A Neurally Constrained Computational Model of Context-Dependent Fear Extinction and Relapse

Abstract

Fear conditioning involves repeatedly pairing a neutral stimulus (conditioned stimulus or CS), like a musical tone, with a naturally threatening stimulus, like a painful shock (unconditioned stimulus or US) until the animal acquires a fear response to the neutral stimulus. The animal’s fear response to the CS can be extinguished by repeatedly presenting the CS without a shock. However, fear extinction is context-specific and temporary, with the fear associated with a CS often relapsing in spatial contexts other than where extinction took place (Fear Renewal), and even within the extinction context following a substantial delay (Spontaneous Recovery). The relapse of fear following extinction is important clinically, particularly for exposure therapy, a common behavioral intervention for anxiety and stress disorders. Exposure therapy relies on fear extinction mechanisms, since it involves gradually exposing the patient to feared stimuli or events to reduce their fear responses. Understanding the mechanisms underlying fear extinction and relapse in the brain could lend important insights into the long term suppression of fear, increasing the effectiveness and longevity of exposure therapy outcomes. These mechanisms have been extensively explored in animals. Recent advances in molecular neuroscience have enabled the detection of distinct memory engrams responsible for encoding fear and extinction memories in rodents, and provided insights into how they might be formed, modified and activated. However, given the invasive nature of these experiments, they cannot be extended to understand parallel mechanisms in humans. Additionally, these experiments tend to study individual mechanisms, or the role of distinct brain regions, in isolation, suggesting the need for a unifying framework to situate their contributions in fear learning. In this dissertation, we introduce a neurally grounded, mechanistic computational model of context-dependent fear extinction and relapse (ConFER). ConFER integrates some of these key animal findings within its assumptions and mechanisms to successfully replicate empirical findings on fear renewal and spontaneous recovery, and also draws on its assumptions to generate novel hypotheses that can be behaviorally tested in humans, with insights for improving clinical interventions. Chapter 1 motivates the need for ConFER, discussing its neuroscientific foundations. It also motivates the need to extend ConFER to incorporate the role of context similarity in determining the degree of fear relapse (ConFER-Sim). Chapter 2 provides an overview of ConFER’s architecture, core assumptions and learning mechanisms. It also provides a detailed technical description of ConFER’s implementation. Chapter 3 describes ConFER’s simulations in replicating empirical findings on fear renewal and spontaneous recovery. It also proposes a novel hypothesis comparing the effectiveness of counterconditioning – pairing a feared CS with a positive US– with fear extinction in suppressing fear relapse. Chapter 4 discusses extending ConFER to incorporate a context-similarity module within it (ConFER-Sim). It includes the neuroscientific and psychological rationale for the context-similarity module, and its technical implementation. We find that ConFER-Sim successfully replicates empirical results on the role of context similarity in fear renewal. Chapter 4 also discusses ConFER-Sim’s novel hypotheses on the interaction of the number of extinction contexts, and the degree of similarity between extinction and test contexts, on suppressing fear renewal. Finally, Chapter 5 engages in a general discussion on ConFER and ConFER-Sim, discussing their contributions, modeling choices, limitations and future directions.