Circuit mechanisms of GPe pauses account for adaptive exploration

Sang Wan Lee, Minryung Song, Shinwoo Kang, Minsu Abel Yang, Doo‐Sup ChoiResearch Square 2025 · DOI

The basal ganglia (BG) serve as the brain's central hub for reinforcement learning and habit formation, orchestrating the balance between exploiting known rewards and exploring new possibilities. Within this complex network, the external globus pallidus (GPe) has long been characterized by a distinctive physiological signature: transient "pauses" in the firing of its high-frequency discharge (HFD) neurons. While these pauses are a conserved feature across species, their functional significance has remained elusive. A recent study introduces a neurophysiologically grounded computational model, the Denoising AutoEncoder-Basal ganglia (DAE-B), to demonstrate how these pauses and the underlying GPe-STN circuit architecture drive adaptive behavior.

The GPe-STN Circuit as a Denoising Autoencoder

Traditional views of the GPe often cast it as a simple relay station. However, the DAE-B model suggests a more sophisticated role. The researchers identified that the GPe-to-subthalamic nucleus (STN) connection is characterized by high convergence, where many GPe neurons project onto a single STN neuron. This architecture mirrors a denoising autoencoder—a type of neural network designed to extract essential features from noisy input.

In this framework, the STN "senses" the average activity across the GPe population. This feedback loop allows the circuit to maintain a high signal-to-noise ratio (SNR) for action-value encodings. By cleaning up the "noise" of background neural activity, the GPe-STN circuit ensures that the brain can clearly distinguish between the values of different potential actions, facilitating decisive behavior during stable conditions.

The Emergence of GPe Pauses

One of the most striking findings of the study is the explanation for GPe pauses. Rather than being discrete signals or "error messages" sent by the brain, the model suggests these pauses are a circuit-level consequence of strong inhibition. When the STN provides powerful inhibitory feedback to the GPe, it can push the activity of HFD neurons down to a biological "floor" of 0 Hz. These periods of silence are the observed pauses.

The model successfully replicated empirical data from previous studies, showing that as animals (and the model) become more proficient at a task, the frequency of these pauses decreases. This suggests that as learning stabilizes, the need for the intense inhibitory feedback that causes pauses diminishes.

Driving Exploration through Contrast Disruption

The true power of this circuit configuration is revealed when the environment changes. In a "reversal learning" scenario—where a previously unrewarded action suddenly becomes the "correct" choice—the brain must quickly shift from exploitation to exploration.

The researchers found that immediately following such a reversal, the activity of GPe neurons projecting to the STN significantly increases. In mouse models, fiber photometry recordings showed a strong negative correlation (r = -0.83) between GPe-STN activity and the time elapsed since the task reversal. This surge in activity has a profound effect: it lowers the baseline excitability of the entire GPe population.

By lowering this baseline, the circuit effectively "flattens" the contrast between different action values. When the difference between the "best" and "second-best" options is minimized, the animal becomes less certain, naturally triggering exploratory behavior. This mechanism allows the GPe-STN circuit to act as a "switch" that promotes flexibility in the face of environmental uncertainty.

The Trade-off Between Flexibility and Proficiency

While a strong GPe-STN connection is essential for adapting to change, the study reveals a surprising trade-off: this same strength can hinder performance during extended, stable training. Simulations showed that if GPe-STN projections remain too potent, the overall proficiency of the system plateaus or even deteriorates.

This finding suggests a functional "division of labor" within the basal ganglia. Associative regions of the BG, which govern goal-directed and flexible behavior, likely maintain strong GPe-STN connectivity. In contrast, sensorimotor regions responsible for well-learned habits may benefit from weakened GPe-to-STN projections, allowing for more efficient, high-contrast action selection without the "interference" of exploratory mechanisms.

Future Directions

While the DAE-B model provides a unifying framework for GPe dynamics, it also highlights areas for further investigation. The current model simplifies the GPe by omitting arkypallidal neurons and internal lateral inhibition, both of which likely play roles in fine-tuning the denoising effect. Furthermore, while the study identifies the GPe-STN circuit as the driver of exploration, the specific "trigger" that increases projection strength during environmental shifts—potentially linked to arousal or stress systems—remains to be identified. Future research using optogenetic tools will be crucial to causally link these specific circuit activities to real-time exploratory choices in behaving animals.

Summary updated

Abstract

The external globus pallidus (GPe) has traditionally been viewed as a relay nucleus within the basal ganglia (BG), but accumulating evidence indicates a more dynamic role in reinforcement learning (RL). One key characteristic of GPe activity-transient pauses in high-frequency discharge (HFD) neurons-is preserved across species, yet its potential implications in RL remains unclear. Here, we developed a neurophysiologically grounded computational model to investigate the origin and role of GPe pauses in RL. Our model successfully replicated a range of empirical observations, including pause dynamics during learning and cue-related activity modulation. We demonstrated that the GPe-subthalamic nucleus (STN) circuit functions analogously to a denoising autoencoder, modulating baseline excitability in downstream BG circuits and that GPe pauses emerge as circuit-level consequences of strong, convergent inhibition from the GPe to STN. Simulations and in vivo recordings revealed that the activity of GPe-STN projecting neurons increases following sudden environmental changes, promoting adaptive exploration by disrupting action value contrast. Intriguingly, this same configuration impairs performance with extended training, suggesting that habitual behavior may benefit from weakened GPe-to-STN projections. These findings provide a unifying framework for understanding GPe pause dynamics and highlight circuit-level distinctions supporting the balance between flexibility and proficiency in RL.
BibTeX
@inproceedings{lee2025circuit,
  author = {Sang Wan Lee and Minryung Song and Shinwoo Kang and Minsu Abel Yang and Doo‐Sup Choi},
  title = {{Circuit mechanisms of GPe pauses account for adaptive exploration}},
  booktitle = {Research Square},
  year = {2025},
  doi = {10.21203/rs.3.rs-7117998/v1},
}
Figure