A computational approach to motivated behaviour and apathy

The loss of motivation and goal-directed behaviour is characteristic of apathy. Across a wide range of neuropsychiatric disorders, including Huntington’s disease (HD), apathy is poorly understood, associated with significant morbidity, and is hard to treat. One of the challenges in understanding the neural basis of apathy is moving from phenomenology and behavioural dysfunction to neural circuits in a principled manner. The computational framework offers one such approach. I adopt this framework to better understand motivated behaviour and apathy in four complementary projects. At the heart of many apathy formulations is impaired self-initiation of goal-directed behaviour. An influential computational theory proposes that “opportunity cost”, the amount of reward we stand to lose by not taking actions per unit time, is a key variable in governing the timing of self-initiated behaviour. Using a novel task, I found that free-operant behaviour in healthy participants both in laboratory conditions and in online testing, conforms to predictions of this computational model. Furthermore, in both studies I found that in younger adults sensitivity to opportunity cost predicted behavioural apathy scores. Similar pilot results were found in a cohort of patients with HD. These data suggest that opportunity cost may be an important computational variable relevant for understanding a core feature of apathy – the timing of self-initiated behaviour. In my second project, I used a reinforcement learning paradigm to probe for early dysfunction in a cohort of HD gene carriers approximately 25 years from clinical onset. Based on empirical data and computational models of basal ganglia function I predicted that asymmetry in learning from gains and losses may be an early feature of carrying the HD gene. As predicted, in this task fMRI study, HD gene carriers demonstrated an exaggerated neural response to gains as compared to losses. Gene carriers also differed in the neural response to expected value suggesting that carrying the HD gene is associated with altered processing of valence and value decades from onset. Finally, based on neurocomputational models of basal ganglia pathway function, I tested the hypothesis that apathy in HD would be associated with the involvement of the direct pathway. Support for this hypothesis was found in two related projects. Firstly, using data from a large international HD cohort study, I found that apathy was associated with motor features of the disease thought to represent direct pathway involvement. Secondly, I tested this hypothesis in vivo using resting state fMRI data and a model of basal ganglia connectivity in a large peri-manifest HD cohort. In keeping with my predictions, whilst emerging motor signs were associated with changes in the indirect pathway, apathy scores were associated with connectivity changes in the direct pathway connectivity within my model. For patients with apathy across neuropsychiatry there is an urgent need to understand the neural basis of motivated behaviour in order to develop novel therapies. In this thesis, I have used a computational framework to develop and test a range of hypotheses to advance this understanding. In particular, I have focussed on the computational factors which drive us to self-initiate, their potential neural underpinnings and the relevance of these models for apathy in patients with HD. The data I present supports the hypothesis that opportunity cost and basal ganglia pathway connectivity may be two important components necessary to generate motivated behaviour and contribute to the development of apathy in HD