Experimental Brain Research

, Volume 191, Issue 2, pp 221–236

Decrease in cortical activation during learning of a multi-joint discrete motor task

Authors

    • Graduate School of EducationUniversity of Tokyo
  • Gentaro Taga
    • Graduate School of EducationUniversity of Tokyo
    • CREST, Japan Science and Technology Agency
Research Article

DOI: 10.1007/s00221-008-1518-2

Cite this article as:
Ikegami, T. & Taga, G. Exp Brain Res (2008) 191: 221. doi:10.1007/s00221-008-1518-2

Abstract

Understanding how the brain learns motor skills remains a very challenging task. To elucidate the neural mechanism underlying motor learning, we assessed brain activation changes on a trial-by-trial basis during learning of a multi-joint discrete motor task (kendama task). We used multi-channel near-infrared spectroscopy (NIRS) while simultaneously measuring upper limb movement changes by using a 3D motion capture system. Fourteen right-handed participants performed the task using their right upper limb while sitting a chair. The task involved tossing a ball connected by a string to the kendama stick (picking up movement) and catching the ball in the cup attached to the stick (catching movement). Participants performed a trial every 20 s for 90 trials. We measured the hemodynamic responses [oxygenated hemoglobin (oxy-Hb) and deoxygenated hemoglobin (deoxy-Hb) signals] around the predicted location of the sensorimotor cortices on both hemispheres. Analysis of the NIRS data revealed that the magnitudes of the event-related oxy-Hb responses to each trial decreased significantly as learning progressed. Analysis of movement data revealed that integrated upper limb muscle torques decreased significantly only for the picking up movements as learning progressed, irrespective of the outcome of the trials. In contrast, dispersion of the movement patterns decreased significantly only for the catching movements in the unsuccessful trials. Furthermore, we found significant positive correlations between the changes in the magnitudes of the oxy-Hb responses and those of the integrated upper limb muscle torques during learning. Our results suggest that the decrease in cortical activation in the sensorimotor cortex reflects changes in motor commands during learning of a multi-joint discrete movement.

Keywords

NIRSMotor learningMulti-joint movementsDiscrete movements

Introduction

Understanding how learning a motor skill is associated with changes in human brain activation remains a challenge. Behavioral studies have suggested that the brain might update the mechanism for sensorimotor control in every trial when participants learn a novel motor task (Thoroughman and Shadmehr 2000; Donchin et al. 2003). These studies have primarily focused on discrete movements, which are defined as the movements that are bounded by identifiable postures (Hogan and Sternad 2007). Discrete movements are different from rhythmic or sequential movements, in that, they require control of starting and stopping a short bout of the single movement (van Mourik and Beek 2004). Reaching to grasp or place an object at a location and kicking, throwing, or striking a ball are general examples of discrete movements, which are ubiquitously found in everyday behavior or sports actions. Despite a large number of behavioral studies on the learning of discrete movements, limited information is available about event-related changes in brain activation associated with a single discrete movement and the time evolution of the activation with practice. The aim of the present study was to examine the brain activation changes on a trial-by-trial basis during learning of a multi-joint discrete motor task by using multi-channel near-infrared spectroscopy (NIRS) while the movements were measured using a 3-dimensional (3D) motion capture system.

We used NIRS to investigate human brain activation. NIRS can noninvasively measure cerebral hemoglobin oxygenation (Jöbsis 1977) and has been used to assess functional brain activation in adults (Chance et al. 1993; Kato et al. 1993; Hoshi and Tamura 1993; Villringer et al. 1993) and in infants (Taga et al. 2003; Taga and Asakawa 2007). NIRS signals reflect the fact that blood flow response to functional activation is larger than the increase in oxygen consumption, similar to the blood oxygenation level-dependent (BOLD) contrast in fMRI (Villringer and Chance 1997). The oxygenation response typically consists of an increase in oxygenated hemoglobin (oxy-Hb) and a slight decrease in deoxygenated hemoglobin (deoxy-Hb) over an activated cortical region (Obrig and Villringer 2003). Because NIRS imposes fewer body constraints, the technique has been used to investigate brain activation in relation to motor function (Colier et al. 1999; Franceschini et al. 2003; Hatakenaka et al. 2007; Hirth et al. 1996, 1997; Huppert et al. 2006; Jasdzewski et al. 2003; Kleinschmit et al. 1996; Maki et al. 1995; Obrig et al. 1996; Sato et al. 2005, 2006; Shimada et al. 2004; Watanabe et al. 1996; Wolf et al. 2002). Recent developments in NIRS technique allow for measurement of cortical activation of natural movements in nonrestrictive environments, such as during daily human activities (Okamoto et al. 2004a), walking (Miyai et al. 2001), or running (Suzuki et al. 2004) and for the detection of cortical activation in relation to a brief event such as a 2-s finger tapping task (Huppert et al. 2006) or a 2-s hand-opening/closing task (Jasdzewski et al. 2003). In the present study, we expected that NIRS could be used to detect cortical activation associated with a single trial of a discrete movement task. We then designed an experiment to monitor changes in activation of the sensorimotor cortex on a trial-by-trial basis while participants practiced the motor task.

