PhD Exit Seminar for Sankirthana Sathiyakumar (Richards Lab)

Neurobiological and Computational Approaches to Behavioural Flexibility

Abstract

Behavioral flexibility is important in a changing environment. Prior experiences are stored as memories which can be utilized to make decisions in novel environments. These memories can promote or reduce behavioural flexibility in novel environments. The hippocampus is intimately involved with learning and memory and its engagement in this process depends on the age of the memory. Taking these ideas into consideration, I developed an automated Y-maze to test behavioural flexibility in response to a variety of changes in the environment. Mice were trained on a random shift paradigm to ensure that they could continuously explore the Y-maze and learn to poke at the end of the arms to receive rewards. Next, they were tested on a textural discrimination paradigm to ensure that they could learn texture-reward associations that were constantly shifting in the maze.

In order to indirectly investigate hippocampal contributions to behavioural flexibility we tested how systems consolidation affects flexibility in response to: (1) value changes and (2) changes in the optimal sequence of actions. Mice were trained to obtain rewards in a Y-maze by switching nose pokes between three arms. During training, all arms were rewarded and mice simply had to switch arms in order to maximize rewards. Then, after a 1 or 28 day delay, we devalued one arm, or we reinforced a specific sequence of pokes. We found that after a 1 day delay mice adapted easily to the changes. In contrast, mice given a 28 day delay struggled to adapt, especially for changes to the optimal sequence of actions. Immediate early gene imaging suggested that the 28 day mice were less reliant on their hippocarnpus and more reliant on their medial prefrontal cortex.

Lastly, I used reinforcement learning to understand the possible computations underlying the facilitation of behavioral flexibility seen in these tasks. Model-free and model-based algorithms were trained and tested in virtual environments that paralleled the setup of the Y-maze experiments. I found that model-based algorithms were able to adapt more efficiently to action devaluations and optimal sequence changes suggesting that the hippocarnpus may be performing model-based reinforcement learning to facilitate behavioural flexibility.

—————————————————————-

Join Zoom Meeting

Thursday, December 16^th, 2021 @ 2:10 pm

https://utoronto.zoom.us/j/85904575365

Meeting ID: 859 0457 5365

Host: Blake A. Richards (blake.richards@utoronto.ca)

—————————————————————-