Instabilities of the resting state in a mathematical model of calcium handling in cardiac myocytes

Abstract

We analyze a recently published model of calcium handling in cardiac myocytes in order to find conditions for the presence of instabilities in the resting state of the model. Such instabilities can create calcium waves which in turn may be able to initiate cardiac arrhythmias. The model was developed by Swietach, Spitzer and Vaughan-Jones [1] in order to study the effect, on calcium waves, of varying ryanodine receptor (RyR)-permeability, sarco/endoplasmic reticulum calcium ATPase (SERCA) and calcium diffusion. We study the model using the extracellular calcium concentration c_e and the maximal velocity of the SERCA-pump v_SERCA as control parameters. In the (c_e, v_SERCA)-domain we derive an explicit function v^* = v^*(c_e), and we claim that any resting state based on parameters that lie above the curve (i.e. any pair (c_e, v_SERCA) such that with v_SERCA > v^*(c_e)) is unstable in the sense that small perturbations will grow and can eventually turn into a calcium wave. And conversely; any pair (c_e, v_SERCA) below the curve is stable in the sense that small perturbations to the resting state will decay to rest. This claim is supported by analyzing the stability of the system in terms of computing the eigenmodes of the linearized model. Furthermore, the claim is supported by direct simulations based on the non-linear model.

Since the curve separating stable from unstable states is given as an explicit function, we can show how stability depends on other parameters of the model.

1 Introduction

The contraction of cardiac myocytes is controlled by a process termed excitation-contraction coupling [2]. An action potential causes influx of calcium, through L-type calcium channels (LCCs), and this ”trigger” calcium elicits further calcium release from the sarcoplasmic reticulum (SR) via the ryanodine receptors (RyR). Calcium then binds to troponin C, which initiates contraction. Finally, calcium is removed from cytosol to facilitate relaxation. Most of the calcium is reuptaken into the SR by the sarco-endoplasmic reticulum calcium ATPase (SERCA) pump. However a fraction of all calcium is taken out of the cell through the sodium-calcium exchanger (NCX). The NCX is a protein that exchanges 1 calcium with 3 sodium ions, creating a net flow of positive charge into the cell.

Under experimental conditions where the extracellular calcium level is increased, pathological calcium waves can be induced [3, 4]. The elevated extracellular calcium is transported across the cell membrane and eventually pumped into SR, via the SERCA pump. Inside the SR, calcium sensitizes the RyRs, which, together with the elevated cytosolic calcium, will eventually trigger calcium release from SR [5, 6]. Calcium overload can also be induced by stimulation of the β-adrenergic pathway [7], which would mimic more natural conditions for the myocyte. Whatever the cause of calcium overload, if the size of this spontaneous release is large enough, a propagating wave will be generated [8]. An SR calcium threshold for waves has been identified and named store-overload-induced calcium release [9]. The released calcium is extracted through the NCX which generates a net flow of positive charge into the cell. If the magnitude of the release is suffciently large, it can trigger an action potential; a so called ectopic beat. Such beats are believed to play a decisive role in the initiation of cardiac arrhythmias [10, 11].

It has been shown that a critical balance between sensitized RyRs, and altered SERCA activity is required to trigger calcium waves [12, 7]. Computational models have been used to shed light on the exact balance between these two parameters for the generation of calcium waves [13, 14, 15]. A model of calcium wave generation and propagation has recently been suggested by Swietach, Spitzer and Vaughan-Jones [1] and used by Stokke et al [16] to probe the generation of calcium waves during pathological conditions. The model is a deterministic system of partial differential equations which is simple enough to be analyzed but also complex enough to represent the processes underlying wave generation. Importantly, features necessary for a complete view of myocyte electrophysiology, such as the generation and shape of an action potential, are not included. However, we are studying generation of calcium waves, which are important for delayed afterdepolarisations, and such waves are triggered during the resting phase of the action potential. Our aim is to understand the model from a mathematical point of view. More specifically, we want to assess the stability of the resting state as a function of model parameters that are known to be important for the generation of calcium waves. If the resting state is unstable, a propagating calcium wave can be generated. Our prime control variables are the extracellular calcium concentration c_e and the strength of the SERCA pump v_SERCA. For standard parameters we theoretically derive a function v^* in the (c_e, v_SERCA) space dividing stable and unstable parameters. This curve explains how the stability of the cell depends on the main control parameters. Furthermore, we demonstrate how this curve is altered as a function of the RyR sensitivity parameter K_α.

The model of Swietach, Spitzer and Vaughan-Jones [1] is a system of reaction-diffusion equations. As shown in e.g. [17, 18, 19] the resting state of such systems can be analyzed by eigenmode analysis. We compute a resting state of the system and derive a linearized model around that state. The linearized model is discretized using a finite difference method leading to a linear system of ordinary differential equations. It is well known, see e.g. Chicone [20], that the stability of resting states for a system of ordinary differential equations can be analyzed by invoking the real parts of the eigenvalues of the system matrix. By using this method for the present model, we show that the curve v^* = v^*(c_e) separates stable from unstable parameters. Since this method is based on a linearization, we present computations based on the full non-linear model showing that the curve we have derived separates stable from unstable states also in the non-linear case. We emphasize, however, that our findings are entirely based on the mathematical model and we do not know if similar properties can be seen in experiments.

The present paper is organized as follows: The mathematical model of [1] and [16] is presented in the next section where we also derive an analytical expression for the curve v^* = v^*(c_e) separating stable and unstable parameters. In Section 3 we present numerical computations based on linear analysis of the system of reaction-diffusion equations, and simulations based on stochastic initial conditions. In Section 4, we present a generalized model for which a similar analysis can be performed, and a conclusion is given in Section 5.