With regard to brain activation changes during motor learning, previous studies using fMRI or PET have shown that the suppression and/or enhancement of brain activation is associated with learning of simple motor tasks in restrictive environments, including sequential finger tasks (Seitz et al. 1990; Karni et al. 1995, 1998), serial reaction time tasks (Pascual-Leone et al. 1994; Berns et al. 1997; Grafton et al. 1998; Honda et al. 1998; Müller et al. 2002), sequential tasks with trial and error (Jenkins et al. 1994; Jueptner et al. 1997a, 1997b; Sakai et al. 1998; Toni et al. 1998), pursuit motor tasks (Grafton et al. 1992, 1994), tracking tasks (Imamizu et al. 2000), and bimanual coordination tasks (Debaere et al. 2004; Puttemans et al. 2005). Although neuroplastic changes during motor learning were associated with widely distributed regions including cortical and subcortical areas (Jueptner et al. 1997a; Hikosaka et al. 1999, 2002; Doyon et al. 2003; Sanes 2003; Ashe et al. 2006), the primary motor or the sensorimotor cortex have been shown to exhibit a variety of learning-related changes in activation. While some studies revealed decreases in magnitude of brain activation during motor learning (Toni et al. 1998; Floyer-Lea and Matthews 2004), other studies revealed the opposite results (Karni et al. 1995, 1998; Debaere et al. 2004; Hluštík et al. 2004). Because these studies have focused on continuous movements or a sequence of simple movements, the neural mechanism underlying discrete movements remains unclear.

In the present study, we chose a kendama task, which represents a multi-joint discrete motor task using a tool. The kendama task consists of a quick and dynamic motion by a single upper limb, whereby a ball connected by a string to a kendama stick is tossed and caught in a cup attached to the stick. Since the task requires appropriate manipulation of a novel tool, precise control of the multi-joint limb, and eye–hand coordination, it is quite difficult for inexperienced participants to perform at the beginning of the experiment; however, dozens of trials improve performance. Thus, the kendama task is suitable for the investigation of short-term motor learning. The purpose of the present study is to examine changes in activation of the sensorimotor cortex on a trial-by-trial basis during learning of a multi-joint discrete movement task and to determine the relationship between the cortical activation changes and the kinematic or kinetic changes in the limb movements.

Methods

Participants

We tested 14 neurologically normal right-handed volunteers (4 females and 10 males, aged 21–39 years). Informed consent was obtained. All experiments were approved by the Ethical Committee of the Graduate School of Education, University of Tokyo.

Task

Participants were directed to hold the kendama with their right hand while sitting on a straight-backed chair. The kendama is made from a stick with a point at one end, a cup (4.2 cm in diameter), and a ball (6 cm in diameter) which is connected by a piece of string (47 cm). The individual masses of both the stick and the ball are 72.5 g. To perform the requested grip, participants needed to use all fingers to hold the stick with the point directed forward and the cup facing up. They performed a kendama task using only their right upper limb to toss the ball and catch it in the cup (Fig. 1). Participants were inexperienced regarding the task. No detailed instructions were dispensed regarding the initial position of the ball or the hand path during the task, but participants were instructed to use only the upper limb and limit head motion as much as possible. The kendama task requires complex hand–eye coordination and precise control of the redundant degree of freedoms inherent to the multi-joint system. Participants performed the task for two sessions (Fig. 2). Each session lasted approximately 15 min and comprised 45 trials. The total number of trials was 90. The interval between the sessions was 3 min 30 s. The participants performed a trial every 20 s according to “go” signals. Upon verbal “ready” signals, they were prepared to perform the task but were asked not to tense or move their right hands until the verbal “go” signal was delivered. Movements were performed as fast as possible after the “go” signals. The interval between the “ready” and “go” signals was 5 s. Regardless of outcome of the trials, the participants were requested to immediately return their hand to the start position (individually determined by each participant at every trial) and to hold the position while focusing on the cup until the next verbal “go” signals were delivered. After the participants performed a trial, an experimenter promptly set the ball under the participants’ right hands.
https://static-content.springer.com/image/art%3A10.1007%2Fs00221-008-1518-2/MediaObjects/221_2008_1518_Fig1_HTML.gif
Fig. 1

Schematic of the experimental configuration showing the locations of reflective markers (filled circles) and joint angles from a sagittal view. The markers are attached at the right shoulder, elbow, wrist, and hand joint. The ball is entirely covered with reflective tape to form a marker and is assumed to be connected to the hand segment through a string. The positive direction of each joint displacement is the flexion direction. Participants started the movements with the ball hanging under their right hands (the starting positions of the upper limb and the ball are depicted by dashed lines)

https://static-content.springer.com/image/art%3A10.1007%2Fs00221-008-1518-2/MediaObjects/221_2008_1518_Fig2_HTML.gif
Fig. 2

