Two computational regimes of a single-compartment neuron separated by a planar boundary in conductance space

Brian N. Lundstrom, Sungho Hong, Matthew H. Higgs, Adrienne L. Fairhall

Abstract

Recent in vitro data show that neurons respond to input variance with varying sensitivities. Here, we demonstrate that Hodgkin-Huxley (HH) neurons can operate in two computational regimes, one that is more sensitive to input variance (differentiating) and one that is less sensitive (integrating). A boundary plane in the 3D conductance space separates these two regimes. For a reduced HH model, this plane can be derived analytically from the V nullcline, thus suggesting a means of relating biophysical parameters to neural computation by analyzing the neuron's dynamical system.

Introduction: The Logic of Ion Channels

In neuroscience, neurons are often categorized by their "ion channel soup"—the specific mix of sodium, potassium, and leak channels they express. However, it has not always been clear how this biophysical makeup translates into specific computational functions. Neurons generally fall into two functional camps:

  1. Integrators: These act as "bean counters," summing inputs over time and firing based on the average input level.
  2. Differentiators: These act as "coincidence detectors," ignoring steady-state inputs and responding primarily to rapid fluctuations or high input variance.
Qualitative response of integrator (middle trace) and differentiator (bottom trace) neurons to step and noisy current inputs (top trace).

The central challenge addressed in this research is whether a clear, mathematical boundary exists that determines when a neuron will switch from one computational mode to the other based purely on its biophysical parameters.

Methods: Analytical and Computational Modeling

The authors combined computational simulations with analytical dynamical systems theory:

  • Simulations: They utilized the standard Hodgkin-Huxley (HH) model, running simulations across 56 distinct combinations of maximal conductances for sodium ($G_{Na}$), potassium ($G_K$), and leak ($G_{Leak}$).
  • Model Reduction: To make the complex 4D HH equations tractable, they used the 2D Abbott-Kepler (AK) model. A 0.75 correction factor was applied to the sodium activation variable in the 2D model to align its excitability with the full 4D simulations.
  • Phase Plane Analysis: By analyzing the "nullclines" (curves where the rate of change of a variable is zero) of the system, they could determine if a neuron would fire repetitively to a constant current (integrator) or remain silent unless "pushed" by noise (differentiator).
  • Linear Regression: Multiple least-squares regression was used to find the best-fit plane in the 3D conductance space that separates the two behaviors.

Key Results: The Planar Boundary in Conductance Space

Despite the highly non-linear nature of neural equations, the study found that the transition between integrating and differentiating is governed by a simple linear plane in 3D conductance space. For the standard HH model, the boundary is defined by:

$$G_{Na} - 2.07G_K - 22.8G_{Leak} = 0$$

When the weighted sum of these conductances is less than zero, the neuron becomes a differentiator and will not fire in response to a noiseless, constant current, regardless of its strength.

Firing rate vs. mean current for different noise levels (SD) in the integrator (b) and differentiator (c) regimes. Note the increased sensitivity to variance in the differentiator regime.

The Role of Sodium and Geometric Transitions

Lowering the maximal sodium conductance ($G_{Na}$) is the most direct way to turn an integrator into a differentiator. Geometrically, this transition occurs when the neuron's "fixed point" (its steady state) moves to the left of the local minimum of the $V$-nullcline. In this region, the system becomes stable for all levels of constant input, meaning it cannot fire unless the input is fluctuating (noisy). This is biologically significant because processes like slow sodium inactivation can effectively lower $G_{Na}$ in real-time, allowing a single neuron to dynamically switch its computational mode.

Discussion and Future Directions

This research provides a bridge between biophysical parameters and neural logic, suggesting that a neuron's "logic" can be predicted from its conductance profile. However, several limitations remain for future study:

  • Single-Compartment Limitation: The study uses a "point neuron" model, which does not capture the complexity of dendritic trees.
  • Static vs. Dynamic: While the paper discusses dynamic switching, the primary analysis treats conductances as static parameters.
  • Experimental Validation: The authors suggest their method could infer biophysical parameters from experimental $I(V)$ plots, but this remains to be rigorously tested in biological settings.

Summary updated

Abstract

Recent in vitro data show that neurons respond to input variance with varying sensitivities. Here, we demonstrate that Hodgkin-Huxley (HH) neurons can operate in two computational regimes, one that is more sensitive to input variance (differentiating) and one that is less sensitive (integrating). A boundary plane in the 3D conductance space separates these two regimes. For a reduced HH model, this plane can be derived analytically from the V nullcline, thus suggesting a means of relating biophysical parameters to neural computation by analyzing the neuron's dynamical system.
BibTeX
@inproceedings{lundstrom2008two,
  author = {Brian N. Lundstrom and Sungho Hong and Matthew H. Higgs and Adrienne L. Fairhall},
  title = {{Two computational regimes of a single-compartment neuron separated by a planar boundary in conductance space}},
  booktitle = {Neural},
  year = {2008},
  doi = {10.1162/neco.2007.05-07-536},
}
Figure