Impaired proactive cognitive control in Parkinson’s disease

Abstract Adaptive control has been studied in Parkinson’s disease mainly in the context of proactive control and with mixed results. We compared reactive- and proactive control in 30 participants with Parkinson’s disease to 30 age matched healthy control participants. The electroencephalographic activity of the participants was recorded over 128 channels while they performed a numerical Stroop task, in which we controlled for confounding stimulus-response learning. We assessed effects of reactive- and proactive control on reaction time-, accuracy- and electroencephalographic time-frequency data. Behavioural results show distinct impairments of proactive- and reactive control in participants with Parkinson’s disease, when tested on their usual medication. Compared to healthy control participants, participants with Parkinson’s disease were impaired in their ability to adapt cognitive control proactively and were less effective to resolve conflict using reactive control. Successful reactive and proactive control in the healthy control group was accompanied by a reduced conflict effect between congruent and incongruent items in midline-frontal theta power. Our findings provide evidence for a general impairment of proactive control in Parkinson’s disease and highlight the importance of controlling for the effects of S-R learning when studying adaptive control. Evidence concerning reactive control was inconclusive, but we found that participants with Parkinson’s disease were less effective than healthy control participants in resolving conflict during the reactive control task.


Items
Supplementary Table 1 List of items used in the numerical Stroop task.We used items with a numerical distance 1 and 2, also previously used by Dadon and Henik 1 .Items were balanced in terms of numerical presentation and overall presentation.For the ISPCE manipulation, either small or large number pairs were manipulated respectively.
Criteria were (1) diagnostic items should not be presented one after each other (in order to have a balanced presentation of diagnostics throughout the block), (2) the same item should not be presented more than twice in a row, (3) inducer items should not be presented more than four times in a row, (4) the same correct response side (left or right) should not be presented on more than three consecutive trials and (5) the same congruence should not be presented more than four times in a row.Item presentation orders were randomized for each participant separately according to these rules.
Supplementary Information 1 Criteria for pseudo-randomization.

Supplementary Equation 1 Shifted log-normal regression model and priors.
For the diagnostic model, we used the posterior distribution of the inducer model to construct informed priors for the effects not expected to differ between conditions (shift parameter, sigma, Congruency, Block/Item, and the random intercepts).We used a Gaussian distribution, with mu equal to the mean of the inducer posterior and sigma defined as the larger absolute difference between the posterior mean and the two 95% credible interval borders of the inducer posterior.For the interaction effects, we used regularized priors with a Gaussian distribution centered around zero, with sigma defined as the larger absolute difference between zero and the two 95% credible interval borders of the posterior distribution of the inducer model.

Supplementary Information 2 Informed prior distributions of the diagnostic models.
~ (  )

Supplementary Equation 2 Logistic regression model and priors
For the analysis of the error data, we performed a logistic regression (Bernoulli distribution with logit-link function) and used the same approach to fit inducer and diagnostic models as outlined for the shifted-log normal model.Priors were weakly informed so that lower error rates had a higher probability, as typically, participants never make more than 10% errors during these kinds of tasks (see supplement Supplementary Equation 2).Log odds of the estimated marginal mean effects were transformed into probabilities for a more meaningful interpretation.

Supplementary Information 3 Description of the logistic regression.
Parameter ~ (−1.3,1.5)  1 ~ (0, 1.5)  2 ~ (0, 1.5)  3 ~ (0, 1.5)  4 ~ (0, 1.5)  5 ~ (0, 1.5)  6 ~ (0, 1.5)  1 ~ (0, 1.5)  2 ~ (0, 1.5) used by Dadon and Henik 1 .Items were balanced in terms of numerical presentation and overall presentation.For the ISPCE manipulation, either small or large number pairs were manipulated respectively.Supplementary Figure 2 Posterior predictive checks of reaction times for the analysis of the LWPCE data.Hollow dots depict the observed data, and full dots represent the posterior predictions of the shifted log-normal model.Light green colored dots show congruent items and dark green dots show reaction times to incongruent items.Rows distinguish reaction times to inducer and diagnostic items.Within each panel, the upper dot reflects the mean of the upper 90 th percentiles of the reaction time distribution.The middle dot reflects the average of the reaction time distribution, and the lower dot is the mean of the 10 th percentile of the reaction time distribution.

1 .
Condition Specific Time Frequency ResultsSupplementary Figure 4 Estimated marginal mean effects of the time-frequency regression analysis for the LWPCE manipulation of the mainly incongruent block.Data on the two left columns are plotted in reference to stimulus onset (SL), and data on the two right columns are referenced to the response (RL).The first two rows show the time-frequency results at channelFCz for incongruent and congruent items by group (HC and PD).The last two rows depict the averaged theta power between 0.3 and 0.7s SL and -0.6s to -0.2s RL by congruence and group.