Impaired savings despite intact initial learning of motor adaptation in Parkinson’s disease
- First Online:
- Cite this article as:
- Leow, L., Loftus, A.M. & Hammond, G.R. Exp Brain Res (2012) 218: 295. doi:10.1007/s00221-012-3060-5
- 286 Views
In motor adaptation, the occurrence of savings (faster relearning of a previously learned motor adaptation task) has been explained in terms of operant reinforcement learning (Huang et al. in Neuron 70(4):787–801, 2011), which is thought to associate an adapted motor command with outcome success during repeated execution of the adapted movement. There is some evidence for deficient savings in Parkinson’s Disease (PD), which might result from deficient operant reinforcement processes. However, this evidence is compromised by limited adaptation training during initial learning and by multi-target adaptation, which reduces the number of reinforced movement repetitions for each target. Here, we examined savings in PD patients and controls following overlearning with a single target. PD patients showed less savings than controls after successive adaptation and deadaptation blocks within the same test session, as well as less savings across test sessions separated by a 24-h delay. It is argued that impaired blunted dopaminergic signals in PD impairs the modulation of dopaminergic signals to the motor cortex in response to rewarding motor outcomes, thus impairing the association of the adapted motor command with rewarding motor outcomes. Consequently, the previously adapted motor command is not preferentially selected during relearning, and savings is impaired.
KeywordsSavingsVisuomotor adaptationParkinson’s diseaseMotor learning
The ability to adapt movement to changing environmental demands is important to everyday functioning, such as adapting the magnitude of wrist movement when using a new computer mouse with different gain. In studies of motor adaptation, the sensory consequence of a movement is systematically perturbed, such that there is a discrepancy between the predicted sensory outcome and the actual sensory outcome (a sensory prediction error). For example, the perturbation can be a 30° clockwise rotation of the visual feedback of a hand movement trajectory during a target reaching task, which leads to visually detected error in movement direction. Motor adaptation is often explained with reference to internal models, which map the relationship between motor commands and their sensory consequences (Miall and Wolpert 1996). When the sensory consequence of executing a motor command is systematically perturbed, the internal model uses the resulting sensory prediction error to generate an updated motor command to reduce the sensory prediction error. This process of updating the motor command based on sensory prediction errors is repeated trial by trial until sensory prediction errors approach pre-perturbation levels.
However, this single timescale process of updating the motor command using sensory prediction errors cannot fully explain the occurrence of savings. Savings, which is evident when the motor system adapts more quickly when readapting to a previously encountered perturbation, occurs even after deadaptation (re-exposure to unperturbed feedback) returns behavior to the unadapted state. Savings has been accounted for by a two-state model, comprising a fast process, which responds strongly to sensory prediction errors and which decays quickly, and a slow process, which responds weakly to sensory prediction errors and which decays slowly (Smith et al. 2006). After adaptation, re-exposure to unperturbed feedback during deadaptation results in large sensory prediction errors. The fast process responds strongly to these sensory prediction errors and returns behavior to the unadapted state, but also decays quickly. However, as the slow process only responds weakly to the sensory prediction errors, it persists despite deadaptation. The presence of savings after deadaptation is therefore thought to be due to the persisting influence of the slow process. Savings occurs following delays of 24 h (Krakauer et al. 2005) to 1 year (Landi et al. 2011), supporting the existence of a slowly decaying process.
A recent re-conceptualization of the slow process posits that the slow process is ‘model-free’ and does not involve trial-by-trial adjustment of an internal model. Two mechanisms are thought to constitute the model-free slow process: first, use-dependent plasticity, whereby repetition of the adapted movement biases subsequent movements in the same direction, and second, operant reinforcement learning, whereby the adapted movement is reinforced by its association with outcome success during initial learning, resulting in its preferential selection during subsequent learning (Huang et al. 2011). Operant reinforcement learning has specifically been invoked to explain the occurrence of savings following deadaptation (Huang et al. 2011).
