m08 · Action potentials and ionic models, from FitzHugh–Nagumo to detailed atrial models
Education only. Not for diagnosis, treatment or any clinical decision.
In 60 seconds
A heart cell’s membrane is a capacitor with ion channels in parallel. Each ion pulls the voltage toward its Nernst potential, and channels open and close through gates whose rates depend on voltage. Fast inward current through a few channels makes the upstroke; slower outward currents bring the voltage back. Two variables are enough to capture what matters for mapping: a fast, voltage-like variable that can jump to an excited state, and a slow recovery variable that sets the refractory period. The FitzHugh–Nagumo model shows threshold and repetitive firing in a phase plane. The Mitchell–Schaeffer model is the cardiac version: its action potential duration (APD) follows from one gate decaying while the cell is excited, and it gives restitution, the shorter action potential after a shorter rest. Detailed atrial models add dozens of currents, fitted to human atrial data, to explain why the action potential changes with rate, drugs and disease.
From the field
- Refractoriness is what pacing protocols measure. An S1–S2 protocol (m11) finds the shortest coupling interval that captures. In the Mitchell–Schaeffer lab it is 217 ms after beats 500 ms apart, and it falls as the action potential shortens.
- Atrial fibrillation shortens refractoriness. In goats kept in atrial fibrillation, the atrial effective refractory period (AERP) at 400-ms pacing fell from 146 ± 19 to 95 ± 20 ms, and the changes reversed within a week of sinus rhythm (Wijffels et al. 1995).
- Detailed models are fitted to diseased tissue. Human atrial data come from patients with underlying heart disease (Koivumäki et al. 2011), so even a detailed model describes a particular population.
- Simple models are calibrated, not derived. Two-variable models reproduce restitution and refractoriness with a handful of parameters, which is why they are used for fitting to mapping data (m33). Their parameters are not ionic quantities.
Derivations
Symbols used throughout. \(V\) membrane voltage (mV); \(C_m\) membrane capacitance per area; \(I_{\text{ion}}\) total ionic current; \(I_{\text{stim}}\) stimulus current; \(R\) gas constant, \(T\) absolute temperature, \(F\) Faraday constant, \(z\) valence; \([X]_o\), \([X]_i\) outside and inside concentrations. \(m\), \(h\) gate variables; \(\alpha\), \(\beta\) opening and closing rates (1/ms); \(\bar g\) maximal conductance. FitzHugh–Nagumo: \(v\), \(w\), \(I\), \(a\), \(b\), \(\varepsilon\) (dimensionless). Mitchell–Schaeffer: \(v \in [0, 1]\), gate \(h\), time constants \(\tau_{\text{in}}\), \(\tau_{\text{out}}\), \(\tau_{\text{open}}\), \(\tau_{\text{close}}\) (ms), threshold \(v_{\text{gate}}\), stimulus \(J_{\text{stim}}\); DI diastolic interval, APD action potential duration.
The membrane as a circuit
Charge conservation across the membrane:
\[ C_m \frac{dV}{dt} = -I_{\text{ion}} + I_{\text{stim}} \tag{1}\]
With only a fixed leak, \(I_{\text{ion}} = V/R\) and a current step charges the membrane exponentially with time constant \(\tau_m = RC_m\). Each ion species is driven toward its Nernst potential, where diffusion and the electric field balance:
\[ E_X = \frac{RT}{zF}\ln\frac{[X]_o}{[X]_i} \tag{2}\]
With the exact SI values of the elementary charge, Boltzmann and Avogadro constants (Tiesinga et al. 2021), \(RT/F\) = 26.73 mV at 37 °C: 61.5 mV per tenfold gradient for a monovalent ion. A potassium-like gradient of 30 to 1, an illustrative ratio, gives −90.9 mV (lab part A).
Key takeaways
- The membrane integrates current; its time constant is \(RC_m\).
- Nernst potentials set where each current pushes the voltage: 61.5 mV per decade at 37 °C.
