Abstract
Rewards influence plasticity of early sensory representations, but the underlying changes in circuitry are unclear. Recent experimental findings suggest that inhibitory circuits regulate learning. In addition, inhibitory neurons are highly modulated by diverse long-range inputs, including reward signals. We, therefore, hypothesise that inhibitory plasticity plays a major role in adjusting stimulus representations. We investigate how top-down modulation by rewards interacts with local plasticity to induce long-lasting changes in circuitry. Using a computational model of layer 2/3 primary visual cortex, we demonstrate how interneuron circuits can store information about rewarded stimuli to instruct long-term changes in excitatory connectivity in the absence of further reward. In our model, stimulus-tuned somatostatin-positive interneurons develop strong connections to parvalbumin-positive interneurons during reward such that they selectively disinhibit the pyramidal layer henceforth. This triggers excitatory plasticity, leading to increased stimulus representation. We make specific testable predictions and show that this two-stage model allows for translation invariance of the learned representation.
Introduction
While it is well-established that "practice with a reward" accelerates learning and sharpens sensory sensitivity, the precise cellular mechanisms that translate a transient reward signal into permanent cortical reorganization remain elusive. In the primary visual cortex (V1), neurons become more responsive to stimuli paired with rewards. However, reward signals—often delivered via neuromodulators like dopamine—are typically spatially broad and short-lived. How does the brain use these signals to create highly specific, long-lasting refinements in excitatory neural networks?
This study proposes that inhibitory interneurons act as the critical "missing link." By capturing reward information and acting as a template, these inhibitory circuits guide the subsequent reorganization of the broader excitatory network.
The Two-Stage Computational Model
The researchers developed a biologically constrained spiking neural network model of Layer 2/3 of the primary visual cortex. The model incorporates four primary cell types: Excitatory Pyramidal Cells (PCs) and three classes of inhibitory interneurons: Parvalbumin-positive (PV), Somatostatin-positive (SST), and Vasoactive Intestinal Peptide-positive (VIP).
The core of the model is a two-stage process governed by Spike-Timing-Dependent Plasticity (STDP) in both excitatory and inhibitory synapses:
- Rewarded Phase: A specific visual stimulus (e.g., a vertical bar) is paired with a top-down reward signal that specifically excites VIP cells.
- Refinement Phase: The reward signal is removed, but the stimulus continues to be presented, allowing the circuit to consolidate changes.
Key Insights: Inhibitory Memory and Disinhibition
The model suggests that reward information is initially stored not in excitatory connections, but in the SST-to-PV inhibitory synapses. During the rewarded phase, VIP cells inhibit SST cells. This suppression, combined with stimulus-driven activity, triggers a plasticity mechanism that strengthens the connections from stimulus-tuned SST cells to PV cells.
Once this SST-to-PV connection is established, it creates a "disinhibitory motif." When the rewarded stimulus is presented again (even without the reward signal), the SST cells strongly inhibit the PV cells. Since PV cells are the primary source of inhibition for Pyramidal cells, this "inhibiting the inhibitor" creates a window of heightened activity and plasticity for the PCs. This mechanism also provides a biological explanation for translation invariance—the ability to recognize a stimulus regardless of its exact retinal position.
Results and Circuit Reorganization
The simulation results demonstrate that this two-stage process leads to significant, long-term changes in the excitatory network. Even after the reward signal is withdrawn, the excitatory PC-PC connections for the rewarded stimulus continue to strengthen during the refinement phase.
As shown in the results, Pyramidal cells become more sharply tuned to the rewarded stimulus, matching observations from in vivo mouse studies. The authors provide specific predictions for experimentalists: PV cells should initially become less responsive to a rewarded stimulus before eventually becoming more responsive once the excitatory circuit catches up, and blocking inhibitory plasticity should prevent long-term excitatory changes.
Discussion and Future Directions
The model provides a robust framework for understanding how top-down signals shape sensory representations. While the current model uses simplified "point neurons," future work could incorporate complex dendritic structures to better capture how SST cells (which target dendrites) and PV cells (which target cell bodies) interact. Additionally, determining which specific neuromodulator drives the top-down VIP signal and investigating how these representations are "unlearned" remain important areas for future research.
Summary updated