|Home | About | Journals | Submit | Contact Us | Français|
Relatively little is known about the interaction between behavioural changes, medication and cognitive function in Parkinson’s disease. We examined working memory, learning and risk aversion in patients with Parkinson’s disease (PD) with and without impulsive or compulsive behaviour and compared to a group of age-matched control subjects. Parkinson patients with impulsive or compulsive behaviour (PD+ ICB) had poorer working memory performance than either controls or PD patients without ICB. PD+ICB patients also showed decreased learning from negative feedback and increased learning from positive feedback off compared to on dopaminergic medication. This interaction between medication status and learning was the opposite of that found in the PD patients without a diagnosis of ICB. Finally, the PD group showed increased risk preference on medication relative to controls and the subgroup of PD+ICB patients with pathological gambling were overall more risk prone than the PD group. Thus, medication status and an impulsive behavioural diagnosis differentially affect several behaviors in PD.
The relationship between dopamine levels and cognitive function in Parkinson’s disease (PD) has been the subject of much recent interest1, 2, but there has been less work examining the interaction between behavioural changes and cognitive function3. Dopamine replacement therapy is effective in treating the motor symptoms of PD, but can lead to impulsive or compulsive behaviours (ICB) in a minority of patients4. These behaviours include pathological gambling (PG), hypersexuality, compulsive shopping and binge eating. Punding and the compulsive overuse of L-dopa termed the dopamine dysregulation syndrome (DDS) are related phenomena also found in patients on dopamine replacement therapy4–6. A number of risk factors for the development of ICBs are known, including male sex, young age at onset (under 45years), high novelty seeking personalities and a previous history of alcoholism or addictive behaviour7.
In early Parkinson’s disease there is an uneven distribution of dopamine loss, with greater depletion in the dorsal striatum than in the ventral striatum8. Effective dopamine replacement in the dorsal striatum designed to reverse bradykinesia might, therefore, over-stimulate the relatively intact ventral striatum and lead to undesirable cognitive changes9. A number of behavioural correlates of medication status (off vs. on) have been documented. For example, PD patients are more sensitive to negative feedback and less sensitive to positive feedback when off medication, whereas they show the opposite behaviour on medication10.
We hypothesized that the behavioural profile off medication in PD+ ICB patients would come to resemble that of PD patients on medication, with relatively enhanced learning from positive feedback compared to negative feedback.
Patients were recruited from a database of attendees at the National Hospital for Neurology and Neurosurgery Queen Square, London. Controls were usually recruited from amongst the patient’s spouses or partners. Written informed consent was obtained from all participants. Patients who scored under 27/30 points on the Mini Mental State Examination (MMSE) 11 were excluded from this study. Four controls performed just the working memory test. Two controls did not perform the computer tests. Two participants (1 control and 1 PD patient) were excluded from the behavioural modelling (described below) because the model(s) failed to converge for their data. All patients were asked to take no anti-Parkinsonian medication overnight and were tested first between 8.00am and 9.00am prior to their morning medication. Patients then took their first L-dopa dose of the day and the tests were repeated 50 minutes later. The therapeutic motor response to L-dopa was assessed by UPDRS scores (PART 3) during “off” and “on” state. All patients had an excellent L-dopa response and had switched “on” at the time of the second test. LEU (Levodopa equivalent units) were calculated as described previously.12 Testing was performed in the patient’s homes using a laptop computer. Distractions were minimized as much as possible, so full attention could be devoted to the task. Controls were tested following a similar sequence, i.e. they were tested once, and then tested again after a 50 minute delay, but received no medication.
The first task was a forward and backward digit span test13 to assess working memory. The second task was an instrumental learning task in which subjects were required, in each of four blocks of trials, to learn which of two stimuli was most often rewarded14, 15. In each trial they selected one stimulus and were then told whether or not they “won” on that trial. Winning probabilities for the two stimuli of 75%/25% and 65%/35% were constant throughout each block and balanced across stimuli across blocks (see supplementary material for additional details).