Gates
Hodgkin and Huxley described each channel’s open fraction with gates that open and close at voltage-dependent rates, and the sodium current as a product of gates (Hodgkin and Huxley 1952):
\[ \frac{dm}{dt} = \alpha_m(V)(1 - m) - \beta_m(V)\,m, \qquad m_\infty = \frac{\alpha_m}{\alpha_m + \beta_m}, \quad \tau_m = \frac{1}{\alpha_m + \beta_m}, \qquad I_{\text{Na}} = \bar g_{\text{Na}}\,m^3 h\,(V - E_{\text{Na}}) \tag{3}\]
At fixed voltage the rates are constant and the gate relaxes exponentially toward \(m_\infty\). Updating gates with that exponential is exact for any time step under voltage clamp; forward Euler is not, and with a 2-ms step and a 2-ms time constant its error is 0.11 (lab part B). Detailed cardiac models are many such gates and currents. The Courtemanche human atrial model was built from currents recorded in human atrial myocytes, and it attributes rate adaptation of the APD to incomplete recovery of the L-type calcium current and incomplete deactivation of the delayed rectifier at fast rates (Courtemanche et al. 1998). Later models add detailed intracellular calcium handling (Koivumäki et al. 2011).
Key takeaways
- A gate is a first-order system with voltage-dependent rate constants.
- Detailed atrial models stack many gated currents; they explain mechanisms but need many fitted parameters.
Excitability in two variables
FitzHugh reduced excitable membranes to two variables, one for excitability and one for refractoriness, and drew their physiological states on a phase plane (FitzHugh 1961). In the form used here:
\[ \frac{dv}{dt} = v - \frac{v^3}{3} - w + I, \qquad \frac{dw}{dt} = \varepsilon\,(v + a - b\,w) \tag{4}\]
The \(v\)-nullcline is a cubic and the \(w\)-nullcline a line; rest is where they cross. Because \(\varepsilon\) is small, \(v\) moves fast and \(w\) slowly. A kick that carries \(v\) past the middle branch of the cubic sends the state around the right branch: an action potential, followed by a slow return during which a second kick fails (refractoriness). With the illustrative values \(a = 0.7\), \(b = 0.8\), \(\varepsilon = 0.08\), a kick of 0.99 times the threshold peaks at \(v = -0.46\) and one of 1.01 times peaks at 1.65 (lab part C). A steady current shifts the rest state. It loses stability when the trace of the Jacobian vanishes:
\[ 1 - v_*^2 - \varepsilon b = 0 \;\Longrightarrow\; v_* = -\sqrt{1 - \varepsilon b}, \qquad I_H = \frac{v_*^3}{3} - v_* + \frac{v_* + a}{b} \tag{5}\]
Above \(I_H\) (0.331 here) the cell fires repeatedly, the impulse trains FitzHugh described for constant currents that make the resting point unstable (FitzHugh 1961).
Simulation
Assumptions
- Model: Equation 4 integrated by fourth-order Runge–Kutta (step 0.05) for 150 time units, starting at rest with a kick added to \(v\). Tested against
epml.membrane. - Parameters: \(a = 0.7\), \(b = 0.8\), \(\varepsilon = 0.08\), kick and current: illustrative, dimensionless.
- What it ignores: real ionic currents and units; FitzHugh–Nagumo has no plateau and no restitution of its own.
Key takeaways
- Threshold, all-or-none response and refractoriness come from a fast variable with a cubic nullcline and a slow recovery variable.
- Enough steady current destabilizes rest (Equation 5), and the cell fires on its own.
