Hybrid Scheme for Modeling Local Field Potentials from Point-Neuron Networks

Espen Hagen, David Dahmen, Maria Stavrinou, Henrik Lindén, Tom Tetzlaff, Sacha J. van Albada, Sonja Grün, Markus Diesmann, Gaute T. Einevoll

TL;DR

This paper introduces a novel "hybrid" modeling scheme that bridges the gap between efficient, large-scale point-neuron network simulations and biophysically detailed models of local field potential (LFP) generation. This approach allows researchers to simulate complex neural activity and predict realistic LFP signals, opening new avenues for understanding brain function.

What Problem Does This Paper Address?

The local field potential (LFP) is a crucial signal for understanding brain activity, but its interpretation is challenging due to the vast number of neurons contributing to it. While simplified "point-neuron" models are excellent for simulating network dynamics (like spiking activity), they inherently cannot generate LFPs because they lack spatial extent. Conversely, detailed "multicompartment" neuron models can accurately capture LFP generation but are computationally too expensive for large-scale network simulations. This creates a dilemma: either sacrifice biological realism for computational tractability or vice-versa. This paper addresses this by proposing a hybrid approach that leverages the strengths of both modeling paradigms.

Methods

The core of the paper is the "hybrid LFP modeling scheme." It works by:

  1. Simulating Network Dynamics with Point Neurons: A large-scale network of simplified "leaky integrate-and-fire" (LIF) point neurons is used to simulate the spiking activity of the neural circuit. This is computationally efficient and allows for large network sizes.
  2. Generating LFPs with Biophysically Detailed Neurons: The spike trains from the point-neuron network are then used as input to a separate set of "multicompartment" neuron models. These models have detailed morphologies and biophysical properties that are essential for accurately generating LFP signals.
  3. Separation of Concerns: Crucially, the spiking dynamics of the point-neuron network do not influence the LFP generation process, and vice-versa. This separation allows for independent investigation of network dynamics and LFP generation.
  4. Forward Modeling: The transmembrane currents generated by the multicompartment neurons are then translated into extracellular potentials (LFPs) using established volume conduction theory.

The authors also introduce a Python package, hybridLFPy, to implement this scheme, making it accessible to other researchers.

Datasets

The paper uses a well-established model of a cortical microcircuit from Potjans and Diesmann (2014) as its primary dataset. This model is based on anatomical data from cat visual cortex. No new datasets were introduced.

Key Insights and Ideas

  • Bridging the Simulation Gap: The hybrid scheme effectively bridges the computational gap between large-scale network simulations and biophysically realistic LFP generation. This is a significant step towards more comprehensive neural modeling.
  • Importance of Spatial Detail for LFPs: The paper strongly emphasizes that accurate LFP prediction requires spatially extended neuron models, as simple point neurons cannot capture the necessary transmembrane current distributions.
  • Role of Synaptic Correlations: The study highlights that synaptic input correlations (both shared input and spike-train correlations) play a critical role in shaping LFP signals and cannot be ignored for accurate predictions, especially in large networks.
  • Computational Efficiency: The "embarrassingly parallel" nature of the LFP generation step (due to the lack of interaction between multicompartment neurons) makes the hybrid scheme computationally feasible even for large networks, requiring significantly fewer computational resources than full multicompartment simulations.

Key Results

  • The hybrid scheme was successfully applied to a ~1 mm² patch of cat visual cortex, simulating approximately 78,000 neurons.
  • The model successfully predicted laminar LFPs for different network states, including spontaneous activity and stimulus-evoked responses.
  • The study demonstrated that LFP amplitudes for stimulus-evoked activity (~1 mV) were an order of magnitude larger than for spontaneous activity (~0.1 mV), aligning with experimental observations.
  • The authors showed that LFP predictions are sensitive to neuron density and synaptic input correlations, underscoring the need for realistic network models.
  • The hybridLFPy software package was released, providing a tool for other researchers to use the hybrid scheme.

Limitations and Future Work

  • Passive Dendrites: The multicompartment neuron models used in this study were passive, meaning they lacked active conductances. The authors acknowledge that active conductances can influence LFPs and suggest this as an area for future exploration.
  • Simplified Synaptic Connectivity: While the scheme allows for arbitrary connectivity, the specific implementation used simplified synapse placement (layer-specific) and did not account for all possible dendritic locations.
  • Soma Positioning: For simplicity, all neurons within a population were assigned the same cortical depth. More realistic distributions of soma depths could lead to smoother CSD profiles.
  • Limited Morphologies: The study used a limited number of reconstructed morphologies, partly due to availability. Future work could incorporate a wider variety of morphologies.
  • Future Work: The authors suggest extending the scheme to model other measurement modalities like ECoG, EEG, and MEG, as well as incorporating more complex network dynamics (e.g., firing rate models) and larger-scale brain circuits. They also propose investigating the impact of active conductances and more detailed synaptic connectivity.

Summary updated

Abstract

Due to rapid advances in multielectrode recording technology, the local field potential (LFP) has again become a popular measure of neuronal activity in both basic research and clinical applications. Proper understanding of the LFP requires detailed mathematical modeling incorporating the anatomical and electrophysiological features of neurons near the recording electrode, as well as synaptic inputs from the entire network. Here we propose a hybrid modeling scheme combining the efficiency of commonly used simplified point-neuron network models with the biophysical principles underlying LFP generation by real neurons. The scheme can be used with an arbitrary number of point-neuron network populations. The LFP predictions rely on populations of network-equivalent, anatomically reconstructed multicompartment neuron models with layer-specific synaptic connectivity. The present scheme allows for a full separation of the network dynamics simulation and LFP generation. For illustration, we apply the scheme to a full-scale cortical network model for a $\sim$1 mm$^2$ patch of primary visual cortex and predict laminar LFPs for different network states, assess the relative LFP contribution from different laminar populations, and investigate the role of synaptic input correlations and neuron density on the LFP. The generic nature of the hybrid scheme and its publicly available implementation in \texttt{hybridLFPy} form the basis for LFP predictions from other point-neuron network models, as well as extensions of the current application to larger circuitry and additional biological detail.
BibTeX
@inproceedings{hagen2016hybrid,
  author = {Espen Hagen and David Dahmen and Maria Stavrinou and Henrik Lindén and Tom Tetzlaff and Sacha J. van Albada and Sonja Grün and Markus Diesmann and Gaute T. Einevoll},
  title = {{Hybrid Scheme for Modeling Local Field Potentials from Point-Neuron Networks}},
  booktitle = {Cerebral},
  year = {2016},
  doi = {10.1093/cercor/bhw237},
}
Figure