A scalable hippocampal code for flexible interval timing through persistent activity

This study identifies a distinct subpopulation of persistently active cells in the dorsal CA1 hippocampus that exhibit scalable ramping dynamics, providing a flexible neural mechanism for interval timing that complements canonical time cell sequences.

The Gist

The Big Picture: How the Brain Keeps Time

Imagine you are waiting for a microwave to finish heating your lunch. You know it takes about 2 minutes. You don't need a clock; your brain just "feels" the time passing so you can grab the food the second the beep sounds.

Scientists have long known that a part of the brain called the hippocampus (usually famous for helping us remember where we are) is also crucial for remembering when things happen. They previously discovered "Time Cells"—neurons that fire like a relay race, where one neuron fires at 1 second, another at 2 seconds, and so on, creating a mental timeline.

The Problem: In this new study, the researchers found that this "relay race" model doesn't work perfectly for all situations. When mice had to wait for a reward, the "Time Cells" were mostly silent during the long wait and only woke up right when the reward arrived. It was like having a clock that only ticks when the alarm goes off, but stays silent while you are waiting.

The Discovery: The team found a new type of neuron, which they call PACs (Persistently Active Cells). These cells act like a dimmer switch rather than a relay race. They turn on when the waiting starts, stay "on" (or slowly change brightness) the entire time you are waiting, and turn off the moment you act.

---

The Experiment: The Mouse and the Virtual Wall

To figure this out, the researchers put mice in a tube in front of a screen showing a virtual world.

1. The Start: A visual cue appears (like a green light).
2. The Wait: A wall appears, and the mouse must wait. The wall moves, but the mouse can't pass it yet.
3. The Reward: After a specific amount of time (say, 4 seconds), a "reward window" opens. If the mouse licks a water spout during this window, it gets a drink.

The mice learned quickly. They stopped licking randomly and started licking just before the window opened, showing they had learned to estimate the time.

The "Aha!" Moments

1. The Missing Middle

When they looked at the brain activity, they saw the "Time Cells" (the relay runners). But these runners were lazy! They mostly only ran when the reward was actually happening. The long, boring wait in the middle was empty. The brain needed something else to fill that gap.

2. The "Dimmer Switch" Neurons (PACs)

Enter the PACs.

• The Analogy: Imagine a long hallway. The "Time Cells" are like people standing at specific spots in the hallway, one at the 1-meter mark, one at the 2-meter mark. If you stop walking, they don't do much.
• The PACs: These are like a single spotlight that starts bright at the beginning of the hallway and slowly fades as you walk down it. Or, in some cases, a light that starts dim and slowly gets brighter until you reach the end.
• What they did: As soon as the "Wait" started, these PACs lit up. They stayed active the whole time the mouse was waiting. When the mouse finally licked (the "Go" signal), the PACs turned off.

3. The Magic of "Stretching" (Temporal Scaling)

This is the coolest part. The brain needs to be flexible. Sometimes the mouse waits 3 seconds; sometimes it waits 5 seconds.

• The Analogy: Think of a rubber band.
  • If the mouse waits a short time, the rubber band is short. The PACs fire quickly and turn off.
  • If the mouse waits a long time, the rubber band stretches. The PACs fire for a longer duration, but the pattern of their firing stretches out to match the time.
• The Result: The brain didn't need a new set of neurons for every different time. It just used the same "rubber band" neurons and stretched them to fit the wait. This explains how we can time anything from a few seconds to a few minutes using the same brain circuit.

4. Learning Makes the Switches Brighter

The researchers watched the mice learn the task over several days.

• Early Days: The mice were confused. They licked randomly. The PACs were quiet or inconsistent.
• Later Days: The mice became experts. They waited perfectly. The researchers found that more PACs were active, and they were firing more reliably.
• The Takeaway: It seems the brain recruits these "dimmer switch" neurons specifically when it needs to learn to time something precisely.

Why Does This Matter?

This study changes how we think about the brain's internal clock.

• Old View: The brain counts time by firing a sequence of different neurons (1, 2, 3, 4...).
• New View: The brain also uses a "sustained signal" that stretches and shrinks like a rubber band to track time.

It's like realizing that to measure a long distance, you don't just need a ruler with marks on it (Time Cells); you also need a stretchy tape measure (PACs) that can adapt to any length. This flexible system allows animals (and humans) to adjust their behavior instantly when the world changes, ensuring we don't act too early or too late.

Summary in One Sentence

The brain doesn't just count seconds with a ticking clock; it also uses a flexible, stretchy "dimmer switch" in the hippocampus that stays active the whole time we wait, helping us know exactly when to act.

Technical Summary

1. Problem Statement

Accurate interval timing (seconds to minutes) is essential for adaptive behavior, allowing animals to anticipate rewards and suppress premature actions. While the hippocampus is known to contain "time cells" that fire sequentially to tile an interval, it remains unclear how hippocampal circuits support flexible timing. Specifically, it is unknown whether sequential time cell activity alone can account for timing when animals exhibit trial-to-trial variability in response times or when the required delay duration changes. The study aims to identify the specific neural dynamics in the dorsal CA1 region that support this flexibility.