2 The mathematical model

The mathematical model under consideration was developed by Swietach, Spitzer and Vaughan-Jones [1], and applied in a slightly modified version by Stokke et al [16]. Here we use the form of the model presented in the online supplement of [16]. The model consists of a one dimensional system of partial differential equations of the following form:

$$\frac{\partial c_{\mathit{cyt}}}{\partial t}=D_{\mathit{cyt}}\frac{{\partial }^2{c}_{\mathit{cyt}}}{\partial x^2}-J_{\mathit{cytbuf}}+\frac{1}{V_{\mathit{cyt}}}(J_{sl}+J_{\mathit{RyR}}-J_{\mathit{SERCA}}),$$ (1)

$$\frac{\partial c_{SR}}{\partial t}=D_{SR}\frac{{\partial }^2{c}_{SR}}{\partial x^2}-J_{\mathit{SRbuf}}+\frac{1}{V_{SR}}(-J_{\mathit{RyR}}+J_{\mathit{SERCA}}),$$ (2)

$$\frac{\partial b_{\mathit{cyt}}}{\partial t}=D_{\mathit{buf}}\frac{{\partial }^2{b}_{\mathit{cyt}}}{\partial x^2}+J_{\mathit{cytbuf}},$$ (3)

$$\frac{\partial b_{SR}}{\partial t}=J_{\mathit{SRbuf}},$$ (4)

$$\frac{\partial p}{\partial t}=k_{on2}\frac{C_{\mathit{SRbuf}}-b_{SR}}{C_{\mathit{SRbuf}}-b_{\mathit{SRO}}}{c}_{\mathit{cyt}}(1-p)-k_{\mathit{off}2}p.$$ (5)

Here c_cyt and c_SR denote the calcium concentration in cytosol and SR; similarly b_cyt and b_SR denote the buffered calcium concentrations, and p denotes the occupancy of the inactivating gate of the RyR. The net sarcolemmal calcium influx is given by

$$J_{sl}=k_{sl}(c_e-c_{\mathit{cyt}})-v_{\mathit{NCX}}\frac{c_{\mathit{cyt}}}{c_{\mathit{cyt}}+K_{\mathit{NCX}}},$$ (6)

and the RyR release/leak is given by

$$J_{\mathit{RyR}}=\left(k_{\mathit{RyR}}max(0,\frac{c_{\mathit{cyt}}}{c_{\mathit{cyt}}+K_{\alpha }}-p)+k_{\mathit{leak}}\right)(c_{SR}-c_{\mathit{cyt}}).$$ (7)

In this system, equations (1–4) are on standard form for models of calcium dynamics, see e.g. [21, 22]. The dynamics of the RyR current are modeled in terms of a threshold; whenever the value of the fast gate, given by $\frac{c_{\mathit{cyt}}}{c_{\mathit{cyt}}+K_{\alpha }}$ is larger than a threshold, given by the slow inactivation gate, p, there is a current. The difference between the fast and the slow gate will then give the size of the current. The fast gate is not included as a state variable in the system of equations, as it is assumed to be in quasi steady state with cytosolic calcium. The slow gate acts as an inactivation gate. From (5) we see that this gate is activated by increased cytosolic calcium c_cyt together with the amount of unbound calcium buffer in SR, (C_SRbuf − b_SR). The former models inactivation of the RyR by cytosolic calcium and the latter models RyR activation of SR calcium and explains why a calcium overloaded cell is more prone to generate calcium waves [5, 6, 23].

The SERCA-pump brings calcium from cytosol into the SR and is modeled by

$$J_{\mathit{SRECA}}=v_{\mathit{SERCA}}\frac{c_{\mathit{cyt}}^2-{(c_{SR}/\delta )}^2}{c_{\mathit{cyt}}^2+{(c_{SR}/\delta )}^2+K_{\mathit{SERCA}}^2}.$$ (8)

Here K_SERCA is the dissociation constant for cytosolic calcium, δ is the ratio between the dissociation constants of calcium in SR and cytosol, and v_SERCA is the maximum velocity of the SERCA-pump. By increasing the activity of the SERCA pump, we increase the SR calcium load, as the increased SERCA activity has to be balanced during rest by an increase in SR calcium leak. The latter can only be achieved by an increase in SR calcium load.

Finally the transition rates from unbound to bound cytosolic and SR buffer are given by

$$J_{\mathit{cytbuf}}=k_{on}(c_{\mathit{cyt}}{C}_{\mathit{cytbuf}}-(c_{\mathit{cyt}}+K_{\mathit{cytbuf}})b_{\mathit{cyt}}),$$ (9)

$$J_{\mathit{SRbuf}}=k_{on}(c_{SR}{C}_{\mathit{SRbuf}}-(c_{SR}+K_{\mathit{SRbuf}})b_{SR}).$$ (10)

Here C_cytbuf and C_SRbuf refer to the total amount of calcium buffer in cytosol and SR respectively. The parameters of the model are taken from [16] and are listed in the Table 1.

Table 1.

The table gives the constants used in the mathematical model. The constants are taken from Stokke et al [16].

kon	200μM−1s−1	kon2	10μM−1s−1	koff2	3s−1
Dbu f	70μm2s−1	Vcyt	0.65	VSR	0.035
Dcyt	500μm2s−1	Ccytbuf	175 μM	Kcytbuf	2μM
DSR	20μm2s−1	CSRbuf	4400 μM	KSRbuf	630μM
kN CX	36μM	KSERCA	0.25μM	ksl	0.0066 s−1
vN CX	2400 μMs−1	Ka	0.35μM	kRyR	60s−1
bSR0	851.1586μM	δ	2000	kleak	0.8s−1

In our analysis the extracellular calcium concentration c_e, and the strength of the SERCA-pump v_SERCA will be used as control parameters; their default values taken from [16] are c_e=1500 μM and v_SERCA=650μMs⁻¹.

We start our analysis by considering the system given by (1–5) with the fluxes and constants described above. In Section 4, we will present a similar analysis for a larger class of models that can be written in the form (1–5).

2.1 Constant solutions at rest

A constant solution in space and time of the system above satisfies the following algebraic system of equations

$$0=-J_{\mathit{cytbuf}}+\frac{1}{V_{\mathit{cyt}}}(J_{sl}+J_{\mathit{RyR}}-J_{\mathit{SERCA}}),$$ (11)

$$0=-J_{\mathit{SRbuf}}+\frac{1}{V_{SR}}(-J_{\mathit{RyR}}+J_{\mathit{SERCA}}),$$ (12)

$$0=J_{\mathit{cytbuf}},$$ (13)

$$0=J_{\mathit{SRbuf}},$$ (14)

$$0=k_{on2}\frac{C_{\mathit{SRbuf}}-b_{SR}}{C_{\mathit{SRbuf}}-b_{SR0}}{c}_{\mathit{cyt}}(1-p)-k_{\mathit{off}2}p.$$ (15)

It follows from this that

$$J_{\mathit{SERCA}}=J_{\mathit{RyR}}$$ (16)

and

$$J_{sl}=0.$$ (17)

From (17) and (6) we get

$$k_{sl}{c}_{\mathit{cyt}}^2+(v_{\mathit{NCX}}+k_{sl}{K}_{\mathit{NCX}}-c_e{k}_{sl})c_{\mathit{cyt}}-c_e{k}_{sl}{K}_{\mathit{NCX}}=0.$$ (18)

By using the parameters values given in Table 1 for k_sl, K_{N CX} and v_{N CX}, and then graph c_cyt as a function of c_e, we note that c_cyt ≈ 10⁻⁴ c_e (see Figure 1). Based on this observation it is reasonable, in general, to seek an approximate relation of the form

Figure 1.

