Neural field dynamics under variation of local and global connectivity and finite transmission speed

https://doi.org/10.1016/j.physd.2009.09.014Get rights and content

Abstract

Spatially continuous networks with heterogeneous connections are ubiquitous in biological systems, in particular neural systems. To understand the mutual effects of locally homogeneous and globally heterogeneous connectivity, we investigate the stability of the steady state activity of a neural field as a function of its connectivity. The variation of the connectivity is implemented through manipulation of a heterogeneous two-point connection embedded into the otherwise homogeneous connectivity matrix and by variation of the connectivity strength and transmission speed. Detailed examples including the Ginzburg–Landau equation and various other local architectures are discussed. Our analysis shows that developmental changes such as the myelination of the cortical large-scale fiber system generally result in the stabilization of steady state activity independent of the local connectivity. Non-oscillatory instabilities are shown to be independent of any influences of time delay.

Introduction

Large-scale neural network models are thought to be involved in the implementation of the cognitive function of the brain [1], [2], [3], [4], [5], [6], [7], [8], [9], [10], [11]. In particular, peripheral areas are more functionally differentiated, whereas the ensuing cognitive integration appears to require more global network activations. The properties of the global network dynamics will then naturally depend on the large-scale, i.e. global, connectivity of the network components, as well as their local connectivity and dynamics [12], [13], [9], [14], [15]. So far, theoretical efforts have focussed almost exclusively on the study of networks with discretely connected nodes and complicated connectivity, but simple local dynamics, in which all neural activity is lumped into a single neural mass ([16], [17], [18], [19], [20], [21], [22]; see [23], [9] for reviews). As an additional complication, large-scale brain networks will be affected by time delays via signal transmission along the connecting pathways. The time delays may reach up to 200 ms [24] for the human brain, which is on a similar time scale as human brain function and hence is not negligible. Complementary efforts focussed on spatially continuous neural fields, which describe the temporal change of neural activity on a local spatial scale, typically within a brain area ([25], [26], [27], [28], [29], [30]; see [23], [9], [31] for reviews). These approaches use a translationally invariant, so-called homogeneous, connectivity and also take time delays into consideration, which, however, given the small spatial extent of a brain area, play a lesser role. To understand the neural basis of cognition, theoretical and analytical means must be developed which are specifically targeted to the properties of large-scale network dynamics integrating local and global contributions. Existing attempts so far include neural field theories which approximate the large-scale components of the connectivity matrix as translationally invariant and exponentially decaying over space, but on a larger spatial scale than is biologically realistic [32], [33], [34], [35], [36], [37]. These approaches have been successful in capturing various phenomena of large-scale brain dynamics including characteristic electroencephalographic power spectra [32], [36], epilepsy [38] and magnetoencephalographic activity during sensorimotor coordination [34], [35]. No systematic investigation, however, has been performed so far testing the validity of the translationally invariant approximation, and this remains an exciting challenge in theoretical neuroscience. Only very recently have computational and theoretical studies been performed exploring the effects of a large-scale connecting pathways upon the local dynamics of coupled brain areas, in particular neural fields, and vice versa; that is, the contributions of local architectures to the entire network dynamics. The first systematic but mostly computational study was performed by Jirsa and Kelso [39], who introduced the notion of a two-point connection embedded in a neural field and studied the spatiotemporal bifurcations as the length of the two-point connection was varied. Later research demonstrated analytically how the stability of the equilibrium point of the neural field depends on the length and coupling strength of the two-point connection [40], [41]. The implementation of multiple two-point connections obeying power-law and small-world distributions in neural fields (though with no time delays) has been shown computationally to result in a rich spatiotemporal dynamics [42], [43]. Other approaches mimicking heterogeneity in neural fields other than connectivity include non-uniform parameter distributions [44]. The equilibrium point solution of large-scale brain networks has received particular attention when biologically realistic anatomical connectivity matrices based on white matter fibre tracts became available to theoretical neuroscientists. Sporns and colleagues [45] studied the synchronization characteristics of the macaque connectivity matrix using chaotic oscillators and no time delays in the context of the brain’s resting state. Jirsa and colleagues [46], [47], [48] demonstrated that the equilibrium point solution of a network with biologically realistic primate connectivity and time delays is of key relevance for the understanding of the emergence of the brain’s spatiotemporal resting state dynamics. Specifically, when noise drives the neural network in the vicinity of the phase space around the stable equilibrium point, then the resulting network dynamics resembles the spatiotemporal brain dynamics observed experimentally in electroencephalography (EEG), magnetoencephalograpy (MEG) and functional magnetic resonance imaging (fMRI) [46]. For these reasons it is imperative to understand the anatomical and physiological determinants that determine the stability of the network’s equilibrium point and its dynamic vicinity in phase space.

