Comparing Real and Ideal Neurons: Lessons From ODEs and PDEs#

Single-Compartment Models#

A single-compartment neuron model ignores any spatial variation of voltage across the neuron. Instead, one lumps the cell’s entire membrane into a single circuit:

1. Membrane Capacitance (C): Stores/accumulates charge, affecting the membrane voltage.
2. Ion Channels (sodium, potassium, leak, etc.): Represented by conductances and reversal potentials.
3. External Current (Iext): Any externally injected current or synaptic input.

In a single-compartment model, membrane voltage V(t) satisfies the basic current-balance equation:

[ C \frac{dV}{dt} = - I_{\text{ion}}(V, t) + I_{\text{ext}}(t). ]

Here, ( I_{\text{ion}} ) could be further decomposed into sodium, potassium, leak, and other terms. Each of these ionic currents is a function of the membrane voltage, and often gating variables that track how channels open or close.

Hodgkin–Huxley Model#

Arguably the most famous neuron model, the Hodgkin–Huxley model was derived from experiments on the squid giant axon in the early 1950s. It captures the interplay of fast sodium and delayed rectifier potassium currents that give rise to the action potential.

In the Hodgkin–Huxley framework, the single compartment’s current-balance equation is:

[ C \frac{dV}{dt} = -g_{\text{Na}} m^3 h (V - V_{\text{Na}}) ; - ; g_{\text{K}} n^4 (V - V_{\text{K}}) ; - ; g_{\text{L}} (V - V_{\text{L}}) ; + ; I_{\text{ext}}. ]

Where:

• ( g_{\text{Na}}, g_{\text{K}}, g_{\text{L}} ) are maximal conductances for sodium, potassium, and leak channels respectively.
• ( V_{\text{Na}}, V_{\text{K}}, V_{\text{L}} ) are reversal (equilibrium) potentials for these channels.
• ( m, h, n ) are gating variables that evolve with ODEs describing their voltage-dependent dynamics.

The gating variable ( m ) corresponds to the activation of sodium channels, ( h ) to the inactivation of sodium channels, and ( n ) to the activation of potassium channels. Each gating variable is governed by equations of the form:

[ \frac{dm}{dt} = \alpha_m(V) (1 - m) - \beta_m(V) , m, ]

with similarly structured equations for ( h ) and ( n ). The functions ( \alpha_m(V) ) and ( \beta_m(V) ) are empirically derived from voltage-clamp data and can be approximated by exponentials or more complex expressions.

Because the Hodgkin–Huxley model reliably reproduces action potential generation, conduction velocity, and refractory periods in a single compartment, it is sometimes considered a “gold standard�?reference for biophysical neuron modeling (at least for the currents it includes). Nonetheless, real neurons use many more ion channels, often distributed non-uniformly across dendrites and axons, and exhibit complex morphologies not captured by a single compartment.

Reduced Models: FitzHugh–Nagumo and Others#

As valuable as the Hodgkin–Huxley model is, it can still be computationally expensive. Sometimes we only want to capture the essence of spiking behavior—i.e., a rapid upstroke (action potential) followed by a slower recovery. Reduced models do exactly this by removing many gating variables or merging them into fewer dimensions. Classic examples:

1. FitzHugh–Nagumo Model: Reduces the Hodgkin–Huxley equations to two dimensions: a fast variable ( V ) (membrane voltage) and a slow “recovery�?variable ( w ).
2. Izhikevich Model: Offers a two-dimensional system with voltage reset rules, capturing diverse firing patterns while remaining computationally cheap.
3. Integrate-and-Fire Models: Even simpler, these approximate the subthreshold dynamics with a linear ODE and impose a “spike and reset�?rule once voltage crosses a threshold.

These simplified ODE-based approaches let us handle large networks, concurrency, and theoretical inquiries into neural coding (e.g., pattern generation, synchronization, etc.). But as we move away from the single-compartment idealized world, we need spatially resolved models—enter PDE-based approaches.

The Cable Equation#

A hallmark of PDE-based neuronal modeling is the cable equation, which describes how voltage changes both in time and along a one-dimensional cable representative of an axon or dendrite. Consider a cylindrical cable of length ( L ). The membrane voltage ( V(x,t) ) at position ( x ) and time ( t ) might follow:

[ C_m \frac{\partial V}{\partial t} = D \frac{\partial^2 V}{\partial x^2} - I_{\text{ion}}(V, x, t) + I_{\text{ext}}(x,t), ]

where

• ( C_m ) is the membrane capacitance per unit length,
• ( D ) is a diffusion-like term related to the intracellular and membrane resistances,
• ( I_{\text{ion}} ) is the sum of all ionic currents per unit length,
• ( I_{\text{ext}}(x,t) ) is any externally applied current along the cable.

If one includes Hodgkin–Huxley-type ionic currents, the PDE becomes more complicated but still manageable with numerical methods. Boundary conditions typically represent sealed ends ((\partial V/\partial x = 0)) or coupling to other compartments or spines.

