Multi-Scale Spiking Network Model of Human Cerebral Cortex

Jari Pronold, Alexander van Meegen, Hannah Vollenbröker, Renan O. Shimoura, Mario Senden, Claus C. Hilgetag, Rembrandt Bakker, Sacha J. van Albada

Abstract

Although the structure of cortical networks provides the necessary substrate for their neuronal activity, the structure alone does not suffice to understand it. Leveraging the increasing availability of human data, we developed a multi-scale, spiking network model of human cortex to investigate the relationship between structure and dynamics. In this model, each area in one hemisphere of the Desikan-Killiany parcellation is represented by a 1 mm² column with a layered structure. The model aggregates data across multiple modalities, including electron microscopy, electrophysiology, morphological reconstructions, and DTI, into a coherent framework. It predicts activity on all scales from the single-neuron spiking activity to the area-level functional connectivity. We compared the model activity against human electrophysiological data and human resting-state fMRI data. This comparison reveals that the model can reproduce aspects of both spiking statistics and fMRI correlations if the cortico-cortical connections are sufficiently strong. Furthermore, we show that a single-spike perturbation propagates through the network within a time close to the limit imposed by the delays.

Introduction: Bridging the Scale Gap

Neuroscience currently suffers from a "scale gap." While we have high-resolution data on individual neurons (micro-scale) and whole-brain activity via fMRI (macro-scale), a coherent framework to connect them has been missing. Most large-scale brain models use "mean-field" approximations—essentially treating brain areas as smooth pools of activity rather than collections of individual pulsing cells.

The challenge is that the brain’s behavior isn't just a sum of its parts; it emerges from the specific, layered way neurons are wired together. To truly understand how a thought or a perturbation (like a single spike) ripples through the cortex, we need models that are "cellularly resolved," meaning every neuron and every synapse is accounted for.

Methods: A Multi-Scale Architecture

The researchers built a model of one human hemisphere divided into 34 areas based on the Desikan-Killiany parcellation.

Overview of the multi-scale model architecture, showing the columnar unit, layered structure, and the integration of local and cortico-cortical connectivity.
  • The Columnar Unit: Each area is represented by a 1 mm² column with a realistic layered structure (Layers II/III, IV, V, and VI).
  • Neuron Modeling: They used Leaky Integrate-and-Fire (LIF) neurons, with parameters derived from the Allen Cell Types Database.
  • Connectivity (The "Mesoconnectome"): This is the model's core innovation. It combines local wiring blueprints, long-range strengths from Diffusion Tensor Imaging (DTI), and predicted laminar specificity based on macaque tracing rules.
  • Simulation: The model was run using the NEST simulator on the JURECA-DC supercomputer, simulating 3.47 million neurons and 42.8 billion synapses.

Key Insights: From Connectivity to Dynamics

A critical finding was the necessity of scaling up cortico-cortical connection strengths. The initial "ground state" failed to replicate human fMRI patterns; only after systematically strengthening long-range connections did the model exhibit modular functional networks, such as the Default Mode Network (DMN), seen in real humans.

Integration of connectivity data: (A) Intrinsic connectivity, (B) DTI-based cortico-cortical connectivity, (C) Synapse-to-soma probability, (D) SLN fit from macaque data, (E) Predicted human SLN, and (F) The resulting mesoconnectome.

The model also demonstrated remarkable stability, suggesting that human-specific neuron densities and wiring proportions contribute to a robust dynamic equilibrium. Furthermore, single-spike perturbation analysis showed that adding just one extra spike to a neuron in the visual cortex could cause a ripple effect reaching the entire hemisphere in under 50 milliseconds, moving at speeds limited primarily by axonal delays.

Key Results

  • Spiking Statistics: The model matched the irregular firing patterns of real human neurons (Coefficient of Variation $\approx$ 0.8–1.0).
  • Functional Connectivity: The model achieved a correlation of 0.47 with empirical fMRI data, successfully capturing major resting-state networks.
  • Propagation Speed: A single-spike perturbation propagated through the network with a mean time of 25.1 ms.

Limitations and Future Work

While groundbreaking, the model has clear boundaries. It currently includes only two classes of neurons (excitatory and inhibitory), omitting the diversity of specialized interneurons. It also lacks subcortical structures like the thalamus and does not yet incorporate synaptic plasticity or neuromodulation. Discrepancies in firing rates in certain layers suggest that local wiring blueprints still need refinement for human-specific architecture.

Summary updated

Abstract

Abstract Although the structure of cortical networks provides the necessary substrate for their neuronal activity, the structure alone does not suffice to understand it. Leveraging the increasing availability of human data, we developed a multi-scale, spiking network model of human cortex to investigate the relationship between structure and dynamics. In this model, each area in one hemisphere of the Desikan-Killiany parcellation is represented by a 1 mm 2 column with a layered structure. The model aggregates data across multiple modalities, including electron microscopy, electrophysiology, morphological reconstructions, and DTI, into a coherent framework. It predicts activity on all scales from the single-neuron spiking activity to the area-level functional connectivity. We compared the model activity against human electrophysiological data and human resting-state fMRI data. This comparison reveals that the model can reproduce aspects of both spiking statistics and fMRI correlations if the cortico-cortical connections are sufficiently strong. Furthermore, we show that a single-spike perturbation propagates through the network within a time close to the limit imposed by the delays.
BibTeX
@inproceedings{pronold2024multi,
  author = {Jari Pronold and Alexander van Meegen and Hannah Vollenbröker and Renan O. Shimoura and Mario Senden and Claus C. Hilgetag and Rembrandt Bakker and Sacha J. van Albada},
  title = {{Multi-Scale Spiking Network Model of Human Cerebral Cortex}},
  booktitle = {Cerebral},
  year = {2024},
  doi = {10.1093/cercor/bhae409},
}
Figure