RationaleDisorders of compulsivity such as stimulant use disorder (SUD) and
obsessive-compulsive disorder (OCD) are characterised by deficits in behavioural
flexibility, some of which have been captured using probabilistic reversal
learning (PRL) paradigms.ObjectivesThis study used computational modelling to characterise the
reinforcement learning processes underlying patterns of PRL behaviour observed in
SUD and OCD and to show how the dopamine D2/3 receptor
agonist pramipexole and the D2/3 antagonist amisulpride
affected these responses.MethodsWe applied a hierarchical Bayesian method to PRL data across three
groups: individuals with SUD, OCD, and healthy controls. Participants completed
three sessions where they received placebo, pramipexole, and amisulpride, in a
double-blind placebo-controlled, randomised design. We compared seven models using
a bridge sampling estimate of the marginal likelihood.ResultsStimulus-bound perseveration, a measure of the degree to which
participants responded to the same stimulus as before irrespective of outcome, was
significantly increased in SUD, but decreased in OCD, compared to controls (on
placebo). Individuals with SUD also exhibited reduced reward-driven learning,
whilst both the SUD and OCD groups showed increased learning from punishment
(nonreward). Pramipexole and amisulpride had similar effects on the control and
OCD groups; both increased punishment-driven learning. These
D2/3-modulating drugs affected the SUD group differently,
remediating reward-driven learning and reducing aspects of perseverative
behaviour, amongst other effects.ConclusionsWe provide a parsimonious computational account of how perseverative
tendencies and reward- and punishment-driven learning differentially contribute to
PRL in SUD and OCD. D2/3 agents modulated these processes
and remediated deficits in SUD in particular, which may inform therapeutic
effects.Electronic supplementary materialThe online version of this article (10.1007/s00213-019-05325-w) contains supplementary material, which is available to authorized
users.