Changes of upper-limb kinematics during practice of a redundant motor task in patients with Parkinson’s disease

Abstract

The ability to learn novel motor skills is essential for patients with Parkinson’s disease (PD) to regain activities of daily living. However, the underlying mechanisms of motor learning in PD remain unclear. To identify motor features that are distinctively manifested in PD during motor learning, we quantified a rich set of variables reflecting various aspects of the learning process in a virtual throwing task. While the performance outcome improved similarly over 3 days of practice for both PD patients and age-matched controls, further analysis revealed distinct learning processes between the two groups. PD patients initially performed with a slow release velocity and gradually increased it as practice progressed, whereas the control group began with an unnecessarily rapid release velocity, which they later stabilized at a lower value. Performance characteristics related to the timing of ball release and the inter-release interval did not show significant group differences, although they were modulated across practice in both groups. After one week, both groups retained the performance outcomes and underlying kinematics developed over practice. This study underscores the importance of analyzing the multi-faceted learning process to characterize motor skill learning in PD. The findings may provide insights into PD pathophysiology and inform rehabilitation strategies.

Introduction

Parkinson’s disease (PD) is characterized by a progressive degeneration of the dopaminergic nigrostriatal pathway in the basal ganglia. The cardinal symptoms primarily manifest as motor-related features, including bradykinesia, resting tremor, rigidity, and postural instability¹, indicating that damage to the basal ganglia impairs various aspects not only in motor control, but also in motor learning^2,3. Consequently, studying the wide spectrum of motor learning deficits in PD has been of utmost importance to researchers and therapists, especially when designing effective interventions to enhance motor learning. Although recent systematic reviews and meta-analyses have suggested some effective motor learning in PD^4,5, it is crucial to exercise caution in drawing definitive conclusions about the intactness of the ability to acquire novel motor skills in PD patients. This caution arises primarily from the multifaceted nature of motor learning processes⁵: Previous reports that suggest intact PD motor learning do not necessarily imply that PD patients learn like individuals without PD.

Studies that assessed PD motor learning with widely accepted experimental paradigms have largely been inconsistent. Learning a mirror-reversed tracing task was reported to be compromised in PD patients by Schnider et al.⁶, but other findings argued the opposite^7–9. Similarly, learning a rotary pursuit task where the rotation speed changes randomly was found intact in PD patients by Harrington et al.⁸, but not by Haaland et al.¹⁰. Cavaco et al. observed that a group of PD patients with damaged striatum was impaired in learning a rotary pursuit task, but not a mirror drawing task, suggesting that the deficit may be task-specific⁹. Learning a serial reaction time (SRT) task is another case where findings regarding PD patients’ learning ability disagree, showing intact^11–14 and impaired PD motor learning¹⁵. Lastly, PD patients were initially identified to have deficits in adapting upper extremity movement to a novel force-field¹⁶, but this finding was not confirmed in a recent study that included a meta-analysis revealing no impairment in PD sensorimotor adaptation¹⁷.

The heterogeneity in research methodologies across different studies might have been more conclusive if these studies had analyzed underlying kinematic variables that represent movement characteristics during motor learning. Most studies have primarily focused on the performance outcome to show whether PD patients can learn a novel motor skill. Given the inconsistent findings on the “whether” and “what” question, focusing on “how” PD patients learn may help us better understand PD motor learning. Such an attempt necessitates monitoring the emergence and the development of characteristics that construct the performance outcome and addresses the multiplicity of motor learning^18,19. Park and colleagues, for example, demonstrated that fine-grained analyses of multiple kinematic details provide insights on long-term processes of motor learning in healthy individuals^20,21. Accordingly, given the inherent motor deficits in PD, the kinematic details may better characterize PD motor features in motor learning, even if their performance outcomes suggest intact motor learning. In the present study, we aimed to identify distinct motor features that uniquely characterize motor learning in PD patients.

To pay tribute to the multifaceted nature of motor learning, it is essential to examine motor learning in a motor task that allows for multiple realizations to achieve the same goal. Such redundancy exists in almost any task in everyday life¹⁸. Hence, we adopted a motor task in a screen-based environment that simulates throwing a ball to a target. The angle and velocity of the ball at release completely determine the performance outcome (distance from the target) based on the physical model that contains task redundancy²². A previous study on PD motor learning focused on the differences of the distributional characteristics of critical variables in a similar virtual throwing task²³. They found that reducing the degree of error sensitivity (Tolerance), contributed most to the performance errors in PD. While acknowledging the contribution of Tolerance in the early stage of motor learning^19,24, we focused on the kinematic details that determined the performance outcome, i.e., the minimum distance between the ball trajectory and the target. We thus examined the change of the angle and velocity at release as process-oriented variables in addition to the change of the performance outcome.