Support for the model-free account of savings comes from findings of savings despite conditions which typically result in interference (adaptation to two rotations in opposite directions) when the fully adapted movement for both rotations was the same (Huang et al. 2011). Subjects adapted first to a 30° counterclockwise rotation when aiming to the first target, deadapted with veridical feedback, and then adapted to a 30° clockwise rotation when aiming to the second target (located 60° counterclockwise from the first target). Manipulation of the target locations ensured that the fully adapted movement for the two opposing rotations was the same. Strikingly, adaptation to the first 30° clockwise rotation resulted in savings when adapting to the second 30° counterclockwise rotation. This finding of savings despite adaptation to rotations in opposite directions is incongruent with an internal model framework, which predicts that adapting to one rotation, deadapting, and then adapting to a rotation in the opposite direction results in the acquisition of two separate internal models for each rotation, each accounting for distinct input–output relationships between motor command and sensory feedback. If separate internal models were acquired for each rotation, learning and subsequently deadapting the first internal model would have no effect on learning the second internal model.
Overlearning, or repetition of the adapted movement after reaching performance asymptote in initial learning, increases savings during relearning. Joiner and Smith (2008) showed that extended overlearning (103 adaptation trials) led to significantly more savings over 24 h than brief overlearning (30 adaptation trials), even though no further error reduction occurred after the first 11 adaptation trials. At first glance, overlearning seems inefficient: as there is no performance improvement, the motor system does not gain new information. When interpreted in terms of model-free processes; however, overlearning makes sense: repeatedly executing the adapted motor command engages use-dependent plasticity and strengthens the association of the motor command with outcome success, thereby biasing re-selection of that motor command at relearning, which increases savings.
A large body of evidence shows that operant reinforcement learning occurs through phasic bursts of the midbrain dopaminergic neurons, which project to the striatum (for a review, see Schultz 1998). If operant reinforcement mechanisms contribute to savings, patients with dysfunctional dopaminergic function such as those with Parkinson’s Disease will show deficient operant reinforcement learning and therefore also show impaired savings. There is some evidence that this occurs in Parkinson’s Disease (PD). PD patients show impaired operant reinforcement learning (Shohamy et al. 2006; Rutledge et al. 2009; Frank et al. 2004), as well as impaired savings within the same test session (Bedard and Sanes 2011) and over a 24 h delay (Marinelli et al. 2009; Bedard and Sanes 2011). There are, however, caveats to the previous reports of impaired savings in PD. First, incomplete error reduction and insufficient overlearning may have contributed to those findings of impaired savings. Studies of savings typically impose enough adaptation trials during initial learning such that the adapted movement is repeated for at least 10 trials after reaching asymptotic performance. In visuomotor adaptation, directional error at performance asymptote is typically 5° given a 30° rotation in visual feedback (Krakauer et al. 2000). In the Bedard and Sanes (2011) study, PD patients and controls only reduced directional error to 17° and 13° respectively at the end of initial adaptation. Given the 30° rotation magnitude, this means less than half the amount of directional error was reduced at the end of initial adaptation; hence, participants were unlikely to have reached performance asymptote. Similarly, although Marinelli et al. (2009) did allow error reduction of up to 7.8° in their Experiment 2, adaptation appeared to have only just reached asymptote, without sufficient overlearning. Hence, while both studies suggest deficient savings in PD, it remains unknown whether this savings deficit will remain evident given sufficient overlearning. Second, because the frequency with which the adapted movement is repeated may affect savings, savings may be reduced with fewer repetitions of the adapted movement, as occurs with multi-target adaptation, in which the adapted movement differs for each target direction, requiring different motor commands. Previous studies required participants to adapt to four (Bedard and Sanes 2011) or eight targets (Marinelli et al. 2009). Their findings might therefore result from insufficient repetition of the adapted movement.
