Inhibitory connections enhance pattern recurrence in networks of neocortical pyramidal cells

R. Stoop, Kaspar Schindler, Leonid BunimovichPhysics 1999 · DOI

TL;DR

This paper proposes a novel "quasistatic" model of neocortical networks where inhibitory connections, surprisingly, enhance pattern recurrence and synchronization, rather than causing desynchronization. By using experimentally derived "phase response functions" from rat neurons, they demonstrate that local chaotic firing patterns induced by inhibition can lead to globally coherent network activity, challenging previous assumptions about the role of inhibition.

What Problem Does This Paper Address?

Understanding how the brain processes information and achieves synchronization is a fundamental challenge. While spiking rates have been a common descriptive tool, they fall short in explaining complex phenomena like synchronization. Recent findings suggest that interspike distributions might follow power laws, hinting at more nuanced coding mechanisms like phase-coding. This paper tackles two key questions: 1) Why does the brain utilize both excitatory and inhibitory connections? and 2) What is the relationship between synchronization and the postulated chaos in the brain? Specifically, they investigate how inhibitory connections, which can induce chaotic firing patterns in individual neurons, influence the overall network behavior and its ability to synchronize.

Methods

The authors introduce a "quasistatic" network model, which simplifies the complex dynamics of biological neurons by assuming "frozen" firing rates and constant excitability over the observation period. This allows them to focus on information encoded in phase properties rather than firing rates.

The core of their method involves:

  • Experimental Phase Response Functions (g-functions): They derive these functions from experimental data on rat neocortical pyramidal cells. These functions describe how a neuron's firing phase is perturbed by external stimuli. They distinguish between inhibitory and excitatory stimulation, noting that inhibition can lead to chaotic firing patterns while excitation does not.
  • Poincaré Return Maps (P-maps): Using the experimental g-functions, they construct discrete maps that predict the phase of the next spike given the phase of the previous perturbation. These P-maps are analogous to circle maps, a well-studied concept in nonlinear dynamics.
  • Network Simulation: They build a lattice of these P-maps representing a network of neurons. This lattice incorporates different scales of input: strong (periodic), medium (diffusive coupling), and weak (decorrelated). They simulate networks with varying percentages of inhibitory connections and coupling strengths.
  • Pattern Recurrence Measurement: To quantify network coherence, they use a metric $\Delta$, which measures the minimum deviation of network states over time.

Datasets

The paper utilizes experimental data from slices of rat brain in an in vitro preparation, specifically focusing on neocortical pyramidal cells. This data is used to derive the phase response functions (g-functions) for both inhibitory and excitatory stimulation. No new benchmarks are introduced; rather, existing experimental data is re-analyzed through a novel modeling framework.

Key Insights and Ideas

  • Inhibition Enhances Synchronization: Contrary to intuition, inhibitory connections are found to be crucial for enhancing pattern recurrence and synchronization in the network. This suggests a more complex role for inhibition than simply disrupting activity.
  • Local Chaos, Global Coherence: Inhibitory stimulation can induce chaotic firing patterns at the individual neuron level (local chaos). However, this local chaos, when integrated within the network, can paradoxically lead to globally synchronized and periodic network behavior.
  • Phase-Coding is Key: The model emphasizes the importance of phase-coding for information processing and synchronization, moving beyond simple firing rate descriptions.
  • "Quasistatic" Model Utility: The simplified "quasistatic" model, by focusing on phase dynamics and experimentally derived response functions, proves effective in capturing essential network behaviors without the computational burden of full differential equation simulations.
  • Generic Properties: The observed phenomena are argued to be "generic" for this type of neuronal system, not artifacts of specific experimental conditions, due to the universality principles of the circle map class.

Key Results

  • Inhibitory Site Maps Generate Chaotic Patterns: The paper demonstrates that inhibitory site maps can generate chaotic spike patterns on a significant parameter set.
  • Pattern Recurrence Driven by Inhibition: Simulations show that inhibitory connections are primarily responsible for enhanced pattern recurrence in quasistatic networks.
  • Optimal Coupling for Coherence: The network exhibits optimal performance (pattern recurrence) at a specific, non-zero coupling strength ($k_2 \approx 0.83$ for $p=0.2$), suggesting a balance between local dynamics and network interaction.
  • Transition from Local to Global Chaos: As coupling strength increases, the network transitions from local chaos (where some sites are periodic and others chaotic) to a "turbulent" state (global chaos), where periodicities break down.
  • Phase-Coincidence Detection: Imposing phase-coincidence detection on the network leads to localized "coding sites" that can discriminate between input patterns, suggesting a mechanism for information processing.

Limitations and Future Work

  • Quasistatic Approximation: The model assumes "frozen" firing rates, which is a simplification. The authors acknowledge that real biological networks involve dynamic changes in firing rates.
  • Experimental Data Limitations: While the experimental data is extensive, it's from specific preparations (rat neocortical pyramidal cells). Generalizability to other neuron types or brain regions may require further investigation.
  • "Turbulent" State: The paper notes that above a certain coupling strength, the network enters a "turbulent" state where it can no longer faithfully return patterns. Understanding this regime and its implications is an area for future work.
  • Hardware Implementation: The authors suggest that these findings have implications for hardware implementation of computational and cognitive tasks, indicating a direction for future research.
  • Dynamic Connectivity: The paper briefly touches upon dynamic connectivity (Hebbian learning) and its potential for temporal pattern storage, suggesting this as a promising avenue for further exploration.

Summary updated

BibTeX
@inproceedings{stoop1999inhibitory,
  author = {R. Stoop and Kaspar Schindler and Leonid Bunimovich},
  title = {{Inhibitory connections enhance pattern recurrence in networks of neocortical pyramidal cells}},
  booktitle = {Physics},
  year = {1999},
  doi = {10.1016/s0375-9601(99)00262-5},
}
Figure