“Timing” is another process-oriented variable key to perform a throwing task. However, the release timing alone does not determine the performance outcome. A recent study involving the same virtual throwing task with healthy young adults revealed that participants not only reduced timing errors, but also adjusted their arm trajectory through practice, resulting in decreased sensitivity of performance outcomes to release time variability²⁵. Hence, to scrutinize motor learning processes of the same task from a different perspective, the present study investigated if PD patients improved timing accuracy and adjusted their trajectory as practice progresses. Going beyond the timing within a single trial, the release time interval between consecutive throws is another factor influencing the performance outcome. As practicing a novel motor skill involves repetition, a series of discrete movements may evolve into rhythmic movements. A recent study with the same throwing task already reported this and found that increased rhythmicity was associated with improved task outcomes in the throwing task²⁶. We analyzed whether PD patients also exhibited this tendency.

Lastly, we decided to assess the extent to which improved performance and altered movement characteristics persisted a week after the end of practice. Previous studies on motor learning in PD have often reported impaired retention of learned motor skill^27,28. Other studies on PD patients’ throwing ability also found that short-term retention was impaired, evidenced by a significant warm-up decrement, even though long-term retention of up to 9 months remained intact^23,29. However, movement characteristics that underlie the performance outcome have not been assessed regarding persistence, prompting our investigation into various aspects of the retention ability of PD patients.

Methods

Participants

Thirty-five participants were recruited for the study between November 2015 and April 2016. Twenty-four (12 females and 12 males) of them were patients with early-stage PD and eleven (8 females and 3 males) were age-matched control participants. A one-way ANOVA confirmed no significant age differences among the 3 groups (p = 0.32). All participants were right-handed. Diagnoses were made by the United Kingdom Parkinson’s Disease Brain Bank Criteria. The Unified Parkinson’s Disease Rating Scale (UPDRS) and Modified Hoehn-Yahr (HY) stage of the patients were reported to indicate their clinical status. UPDRS part III, the motor examination part, and the following experiment were all performed at medication on condition at which the patients best responded to the individualized levodopa dosage (see Table 1). Global cognitive status was assessed with the Korean version of the Mini-Mental State Examination (K-MMSE). To avoid potential confounding effects between cognitive and motor learning, patients with the K-MMSE score of 25 or higher were recruited. Out of the 24 patients, two patients who practiced the task in the pilot study were excluded from the analysis to control the total amount of practice. Patients were divided into two groups based on their HY stage. Patients with 1 or 1.5 HY stage, mild unilateral involvement, were classified into the PD1 group (6 females and 5 males) and those with 2 or 2.5 HY stage, bilateral involvement, were classified into the PD2 group (5 females and 6 males). Age-matched control participants were recruited from a local community.

Table 1.

Demographic and clinical characteristics of the PD groups and the control group.

Group	PD1 (n = 11)			PD2 (n = 11)			Control (n = 11)
	M	SD	R	M	SD	R	M	SD	R
Age (years)	60.5	5.05	50–67	63.2	5.47	49–69	63.1	5.52	52–68
Duration of PD (years)	6	6.47	3–11	6.1	1.85	3–9
UPDRS III	7.4	5.66	1–10	8.79	6.01	2–20.5
HY stage	1.14	0.23	1–1.5	2.23	0.26	2–2.5
K-MMSE	28.73	1.39	27–30	29	1.23	25–30
LEDD (mg/day)	583	187	280–887.5	750	270	450–1260

PD1 patients with Hoehn-Yahr stage 1 and 1.5, PD2 patients with Hoehn-Yahr stage 2 and 2.5, C control group, M mean, SD standard deviation, R range, UPDRS Unified Parkinson’s Disease Rating Scale, HY stage Hoehn-Yahr stage, K-MMSE Korean-Mini Mental Status Examination, LEDD Levodopa Equivalent Daily Dose.

Demographics and clinical characteristics of participants are shown in Table 1. None of the patients reported previous or current neurological disorders other than PD at the time of participation. We also excluded patients with dementia, disabling motor fluctuations or dyskinesia, or any other advanced disease with HY stage 3 or more. Participants in the control group self-reported that they had no problem in their activities of daily living and had no history of diagnosis with any neurological disease at the time of study participation. All participants read and signed the written informed consent prior to the experiment.

All experimental procedures were approved by Seoul National University Boramae Hospital Institutional Review Board (IRB no. 26-2015-124). All experiments were performed in accordance with all ethical requirements of the institution and with the 1964 Helsinki Declaration and its later amendments.