Schematic representations of the experimental procedure. Participants performed the task for two sessions. Each session comprised 45 trials and the total number of trials was 90. The interval between the sessions was 3 min 30 s. The inter-trial interval was 20 s. A rectangle represents a single trial. On verbal “ready” signals, participants were prepared to perform the task and executed the task as fast as possible after the delivery of “go” signals. The interval between “ready” signals and “go” signals was 5 s

NIRS recordings

A NIRS instrument (ETG-100, Hitachi Medical Corporation, Tokyo, Japan) was used to measure activation of the predicted locations of the sensorimotor cortices of both hemispheres. The instrument was placed behind participants. The optical fiber bundles were suspended from an aluminum rod attached to the NIRS instrument to avoid motion-related artifacts.

The NIRS instrument generated two wavelengths of NIR light (780 and 830 nm) and measured time courses of levels of oxygenated and deoxygenated hemoglobin (oxy-Hb and deoxy-Hb) at measurement channels with 0.1-s time resolution. The NIR light generated by 20 laser diodes (10 per wavelength) was modulated at different frequencies to prevent crosstalk between the channels and the wavelengths. Intensity of illumination at incident position was 1.5 mW for each wavelength. Light traveling from the instruments to the brain tissue and back to the instrument was guided by optical fiber bundles. Received light was detected by 8 avalanche photodiodes and separated into individual light sources with each wavelength by 48 lock-in amplifiers. We placed a pair of 3 × 3 arrays with five incident and four detection fibers, mounted on a flexible holder, over the predicted location of the sensorimotor cortices of each participant’s head. Each pair of the adjacent incident and detection fibers defined a single measurement channel, which allowed for the measurement of the oxy-Hb and deoxy-Hb changes at 24 channels (Fig. 3). The distance between the incident and detection fibers was set at 3 cm. The measurement regions in each hemisphere were correctly positioned by using each participant’s external auditory pores and vertex as landmarks. The midpoints between CH4 and CH5 and between CH15 and CH16 were set on the C3 and C4 positions of the international 10–20 system of electrode placement, respectively. CH2, CH7, and CH12 on the left hemisphere and CH13, CH18, and CH23 on the right hemisphere were aligned along the line between the external auditory pore and the vertex. A previous study demonstrated that the C3 and C4 positions of the international 10–20 system are projected over the cerebral cortical surface of the precentral or postcentral gyri in the Talairach stereotactic coordinates (Okamoto et al. 2004b). Thus, we measured the cortical activation around the predicted location of the sensorimotor cortex. Since the exact optical path-length of the light traveling through the brain tissue was not known, the unit of these values was molar concentration multiplied by length (mM mm).
https://static-content.springer.com/image/art%3A10.1007%2Fs00221-008-1518-2/MediaObjects/221_2008_1518_Fig3_HTML.gif
Fig. 3

Arrangement of measurement channels over left and right sensorimotor cortex. The measurement regions of both hemispheres were correctly positioned by using landmarks of the external auditory pores and the vertex in each participant. CH2, CH7, and CH12 on the left hemisphere, and CH13, CH18, and CH23 on the right hemisphere were aligned as the line between the external auditory pores and the vertex. The midpoint between CH4 and CH5 and the midpoint between CH15 and CH16 were set on the C3 and C4 positions of the international 10-20 system of electrode placement, respectively. Red points represent light irradiation positions and blue points represent light detection positions

Motion recordings

Upper limb movements and ball movements were recorded using a real time 3D motion capture system (Motion Analysis Co., Santa Rosa, California). Five CCD monochrome shuttered cameras (60 Hz, Falcon) with electronically shuttered infrared and red LED synchronized strobe lights were placed around the participant. The video images captured by the cameras were digitized by video processors and the 3D positions of markers were tracked in real time (EvaRT 4.0, Motion Analysis Co.). Spherical reflective markers with a diameter of 2.0 cm and a weight of approximately 5 g were attached to the glenohumeral (shoulder), elbow, wrist, and thumb metacarpophalangeal (hand) joints. The ball, entirely covered with reflective tape, was also a marker (Fig. 1). We measured the displacement positions of the shoulder, elbow, wrist, and hand joints and of the ball. In addition to recordings by the motion capture system, a video camera was used to record the movements.

Data analysis

Behavioral data

To evaluate behavioral performance changes during learning, we calculated kendama task success rates. A success trial occurred when the participant caught the ball in the cup and held it for 2 s. If this criterion was not met, the trial was defined as a failure. We divided all trials into six practice blocks (each block comprising 15 trials) and examined the changes in the mean success rates across the practice blocks. We performed one-way repeated-measures ANOVAs to detect the significant differences across the practice blocks. Additionally, we conducted post hoc Tukey tests to reveal the significant differences between the practice blocks. The statistical threshold was set at < 0.05.