The accuracy of the linear approximation (24) is illustrated using three sets of parameters. Upper panel: The green graphs show the exact solution (solid line) of (18) and the linear approximation (dotted line) given by (24) for standard parameters. Similar results are given by the red graphs where the value of k_sl is multiplied by two, and in the blue graphs where the value of k_sl is divided by two. Center panel: Same as the upper panel, but the value of v_{N CX} is varied; standard (green), two times standard (red), and half of standard (blue). Lower panel: Similar to upper and center but the values of K_{N CX} is varied. We conclude that for all parameter sets, the approximation given by (24) is accurate.

$$c_{\mathit{cyt}}=\phi c_e$$ (19)

where φ ~ 10⁻⁴. We get the equation

$$k_{sl}{\phi }^2{c}_e^2+(v_{\mathit{NCX}}-k_{sl}{K}_{\mathit{NCX}}-c_e{k}_{sl})\phi c_e-c_e{k}_{sl}{K}_{\mathit{NCX}}=0,$$ (20)

and thus, by ignoring higher order terms, we get

$$\phi \approx \frac{k_{sl}{K}_{\mathit{NCX}}}{v_{\mathit{NCX}}+k_{sl}{K}_{\mathit{NCX}}-c_e{k}_{sl}}.$$ (21)

From the parameters given in Table 1, we have

$$\frac{k_{sl}{K}_{\mathit{NCX}}}{v_{\mathit{NCX}}+k_{sl}{K}_{\mathit{NCX}}-c_e{k}_{sl}}\approx \frac{k_{sl}{K}_{\mathit{NCX}}}{v_{\mathit{NCX}}+k_{sl}{K}_{\mathit{NCX}}}+2.722\times {10}^{-10}{c}_e+7.4847\times {10}^{-16}{c}_e^2$$ (22)

and thus we use the approximation

$$\phi =\frac{k_{sl}{K}_{\mathit{NCX}}}{v_{\mathit{NCX}}+k_{sl}{K}_{\mathit{NCX}}}$$ (23)

so

$$c_{\mathit{cyt}}(c_e)=\phi c_e=\frac{k_{sl}{K}_{\mathit{NCX}}}{v_{\mathit{NCX}}+k_{sl}{K}_{\mathit{NCX}}}{c}_e.$$ (24)

In Figure 1 we have graphed c_cyt as a function of c_e given as the solution of (18) and compared it to the linear approximation given by (24) for four different sets of the parameters (k_sl, K_{N CX}, v_{N CX}). We observe that the linear approximation given by (24) is accurate for all four parameter sets and it will therefore be used below.

In the system (11) – (15) we seek a solution that is constant in space and time, and we seek a solution at a point in the state-space where the RyR-release is about to activate. From (7) it follows that the transition from inactive to active RyR-release is characterized by the equation

$$\frac{c_{\mathit{cyt}}}{c_{\mathit{cyt}}+K_{\alpha }}-p=0,$$ (25)

and thus, from (24), we get

$$p(c_e)=\frac{\phi c_e}{\phi c_e+K_{\alpha }}$$ (26)

where φ is given by (23).

From (13) and (9) we have

$$c_{\mathit{cyt}}{C}_{\mathit{cytbuf}}-(c_{\mathit{cyt}}+K_{\mathit{cytbuf}})b_{\mathit{cyt}}=0$$ (27)

so by applying the approximation (24) we get

$$b_{\mathit{cyt}}(c_e)=\frac{C_{\mathit{cytbuf}}\phi c_e}{\phi c_e+K_{\mathit{cytbuf}}}$$ (28)

where again φ is given by (23). Furthermore, from (15), we get

$$\frac{k_{on2}}{k_{\mathit{off}2}}\frac{C_{\mathit{SRbuf}}-b_{SR}}{C_{\mathit{SRbuf}}-b_{SR0}}{c}_{\mathit{cyt}}=\frac{p}{1-p}$$ (29)

and since $p=\frac{c_{\mathit{cyt}}}{c_{\mathit{cyt}}+K_{\alpha }}$ (see (25)) we have

$$\frac{k_{on2}}{k_{\mathit{off}2}}\frac{C_{\mathit{SRbuf}}-b_{SR}}{C_{\mathit{SRbuf}}-b_{SR0}}=\frac{1}{K_a}$$ (30)

We define

$$b_{SR1}=C_{\mathit{SRbuf}}-\frac{k_{\mathit{off}2}}{k_{on2}{K}_a}(C_{\mathit{SRbuf}}-b_{SR0}).$$ (31)

and conclude that

$$b_{SR}=b_{SR1}$$ (32)

which is a constant independent of c_e.

From (14) and (10) we have

$$c_{SR}{C}_{\mathit{SRbuf}}-(c_{SR}+K_{\mathit{SRbuf}})b_{SR}=0$$ (33)

and then

$$c_{SR}=\frac{K_{\mathit{SRbuf}}{b}_{SR}}{C_{\mathit{SRbuf}}-b_{SR}}$$ (34)

Since b_SR = b_SR1, we have

$$c_{SR}=c_{SR1}$$ (35)

where

$$c_{SR1}=\frac{K_{\mathit{SRbuf}}{b}_{SR1}}{C_{\mathit{SRbuf}}-b_{SR1}}$$ (36)

which is also independent of c_e.

It follows from (16) that J_RyR = J_SERCA and because of (25) it follows from (7) and (8) that

$$v_{\mathit{SERCA}}=\frac{k_{\mathit{leak}}(c_{SR}-c_{\mathit{cyt}})\left(c_{\mathit{cyt}}^2+{(c_{SR}/\delta )}^2+K_{\mathit{SERCA}}^2\right)}{c_{\mathit{cyt}}^2-{(c_{SR}/\delta )}^2}.$$ (37)

We have found the critical strength of the SERCA-pump as a function of the extracellular calcium concentration c_e at the point in state-space where the RyR-release activates. From the approximation (24) and the fact that c_SR is a constant given by (36) we get

$$v^{\ast }(c_e)=\frac{k_{\mathit{leak}}(c_{SR1}-\phi c_e)\left({(\phi c_e)}^2+{(c_{SR1}/\delta )}^2+K_{\mathit{SERCA}}^2\right)}{{(\phi c_e)}^2-{(c_{SR1}/\delta )}^2}$$ (38)

where

$$\phi =\frac{k_{sl}{K}_{\mathit{NCX}}}{v_{\mathit{NCX}}+k_{sl}{K}_{\mathit{NCX}}}$$ (39)

In the (c_e, v_SERCA)– domain, we claim that any resting state based on parameters below this graph given by v_SERCA = v*(c_e) is stable and any resting state based on parameters above the graph is unstable. The validity of this claim will be addressed in computations below.

