Review
Stepping out of the box: information processing in the neural networks of the basal ganglia

https://doi.org/10.1016/S0959-4388(01)00270-7Get rights and content

Abstract

The Albin-DeLong ‘box and arrow’ model has long been the accepted standard model for the basal ganglia network. However, advances in physiological and anatomical research have enabled a more detailed neural network approach. Recent computational models hold that the basal ganglia use reinforcement signals and local competitive learning rules to reduce the dimensionality of sparse cortical information. These models predict a steady-state situation with diminished efficacy of lateral inhibition and low synchronization. In this framework, Parkinson's disease can be characterized as a persistent state of negative reinforcement, inefficient dimensionality reduction, and abnormally synchronized basal ganglia activity.

Introduction

The basal ganglia (BG) are a complicated interconnected network of neuronal elements that process motor, cognitive and motivational (limbic) cortical information 1., 2•.. The clinical manifestations of neuronal disorders of the BG, including hypokinetic movement disorders such as Parkinson's disease (PD) and hyperkinetic movement disorders, such as Hemiballismus and Huntington's disease, suggest that the BG use this multi-dimensional cortical information to generate, or to control, action. Many computational models of the BG function have been developed (see reviews in 3., 4.). These models have generated testable hypotheses, and enable greater insights into the physiology and pathophysiology of the BG and human diseases. In this review, we use this background to construct a better understanding of normal and pathological information processing in the BG cortical circuits.

Section snippets

The classical ‘box and arrow’ view of the BG

Information processing in any neuronal system is bound by the underlying anatomical substrate. One of the first modern models (the Albin-DeLong model 5., 6.) of BG function was inspired by the dominant anatomical connections of BG nuclei and their neurochemistry. A major pathway in the BG circuitry leads from most cortical areas to the striatum. Subsequent projections link striatal neurons to the BG output stage (i.e. the globus pallidus, internal segment [GPi] and the substantia nigra, pars

The action–selection paradigm and lateral inhibition models of the basal ganglia

The assumption of separate direct–inhibitory and indirect–excitatory striato–pallidal pathways leads to two different views of the BG. The first assumes that the two pathways converge on the same pallidal neurons, therefore enabling temporal scaling of their activity. The second view assumes that the two pathways project to different populations of pallidal neurons. When actions or voluntary movements are generated by cortical mechanisms, the indirect pathway acts broadly, mainly through the

Alterations in discharge rate of BG neurons and pathophysiology of movement disorders

Despite the arguments regarding the precise nature of BG processing, the Albin-DeLong ‘box and arrow’ model has generally been accepted as the core model for BG function. The main achievement of this model lies in accounting for pathophysiological mechanisms of both hypokinetic and hyperkinetic movement disorders. The model predicts an enhanced tonic inhibition of the thalamo–cortical circuitry in hypokinetic disorders and a diminished amount of inhibition of these circuits in hyperkinetic

The BG network is more complicated than the BG models

Although many experimental findings are in agreement with the Albin-DeLong model, accumulating evidence challenges this classical view in several respects. First, neurons in the BG show extensive collateral connectivity [12•] and additional internal and external (e.g. to brainstem nuclei) projections [2•] that are incompatible with the simplified classical view. Second, recent studies indicate that: D1 and D2 receptors co-localize on striatal neurons [18••]; all striatal neurons projecting to

Sparse information is transmitted from the cortex to the BG

The mutual inhibition models and the focusing (action–selection) models predict strong lateral inhibitory interactions between BG neurons. These inhibitory processes should be characterized by inhibitory postsynaptic potentials, and by negative correlation or suppression of firing of one neuron by the firing of another neuron in multiple neuron recordings. This prediction, however, has not been borne out by physiological intracellular studies. No evidence has been found for functional synaptic

Reinforcement learning models of the BG

Most brain dopamine is generated by midbrain dopaminergic neurons, projecting to the striatum. The central role of dopamine in controlling motivation and learning has been known for many years [33], however, most ‘box and arrow’ models of the BG have overlooked the relationships between dopamine and learning in normal BG function. The outstanding series of physiological experiments by Schultz (see [34]) revealed that the dopaminergic signal is best characterized as relating to the differences

