The effects of graded levels of calorie restriction: IV. Non-linear change in behavioural phenotype of mice in response to short-term calorie restriction

David Lusseau, Sharon E. Mitchell, Ceres Barros, Davina Derous, Cara Green, Luonan Chen, Jing‐Dong J. Han, Yingchun Wang, Daniel Promislow, Alex Douglas, John R. SpeakmanScientific 2015 · DOI

TL;DR

This paper introduces a novel approach using Hidden Markov Models (HMMs) to analyze mouse behavior under varying levels of calorie restriction (CR). They discovered that mice don't just change what they do, but rather how they perform their activities, with significant shifts in behavioral phenotypes observed at higher CR levels (30-40%). This work provides a new way to link CR to healthspan by detailing the nuanced behavioral adaptations mice employ.

What Problem Does This Paper Address?

Understanding how animals adapt their behavior to environmental challenges, particularly food scarcity, is crucial for comprehending their physiology and survival strategies. Caloric restriction (CR) is known to extend lifespan and confer health benefits, but the underlying mechanisms, especially how behavior mediates these effects, are not fully understood. Previous studies often looked at specific behaviors or reactions to stimuli. This paper aims to provide a more holistic view of behavioral changes under CR by developing a method to capture the dynamic, state-dependent nature of an animal's daily activity patterns. They specifically wanted to see if CR leads to a simple reduction in activity or a more complex shift in how activities are performed.

Methods

The core of this research lies in the application of Hidden Markov Models (HMMs) to analyze continuous behavioral data from mice.

  • Data Collection: Mice were fitted with telemetry devices to record movement (activity count) and body temperature at 15-minute intervals over a three-month period.
  • Behavioral Phenotyping: HMMs were used to model the observed movement and temperature data as emanating from underlying, unobserved "activity states." These models allow for the characterization of distinct behavioral states (e.g., active, inactive, torpor) and the probabilities of transitioning between them.
  • Model Development:
    • HMM1 (Baseline Model): This initial model assumed a fixed number of states and estimated their characteristics (mean and variance of movement and temperature) and the transition probabilities between them.
  • Covariate Models (HMM2-HMM5): To account for changes over time, they developed models that incorporated covariates like time elapsed since CR started (t) and body weight (BW). These models allowed for changes in either the characteristics of the activity states (MVN parameters) or the transition probabilities (tpm) over time or with changes in body weight.
  • Model Selection: The Akaike Information Criterion (AIC) was used to select the best-fitting models for each individual mouse.
  • Interpretation: State characteristics were visualized by simulating observations from the fitted models. The effect of CR on phenotypic changes was tested by fitting multivariate linear models to the paired differences in state characteristics between the baseline and treatment periods.

Datasets

The study utilized data collected from C57BL/6 male mice.

  • Experimental Groups: Mice were divided into six groups: a control group with 24-hour food access (24AL), a control group with 12-hour food access (12AL), and four CR groups with 10%, 20%, 30%, and 40% calorie reduction.
  • Data Type: Continuous telemetry data (movement counts and body temperature) collected over a three-month period.
  • No New Benchmarks: The paper does not introduce a new benchmark dataset but rather applies novel analytical methods to existing types of behavioral data.

Key Insights and Ideas

  • Activity Characteristics vs. Activities: The most striking insight is that mice under CR primarily changed how they performed their activities (i.e., the characteristics of their behavioral states) rather than simply changing which activities they engaged in or the overall duration of those activities. This suggests a more subtle adaptation than a simple reduction in movement.
  • Non-Linear Response to CR: The behavioral changes were not linear with increasing CR levels. While 10% and 20% CR showed minimal behavioral shifts compared to controls, 40% CR induced significant changes in all mice, and 30% CR affected some mice. This highlights a potential threshold effect.
  • Emergence of Torpor as a State Shift: Torpor, characterized by very low activity and body temperature, wasn't a distinct new state but rather an extreme manifestation of changes within an existing state (state 3), becoming more pronounced with higher CR.
  • Body Weight as a Key Driver: Changes in body weight were strongly correlated with shifts in behavioral phenotypes, suggesting that the energetic consequences of CR, reflected in weight loss, are a primary driver of behavioral adaptation.
  • Holistic Behavioral Phenotyping: The HMM approach provides a powerful, integrated way to characterize an animal's overall behavioral phenotype, moving beyond single-measure metrics to capture complex temporal dynamics and state transitions.

Key Results

  • State Characterization: The best-fitting models consistently identified three activity states for mice, characterized by different levels of movement and body temperature. State 1 was high activity/high temperature, State 2 was lower activity/lower temperature, and State 3 was very low activity/very low temperature (approaching torpor).
  • CR-Induced Changes:
    • At 40% CR, all mice exhibited a significant shift in their behavioral phenotype.
  • At 30% CR, some mice showed significant behavioral shifts, indicating variability in response.
  • 10% and 20% CR had minimal impact on behavioral phenotype compared to control groups.
  • Torpor: Torpor was observed as a progressive change within State 3, particularly at 30% and 40% CR, rather than a completely new behavioral state.
  • Transition Probabilities: While transition probabilities between states didn't always change consistently with CR levels, the characteristics of the states themselves did, especially at higher CR.
  • Body Weight Correlation: Changes in activity state characteristics were significantly influenced by body weight changes, particularly for individuals experiencing greater weight loss under CR.

Limitations and Future Work

  • Short-Term CR: The study focused on short-term CR (three months). The authors suggest investigating whether progressive changes observed at higher CR levels might also emerge at lower CR levels if exposure is prolonged.
  • Variability at 30% CR: The observed variability in response at 30% CR warrants further investigation at a molecular level to understand the mechanisms behind this divergence.
  • Diurnal Patterns: While the models captured diurnal cycles as emergent properties, they were not explicitly incorporated as covariates. Future work could explore how CR affects these inherent daily rhythms more directly.
  • Healthspan Link: While the paper provides a new avenue for understanding CR's link to healthspan through behavioral changes, further research is needed to directly connect these specific behavioral adaptations to improved health outcomes.

Summary updated

Abstract

Animals have to adjust their activities when faced with caloric restriction (CR) to deal with reduced energy intake. If CR is pronounced, allostasis can push individuals into alternate physiological states which can result in important health benefits across a wide range of taxa. Here we developed a new approach to determine the changes in behavioural phenotype associated with different levels of CR. We exposed C57BL/6 male mice to graded CR (from 0 to 40%) for three months and defined their behavioural phenotype using hidden Markov models of their movement and body temperature. All 40% CR mice exhibited a state-shift in behavioural phenotype and only some exposed to 30% CR did. We show for the first time that mice changed their activity characteristics rather than changed their activities. This new phenotyping approach provides an avenue to determine the mechanisms linking CR to healthspan.
BibTeX
@inproceedings{lusseau2015effects,
  author = {David Lusseau and Sharon E. Mitchell and Ceres Barros and Davina Derous and Cara Green and Luonan Chen and Jing‐Dong J. Han and Yingchun Wang and Daniel Promislow and Alex Douglas and John R. Speakman},
  title = {{The effects of graded levels of calorie restriction: IV. Non-linear change in behavioural phenotype of mice in response to short-term calorie restriction}},
  booktitle = {Scientific},
  year = {2015},
  doi = {10.1038/srep13198},
}
Figure