3 Numerical computations

Above, we computed a curve v^* in the (c_e, v_SERCA)-domain and we claim that this curve divides parameters leading to stable and unstable resting states. The main purpose of this section is to provide computational evidence that this claim holds. We do that by assessing the stability of the resting state by invoking the real parts of the eigenvalues of an associated linear problem. This technique has recently been applied to assess the stability of resting state in cardiac tissue, see [17, 18, 19]. We start this section by repeating the basic steps of linear analysis of a system of reaction diffusion equations. In the rest of the section we present numerical experiments discussing the validity of our claim regarding the stability of the problem. In the numerical computations, we start by using linear theory to investigate the stability as a function of extracellular calcium concentration and the strength of the SERCA-pump. Next we consider the influence of molecular diffusion, and the magnitude of K_α representing the dissociation constant of the fast activating RyR-site. We study how the magnitude of the principal eigenvalue depends on the RyR channel flux constant k_RyR, and how a change in sign of an eigenvalue alters the behavior of the solution. Finally, we discuss the accuracy of linear analysis.

3.1 Linear analysis revisited

The system (1) – (5) can be written in the form

$$z_t={dz}_{xx}+f(z)$$ (40)

where z is a vector containing the variables (c_cyt, c_SR, b_cyt, b_SR, p), the function f carries the fluxes and functions defined in (6) – (10) and d denotes a diagonal matrix of the form d =diag(D_cyt, D_SR, D_buf, 0, 0). We discretise the system by introducing the nodes x_j = jΔx, where Δx = L/n. Here L denotes the length of the cell (set to 100 μm in our computations) and n denotes the number of computational nodes (set to 50 in our computations). Let z_j (t) denote the approximate solution at x = x_j defined as the solution of the following system of ordinary differential equations,

$$z_j^{\prime }(t)=d\frac{z_{j-1}-2z_j+z_{j+1}}{{(\mathrm{\Delta }x)}^2}+f(z_j)$$ (41)

for j = 0, …, n. Suppose a stationary solution is given by z⁰, i.e. f (z⁰) = 0. Then we can introduce y = z − z⁰, and find that, up to linear terms, y is governed by a system of the form

$$y^{\prime }=Ay.$$ (42)

It is well known from the theory of ordinary differential equations (see e.g. Chicone [20]) that the stability of the resting state z⁰ is determined by the real parts of the eigenvalues of the system matrix A. More specifically, the system is termed unstable if the largest real part is greater than zero. Observe that if λ is a real-valued eigenvalue of A and r is the associated eigenvector, then e^λtr solves the system (42) when the initial condition is given by y(0) = r. Thus an initial perturbation in terms of an eigenvector will grow or decay at exponential rate depending on the sign of the associated eigenvalue. Due to the inherit rhythm of cardiomyocytes, the magnitude of the eigenvalues are of great importance; all the cells in the tissue are reset after a few hundred milliseconds and thus a perturbation that grows slowly, will not be able to create a calcium wave. In the computations presented below we will see that in the present model the eigenvalues of the unstable cases are rather big and can therefore be able to initiate waves.

3.2 Stability as a function of extracellular calcium and the strength of the SERCA-pump

To assess the susceptibility of the model to generating calcium waves we examine how the level of extracellular calcium c_e and the strength of the SERCA-pump v_SERCA impact the stability of the associated resting state. For the parameters given in Table 1, we show the stability as a function of c_e and v_SERCA in Figure 2. Given a point (c_e, v_SERCA), we compute the resting state of the system (1) – (5) and the largest real part of the eigenvalues of the associated linearized system (42). In the figure, the light grey area (upper part) denotes an area where the principal eigenvalue has a positive real part. Thus for sufficiently large values of c_e and v_SERCA the resting state is unstable. Correspondingly, the dark grey area represents a stable area in the (c_e, v_SERCA)–space. These areas are found by using the linearization described above. Furthermore, the transition from negative to positive values are found by using a root-finding function in Matlab and are marked by ×. The solid line is given by the function (38) derived above and we observe that the solid line coincides with the transition from the stable to the unstable region, and thus provides support for the claim that the curve defined by (38) separates stability from instability.

Figure 2.

The solid line represents the estimated transition from stable to unstable equilibrium point given by (38). The dark grey area represents a region where the real part of all eigenvalues of the linearized system is negative, and in the light grey area, at least one eigenvalue has a positive real part. Furthermore, the transition from negative to positive values are found by using a root-finding function in Matlab and are marked by ×. We observe that the curve computed by (38) coincides completely with the curve separating negative and positive principal eigenvalues.

The curve in Figure 2 illustrates that increasing either c_e or v_SERCA (or both) will make the model less stable. In the unstable region, a small perturbation will grow and may turn into a full calcium wave. By increasing c_e we increase c_cyt bringing the RyR current closer to the release threshold, see (7). Although increasing v_SERCA does not alter c_cyt, it lowers the threshold variable, p, by increasing steady state SR calcium load. The latter is indirectly accomplished by lowering the amount of unbound calcium buffer in SR, which regulates p.

3.3 Stability as a function of diffusion

In Figure 3 we study how the stability of a resting state depends on the diffusion coefficients of the system (1)–(5) given by the three parameters (D_cyt, D_SR, D_buf). More specifically, we have plotted the third largest real part of the eigenvalues (the two largest seem to be independent of diffusion) for three values of molecular diffusion; one standard given in Table 1 and denoted D in the figure, and two cases where we either multiply or divide D = (D_cyt, D_SR, D_buf) by 10.0. We observe that the stable region seems to be rather insensitive to the magnitude of molecular diffusion. We graph the real part of the principal eigenvalue as a function of c_e and we observe that as c_e passes the critical point from the stable region to the unstable region, the eigenvalue changes sign and the magnitude increases dramatically. This indicates that for unstable parameters, even a small perturbation will diverge rapidly from the resting state. The sudden change in largest eigenvalue is a result of the threshold for the RyR current being passed.

