Predicting effects of atomoxetine and citalopram in Parkinson's disease
1. Accurate Predictive Models of Atomoxetine’s and Citalopram’s
Effect on Motor Inhibition in Parkinson’s Disease
Z Ye1,6, CL Rae1,2, C Nombela1, T Ham1, T Rittman1, PS Jones1, P Vázquez Rodríguez1, Ian Coyle-Gilchrist1, R Regenthal3, E
Altena1, CR Housden1, H Maxwell1, BJ Sahakian4,5, RA Barker1, Trevor Robbins1,4, JB Rowe1,2,4
1University of Cambridge Department of Clinical Neurosciences 2MRC Cognition and Brain Sciences Unit 3University of Leipzig 4Behavioural and Clinical Neuroscience
Institute 5University of Cambridge Department of Psychiatry 6Donders Institute for Brain, Cognition and Behaviour
Contact Dr Rowe
James.Rowe@mrc-cbu.cam.ac.uk
Measures PD Control Diff.
Sex (male:female) 21:13 23:19 n.s.
Age (years) 67 (7) 67 (7) n.s.
Cognitive impairment (MMSE) 29 (2) 29 (1) n.s.
Duration of symptoms (years) 9 (6) -- --
Hoehn & Yahr 2.0 (0.6) -- --
Disease severity (UPDRS-III) 23 (8) -- --
L-dopa equivalent dose
(mg/day)
547 (304) -- --
Atomoxetine level (ng/ml) 402 (179) -- --
Citalopram level (ng/ml) 40 (16) -- --
Background
Motor inhibition deficits occur in Parkinson’s disease (PD) even in
the absence of impulse control disorders. Their aetiology is
multifactorial, including the loss of noradrenergic and seroton-
ergic projects to the forebrain and structural change in fronto-
striatal circuits1-3.
Our recent studies have showed that selective noradrenaline
(atomoxetine) and serotonin (citalopram) reuptake inhibitors
improve motor inhibition in a subgroup of PD patients, in
terms of both performance and brain activity.4,5. However, the
impact of treatment depends on individual differences in
clinical characteristics and brain imaging.
Go
Stop
Variable
delay
max. 1000 ms
Control
(stop>Go)
0 5
Primarily funded by
References
1. Goldstein et al. 2011. Eur J Neurol
2. Politis et al. 2010. Neurob Dis
3. Rae et al. 2014. Neuroimage
4. Ye et al. 2014. Biol Psyhiat
5. Ye et al. 2014. Brain
Methods & Results
• Double-blind randomized placebo-controlled crossover study
with 34 PD patients (three sessions: 40mg atomoxetine, 30mg
citalopram or placebo) and 42 controls (1 session, no drug).
• Stop task: 360 Go trials, 80 stop trials (tracking algorithm to
50% inhibition, figure); SSRT estimated by integration method;
confirmed poor motor inhibition in PD (figure). Large individual
variability led to no main effect of either drug.
• DTI: 3-T, 63 directions, analysed with FSL; fractional anisotropy
and mean diffusivity extracted from the anterior internal
capsule (frontostriatal connection).
• fMRI during the stop task: 3-T, whole-brain, 3mm resolution,
analyzed with SPM. See figure for main effects of task and PD;
parameter estimates extracted from the right inferior frontal
gyrus, bilateral pre-supplementary motor areas and striatum.
• Benchmark for behavioural efficacy: 30% reduction in the
impact of PD on the SSRT.
• Support vector machine with a radial function kernel; leave-
one-out cross validation; inferences from 5000 permutations;
79-85% accuracy in predicting the drug response.
This Study
We developed two predictive models to identify patients who are
most likely to benefit from these new candidate therapies.
Clinical model which used demographic (sex, age) and clinical
measures (cognitive impairment, disease severity, levodopa
equivalent dose, plasma concentration of each drug), in
combination with structural brain imaging data.
Mechanistic model which used the demographic and clinical
data, plus functional brain imaging measures.
400
500
600
Ctrl PD-PLA PD-ATO PD-CIT
GoRT(ms)
*
Disease effects
PD-PLA<Control
(stop>Go)
Optimal models for 30% improvement Accuracy Perm.
Features for optimal Atomoxetine prediction
Clinical: R fractional anisotropy, drug
concentration, levodopa dose
79.4% <0.005
Mechanistic: activation in R caudate
nucleus, L caudate nucleus, R pre-SMA
85.3% <0.001
Features for optimal Citalopram prediction
Clinical: R fractional anisotropy, age, R
mean diffusivity, cognition
79.4% <0.005
Mechanistic: activation in L caudate
nucleus, R putamen, R pre-SMA
85.3% <0.005
Conclusion
• Increasing noradrenaline or serotonin improves stopping
efficiency of a subgroup of patients with Parkinson’s disease.
• The clinical model is of relevance to prospective trialists and
clinicians, to enable patient stratification in future clinical trials
and for potential treatment decision for individual patients.
• The mechanistic model provides the link to preclinical and
comparative studies noradrenergic and serotonergic systems
for cognitive control.
110
160
210
Ctrl PD-PLA PD-ATO PD-CIT
SSRT(ms)
*
90
80
70
60
50
10 20 30 40 50
Benchmark (30%)
10 20 30 40 50
Cross-validation
accuracy(%)
RobustnessBest model performance
ATO-clinical
ATO-mechanistic
CIT-clinical
CIT-mechanistic
Stop-signal task
Predictive models