The final task was a gambling task which was designed to probe the risk aversion of the subjects and programmed to match the description given of the task in Huettel et al 16. In each trial subjects were given a choice between two gambling options which were presented on the left and right of the screen. Each option had either a single sure outcome, or two possible outcomes. The probabilities associated with each outcome were represented by a pie. For example, if the subjects had a 25% chance of winning £20 and a 75% chance of winning £5, the pie would be split 75/25, with the winning amount represented in each pie section. The sure options were simply solid circles, representing the 100% outcome. After selecting their preferred gambling choice subjects were told which of the two possibilities for the chosen gamble they had “won”. Despite telling participants that their reward depended on their performance they all received a modest financial reward (£20) after completing the study.
Patients also filled out a self rating questionnaire and rated themselves on a 1–5 point rating scale for alertness, attentiveness and interest, where 1 is associated with “not at all” and 5 with “extremely. On average patients scored 3.2 points on alertness, 3.5 on attentiveness and 4 on interest prior to treatment and 3.8 points for alertness, 3.9 points for attentiveness and 4.3 points for interest one hour after L-dopa treatment.
Random effects ANOVA models were fit to all behavioural variables. For the working memory task, the raw working-memory scores were converted into scaled scores according to published normative tables13. For both the learning and risk tasks, ANOVAs were carried out on parameters from computational models fit to the behavioural data of individual subjects (see supplementary methods for details of the models). For the learning task, we extracted two parameters which characterized how much each subject learned from positive and negative feedback. The first parameter characterized the amount that positive feedback after selecting one of the stimuli caused that stimuli to be chosen more often in subsequent trials, and the second parameter characterized the amount that negative feedback caused the corresponding stimulus to be selected less often in subsequent trials. Thus, these parameters characterized how much each subject learned from positive and negative feedback, and our ANOVA was carried out with these parameters as the dependent measure. Similarly, for the risk task, we fit a single parameter to the choices of the subject, which characterized how much the subjects valued large vs. small rewards. Larger positive values of this parameter imply that subjects prefer small, sure rewards to large rewards with a lower probability. Negative values imply that subjects prefer larger rewards with low probabilities. Thus, this parameter characterizes the amount of risk subjects are prone too. For the risk analysis, the ANOVA was carried out with this parameter as the dependent variable.
All patients fulfilled the Queen Square Brain Bank criteria for PD 17 and were taking L-dopa (Table 1 for demographics). Twelve patients with idiopathic Parkinson’s disease without ICB (3/12 female) and 18 PD patients with ICBs (PD+ICB) (5/18 female) were compared against 22 healthy controls (10/22 female). All PD+ ICB patients had at least 2 impulsive compulsive behaviours. PD+ ICB patients had an earlier disease onset (t-test, p < 0.05). The average time lag between the diagnosis of an ICB and the testing was 5.6 months.
All patients were screened for sub-classes of impulse control behaviors (Table 1). Nine PD+ ICB were tested during reduction of their dopamine agonist medication. Seven patients had already reduced their dopamine agonist medication, which had improved their impulsive behaviour. However, they still fulfilled the criteria of ICB. Two PD+ ICB patients denied having active impulsive or compulsive behaviour at the time of testing but both had significant behavioural abnormalities less than 1 year previous. All patients with ICBs developed their behavioural abnormalities as a result of medication.
Although there was no significant age difference between the 3 groups (ANOVA F2,46 = 2.96, p > 0.05), the PD group showed a trend to be older than PD+ ICB group. There was no significant difference in the morning (t-test: p = 0.1 95% CI of the difference = −83 to +56 mg) and daily L-dopa dose between the patient groups (t-test: p = 0.2, 95% CI of the difference = −407 to +112 mg). We assessed years of education in 17/22 controls, 9/12 PD patients and 14/18 ICB patients and found that it was not significantly different (ANOVA F2,37 = 1.98, p > 0.05). The timing of the last dopaminergic medication was also not significantly different between the patient groups (t-test p=0.2, 95% CI of the difference = −0.7 to +3.3 hours).