Task and apparatus

The participants performed a virtual throwing task^22,30. While seated on a chair, they rested their dominant forearm on a manipulandum fixed to a table adjusted to a comfortable height for each individual. The manipulandum was an aluminum bar (40 cm in length) with a pivot on one side and a ball-shaped handle on the other side. A force sensor was attached to the ball. To measure the rotation of the arm, an optical encoder (H25 Absolute Encoder, BEI Sensors, Goleta, CA) was placed under the pivot point of the manipulandum. The encoder was affixed to a base plate that was attached to the table. Measured data were recorded by a data acquisition board at the sampling rate of 1000 Hz (NI USB-6343, National Instruments, Austin, TX). With the participants’ elbow placed above the encoder and their forearm fastened by a Velcro strap, they grasped the handle and pressed the force sensor with their index finger.

The task goal and the ball dynamics of the motor task resembled a British pub game, Skittles, or the American tetherball game on playgrounds. Its visualization is analogous to a top-down view of the real-world game. An LCD monitor (width: 121.76 cm, height: 68.49 cm) was placed in front of the participants. The distance between the monitor and the participant’s shoulder was 100 cm. Figures 1a,b illustrate the entire setup. In the center of the computer-generated display was a red circle of 3.3 cm in diameter, marking the fixed center post. A circular target of 1.1 cm in diameter was presented, with its center 5.5 cm to the left and 3.3 cm above the center post. The virtual arm was represented as a solid bar of 4.4 cm in length, fixed at one end. The fixed end (pivot point) of the virtual arm was centered at 16.5 cm below the center post (Fig. 1c).

Fig. 1.

Experimental Setup. (a) A top-down view of a participant sitting in front of the monitor. The elbow angle (a) is defined as the angle between a line parallel to the trunk and the forearm. (b) A side view of the experimental setup. (c) An example display of the task before the virtual ball (white circle) is released. The yellow circle is the target, the red circle is the center post, and the purple bar is the rotating forearm controlled by the participant. (d) A schematic of the ball trajectory and the definition of the error.

Participants were instructed to throw a virtual ball to hit the target. Once they pressed the sensor, a virtual ball (1.1 cm in diameter) appeared at the distal end of the virtual arm. When participants extended their forearm in a horizontal plane, the virtual ‘arm’ on the screen moved synchronously. Releasing the finger from the force sensor triggered the release of the virtual ball, which traversed a trajectory initialized by the angle and velocity of the participant’s arm at release. The dynamics of the ball trajectory after release were simulated by the two-spring model described in a previous study²². Both the movements of the arm and the simulated trajectory of the ball were displayed on the screen in real-time. All participants were asked to release the virtual ball as they rotated the elbow in the clockwise direction. Although the participants did not explicitly acknowledge the exact model or the underlying dynamics of the ball trajectory displayed on the monitor, they were informed that the ball after release would trace out an elliptic trajectory as the ball traversed around the center post to hit the target. Performance error was visualized to the participants after every trial as a partial trajectory around the target was displayed on the computer screen. Figure 1d depicts the execution (angle and velocity at release) and result variable (performance error) shown on the screen. This virtual task was designed such that the angle and angular velocity of the rotating arm at release fully determined the ball trajectory and thereby its error from the target.

Experimental procedure

The experiment consisted of 3 practice sessions and 1 retention session. The 3 practice sessions took place for 3 consecutive days, with a single session per day. Each session included 6 blocks of 60 throwing trials (360 trials in total). There was a 24 h break between each practice session and a week-long break between the end of the practice and the retention session. On the first day, all participants received the instructions, including how to position the forearm on the manipulandum, how to grasp the ball while simultaneously applying pressure to the force sensor with the index finger, and how to release the virtual ball. However, no guidance was provided regarding how to accurately target the object, i.e., the angle for ball release and the required arm movement speed. A one-minute break followed the completion of each block, and additional breaks were permitted at any point during the task upon the participants’ request.

Performance error

As the outcome measure, performance error was defined as the minimum distance between the ball trajectory after release and the center of the target. This minimum could have occurred at different locations, depending on the specific combination of release angle and velocity. For the trials that hit the center post, the trajectory until the ball hit the center post was used for calculating the minimum distance.

Angle and velocity at release

This virtual task was designed such that the angle and angular velocity of the rotating arm at release fully determined the ball trajectory and thereby its error from the target based on the physical model of the task²². As the two execution variables determined one performance outcome, a mathematically infinite number of angle-velocity pairs yield the same amount of error. Figures 2a–f showcase the relationship between the execution variables and the performance outcome. The solution space is spanned by the two execution variables. The turquoise U-shape area on the solution space depicts the set of angle-velocity pairs leading to a target hit. Hollow circles overlayed are the errors mapped onto the solution space as a function of release angle and velocity. The experimental details of the online calculation were depicted in a previous study²⁵.

Fig. 2.