NIRS data

We focused on the time evolution of the magnitudes of event-related hemodynamic responses to each trial during learning. In each individual set of the oxy-Hb and deoxy-Hb data, we initially extracted the single trial data ranging from the “go” signal onset to 12 s after the signal from the raw time course data. Drifts of the baseline signals were subsequently corrected using linear fitting in each single trial data.

Based on analysis of sharp changes in the time courses of total-Hb data (sum of oxy-Hb and deoxy-Hb), measurement channels showing a low signal-to-noise ratio or single trial data including motion artifacts caused by abrupt head movements were removed. Participant measurement channels that contained a minimum of 46 good single trial data were used for analysis. The number of the measurement channels used for analysis differed among participants. The number of the trials used for analysis also differed among measurement channels of each participant. The average number of measurement channels used for analysis among participants was 20. The average number of trials used for analysis among measurement channels among participants was 66. Because the highest row of the measurement channels (CH1, CH2, CH13, and CH14) showed low signal-to-noise ratios, many trials in these measurement channels were excluded from analysis, resulting in a decrease in the average number of trials used for analysis.

We obtained the event-related hemodynamic responses at each measurement channel by averaging the signal changes among trials. To elucidate the channels that revealed task-related responses, we analyzed the participant-averaged event-related hemodynamic responses of oxy-Hb and deoxy-Hb for the first 30 trials on a channel-by-channel basis. In addition to the abovementioned criterion for exclusion of measurement channels, measurement channels for each participant that contained less than ten good single trial data were also excluded from this analysis. We detected a time-point for the peak value in the time courses of the event-related hemodynamic responses of oxy-Hb for the first 30 trials, averaged among all measurement channels among all participants. We used the same procedure to detect a time-point for the low value in the time courses of the event-related hemodynamic responses of deoxy-Hb. The oxy-Hb peak time and deoxy-Hb low time responses were 5.0 and 6.6 s after the go signal onset, respectively. We subsequently defined the 3 s-period centered on the peak time and the low time as the time windows for analysis of the oxy-Hb and deoxy-Hb responses to each trial, respectively. We calculated the average values within the time windows as the magnitudes of the event-related hemodynamic responses of oxy-Hb and deoxy-Hb at each measurement channel. To identify the significant activation regions, we considered the individual data as random effects and performed student t tests against baseline for the magnitudes of the event-related hemodynamic responses of oxy-Hb and deoxy-Hb at each measurement channel. Although the statistical threshold was < 0.0021 after considering Bonferroni’s correction for multiple comparisons among 24 channels, we employed a more conservative threshold of < 0.001.

It should be noted that the Hb signals reflect the products of the molar concentration changes and the optical path lengths. The optical path lengths may vary between brain regions. Since we performed the group analysis on a channel-by-channel basis, the statistical results were not affected by the region-dependent differences of the optical path lengths. Although the optical path lengths may vary between participants, consistent changes in the Hb signals at a specific measurement channel among participants should indicate that the changes in the Hb signals are good approximations of those in the Hb concentrations that are induced by the brain activation.

Next, we examined the time evolution of the magnitudes of the event-related hemodynamic responses to each trial at the measurement channels that showed the significant task-related cortical activation. To perform group analysis, we divided all trials into six practice blocks (each block including 15 trials) and examined the changes in the mean magnitudes of the event-related hemodynamic responses across the practice blocks. We performed one-way repeated-measures ANOVAs to detect significant differences across the practice blocks. Additionally, we conducted post hoc Tukey tests to reveal significant differences between the practice blocks. The statistical threshold was set at < 0.05. Additionally, to compare the changes in the mean magnitudes of the event-related hemodynamic responses between the successful and unsuccessful trials, we performed 2 × 6 (trial types: successful and unsuccessful trials × practice blocks: from block 1 to block 6) ANOVAs. Both factors were within-subjects. Ryan’s test was used as the post hoc test. The statistical threshold was set at < 0.05.

Motion data

We assumed that the upper limb was three interconnected rigid links (the upper arm, forearm, and hand) with frictionless joints at the shoulder, elbow, and wrist as shown in Fig. 1. Additionally, we assumed that the kendama stick mass was included in the hand segment mass, and that the center of mass of the stick corresponded to the center of mass of the hand segment. The kendama ball was assumed to be connected to the hand segment via the string. We considered the projections of markers to the sagittal plane. Angular displacements at each joint were obtained in this plane. The definition of the joint angles is shown in Fig. 1. The direction of the arrow in the stick picture represents the positive direction of the angular displacement and corresponds to flexion direction. All linear and angular displacement data were smoothed with the use of the second order Butterworth low-pass filter (cut-off frequency = 6 Hz). The first and second derivatives were calculated to obtain velocity and acceleration. We measured the upper arm, forearm, and hand lengths of each participant from the captured data and estimated the mass, the center of mass, and the moment of inertia of each segment based on the segment length, the body segment parameters for Japanese athletes data (Ae et al. 1992), and the total body weight of each participant.