Figure 3.

Left panel: The figure shows the magnitude of the largest and third largest real part of the eigenvalues, and how they depend on the diffusion coefficients. The largest and second largest real part of eigenvalues are independent of diffusion, but the third largest depends on diffusion; the magnitude is reduced as diffusion is increased. Right panel: Same data as on the left, zoomed in around the critical interval.

Although the sign of the real part of the eigenvalue seems to be almost independent of the size of the molecular diffusion, the magnitude of the real parts of the eigenvalues may still depend on it. This is also illustrated in Figure 3 were we observe that the magnitude of the third largest eigenvalue in the unstable region decreases as the magnitude of molecular diffusion is increased. This is because a larger diffusion will lower the regenerative effect of the RyR current as calcium is faster transported away from the local release site. So more diffusion stabilizes the stationary solutions of (1) – (5), which is generally expected to be the case, see e.g. [17]. This effect is highlighted in the right panel of Figure 3. In summary we have observed that the magnitude of molecular diffusion can affect the magnitude of the real part of the eigenvalue, but it does not seem to affect the sign of the real part of the principal eigenvalue.

3.4 Stability as a function of the dissociation constant of the fast, activating RyR-site; K_α

In Figure 4 we study how the regions of stability/instability depend on the dissociation constant of the fast, activating RyR-site called K_α; see (7). This constant represents the calcium affinity of the fast activating gate. The default value is given by K_α = 0.35 (green graph), and in the figure we also consider the case of K_α = 0.9·0.35 (blue graph) and K_α = 1.1·0.35 (red graph). We first note that the results obtained by the theoretical estimate (38) again coincide with the results obtained through the eigenvalue analysis described above. Second, we observe that the stable region is sensitive to changes in K_α; the size of the stable region is significantly increased by increasing the value of K_α. This is because the activating gate becomes less sensitive to calcium and a larger c_cyt and hence larger c_e is needed to trigger the RyR current.

Figure 4.

Left panel: The figure shows that the stable region depends heavily on the dissociation constant of the fast, activating RyR-site K_α; the stable region is increased when the value of K_α is increased. The figure also illustrates that the theoretical estimate (solid line) provides a good approximation to the transition from stable to unstable states; × illustrates points in the (c_e, v_SERCA)-domain where the largest real part of the eigenvalues goes from negative to positive. Right panel: The figure shows the largest real part of the eigenvalues as a function of the extracellular Calcium concentration c_e. Standard parameters are used (see Table 1), and v_SERCA = 650μMs⁻¹. Observe that there is a large jump in the magnitude of the eigenvalue as it goes from being mildly negative to strongly positive.

In the right panel of Figure 4 we observe, again, a significant jump in the magnitude of the principal eigenvalue as the extracellular concentration is changed from the stable to the unstable region. Interestingly, the largest eigenvalue gets larger as K_α increases. This is because a higher K_α will cause an elevated SR calcium content, which results in a larger gradient for the RyR current once the release starts.

3.5 The magnitude of the principal eigenvalue as a function of the RyR channel flux constant K_RyR

The RyR current flux constant k_RyR determines the size of the release current once it is triggered, see (7). A larger release will, through positive feedback via an elevated c_cyt, act to re-enforce itself. Hence we would expect the principal eigenvalue of an unstable system to increase with k_RyR. In Figure 5 we illustrate that this turns out to be the case.

Figure 5.

The figure illustrates how the largest real part of the eigenvalues depends on the RyR channel flux constant k_RyR; it is strongly increasing. The × marks the default value of k_RyR. We have used the standard parameters given in Table 1, and we have put v_SERCA = 650μMs⁻¹.

3.6 Stability as a function of the strength of the Na/Ca-exchanger

In the model above, the Na/Ca-exchanger (NCX) flux is given by

$$J_{\mathit{NCX}}=v_{\mathit{NCX}}\frac{c_{\mathit{cyt}}}{c_{\mathit{cyt}}+K_{\mathit{NCX}}},$$ (43)

see latter term of J_sl in Equation (6). The strength of the flux is governed by the two parameters v_{N CX} and K_{N CX}, and it is of interest to understand how the stability of the problem depends on these parameters. The stability curve (38) is given by

$$v^{\ast }(c_e;K_{\mathit{NCX}},v_{\mathit{NCX}})=\frac{k_{\mathit{leak}}(c_{SR1}-\phi c_e)\left({(\phi c_e)}^2+{(c_{SR1}/\delta )}^2+K_{\mathit{SERCA}}^2\right)}{{(\phi c_e)}^2-{(c_{SR1}/\delta )}^2}$$ (44)

where we recall that

$$\phi =\frac{k_{sl}{K}_{\mathit{NCX}}/v_{\mathit{NCX}}}{1+k_{sl}{K}_{\mathit{NCX}}/v_{\mathit{NCX}}}.$$ (45)

From the form of φ, we immediately see that the stability curve is a function of the ratio K_{N CX}/v_{N CX}. The changes in the stable region as a function of K_{N CX} and v_{N CX} are analyzed using numerical computations in Figure 6. We consider four cases; A: (K_{N CX}, v_{N CX}), B: (1.5K_{N CX}, v_{N CX}), C: (K_{N CX}, 1.5v_{N CX}), D: (1.5K_{N CX}, 1.5v_{N CX}) where the default case is given by (K_{N CX}, v_{N CX}) = (36, 2400). The numerical computations show that the stability regions are identical for the two cases A and D. This is what we expected since the value of φ is equal for these cases. Furthermore, $v_{\mathit{SERCA}}^{\ast }$ is smaller than default for case B and larger than default for case C, so the stable region is smaller than default in case B and larger than default in case C. Again, we note that the theoretical estimate of the boundary between stable and unstable regions coincides with the curve where the largest real part of the eigenvalues changes sign.

Figure 6.

The figure shows how changes in the parameters K_{N CX} and v_{N CX} (see Equation (43)) governing the strength of the NCX-current affect the stable region. We observe that case A (default parameters) and D (both K_{N CX} and v_{N CX} multiplied by 1.5) are identical. If we increase K_{N CX} by a factor of 1.5 and keep v_{N CX} at the default value (case B), we observe that the stable region becomes smaller, and thus the stability decreases when K_{N CX} is increased. On the other hand, if we increase v_{N CX} by a factor of 1.5 and keep K_{N CX} constant (case C), we observe that the stable region becomes larger, and hence the stability increases when v_{N CX} is increased. Similarly to Figure 2, the solid lines show the theoretical estimate, while transitions from negative to positive eigenvalues are marked with ×.

