Age-related changes in behavioral and neural variability in a decision-making task

This study utilizes large-scale neural recordings across 16 brain regions in mice to demonstrate that aging increases behavioral response variability and disrupts neural variability quenching, characterized by globally elevated firing rates and region-specific alterations in the visual, motor, striatal, and thalamic areas.

The Gist

The Big Picture: Why Do We Get "Slower" and "Clunkier" as We Age?

You've probably noticed that as people get older, they don't just get slower; they get more unpredictable. Sometimes an older person reacts instantly, and other times they take a long time to make a simple decision. Scientists have long suspected that this isn't just because the brain is "slowing down," but because the brain is getting noisier.

Imagine your brain is a radio station. When you are young, the signal is clear and steady. As you age, static (or "noise") creeps in, making the signal harder to tune in. This paper asks: Is the brain actually getting noisier, and where does this noise come from?

To find out, the researchers didn't just look at human brains (which is hard to do directly); they looked at 149 mice ranging from "young adults" to "senior citizens" (roughly ages 3 to 20 months, which is like humans aged 23 to 70). They used a super-advanced tool called Neuropixels probes—think of these as microscopic "microphone arrays" that can listen to the conversations of over 18,000 individual brain cells at once across 16 different parts of the brain.

The Experiment: A Visual Game of Whack-a-Mole

The mice played a video game. A light would flash on the left or right side of a screen. The mouse had to turn a steering wheel to move the light to the center to get a water reward.

• The Young Mice: They were consistent. They saw the light, thought about it, and turned the wheel with a steady rhythm.
• The Older Mice: They were a bit more scattered. They didn't necessarily take longer on average, but their reaction times were all over the place. One second they were fast, the next they were slow.

The Analogy: Imagine a group of runners. The young runners all cross the finish line within a 1-second window. The older runners might cross in the same average time, but one crosses in 10 seconds, the next in 11, and the next in 9. The variability is higher, even if the average speed is similar.

The Discovery: The Brain is "Louder" and Less Focused

When the scientists listened to the neurons (the brain cells), they found three major changes in the older mice:

1. The "Volume" is Turned Up Too High

In many parts of the brain (like the visual cortex and the motor areas), the older neurons were firing more often than the young ones.

• The Analogy: Imagine a library. In a young library, people whisper when they need to talk. In the old library, everyone is shouting, even when they don't need to. This constant "shouting" (high firing rates) makes it harder to hear the specific message you need. Interestingly, in the "relay stations" of the brain (the thalamus), the volume was actually turned down, suggesting the brain is trying to compensate in some areas while over-compensating in others.

2. The "Static" Increases After a Signal

When a stimulus (the light) appears, a healthy brain usually calms down. It stops the random chatter and focuses all its energy on the task. This is called "Variability Quenching."

• The Analogy: Think of a chaotic party. When the DJ drops the beat (the stimulus), everyone stops talking and starts dancing in sync. The noise of conversation disappears.
• The Problem: In the older mice, the party didn't quiet down as much. Even after the light appeared, the neurons kept chattering randomly. They failed to "quench" the noise. This means the brain is still trying to process the signal while simultaneously dealing with a lot of background static.

3. The "Focus" is Weaker

Because the brain couldn't quiet the noise, the older mice had a harder time making consistent decisions.

• The Analogy: If you are trying to read a book in a quiet room (young brain), it's easy. If you are trying to read the same book in a room where people are shouting and the TV is blaring (old brain), you might get the gist of the story, but you'll stumble over words and take longer to understand the plot. The "signal-to-noise ratio" has dropped.

Why Does This Matter?

The researchers checked to make sure this wasn't just because the older mice were moving their bodies differently or had been training longer. They ruled those out. The changes were truly about age.

This study is a big deal because:

1. It confirms a theory: It proves that "neural noise" really does increase with age, and this noise is linked to the inconsistent behavior we see in older adults.
2. It maps the noise: They didn't just say "the brain is noisy." They showed where the noise is. It's mostly in the visual and motor areas (where the mice saw the light and moved the wheel) and the striatum (a decision-making hub).
3. It opens the door for cures: By understanding exactly which brain circuits are getting "noisy," scientists can eventually design drugs or therapies to help "tune the radio" back to a clear signal, potentially helping older people maintain their sharpness.

The Takeaway

Aging doesn't just mean your brain is "slower." It means your brain is less consistent. It's like an orchestra where the musicians are still playing the right notes, but they are playing them at slightly different times and volumes, creating a jumbled sound. The goal of future research is to help the orchestra get back in sync.

Technical Summary

1. Problem Statement

Age-related cognitive decline is often hypothesized to stem from increased neural "noise" or variability, leading to reduced signal-to-noise ratios and impaired decision-making. However, existing literature presents conflicting evidence:

• Human Studies: fMRI and EEG data show contradictory results, with some studies reporting increased BOLD variability and others reporting decreased variability in older adults. EEG studies suggest increased baseline noise (flatter 1/f power spectra), but these measures can be confounded by vascular and cardiac factors.
• Animal Studies: Previous single-neuron recordings in primates (often anesthetized) showed increased spiking variability (Fano Factor) in older animals, but these datasets were small, limited to specific cortical areas (V1, MT), and lacked behavioral context.
• Gap: There is a lack of large-scale, brain-wide neural recordings in awake, behaving animals that systematically characterize how aging affects single-neuron variability, firing rates, and stimulus-induced "variability quenching" across diverse brain regions.