Note that the kendama task consists of two movements, which are sequentially produced: picking up the ball and catching it. Participants perform the task by coordinating shoulder, elbow, and wrist flexions and extensions. Figure 4 shows the time courses of ball displacement in the z axial direction (vertical direction), angular velocity at each joint, and the muscle torque for a single trial of a representative participant. The time course of the angular velocity clearly demonstrates that the participant initially lowered his or her right upper limb, subsequently swung upward to pick up the ball, and pulled back to catch it. Therefore, to define the two movements, a transition point between the picking up and catching movement was determined: this point was when ball acceleration along the z-axis was maximum. In this study, the picking up, catching, and total movements (sum of the two movements) in the kendama task were analyzed. The durations of the picking up and catching movements are depicted by solid lines and dotted lines in Fig. 4, respectively. The duration of the total movement was the sum of the durations of the two-abovementioned movements. The time course of the muscle torque indicates that the peak of the flexion torques of the shoulder and elbow that are essential for picking up the ball are included in the picking up movement. Thus, for analysis, the time course of individual motion data at each trial was divided into those of picking up and catching movements. Trials that lacked data due to markers going beyond a measurable space and, therefore, not being captured by cameras were excluded from analysis. Approximately 2% of trials were excluded from the motion data analysis.
https://static-content.springer.com/image/art%3A10.1007%2Fs00221-008-1518-2/MediaObjects/221_2008_1518_Fig4_HTML.gif
Fig. 4

Time courses of ball displacement in z axial direction (vertical direction), angular velocity, and muscle torque at the shoulder (red lines), elbow (blue lines) and wrist (black lines) joints for a representative participant. The kendama task was analyzed for three movements: picking up, catching, and total movements. Solid lines and dotted lines represent the picking up and catching movements, respectively. The total movement is the sum of the picking up and catching movements. For angular velocity, positive/negative values indicate flexion/extension velocities. For muscle torque, positive/negative values indicate extension/flexion torques

Dispersion of movement patterns

For the kinematic analysis of limb movements, we first quantified the dispersion of the movement patterns for each joint as follows. We determined the time course of the angular displacement at each joint. Next, the time courses were normalized for each movement (picking up, catching, and total movements) and differentiated to obtain normalized time courses of the angular velocity. After we divided all trials into six practice blocks (each block comprising 15 trials), we calculated the within-block standard deviations of the normalized time courses for each angular velocity at each time point and added them over all the time points of each movement. Thus, we obtained an index of the dispersion of the movement patterns at the shoulder, elbow, and wrist joints (hereafter referred to as shoulder, elbow, and wrist dispersions).

Second, we quantified the dispersion of the multi-joint movements for the upper limb. We used a phase space composed of dimensions of the shoulder, elbow, and wrist angular velocity to determine the time courses of the phase trajectories for every trial. After we divided all trials into six practice blocks (each block comprising 15 trials), we calculated the within-block standard deviations of the phase trajectories at each time point and added them over all the time points of each movement. Thus, we obtained an index of the dispersion of the multi-joint movement patterns for the upper limb (hereafter referred to as upper limb dispersion).

Third, to evaluate the dispersion of the ball trajectories, we calculated the within-block standard deviations of the normalized time courses of the ball velocity at each time point and added them over all the time points of each movement. Thus, we obtained an index of the dispersion of the ball trajectories (hereafter referred to as ball dispersion).

We performed one-way repeated-measures ANOVAs to detect significant differences in the mean dispersion across the practice blocks for each movement. Additionally, we conducted post hoc Tukey tests to examine significant differences between the practice blocks. The statistical threshold was set at < 0.05. Furthermore, to compare the changes in the dispersion between the successful and unsuccessful trials, we performed 2 × 6 (trial types: successful and unsuccessful trials × 6 practice blocks: from block 1 to block 6) ANOVAs for each movement. Both factors were within-subjects. Ryan’s test was used as the post hoc test. The statistical threshold was set at < 0.05.

Integrated muscle torque

To analyze limb movements kinetically, we calculated the integrated muscle torques at the shoulder, elbow, and wrist joints and for the entire upper limb. First, we obtained the time course of the muscle torque at each joint. To calculate the muscle torques, we used the Newton–Euler inverse dynamic equation of a three-joint arm model to which a ball was connected by a string (Fig. 1). It is noted that the muscle torque includes not only the mechanical contribution of muscle contraction at the joint but also passive contributions from the muscles, tendons, articular capsules, and other connective tissues. Because the effects of active muscular force are embedded within this component, the muscle torque comprises the actively controlled elements of limb-trajectory motor production (Schneider et al. 1989). To cancel the muscle torque required for maintaining the static posture during the holding of the kendama stick, we subtracted the value of the torque at the starting point from the entire time course of each muscle torque (Fig. 4). Second, we expressed the time courses of each muscle torque as absolute values, regardless of flexion or extension torque, to evaluate the muscle torques requiring the execution of each trial. Third, we obtained the time course of the entire upper limb muscle torque by adding the time courses of the shoulder, elbow, and wrist muscle torques. Finally, we integrated the values of the time course over each movement (picking up, catching, and total movements) and obtained four integrated muscle torques for the shoulder, elbow, and wrist joints and the upper limb.