3.7 An illustration of the effect of change of sign of an eigenvalue

In Figure 7 we consider two possible perturbations of the resting state. In order to do the comparison, we introduce the L²–norm given by

Figure 7.

The figure illustrates how a perturbation develops as a function of time measured in the L²–norm. In the computations we have used (c_e, v_SERCA) = (3000, 650). If we perturb the constant state using an eigenvector associated with an eigenvalue with a positive real part, the magnitude of the perturbation grows. And similarly, if we perturb the resting state using an eigenvector associated with an eigenvalue with a negative real part, the magnitude of the perturbation decays. This behavior is what we expect based on linear theory.

$$\left|\right|z\left|\right|=\sum _i{({\int }_0^L{(z^i(x))}^{2}dx)}^{1/2}$$

where i denotes the components of the vector z = (c_cyt, c_SR, b_cyt, b_SR, p). The blue graph illustrates the L²–norm of the deviation between the resting state and the solution of the system (1) – (5) when the initial condition is perturbed using the eigenvector associated with eigenvalue number 100 (arranged according to the size of real part). The eigenvalue is positive, and the solution deviates from equilibrium at exponential rate. This computation is based on the solution of the fully non-linear model (1) – (5) and illustrates that the linear analysis provides reasonable prediction about the solution of the problem. In the green line we illustrate the non-linear solution of a problem where the initial condition is defined by a perturbation of the resting state using the eigenvector associated with eigenvalue number 101. We observe that the eigenvalue is negative and a rather big perturbation decays in agreement with linear theory.

3.8 Accuracy of linear analysis

In general terms, linear analysis is believed to provide accurate predictions for small perturbations and over small time-intervals. This was illustrated in the example mentioned above. However, in the present case it is extremely hard to provide accurate analytical assessments of how small the perturbation must be and for how small time-intervals the analysis holds. To shed some light on this we compare the development of the deviation from equilibrium, measured in the L²–norm, for the solution of the non-linear system (1) – (5) and the deviation as predicted by linear analysis. We use an eigenvector r as the initial perturbation and then linear theory predicts that the deviation will evolve like e^λtr where λ is the associated eigenvalue. In Figure 8 we compare the deviation from equilibrium measured in the L²–norm for the linear and non-linear cases. We observe that linear analysis is extremely accurate in the stable case (upper panel), and in the unstable case it gives accurate prediction for the first few milliseconds.

Figure 8.

The figure illustrates the accuracy of linear analysis. In the upper panel we compare the development of the L²–norm of the solution of the nonlinear system (1 – 5) and the L²–norm as predicted by linear theory. The initial condition is defined by the eigenvector, scaled to unit length, associated with the principal eigenvalue. In the upper case, we assess a stable problem and the two graphs are almost identical. In the unstable case below, we note that the estimate based on linear theory is good initially and then deviates more and more from the non-linear solution.

3.9 Assessment of stability in terms of non-linear analysis

Finally, we use direct simulations based on the system (1) – (5) to evaluate whether the stability predictions given above hold in the fully nonlinear case. In Figure 9 we consider simulations for four different data-sets using five different values of the extracellular calcium concentration c_e. The initial condition is provided by the associated equilibrium solutions. For each 10 ms the numerical approximation of c_cyt is perturbed, at each computational node, by adding a uniformly distributed random variable in the interval plus/minus 0.05μM. In the left column we have used default data (see Table 1) and we have used v_SERCA = 650μMs⁻¹. By solving the algebraic equation v*(c_e) = 650 we find that the problem should become unstable for c_e > c* ≈ 2743μM, or ≈ 2724μM if we use the fully-non-linear version of c_cyt given as the solution (18). We observe from the figure that for small values of c_e the resting state is stable in the sense that no wave is created from the perturbations. This means that each perturbation simply decays and the solution converges to equilibrium, which now defines a resting state. But when c_e ≥ 3mM, the equilibrium state becomes unstable and we observe that perturbations lead to calcium waves that do not return to the equilibrium.

Figure 9.

The simulations are based on c_e = 1, 2, 3, 4, 5mM from top to bottom; time is horizontal and space is the vertical axis. Each simulation uses initial conditions equal to the equilibrium solution for the given parameter choice (indicated in the titles, relative default values). Plot shows deviation from the initial calcium distribution where blue is zero and red is 1μM. Each 10ms, c_cyt is perturbed by adding a uniformly distributed variable in the interval plus/minus 0.05μM. The simulations go on for 50ms, the vertical axis is 100μm, and Δx = 2μm.

In the second column we have reduced the value of v_SERCA and the theory predicts that we should have unstable solutions for c_e > c* ≈ 4500μM. From the figure, we observe that perturbations do not create waves for c_e smaller than 4mM, and that we do get calcium waves for c_e = 5mM. In the two latter columns we change other parameters and we observe that these changes in the non-linear case behave as predicted by linear theory. The computations presented in Figure 9 support the claim that the curve given in (38) separates stable and unstable resting states of the system (1) – (5).

4 A generalized mathematical model

We have derived a sharp estimate of a stable region for the model stated above. In this section, we will consider the problem of generalizing the result to other models that can be written on the same form. We start with the system of the form

$$\frac{\partial c_{\mathit{cyt}}}{\partial t}=D_{\mathit{cyt}}\frac{{\partial }^2{c}_{\mathit{cyt}}}{\partial x^2}-J_{\mathit{cytbuf}}+\frac{1}{V_{\mathit{cyt}}}(J_{sl}+J_{\mathit{RyR}}-J_{\mathit{SERCA}}),$$ (46)

$$\frac{\partial c_{SR}}{\partial t}=D_{SR}\frac{{\partial }^2{c}_{SR}}{\partial x^2}-J_{\mathit{SRbuf}}+\frac{1}{V_{SR}}(-J_{\mathit{RyR}}+J_{\mathit{SERCA}}),$$ (47)

$$\frac{\partial b_{\mathit{cyt}}}{\partial t}=D_{\mathit{buf}}\frac{{\partial }^2{b}_{\mathit{cyt}}}{\partial x^2}+J_{\mathit{cytbuf}},$$ (48)

$$\frac{\partial b_{SR}}{\partial t}=J_{\mathit{SRbuf}},$$ (49)

$$\frac{\partial p}{\partial t}=k_{on2}F(b_{SR})c_{\mathit{cyt}}(1-p)-k_{\mathit{off}2}p.$$ (50)

