Spiking and neuromodulation during active experience shape visuomotor integration in V1 layer 2/3 neurons

This study demonstrates that in mouse V1 layer 2/3, active spiking during visuomotor experience drives firing rate-dependent plasticity to refine prediction error minimization, a process that is further facilitated by locus coeruleus-mediated neuromodulation.

The Gist

Imagine your brain is a highly sophisticated weather forecast center. Its main job isn't just to see what's happening right now (the rain, the wind); it's to constantly predict what will happen next based on your movements.

If you decide to walk forward, your brain predicts, "Okay, the scenery should be rushing past my eyes." When the scenery actually rushes past exactly as predicted, your brain says, "All clear, nothing new to learn," and it tunes out the noise. This is called prediction error minimization.

However, if you walk forward but the scenery stays still (or moves sideways), your brain screams, "Wait a minute! That's wrong!" This "wrongness" is a prediction error. The brain needs to fix its forecast so it doesn't make the same mistake twice.

This paper investigates how the brain fixes these mistakes in the visual cortex (the part of the brain that sees), specifically in a layer of neurons called Layer 2/3. Here is the story of their discovery, broken down simply:

1. The Experiment: A Virtual Reality Gym

The researchers put mice in a virtual reality tunnel. The mice could run on a wheel, which made the walls of the tunnel scroll by, simulating running through a forest.

They wanted to see how the brain learns to cancel out the "expected" visual blur of running. To do this, they used a clever trick: electric current injection.

• The "Feedback-Driven" Group: They zapped the neurons with electricity only when the mouse was running. This forced the neurons to fire (send signals) exactly when the mouse saw the world moving. This simulated a "confused" brain that thinks, "I'm running, but I'm seeing something unexpected!"
• The "Stationary-Driven" Group: They zapped the neurons when the mouse was sitting still, and silenced them when the mouse ran. This simulated the opposite confusion.

2. The Discovery: The Brain Rewires Itself

After the mice ran for a while with these artificial signals, the researchers checked the neurons again. They found something amazing: The brain had physically rewired itself to cancel out the confusion.

• The Analogy: Imagine you are trying to listen to a friend (the visual signal) while a loud fan is humming (the movement signal). If the fan hums in a way that cancels out your friend's voice, you can't hear them.
• The Result: When the neurons fired during running (the "Feedback-Driven" group), the brain learned to turn up the "fan" (the inhibitory signal from movement) specifically to cancel out the "voice" (the visual signal).
• The Twist: It wasn't a one-size-fits-all fix.
  • If a neuron was naturally very sensitive to visual motion, the brain increased the "movement cancellation" signal to quiet it down.
  • If a neuron was not sensitive to visual motion, the brain did the opposite, adjusting the signals to balance things out differently.

Key Takeaway: The brain doesn't just learn; it learns differently depending on the specific "personality" of each neuron. The more a neuron fired during the confusion, the more the brain adjusted its internal wiring to fix it.

3. The Secret Ingredient: The "Focus" Chemical

The researchers also looked at a second set of data involving a different group of mice. In this group, they activated the Locus Coeruleus (LC)—a tiny part of the brain that acts like a spotlight of attention or a "wake-up call" chemical (noradrenaline).

• Without the Spotlight: The brain tried to learn, but the changes were weak and messy.
• With the Spotlight: When the LC was activated during the running, the brain's rewiring was strong, fast, and precise.

The Metaphor: Think of the brain as a student trying to study.

• Spiking (Firing): The student is actively reading the book.
• Neuromodulation (LC): The student is drinking a strong cup of coffee.
• The Result: You can read the book all day (spiking), but without the coffee (neuromodulation), you might not remember the details. With the coffee, the learning sticks instantly. The study shows that the brain needs this "chemical coffee" to effectively rewire itself based on prediction errors.

4. The Big Picture: Why Does This Matter?

This study solves a long-standing mystery in neuroscience. We knew the brain could predict the world, and we knew it made "error signals" when predictions failed. But we didn't know how those error signals actually changed the brain's wiring to make future predictions better.