We divided all trials into six practice blocks (each block including 15 trials) and examined the changes of the mean integrated muscle torques across the practice blocks for each movement. We performed one-way repeated-measures ANOVAs to detect significant differences across the practice blocks. Additionally, we conducted post hoc Tukey tests to reveal significant differences between the practice blocks. The statistical threshold was set at < 0.05. Additionally, to compare the changes in the integrated muscle torques between the successful and unsuccessful trials, we performed 2 × 6 (trial types: successful and unsuccessful trials × practice blocks: from block 1 to block 6) ANOVAs for each movement. Both factors were within-subjects. Ryan’s test was used as the post hoc test. The statistical threshold was set at < 0.05.

Results

Success rate

The success rates increased from 41.0 to 60.5% with practice (Fig. 5). A group analysis (one-way repeated-measures ANOVAs) revealed that the success rates demonstrated the main effect for practice blocks (F5,65 = 4.17, < 0.005). Post hoc Tukey tests revealed significant increases in the success rates from block 1 to blocks 3, 4, 5, and 6.
https://static-content.springer.com/image/art%3A10.1007%2Fs00221-008-1518-2/MediaObjects/221_2008_1518_Fig5_HTML.gif
Fig. 5

Changes of success rates across practice blocks. The learning process is shown as the increase in success rate. Vertical bars represent the standard errors among participants

NIRS data

Figure 6a shows a portion of the continuous time courses of oxy-Hb and deoxy-Hb at CH4 for a representative participant, demonstrating that the oxy-Hb signal increased after the “go” signals while the deoxy-Hb signal decreased to a lesser extent. Figure 6b shows the trial-by-trial changes of the magnitudes of the oxy-Hb responses during learning at the same measurement channel for the same participant as shown in Fig. 6a, demonstrating that the cortical activation for each trial gradually decreased.
https://static-content.springer.com/image/art%3A10.1007%2Fs00221-008-1518-2/MediaObjects/221_2008_1518_Fig6_HTML.gif
Fig. 6

a Continuous hemodynamic changes of oxy-Hb (red line) and deoxy-Hb (blue line) at CH4 over the left sensorimotor cortex for a representative participant. A gray dashed line represents the time point of a “go” signal. The data includes nine trials (from trial 37 to trial 45 for the participant). b Trial-by-trial changes of the magnitudes of oxy-Hb changes during learning for the same representative participant. Each circle represents the magnitude in each trial. The least square regression line (red line) shows the decrease in the cortical activation changes. A red rectangle and black rectangle represent a successful trial and an unsuccessful trial, respectively

Figure 7 shows the group-averaged event-related hemodynamic responses of oxy-Hb and deoxy-Hb for the first 30 trials at all measurement channels over the sensorimotor cortices of the both hemispheres. From the statistical analysis for the task-related cortical activation, we found the significant oxy-Hb increases that surpass P < 0.001 (student t test, n = 14) in four measurement channels: CH4 (t = 4.26), CH9 (t = 5.93), CH10 (t = 4.26), and CH12 (t = 5.66). The four measurement channels were located in the left sensorimotor cortex, which was contralateral to the hand performing the movements (Fig. 8). The same four measurement channels showed significant increases in a right hand tapping task during a pilot study conducted before the present study. Conversely, we did not find a significant change in deoxy-Hb in any measurement channel.
https://static-content.springer.com/image/art%3A10.1007%2Fs00221-008-1518-2/MediaObjects/221_2008_1518_Fig7_HTML.gif
Fig. 7

Task-related cortical activation of the kendama task. The group-averaged time courses of the event-related hemodynamic responses of oxy-Hb (red lines) and deoxy-Hb (blue lines) for the first 30 trials in 24 measurement channels over the sensorimotor cortices of both hemispheres are represented. Zero second of each panel is set at the “go” signal onset. The unit of oxy-Hb is a relative change from an arbitrary zero baseline at the measurement period. Note that the number of participants used for analysis differed among the measurement channels

https://static-content.springer.com/image/art%3A10.1007%2Fs00221-008-1518-2/MediaObjects/221_2008_1518_Fig8_HTML.gif
Fig. 8

Statistical map of event-related cortical activation for the first 30 trials. The filled circles indicate channels that showed significant activation (< 0.001)