Changes of the performance outcome and the execution variables over practice and retention. (a) The solution space of the task with a block of trials performed by a control participant on Day 1 (hollow circles). The color code on the 2D space represents the performance error. The turquoise U-shape area indicates a set of angle-velocity pairs that lead to a target hit. The s circles represent individual trials. (b) The solution space of the task with a block of trials performed by the same participant as in panel (a) on Day 3. (c) The solution space of the task with a block of trials performed by a participant in PD1 on Day 1. (d) The solution space of the task with a block of trials performed by the same participant as in panel (c) on Day 3. (e) The solution space of the task with a block of trials performed by a participant in PD2 on Day 1. (f) The solution space of the task with a block of trials performed by the same participant as in panel (e) on Day 3.

Timing error and timing window

We calculated both the timing error and the timing window of each throw²⁵. The difference in time between the actual release and the ideal release for a given trajectory rendered the timing error (Figs. 3a,b). The ideal release of a trajectory was found with the following steps. An arm trajectory was first converted from a sequence of angle-velocity pairs to a sequence of the corresponding performance error, assuming each time point of the arm trajectory was a ball release. The ideal release was the time of the minimum performance error (Figs. 3c,d).

Fig. 3.

Changes of the timing error and window. (a) The solution space of the task with the arm trajectory (white line) of a single trial performed by a participant on Day 1. The turquoise U-shape band is the area that yields a target hit. The white trajectory depicts the elbow angle profile in a single trial and the black circle indicates the moment at which the virtual ball was released. (b) The solution space of the task with the arm trajectory (white line) of a single trial performed by the same participant on Day 3. (c, d) The definition of the timing error and timing window from the same trial as in (a) and (b), respectively.

The timing window refers to the time interval of the arm trajectory during which a ball release would hit the target (Figs. 3c,d). This corresponds to the interval when a trajectory was within a band around the solution manifold. When a trajectory crossed the manifold twice, the segment closer to the ideal release time was chosen. This measure quantifies the sensitivity of the trajectory to timing error. A trajectory with a longer timing window allows for a broader range of ball releases acceptable as a target hit. The unit of the two measures was milliseconds, and 60 timing error and timing window values from one block were each averaged to quantify block-wise measures.

Figures 3a,b exhibit two representative arm trajectories of a PD2 participant. The trajectory recorded on Day 1 shows minimal overlap with the solution manifold, and the actual release is outside a narrow timing window. In comparison, the trajectory on Day 3 aligns more closely with the solution manifold. The timing window expanded, and the actual release got closer to the ideal release, reducing the timing error.

Inter-release interval

Periodicity is the essential characteristic of rhythmic movement, i.e., the same posture or event should recur at invariant intervals³¹. However, strict periodicity is unlikely in human behavior, and thus rhythmicity implies some variation around a constant interval. Figures 4a,b display the time series of arm trajectories of one PD2 participant on Day 1 and Day 3. The vertical lines mark the moments of ball release. A comparison between Day 1 and Day 3 delineates successive throws that developed into an approximately periodic sequence. To quantify the degree of periodicity, the moments of ball release served as landmarks; the interval between two consecutive releases defined the inter-release interval (IRI). The median of all IRIs within a block characterized each participant’s block-wise performance, as the distribution of IRI was highly leptokurtic and skewed³². IRIs longer than 10 s were removed prior to the calculation, as they indicated voluntary rest. The rhythmicity of throws was estimated by the variability of IRI, quantified by the quartile variation coefficient (QVC) of IRI for each block, as the distribution of IRI was highly leptokurtic and skewed³².

Fig. 4.

Changes of the Inter-release interval over practice and retention. (a) The velocity profile of elbow rotation in a series of trials performed by a PD2 participant on Day 1. The black circles and the vertical lines indicate the moment of release. (b) The velocity profile of elbow rotation in a series of trials performed by the same participant on Day 3.

Statistical analysis

To examine the change of performance between the first and the last practice session and to test whether learning differed across the three groups, each dependent variable in the first and the last practice session for all three groups was submitted to a 2 × 3 mixed ANOVA with Session (Day 1 and 3) as a within-subject effect and Group (Control, PD1 and PD2) as a between-subject effect. When the ANOVA reported a significant interaction, we fitted nonlinear mixed models to take within- and between-subject variabilities into account using block-by-block changes across the three practice sessions. The growth curve fitted to the data was $y=a-\left(a,-,b\right)\ast e_{}^{-\left(t,-,1\right)\ast k}$, where t is the block number, a is the asymptote, b is the initial value of the dependent variable y, and k is the rate of change. Note that y = b when t = 1. The model had Group as its fixed effect and allowed random slope and intercept per subject. The mixed-effect model provides a flexible framework in which population characteristics were modeled as fixed effects and individual-specific variation was modeled as random effects. All statistical analyses were conducted with the statistical software R 3.3.1 and its package nlme³³.

In case of significant changes of a dependent variable after practice shown by the ANOVA or the nonlinear mixed-effect model, we tested if the change persisted after a week of retention interval. To test the retention effect, we examined whether each dependent variable in the first block of the retention session differed from the last block in the last practice session with respect to the three groups. Prior to each ANOVA, we conducted a normality check by creating Q-Q plots. When the normality assumption was not met, we replaced the analysis with a robust ANOVA using the wrs2 package in R³⁴.

