paper_pdf/Li_et_al_Supporting_Information.pdf. These materials are not a part of this manuscript and have not undergone peer review by the American Physiological Society (APS). APS and the journal editors take no responsibility for these materials, for the website address, or for any links to or from it.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the author(s).

AUTHOR CONTRIBUTIONS

Author contributions: Q.L., D.S., and L.Z. performed experiments; Q.L., D.S., and L.Z. analyzed data; Q.L., B.O., S.M.P., and L.Z. interpreted results of experiments; Q.L. and L.Z. prepared figures; Q.L., D.S., and L.Z. drafted manuscript; B.O., S.M.P., and L.Z. edited and revised manuscript; L.Z. conception and design of research; L.Z. approved final version of manuscript.