Dimensionality reduction neural networks

Reinforcement learning models emphasize the position of the BG in normal behavior; however, the role of the BG in the pathophysiology of movement disorders has been overlooked. A model that combines most of the anatomical, physiological and computational approaches cited above has recently been suggested [45••] (Fig. 2, and see 14••., 46•. for related approaches). The model assumes that the BG perform efficient dimensionality reduction 47., 48. and decorrelation of the large information space

Closing the loop, sequential behavior and conclusions

The output of the BG is directed mainly towards the thalamus. Most models of the BG network assume that the thalamus acts as a simple relay station between the GPi and the frontal cortex. However, the projections from GPi to several thalamic nuclei, the heavy back projections from the cortex to the thalamus and to the reticular nucleus, the thalamo-striatal projections 52•., 53••., and finally the complex thalamic network, suggest that the thalamus serves a more complicated role. In any case,

Acknowledgments

This study was supported in part by the Israeli Academy of Science and the US–Israel Bi-national Science Foundation. We thank Opher Donchin, Genela Morris and Eilon Vaadia for their critical reading and helpful suggestions. We thank Aeyal Raz, Gali Heimer, Joshua A Goldberg, Sharon Maraton, Thomas Boroud, Rony Paz, David Arkadir and Genella Morris for their physiological studies that form the basis of this manuscript.

References and recommended reading

Papers of particular interest, published within the annual period of review,have been highlighted as:

  • • of special interest

  • •• of outstanding interest

References (57)

  • P. Calabresi et al.

    Acetylcholine-mediated modulation of striatal function

    Trends Neurosci

    (2000)
  • D. Joel et al.

    The organization of the basal ganglia-thalamocortical circuits: open interconnected rather than closed segregated

    Neuroscience

    (1994)
  • O. Hikosaka et al.

    Parallel neural networks for learning sequential procedures

    Trends Neurosci

    (1999)
  • J.P. Bolam et al.

    Synaptic organisation of the basal ganglia

    J Anat

    (2000)
  • A. Gillies et al.

    Computational models of the basal ganglia

    Mov Disord

    (2000)
  • G.E. Alexander et al.

    Parallel organization of functionally segregated circuits linking basal ganglia and cortex

    Annu Rev Neurosci

    (1986)
  • G. Percheron et al.

    The basal ganglia related system of primates: definition, description and informational analysis

  • G.S. Berns et al.

    How the basal ganglia make decisions

  • T. Koos et al.

    Inhibitory control of neostriatal projection neurons by GABAergic interneurons

    Nat Neurosci

    (1999)
  • J.R. Wickens et al.

    Neuronal dynamics and surround inhibition in the neostriatum: a possible connection

  • T. Boraud et al.

    Ratio of inhibited to activated pallidal neurons decreases dramatically during passive limb movement in the MPTP-treated monkey

    J Neurophysiol

    (2000)
  • H. Bergman et al.

    Reversal of experimental parkinsonism by lesions of the subthalamic nucleus

    Science

    (1990)
  • A.L. Benabid et al.

    Deep brain stimulation of the subthalamic nucleus for Parkinson's disease: methodologic aspects and clinical criteria

    Neurology

    (2000)
  • O. Aizman et al.

    Anatomical and physiological evidence for D1 and D2 dopamine receptor colocalization in neostriatal neurons

    Nat Neurosci

    (2000)
  • S.M. Nicola et al.

    Dopaminergic modulation of neuronal excitability in the striatum and nucleus accumbens

    Annu Rev Neurosci

    (2000)
  • J.A. Obeso et al.

    Pathophysiology of levodopa-induced dyskinesias in Parkinson's disease: problems with the current model

    Ann Neurol

    (2000)
  • D. Jaeger et al.

    Surround inhibition among projection neurons is weak or nonexistent in the rat neostriatum

    J Neurophysiol

    (1994)
  • A. Raz et al.

    Firing patterns and correlations of spontaneous discharge of pallidal neurons in the normal and the tremulous 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine vervet model of parkinsonism

    J Neurosci

    (2000)
  • Cited by (0)

    View full text