2. Methodology

The study leveraged the International Brain Laboratory (IBL) standardized protocols to conduct a lifespan analysis of neural activity in mice.

• Subjects: A total of 149 mice (C57BL/6 strain) were analyzed, spanning an age range of 3 to 20 months (roughly equivalent to human ages 23–70 years).
  • Young Group: N=97 (Mean age ~5.5 months).
  • Old Group: N=52 (Mean age ~11.6 months).
  • Note: Age was treated as a continuous variable in all statistical models.
• Task: Mice performed a standardized visual perceptual decision-making task. They discriminated the location (left vs. right) of a Gabor patch with varying contrast levels (0% to 100%) by rotating a steering wheel.
• Neural Recording:
  • Technique: Neuropixels probes (up to 2 per session) recorded extracellular activity.
  • Scope: 16 brain regions across the cortex (visual, motor, frontal), striatum, thalamus, midbrain, hippocampus, and olfactory areas.
  • Data Volume: >18,000 "good" single neurons were identified from 367 recording sessions and 503 probe insertions after rigorous quality control (QC).
• Analysis Metrics:
  • Behavioral: Reaction Time (RT) variability (Coefficient of Variation), psychometric parameters (bias, threshold, lapse rate).
  • Neural:
    • Firing Rates: Pre- and post-stimulus.
    • Fano Factor (FF): Ratio of spike-count variance to mean spike count, used to quantify trial-to-trial variability.
    • Variability Quenching: The reduction in FF upon stimulus onset (Post-stimulus FF minus Pre-stimulus FF).
    • Contrast Modulation: Slope of neural response changes relative to stimulus contrast.
  • Controls: Analyses controlled for training duration and movement covariates (paw speed, nose-tip speed) derived from DeepLabCut video tracking. Statistical inference used Bayes Factors and permutation tests.

3. Key Contributions

1. Largest Scale: This is the first study to record from >18,000 neurons across 16 brain regions in behaving mice to map age-related neural changes.
2. Behavioral-Neural Link: It explicitly links increased behavioral RT variability in aging mice to specific neural mechanisms (attenuated variability quenching).
3. Regional Specificity: It moves beyond global averages to reveal that aging effects are highly region-specific (e.g., increased firing in cortex vs. decreased in thalamus).
4. Open Science: The authors released a curated, open dataset and code to facilitate further research into the computational mechanisms of cognitive decline.

4. Key Results

A. Behavioral Findings

• Increased RT Variability: Older mice did not respond significantly slower on average, but their Reaction Times (RTs) were significantly more variable (higher Coefficient of Variation). This replicates a well-known finding in human aging.
• Slight Performance Decline: Older mice showed increased absolute bias, higher perceptual thresholds, and higher lapse rates, indicating slightly worse task performance.
• Learning: Older mice learned the task at similar rates to younger mice; the age variability at recording was primarily due to the age at which training started, not the duration of training.

B. Neural Findings

• Global Firing Rate Changes:
  • Increased Rates: Older animals showed higher firing rates in visual/motor cortex, striatum, midbrain, and hippocampus.
  • Decreased Rates: Older animals showed lower firing rates in thalamic areas (LP and PO).
  • No Change: Posterior parietal cortex (PPC) showed no significant age-related firing rate change.
• Increased Post-Stimulus Variability:
  • Baseline (pre-stimulus) variability (Fano Factor) was unchanged with age.
  • Post-stimulus variability increased significantly in older animals, particularly in the midbrain and thalamic areas.
• Attenuated Variability Quenching:
  • In young animals, neural variability typically drops ("quenches") immediately after a stimulus.
  • Older animals showed attenuated quenching (the reduction in variability was weaker). This effect was most prominent in the visual/motor cortex, striatum, and thalamus (LP).
• Contrast Modulation: There was no significant global age effect on how firing rates or variability changed with stimulus contrast, suggesting that the loss of selectivity (neural dedifferentiation) was not the primary driver of the observed effects in this task.

C. Control Analyses

• Training Duration: The observed neural effects persisted even after controlling for the duration of training, confirming they are linked to chronological age rather than training history.
• Movement: Controlling for task-relevant (paw) and task-irrelevant (nose) movement did not eliminate the age-related neural differences, suggesting the effects are intrinsic to neural processing rather than motor artifacts.

5. Significance and Implications

• Mechanism of Decline: The study suggests that age-related cognitive decline is not merely due to a general increase in "noise" or slower processing speed. Instead, it is characterized by a failure to stabilize neural representations upon stimulus onset (attenuated quenching). This lack of temporal precision in neural circuits likely drives the increased trial-to-trial variability in behavior (RTs).
• Regional Heterogeneity: Aging does not affect the brain uniformly. The divergence between increased firing in cortical/striatal areas and decreased firing in thalamic areas highlights complex, circuit-specific adaptations or failures in the aging brain.
• Validation of Mouse Models: The replication of human-like behavioral variability (increased RT CV) in mice validates the mouse as a robust model for studying the neural basis of cognitive aging.
• Future Directions: The findings point toward inhibitory and neuromodulatory system degradation as potential biological mechanisms. The open dataset provides a foundation for investigating how single-neuron variability propagates to network dynamics and ultimately drives behavioral deficits.