2. Methodology

The researchers employed a multi-modal approach combining behavioral tasks, pharmacological manipulations, and advanced neurophysiological recording techniques in mice:

• Behavioral Task (Time-Estimation Task): Head-fixed mice were trained to estimate a delay period (initially 4s) following a visual cue. They received a water reward only if they licked within a 1-second unmarked window after the delay.
  • Perturbations: The study included "no-cue" sessions (visual screens off), reward omission trials (10% of trials), and block-delay sessions (alternating 3s and 5s delays).
  • Learning Protocol: Mice were trained over days, with the delay gradually increasing from 2s to 4s and reward delivery shifting from automatic to lick-triggered.
  • Control Sessions: "Random licking" sessions were conducted where the delay was extended to 40s, screens were off, and rewards were automatic, disrupting the timing requirement.
• Neurophysiology:
  • Two-Photon Calcium Imaging: Used to record activity from dorsal CA1 pyramidal neurons (expressing jGCaMP7f or jGCaMP8s) in head-fixed mice.
  • Extracellular Recordings: 64-channel silicon probes were used to validate findings at the single-unit spike level.
  • Pharmacology: Local inactivation of the hippocampus using muscimol confirmed the necessity of the region for task performance.
• Data Analysis:
  • Cell Classification: Neurons were classified as "Time Cells" (sequential firing peaks) or "Persistently Active Cells" (PACs) based on trial-averaged activity aligned to specific events (last lick of previous trial vs. first lick of current trial).
  • Temporal Scaling: Activity profiles were analyzed for "time warping" (stretching/compressing) relative to the animal's response time.
  • Decoding: Bayesian decoding was used to determine if elapsed time could be predicted from the population activity of PACs.

3. Key Contributions

The paper introduces a novel subpopulation of hippocampal neurons and redefines the understanding of hippocampal timing mechanisms:

1. Identification of PACs: Discovery of "Persistently Active Cells" (PACs), a distinct subpopulation of CA1 pyramidal neurons that maintain sustained activity from the onset of the timing interval (defined by the previous trial's reward consumption) until the animal's behavioral response (first lick).
2. Complementarity to Time Cells: The study demonstrates that canonical time cells are sparse during the waiting period and biased toward the reward event, whereas PACs fill the temporal gap, providing a continuous code for the interval.
3. Scalability: PACs exhibit temporal scaling, meaning their activity profiles stretch or compress to match both trial-to-trial variations in response time and changes in the required delay duration.
4. Learning Correlation: The prevalence of PACs increases significantly as animals learn the task, suggesting a direct link between these dynamics and behavioral proficiency.

4. Key Results

• Timing Reference Point: Behavioral perturbations revealed that the "start" of the timing interval is not the visual cue onset, but rather the reward consumption event (last lick) of the previous trial. Removing the visual cue did not affect performance, but omitting the reward disrupted the timing of the subsequent trial.
• Sparse Time Cell Coverage: While ~25% of CA1 neurons were classified as time cells, their firing peaks were heavily concentrated within the 1-second period following reward consumption. They were notably sparse during the 3–4 seconds leading up to the animal's predictive licking.
• Characterization of PACs:
  • Prevalence: PACs comprised ~19.5% of the pyramidal population.
  • Dynamics: PACs increased activity after the "last lick" and decreased activity at the "first lick."
  • Subtypes: K-means clustering revealed two complementary subgroups: "ramp-down" cells (peaking early and declining) and "ramp-up" cells (gradually increasing until the response).
• Temporal Scaling:
  • Trial-to-Trial: On trials where mice licked earlier (3.5–4.5s), PAC activity compressed; on late-lick trials (5–6s), activity stretched. Time-warping the interval aligned the activity profiles perfectly.
  • Delay Duration: In block-delay experiments (3s vs. 5s), PACs scaled their duration to match the new delay, maintaining the same proportion of the interval.
• Decodability: Bayesian decoding using only PAC population activity successfully predicted elapsed time on single trials with significantly lower error than shuffled controls.
• Behavioral Specificity:
  • In "random licking" sessions (where timing was not required), the percentage of PACs dropped significantly, and their sustained activity was lost.
  • During learning, the percentage of PACs increased significantly from early to late training, while time cell prevalence and "lick-on" cells (opposite temporal profile) remained stable.

5. Significance

This study fundamentally shifts the understanding of hippocampal contributions to interval timing:

• Beyond Sequences: It challenges the view that hippocampal timing relies solely on sequential "time cell" firing. Instead, it proposes a dual-code system where sustained, scalable persistent activity complements sequential codes.
• Mechanism for Flexibility: The discovery of temporal scaling in PACs provides a neural mechanism for how the brain handles the natural variability of behavior. Unlike rigid sequences, scalable persistent activity allows the brain to maintain a representation of time that adapts dynamically to when an action is actually executed.
• Learning and Plasticity: The correlation between PAC prevalence and learning suggests that the recruitment of these specific dynamics is a hallmark of acquiring interval timing skills.
• Generalizability: The findings extend to other tasks (e.g., virtual reality running) and align with ramping dynamics observed in the frontal cortex and striatum, suggesting a conserved computational principle for timing across brain regions, but with a unique implementation in the hippocampus.

In conclusion, the paper establishes that the hippocampus supports flexible interval timing not just by marking specific moments in time, but by maintaining a scalable, evolving population trajectory that bridges the gap between a behavioral event and the subsequent action.