Cortex-wide voltage imaging in a sensory cued reaching task reveals fast subnetwork dynamics

Yunmiao Wang, Soon Ho Kim, Eric He, Brune Le Chatelier, Hannah Choi, Dieter JaegerbioRxiv 2025 · DOI

Abstract

Oscillatory activities and network dynamics are fundamental to sensorimotor processing, motor control, and cognitive functions. Capturing the fast cortical dynamics underlying sensorimotor behaviors demands temporal resolution beyond conventional imaging methods. We employed pan-cortical JEDI-1P voltage imaging of layer 2/3 activity in a skilled reaching task to reveal previously inaccessible rapid networks dynamics. Our approach uncovered distinct spectral signatures including reward-related gamma oscillations in M2, global low-frequency suppression during movement execution, and 8 Hz large-amplitude oscillations in sensorimotor cortex during task disengagement. Using independent component analysis, we were able to separate multiplexed functional networks with fast temporal features shared among sensorimotor regions and slower dynamics such as a global preparatory ramp. These distributed activity patterns successfully predicted both sensorimotor parameters and trial outcomes. Our study bridges critical gaps in understanding rapid temporal organization of cortical networks during complex behaviors, revealing how fast neural dynamics coordinate to produce motor actions across distributed cortical regions.

Introduction

To understand how the brain coordinates complex movements, researchers often utilize "widefield" imaging to observe the entire cortex simultaneously. Historically, this has relied heavily on calcium imaging. However, calcium signals serve as slow proxies for neural activity, effectively acting as a "low-pass filter" that blurs out fast oscillations—such as gamma or beta rhythms—and rapid transitions occurring on a millisecond scale.

Because of this temporal blur, a clear picture of how fast, distributed cortical subnetworks coordinate during sensorimotor integration has remained elusive. This study addresses this gap by employing a new generation of voltage sensors to capture the high-frequency dynamics of the cortex during a goal-directed behavior, revealing the rapid temporal organization required to produce skilled motor actions.

Methodology

High-Speed Voltage Imaging

The researchers utilized JEDI-1P, a cutting-edge voltage sensor expressed in Layer 2/3 pyramidal neurons across the mouse dorsal cortex. This allowed for recording at 200 Hz, a sampling rate sufficient to track oscillations up to 100 Hz—well beyond the capabilities of standard calcium indicators. Data was collected from 7 EMX1-Cre mice across a total of 6,777 successful trials and 1,629 no-response trials.

Behavioral Task and Tracking

Mice were trained on a bilateral reaching task where they reached for a water droplet following a whisker air-puff stimulus and a variable delay. To provide high-resolution ground truth for neural data, DeepLabCut was used to track limb positions and pupil diameter from multi-angle video recordings.

Computational Analysis

Two primary computational approaches were used to interpret the high-dimensional data:

  1. Temporal Independent Component Analysis (t-ICA): A dimensionality reduction technique used to "unmix" complex cortical signals into distinct functional networks based on unique timing patterns.
  2. Deep Learning Decoders: Feedforward neural networks were trained to classify trial outcomes (success vs. failure) and stimulus side (left vs. right) based on recorded voltage activity.

Key Insights: Spectral Signatures of Behavior

The 8 Hz "Brake"

The most striking finding was a large-amplitude 8 Hz oscillation localized in the sensorimotor cortex. When this rhythm appeared, mice were "akinetic," completely ceasing paw movement. Interestingly, these mice often exhibited dilated pupils, indicating high arousal. This suggests the 8 Hz rhythm represents a specific "task-disengaged" state rather than simple sleepiness or low arousal.

The researchers discovered a gamma-band (30–80 Hz) increase in the secondary motor cortex (M2) triggered upon reward delivery. This suggests that M2 plays a critical role in rule updating or attentional shifts during successful task completion.

Multiplexed Dynamics and Symmetry

Using t-ICA, the study demonstrated that the cortex simultaneously processes "fast" dynamics (such as stimulus-locked oscillations) and "slow" dynamics (such as a global preparatory ramp-up of activity). While raw signals often appeared bilaterally symmetric, ICA successfully extracted specific contralateral subnetworks that accurately reflected which limb the mouse was using.

Predictive Modeling and Results

Outcome Prediction

Neural networks were able to predict whether a trial would result in success or a "no-response" failure with significant accuracy even before the stimulus was presented. This suggests that the pre-trial state of the cortical network is a primary determinant of behavioral performance.

Low-Dimensional Control

The study found that just six independent components were sufficient to reconstruct complex cortical activity with over 95% accuracy. This proves that cortical dynamics during this reaching task are relatively low-dimensional, despite the complexity of the behavior.

Frequency Specificity in Decoding

The decoding of different task parameters relied on different frequency bands. The stimulus side was best decoded using high-frequency activity (30-40 Hz) immediately following the air-puff. In contrast, trial outcome was best predicted using lower frequencies (4-8 Hz) during the response period.

Discussion and Future Directions

While this study provides a high-resolution map of cortical dynamics, it focuses primarily on Layer 2/3. These networks are likely coupled to deeper cortical layers and subcortical loops—such as the basal ganglia and thalamus—which remain invisible to widefield imaging. Furthermore, as an observational study, the results show strong correlations but do not definitively prove causality.

Future research utilizing optogenetics to drive or suppress specific rhythms (like the 8 Hz or gamma oscillations) will be necessary to confirm their functional roles. Additionally, combining widefield voltage imaging with deep-tissue recordings will be essential to understanding the "vertical" organization of these fast networks across the entire brain.

Summary updated

Abstract

Oscillatory activities and network dynamics are fundamental to sensorimotor processing, motor control, and cognitive functions. Capturing the fast cortical dynamics underlying sensorimotor behaviors demands temporal resolution beyond conventional imaging methods. We employed pan-cortical JEDI-1P voltage imaging of layer 2/3 activity in a skilled reaching task to reveal previously inaccessible rapid networks dynamics. Our approach uncovered distinct spectral signatures including reward-related gamma oscillations in M2, global low-frequency suppression during movement execution, and 8 Hz large-amplitude oscillations in sensorimotor cortex during task disengagement. Using independent component analysis, we were able to separate multiplexed functional networks with fast temporal features shared among sensorimotor regions and slower dynamics such as a global preparatory ramp. These distributed activity patterns successfully predicted both sensorimotor parameters and trial outcomes. Our study bridges critical gaps in understanding rapid temporal organization of cortical networks during complex behaviors, revealing how fast neural dynamics coordinate to produce motor actions across distributed cortical regions.
BibTeX
@misc{wang2025cortex,
  author = {Yunmiao Wang and Soon Ho Kim and Eric He and Brune Le Chatelier and Hannah Choi and Dieter Jaeger},
  title = {{Cortex-wide voltage imaging in a sensory cued reaching task reveals fast subnetwork dynamics}},
  year = {2025},
  doi = {10.1101/2025.04.24.650533},
  howpublished = {bioRxiv},
}
Figure