Decision-making invokes two fundamental axes of control: affect or valence, spanning reward and punishment, and effect or action, spanning invigoration and inhibition. We studied the acquisition of instrumental responding in healthy human volunteers in a task in which we orthogonalized action requirements and outcome valence. Subjects were much more successful in learning active choices in rewarded conditions, and passive choices in punished conditions. Using computational reinforcement-learning models, we teased apart contributions from putatively instrumental and Pavlovian components in the generation of the observed asymmetry during learning. Moreover, using model-based fMRI, we showed that BOLD signals in striatum and substantia nigra/ventral tegmental area (SN/VTA) correlated with instrumentally learnt action values, but with opposite signs for go and no-go choices. Finally, we showed that successful instrumental learning depends on engagement of bilateral inferior frontal gyrus. Our behavioral and computational data showed that instrumental learning is contingent on overcoming inherent and plastic Pavlovian biases, while our neuronal data showed this learning is linked to unique patterns of brain activity in regions implicated in action and inhibition respectively.

Translating advances in neuroscience into benefits for patients with mental illness presents enormous challenges because it involves both the most complex organ, the brain, and its interaction with a similarly complex environment. Dealing with such complexities demands powerful techniques. Computational psychiatry combines multiple levels and types of computation with multiple types of data in an effort to improve understanding, prediction and treatment of mental illness. Computational psychiatry, broadly defined, encompasses two complementary approaches: data driven and theory driven. Data-driven approaches apply machine-learning methods to high-dimensional data to improve classification of disease, predict treatment outcomes or improve treatment selection. These approaches are generally agnostic as to the underlying mechanisms. Theory-driven approaches, in contrast, use models that instantiate prior knowledge of, or explicit hypotheses about, such mechanisms, possibly at multiple levels of analysis and abstraction. We review recent advances in both approaches, with an emphasis on clinical applications, and highlight the utility of combining them.

When planning a series of actions, it is usually infeasible to consider all potential future sequences; instead, one must prune the decision tree. Provably optimal pruning is, however, still computationally ruinous and the specific approximations humans employ remain unknown. We designed a new sequential reinforcement-based task and showed that human subjects adopted a simple pruning strategy: during mental evaluation of a sequence of choices, they curtailed any further evaluation of a sequence as soon as they encountered a large loss. This pruning strategy was Pavlovian: it was reflexively evoked by large losses and persisted even when overwhelmingly counterproductive. It was also evident above and beyond loss aversion. We found that the tendency towards Pavlovian pruning was selectively predicted by the degree to which subjects exhibited sub-clinical mood disturbance, in accordance with theories that ascribe Pavlovian behavioural inhibition, via serotonin, a role in mood disorders. We conclude that Pavlovian behavioural inhibition shapes highly flexible, goal-directed choices in a manner that may be important for theories of decision-making in mood disorders.

Hard-wired, Pavlovian, responses elicited by predictions of rewards and punishments exert significant benevolent and malevolent influences over instrumentally-appropriate actions. These influences come in two main groups, defined along anatomical, pharmacological, behavioural and functional lines. Investigations of the influences have so far concentrated on the groups as a whole; here we take the critical step of looking inside each group, using a detailed reinforcement learning model to distinguish effects to do with value, specific actions, and general activation or inhibition. We show a high degree of sophistication in Pavlovian influences, with appetitive Pavlovian stimuli specifically promoting approach and inhibiting withdrawal, and aversive Pavlovian stimuli promoting withdrawal and inhibiting approach. These influences account for differences in the instrumental performance of approach and withdrawal behaviours. Finally, although losses are as informative as gains, we find that subjects neglect losses in their instrumental learning. Our findings argue for a view of the Pavlovian system as a constraint or prior, facilitating learning by alleviating computational costs that come with increased flexibility.

