Cell-type-specific sustained value representations in the claustrum

This study demonstrates that the claustrum contains distinct cell populations that sustain reward-rate-dependent activity to guide flexible decision-making and reaction time adjustments in mice, thereby establishing it as a key subcortical locus for stable value representations in value-based decision circuits.

The Gist

Imagine your brain is a bustling, high-tech city. In this city, the Frontal Cortex is the CEO's office—the place where big decisions are made, like "Should I turn left or right?" or "Is it worth waiting for a better deal?"

For a long time, scientists thought the CEO's office did all the heavy lifting alone. But this new research discovers a crucial, hidden assistant in the basement of the city: the Claustrum.

Here is the story of what this paper found, explained simply.

The Job: The "Value Detective"

The researchers put mice in a video-game-like situation. The mice had to choose between two water spouts. Sometimes one spout gave water often, and the other rarely. The catch? The "good" spout would switch randomly. The mice had to keep learning and adapting to find the best water source.

The scientists wanted to know: How does the brain keep track of how "valuable" the world is right now? Is the water plentiful? Is it scarce?

They found that the Claustrum acts like a real-time weather report for rewards.

• When the environment is full of rewards (lots of water), the Claustrum says, "Great! Things are good!"
• When the environment is dry (no water), the Claustrum says, "Uh oh, things are tough."

The Superpower: The "Memory Stick"

Here is the coolest part. Usually, neurons fire when something happens (like a mouse licks a spout) and then stop. But these Claustrum neurons are different.

Imagine you are waiting for a bus. You know the schedule. Even though the bus hasn't arrived yet, you are thinking about it.
The Claustrum neurons keep "thinking" about the value of the water for seconds after the mouse has finished drinking and before the next trial starts. They hold onto that information like a sticky note that doesn't fall off. This "sticky note" tells the brain: "Remember, the water was scarce last time. Be ready to switch strategies."

This persistent signal helps the mouse decide:

1. How fast to react: If the world is rich with rewards, the mouse moves fast. If it's poor, the mouse slows down and thinks harder.
2. Whether to switch: If the current spout isn't giving water, the Claustrum signals, "Time to try the other one!"

The Plot Twist: Two Different Types of Workers

The researchers discovered that the Claustrum isn't just one big blob of cells. It has two distinct teams of workers, and they do opposite jobs:

1. The "Excited" Team (Narrow-Spiking Neurons)

• What they do: These guys get hyped up when the mouse is actually doing the task (licking, choosing).
• Their vibe: They are like the cheerleaders. They fire a lot when the mouse is working, and they react to whether the mouse got a reward or not. They are the "in-the-moment" workers.

2. The "Suppressed" Team (Wide-Spiking Neurons)

• What they do: These guys are the managers. They actually calm down (stop firing) when the mouse is busy working.
• The Magic: But here is the trick. When the mouse is waiting between trials, these managers start talking.
  • If the world is rich with rewards, they stay quiet.
  • If the world is poor (low value), they start firing like crazy.
• The Connection: The scientists proved that these specific "manager" neurons are the ones sending a direct line to the CEO's office (the Frontal Cortex). They are the ones whispering, "Hey boss, the water is running low. We need to change our plan!"

The Big Picture

Think of the brain's decision-making process like a car driving through a foggy city.

• The Frontal Cortex is the driver.
• The Claustrum is the GPS system.

The GPS doesn't just tell the driver "Turn Left." It constantly updates the driver on the overall traffic conditions (the total value).

• If the GPS says, "Traffic is light, plenty of gas stations nearby," the driver speeds up and sticks to the current route.
• If the GPS says, "Traffic is gridlocked, no gas for miles," the driver slows down and starts looking for a new route.

Why This Matters

This paper changes how we see the brain. We used to think the "thinking" part of the brain (the cortex) did everything. Now we know there is a subcortical "GPS" (the Claustrum) that constantly monitors the value of our environment and sends a steady, stable signal to the cortex to help us make flexible, smart decisions.

It's the difference between a robot that just reacts to what's in front of it, and a smart animal that knows the big picture and adjusts its behavior seconds before it even needs to act.

Technical Summary

1. Problem Statement

Flexible decision-making relies on the interaction between the frontal cortex and subcortical structures. While the frontal cortex is known to track recent experience to inform future decisions, the specific role of the claustrum—a subcortical nucleus highly interconnected with the frontal cortex—in representing sustained value signals during dynamic decision-making remains unclear.
Key questions addressed include:

• Does the claustrum represent stable value estimates (specifically "total value" or reward availability) between trials?
• How does this activity relate to behavioral adjustments (reaction time, choice switching)?
• Are there distinct cell types within the claustrum that differentially encode these signals and project to the frontal cortex?

2. Methodology

The study employed a combination of in vivo electrophysiology, optogenetics, behavioral modeling, and anatomical tracing in mice.