We subsequently performed a group analysis of the time evolution of the magnitudes of the event-related oxy-Hb responses to each trial at the four measurement channels over the left sensorimotor cortex. A group analysis (one-way repeated-measures ANOVAs) revealed the main effects for practice blocks at 3 out of the 4 measurement channels: CH4 (F5,65 = 3.03, < 0.025), CH9 (F5,65 = 4.04, < 0.005), CH10 (F5,65 = 1.41, > 0.05), and CH12 (F5,65 = 7.79, < 0.001) (Fig. 9). Post hoc Tukey tests showed significant decreases for CH4 from block 1 to block 4; for CH9 from block 2 to blocks 5 and 6; and for CH12 from block 1 to blocks 4, 5, and 6, from block 2 to blocks 4, 5, and 6, and from block 3 to blocks 4, 5, and 6. Furthermore, the 2 × 6 (trial types: successful trials and unsuccessful trials × practice blocks: from block 1 to block 6) ANOVAs revealed the main effects for the practice blocks at 3 out of the 4 measurement channels: CH4 (F5,65 = 3.74, < 0.005), CH9 (F5,65 = 6.72, < 0.001), CH10 (F5,65 = 1.75, > 0.05), and CH12 (F5,65 = 6.56, < 0.001), but not for the trial types (Fig. 9). There were no statistically significant interactions between the trial types and practice blocks. Ryan’s multiple comparisons for the practice blocks showed that the magnitudes of the event-related oxy-Hb responses to each trial decreased with learning, irrespective of the outcome of the trial. The average number of trials for the analysis among these four measurement channels among participants was 73 (>81% of the total trials).
https://static-content.springer.com/image/art%3A10.1007%2Fs00221-008-1518-2/MediaObjects/221_2008_1518_Fig9_HTML.gif
Fig. 9

Changes of the magnitudes of oxy-Hb responses across practice blocks at the measurement channels that showed significant task-related activation. Black lines represent the magnitudes of the oxy-Hb responses in the total trials (top row). The red and blue lines represent the magnitudes of the oxy-Hb responses in the successful and unsuccessful trials, respectively (bottom row). Vertical bars represent the standard errors of the means among participants

Motion data

Dispersion of movement patterns

To quantify the kinematic aspects of the movement changes in the upper limb during learning for each type of movement (picking up, catching, and total movements), we examined the changes in the dispersions of each type of movements for the shoulder, elbow, and hand joint and the dispersion of the multi-joint movements for the upper limb (upper limb dispersion). A group analysis (one-way repeated-measures ANOVAs) was performed to detect significant differences in the dispersions across practice blocks. For the picking up and total movements, none of the dispersion values showed significant changes (Fig. 10a–d). For the catching movements, the main effects for the practice blocks were shown for the shoulder (F5,65 = 8.22, < 0.001), elbow (F5,65 = 6.00, < 0.001), and the upper limb (F5,65 = 4.30, < 0.001) dispersions, but not for the wrist dispersion. Post hoc Tukey tests showed significant decreases in the dispersion for the shoulder joint from block 1 to blocks 2, 3, 4, 5, and 6; in that for the elbow joint, from block 1 to blocks 2, 3, 5, and 6; and in that for the upper limb, from block 1 to blocks 2, 3, 5, and 6.
https://static-content.springer.com/image/art%3A10.1007%2Fs00221-008-1518-2/MediaObjects/221_2008_1518_Fig10_HTML.gif
Fig. 10

Changes in the dispersions of the movement patterns across practice blocks for the a shoulder joint, b elbow joint, c wrist joint, d upper limb, and e the ball during the picking up (top row), catching (middle row), and total movements (bottom row). Dispersion values are given in arbitrary units. Vertical bars represent the standard errors of the means among participants. Black lines represent changes in dispersion of total trials (left column). Red and blue lines represent changes in dispersion in the successful and unsuccessful trials, respectively (right column)

In addition, we examined the changes in the dispersion of the ball trajectories during learning. For the picking up and total movements, the ball dispersions did not show significant changes (Fig. 10e). For the catching movements, the main effect for the practice blocks was shown (F5,65 = 6.84, < 0.001). Post hoc Tukey tests showed significant decreases in the ball dispersion from block 1 to blocks 2, 3, 5, 6. These results indicated that the dispersions of the upper limb’s motions and ball trajectories during the catching movements decreased with practice.