Compared to the system (1) – (5), we allow more general functions and fluxes to be involved. We keep the form of J_cytbuf and J_SRbuf unchanged (see (9, 10)) and for the remaining fluxes and functions we make the following assumptions:

1. There is a function $J_{sl}^{-1}$ such that

   $$J_{sl}(c_e,c_{\mathit{cyt}})=0$$
   implies that

   $$c_{\mathit{cyt}}(c_e)=J_{sl}^{-1}(c_e).$$
2. The RyR-term is given by

   $$J_{\mathit{RyR}}(z)=\left(k_{\mathit{RyR}}(z)max(0,\frac{c_{\mathit{cyt}}^m}{c_{\mathit{cyt}}^m+K_{\alpha }^m}-p)+k_{\mathit{leak}}(z)\right)(c_{SR}-c_{\mathit{cyt}}),$$ (51)

   where k_RyR = k_RyR(z) can be any strictly positive function of any of the variables; z denotes the vector (c_cyt, c_SR, b_cyt, b_SR, p), and where the coefficient m can be any positive real number. Furthermore, the function k_leak = k_leak(z) can be any positive function of all the variables.

3. There is a function F⁻¹ such that F (b) = b₀ implies b = F⁻1(b₀).

4. The SERCA pump is given by

   $$J_{\mathit{SERCA}}=v_{\mathit{SERCA}}S(c_{\mathit{cyt}},c_{SR}),$$ (52)

   where S can be any strictly positive function.

Note that in the original paper [1] it is pointed out that in some cases k_leak is given by:

$$k_{\mathit{leak}}(c_{SR})=0.8{\left(\frac{c_{SR}}{500}\right)}^{1.7}.$$ (53)

We therefore only use k_leak = k_leak(c_SR), but the more general case of k_leak = k_leak(z) can be handled in the same manner. Furthermore, we can also have S = S(z) as long as S is strictly positive.

4.1 Constant solutions at rest

Again we start by seeking a constant and stationary solution of the system 1

$$0=-J_{\mathit{cytbuf}}+\frac{1}{V_{\mathit{cyt}}}(J_{sl}+J_{\mathit{RyR}}-J_{\mathit{SERCA}}),$$ (54)

$$0=-J_{\mathit{SRbuf}}+\frac{1}{V_{SR}}(-J_{\mathit{RyR}}+J_{\mathit{SERCA}}),$$ (55)

$$0=J_{\mathit{cytbuf}},$$ (56)

$$0=J_{\mathit{SRbuf}},$$ (57)

$$0=k_{on2}F(b_{SR})c_{\mathit{cyt}}(1-p)-k_{\mathit{off}2}p.$$ (58)

As above, it follows from this that

$$J_{\mathit{SERCA}}=J_{\mathit{RyR}}$$ (59)

and

$$J_{sl}=0.$$ (60)

Therefore, by Assumption 1, we have

$$c_{\mathit{cyt}}(c_e)=J_{sl}^{-1}(c_e).$$ (61)

Since we are seeking stationary and constant in space solutions where the RyR-release is about to activate, it follows from (51) that

$$\frac{c_{\mathit{cyt}}^m}{c_{\mathit{cyt}}^m+K_{\alpha }^m}-p=0,$$ (62)

and thus

$$p(c_e)=\frac{c_{\mathit{cyt}}^m(c_e)}{c_{\mathit{cyt}}^m(c_e)+K_{\alpha }^m}$$ (63)

$$=\frac{J_{sl}^{-m}(c_e)}{J_{sl}^{-m}(c_e)+K_{\alpha }^m}$$ (64)

where we have used the convention that $J_{sl}^{-m}(c)={(J^{-1}(c))}^m$.

From (56) we have

$$J_{\mathit{cytbuf}}=0$$ (65)

and then from (9) we get

$$c_{\mathit{cyt}}{C}_{\mathit{cytbuf}}-(c_{\mathit{cyt}}+K_{\mathit{cytbuf}})b_{\mathit{cyt}}=0$$ (66)

so

$$b_{\mathit{cyt}}(c_e)=\frac{C_{\mathit{cytbuf}}{c}_{\mathit{cyt}}}{c_{\mathit{cyt}}+K_{\mathit{cytbuf}}}$$ (67)

$$=\frac{C_{\mathit{cytbuf}}{J}_{sl}^{-1}(c_e)}{J_{sl}^{-1}(c_e)+K_{\mathit{cytbuf}}}.$$ (68)

Furthermore, from (58), we get

$$k_{on2}F(b_{SR})c_{\mathit{cyt}}(1-p)=k_{\mathit{off}2}p$$ (69)

Then, by using (61) and (62) we get

$$\frac{k_{on2}}{k_{\mathit{off}2}}F(b_{SR})J_{sl}^{-m}(c_e)=\frac{p}{1-p}=\frac{J_{sl}^{-m}(c_e)}{K_{\alpha }^m}$$ (70)

so, using Assumption 3, we arrive at

$$b_{SR}=F^{-1}\left(\frac{k_{\mathit{off}2}}{k_{on2}{K}_{\alpha }^m}\right).$$ (71)

We observe that also in this quite general case, b_SR is again a constant not depending on c_e.

From (57) we have

$$c_{SR}{C}_{\mathit{SRbuf}}-(c_{SR}+K_{\mathit{SRbuf}})b_{SR}=0$$ (72)

so

$$c_{SR}=\frac{K_{\mathit{SRbuf}}{b}_{SR}}{C_{\mathit{SRbuf}}-b_{SR}}$$ (73)

and thus

$$c_{SR}=\frac{K_{\mathit{SRbuf}}{F}^{-1}\left(\frac{k_{\mathit{off}2}}{k_{on2}{K}_{\alpha }^m}\right)}{C_{\mathit{SRbuf}}-F^{-1}\left(\frac{k_{\mathit{off}2}}{k_{on2}{K}_{\alpha }^m}\right)}.$$ (74)

We note that also c_SR is a constant independent of c_e.