A two-current cardiac model
Mitchell and Schaeffer wrote the cardiac action potential with one gated inward current and one outward current (Mitchell and Schaeffer 2003):
\[ \frac{dv}{dt} = \frac{h\,v^2(1 - v)}{\tau_{\text{in}}} - \frac{v}{\tau_{\text{out}}} + J_{\text{stim}}, \qquad \frac{dh}{dt} = \begin{cases} (1 - h)/\tau_{\text{open}}, & v < v_{\text{gate}} \\ -h/\tau_{\text{close}}, & v \ge v_{\text{gate}} \end{cases} \tag{6}\]
Setting \(dv/dt = 0\) with \(v > 0\) gives \(v(1 - v) = \tau_{\text{in}}/(h\,\tau_{\text{out}})\). Since \(v(1 - v) \le 1/4\), the excited state exists only while \(h \ge h_{\min} = 4\tau_{\text{in}}/\tau_{\text{out}}\). During the action potential \(h\) decays as \(e^{-t/\tau_{\text{close}}}\), and during diastole it recovers toward 1 with time constant \(\tau_{\text{open}}\). Together:
\[ \text{APD} \approx \tau_{\text{close}}\ln\frac{h_0}{h_{\min}}, \qquad h_{\min} = \frac{4\tau_{\text{in}}}{\tau_{\text{out}}}, \qquad h_0 = 1 - (1 - h_{\min})\,e^{-\text{DI}/\tau_{\text{open}}} \tag{7}\]
Here \(h_0\) is the gate at the upstroke, assuming it closed to \(h_{\min}\) at the end of the previous beat. We use the parameter values Nieto Ramos and colleagues report as based on the original paper: \(\tau_{\text{in}}\) = 0.3 ms, \(\tau_{\text{out}}\) = 6 ms, \(\tau_{\text{open}}\) = 150 ms, \(\tau_{\text{close}}\) = 120 ms, \(v_{\text{gate}}\) = 0.13 (Nieto Ramos et al. 2023). Then \(h_{\min}\) = 0.2, and a rested cell loses its excited state after \(120\ln 5\) = 193.1 ms; the simulation agrees to 0.1 ms. The voltage then lingers near the vanished state and crosses \(v_{\text{gate}}\) 25–40 ms later, so measured APDs run that much longer than the formula. The slope of restitution at short intervals is \(\tau_{\text{close}}(1 - h_{\min})/(\tau_{\text{open}}h_{\min})\) = 3.2, steep enough to matter for alternans (m10).
Simulation
Assumptions
- Model: one Mitchell–Schaeffer cell (Equation 6) by forward Euler, step 0.02 ms; four S1 beats 500 ms apart and one S2; stimuli of 0.5 per ms for 1 ms. APD and DI are measured at \(v_{\text{gate}}\). Tested against
epml.membrane. - Parameters: as reported by Nieto Ramos et al. (2023), based on Mitchell and Schaeffer (2003); the closing time constant slider and the stimulus are free choices for illustration.
- What it ignores: ionic detail, calcium handling, electrotonic coupling to neighbors (m09) and memory beyond the previous beat.
Key takeaways
- The APD is set by how long the inward gate takes to close to \(h_{\min}\); restitution comes from how far it reopened.
- Restitution is steep at short diastolic intervals.
Clinical correlate
This section describes published findings only. It is not guidance for patient care, and it has not yet been reviewed by an EP physician.
Where this shows up in EP. Refractoriness and its rate dependence are what an EP study measures with extrastimuli (m11), and they set the wavelength of reentry (m10). In chronically instrumented goats, maintaining atrial fibrillation shortened the AERP by 35%, reversed its normal rate adaptation, and made atrial fibrillation easier to induce and longer lasting; all changes reversed within a week of sinus rhythm (Wijffels et al. 1995). Mathematical models of the human atrial action potential connect such changes to specific currents (Courtemanche et al. 1998; Koivumäki et al. 2011), and two-variable models are fitted to mapping data in digital twins (m33).
What the evidence is. An animal model of atrial fibrillation (Wijffels et al. 1995) and computational models built on human atrial cell data (Courtemanche et al. 1998; Koivumäki et al. 2011).
What a reader should ask:
- At which pacing rate was refractoriness measured? It depends on rate (Equation 7).
- Which model, fitted to which data? Detailed models carry the population they were fitted to.
- Is a model parameter a physiological quantity? In two-variable models it usually is not (Equation 6).
Visual summary
Schematic
Assumptions
- Schematic: not to scale, no quantitative claim. Labels follow Equation 1, Equation 4 and Equation 7.
How experts think
Before using an action potential model, ask:
- What question needs answering? Ionic mechanism needs a detailed model; wave dynamics often needs only restitution and refractoriness.
- What data was the model fitted to? Species, chamber, disease and pacing rate all matter.
- Does it reproduce restitution? Wave stability depends on it (Equation 7, m10).
- How is it integrated? Gates need exponential updates or small steps (Equation 3).
- Which parameters are physical? Phenomenological parameters are fitted shapes, not conductances (Equation 6).
One sentence to keep: the action potential is a fast switch with a slow reset; most of mapping happens on the reset’s time scale.