BackgroundDepression is characterised partly by blunted reactions to reward. However, tasks probing this deficiency have not distinguished insensitivity to reward from insensitivity to the prediction errors for reward that determine learning and are putatively reported by the phasic activity of dopamine neurons. We attempted to disentangle these factors with respect to anhedonia in the context of stress, Major Depressive Disorder (MDD), Bipolar Disorder (BPD) and a dopaminergic challenge.MethodsSix behavioural datasets involving 392 experimental sessions were subjected to a model-based, Bayesian meta-analysis. Participants across all six studies performed a probabilistic reward task that used an asymmetric reinforcement schedule to assess reward learning. Healthy controls were tested under baseline conditions, stress or after receiving the dopamine D2 agonist pramipexole. In addition, participants with current or past MDD or BPD were evaluated. Reinforcement learning models isolated the contributions of variation in reward sensitivity and learning rate.ResultsMDD and anhedonia reduced reward sensitivity more than they affected the learning rate, while a low dose of the dopamine D2 agonist pramipexole showed the opposite pattern. Stress led to a pattern consistent with a mixed effect on reward sensitivity and learning rate.ConclusionReward-related learning reflected at least two partially separable contributions. The first related to phasic prediction error signalling, and was preferentially modulated by a low dose of the dopamine agonist pramipexole. The second related directly to reward sensitivity, and was preferentially reduced in MDD and anhedonia. Stress altered both components. Collectively, these findings highlight the contribution of model-based reinforcement learning meta-analysis for dissecting anhedonic behavior.

The acquisition of reward and the avoidance of punishment could logically be contingent on either emitting or withholding particular actions. However, the separate pathways in the striatum for go and no-go appear to violate this independence, instead coupling affect and effect. Respect for this interdependence has biased many studies of reward and punishment, so potential action-outcome valence interactions during anticipatory phases remain unexplored. In a functional magnetic resonance imaging study with healthy human volunteers, we manipulated subjects' requirement to emit or withhold an action independent from subsequent receipt of reward or avoidance of punishment. During anticipation, in the striatum and a lateral region within the substantia nigra/ventral tegmental area (SN/VTA), action representations dominated over valence representations. Moreover, we did not observe any representation associated with different state values through accumulation of outcomes, challenging a conventional and dominant association between these areas and state value representations. In contrast, a more medial sector of the SN/VTA responded preferentially to valence, with opposite signs depending on whether action was anticipated to be emitted or withheld. This dominant influence of action requires an enriched notion of opponency between reward and punishment.

Pavlovian biases influence learning and decision making by intricately coupling reward seeking with action invigoration and punishment avoidance with action suppression. This bias is not always adaptive; it can oftentimes interfere with instrumental requirements. The prefrontal cortex is thought to help resolve such conflict between motivational systems, but the nature of this control process remains unknown. EEG recordings of mid-frontal theta band power are sensitive to conflict and predictive of adaptive control over behavior, but it is not clear whether this signal would reflect control over conflict between motivational systems. Here we utilized a task that orthogonalized action requirements and outcome valence while recording concurrent EEG in human participants. By applying a computational model of task performance, we derived parameters reflective of the latent influence of Pavlovian bias and how it was modulated by mid-frontal theta power during motivational conflict. Between subjects, individuals who performed better under Pavlovian conflict exhibited higher mid-frontal theta power. Within subjects, trial-to-trial variance in theta power was predictive of ability to overcome the influence of the Pavlovian bias, and this effect was most pronounced in individuals with higher mid-frontal theta to conflict. These findings demonstrate that mid-frontal theta is not only a sensitive index of prefrontal control, but it can also reflect the application of top-down control over instrumental processes.

Subjects with schizophrenia are impaired at reinforcement-driven reversal learning from as early as their first episode. The neurobiological basis of this deficit is unknown. We obtained behavioral and fMRI data in 24 unmedicated, primarily first episode, schizophrenia patients and 24 age-, IQ- and gender-matched healthy controls during a reversal learning task. We supplemented our fMRI analysis, focusing on learning from prediction errors, with detailed computational modeling to probe task solving strategy including an ability to deploy an internal goal directed model of the task. Patients displayed reduced functional activation in the ventral striatum (VS) elicited by prediction errors. However, modeling task performance revealed that a subgroup did not adjust their behavior according to an accurate internal model of the task structure, and these were also the more severely psychotic patients. In patients who could adapt their behavior, as well as in controls, task solving was best described by cognitive strategies according to a Hidden Markov Model. When we compared patients and controls who acted according to this strategy, patients still displayed a significant reduction in VS activation elicited by informative errors that precede salient changes of behavior (reversals). Thus, our study shows that VS dysfunction in schizophrenia patients during reward-related reversal learning remains a core deficit even when controlling for task solving strategies. This result highlights VS dysfunction is tightly linked to a reward-related reversal learning deficit in early, unmedicated schizophrenia patients.