The mixed model ANOVA using the z-score transformations of working memory performance as the dependent variable found a main effects of group (F2,47 = 6.9, p = 0.02) and task (F1,131 = 16.0, p < 0.001), and a significant interaction between these factors (F2,131 = 3.3, p < 0.05), but no effect of “off “ versus “on” (F1,131 =0.007, p > 0.9). To examine these effects in more detail two additional ANOVAs with post-hoc comparisons were carried out, which revealed that the overall working memory (digit forward + backward span) was significantly impaired in the PD+ ICB group compared to both the control and PD groups (p < 0.01), but there was no difference between the PD group and controls (p > 0.05; Fig 1A). More specifically PD+ ICB patients performed significantly worse on the forward task than the PD and control groups (both p < 0.001) and also performed significantly worse on the backward task than the PD and control groups (p < 0.01). We performed a further analysis splitting the results for PD+ ICB patients depending on their type of impulsive compulsive behaviour (eg. PG versus hypersexuality) but did not find any significant difference (p > 0.05). There were no significant differences between PD and controls in the forwards task (p > 0.05) but the control group was significantly better than the PD group in the backwards task (p = 0.01) (Fig. 1B).
Overall, subjects found the learning task to be challenging. The number of times that subjects picked the most rewarded image (Fig. 2) was similar between on and off conditions. Averaged across each group, subjects were able to identify the best stimulus (i.e. the stimulus which was being most often rewarded) in each block and picked the stimulus which was most often rewarded in the block at above chance (50%) levels.
Learning from positive and negative feedback was compared, among groups and sessions, using the estimates of the amount learned by positive and negative feedback derived from the model fit to individual subjects. All analyses were done within subject. Separate ANOVAs were initially carried out on each group. There was a significant interaction between session (off vs. on drug) and type of feedback for the PD+ICB group (Fig. 4A; F1,19 = 4.8, p < 0.05, one-sided), as well as a significant main effect of feedback valence (F1,19 = 12.25, p < 0.05). The PD group showed a main effect of feedback valence (Fig. 3B; F1,11 = 14.4, p < 0.05), but no interaction (F1,11 = 2.89, p > 0.05). There was a significant main effect of valence in the control group (Fig. 4C; F1,16 = 10.2, p < 0.05) but there was no significant interaction with session (F1,16 = 0.09, p > 0.05). An ANOVA was then carried out between PD and PD +ICB groups and a significant group by session (off vs. on) by valence interaction was found (F1,30 = 6.55, p < 0.05). Specifically, the difference between learning from positive vs. negative feedback on and off medication differed (Fig. 3D) between PD and PD+ICB groups.
Participants were generally risk averse (Fig. 4A). There were several salient differences between groups, however. First, dopamine medication led to an increase in risk preference in both the PD and PD+ ICB groups relative to controls tested twice (Fig. 4B): controls showed a decrease in risk preference in the second test session, whereas both patient groups showed an increase in risk preference in the second session relative to the first session (F2,48 = 4.2, p < 0.05). Post-hoc comparisons showed that the controls were significantly different than the PD subjects (p < 0.05) but did not quite differ significantly from the ICB group (p > 0.05). Subsequently, we carried out the same analysis on the subset of ICB patients that had pathological gambling (n = 10, ICB gamblers). This group was significantly more risk prone across both the first and second session than the normal PD group (F1,42 = 5.58, p < 0.05).
Previous studies have shown that dopaminergic replacement improves learning from positive feedback but impairs learning from negative feedback while withdrawal from anti-Parkinsonian medication leads to the reverse profile10, 19. Contrary to this, we found that PD+ ICB patients showed increased learning from positive vs. negative feedback off medication compared to on medication, whereas PD patients without ICB in our study showed a trend towards the previously described learning effects10. Furthermore, the group by session by valence interaction was significant, such that these learning effects were significantly complimentary.
PET studies of dopamine release have shown that dopamine medication leads to elevated ventral striatal dopamine release in PD+ ICB patients relative to PD patients without ICB 20, 21. Furthermore, studies in healthy human subjects have shown that PET measures of dopamine synthesis rates in the striatum correlate with increased learning from positive feedback relative to negative feedback and deleterious effects of dopamine agonists in normal subjects with high baseline dopamine synthesis rates have also been reported 22. These observations and our results are consistent with the hypothesis that PD+ ICB patients have elevated baseline dopamine levels in the ventral striatum, and that dopamine replacement therapy increases the levels further, causing reduced learning from positive feedback. This might be explained by the “inverted U” shape hypothesis 23, 24 where cognitive performance and the ability to pick the rewarded stimulus might be impaired when PD+ ICB subjects are pushed off the upper end of the curve by their medication.