The Conclusion:

1. Action Drives Change: When neurons fire during a confusing moment (a prediction error), they trigger a physical change in their connections.
2. Precision: The brain adjusts the "volume" of movement signals to perfectly cancel out visual signals, making the world feel stable when you move.
3. Chemical Boost: This learning process needs a boost from the brain's attention system (the LC) to work efficiently.

In short, your brain is a dynamic, self-correcting machine. Every time you move and see the world, it's constantly tweaking its internal dials, using your own activity and a chemical "focus" signal to ensure that tomorrow, it will predict your world even more accurately than today.

Technical Summary

1. Problem Statement

Predictive processing theories propose that the brain continuously updates sensory predictions to minimize "prediction errors" (the discrepancy between expected and actual sensory input). In the mouse primary visual cortex (V1), layer 2/3 pyramidal neurons exhibit spiking patterns consistent with prediction errors during locomotion and visual flow. However, a critical gap exists in the literature: there is limited direct evidence that these prediction-error-related spikes drive the synaptic plasticity required to minimize future errors.

Specifically, it is unclear:

• Whether spiking during active visuomotor experience drives the reorganization of inputs (sensory vs. predictive) to improve cancellation.
• How the direction of this plasticity is determined (e.g., does it depend on the neuron's specific visual responsiveness?).
• Whether neuromodulation (specifically from the Locus Coeruleus, LC) is required to facilitate this activity-dependent plasticity.

2. Methodology

The authors employed a multi-modal approach combining in vivo whole-cell electrophysiology and analysis of a published two-photon calcium imaging dataset.

A. In Vivo Whole-Cell Recordings (Electrophysiology)

• Subjects: C57BL/6J mice running head-fixed in a virtual reality (VR) tunnel with square gratings.
• Target: Layer 2/3 pyramidal neurons in V1 (n=37).
• Experimental Paradigm:
  1. Baseline: Brief visual flow stimuli (1s) presented independently of locomotion to measure initial visual and locomotion-related inputs.
  2. Visuomotor Coupling (5 mins): Visual flow speed was coupled to locomotion speed (predictable feedback).
  3. Post-Coupling: Visual flow stimuli presented again to measure changes.
• Closed-Loop Current Injection: To test the causal role of spiking, the authors manipulated firing rates during the coupling period using two conditions:
  • Feedback-driven: Depolarizing current injected proportional to locomotion speed (spiking correlated with feedback).
  • Stationary-driven: Tonic depolarization during stationary periods and hyperpolarization during locomotion (spiking anticorrelated with feedback).
• Control Group: A "locomotion-driven" group received current injection but no visual flow feedback (static walls) to isolate the role of visual input in plasticity.
• Measurements: Subthreshold membrane potential ($V_m$) cross-correlated with locomotion speed to quantify locomotion-related inputs; visual flow responses measured via $V_m$ deflections.

B. Two-Photon Calcium Imaging Analysis

• Dataset: Re-analysis of a published dataset (Jordan and Keller, 2023) involving jGCaMP8m expression in V1 layer 2/3.
• Manipulation: In this dataset, Locus Coeruleus (LC) axons expressing ChrimsonR were optogenetically stimulated during visuomotor coupling to enhance neuromodulation.
• Analysis: Population-level calcium activity was analyzed to assess changes in visuomotor integration (opposing slopes of visual vs. locomotion responses) and self-generated activity reduction, stratified by feedback activity levels and visual responsivity.

