Mesoscopic description of hippocampal replay and metastability in spiking neural networks with short-term plasticity

Bastian Pietras, Valentin Schmutz, Tilo Schwalger

Abstract

Bottom-up models of functionally relevant patterns of neural activity provide an explicit link between neuronal dynamics and computation. A prime example of functional activity pattern is hippocampal replay, which is critical for memory consolidation. The switchings between replay events and a low-activity state in neural recordings suggests metastable neural circuit dynamics. As metastability has been attributed to noise and/or slow fatigue mechanisms, we propose a concise mesoscopic model which accounts for both. Crucially, our model is bottom-up: it is analytically derived from the dynamics of finite-size networks of Linear-Nonlinear Poisson neurons with short-term synaptic depression. As such, noise is explicitly linked to spike noise and network size, and fatigue is explicitly linked to synaptic dynamics. To derive the mesosocpic model, we first consider a homogeneous spiking neural network and follow the temporal coarse-graining approach of Gillespie ("chemical Langevin equation"), which can be naturally interpreted as a stochastic neural mass model. The Langevin equation is computationally inexpensive to simulate and enables a thorough study of metastable dynamics in classical setups (population spikes and Up-Down states dynamics) by means of phase-plane analysis. This stochastic neural mass model is the basic component of our mesoscopic model for replay. We show that our model faithfully captures the stochastic nature of individual replayed trajectories. Moreover, compared to the deterministic Romani-Tsodyks model of place cell dynamics, it exhibits a higher level of variability in terms of content, direction and timing of replay events, compatible with biological evidence and could be functionally desirable. This variability is the product of a new dynamical regime where metastability emerges from a complex interplay between finite-size fluctuations and local fatigue.

Introduction

In neuroscience, neural populations often exhibit "metastability," characterized by long periods in a stable state (e.g., a low-activity "Down" state) punctuated by sudden transitions to another state (e.g., a high-activity "Up" state or traveling waves). A prominent example is hippocampal replay, where sequences of place cells fire in a specific order to re-simulate paths taken during wakefulness, a process vital for memory consolidation.

Modeling these dynamics presents a significant challenge. Microscopic models that track individual spikes are computationally prohibitive and mathematically opaque. Conversely, macroscopic mean-field models often assume infinite network sizes, thereby averaging out the stochastic spiking noise inherent in biological systems. This paper bridges this gap by proposing a mesoscopic description that retains the essential noise of finite-sized networks while remaining computationally efficient and analytically tractable.

Methodology

The authors derive a bottom-up mesoscopic model starting from a microscopic network of Linear-Nonlinear Poisson (LNP) neurons incorporating short-term plasticity (STP), specifically synaptic depression.

Derivation of a mesoscopic population model from a microscopic spiking network with short-term plasticity (STP).

To transition from thousands of individual neuronal equations to a concise population-level description, the following techniques were employed:

  • Temporal Coarse-Graining: Utilizing the Gillespie approach from chemical kinetics, the authors derived a Chemical Langevin Equation. This results in a stochastic neural mass model where noise is explicitly linked to the network size ($N$) and spiking rate.
  • Jump-Diffusion Model: For small networks where Gaussian approximations fail, a hybrid model combining Poisson shot noise with Gaussian diffusion was developed to capture the "jumpy" nature of discrete spikes.
  • Phase-Plane Analysis: The low-dimensional nature of the mesoscopic model allows for geometric analysis of the "separatrix," identifying the tipping points where noise-induced fluctuations trigger state transitions.

Validation was performed against high-fidelity microscopic simulations and synthetic ring attractor networks, with results compared to rodent experimental benchmarks.

Key Insights and Results

The model reveals that in a "fluctuation-driven" regime, replay events are triggered by random spike clusters in finite networks, rather than purely by deterministic fatigue mechanisms. A key finding is the $1/N$ scaling, which mathematically links neuron count to fluctuation intensity; smaller networks exhibit louder "noise," leading to more frequent state transitions.

The mesoscopic model demonstrates high statistical fidelity, matching power spectra and inter-spike interval (ISI) distributions of microscopic simulations even for $N=30$. Furthermore, it successfully replicates Sharp Wave Ripples (SWRs) and captures the high variability in replay content, direction, and timing observed in biological data—variability that deterministic models like the Romani-Tsodyks model fail to reproduce.

Discussion and Future Work

While the model provides a powerful tool for studying metastable dynamics, it currently omits neuronal refractoriness, which would alter noise statistics. Future work could also explore more complex, sparse connectivity patterns and incorporate synaptic stochasticity (vesicle noise) to further refine the description of metastability. By mathematically grounding the reduction of complex neural models, this work offers a new framework for understanding how finite-size noise drives fundamental cognitive processes like memory replay.

Summary updated

Abstract

Bottom-up models of functionally relevant patterns of neural activity provide an explicit link between neuronal dynamics and computation. A prime example of functional activity pattern is hippocampal replay, which is critical for memory consolidation. The switchings between replay events and a low-activity state in neural recordings suggests metastable neural circuit dynamics. As metastability has been attributed to noise and/or slow fatigue mechanisms, we propose a concise mesoscopic model which accounts for both. Crucially, our model is bottom-up: it is analytically derived from the dynamics of finite-size networks of Linear-Nonlinear Poisson neurons with short-term synaptic depression. As such, noise is explicitly linked to spike noise and network size, and fatigue is explicitly linked to synaptic dynamics. To derive the mesosocpic model, we first consider a homogeneous spiking neural network and follow the temporal coarse-graining approach of Gillespie ("chemical Langevin equation"), which can be naturally interpreted as a stochastic neural mass model. The Langevin equation is computationally inexpensive to simulate and enables a thorough study of metastable dynamics in classical setups (population spikes and Up-Down states dynamics) by means of phase-plane analysis. This stochastic neural mass model is the basic component of our mesoscopic model for replay. We show that our model faithfully captures the stochastic nature of individual replayed trajectories. Moreover, compared to the deterministic Romani-Tsodyks model of place cell dynamics, it exhibits a higher level of variability in terms of content, direction and timing of replay events, compatible with biological evidence and could be functionally desirable. This variability is the product of a new dynamical regime where metastability emerges from a complex interplay between finite-size fluctuations and local fatigue.
BibTeX
@article{pietras2022mesoscopic,
  author = {Bastian Pietras and Valentin Schmutz and Tilo Schwalger},
  title = {{Mesoscopic description of hippocampal replay and metastability in spiking neural networks with short-term plasticity}},
  journal = {PLOS Computational Biology},
  year = {2022},
  doi = {10.1371/journal.pcbi.1010809},
}
Figure