Functional reorganization of motor cortex connectivity during learning

Kayvon Daie, Kyle Aitken, Márton Rózsa, Matthew S. Bull, Peter C. Humphreys, Ziheng Christina Wang, Lucas W. Kinsey, Mohit Rameshchandra Kulkarni, Kimberly Stachenfeld, Maria K. Eckstein, Zeb Kurth-Nelson, Clopath Claudia, Timothy Lillicrap, Matthew Botvinick, Matthew Golub, Stefan Mihalas, Karel Svoboda2026 · DOI

TL;DR

This study reveals how the brain rapidly rewires its neural connections in the motor cortex to learn new tasks. By combining advanced imaging techniques with a brain-computer interface (BCI) in mice, researchers found that learning involves strengthening specific connections between neurons that are active before a movement, rerouting this preparatory activity to control the learned task. This work provides crucial insights into the neural mechanisms underlying rapid motor learning and adaptation.

What Problem Does This Paper Address?

The brain's ability to learn new skills is fundamental to our adaptability. While we know that learning involves changes in neural activity and connections, the precise mechanisms by which these changes occur, especially during rapid learning, remain elusive. Specifically, how does the brain dynamically reshape the communication pathways between neurons in the motor cortex to support new motor behaviors? Previous research has shown that motor cortex activity reorganizes during learning, but it's been unclear whether this plasticity occurs locally within the motor cortex or is driven by upstream brain regions, and how it specifically alters the flow of information. This paper tackles this by investigating how connectivity in the mouse motor cortex changes during rapid learning of a BCI task, aiming to pinpoint where and how these changes happen.

Methods

The researchers employed a sophisticated combination of techniques:

  • Optical Brain-Computer Interface (BCI): Mice were trained to control a reward port by modulating the activity of a single neuron (the "conditioned neuron" or CN) in their motor cortex. This allowed for precise control over the learning task and the neural population being studied.
  • Two-Photon Calcium Imaging: This technique allowed for real-time visualization of neural activity in layer 2/3 of the mouse motor cortex across a large field of view (1mm x 1mm) as mice performed the BCI task.
  • Two-Photon Photostimulation: This method was used to precisely activate specific neurons or small groups of neurons. By observing how this artificial stimulation affected the activity of other neurons, the researchers could map "causal connectivity" – the directed influence one neuron has on another.
  • Longitudinal Measurements: Connectivity was mapped before and after BCI learning sessions, allowing for the direct observation of changes over time.
  • Recurrent Neural Network (RNN) Modeling: The researchers built computational models to simulate different scenarios of plasticity (local motor cortex vs. upstream) and to test hypotheses about how connectivity changes support learning.

Datasets

Not discussed in the paper. The study relies on data collected from experiments performed on mice.

Key Insights and Ideas

  • Sparse but Structured Learning: While the changes in neural activity during learning were sparse (affecting only a small subset of neurons, primarily the CN), these changes were not random. They were organized and targeted.
  • Preparatory Activity is Key: The most significant changes in connectivity occurred between neurons that were active before the trial began (pre-trial activity). This suggests that learning leverages and reshapes the brain's existing preparatory mechanisms.
  • Rerouting Information Flow: Plasticity in the motor cortex doesn't just strengthen existing pathways; it actively reroutes preparatory signals. These signals are redirected to neurons that become active later in the trial and are crucial for controlling the learned task (i.e., the CN).
  • Local Plasticity is Sufficient: The study provides strong evidence that local synaptic plasticity within the motor cortex is sufficient to drive rapid learning, rather than relying solely on changes in upstream brain areas.
  • Connectivity Reflects Task Tuning: The way neurons are connected is not arbitrary but is organized around their specific roles in different phases of the task (pre-trial, early trial, late trial, reward).

Key Results

  • Mice learned the BCI task rapidly, within minutes, showing significant improvements in hit rates and reductions in trial duration.
  • The conditioned neuron (CN) showed a marked increase in activity during learning, but this change was sparse across the broader population.
  • Mapping connectivity before and after learning revealed significant changes, particularly in connections involving neurons active during the pre-trial epoch.
  • RNN models demonstrated that both local motor cortex plasticity and upstream plasticity could explain observed activity changes, but experimental data favored local plasticity.
  • Changes in causal connectivity were strongly predicted by the task tuning of neurons, especially those involved in pre-trial activity. Specifically, strengthening connections from pre-trial tuned neurons to other pre-trial tuned neurons was a key finding.

Limitations and Future Work

  • Model Simplification: The RNN models used were simplified representations of the brain, focusing on specific aspects of plasticity. While useful for testing hypotheses, they don't capture the full complexity of neural circuits.
  • Upstream Contributions: While local plasticity was shown to be sufficient, the study doesn't exclude the possibility that upstream areas also contribute to learning-related changes in motor cortex activity.
  • Cell-Type Specificity: The study focused on layer 2/3 neurons but didn't delve deeply into specific cell types within this layer. Future work could explore how different neuronal populations contribute to this plasticity.
  • Consolidation: The study focused on rapid learning. Future research could investigate how these initial plastic changes are stabilized and consolidated into more permanent motor memories.
  • Generalizability: While this study provides a detailed look at motor cortex plasticity during BCI learning, further research is needed to understand if these mechanisms generalize to other forms of motor learning and other brain regions.

Summary updated

Abstract

Learning new tasks requires the brain to reshape the flow of neural activity, but how these changes arise from dynamic neural connectivity remains unclear. Here, we used two-photon photostimulation and calcium imaging to map learning-related changes in connectivity in layer 2/3 of mouse motor cortex, induced by learning of an optical brain-computer interface (BCI) task. Mice rapidly (within minutes) learned to change activity in a conditioned neuron to earn rewards. Activity changes were sparse; the conditioned neuron increased activity more than surrounding neurons. Mapping connectivity before and after learning revealed changes in motor cortex connectivity, enriched in neurons that were active before trial initiation, analogous to motor cortex populations that are active preceding movement. Motor cortex plasticity reroutes preparatory activity to neurons that are active later and control the conditioned neuron. Our findings show how rapid learning can be achieved through structured changes in motor cortex connectivity.
BibTeX
@misc{daie2026functional,
  author = {Kayvon Daie and Kyle Aitken and Márton Rózsa and Matthew S. Bull and Peter C. Humphreys and Ziheng Christina Wang and Lucas W. Kinsey and Mohit Rameshchandra Kulkarni and Kimberly Stachenfeld and Maria K. Eckstein and Zeb Kurth-Nelson and Clopath Claudia and Timothy Lillicrap and Matthew Botvinick and Matthew Golub and Stefan Mihalas and Karel Svoboda},
  title = {{Functional reorganization of motor cortex connectivity during learning}},
  year = {2026},
  doi = {10.64898/2026.03.03.709199},
}
Figure