Results

The characteristics of the study participants are summarized in Table 1 and all participants completed the task successfully.

Performance errors

We first analyzed the change in the performance error, the performance outcome explicitly instructed to the participants, and the persistence of the changes. Figure 5a and Supplementary Fig. S1 show a reduction of the performance errors over 3 days of practice and their retention after a 1-week retention period for all three groups. A main effect of a 2 (Session) × 3 (Group) mixed ANOVA model was significant only in Session (Session: F_1,30 = 34.52, p = 1.98 × 10^–6, η² = 0.28; Group: F_2,30 = 1.72, p = 0.196, η² = 0.07) without a significant Session × Group interaction (F_2,30 = 1.71, p = 0.197, η² = 0.04), indicating that all three groups improved the throwing performance over three days. Another 2 × 3 mixed ANOVA that examined retention showed no significant main effect (Session: F_1,30 = 0.002, p = 0.97, η² = 2.89 × 10^–5; Group: F_2,30 = 0.11, p = 0.90, η² = 3.88 × 10^–3) or an interaction (F_2,30 = 0.013, p = 0.99, η² = 3.92 × 10^–4), indicating that the reduced error over practice remained low after a week of retention interval. Altogether, all three groups displayed signs of learning and retention of their throwing performance based on the performance error.

Fig. 5.

Changes of the performance outcome and release angle and velocity of the three groups over practice and retention. (a) Changes of the performance error. Colored shades in this and all other panels indicate the standard error of the mean of each group. (b) Changes of the mean release angle. (c) Changes of the release velocity. The black lines indicate the exponential fits of the groups.

Kinematics in the release velocity and angle

Having shown the ability to improve and retain the performance outcome in PD patients, we analyzed the kinematic aspects underlying the performance outcome. Figures 2a,b compared the change of performance of a control participant in solution space from Day 1 to Day 3. Figures 2c,d showed the change in performance of a PD1 patient and Figs. 2e,f showed the change in performance of a PD2 patient. While all improved the performance outcome, their “learning paths” differed from each other. To examine the group difference, Fig. 5b and Supplementary Fig. S1 show that the angle at release did not change throughout the 3-day practice and 1-week retention session for all three groups. By contrast, Fig. 5c and Supplementary Fig. S1 show that the velocity at release, while exhibiting two different values, converged into a certain asymptote.

The same two-way mixed ANOVA confirmed the contrast between the two executive variables. For angle at release, neither Session nor Group effect was significant (Session: F_1,30 = 1.237, p = 0.275, η² = 0.04; Group: F_2,30 = 0.004, p = 0.337, η² = 0.07) nor was there a significant Session × Group interaction (F_2,30 = 0.004, p = 0.996, η² = 2.67 × 10^–4). The standard deviation (SD) of angle at release decreased during the three practice sessions, whereas the three groups did not show a significant difference (Session: F_1,30 = 9.946, p = 0.004, η² = 0.07; Group: F_2,30 = 1.19, p = 0.317, η² = 0.079; Session × Group interaction: F_2,30 = 1.068, p = 0.356, η² = 0.018). The reduced SD remained low after the one-week-long retention period (Session: F_1,30 = 0.002, p = 0.962, η² = 2.42 × 10^–5; Group: F_2,30 = 0.80, p = 0.457, η² = 0.035; Session × Group interaction: F_2,30 = 0.549, p = 0.583, η² = 0.012).

For velocity at release, while both Session and Group main effects were significant (Session: F_1,30 = 5.01, p = 0.03, η² = 0.038; Group: F_2,30 = 5.06, p = 0.012, η² = 0.20), and the Session × Group interaction was also significant (F_2,30 = 3.58, p = 0.04, η² = 0.054). To further analyze the characteristics of the change across the three groups, a nonlinear parameter estimation was performed. The exponential curve (specified in Methods) fitted to individual mean release velocity over the 18 blocks (6 blocks × 3 days) showed that k, the rate of change across the blocks was significantly higher for the Control group (1.18) compared to the two PD groups (both 0.09; F_2,553 = 5.389, p = 0.005). This implies that the release velocity in the Control group plateaued earlier compared to PD1 and PD2. As depicted in Fig. 5c, the Control group started from a relatively high release velocity and converged into a lower value. In contrast, both PD groups started from a lower value and converged into a high release velocity. The nonlinear mixed-effect analysis confirmed these observations: The value of a, the asymptote, did not show a significant difference across the three groups (Control: 279, PD1: 257, PD2: 266; F_2,553 = 0.700, p = 0.497) while the value of b, the initial value, showed a significant group effect (Control: 300; PD1: 215, PD2: 179; F_2,553 = 6.114, p = 0.002).

