Strain- and age-dependent divergence in mouse appetitive spatial learning and decision strategies

This study demonstrates that mouse strain (CBA/CaOlaHsd vs. C57BL/6) and age significantly interact to shape spatial learning performance and decision-making strategies in a T-maze task, with CBA/CaOlaHsd mice exhibiting faster learning, greater resilience to reward uncertainty, and distinct reinforcement learning policies compared to C57BL/6 mice.

The Gist

The Big Picture: Two Mouse Families, Two Different Ways to Learn

Imagine you are trying to teach two different families of mice how to find a delicious cheese treat hidden in a maze. You want to see how fast they learn, how smart they are, and how they react when the rules get a little tricky.

The researchers compared two very famous mouse "families" (strains):

1. The C57BL/6 mice: These are the "standard" lab mice, like the Toyota Camry of the rodent world—very common, but they have a known quirk: they start losing their hearing early in life.
2. The CBA/CaOlaHsd mice: These are the "senior citizens" of the mouse world regarding hearing; they keep their hearing sharp for much longer.

The scientists also tested these mice at two different ages: Young adults (2–3 months old) and Middle-aged adults (7–8 months old).

The Game: The "T-Maze" Treasure Hunt

The mice were put in a T-shaped maze. At the end of one arm, there was a food reward.

• Phase 1 (The Easy Mode): For the first few days, the reward was always there. If the mouse picked the right arm, it got a treat 100% of the time. This was like learning a song where the notes never change.
• Phase 2 (The Hard Mode): Then, the scientists made it tricky. They kept the reward in the same spot, but they stopped giving it out every time. Sometimes the mouse picked the right arm and got a treat; other times, it picked the right arm and got nothing. The reward became unpredictable, like a slot machine that pays out less often but gives a bigger jackpot when it does.

What They Discovered

The study found that the two mouse families didn't just learn at different speeds; they used completely different mental strategies to solve the puzzle.

1. The "Confident Learners" (CBA Mice)

The CBA mice were like confident explorers.

• Speed: They figured out the maze faster than the C57 mice.
• Strategy: Once they learned the right path, they stuck to it. Even when they didn't get a treat for a few tries (because the scientists stopped giving them out), the CBA mice didn't panic. They thought, "I know this is the right path; the machine is just glitching today. I'll keep going."
• The Takeaway: They built a strong "mental map" and trusted it, even when the feedback was noisy.

2. The "Reactive Learners" (C57 Mice)

The C57 mice were like nervous gamblers.

• Speed: They took longer to figure out the maze.
• Strategy: They were very sensitive to immediate results. If they picked the right arm and didn't get a treat, they immediately thought, "Oh no, I must be wrong!" and switched to the other arm.
• The Takeaway: They relied too much on the very last thing that happened. When the reward became unpredictable, they kept changing their minds, which made them slower and less accurate.

3. The Effect of Age (Getting Older)

As the mice got "middle-aged" (7–8 months):

• Slower Feet: Both older groups took longer to run the maze, which makes sense as they aren't as spry as the young ones.
• Stiffer Minds: The older mice, regardless of their family, became less likely to change their minds after a mistake. They were more stubborn. Interestingly, for the C57 mice, this stubbornness actually helped them in the "Hard Mode" because they stopped switching arms so frantically.

The Secret Sauce: How Their Brains Processed "Win" vs. "Loss"

The researchers used a computer model to peek inside the mice's decision-making brains. They found a key difference in how the two families processed rewards:

• CBA Mice: Their brains had a super-charged "Reward Detector." When they got a treat, their brain said, "YES! Remember this path!" very strongly. This made their memory of the correct path very sticky and hard to shake, even when they didn't get a treat for a while.
• C57 Mice: Their "Reward Detector" was weaker. They didn't learn as strongly from the good times. Instead, they were hyper-sensitive to the bad times (getting no treat). A single "no reward" made them doubt their entire strategy.

Why Does This Matter? (The "So What?")

This study is a huge reminder that not all mice are created equal.