It follows from (59) that J_RyR = J_SERCA and because of (62) it follows from (51) that

$$v_{\mathit{SERCA}}S(c_{\mathit{cyt}},c_{SR})=k_{\mathit{leak}}(z)(c_{SR}-c_{\mathit{cyt}}),$$

so

$$v_{\mathit{SERCA}}(c_e)=\frac{k_{\mathit{leak}}(z)(c_{SR}-c_{\mathit{cyt}})}{S(c_{\mathit{cyt}},c_{SR})}$$

Again, we have found the strength of the SERCA-pump as a function of the extracellular calcium concentration c_e at the point in state-space where the RyR-release activates,

$$v^{\ast }(c_e)=\frac{k_{\mathit{leak}}\left(\frac{K_{\mathit{SRbuf}}{F}^{-1}\left(\frac{k_{\mathit{off}2}}{k_{on2}{K}_{\alpha }^m}\right)}{C_{\mathit{SRbuf}}-F^{-1}\left(\frac{k_{\mathit{off}2}}{k_{on2}{K}_{\alpha }^m}\right)}\right)\left(\frac{K_{\mathit{SRbuf}}{F}^{-1}\left(\frac{k_{\mathit{off}2}}{k_{on2}{K}_{\alpha }^m}\right)}{C_{\mathit{SRbuf}}-F^{-1}\left(\frac{k_{\mathit{off}2}}{k_{on2}{K}_{\alpha }^m}\right)}-J_{sl}^{-1}(c_e)\right)}{S\left(J_{sl}^{-1}(c_e),\frac{K_{\mathit{SRbuf}}{F}^{-1}\left(\frac{k_{\mathit{off}2}}{k_{on2}{K}_{\alpha }^m}\right)}{C_{\mathit{SRbuf}}-F^{-1}\left(\frac{k_{\mathit{off}2}}{k_{on2}{K}_{\alpha }^m}\right)}\right)}.$$ (75)

This demonstrates that we can compute the critical curve for the general class of models written in the form given by (54) – (58) provided that Assumptions 1–4 hold. We have, however, not performed numerical investigations covering cases other than those reported above.

5 Discussion

The physiological relevance of our findings are limited to the physiological relevance of the model we have analyzed, which as mentioned above is relatively simple. However, this model was constructed specifically to interrogate the determinants of calcium waves, and at least qualitatively, our analysis regarding the sensitivity of stability transition to specific parameter modifications are in good agreement with experimental data.

5.1 Species-Specific Calcium Stability

Many of the parameter alterations we have imposed are analogous to the differences underlying species-specific characteristics of myocyte calcium-handling, and it is useful to frame the following discussions with a comment on those differences and how they may impact stability. First, rodents exhibit a greater reliance upon SR calcium release for contraction than do higher order species, including rabbits and humans. This increased SR release is fueled largely by a higher SR calcium load, which itself results from enhanced SERCA expression, and reduced NCX function [2]. Both of these differences shift the competitive balance between NCX-mediated calcium extrusion and SR reuptake strongly in favor of reuptake. Based upon our model results, this shift in balance would be expected to promote a less stable diastolic state in the rodent, but it has been clearly shown that these events also occur in higher order species, particularly the rabbit [24, 25]. So our overall expectation is that the model analysed here may become unstable with less severe changes to v_SERCA and c_e, than in a rabbit- or human-specific model. However, rabbit and human models would also be expected to exhibit a greater arrhythmogenic current response to those instabilities due to greater forward mode Na/Ca exchange.

5.2 The Ryanodine Receptor

Mutation of the ryanodine receptor is capable of promoting calcium waves and ventricular arrhythmia. The best-described pathology of this type is the autosomal dominant form of catecholaminergic polymorphic ventricular tachycardia (CPVT). In this syndrome the RyR pool exhibits heightened sensitivity to cytosolic calcium, as reflected by increased open probability [26]. This reduces the SR calcium load required to initiate a calcium wave, and causes these cardiac cells to be unstable during β-adrenergic stimulation. Our modifications to K_α compare well with experimental data describing CPVT. This can be seen in Figure 4, where increasing K_α (decreasing RyR calcium sensitivity) increases the stable range of cellular and SR calcium load.

5.3 SERCA

Experimental manipulations that enhance SR calcium reuptake and SERCA function promote calcium waves, but generally require simultaneous calcium overload [27, 28]. Both of these alterations occur during β-adrenergic stimulation, and for this reason it is not surprising that β-adrenergic agonists produce calcium waves and arrhythmia in CPVT and heart failure [26, 24]. The results of the model suggest that when extracellular calcium concentration is increased ~ 30% increased from baseline, maximal SERCA pump velocity must still more than double before calcium handling becomes unstable. This finding may explain why recent experimental investigations involving SERCA overexpression [29, 30] did not notice increased wave propensity. These studies achieved approximately 37% and 150% increases in SERCA protein, and may therefore not have achieved an SR calcium load sufficient to elicit waves.

5.4 Sodium-Calcium Exchange

The manipulation of K_NCX applied in Figure 6B is roughly analogous to diminishing the thermodynamic gradient for forward-mode Na/Ca exchange (as would occur under conditions of sodium overload). It is well known that intra-cellular sodium overload promotes calcium retention and elevates SR calcium load, particularly in rodent and failing cardiac myocytes [31, 32]. Both of these effects would be expected to reduce diastolic calcium-handling stability, in line with our analytic observations. The prescribed manipulation of v_{N CX} in Figure 6C is more directly analogous to increased NCX protein expression, which is also a well-described manifestation of heart failure [33, 34] and clearly contributes to arrhythmia associated with calcium waves [35, 24, 25]. Interestingly, our simple analysis lends quantitative support to the idea that elevated NCX expression is an effective early compensation in heart failure. By inhibiting calcium wave initiation (i.e. extending the stable region in c_e, V_SERCA), increased NCX expression may improve diastolic stability in the short term. However, under acute β-adrenergic challenge, or with further pathological development, diastolic stability may still be exceeded, and in this circumstance increased v_{N CX} will clearly magnify the resulting arrhythmogenic current development.

6 Conclusion

We have analyzed a mathematical model of calcium handling in cardiomyocytes and found explicit formulas separating parameters leading to stable and unstable resting states of the model. The suggested stability curve is verified through a series of computations based on linear analysis where the eigenvalues of a linearized model are analyzed. Furthermore, the result is confirmed by using fully non-linear computations. The model we have analyzed is intentionally simple and does not include the action potential. By choosing this simple model we have been able to describe the transition from an unstable to a stable resting state for calcium handling with an analytic expression.

Highlights.

• A mathematical model of calcium concentrations in cardiomyocytes is analyzed.

• The main control parameters are the strength of the SERCA-pump and the extracellular calcium concentration.

• A stability curve separating stable and unstable parameters is derived.

• The theory is supported by computational analysis.