From Cable Equation to Multi-Compartment Models#

While the continuous cable equation is theoretically elegant, simulating it directly can be computationally dense. Neuroscientists often approximate a long dendrite or axon with discrete compartments, each governed by an ODE:

[ C_i \frac{dV_i}{dt} = \frac{V_{i+1} - V_i}{R_{i,i+1}} + \frac{V_{i-1} - V_i}{R_{i,i-1}} - I_{\text{ion},i}(V_i, t), ]

where ( R_{i,j} ) is the axial resistance between compartments ( i ) and ( j ). This approach is sometimes called the multi-compartment method. In effect, the PDE has been discretized in space, leading to a system of ODEs in time. For complicated dendritic trees, each branch can be subdivided into multiple compartments, forming a branching tree system.

Dendritic Processing and Nonlinearities#

Because dendrites contain active ion channels, nonlinearities can arise in the PDE-based cable equation. This leads to phenomena such as:

1. Back-propagation of Action Potentials: Spikes initiated at the axon hillock can travel back into dendrites, modulating synaptic plasticity.
2. Forward-Propagation of Dendritic Spikes: Under certain circumstances, dendrites can generate localized spikes that then propagate to the soma.
3. Complex Synaptic Integration: Spatially distributed inputs combine sublinearly or supralinearly, depending on channel distributions and dendritic properties.

These PDE-based or PDE-derived approaches bring us closer to “real�?neurons by including morphological effects. However, they can become very large models if every branch and channel distribution is included. Balancing detail and computational feasibility is a core challenge in modern computational neuroscience.

Table: Key Differences#

Reasons to Use Idealized vs. Biophysical Models#

• Idealized Models (ODEs):

  • Large network simulations where each neuron is simplified.
  • Theoretical analysis of spike timing, oscillator coupling, and basic coding.
  • Rapid prototyping when investigating fundamental properties of neural systems.

• Biophysical Models (PDEs):

  • Detailed single-neuron analyses of dendritic processing, synaptic integration.
  • Evaluating the impact of subcellular channel localization on neuron firing.
  • Understanding conduction velocity changes along axons under perturbations or disease models.

Real neurons justify PDE-based detail, but ideal neurons remain essential for high-level conceptual insights and computational efficiency.

ODE Example: Integrate-and-Fire Neuron in Python#

This simplistic code snippet demonstrates a basic “leaky integrate-and-fire�?neuron model:

1
import numpy as np
2
import matplotlib.pyplot as plt
3

4
# Parameters
5
T = 100.0        # total time (ms)
6
dt = 0.1         # time step (ms)
7
time = np.arange(0, T, dt)
8

9
R = 10.0         # membrane resistance (M ohms)
10
C = 1.0          # membrane capacitance (uF)
11
tau = R*C        # membrane time constant (ms)
12
V_reset = -65.0  # reset potential (mV)
13
V_th = -50.0     # threshold potential (mV)
14
I_ext = 1.5      # external input current (nA)
15

16
# Initialize
17
V = np.zeros_like(time)
18
V[0] = V_reset
19

20
spike_times = []
21

22
# Simulation
23
for i in range(1, len(time)):
24
    dV = (-(V[i-1] - V_reset) + R * I_ext) / tau
25
    V[i] = V[i-1] + dV * dt
26

27
    # Check for spike
28
    if V[i] >= V_th:
29
        V[i] = V_reset
30
        spike_times.append(time[i])
31

32
# Plot results
33
plt.figure(figsize=(8, 4))
34
plt.plot(time, V, label='Membrane potential')
35
plt.axhline(y=V_th, color='r', linestyle='--', label='Threshold')
36
plt.xlabel('Time (ms)')
37
plt.ylabel('Voltage (mV)')
38
plt.title('Leaky Integrate-and-Fire Neuron')
39
plt.legend()
40
plt.show()
41

42
print("Spike times (ms):", spike_times)

Here, the membrane voltage evolves according to:

[ \tau \frac{dV}{dt} = -(V - V_{\text{rest}}) + R \cdot I_{\text{ext}}, ]

with a “hard�?threshold. This code helps illustrate how to implement an idealized single-compartment ODE neuron.