To do so, here we take the following approach: We identify a convenient mathematical representation for the network dynamics with a locally invariant (homogeneous) and globally variant (heterogeneous) architecture via a one-dimensional spatiotemporally continuous integral–differential field equation with space-dependent delay. In particular, we prove that the subsequent discussions apply to arbitrary fixed point states of the entire network dynamics. Then we discuss the specific case of diffusive coupling, which expresses the locally invariant short-range (homogeneous) connectivity, and perform a complete analysis thereof. Diffusion is biologically unrealistic, but it nicely illustrates our approach, because it reduces the discussion to the analysis of the well-known Ginzburg–Landau equation with time delay. In particular, we generalize the analysis to include distributed transmission speeds. Next, we consider various biologically more realistic local architectures and expand our stability analysis to these cases. Specifically, we discuss the developmentally important effect of non-identical transmission speeds for local and global connecting pathways.

Section snippets

Mathematical framework for the neural field dynamics with local and global connectivity

Let ψ(x,t) be the neural field capturing the population activity at time point t and position x. The dynamics of the neural field can then be described by the following integro–differential equation: ψ̇(x,t)=ϵψ(x,t)+αΓWhom(|xy|)S[ψ(y,t|xy|/c)]dy+ΓWhet(x,y)S[ψ(y,t|xy|/v)]dy where 1/ϵ>0R represents the intrinsic time scale and α>0 is a scaling constant. The dot indicates the first time derivative. The spatial domain of the neural field is denoted by Γ, where xΓ=[0,L] and L is the

Neural fields with local diffusion and global two-point heterogeneous connections

This example illustrates the mutual effects of local homogeneous and global heterogeneous connections. Consider a one-dimensional spatially continuous diffusive field whose dynamics is given by the RGLE. Additionally, a heterogeneous connection is embedded between two points of the field where the activity is transmitted from one point to another, as shown in Fig. 1. The system can then be described as follows: ψ̇(x,t)=(ε̃+D2)ψ(x,t)ψ3(x,t)+i,j=12μijδ(xxi)S[ψ(xj,t|xxj|/v)]. The

Dispersion and distributed transmission velocities

The neural field ψ(x,t) captures the neural population activity at a given location x. The axonal transmission speed depends mainly on the degree of myelination of the axons, which is diverse along the axis of the axon and will include a degree of variation. In a neural population model, the total axonal input into the population is the sum over all axonal inputs and will hence have a distributed propagation time. As shown in [21], distributed propagation times are equivalent to the distributed

Neural fields with locally general and globally heterogeneous two-point heterogeneous connectivity

In the previous sections we discussed the biologically unrealistic case that the effects of the local connectivity function are captured by a diffusion process. We now wish to generalize our discussion to various locally more realistic connectivity kernels. Additional consideration will be given to the fact that transmission speeds through the heterogeneous fiber system are generally by a factor 10 greater than the speeds in the local unmyelinated architecture. A suitable form of the

Summary and discussion

In this article we have developed a toy problem which addresses a key issue in biologically realistic large-scale networks, that is the interplay between local and global architectures leading to a large-scale network dynamics. We studied the contributions of the global connectivity by way of a two-point connection with a finite transmission speed. Our specific objective was to identify the conditions under which a fixed point solution loses its stability. We summarize the results of our

Acknowledgements

We would like to thank Felix Almonte and Arpan Banerjee for many valuable discussions and computational advice. This research was funded by the grants Brain NRG JSM22002082 and ATIP (CNRS).

References (61)

  • V.K. Jirsa et al.

    A derivation of a macroscopic field theory of the brain from the quasi-microscopic neural dynamics

    Physica D

    (1997)
  • S. Coombes et al.

