Energetic Constraints Produce Self-sustained Oscillatory Dynamics in Neuronal Networks

Javier Burroni, Patrick Taylor, Cassian Corey, Tengiz Vachnadze, Hava T. SiegelmannFrontiers in Neuroscience 2017 · DOI

TL;DR

This paper introduces a novel computational model for spiking neurons that incorporates realistic energy constraints, demonstrating that these limitations can spontaneously generate oscillatory dynamics. This finding suggests that energy considerations are crucial for biologically realistic neural network models and could have implications for understanding brain function and designing energy-aware artificial systems.

What Problem Does This Paper Address?

Current computational models of spiking neural networks, while aiming for biological realism, largely overlook the fundamental energetic costs associated with neuronal activity. Neurons, like all biological cells, require energy to function, and this energy is not unlimited. Metabolic states significantly impact brain function and neurological diseases, yet existing models typically focus on membrane potential as the sole driver of spiking. This paper addresses the gap by exploring how direct, realistic energy costs influence the behavior of spiking neural networks, moving beyond purely biophysical dynamics to incorporate metabolic constraints.

Methods

The authors developed a new neuron model called the "Spiking Neuron Energy Pool" (SNEP). This model extends a standard Spike Response Model (SRM) by introducing a local energy pool for each neuron. Key features of the SNEP model include:

  • Energy Cost for Spiking: Each spike incurs a specific energy cost, which depletes the neuron's energy pool.
  • Energy Regeneration: Energy pools are replenished over time, either at a constant or oscillating rate.
  • Spiking Condition: A neuron can only spike if its membrane potential crosses the firing threshold and its energy pool contains sufficient resources.
  • Network Structure: The model utilizes a large-scale network structure derived from human connectome data, with connections weighted and directed.
  • Simulation Environment: A user-friendly, web-based simulation environment called "BrainPower" was developed, allowing for flexible parameter manipulation and real-time visualization.

The core innovation lies in explicitly modeling the energy pool as a critical resource that directly gates neuronal firing, in addition to membrane potential.

Datasets

The paper utilizes structural network data derived from in-vivo human connectome data (rsfMRI and DTI) as described in previous work (Nooner et al., 2012; Taylor et al., 2015). This data provides the basis for the network's connectivity matrix and region partitioning. No new benchmarks or datasets were introduced.

Key Insights and Ideas

  • Energy as a Driver of Oscillations: The most striking finding is that energy limitations, even with constant regeneration rates, can spontaneously induce self-sustained oscillatory dynamics in the neuronal network. This is a significant departure from models that only consider membrane potential.
  • Lotka-Volterra Analogy: The emergent oscillatory behavior can be effectively analyzed using Lotka-Volterra predator-prey equations, where available energy acts as the "prey" and neuronal spikes act as the "predators." This analogy provides a powerful framework for understanding the dynamics.
  • Modulation of Oscillations by Energy Cost: The energetic cost of spiking (re) directly modulates the degree and type of oscillations. Increasing this cost can lead to more synchronized activity and eventually to the breakdown of oscillations when energy becomes too scarce.
  • Harmonics and Energetic Limits: At high energy requirements, secondary components (harmonics) emerge in the power spectrum, suggesting complex interactions between the limit cycle dynamics and energy regeneration.
  • Non-linear Relationship between Energy and Activity: The relationship between spike cost and overall network activity is non-linear and concave. This implies that as energy costs increase, activity decreases, but at a diminishing rate, eventually approaching an asymptote. This suggests a potential "sweet spot" for efficient, sustained activity.

Key Results

  • Emergent Oscillations: When the energy cost per spike (re) was increased from zero, the network transitioned from random, noisy activity to synchronized, oscillatory behavior. At re = 0.227γ, a clear limit cycle emerged, and at re = 0.394γ, the system began to break down due to energy depletion.
  • Frequency-Cost Relationship: The dominant frequency of network oscillations was found to be inversely related to the spike cost (re). As re increased, the frequency decreased, and the period of oscillation increased linearly.
  • EEG-like Frequency Distributions: For a range of re values (0.006γ to 0.227γ), the network's firing rate distributions closely matched the fat-tailed geometric distributions observed in human EEG data.
  • Energy Consumption Asymptote: Average energy consumption per millisecond showed a concave relationship with re, approaching an asymptote. This suggests that beyond a certain energy cost threshold (around 0.15γ), the system reaches a state where energy dominates dynamics, and activity levels stabilize.

Limitations and Future Work

  • Abstraction of Energy: The model uses an abstract "energy pool" (γ) rather than specific biological molecules like ATP. While this allows for generalizability, it abstracts away the complex biochemical details of neuronal metabolism.
  • Fixed Network Structure and Weights: The network structure and synaptic weights were fixed and not subject to learning or plasticity, which are crucial aspects of biological neural networks.
  • Limited Exploration of Parameters: While the study systematically mapped some parameter spaces, a comprehensive exploration of all possible combinations of energy and neuronal parameters was not undertaken.
  • Future Work: The authors suggest exploring:
  • More detailed biological energy models, including specific metabolites and glia-neuron interactions.
  • Incorporating plasticity into the network to study how energy constraints affect learning.
  • Investigating the emergent oscillatory effects in other types of abstract graphs and technological applications, extending the findings beyond neuroscience.
  • Further analysis of the harmonics and their biological relevance.
  • Matching the model's parameters to specific biological conditions or diseases.

Summary updated

Abstract

<b>Overview:</b> We model energy constraints in a network of spiking neurons, while exploring general questions of resource limitation on network function abstractly. <b>Background:</b> Metabolic states like dietary ketosis or hypoglycemia have a large impact on brain function and disease outcomes. Glia provide metabolic support for neurons, among other functions. Yet, in computational models of glia-neuron cooperation, there have been no previous attempts to explore the effects of direct realistic energy costs on network activity in spiking neurons. Currently, biologically realistic spiking neural networks assume that membrane potential is the main driving factor for neural spiking, and do not take into consideration energetic costs. <b>Methods:</b> We define local energy pools to constrain a neuron model, termed Spiking Neuron Energy Pool (SNEP), which explicitly incorporates energy limitations. Each neuron requires energy to spike, and resources in the pool regenerate over time. Our simulation displays an easy-to-use GUI, which can be run locally in a web browser, and is freely available. <b>Results:</b> Energy dependence drastically changes behavior of these neural networks, causing emergent oscillations similar to those in networks of biological neurons. We analyze the system via Lotka-Volterra equations, producing several observations: (1) energy can drive self-sustained oscillations, (2) the energetic cost of spiking modulates the degree and type of oscillations, (3) harmonics emerge with frequencies determined by energy parameters, and (4) varying energetic costs have non-linear effects on energy consumption and firing rates. <b>Conclusions:</b> Models of neuron function which attempt biological realism may benefit from including energy constraints. Further, we assert that observed oscillatory effects of energy limitations exist in networks of many kinds, and that these findings generalize to abstract graphs and technological applications.
BibTeX
@inproceedings{burroni2017energetic,
  author = {Javier Burroni and Patrick Taylor and Cassian Corey and Tengiz Vachnadze and Hava T. Siegelmann},
  title = {{Energetic Constraints Produce Self-sustained Oscillatory Dynamics in Neuronal Networks}},
  booktitle = {Frontiers in Neuroscience},
  year = {2017},
  doi = {10.3389/fnins.2017.00080},
}
Figure