Previous authors have described the premorbid Parkinsonian personality as one characterised by caution, risk aversion and anhedonia 25. In contrast PD+ ICB patients have a behavioural profile characterized by increased impulsiveness or novelty seeking 26 similar to subjects prone to substance abuse and behavioural addictions 27. Impulsive individuals are more sensitive to reward than punishment 28 and resemble the unmedicated PD+ ICB patients in this study.
The risk task was designed to test the hypothesis that patients with ICB are more risk-prone than non- ICB patients 26. Overall PD+ ICB patients showed a trend to be more risk prone relative to normal PD, which did not reach significance. However those PD+ ICB patients who had PG were significantly more risk prone compared to the PD group. Impaired decision making with an increased tendency towards risky behaviour has also been found in pathological gamblers 29. Furthermore dopaminergic medication led to increased risk preference in the PD patients relative to the healthy controls, and just missed significance in the ICB patients vs. healthy controls. This is particularly interesting since risk taking decreases with age 30 and the PD group without ICB showed a trend to be older compared to the other two groups. These findings are consistent with two recently published studies which showed that dopamine agonist treatment lead to an increased novelty and reward seeking behaviour and in a reduction of negative feedback learning 31, 32
Working memory in the forward and backward digit span was significantly reduced in PD+ ICB patients compared to the PD and the control group. PD patients without impulsive behaviour showed impairment in the digit backward span test compared to healthy controls, which is in agreement with a recent publication 33. We did not find any improvement of WM after L-dopa administration. Previous studies have shown that working memory is reduced in impulsive patients with attention deficit/hyperactivity disorder and healthy controls who scored highly on an impulsivity questionnaire and that these subjects had lower total striatal dopamine levels which seem to be associated with lower working memory capacity 34, 35. These Other studies have shown impaired spatial memory in patients with impulse control disorders36.
We have demonstrated a different learning profile between PD patients with and without ICB and healthy controls. These differences could be explained by higher ventral striatal dopamine levels in PD+ ICB patients. In addition PD patients with pathological gambling were more risk prone compared to normal PD patients and healthy controls. These findings may have therapeutic and clinical implications. The reduction in the overall anti- parkinsonian medication with positive reinforcement of non impulsive behaviour is likely to be more beneficial than aversion therapy in PD patients with impulsive compulsive behaviours.
The authors wish to thank Karen Shaw for advice with the ethics amendment for this study. We would also like to thank all patients and their partners who participated on this study. This work was undertaken at UCLH/UCL who received a proportion of funding from the Department of Health's NIHR Biomedical Research Centres funding scheme.
Potential conflict of interest: The authors declare no financial or other conflict of interest
Contribution of Authors:Atbin Djamshidian: recruitment of patients, project design, data analysis and manuscript preparation
Ashwani Jha: contribution to data analysis and manuscript preparation
Sean S. O’Sullivan: recruitment and reviewing manuscript
Laura Silveira-Moriyama: contribution to data analysis and reviewing manuscript
Clare Jacobson: contribution to data analysis and reviewing manuscript
Peter Brown: supervising, reviewing manuscript
Andrew Lees: supervising, reviewing manuscript and manuscript preparation
Bruno B. Averbeck: project design,developing tests, supervision, analysing data and manuscript preparation.
|Stock Ownership in medically-related fields||none||none||none||none||none||none||none||none|
|Advisory Boards||None||None||None||None||None||Medtronic||Novartis,Teva, Meda,Boehringer Ingelheim,GSK, Ipsen,Lundbeck, Allergan, Orion||None|
|Honoraria||none||none||Britannia Pharmaceuticals||Britannia Pharmaceuticals||None||Medtronic||Novartis,Teva, Meda,Boehringer Ingelheim,GSK, Ipsen,Lundbeck,Allergan, Orion||None|
|Grants||none||UK Parkinson’s Disease Society||UK Parkinson’s Disease Society||none||None||none||PSP Association Weston Trust, The Reta Lila Howard Foundation||Wellcome, Intramural research program of the National Institute of Mental Health|
|Intellectual Property Rights||none||none||None||None||None||none||none||none|
|Employment||Reta Lila Weston Foundation||UCL/UCLH||Barts and the London NHS Trust, UK||Reta Lila Weston Foundation||UCL/UCLH||UCL/UCLH||UCL/UCLH||UCL/UCLH|