Effects of dopamine on reinforcement learning and consolidation in Parkinson’s disease

  1. John P Grogan  Is a corresponding author
  2. Demitra Tsivos
  3. Laura Smith
  4. Brogan E Knight
  5. Rafal Bogacz
  6. Alan Whone
  7. Elizabeth J Coulthard  Is a corresponding author
  1. University of Bristol, United Kingdom
  2. North Bristol NHS Trust, United Kingdom
  3. University of Oxford, United Kingdom
8 figures, 2 tables and 1 additional file

Figures

Diagram of the learning trials of the Probabilistic Selection Task.

In each pair one card is more likely to be rewarded (shown ‘Correct’ feedback) than the other, with card A in pair AB rewarded 80% of trials, and card B on 20% of trials. For pairs CD and EF, the …

https://doi.org/10.7554/eLife.26801.003
Mean difference in memory block accuracy between 24 hr and 30 min, for each condition in experiment 1.

ON-OFF had significantly greater increases in memory score over this time than OFF-OFF (* = p=0.0106) and indeed both day 1 ON conditions (blue bars) had a mean increase in accuracy while both day 1 …

https://doi.org/10.7554/eLife.26801.004
Figure 2—source data 1

Summary statistics for Figure 2, the difference in memory scores between 30 min and 24 hr tests for each condition.

https://doi.org/10.7554/eLife.26801.005
Figure 3 with 2 supplements
The mean percentages of choose-A and avoid-B selections at the 24 hr novel pairs tests in experiment 1, split by a) day 1 and b) day 2 conditions.

There were no significant effects of day 1 or day 2 medication state (p>0.28). Error bars are SEM. Figure 3—figure supplement 1 shows the data when filtered by performance on the 80–20% pair for day …

https://doi.org/10.7554/eLife.26801.006
Figure 3—source data 1

Summary statistics for Figure 3, the percentages of choose-A and avoid-B selections in the experiment 1 novel pairs test.

https://doi.org/10.7554/eLife.26801.007
Figure 3—figure supplement 1
The mean percentages of choose-A and avoid-B selections for the filtered data in the experiment 1 novel pairs test split by day 1 conditions.

As the filtering removed data from different participants, paired samples statistics were no longer suitable. Independent-samples t-tests showed no significant effects of day 1 medication state (df =…

https://doi.org/10.7554/eLife.26801.008
Figure 3—figure supplement 1—source data 1

Summary statistics for Figure 3—figure supplement 1, percentages of choose-A and avoid-B behaviours for experiment 1 split by Day 1 condition, after data filtering.

https://doi.org/10.7554/eLife.26801.009
Figure 3—figure supplement 2
The mean percentages of choose-A and avoid-B selections for the filtered data in the experiment 1 novel pairs test split by day 2 conditions.

Between-subjects t-tests showed no effect of day 2 medication state (df = 57, p=0.151, d = 0.1780; p=0.928, d = 0.0061). Error bars are SEM.

https://doi.org/10.7554/eLife.26801.010
Figure 3—figure supplement 2—source data 1

Summary statistics for Figure 3—figure supplement 2, percentages of choose-A and avoid-B behaviours for experiment 1 split by day 2 condition, after data filtering.

https://doi.org/10.7554/eLife.26801.011
Figure 4 with 2 supplements
The mean percentages of selections on the novel pairs test for a) experiment 2 and b) experiment 3.

There were no significant effects of disease state or medication condition on either selection in either experiment 2 (p>0.2) or experiment 3 (p>0.3). Error bars are SEM. Figure 4—figure supplement 1

https://doi.org/10.7554/eLife.26801.012
Figure 4—source data 1

Summary statistics for Figure 4, the percentages of choose-A and avoid-B selections in experiments 2 and 3 novel pairs tests.

https://doi.org/10.7554/eLife.26801.013
Figure 4—figure supplement 1
The mean percentage of selections for the filtered data for experiment 2.

Independent-samples t-tests showed no significant effects on choose-A or avoid-B for disease state (df = 47, p=0.361, d = 0.2768; p=0.126, d = 0.4681) or medication state (df = 30, p=0.473, d = 0.257…

https://doi.org/10.7554/eLife.26801.014
Figure 4—figure supplement 1—source data 1

Summary statistics for Figure 4—figure supplement 1, percentages of choose-A and avoid-B behaviours for experiment 3, after data filtering.

https://doi.org/10.7554/eLife.26801.015
Figure 4—figure supplement 2
The mean percentage of selections for the filtered data for experiment 3.

Independent-samples t-tests showed no effects of disease state (df = 31, p=0.509, d = 0.2465; p=0.926, d = 0.0345) or medication state (df = 20, p=0.579, d = 0.2415; p=0.555, d = 0.25658). Error …

https://doi.org/10.7554/eLife.26801.016
Figure 4—figure supplement 2—source data 1

Summary statistics for Figure 4—figure supplement 2, percentages of choose-A and avoid-B behaviours for experiment 2, after data filtering.

https://doi.org/10.7554/eLife.26801.017
Mean final learning block accuracies across the three experiments.