• Behavioral Task: Mice performed a dynamic foraging task. They chose between two water ports with reward probabilities ($P(R) \in \{0.1, 0.5, 0.9\}$) that fluctuated independently every 20–35 trials. The task included variable intertrial intervals (ITIs) up to 21 seconds, allowing for the study of sustained activity in the absence of immediate sensory input.
• Electrophysiology:
  • Tetrode Recordings: 341 neurons were recorded from the anterior claustrum of 9 mice using tetrodes tethered to an optic fiber.
  • Neuropixels & Optogenetics: In a separate cohort (7 mice), Neuropixels probes were used alongside antidromic collision testing. Mice were injected with retrograde Cre (rAAV2-retro-iCre) in the medial frontal cortex (ACC/RSC) and Cre-dependent Chronos-GFP in the claustrum. Photostimulation of cortical axons was used to identify projection neurons.
• Anatomical Tracing: Retrograde tracing with Cholera Toxin Subunit B (CTB) combined with immunohistochemistry for Parvalbumin (PV) and GAD67 to characterize the cell types projecting to the frontal cortex.
• Computational Modeling:
  • Q-Learning: A reinforcement learning (RL) model was fitted to behavioral data to estimate action values ($Q_l, Q_r$), Total Value ($\Sigma Q = Q_l + Q_r$), and Relative Value ($\Delta Q = Q_l - Q_r$).
  • Regression & Decoding: Linear regression models and Support Vector Machines (SVM) were used to correlate neural firing rates with task variables (lick rate, choice, outcome, history) and to decode value signals. Cross-temporal decoding (CTD) was used to assess signal stability.

3. Key Contributions

1. Identification of Sustained Value Signals: Demonstrated that the claustrum encodes a sustained representation of total value ($\Sigma Q$) (global reward availability) that persists for seconds during ITIs, predicting trial-by-trial behavioral adjustments.
2. Cell-Type Specificity: Identified two distinct electrophysiological populations in the claustrum: Narrow-spiking (NS) and Wide-spiking (WS) neurons.
3. Projection Mapping: Established that WS neurons are the primary population projecting to the medial frontal cortex, whereas NS neurons (likely local interneurons) do not project to these cortical areas.
4. Differential Encoding: Revealed that while both cell types represent value, WS neurons (the output to the cortex) specifically scale their activity inversely with total value (increasing firing as reward availability decreases), providing a graded signal for behavioral adjustment.

4. Key Results

A. Claustrum Encodes Choice, Outcome, and History

• During task execution, ~48% of neurons showed choice selectivity, and ~58% showed outcome selectivity.
• During ITIs, neural activity was significantly correlated with past choice and outcome history.
• A "History Model" (exponentially weighted past trials) explained neural activity significantly better than a "Last-Trial Model," indicating the claustrum integrates information over multiple trials.

B. Sustained Total Value ($\Sigma Q$) Representations

• Approximately 43% of recorded neurons showed firing rates significantly correlated with Total Value ($\Sigma Q$).
• These signals were action-independent (similar correlation regardless of left/right choice) and persisted for >10 seconds during ITIs.
• Behavioral Correlation:
  • Reaction Time: Decreased monotonically as $\Sigma Q$ increased (high value = faster response).
  • Switching Probability: Decreased as $\Sigma Q$ increased (high value = less switching).
  • The firing rate of $\Sigma Q$-encoding neurons predicted these behavioral adjustments on a trial-by-trial basis.
• Stability: Cross-temporal decoding confirmed that individual neurons maintained stable representations of $\Sigma Q$ throughout the ITI, with a gradual decay that correlated with behavioral slowing over long ITIs.

C. Distinct Cell Types: NS vs. WS

• Waveform Classification: Neurons were split into Narrow-spiking (NS) (mean width ~0.22 ms) and Wide-spiking (WS) (mean width ~0.61 ms).
• Projection Identity (Collision Test):
  • WS neurons: 24% (14/59) of testable WS neurons were collision-positive, confirming they project to the frontal cortex.
  • NS neurons: 0% (0/29) were collision-positive.
  • Anatomy: Retrograde tracing confirmed that <1% of claustrocortical neurons expressed PV (a marker for NS interneurons), supporting the finding that projection neurons are excitatory WS cells.
• Response Properties:
  • NS Neurons: Predominantly excited during task execution; showed bidirectional scaling with $\Sigma Q$ (both positive and negative correlations) during ITIs.
  • WS Neurons: Predominantly suppressed during task execution; showed a strong inverse correlation with $\Sigma Q$ during ITIs (firing rate increases as total value decreases).

D. Functional Implication of WS Activity

The WS population, which projects to the frontal cortex, exhibits a monotonic increase in firing rate as total value declines. This suggests the claustrum broadcasts a signal to the frontal cortex indicating "low reward availability," which drives:

1. Increased reaction times (slowing down).
2. Increased probability of switching choices (exploration).

5. Significance

• Subcortical Role in Value: This study positions the claustrum not just as a relay, but as a critical subcortical locus for stable, persistent value representations that bridge trials in dynamic environments.
• Mechanism for Flexible Behavior: The findings provide a mechanistic link between subcortical value estimation and cortical decision-making. The claustrum conveys a continuous, graded estimate of environmental reward availability to the frontal cortex via WS neurons.
• Circuit Specificity: By distinguishing between NS (local) and WS (projection) populations, the study clarifies how the claustrum processes information: local NS circuits may handle rapid task-related modulation, while WS circuits provide a sustained, global value signal to the cortex to regulate behavioral vigor and exploration strategies.
• Theoretical Integration: The results integrate the claustrum into reinforcement learning frameworks, suggesting it acts as a "global richness estimator" that modulates the trade-off between exploitation (staying with a high-value option) and exploration (switching when value drops).