Long-Term Latent Dynamics and Cortical Interactions in a Dynamic Bandit Task
Open access
Author
Date
2023Type
- Doctoral Thesis
ETH Bibliography
yes
Altmetrics
Abstract
Value-based decision-making is a complex process involving multiple brain areas. The medial prefrontal cortex (mPFC) and the ventrolateral orbitofrontal cortex (vlOFC) are two cortical areas central to value encoding, reward processing, and action selection. Despite ongoing research, the specific roles of these regions in encoding decision-related variables remain elusive. Furthermore, the nature of their interaction during the decision-making process is still a subject of investigation. In this dissertation, I present a series of studies investigating how value, choice, and reward are encoded in the mPFC and the vlOFC. Using simultaneous, large- scale recordings of single neurons in the mPFC and vlOFC of rats performing a two-armed bandit task, I characterized how each decision variable is encoded at the neuronal and population levels. Most critically, by utilizing a novel analysis method for high- dimensional neural data, I examined the interplay between the mPFC and the vlOFC during decision-making processes. The findings provided insights into the inter-population interaction dynamics of these brain areas, and advanced our understanding of neural mechanisms underlying decision-making.
Chapter 2 discusses the implementation and validation of a novel recording paradigm for long-term stable, large-scale single neuron recordings in rats across multiple brain areas using ultraflexible arrays. Utilizing ultraflexible arrays, I managed to track the activity of the same populations of neurons from mPFC and vlOFC for more than three weeks.
Chapter 3 explores how decision-related variables are encoded in the mPFC and vlOFC, and how inter-population correlation dynamics change during value-based decision-making. The results reveal distinct preferences of neuronal populations in the mPFC and vlOFC for encoding choice, outcome, and inferred values. The
1
mPFC demonstrates a more persistent and pronounced encoding of choice across the task compared to the vlOFC, whereas the vlOFC more robustly encodes for reward. Action values are encoded in single neuron activity and the latent population dynamics of both the mPFC and vlOFC. The inter-population correlation dynamics between the mPFC and vlOFC vary with task stages, demonstrating high correlation during action execution and low correlation during action selection and reward processing. There are temporal lags between the population activities in the mPFC and vlOFC, which vary in directionality at certain behavioral stages. The findings suggest variable temporal engagement of mPFC and vlOFC during different stages of the task.
Chapter 4 examines the long-term characteristics and cross-animal generalizability of neural representations encoding choice and outcome. Single neuron responses and latent population dynamics across consecutive recording sessions are closely correlated. However, the degree of correlation between sessions appears to decrease gradually over the course of days. Latent population dynamics are generalizable across animals, as choice and reward information can be reliably decoded from aligned latent population dynamics of two different rats.
Taken together, this dissertation offers insights into how the mPFC and vlOFC encode decision variables and interact during value- based decision-making, and the long-term stability and generalizability of these neural representations. These findings enhance our understanding of inter-cortical interactions that occur during adaptive, reward-based behaviors in dynamic environments. Show more
Permanent link
https://doi.org/10.3929/ethz-b-000650608Publication status
publishedExternal links
Search print copy at ETH Library
Publisher
ETH ZurichOrganisational unit
09474 - Yanik, Mehmet Fatih / Yanik, Mehmet Fatih
More
Show all metadata
ETH Bibliography
yes
Altmetrics