Cheat sheet
| Quantity | Result |
|---|---|
| Membrane (Equation 1) | \(C_m\,dV/dt = -I_{\text{ion}} + I_{\text{stim}}\); passive \(\tau_m = RC_m\) |
| Nernst (Equation 2) | \(RT/F\) = 26.73 mV at 37 °C; 61.5 mV per decade for \(z = 1\) |
| Gates (Equation 3) | \(m_\infty = \alpha/(\alpha + \beta)\), \(\tau = 1/(\alpha + \beta)\); \(I_{\text{Na}} = \bar g m^3h(V - E)\) |
| FitzHugh–Nagumo (Equation 4) | cubic and linear nullclines; threshold at the middle branch |
| Hopf onset (Equation 5) | \(v_* = -\sqrt{1 - \varepsilon b}\); 0.331 for \(a = 0.7\), \(b = 0.8\), \(\varepsilon = 0.08\) |
| Mitchell–Schaeffer (Equation 6) | excited state needs \(h \ge 4\tau_{\text{in}}/\tau_{\text{out}}\) |
| APD and restitution (Equation 7) | \(\tau_{\text{close}}\ln(h_0/h_{\min})\); slope at DI → 0: \(\tau_{\text{close}}(1 - h_{\min})/(\tau_{\text{open}}h_{\min})\) |
Values used in this module
| Value | Source |
|---|---|
| Elementary charge, Boltzmann and Avogadro constants (exact) | Tiesinga et al. (2021) |
| Mitchell–Schaeffer parameters 0.3, 6, 150, 120 ms and 0.13 | Nieto Ramos et al. (2023), based on Mitchell and Schaeffer (2003) |
| Goat AERP 146 ± 19 → 95 ± 20 ms with sustained atrial fibrillation; reversible within a week | Wijffels et al. (1995) |
| Concentrations, FitzHugh–Nagumo parameters, stimuli | illustrative |
| Thresholds, APDs, refractory periods, restitution gaps | our own code |
Interview questions
Try each question before you open its model answer. In a real interview, say which assumptions you are making.
Q1. Why is the membrane modeled as a capacitor?
Model answer
The lipid bilayer separates charge across a thin insulator; ionic currents charge and discharge it, so \(C_m\,dV/dt = -I_{\text{ion}} + I_{\text{stim}}\) (Equation 1).
Q2. What is a Nernst potential, and how big is one decade of gradient?
Model answer
The voltage at which diffusion and electric drift of one ion balance, \((RT/zF)\ln([X]_o/[X]_i)\): 61.5 mV per tenfold gradient for a monovalent ion at 37 °C (Equation 2).
Q3. What does a Hodgkin–Huxley gate do under voltage clamp?
Model answer
It relaxes exponentially toward \(\alpha/(\alpha + \beta)\) with time constant \(1/(\alpha + \beta)\) (Equation 3) (Hodgkin and Huxley 1952).
Q4. Why do simulators update gates with an exponential rather than forward Euler?
Model answer
At fixed voltage the exponential update is exact for any step; Euler becomes inaccurate or unstable when the step is comparable with the gate’s time constant (Equation 3; lab part B).
Q5. Explain threshold in the FitzHugh–Nagumo phase plane.
Model answer
The middle branch of the cubic \(v\)-nullcline separates states that return to rest from those that jump to the right branch; a kick past it gives a full action potential (Equation 4) (FitzHugh 1961).
Q6. Where does refractoriness come from in a two-variable model?
Model answer
The slow variable is still high after an action potential, so the threshold is farther away and a second stimulus fails or gives a smaller response (Equation 4).
Q7. When does a FitzHugh–Nagumo cell fire on its own?
Model answer
When a steady current moves the rest state to where the Jacobian’s trace vanishes, \(v_* = -\sqrt{1 - \varepsilon b}\); beyond that the cell fires repeatedly (Equation 5).
Q8. What sets the APD in the Mitchell–Schaeffer model?
Model answer
The excited state exists only while \(h \ge 4\tau_{\text{in}}/\tau_{\text{out}}\); \(h\) decays with \(\tau_{\text{close}}\), so the excited state ends after \(\tau_{\text{close}}\ln(h_0/h_{\min})\) (Equation 7) (Mitchell and Schaeffer 2003).
Q9. What is restitution, and why is it steep at short intervals?
