Predicting beneficial effects of atomoxetine and citalopram on response inhibition in Parkinson's disease with clinical and neuroimaging measures.

Recent studies indicate that selective noradrenergic (atomoxetine) and serotonergic (citalopram) reuptake inhibitors may improve response inhibition in selected patients with Parkinson's disease, restoring behavioral performance and brain activity. We reassessed the behavioral efficacy of these drugs in a larger cohort and developed predictive models to identify patient responders. We used a double-blind randomized three-way crossover design to investigate stopping efficiency in 34 patients with idiopathic Parkinson's disease after 40 mg atomoxetine, 30 mg citalopram, or placebo. Diffusion-weighted and functional imaging measured microstructural properties and regional brain activations, respectively. We confirmed that Parkinson's disease impairs response inhibition. Overall, drug effects on response inhibition varied substantially across patients at both behavioral and brain activity levels. We therefore built binary classifiers with leave-one-out cross-validation (LOOCV) to predict patients' responses in terms of improved stopping efficiency. We identified two optimal models: (1) a "clinical" model that predicted the response of an individual patient with 77-79% accuracy for atomoxetine and citalopram, using clinically available information including age, cognitive status, and levodopa equivalent dose, and a simple diffusion-weighted imaging scan; and (2) a "mechanistic" model that explained the behavioral response with 85% accuracy for each drug, using drug-induced changes of brain activations in the striatum and presupplementary motor area from functional imaging. These data support growing evidence for the role of noradrenaline and serotonin in inhibitory control. Although noradrenergic and serotonergic drugs have highly variable effects in patients with Parkinson's disease, the individual patient's response to each drug can be predicted using a pattern of clinical and neuroimaging features.The BCNI is supported by the Wellcome Trust and Medical Research Council. We are grateful to Dr Gordon Logan for advice on stop-signal reaction time estimation and to Dr Marta Correia for advice on diffusion-weighted imaging data analysis. Conflict of interest: Prof. Sahakian has received grants from Janssen/J&J, personal fees from Cambridge Cognition, personal fees from Lundbeck, and personal fees from Servier, outside the submitted work. Prof. Robbins has received personal fees and royalties from Cambridge Cognition, personal fees and grants from Eli Lilly Inc, personal fees and grants from Lundbeck, grants from GSK, personal fees from Teva Pharmaceuticals, personal fees from Shire Pharmaceuticals, grants from Medical Research Council, editorial honorarium from Springer Verlag Germany, and personal fees from Chempartners, outside the submitted work. Prof. Rowe has received grant funding from AZ-Medimmune unrelated to the current work. Dr Housden is an employee of Cambridge Cognition. Other authors reported no biomedical financial interests or potential conflict of interest.This is the final version of the article. It was first available from Wiley via http://dx.doi.org/10.1002/hbm.2308