If you are a scientist studying memory or learning:

• If you want to study stable, long-term memory (like remembering a route home), the CBA mouse might be your best friend because they stick to what they know.
• If you want to study how animals react to sudden changes or bad news, the C57 mouse is perfect because they react strongly to every little fluctuation.

The Bottom Line:
Genetics and age don't just change how fast an animal learns; they change how they think. The CBA mice were the steady, confident navigators who trusted their map. The C57 mice were the jittery navigators who kept checking the map every time the wind blew. Both learned the maze, but they took very different mental journeys to get there.

Technical Summary

1. Problem Statement

While it is well-established that mouse genetic background (strain) and age significantly influence behavior, sensory function, and hippocampal plasticity, the specific mechanisms by which these factors interact to shape spatial appetitive learning and decision-making strategies remain poorly defined.

• The Gap: Most studies treat mouse strain as a source of experimental "noise" rather than a biological variable. Furthermore, the interplay between genetic background (specifically comparing C57BL/6 and CBA/CaOlaHsd) and moderate aging (7–8 months) in the context of learning under uncertainty (probabilistic rewards) has not been thoroughly characterized.
• The Question: How do strain and age interact to affect the acquisition of spatial rules, the stability of learned behaviors when reward probability decreases, and the underlying trial-by-trial decision policies?

2. Methodology

Subjects and Design

• Animals: Male and female mice from two strains: C57BL/6 (n=28 total) and CBA/CaOlaHsd (n=26 total).
• Age Groups: Young adults (2–3 months) and moderately aged adults (7–8 months).
• Task: A 5-day T-maze appetitive spatial learning task.
  • Habituation: 5 days of free exploration and food association.
  • Training: 5 days, 20 trials/day, divided into 4 blocks of 5 trials.
  • Phases:
    1. Deterministic Phase (Days 1–3): 100% reward probability for the correct arm.
    2. Probabilistic Phase (Days 4–5): Reward probability progressively decreased (80% $\to$ 30%), while reward magnitude increased proportionally to maintain motivation. This created "correct but unrewarded" trials to test perseverance.

Behavioral Metrics

• Performance: Percentage of correct choices (accuracy) across blocks.
• Latency: Time taken to reach the goal zone.
• Decision Strategies: Trial-by-trial analysis of Stay/Shift behaviors conditional on previous outcomes (Win/Lose):
  • Win-Stay / Win-Shift
  • Lose-Stay / Lose-Shift

Computational Modeling

• Reinforcement Learning (RL) Model: A Rescorla-Wagner/Q-learning model with separate learning rates for positive and negative outcomes was fitted to individual trial data.
  • Parameters:
    • $\alpha_+$: Learning rate for rewarded outcomes.
    • $\alpha_-$: Learning rate for unrewarded outcomes (omissions/errors).
    • $k$: "Stickiness" (tendency to repeat or alternate choices independent of reward).
    • $\beta$: Inverse temperature (choice determinism).
    • $\epsilon$: Lapse rate (random errors).
• Validation: Parameter recovery was confirmed using synthetic datasets to ensure the model could accurately estimate latent variables.

Statistical Analysis

• Two-way ANOVA (fANOVA) and repeated-measures ANOVA (rANOVA) to test main effects of Strain, Age, and their interactions.
• Post-hoc Tukey tests for pairwise comparisons.
• Correlation analyses between model parameters and behavioral metrics.

3. Key Results

A. Spatial Learning Performance

• Strain Difference: CBA/CaOlaHsd mice learned significantly faster and achieved higher accuracy than C57BL/6 mice.
  • CBA mice outperformed C57BL/6 by the end of Day 1 and maintained this advantage.
  • Crucially, when reward probability dropped (Days 4–5), C57BL/6 performance declined, whereas CBA/CaOlaHsd mice maintained stable performance.
• Age Effect: Age alone did not significantly alter overall accuracy but interacted with strain. Young C57BL/6 mice showed the most severe performance drop in the final probabilistic blocks.

