Assessing Selectivity in the Basal Ganglia: The “Gearbox” Hypothesis

Zafeirios Fountas, Murray ShanahanbioRxiv 2017 · DOI

TL;DR

This paper proposes a novel "gearbox" hypothesis for action selection in the brain, suggesting that the basal ganglia (BG) act as a flexible mechanism controlled by cortical oscillations. Using a detailed spiking neural model, the authors demonstrate that cortical frequency and dopamine levels dictate the BG's strategy, switching between exploitation, exploration, or halting the selection process.

What Problem Does This Paper Address?

The basal ganglia (BG) are widely believed to be crucial for action selection – choosing the most appropriate action from competing options. However, the precise mechanisms by which the BG achieve this, and how external factors influence this process, remain unclear. Specifically, the role of cortical oscillations, rhythmic electrical activity in the brain's outer layer, in modulating BG selectivity has been largely unexplored. This paper aims to bridge this gap by investigating how cortical oscillations, in conjunction with dopamine levels, influence the BG's ability to select actions.

Methods

The authors employed a sophisticated, large-scale spiking neural network model of the basal ganglia circuitry, building upon previous work. This model incorporates detailed neurophysiological properties and anatomical connectivity. To assess action selection, they developed quantitative metrics for "effectiveness" (selecting the most salient action) and "exploration" (selecting alternative options). They then systematically varied parameters such as cortical input frequency, phase offset between inputs, and dopamine concentration to generate "selectivity portraits" that illustrate the BG's behavior under different conditions. They also explored simplified versions of the model to assess the robustness of their findings.

Datasets

Not discussed in the paper. The study relies on a computational model and simulated data.

Key Insights and Ideas

  • The "Gearbox" Hypothesis: The BG are not simply executing a decision but rather acting as a flexible "gearbox" controlled by the cortex. Cortical oscillations act as the "gear lever," determining if and what type of selection occurs, while dopamine acts as the "control pedal," initiating or halting the process.
  • Cortical Frequency as the Primary Controller: While dopamine plays a role, cortical frequency emerges as a more decisive factor in the exploration-exploitation trade-off. Different frequency bands (alpha, beta, gamma) promote distinct selection strategies.
  • Alpha and Theta for Exploration: Low-frequency oscillations (alpha and theta) appear to promote exploration of less salient options, potentially by temporarily inhibiting the strongest input. This suggests a mechanism for dealing with uncertainty.
  • Beta for Neutrality/Criticality: Beta frequencies (around 20 Hz) seem to put the BG in a neutral state, halting both exploitation and exploration, unless there's a strong difference between competing inputs. This frequency band acts as a critical borderline.
  • Gamma for Effectiveness: Gamma oscillations appear to facilitate effectiveness, particularly in steady-state conditions when a salient channel is clearly dominant.
  • A ~200ms Selection Cycle: The model suggests a consistent selection cycle of approximately 200ms, which has implications for the timing of cognitive processes.

Key Results

  • Selectivity Portraits: The study generated detailed "selectivity portraits" showing how BG effectiveness and exploration vary with cortical frequency and dopamine levels. These portraits revealed distinct behavioral regimes associated with different frequency bands.
  • Cortical Control: The cortex, through its oscillatory activity, dictates whether selection occurs and what strategy (exploitation, exploration, or halt) the BG employs.
  • Dopamine's Role: Dopamine acts more as a trigger or modulator, influencing the initiation and strength of selection rather than dictating the fundamental strategy.
  • Frequency-Specific Strategies:
    • Low frequencies (alpha, theta) promote exploration.
  • Beta frequencies (around 20 Hz) lead to a neutral state or criticality.
  • Gamma frequencies facilitate effectiveness.
  • Selection Cycle: A consistent selection cycle of around 200ms was identified, which aligns with some cognitive architectures but challenges others.
  • Robustness: The core findings were largely maintained even in simplified versions of the BG model, suggesting their biological relevance.

Limitations and Future Work

  • Model Simplification: While simplified models were used to test robustness, the authors acknowledge the inherent limitations of computational models and the need for further investigation into the impact of specific features.
  • Limited Brain Regions: The model focuses on the BG and its immediate cortical inputs. Future work could incorporate a wider range of brain structures to better capture complete cognitive behaviors.
  • Understanding Gamma's Role: The precise mechanism behind gamma oscillations' effect on selectivity requires further investigation.
  • Cognitive Architecture Alignment: While the 200ms cycle is a significant finding, further analysis is needed to fully reconcile it with the diverse timing assumptions of various cognitive architectures.
  • Exploratory Behavior: The exact mechanism by which low-frequency oscillations facilitate exploration needs deeper exploration.

Summary updated

Abstract

Abstract Despite experimental evidence, the literature so far contains no systematic attempt to address the impact of cortical oscillations on the ability of the basal ganglia (BG) to select. In this study, we employed a state-of-the-art spiking neural model of the BG circuitry and investigated the effectiveness of this circuitry as an action selection device. We found that cortical frequency, phase, dopamine and the examined time scale, all have a very important impact on this process. Our simulations resulted in a canonical profile of selectivity, termed selectivity portraits, which suggests that the cortex is the structure that determines whether selection will be performed in the BG and what strategy will be utilized. Some frequency ranges promote the exploitation of highly salient actions, others promote the exploration of alternative options, while the remaining frequencies halt the selection process. Based on this behaviour, we propose that the BG circuitry can be viewed as the “gearbox” of action selection. Coalitions of rhythmic cortical areas are able to switch between a repertoire of available BG modes which, in turn, change the course of information flow within the cortico-BG-thalamo-cortical loop. Dopamine, akin to “control pedals”, either stops or initiates a decision, while cortical frequencies, as a “gear lever”, determine whether a decision can be triggered and what type of decision this will be. Finally, we identified a selection cycle with a period of around 200ms, which was used to assess the biological plausibility of the popular cognitive architectures. Author summary Our brains are continuously called to select the most appropriate action between alternative competing choices. A plethora of evidence and theoretical work indicates that a fundamental brain region called the basal ganglia might be the locus where this competition occurs. But how is the winning choice determined each time? Using a detailed computational model, based on neurophysiological properties of this region, we suggest that, whereas the basal ganglia might indeed contain the circuitry of action selection, the cerebral cortex is, in fact, the brain region that dictates this process. Similarly to a gearbox in a car, the basal ganglia provide modes for the exploitation of the safest option (forward gears), exploration of alternative options (reverse gear) and a neutral state, in case that the selection process needs to be halted. Our results further indicate that the instructions for mode-switching are relayed to the basal ganglia through specific low frequencies of oscillations within cortical areas. Finally, we provide estimations for the frequency ranges that can be used to activate each selectivity mode, as well as the duration of the selection process under various conditions.
BibTeX
@misc{fountas2017assessing,
  author = {Zafeirios Fountas and Murray Shanahan},
  title = {{Assessing Selectivity in the Basal Ganglia: The “Gearbox” Hypothesis}},
  year = {2017},
  doi = {10.1101/197129},
  howpublished = {bioRxiv},
}
Figure