The SD of velocity at release also decreased over the three practice sessions whereas the three groups did not show a significant difference (Session: F_1,30 = 9.946, p = 0.004, η² = 0.07; Group: F_2,30 = 1.19, p = 0.317, η² = 0.079; Session × Group interaction: F_2,30 = 1.068, p = 0.356, η² = 0.018). The reduced SD of release velocity stayed low after a week (Session: F_1,16.16 = 2.48, p = 0.135, η² = 0.019; Group: F_2,11.33 = 0.54, p = 0.457, η² = 0.036; Session × Group interaction: F_2,11.44 = 0.735, p = 0.501, η² = 4.84 × 10^–4).

Timing error and window

The change of timing error and timing window is illustrated with a single trial from a PD2 patient in Figs. 3a–d, respectively. The participant not only reduced the timing error, but also expanded the timing window by adjusting their arm trajectory. Figures 6a,b and Supplementary Fig. S2 show the group results for two timing variables. It appears that PD1 and PD2 started with higher timing error and smaller timing window compared to Control and the values for the PD patients became indistinguishable from the Control participants at the end of practice. However, two-way mixed ANOVAs did not show a significant Group effect for either of the two variables (Timing error: F_2,30 = 2.10, p = 0.14, η² = 0.08; Timing window: F_2,30 = 2.47, p = 0.10, η² = 0.10). There was no Group × Session interaction (Timing error: F_2,30 = 1.602, p = 0.218,η² = 0.038; Timing window: F_2,30 = 0.212, p = 0.810, η² = 0.004). A significant Session effect was found in both variables (Timing error: F_1,30 = 10.267, p = 3.0 × 10^–3,η² = 0.112; Timing window: F_1,30 = 18.24, p = 2.0 × 10^–4, η² = 0.16). For retention, two-way mixed ANOVAs revealed that the improved timing accuracy at the end of practice stayed low after one week of no practice for both variables. There was no Session effect for both variables (Timing error: F_1,30 = 0.044, p = 0.834,η² = 6.12 × 10^–4; Timing window: F_1,30 = 0.0019, p = 0.965, η² = 2.89 × 10^–5). No Group effect was found for both variables (Timing error: F_2,30 = 0.096, p = 0.909, η² = 3.74 × 10^–3; Timing window: F_2,30 = 0.105, p = 0.900, η² = 3.88 × 10^–3) without an interaction (Timing error: F_2,30 = 0.260, p = 0.773,η² = 7.14 × 10^–3; Timing window: F_2,30 = 0.013, p = 0.987,η² = 3.92 × 10^–4). These results indicate that the timing metrics, timing error and timing window, did not exhibit differences between the three groups and all groups improved the temporal aspects of the task.

Fig. 6.

Changes of the timing variables of the three groups over practice and retention. (a) Changes of the timing error. Colored shades in this and all other panels indicate the standard error of the mean of each group. (b) Changes of the timing window. (c) Changes of the median inter-release interval. (d) Changes of the quartile variance coefficient of the inter-release interval.

Inter-release intervals and variability

To investigate the periodicity of repetitive throwing, we analyzed the IRIs and their variability. Figure 6c and Supplementary Fig. S2 show the median IRI over practice and retention. For most participants, the interval did not vary across practice and retention for all three groups. Figure 6d shows that the QVC of the IRIs decreased over practice and remained low in the retention session. Two-way mixed ANOVAs for the practice sessions revealed that the IRI-QVC decreased over practice for all three groups, whereas the median IRI did not change across practice (Group effect for IRI: F_2,30 = 0.461, p = 0.64, η² = 0.023; Group effect for IRI-QVC: F_2,30 = 0.337, p = 0.717, η² = 0.019; Session effect for IRI: F_1,30 = 0.911, p = 0.347,η² = 0.007; Session effect for IRI-QVC: F_1,30 = 3.657, p = 0.006, η² = 0.016; interaction for IRI: F_2,30 = 0.626, p = 0.541,η² = 0.009; interaction for IRI-QVC: F_2,30 = 0.669, p = 0.520,η² = 5.85 × 10^–3). The reduced IRI-QVC remained low after a week without practice (Group: F_2,30 = 1.239, p = 0.304, η² = 0.076; Session: F_1,30 = 0.368, p = 0.549, η² = 0.012; Session × Group: F_2,30 = 2.532, p = 0.096, η² = 0.144). Therefore, participants in all three groups reduced the variability of the IRI rather than the IRI per se, as the median and the variability did not significantly differ between the three groups.