B. Latency and Execution Speed

• Age: Older mice had longer latencies overall, particularly in early training blocks.
• Strain: CBA/CaOlaHsd mice were consistently faster (lower latencies) than C57BL/6 mice during the deterministic phase, indicating more efficient task acquisition.

C. Trial-by-Trial Decision Strategies

• Adaptive Policies: CBA/CaOlaHsd mice adopted Win-Stay/Lose-Shift strategies earlier (by Day 1) compared to C57BL/6 mice.
• Probabilistic Phase (Uncertainty):
  • CBA/CaOlaHsd: Showed high Win-Stay and low Lose-Shift (specifically after correct but unrewarded trials). They tolerated omissions and persisted with the learned rule.
  • C57BL/6: Exhibited higher Lose-Shift probabilities, meaning they were more likely to abandon the correct arm after a single omission.
  • Age Interaction: Younger mice were more sensitive to omissions (higher shifting after losses) than older mice, who showed greater behavioral stability.

D. Reinforcement Learning Model Findings

• Value Representation: CBA/CaOlaHsd mice developed a larger value separation ($\Delta Q$) between the correct and incorrect arms, indicating a stronger internal representation of the correct rule.
• Learning Rates ($\alpha$):
  • $\alpha_+$ (Positive Outcomes): Significantly higher in CBA/CaOlaHsd. This indicates they learn more effectively from rewards, rapidly strengthening the preference for the correct arm.
  • $\alpha_-$ (Negative Outcomes): Younger CBA/CaOlaHsd mice showed higher $\alpha_-$ than younger C57BL/6, suggesting they also update strategies faster after errors.
• Stickiness ($k$): CBA/CaOlaHsd mice had more negative $k$ values, indicating a higher intrinsic tendency to alternate choices (exploration) independent of reward, yet they maintained the correct rule due to strong reward-driven updating.
• Correlation: High $\alpha_+$ correlated strongly with Win-Stay behavior and the ability to maintain the correct choice during the probabilistic phase.

4. Key Contributions

1. Strain-Specific Cognitive Profiles: The study provides quantitative evidence that CBA/CaOlaHsd mice are superior in forming stable reference memories for spatial tasks and maintaining them under uncertainty, whereas C57BL/6 mice are more reactive to immediate feedback fluctuations (omissions).
2. Mechanistic Insight via RL Modeling: By separating learning rates for positive and negative outcomes, the authors demonstrated that the superior performance of CBA mice is driven primarily by a higher learning rate for rewarded outcomes ($\alpha_+$), leading to stronger value accumulation for the correct choice.
3. Age-Strain Interaction: The study reveals that moderate aging (7–8 months) does not simply degrade performance but shifts decision policies. Older mice showed reduced sensitivity to omissions (less switching), which was beneficial in this stable-but-probabilistic task but might be detrimental in tasks requiring rapid reversal learning.
4. Sensory-Plasticity Link: The results support the hypothesis that the well-documented early-onset hearing loss and sensory deficits in C57BL/6 mice contribute to their altered hippocampal plasticity and less stable spatial learning strategies compared to the sensorily intact CBA strain.

5. Significance and Implications

• Experimental Design: The findings argue against treating mouse strain as a mere control variable. Researchers must carefully select strains based on the specific cognitive process under investigation:
  • CBA/CaOlaHsd: Ideal for studying stable reference memory, reinforcement learning, and decision-making under uncertainty.
  • C57BL/6: More suitable for studying sensitivity to immediate feedback, outcome fluctuations, and the impact of sensory decline on cognition.
• Aging Research: The study highlights that "aging" effects are not uniform; they depend heavily on the genetic background and the specific cognitive demands of the task (e.g., stability vs. flexibility).
• Neurobiological Mechanisms: The divergence in learning rates ($\alpha_+$) suggests fundamental differences in how these strains process dopaminergic or noradrenergic signals related to reward salience and synaptic plasticity in the hippocampus.

In conclusion, this paper demonstrates that genetic background and age fundamentally reshape not just how well mice learn, but how they learn—specifically regarding the balance between exploiting known rules and exploring new options in the face of uncertainty.