    Waves and bumps in neuronal networks with axo-dendritic synaptic interactions

    Physica D

    (2003)
  • S. Grossberg

    Nonlinear neural networks: Principles, mechanisms, and architectures

    Neural Netw.

    (1988)
  • V.B. Mountcastle

    Perceptual Neuroscience: The Cerebral Cortex

    (1998)
  • M.M. Mesulam

    From sensation to cognition

    Ann. Neurol

    (1998)
  • A.R. McIntosh

    Towards a network theory of cognition

    Neural Netw.

    (2000)
  • S.L. Bressler

    Understanding cognition through large-scale cortical networks

    Curr. Dir. Psych. Sci.

    (2002)
  • S.L. Bressler

    Cortical coordination dynamics and the disorganization syndrome in schizophrenia

    Neuropsychpharmacology

    (2003)
  • V.K. Jirsa

    Connectivity and dynamics of neural information processing

    Neuroinformatics

    (2004)
  • S.L. Bressler et al.

    The role of neural context in large-scale neurocognitive network operations

  • O. Sporns et al.

    Classes of network connectivity and dynamics

    Complexity

    (2002)
  • O. Sporns

    Complex neural dynamics

  • O. Sporns et al.

    Structural determinants of functional brain dynamics

  • J.M. Beggs et al.

    Connectivity and dynamics in local cortical networks

  • F.H. Lopes da Silva et al.

    Model of brain rhythmic activity. The alpha-rhythm of the thalamus

    Kybernetik

    (1974)
  • W.J. Freeman

    Mass Action in the Nervous System

    (1975)
  • M.A. Tagamets et al.

    Integrating electrophysiological and anatomical experimental data to create a large-scale model that simulates a delayed match-to-sample human brain imaging study

    Cereb. Cortex

    (1998)
  • V.K. Jirsa et al.

    Will a large complex system with time delays be stable?

    Phys. Rev. Lett.

    (2004)
  • B. Ermentrout

    Neural networks as spatio-temporal pattern-forming systems

    Rep. Prog. Phys.

    (1998)
  • P. Nunez

    Neocortical Dynamics and Human EEG Rhythms

    (1995)
  • Cited by (13)

    • Spatiotemporal multi-resolution approximation of the Amari type neural field model

      2013, NeuroImage
      Citation Excerpt :

      This work is significant, as the ability to create patient-specific neural field models has the potential to contribute to our understanding and improve treatment of diseases resulting from abnormal neurodynamics. Other groups have previously highlighted the importance of a multi-resolution approach in neural field modeling, where the dynamics and connectivity structure differs at different spatial scales (Breakspear and Stam, 2005; Jirsa, 2004; Jirsa et al., 2002; Qubbaj and Jirsa, 2009; Schultze-Kraft et al., 2011). In order to apply estimation algorithms to neural field models a finite-dimensional state-space model must be formed.

    • Dynamic causal modeling with neural fields

      2012, NeuroImage
      Citation Excerpt :

      Here, we address this challenge using neural field models. Neural fields model current fluxes as continuous processes on the cortical manifold, using partial differential equations (PDEs) (please see Deco et al., 2008 for a review and also Atay and Hutt, 2006; Breakspear et al., 2006; Bressloff, 1996,2001; Coombes et al., 2003, 2007; Coombes, 2005; Freeman, 2003,2005; Ghosh et al., 2008a,b; Golomb and Amitai, 1997; Jirsa and Haken, 1996; Jirsa and Kelso, 2000; Jirsa, 2009; Lopes da Silva and Storm van Leeuwen, 1978; Liley et al., 2002; Nunez, 1995,1996; Nunez and Srinivasan, 2006; O'Connor and Robinson, 2005; Qubbaj and Jirsa, 2009; Robinson et al., 2001, 2002, 2003, 2004, 2005; Robinson, 2006; Rennie et al., 2000; Roberts and Robinson, 2008; Rowe et al., 2004). The key advance that neural field models offer, over conventional neural mass models, is that they embody spatial parameters (like the density and extent of lateral connections).

    • Neural fields, masses and Bayesian modelling

      2014, Neural Fields: Theory and Applications
    • Heterogeneous connectivity in neural fields: A stochastic approach

      2014, Neural Fields: Theory and Applications
    View all citing articles on Scopus
    View full text