Discussion

Like most other motor tasks in everyday life, the goal of a throwing task can be achieved in infinitely different ways, i.e., the task is redundant. We exploited this redundancy to test whether PD exhibited differences in motor learning processes. We found slow arm velocities at ball release during the early stage of motor learning to be the only feature that was different in PD patients. Movement slowness, or bradykinesia, has been known to be one of the most prominent motor features in PD³⁵. However, it is noteworthy that the present study demonstrated the improvement of the velocity at release during practice together with performance after a 1-week retention interval. While overcoming slow movement in PD has been previously reported^36,37, it is intriguing to note that the participants in the present study were not explicitly instructed to increase the release velocity.

What then made the PD patients throw faster? The successful learning and retention could be in line with kinesia paradoxa, a phenomenon that refers to PD patients suddenly performing complex movements fluently while struggling to perform simple movements³⁸. Such paradoxical movements may be provoked by visual stimuli^39,40. More generally, external sensory cueing has been suggested as an effective physical therapy strategy⁴¹. Hence, we speculate that presenting the external goal such as the target and the ball trajectory provided the PD patients with sufficient motivation to throw balls faster. Nevertheless, further studies are necessary to identify conditions that led to an increase of ball release velocity.

What are the neural underpinnings for increasing the release velocity across practice? First, PD patients exhibited initial difficulties in releasing the virtual ball while moving the arm, i.e., the multi-joint coordination, which led to a low release velocity. Mazzoni and colleagues suggested that PD patients are more sensitive to the effort associated with faster movements and thus less motivated to make such movements³⁷. While bradykinesia is closely related to the slow ball release⁴², rigidity could also be responsible for the initial low release velocity⁴³. Previous studies suggested that compromised modulation of striatal dopamine, particularly the tonic release of the neurotransmitter, could be the source of decreased motivation^44,45. Moreover, recent animal studies on dopaminergic neurons in the basal ganglia suggested that dopamine depletion is associated with reducing movement vigor^46,47. Accordingly, deficits in the cortico-thalamo-striatal circuit that exacerbate the striatal damage could be a potential cause of the decreased motivation to make faster movements in PD patients⁴⁸.

On the other hand, PD patients still did increase overall release velocity as they improved their performance with practice. Such intactness could be the result of relatively less impaired basal ganglia and relevant circuits in mild PD patients. However, it is also possible that there is another circuit that can compensate for the damage to the cortico-striatal circuit. A candidate is the cortico-cerebellar circuit. Mentis and colleagues observed that PD patients showed a greater volume of cerebellar activation as learning progressed⁴⁹, but this was reported in a sequence learning task that arguably involves more cognitive contributions. It has also been suggested that the functional interaction between the cortico-cerebellar circuit and the cortico-striatal circuit is critical for learning a new motor skill^5,50. Additionally, the paradoxical movement has been explained by the utilization of the cortico-cerebellar circuit^39,51. Nevertheless, it is only speculative at this point whether the cortico-cerebellar circuit is involved in a compensatory mechanism. For example, a recent functional magnetic resonance imaging study showed co-activation of the two circuits when PD patients performed ankle dorsiflexion in the more affected side⁵². A systematic review of the effects of transcranial direct current stimulation on upper limb motor learning also supported the non-significant effect of stimulating the cerebellum⁵³, making the role of the cortico-cerebellar circuit in PD patients’ motor learning more elusive. A future study with participants having cerebellar damage using the same experimental task may reveal more about the relationship between the two circuits.

The timing of ball release in the arm trajectory allows us to view yet another aspect of the learning process from the angle and velocity at release. Achieving the goal of hitting the target consistently would thus entail finger extension movements temporally coordinated with elbow rotations. Both PD and age-matched control participants improved their release timing across practice. Our finding is consistent with previous motor learning studies with PD patients in the virtual throwing task^23,29. However, while Pendt and colleagues found delays in release timing at the beginning of each daily session, the present study did not observe such a warm-up decrement²⁹. Additionally, we found that PD patients also adjusted the arm trajectory to make the timing error insensitive to the performance outcome. Further studies are necessary to elucidate the relationship between the release timing and the PD symptom severity.

The increased regularity of the IRI shown in the present study is analogous to a recent virtual throwing study that shows the tendency to transition to rhythmic movement as repetitive trials progress²⁶. In a kinematic profile, rhythmicity was also determined by the dwell times between two consecutive movements³¹. However, we were unable to record the angle profile between the end of a trial and the beginning of the subsequent trial. A recent study by Breska and Ivry showed a double dissociation between impaired regions of the brain (cerebellar vs. basal ganglia degeneration) and the performance on the interval-based and rhythm-based timing prediction tasks⁵⁴. PD patients did not benefit from repetitive cues, but they improved temporal prediction when a single interval was presented. Nevertheless, our results suggest that the degeneration in the basal ganglia did not influence rhythmic pattern generation that emerged from the sequence of discrete movements. Although numerous studies have shown impaired rhythmic pattern generation in PD patients^55–57, the present study demonstrated that the originally discrete task became rhythmic as PD patients improved their performance and optimized the underlying kinematics.