Rationale and hypothesis
Previous findings of intact error reduction but deficient savings in PD (Bedard and Sanes 2011; Marinelli et al. 2009) suggest intact model-based fast learning but impaired model-free slow learning in PD. However, insufficient overlearning and multiple-target adaptation may have contributed to these findings of deficient savings in PD. Here, we examined savings in PD patients and neurologically intact older adult controls following overlearning and single-target adaptation. Savings was examined within the same test session (Experiment 1) and over a 24-h delay (Experiment 2). In Experiment 1, PD patients and controls adapted to rotated visual feedback, and then deadapted with veridical feedback trials in 4 successive blocks of 25 adaptation and 25 deadaptation trials. In Experiment 2, PD patients and controls completed a first block of 50 adaptation trials with rotated visual feedback of the movement trajectory. After a 24-h delay, participants completed a second block of 50 adaptation trials with the same perturbation. In both studies, savings was quantified as block-to-block increases in percent adaptation averaged over Trials 2–15. Because Marinelli et al. (2009) suggested that sleep deficiencies in PD may have contributed to their findings of deficient savings over 24 h, we explored the role of sleep deficiencies in Experiment 2 with the Parkinson’s Disease Sleep Scale (Chaudhuri et al. 2002). Given the deficits in operant reinforcement learning in PD, and the proposed role of operant reinforcement in savings, it was predicted that PD patients would show deficient savings within the same test session, and between test sessions separated by 24 h, despite overlearning after reaching asymptotic performance with single-target adaptation.
A total of 16 patients with mild-to-moderate Parkinson’s disease and 16 neurologically intact older adult controls were recruited from the Parkinson’s Western Australia newsletter and local newspapers. This study was approved by the Human Research Ethics Committee at The University of Western Australia. All participants provided written informed consent.
All PD patients were tested on-peak of their medication schedule. None of the participants showed cognitive impairment, scoring within the normal range of greater than 24 on the Montreal Cognitive Assessment (Nasreddine 2005). All participants had normal or corrected-to-normal vision and were naïve to the experimental design. PD patients’ disease severity was rated according to the motor subscale of the Movement Disorders Society Sponsored Revised Unified Parkinson’s Disease Rating Scale (MDS-UPDRS) (Goetz et al. 2007).
Eight PD patients (mean age 65 ± 8, 6 women) and eight older adult controls (mean age 69 ± 10, 7 women) participated in Experiment 1. All but two of the PD patients were on Levodopa (mean daily Levodopa dose: 519 ± 200 mg). Four PD patients were also on the dopamine agonist Pramipexole (mean daily dose 2.25 ± 1.06 mg), while 1 PD patient was on 4 mg of Cabergoline per day. Mean disease duration was 6.6 ± 5.1 years, and the mean UPDRS-III score was 30.3 ± 6.4.
Eight PD patients (mean age 66 ± 8, 4 women) and eight older adult controls (mean age 69 ± 9, 8 women) participated in Experiment 2. All PD patients were on Levodopa (mean daily Levodopa dose 584 ± 422). Five of the 8 PD patients were on the dopamine agonist Pramipexole (mean daily dose: 1.1 ± 1.5 mg). Mean disease duration was 8.3 ± 6.9 years, and mean UPDRS-III score was 28 ± 13.
Participants were seated in front of a laptop computer placed approximately 50 cm away from their midline. Participants held a digitizing pen (15.95 cm long, 1.4 cm wide, 17 g) on a WACOM Intuos 2 digitizing tablet (size: 30.48 cm x 30.48 cm, resolution ± 0.025 mm). The pen’s position on the tablet (XY coordinates) was sampled at 100 Hz and displayed on the computer monitor as a circular cursor with a 5 pixel radius (1.25 mm). Custom software written in LabVIEW 7.0 (National Instruments) was used for data acquisition. Direct vision of the hand was prevented by placing the tablet and the hand directly beneath a monitor stand, with the laptop placed atop the stand.
General experimental procedure
The experimental task required participants to move the on-screen cursor from the start circle to the target circle by moving the digitizing pen on the digitizing tablet. Participants were first instructed to move the cursor representing the pen’s position into the start circle. After the cursor was within the start circle for 2 s, a target circle of radius 23 pixels (6.08 mm) appeared 75 mm at 45° from the target. A tone sounded immediately after the target circle appeared, signaling participants to move the cursor to the target. Participants were instructed to move the cursor from the start circle to the target circle as accurately and as quickly as possible, in a single, uncorrected movement. In Experiment 1, visual feedback of the movement trajectory was shown on-screen 100 ms after movement completion to prevent online correction of movement. In Experiment 2, visual feedback of the movement trajectory was shown concurrently during movement in Experiment 2. This was because pilot testing showed that PD patients and controls were largely able to follow instructions to avoid online correction of movement and execute straight, uncorrected movements. After movement completion, visual feedback of the movement trajectory was shown for 1,000 ms in both Experiment 1 and 2. Only one target position was used in all phases of both experiments. The key measure of interest was error in movement direction (directional error).