3. Key Contributions

1. Causal Link between Spiking and Plasticity: Demonstrated that spiking during active visuomotor experience directly drives rate-dependent changes in locomotion-related inputs.
2. Visual Responsivity Dictates Plasticity Sign: Revealed that the direction of plasticity (enhancement of inhibition vs. excitation) depends on the neuron's intrinsic visual responsiveness ($V_{high}$ vs. $V_{low}$).
3. Role of Visual Feedback: Showed that visual flow feedback is strictly required for this specific form of plasticity; locomotion alone (without visual feedback) does not induce the same reorganization.
4. Neuromodulatory Facilitation: Provided evidence that LC-mediated neuromodulation is a necessary gate for these activity-dependent reorganizations at the population level.
5. Mechanism of Error Minimization: Linked the emergence of opposing visuomotor inputs to a reduction in self-generated neuronal activity, a hallmark of prediction error minimization.

4. Key Results

A. Spiking Drives Rate-Dependent Plasticity

• High firing rates during visuomotor feedback correlated with larger absolute changes in locomotion-related inputs ($V_m$-locomotion correlation).
• Multiple linear regression identified firing rate during feedback as the strongest predictor (30% variance explained) of these changes, outperforming locomotion speed or physiological drift.

B. Visual Responsivity Determines the Sign of Change

• Neurons were categorized as $V_{high}$ (strong visual depolarization) or $V_{low}$ (weak/no visual response).
• In Feedback-driven neurons:
  • $V_{high}$ neurons: Showed a significant negative change in locomotion-$V_m$ correlation (increased locomotion-related inhibition). This aligns with prediction error minimization: if a neuron fires due to unexpected visual input, the brain strengthens the predictive (locomotion) inhibition to cancel it out.
  • $V_{low}$ neurons: Showed a positive change (increased locomotion-related excitation).
• In Stationary-driven neurons: The pattern was reversed, confirming that the timing of spikes relative to feedback determines the plasticity direction.

C. Necessity of Visual Flow Feedback

• In the locomotion-driven group (current injection + locomotion, but no visual flow), the differential plasticity based on visual responsivity disappeared. Both $V_{high}$ and $V_{low}$ neurons showed negligible changes.
• Conclusion: The coincidence of spiking and visual flow input is required to drive the reorganization of locomotion-related inputs.

D. Emergence of Opposing Visuomotor Inputs

• After visuomotor coupling, feedback-driven neurons developed a significant negative relationship (anticorrelation) between visual flow responses and locomotion-related inputs.
• This "opposing" integration is the neural signature of improved cancellation (prediction error minimization).
• Bootstrapping analysis confirmed that changes in locomotion-related inputs were the primary driver of this slope change, though visual response changes contributed.

E. Population-Level Validation and Neuromodulation

• Analysis of the calcium imaging dataset recapitulated the whole-cell findings:
  • Neurons with high feedback activity showed enhanced opposing visuomotor slopes.
  • The sign of locomotion-related changes depended on visual responsivity ($V_{high}$ vs. $V_{low}$).
• Crucial Finding: These effects were absent in control mice lacking LC optogenetic stimulation. LC activation was required to facilitate the activity-dependent reorganization.

F. Reduction in Self-Generated Activity

• In the LC-stimulated dataset, neurons with high feedback activity levels showed a significant reduction in their response to self-generated feedback onsets (early vs. late in the session).
• This reduction paralleled the enhancement of opposing visuomotor inputs, confirming that the plasticity leads to a suppression of predictable activity.

5. Significance

This study provides the first direct experimental evidence supporting the prediction error minimization hypothesis in cortical layer 2/3. It demonstrates that:

1. Spiking is the driver: Prediction errors (represented by spiking) actively drive synaptic plasticity to refine future predictions.
2. Context matters: The specific nature of the plasticity (inhibitory vs. excitatory) is dictated by the neuron's existing sensory tuning ($V_{high}$ vs. $V_{low}$), suggesting a sophisticated, cell-type-specific learning rule.
3. Neuromodulation is a gate: The locus coeruleus (noradrenergic system) acts as a critical facilitator, enabling the brain to update its internal models during active behavior.

These findings bridge the gap between theoretical predictive coding models and biological reality, showing how the cortex dynamically rewires itself to cancel out predictable sensory inputs during active exploration.