Our follow-up tests revealed that the PD patients performed at a consistent level immediately after a one-week retention interval. This contrasts with other studies that demonstrated a lack of retention of an acquired motor skill²⁷. Particularly, previous studies with a similar virtual throwing setting showed that PD patients improved their throwing performance largely over practice; at the beginning of a retention session after one and five days they reported a performance drop^23,29. Contrary to these previous studies, the successful one-week retention in our study is evidence of robust motor learning in PD patients. Not only did we see the retention effect in the primary performance metric, but other underlying kinematics that changed over practice also demonstrated persistence after one week. Two factors regarding the experimental setup may explain the difference. First, the patients in the present study had on average lower HY scale and UPDRS scores. Second, the present study trained with 360 throws each day for three days, whereas Pendt and colleagues trained with only 200 throws per day²⁹. Future studies may investigate the interplay between the neurophysiological representation of clinical scoring of PD and the benefits of practice intensities in their effect on retention.

We acknowledge several limitations in the study design and interpretation of the results. In the present study, we recruited PD patients whose HY scores were less than 3 and had excellent responses to dopaminergic therapy. Hence, the results cannot be generalized to patients with severe PD symptoms and sub-optimal response to treatment. On the other hand, it is worth considering that patients with higher HY scores tend to have non-dopaminergic degenerations that are known to interfere with motor control and learning. By narrowing the range of the HY scores, we focused on the association between nigrostriatal dysfunction and motor learning. Future studies that include a cohort of PD patients with a broader range of severity and a larger sample size will enable us to examine the correlation between the kinematic characteristics and the clinical status or the amount of dopamine dosage. Related to this, a comparison between learning with the more affected side and learning with the less affected side would be of interest. As a preliminary test, we compared the degree of motor learning and performance between patients who performed with the more affected limb (6 from PD1 and 6 from PD2) and those with the less affected limb (5 from PD1 and 5 from PD2) within each PD group. None of the metrics showed a significant group difference (p-values of 0.0954 or higher, Mann-Whitney-U tests, uncorrected), which warrants a larger cohort study.

Another limitation was that all patients were on dopaminergic medication cycles (Levodopa). Consequently, we could not observe a significant correlation between the performance variables and their symptom severity. We suspect that the treatment reversed the degeneration of the dopaminergic system to some degree, altering the patients’ motor learning capacity. Future studies can compare the test results of de novo PD patients comparable to our sample before and after medication. Such studies can articulate the effect of neurophysiological changes on their motor learning in PD patients, not moderated by treatments restoring nigrostriatal circuits. The amount of restoration and the corresponding gain in motor learning capacity can also be delineated. This approach would enable us to further elucidate possible strategies for PD rehabilitation.

In summary, we investigated multiple process-oriented measures in a goal-directed movement throughout the motor learning process. The need for studying such kinematic details has been acknowledged in the motor learning literature⁵⁸ but has been understudied, especially in individuals with movement disorders such as PD. Results suggest that, although performance improvement of PD patients could be commensurate with that of age-matched controls, the arm velocity at ball release reflects initial difficulties in multi-joint coordination between the finger and the elbow in PD. The practice-induced motor learning demonstrated despite the initial differences in PD is noteworthy.

To obtain more insights into designing effective therapeutic interventions for PD patients, it is imperative for more PD motor learning studies to investigate the relationship between performance outcomes and movement characteristics. Based on our findings, we can propose two ideas for designing a task in a rehabilitation program for PD patients. First, the task should have redundancy. Allowing multiple combinations of kinematic variables for success is analogous to providing performers with options. Again, depending on the severity of symptoms, patients may adopt strategies that differ from those of unimpaired participants, but without loss in the primary performance outcome. Constraining strategies to one solution would restrict patients from finding their own strategy to learn a new motor skill. Second, consider a discrete task that can be repeated many times without losing the performer’s interest. Motivation in PD patients is an important factor for a successful experiment where they reveal their potential. All too often laboratory tasks only excite the scientists. Many advances in game design can help to make the experimental research not only more entertaining, but also more revealing⁵⁹.

Author contributions

S.-W.P., J.-K.R. and K.-M.L. conceived the study. K.-M.L., J.-K.R., S.-W.P. and D.S. provided experimental support and supervision. J.-Y.L. and K.-M.L. provided their clinical expertise for study protocols. J.O. performed the experiment protocols for data procurement. S.-W.P. and J.O. analyzed and visualized data. J.O., S.-W.P. and M.S. wrote the original draft. All authors reviewed and edited the manuscript and approved the final version.

Data availability

All de-identified data generated and analyzed during the present study are available from the corresponding author upon reasonable request.

Competing interests

S.-W.P is an Editorial Board Member in Scientific Reports. Other authors declare no competing interests.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-024-76015-7.