Prior to adaptation, all participants completed a minimum of 30 practice trials with veridical feedback of the movement trajectory, until three out of four consecutive movements were made with directional error of less than or equal to 3° and movement time less than 1,000 ms. Once the practice criteria were met, the test phase commenced. During the test phase, participants completed adaptation trials in which the visual feedback of the movement trajectory was rotated 30° counterclockwise relative to the start circle, while the position of the start circle and the target circle remained unchanged. This resulted in an error in the movement direction (directional error). To hit the target, participants had to compensate by moving 30° clockwise to the original movement direction.
Experiment 1 design
After the practice phase, participants completed four blocks of 25 adaptation trials with the 30° counterclockwise rotation in visual feedback. Each adaptation block was followed by a deadaptation block of 25 trials with veridical feedback of the movement trajectory. Adaptation and deadaptation blocks were continuous, without breaks between blocks.
Experiment 2 design
After the practice phase, participants completed a first adaptation block of 50 adaptation trials with the 30° counterclockwise rotation in visual feedback. After a 24 h delay, participants completed a second adaptation block of 50 adaptation trials with the same 30° counterclockwise rotation. Subsequently, participants were deadapted with 20 trials with veridical feedback of the movement trajectory.
Cartesian XY coordinates were recorded and used to plot movement trajectory. Directional error was scored at either (i) 100 ms into the movement after moving at least 5 mm (Bedard and Sanes 2011) or (ii) at 25 % of movement trajectory, whichever came earlier. Directional error was calculated as the angular difference between this movement direction and an idealized movement direction starting from the start circle to the target circle.
Between group comparisons
Figure 1 shows the trial-by-trial directional error averaged across PD patients (clear circles) and controls (filled circles) in Blocks 1–4. In Block 1, at the onset of 30° counterclockwise rotation of visual feedback, directional errors were in the expected direction (negative, i.e., counterclockwise) and of the expected size (approximately 30°). PD patients and age-matched controls showed similar trial-by-trial error reduction, reducing error at an exponential rate. To examine whether adaptation rate in Block 1 was similar in PD patients and controls, mixed-ANOVAs with a between-subjects factor of Group (PD, Controls) and a within-subjects factor of Trial (Trials 2–15) were conducted for trials during rapid error reduction (Trials 2–15). There was no significant main effect of Group [F(1, 14) = 0.24, p = 0.6], nor any significant Group-by-Trial interaction. The same analysis on Trials 2–6 also did not result in a significant main effect of Group [F(1, 14) = 0.86, p = 0.37]. In addition, individual data sets for Block 1 were fitted to the exponential function to obtain the rate constant for each participant. Rate constant was similar in PD patients (0.25 ± 0.04) and in controls (0.28 ± 0.05), [t(12) = 0.39, p = 0.6, d = 0.2]. Thus, PD patients and controls did not differ significantly in adaptation rate in Block 1. Asymptotic directional error for Block 1 (estimated from mean of the last 10 trials) was not significantly different in PD patients (−6.4 ± 1.9°) and controls (−4.2 ± 1.5°), [t(14) = 1.15, p = 0.3, d = 0.4].
After each deadaptation block, PD patients appeared to show larger directional error at performance asymptote than controls, particularly in Block 2 (see Fig. 1). Mixed-ANOVAs with between-subjects factor Group (PD, Controls) and within-subjects factors Block (Blocks 2, 3, and 4) and Trial (last 10 trials of each block) revealed a significant main effect of Group on asymptotic directional error [F(1, 14) = 6.24, p = 0.026]. Asymptotic directional error, estimated through mean directional error from the last 10 trials of each adaptation block, was significantly larger in PD than in controls in Block 2 (PD: −8.0 ± 1.2°; controls: −3.9 ± 1.0°), [t(14) = 2.46, p = 0.02, d = 1.22]. Mean asymptotic directional error was also larger in PD than in controls at Block 3 (PD: −7.1 ± 1.1°, controls: −5.1 ± 1.1°) and Block 4 (PD: −5.4 ± 0.8°, controls: −3.7 ± 0.7°), but this difference was not significant at Block 3 [t(14) = 1.25, p = 0.2, d = 0.6] or Block 4 [t(14) = 1.62, p = 0.13, d = 0.8].
