Inhibitory cell type heterogeneity in a spatially structured mean-field model of V1

Soon Ho Kim, Hannah ChoibioRxiv 2025 · DOI

Abstract

Inhibitory interneurons in the cortex are classified into cell types differing in their morphology, electrophysiology, and connectivity. Although it is known that parvalbumin (PV), somatostatin (SST), and vasoactive intestinal polypeptide-expressing neurons (VIP), the major inhibitory neuron subtypes in the cortex, have distinct modulatory effects on excitatory neurons, how heterogeneous spatial connectivity properties relate to network computations is not well understood. Here, we study the implications of heterogeneous inhibitory neurons on the dynamics and computations of spatially-structured neural networks. We develop a mean-field model of the system in order to systematically examine excitation-inhibition balance, dynamical stability, and cell-type specific gain modulations. The model incorporates three inhibitory cell types and excitatory neurons with distinct connectivity probabilities and recent evidence of long-range spatial projections of SST neurons. Position-dependent firing rate predictions are validated against simulations, and balanced solutions under Gaussian assumptions are derived from scaling arguments. Stability analysis shows that while long-range inhibitory projections in E-I circuits with a homogeneous inhibitory population result in instability, the heterogeneous network maintains stability with long-range SST projections. This suggests that a mixture of short and long-range inhibitions may be key to providing diverse computations while maintaining stability. We further find that conductance-based synaptic transmissions are necessary to reproduce experimentally observed cell-type-specific gain modulations of inhibition by PV and SST neurons. These gain modulations are distance-dependent, and a linear response analysis suggested that shifts in the excitation-inhibition balance of the network underlie these effects. Our theoretical approach offers insight into the computational function of cell-type-specific and distance-dependent network structure.

Introduction

Theoretical neuroscience has long grappled with a "stability-diversity" paradox. Standard neural network models often assume inhibitory populations are homogeneous. However, in these simplified models, if inhibitory connections reach further than excitatory ones—a common biological feature—the network becomes mathematically unstable, leading to unrealistic runaway activity. In the mouse visual cortex (V1), Somatostatin (SST) neurons are known to have long-range projections that extend significantly further than local excitatory connections. This paper investigates how the brain utilizes these long-range "brakes" without sacrificing stability and how distinct cell types contribute to complex computations like gain modulation.

Methods: Spatially Structured Mean-Field Modeling

The researchers developed a spatially structured mean-field model representing neurons arranged on a 1D ring, simulating a cortical layer. The model incorporates four primary cell types: Excitatory (E) neurons and three inhibitory subtypes—Parvalbumin (PV), Somatostatin (SST), and Vasoactive Intestinal Polypeptide (VIP).

Circuit architecture and spatial connectivity kernels for the E-PV-SST-VIP network.

Connectivity is defined by Gaussian spatial kernels where the probability of connection decays with distance. Crucially, the model assigns SST neurons a wider kernel ($\sigma_s = 0.2$) compared to PV or VIP neurons ($\sigma_p = \sigma_v = 0.1$). The authors employed Fokker-Planck equations to derive stationary firing rates and linear response theory to analyze network perturbations. These mean-field predictions were validated against large-scale simulations of individual Leaky Integrate-and-Fire (LIF) spiking neurons.

Key Results

Heterogeneity as a Stabilizer

A central finding is that cellular heterogeneity is the key to maintaining network stability. In a homogeneous inhibitory population (E-I circuit), long-range projections typically cause the system to crash when inhibitory reach exceeds excitatory reach ($\sigma_i > \sigma_e$). However, the heterogeneous model (E-P-S-V circuit)—mixing short-range PV and long-range SST inhibition—remains stable even when the SST reach is double that of the excitatory neurons.

Stability analysis comparing homogeneous E-I circuits with heterogeneous E-P-S-V circuits.

Validation and Gain Modulation

The mean-field model's firing rate predictions closely matched spiking simulations, with the normalized mean squared error (NMSE) decreasing as the network size ($N$) increased toward the thermodynamic limit. Furthermore, the study found that conductance-based synaptic transmissions are necessary to replicate the "divisive" gain control observed experimentally. Unlike simple current-based models that merely subtract from input, conductance-based models allow inhibition to scale the slope of a neuron's response curve based on membrane voltage. The model explains that distal SST activation causes a divisive shift in excitatory responses, a critical mechanism for processing objects against complex backgrounds.

Discussion

While the 1D ring model provides significant insights, it remains an abstraction of the 3D cortical structure. The model also treats neurons as point sources, omitting the complexities of dendritic integration. Future research could expand these findings into 2D spatial structures or incorporate more detailed biophysical features like synaptic plasticity to understand how these stable, balanced states emerge during development. This work provides a vital theoretical bridge between biological diversity and mathematical models of cortical computation.

Summary updated

Abstract

Inhibitory interneurons in the cortex are classified into cell types differing in their morphology, electrophysiology, and connectivity. Although it is known that parvalbumin (PV), somatostatin (SST), and vasoactive intestinal polypeptide-expressing neurons (VIP), the major inhibitory neuron subtypes in the cortex, have distinct modulatory effects on excitatory neurons, how heterogeneous spatial connectivity properties relate to network computations is not well understood. Here, we study the implications of heterogeneous inhibitory neurons on the dynamics and computations of spatially-structured neural networks. We develop a mean-field model of the system in order to systematically examine excitation-inhibition balance, dynamical stability, and cell-type specific gain modulations. The model incorporates three inhibitory cell types and excitatory neurons with distinct connectivity probabilities and recent evidence of long-range spatial projections of SST neurons. Position-dependent firing rate predictions are validated against simulations, and balanced solutions under Gaussian assumptions are derived from scaling arguments. Stability analysis shows that while long-range inhibitory projections in E-I circuits with a homogeneous inhibitory population result in instability, the heterogeneous network maintains stability with long-range SST projections. This suggests that a mixture of short and long-range inhibitions may be key to providing diverse computations while maintaining stability. We further find that conductance-based synaptic transmissions are necessary to reproduce experimentally observed cell-type-specific gain modulations of inhibition by PV and SST neurons. These gain modulations are distance-dependent, and a linear response analysis suggested that shifts in the excitation-inhibition balance of the network underlie these effects. Our theoretical approach offers insight into the computational function of cell-type-specific and distance-dependent network structure.
BibTeX
@misc{kim2025inhibitory,
  author = {Soon Ho Kim and Hannah Choi},
  title = {{Inhibitory cell type heterogeneity in a spatially structured mean-field model of V1}},
  year = {2025},
  doi = {10.1101/2025.03.13.643046},
  howpublished = {bioRxiv},
}
Figure