Model answer
APD as a function of the preceding diastolic interval. After a short rest the gate has barely reopened, and the APD depends on the log of a small quantity, so it changes fast (Equation 7).
Q10. What did sustained atrial fibrillation do to refractoriness in goats?
Model answer
AERP at 400-ms pacing fell from 146 to 95 ms (−35%), and its normal rate adaptation reversed; all changes reversed within a week of sinus rhythm (Wijffels et al. 1995).
Q11. When would you choose a detailed atrial model over a two-variable one?
Model answer
When the question is about specific currents, drugs or remodeling mechanisms (Courtemanche et al. 1998; Koivumäki et al. 2011). For wave dynamics or fitting to mapping data, a two-variable model with the right restitution is often enough.
Q12. A colleague says the model’s \(\tau_{\text{close}}\) is the potassium current’s time constant. Comment.
Model answer
It is not: in the two-current model it is a phenomenological rate that sets the APD (Equation 6). Mapping it to a specific current needs a detailed model.
Code lab
The lab builds from a circuit to a restitution curve.
| Part | What you do | Data |
|---|---|---|
| A | The membrane as a circuit; Nernst potentials (Equation 1, Equation 2) | synthetic |
| B | Gates under voltage clamp; exact and Euler updates (Equation 3) | synthetic |
| C | FitzHugh–Nagumo: threshold, refractoriness, repetitive firing (Equation 4, Equation 5) | synthetic |
| D | Mitchell–Schaeffer: the action potential and S1–S2 restitution (Equation 6, Equation 7) | synthetic |
| E | Shorter action potentials, as in electrical remodeling (Equation 7) | synthetic |
- Python notebook: labs/python/m08_lab.ipynb. Run it from the course repository after
pip install -r requirements.txt, or open it in Google Colab. - Data: none to download.
Each part ends with checks: lines that pass with the code as given. If you change something and a check fails, that is information, not an error to hide.
Self-check
Short checks you can do in your head. Open each answer after you have tried it.
S1. \(R\) = 10 MΩ, \(C\) = 100 pF. Membrane time constant?
Answer
1 ms (Equation 1).
S2. A divalent ion with a 100-fold gradient at 37 °C. Nernst potential magnitude?
Answer
\(2 \times 61.5/2\) = 61.5 mV (Equation 2).
S3. \(\alpha\) = 0.3 /ms, \(\beta\) = 0.1 /ms. \(m_\infty\) and \(\tau\)?
Answer
0.75 and 2.5 ms (Equation 3).
S4. \(\tau_{\text{in}}\) = 0.3 ms, \(\tau_{\text{out}}\) = 6 ms. \(h_{\min}\)?
Answer
0.2 (Equation 7).
S5. With \(h_{\min}\) = 0.2 and \(\tau_{\text{close}}\) = 120 ms, the longest APD by the closed form?
Answer
\(120\ln 5 \approx 193\) ms (Equation 7).
S6. You halve \(\tau_{\text{close}}\). What happens to every APD by the closed form?
Answer
It halves (Equation 7).
Literature figures
These figures are worth looking at alongside this module. We link to them instead of copying them: some licenses do not allow reuse here, and others we could not confirm. Each entry says what to look for and where this module explains it.
- The physiological state diagram. FitzHugh divides the phase plane into resting, active and refractory regions (FitzHugh 1961). This module: excitability in two variables.
- AERP shortening. Wijffels and colleagues plot refractory periods and fibrillation intervals over the first days of atrial fibrillation (Wijffels et al. 1995). This module: clinical correlate.
- A human atrial action potential model. Courtemanche and colleagues show model action potentials against human atrial recordings and at different rates (Courtemanche et al. 1998). This module: gates.
Further reading
- Foundations. Gates and the sodium current (Hodgkin and Huxley 1952); the two-variable model (FitzHugh 1961).
- Cardiac models. The two-current model (Mitchell and Schaeffer 2003; Nieto Ramos et al. 2023); human atrial models (Courtemanche et al. 1998; Koivumäki et al. 2011).
- Remodeling. Atrial fibrillation begets atrial fibrillation (Wijffels et al. 1995).
- Related modules. m09 (propagation), m10 (arrhythmia mechanisms), m11 (the EP study), m33 (digital twins).