Larger asymptotic directional error was evident despite similar extent of deadaptation in PD patients and controls in each preceding block of deadaptation trials (supplementary electronic materials, S2). Asymptotic directional error during deadaptation (estimated from directional error in the last 10 deadaptation trials) did not differ significantly between PD patients and controls: mixed-ANOVA with between-subjects factor Group (PD, Controls) and within-subjects factor Block (deadaptation blocks 1–3), and Trial (last 10 trials of each block) showed no significant main effect of Group, and no significant Group-by-Trial interaction.
Within-group comparisons (savings)
In controls, percent adaptation stabilized in subsequent adaptation blocks. Percent adaptation increased an averaged 5.7 % from Block 2 to Block 3 [t(7) = 1.03, p = 0.3, d = 0.27], and 14.6 % from Block 3 to Block 4 [t(7) = 1.9, p = 0.09, d = 0.6] (see insets, Fig. 2a middle and right panels). Similarly, repeated-measures ANOVAs comparing Blocks 2 to Block 3, and Block 3 to Block 4 yielded no significant main effect of Block, and no significant Block-by-Trial interactions.
PD patients did show block-to-block reduction in directional error from Block 2 to Block 3 (Fig. 2b middle panel), as percent adaptation increased significantly by an averaged 10.5 % (inset), [t(7) = 4.5, p = 0.003, d = 0.7], indicating savings. Supporting this result, a RM-ANOVA with within-subjects factors Block (Block 2, 3) and Trials (Trials 2–15) yielded a significant main effect of Block [F(1,7) = 11.2, p = 0.01]. The Block-by-Trial interaction was not significant. PD patients only increased percent adaptation by an average 0.4 % from Block 3 to Block 4 [t(7) = 0.8, p = 0.2, d = 0.03]. Similarly, A RM-ANOVA Block-by-Trial ANOVA yielded no significant main effect of Block, nor any significant Block-by-Trial interaction.
Between group comparisons
Figure 3 (right panel) shows Day 2 trial-by-trial directional error averaged across PD patients and controls. Controls (filled circles) showed significantly smaller mean directional error than PD patients (clear circles) in the first trial at the start of Day 2 (controls: −21.1 ± 2.7°, PD: 33.8 ± 3.2°), [t(14) = 3.04, p = 0.009]. Mean directional error also remained smaller in controls than in PD patients throughout Day 2. This observation was supported by a mixed-ANOVA with between-subjects factor Group (PD, Controls) and within-subjects factor Trial (Trial 1–25), which revealed a significant main effect of Group, [F(1, 14) = 9.72, p = 0.008]. The Group-by-Trial interaction was not significant.
Figure 3b also suggests that PD patients show larger asymptotic directional error than controls on Day 2 (PD: −6.2 ± 1.5°, controls: −2.6 ± 1.2°). Mixed-ANOVAs with between-subjects factor Group (PD, Controls) and within-subjects factor Trial (Trials 16–25) showed a main effect of Group [F(1, 14) = 3.78] that just missed significance (p = 0.07) and no significant Group-by-Trial interaction.
Within-group comparisons (savings)
Figure 4 (right panel) depicts mean trial-by-trial directional error on Day 1 and Day 2 in PD patients. Mean directional error appears similar on Day 1 and Day 2. Percent adaptation in PD patients (Fig. 4 right panel, inset) improved an averaged 5.0 % from Day 1 to Day 2, and this difference missed significance [t(7) = 2.0, p = 0.08, d = 0.5]. Supporting this result, a RM-ANOVA with between-subjects factor Block (Block 1, Block 2) and within-subjects factor Trial (Trials 2–15) showed a non-significant main effect of Block [F(1,7) = 0.77, p = 0.4], indicating no significant change in directional error from Day 1 to Day 2 in PD patients. The Block-by-Trial interaction was not significant. The small 5.0 % increase in percent adaptation in PD patients is less than the 20 % increase shown in controls, suggesting deficient savings in PD.
