Spikes can transmit neurons' subthreshold membrane potentials

Valentin SchmutzarXiv 2025 · DOI · arXiv

Abstract: Neurons primarily communicate through the emission of action potentials, or spikes. To generate a spike, a neuron's membrane potential must cross a defined threshold. Does this spiking mechanism inherently prevent neurons from transmitting their subthreshold membrane potential fluctuations to other neurons? We prove that, in theory, it does not. The subthreshold membrane potential fluctuations of a presynaptic population of spiking neurons can be perfectly transmitted to a downstream population of neurons. Mathematically, this surprising result is an example of concentration phenomenon in high dimensions.

Introduction: The "Iceberg" Problem in Neural Coding

In neuroscience, a long-standing challenge is the "iceberg" problem. A neuron’s membrane potential ($V_m$) fluctuates constantly in response to synaptic inputs, yet the neuron only communicates with its neighbors by firing a discrete "spike" (action potential) when that potential crosses a specific threshold. To a single downstream observer, anything happening below that threshold appears invisible. This suggests a massive loss of information: if a neuron spends the vast majority of its time below threshold, is all that subthreshold data—which reflects the total synaptic input—simply lost? While previous research indicated that a single neuron's potential can only be partially recovered from its spikes, this work explores whether a population of neurons can perfectly transmit these hidden signals.

Methodology

The study utilizes a mathematical framework to demonstrate that perfect transmission is theoretically possible.

The Model and Input Structure

The research employs a population of $N$ "Linear-Nonlinear-Poisson" (LNP) neurons. Each neuron $i$ has a time-varying potential $V_i(t)$ and fires spikes stochastically based on a nonlinear transfer function. The membrane potentials are modeled as stationary stochastic processes with a specific covariance structure sampled from a Wishart distribution. This setup induces "weak correlations" between neurons, a state frequently observed in biological neural networks.

Linear Readout and Mathematical Proof

A downstream population recovers the signal using a linear readout, which is essentially a weighted sum of the incoming spikes. The core of the proof relies on concentration of measure in high dimensions. By performing a bias-variance decomposition of the reconstruction error, the author proves that as the population size ($N$) and the dimensionality of the signal ($P$) grow, the error vanishes. To validate the theory, synthetic membrane potential data were generated using Stochastic Differential Equations (SDEs) to simulate stationary Gaussian processes, with simulations involving up to $N = 1,000,000$ neurons.

Key Results

The primary theoretical contribution is a formal proof that the Mean Squared Error (MSE) of signal recovery approaches zero in the limit where $N \to \infty$ and $P \to \infty$, provided the signal dimensionality $P$ grows slower than the population size $N$ (specifically $P/N \to 0$).

Figure 1: Illustration of subthreshold potential reconstruction. Spikes from a population of neurons (left) are combined via a linear readout to reconstruct the original membrane potential fluctuations (right) with high fidelity.

Simulations demonstrate that even with sparse spiking (average firing rates of ~1 Hz), a linear readout can track the subthreshold fluctuations of presynaptic neurons with near-exact precision. Furthermore, the synaptic weight scaling required for this recovery ($N^{-0.59}$) aligns remarkably well with empirical measurements from real cortical neuron cultures.

Discussion: Insights into Neural Communication

Several surprising insights emerge from this high-dimensional analysis:

  • Nonlinearity Becomes Immaterial: In high dimensions, the specific shape of the spiking threshold—whether a sharp step or a smooth curve—does not prevent the downstream population from recovering the underlying analog signal.
  • The Power of Weak Correlations: While correlations are often viewed as noise, this work shows that the specific structure of weak correlations is the mechanism allowing subthreshold signals to be distributed across a population and reconstructed.
  • Concentration Phenomena: The result hinges on the fact that in high-dimensional spaces, random variables concentrate around their expectation. The aggregate "noise" of millions of individual spiking neurons cancels out, leaving the pure subthreshold signal.

Limitations and Future Directions

While the theory is robust, it faces biological constraints. The current proof assumes "dense" connectivity, whereas real brain connections are sparse. Adapting the theorem to sparse matrices remains a mathematical challenge. Additionally, the model does not yet account for Dale’s Law, which dictates that neurons are typically either purely excitatory or inhibitory. Future experimental validation could involve testing the prediction that membrane potentials are more accurately recoverable from spikes when the brain is in a "desynchronized" (high-dimensional) state compared to a "synchronized" (low-dimensional) state.

Summary updated

Abstract

Neurons primarily communicate through the emission of action potentials, or spikes. To generate a spike, a neuron's membrane potential must cross a defined threshold. Does this spiking mechanism inherently prevent neurons from transmitting their subthreshold membrane potential fluctuations to other neurons? We prove that, in theory, it does not. The subthreshold membrane potential fluctuations of a presynaptic population of spiking neurons can be perfectly transmitted to a downstream population of neurons. Mathematically, this surprising result is an example of concentration phenomenon in high dimensions.
BibTeX
@misc{schmutz2025spikes,
  author = {Valentin Schmutz},
  title = {{Spikes can transmit neurons' subthreshold membrane potentials}},
  year = {2025},
  eprint = {2501.13845},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2501.13845},
  howpublished = {arXiv},
}
Figure