PDE Example: Simple Cable Equation Simulation#

For a PDE-based approach, let’s approximate the cable equation in 1D. Below is a conceptual example using finite differences:

1
import numpy as np
2
import matplotlib.pyplot as plt
3

4
# Parameters
5
L = 1.0          # length of cable (cm)
6
dx = 0.01        # spatial step (cm)
7
T = 5.0          # total time (ms)
8
dt = 0.001       # time step (ms)
9

10
x = np.arange(0, L+dx, dx)
11
time = np.arange(0, T+dt, dt)
12

13
# Biophysical constants (simplified)
14
Cm = 1.0     # membrane capacitance (uF/cm^2)
15
Rm = 10.0    # membrane resistance (kOhm*cm)
16
Ri = 100.0   # axial (intracellular) resistance (Ohm*cm)
17
D = 1.0 / (Rm * Ri)   # simplified "diffusion" coefficient
18

19
# External stimulation
20
Iext = np.zeros_like(x)
21
# Inject current at the center
22
Iext[len(x)//2] = 1.0
23

24
V = np.zeros((len(time), len(x)))  # voltage array
25

26
# PDE simulation with finite differences
27
for n in range(0, len(time) - 1):
28
    # Use explicit Euler for demonstration
29
    for i in range(1, len(x) - 1):
30
        d2V_dx2 = (V[n,i+1] - 2*V[n,i] + V[n,i-1]) / (dx**2)
31
        dV_dt = D * d2V_dx2 / Cm + Iext[i]/Cm
32
        V[n+1,i] = V[n,i] + dt * dV_dt
33

34
    # Boundary conditions: no flux at the ends
35
    V[n+1,0] = V[n+1,1]
36
    V[n+1,-1] = V[n+1,-2]
37

38
# Plot voltage distribution at final time
39
plt.figure(figsize=(8,4))
40
plt.plot(x, V[-1,:])
41
plt.xlabel('Position (cm)')
42
plt.ylabel('Voltage (mV) [arbitrary scale]')
43
plt.title('Simple Cable Simulation - Final Voltage Profile')
44
plt.show()

In this contrived example:

• We treat the cable equation as ( \partial V / \partial t = D , \partial^2 V / \partial x^2 + I_{\text{ext}} / C_m ) for simplicity (ignoring ionic currents).
• The boundary conditions at ( x=0 ) and ( x=L ) approximate sealed ends (zero flux), so ( \partial V / \partial x = 0 ) at boundaries.
• A small current is injected at the center of the cable, and we observe how the voltage diffuses away over time.

This PDE approach allows us to examine how voltage depends on space in addition to time.

Advanced PDE Example: Nonlinear Cable Model#

To include Hodgkin–Huxley currents in a spatially extended cable, you’d have to attach the gating variables at each spatial point, forming a large set of PDEs (or a large set of coupled ODEs after discretization). Conceptually, you might do:

[ C_m \frac{\partial V}{\partial t} = D \frac{\partial^2 V}{\partial x^2} - I_{\text{ion}}(V,m,h,n) + I_{\text{ext}}(x,t), ] [ \frac{\partial m}{\partial t} = \alpha_m(V)(1-m) - \beta_m(V)m, ] and similarly for (h) and (n). Then you discretize along ( x ), for instance with a finite difference or finite element method. Each grid point has its own set of gating variables. The computational demands grow quickly, but the fidelity to real neuronal behavior also increases dramatically.

Detailed Ion Channel Modeling#

Real neurons often have tens or hundreds of different channel subtypes, each regulated by second messengers, phosphorylation states, or local microenvironments. Biophysically detailed models may incorporate:

1. Multiple Channel Subtypes: E.g., transient sodium, persistent sodium, A-type potassium, calcium-dependent potassium, HCN channels, etc.
2. Modulatory Effects: Activation of channels by neuromodulators (e.g., dopamine, acetylcholine) can shift gating dynamics.
3. Spatial Variations: Ion channels vary in density along dendrites or axons, requiring PDE or multi-compartment approaches to capture local changes in excitability.

Such detail helps replicate experimental recordings more precisely, but it comes at high computational and parameter-fitting cost.

Network Models and PDE Approaches#

PDE-based approaches are sometimes extended from single neurons to entire networks when:

• Analyzing population activity in a spatial domain, like cortical tissue.
• Studying traveling waves, oscillations, or spatiotemporal patterns in large-scale brain areas.

Emerging PDE-based frameworks like neural field equations capture the average activity of large populations of neurons with continuous spatial coordinates. Meanwhile, local features (like synaptic microcircuits) might still be approximated or integrated into the PDE system.

Conduction Delays and Multidimensional PDEs#

For complex morphological neurons (3D branching structures), conduction delays can become crucial. As an action potential travels, the speed depends on axon diameter, myelination, and channel distribution. Multidimensional PDEs or specialized PDEs for fiber tracts can approximate realistic conduction velocities. In a 2D or 3D model, each segment of the neuron or fiber track is an element in the partial differential domain.

Electrodiffusion and Metabolic Coupling#

Beyond modeling the membrane potential, advanced PDE-based models may incorporate electrodiffusion of ions in the intracellular and extracellular space, capturing concentration gradients that shift with firing activity. For instance, changes in extracellular potassium concentration can influence subsequent neuronal excitability. Additionally, metabolic processes that generate or consume ATP can feed back into channel functions and homeostasis, requiring further PDE or PDE-ODE coupled modeling to track ionic and metabolic states over time and space.