Thus, it appears that while controls improved percent adaptation from Day 1 to Day 2, PD patients did not. Supporting this observation, a Group (PD, controls) by Block (Block 1, Block 2) mixed-ANOVA for percent adaptation showed a significant Block-by-Group interaction, [F(1,14) = 6.07, p = 0.03].
Parkinson’s Disease Sleep Scale scores and savings
Mean PDSS scores were significantly lower in PD patients (100 ± 24) than in controls (136 ± 14.1), [t(14) = 3.53, p = 0.003]. To explore the relationship between sleep and 24 h savings, Pearson’s correlations were conducted on PDSS scores and percent improvement separately for PD patients and controls. Percent improvement was estimated for each participant as the difference in percent adaptation on Day 1 and percent adaptation on Day 2. Percent improvement did not correlate significantly with PDSS sleep scores in PD patients [r = −0.4, p = 0.2] or in controls [r = −0.08, p = 0.8].
The current study showed impaired savings in PD patients within one test session (Experiment 1) and between test sessions separated by 24 h (Experiment 2), with overlearning beyond performance asymptote at initial learning, and with single-target adaptation to ensure repetition of the adapted movement. Experiment 1 showed that despite adapting to the same extent as controls during initial adaptation in Block 1, PD patients showed less savings than controls across successive adaptation blocks. Experiment 2 showed that despite adapting to the same extent as controls on Day 1, PD patients showed less savings than controls on Day 2. Hence, the current study replicates and extends previous findings of impaired savings in PD (Bedard and Sanes 2011; Marinelli et al. 2009).
The findings of deficient savings in PD can be accounted for by assuming that a reduced range of dopaminergic signals in the striatum in PD impairs operant reinforcement learning processes (Frank 2005). PD does not however result in an absolute deficit in savings: in Experiment 1, while PD patients failed to show savings from Block 1 to Block 2, further adaptation trials in Block 2 did result in significant, albeit attenuated savings from Block 2 to Block 3. As PD does not result in absolute dopamine denervation, continued repetition of the adapted motor command is likely to eventually associate it with outcome success. It is noteworthy that deficient savings have been shown even in de novo PD patients (Marinelli et al. 2009) in whom the dopaminergic deficiency is confined to the dorsal striatum, with the ventral striatum intact (Kish et al. 1988). The dorsal striatum is thought to form associations between action and reward (O’Doherty et al. 2004), and an impairment of this process might impair the slow-learning processes responsible for the occurrence of savings in PD patients.
Neural mechanisms underlying savings
The fast-learning component of adaptation learning is thought to be supported by the cerebellum and the posterior parietal cortex (Tanaka et al. 2009), areas with relatively intact function in PD, while the slow-learning component is thought to be supported by the primary motor cortex (M1) (Li et al. 2001; Galea et al. 2010; Cothros et al. 2006; Richardson et al. 2006). How does M1’s proposed role in the slow process fit with the operant reinforcement account of savings? It is possible that when the adapted motor command is executed and results in a rewarding motor outcome, midbrain dopaminergic neurons respond by firing in a phasic fashion, which then modulate M1 activity via the direct and indirect dopaminergic projections from the basal ganglia to M1 (Luft and Schwarz 2009). This in turn may lead to the encoding of longer-term representations of motor learning in M1. The hypothesis that reward-related midbrain dopaminergic signals modulate M1 activity is supported by findings that reward modulates M1 excitability in neurologically intact adults (Thabit et al. 2011) but not in PD patients (Kapogiannis et al. 2011). In PD, a rewarded motor outcome that occurs as a result of executing the adapted movement might elicit a blunted response from midbrain dopaminergic neurons and lead to attenuated modulation of M1 excitability and thus impaired encoding of longer-term representations of motor learning in M1. Future studies examining how M1 excitability is modulated by reward during adaptation learning may help elucidate M1’s role in the model-free slow-learning processes.
Larger asymptotic directional error
An incidental finding of Experiment 1 was that PD patients showed significantly larger asymptotic directional error than controls during relearning but not during initial learning. This finding is reminiscent of a previous report that disrupting M1 activity with single-pulse TMS after movement completion impaired asymptotic performance in a force-field adaptation task (Orban de Xivry et al. 2010). This effect was absent in a gradual adaptation condition in which the strength of the perturbation was incrementally increased from trial to trial, thereby preventing repetition of a fully adapted movement. The finding was interpreted as evidence of M1 involvement in a repetition-dependent process that determined the level of asymptotic performance. While it is unknown why PD patients showed larger asymptotic directional error than controls during relearning in the current study, we suggest that impaired M1 processing caused by deficient dopaminergic signals to M1 led to this pattern of results. Hence, dopamine may play a role in the process that determines the level of asymptotic performance.