Experiment 3’s final accuracy was significantly lower than experiments 1 and 2’s (p<0.000001). Error bars are SEM. Figure 5—source data 1 shows the summary statistics.

https://doi.org/10.7554/eLife.26801.018
Figure 5—source data 1

Summary statistics for Figure 5, the mean final learning block accuracies across the three experiments.

https://doi.org/10.7554/eLife.26801.019
A diagram of the timing of PD medication withdrawal for all four conditions in experiment 1.

Blue is when patients were ON medication, red hatched bars when they were OFF, and yellow bars the PST phases. In order for patients to be fully OFF medication during testing, they were withdrawn …

https://doi.org/10.7554/eLife.26801.020
The modified Probabilistic Selection Task procedure.

Participants saw two symbols on the screen, and selected one with a button press. If no response was made within 2 s, they were shown a ‘GO’ prompt. Feedback was determined probabilistically, was …

https://doi.org/10.7554/eLife.26801.022
Appendix 1—figure 1
The mean percentages of win-stay and lose-shift during the learning trials for experiments 1 (a), 2 (b) and 3 (c), each split by day 1 condition.

Appendix 1—figure 1—source data 1 shows the summary statistics.

https://doi.org/10.7554/eLife.26801.023
Appendix 1—figure 1—source data 1

Summary statistics for the percentage of win-stay and lose-shift behaviours during learning trials for each experiment.

https://doi.org/10.7554/eLife.26801.024

Tables

Table 1

The means and SEM for each experiment for PD patients and HC on all measures taken. Within each experiment, one-way ANOVAs were run between PD patients and HC (χ2 for the genders), and paired …

https://doi.org/10.7554/eLife.26801.021
Experiment123
MeasurePD patientsHCPD patientsHCPD patientsHC
Number181818201818
Gender (M/F)15/3**7/1116/2***5/1511/711/7
Age71.56 (2.06)71.19 (2.52)67.39 (2.10)66.05 (2.05)69.11 (1.44)71.61 (2.05)
Years Education13.50 (0.66)12.93 (0.89)14.83 (0.91)13.75 (0.56)11.94 (0.52)*14.72 (0.65)
MoCA29.44 (0.12)29.63 (0.13)28.72 (0.50)*26.85 (0.53)27.61 (0.54)26.78 (0.56)
DASS21.71 (2.85)*12.19 (2.76)15.39 (2.54)20.05 (5.50)29.13 (4.53)**10.44 (2.29)
 Depression6.35 (0.81)**2.88 (0.87)4.94 (1.13)5.55 (1.958)7.13 (1.71)***2.78 (0.75)
 Anxiety7.88 (1.27)**3.25 (0.88)5.67 (0.84)5.50 (1.80)10.33 (1.58)***2.61 (0.69)
 Stress7.47 (1.41)6.06 (1.32)4.78 (1.09)9.00 (2.02)11.67 (1.73)**5.06 (1.25)
BIS14.94 (2.36)15.31 (3.07)53.50 (2.47)51.90 (2.35)53.56 (2.70)51.00 (1.88)
LARS−20.22 (1.36)***−27.44 (1.24)−23.50 (1.71)*−30.00 (1.58)−22.44 (2.06)*−27.06 (1.16)
UPDRS ON18.67 (2.69)18.78 (2.85)26.50 (2.73)
UPDRS OFF24.56 (3.41)***23.28 (2.97)30.44 (2.52)***
Years since diagnosis4.44 (1.21)4.39 (0.90)5.00 (1.02)
Years since symptoms5.18 (1.28)4.78 (0.86)6.44 (1.10)
LDE (mg)566.41 (61.18)543.70 (67.36)653.00 (93.96)
# levodopa/ dopamine agonists/both12/1/510/2/610/0/8
# on XL meds3109
  1. MoCA=Montreal Cognitive Assessment, DASS=Depression, Anxiety and Stress Scale, BIS=Barratt Impulsivity Scale, LARS=Lille Apathy Rating Scale, UPDRS=Unified Parkinson’s Disease Rating Scale, LDE=Levodopa Dose Equivalence.

Appendix 2—table 1

The mean BIC and model parameter values for each model for each experiment. The bolded lines show the models with the smallest BIC, which was model number 2, the dual-learning rate model without …

https://doi.org/10.7554/eLife.26801.025
ExperimentModelBIConoffα+ONα+OFFα-ONα-OFFβ
11996.2430.100660.8972
2885.94340.10240.00419.678
 3992.39850.08080.098836.866
 4891.55220.11840.15930.01840.00977.3885
21520.60290.1131231.6615
2485.01050.19220.031921.278
 3510.50980.12480.0616116.4691
 4487.68250.31440.18250.04380.12737.2812
31807.03220.132101.1408
2737.41450.20690.02535.9512
 3802.91930.10590.0624163.4829
 4746.79260.29180.18460.05210.05554.104

Additional files

Appendix 1—figure 1—source data 1

Summary statistics for the percentage of win-stay and lose-shift behaviours during learning trials for each experiment.

https://doi.org/10.7554/eLife.26801.024

Download links