Smaller directional error at the beginning of Day 2
Unlike previous studies (Bedard and Sanes 2011; Marinelli et al. 2009), there were no trials with veridical feedback prior to the relearning phase on Day 2 in Experiment 2. This resulted in a novel finding: reduced directional error in the first trial of relearning on Day 2 was present in controls but not in PD patients. Smaller directional error at the beginning of relearning after a 24-h delay has been shown in previous studies (Krakauer et al. 2005) and has been interpreted as evidence of partial maintenance of adaptation learning over 24 h. The absence of this pattern of results in PD suggests a dopaminergic role and might be explained with an operant reinforcement account. In controls, as the adapted motor command was successful on Day 1, a similar motor command was selected on the first trial of Day 2, thereby resulting in smaller directional error. Impaired operant reinforcement mechanisms in PD, however, would result in deficits associating the adapted motor command with outcome success at initial learning on Day 1, thus PD patients do not select a similar motor command on Day 2, which results in large directional errors.
It remains unclear whether processes occurring during the 24-h off-task period contributed to the finding of deficient savings in PD in Experiment 2. Post-learning periods of wakefulness (Brashers-Krug et al. 1996) and sleep (Huber et al. 2004) are thought to support the consolidation of adaptation learning to a longer-term state, conceivably by homeostatically regulating synaptic strength after synaptic potentiation during learning (Landsness et al. 2009). Slow-wave activity in the posterior parietal cortex after initial learning influences the amount of savings shown on the next day, as deprivation of slow-wave sleep reduces savings shown 24 h later (Landsness et al. 2009). Our PD patients did show significantly lower PDSS scores than controls, but these scores did not correlate with savings, giving no support to the proposal that sleep deficits are related to impaired savings in PD. Nonetheless, as the PDSS is a broad clinical measure of sleep impairment in PD that lacks specificity on slow-wave sleep, the possibility that impaired slow-wave sleep contributed to deficient savings over 24 h in Experiment 2 cannot be ruled out. Nevertheless, current and previous findings (Bedard and Sanes 2011) show deficient savings in PD within the same test session, which implies that deficient savings in PD cannot be attributed entirely to deficient slow-wave sleep.
The type and dose of dopaminergic medication was also not experimentally controlled. PD patients were all tested on their usual medication schedules, as going off-medication causes significant discomfort and may have reduced willingness to participate. Dopaminergic drugs such as the dopamine precursor Levodopa and dopamine agonists taken by PD patients may have conflicting effects on reinforcement learning, and thus cause mixed effects on savings. Levodopa ameliorates reinforcement learning deficits in PD (Frank et al. 2004; Rutledge et al. 2009), ostensibly by increasing the availability of dopamine and increasing the range of dopaminergic neurotransmission in response to reward (Frank 2005). Conversely, low doses of the D2/D3 dopamine agonist Pramipexole impairs reinforcement learning in neurologically intact adults (Pizzagalli et al. 2008; Santesso et al. 2009), probably by tonically increasing dopamine levels by binding to pre- and post-synaptic dopamine receptors, thus reducing the potential for reward to phasically modulate dopaminergic neurotransmission (Frank 2005). We speculate that in the current study, while Levodopa may attenuate the savings deficit by increasing the potential range of the dopaminergic response, Pramipexole may exacerbate the savings deficit by blunting the dopaminergic response to reward. Future research should explore potential differential effects of Levodopa and dopamine agonists on savings, as well as dose-dependent effects of dopaminergic medication on savings.
In summary, despite continued training after attaining asymptotic levels of performance with single-target adaptation, PD patients showed impaired savings at relearning, both within the same test session and between test sessions separated by 24 h. Impaired savings in PD might result from impaired operant reinforcement mechanisms caused by blunted dopaminergic reward signals.
Conflict of interest
The authors declare that they have no financial disclosures or potential conflicts of interest.