To compare the changes in the dispersions between the successful and unsuccessful trials, we performed 2 × 6 (trial types: successful and unsuccessful trials × practice blocks: from block 1 to block 6) ANOVAs for each movement. Because, the above results indicated that the dispersions of the catching movements changed significantly during learning, we focused on the results of dispersion for the catching movements. The shoulder dispersions showed significant main effects for the practice blocks (F5,65 = 6.77, < 0.001), but not for the trial types (Fig. 10a). There were no significant interactions between the trial types and practice blocks. Ryan’s multiple comparisons for the practice blocks revealed significant decreases from block 1 to blocks 2, 3, 4, 5, and 6. The elbow dispersions showed significant main effects for the trial types (F1,13 = 6.96, < 0.05) and practice blocks (F5,65 = 4.69, < 0.005) (Fig. 10b). Although there were no significant interactions between the trial types and practice blocks (F5,65 = 2.33, = 0.0526), the elbow dispersions in the unsuccessful trials, but not those in the successful trials, appeared to decrease with practice. The wrist dispersions did not show significant main effects for the trial types or practice blocks, but showed significant interaction effects between the trial types and practice blocks (F5,65 = 2.92, < 0.05) (Fig. 10c). With regard to the interaction, there were both significant simple main effects of the trial types on the practice blocks and of the practice blocks on the trial types. Ryan’s multiple comparisons for the trial types in block 1 revealed that the dispersions in the unsuccessful trials were significantly greater than those in the successful trials. Ryan’s multiple comparisons for the practice blocks in the unsuccessful trials revealed significant decreases from block 1 to block 5. The upper limb dispersions showed significant main effects for the trial types (F1,13 = 7.27, < 0.05) and practice blocks (F 5,65 = 3.25, < 0.05) (Fig. 10d). In addition, the upper limb dispersions showed significant interactions between the trial types and practice blocks (F 5,65 = 2.84, < 0.05). With regard to the interaction, there were both significant simple main effects of the trial types on the practice blocks and of the practice blocks on the trial types. Ryan’s multiple comparisons for the trial types in block 1 revealed that the dispersions in the unsuccessful trials were significantly greater than those in the successful trials. Ryan’s multiple comparisons for the practice blocks in the unsuccessful trials revealed significant decreases from block 1 to blocks 2, 3, 4, 5, and 6. The ball dispersions showed significant main effects for the trial types (F1,13 = 17.08, < 0.005) and practice blocks (F5,65 = 5.04, < 0.001) (Fig. 10e). In addition, there were significant interactions between the trial types and practice blocks (F5,65 = 6.23, < 0.001). With regard to the interaction, there were both significant simple main effects of the trial types on the practice blocks and of the practice blocks on the trial types. Ryan’s multiple comparisons for the trial types in blocks 1 and 2 revealed that the dispersions in the unsuccessful trials were significantly greater than those in the successful trials. Ryan’s multiple comparisons for the practice blocks in the unsuccessful trials revealed significant decreases from block 1 to blocks 2, 3, 4, 5, and 6.

The results for the catching movements are summarized as follows. The shoulder dispersions showed significant decreases, irrespective of the outcome of the trials. On the other hand, the elbow and wrist dispersions in the unsuccessful trials decreased with practice, while those in the successful trials did not change. The upper limb dispersions in the unsuccessful trials decreased with learning, similar to the elbow and wrist dispersions, while those in the successful trials did not change. The ball dispersions in the unsuccessful trials also decreased with practice, while those in the successful trials did not change. These results of ball dispersion were very similar to those for the upper limb dispersion, as shown in Fig. 10.

Integrated muscle torque

For the kinetic analysis of limb movements during learning, we examined the changes in the integrated muscle torques of the shoulder, elbow, and hand joints and the sum of these muscle torques (upper limb integrated muscle torque) for each type of movement (picking up, catching, and total movements). A group analysis (one-way repeated-measures ANOVAs) was performed to detect significant differences in the integrated muscle torques across practice blocks. For the picking up movements, the main effects for the practice blocks were shown for the shoulder (F5,65 = 3.49, < 0.01), elbow (F5,65 = 2.71, < 0.05), and upper limb (F5,65 = 3.63, < 0.01) integrated muscle torques, but not for the wrist integrated muscle torque (Fig. 11a–d). Post hoc Tukey tests showed significant decreases in the integrated muscle torque for the shoulder joint from block 1 to blocks 4 and 6; in that for the elbow joint, from block 1 to block 4 and from block 2 to block 4; and in that for the upper limb, from block 1 to blocks 4 and 6 and from block 2 to block 4. With regard to the catching movements, none of the muscle torques showed significant changes. For the total movements, the shoulder integrated muscle torques showed the main effect for the practice blocks (F5,65 = 2.42, < 0.05). Post hoc Tukey tests showed that the shoulder integrated muscle torques tended to decrease as learning progressed, although this decrease was not statistically significant. These results indicate that the integrated muscle torques of the upper limb required for picking up a ball significantly decrease as learning progresses. In particular, the decrease in the upper limb integrated muscle torques could be primarily attributed to the decrease in the shoulder integrated muscle torque, which was notable.
https://static-content.springer.com/image/art%3A10.1007%2Fs00221-008-1518-2/MediaObjects/221_2008_1518_Fig11_HTML.gif
Fig. 11

Changes in the a shoulder, b elbow, c wrist, and d upper limb integrated muscle torques for each block during the picking up (top row), catching (middle row), and total movements (bottom row). Black lines represent changes in muscle torques of total trials (left column). Red and blue lines represent changes in muscle torques in successful and unsuccessful trials, respectively (right column). Vertical bars represent the standard errors of the means among participants

To compare the changes in the integrated muscle torques between the successful and unsuccessful trials, we performed a 2 × 6 (trial types: successful and unsuccessful trials × practice blocks: from block 1 to block 6) ANOVAs for each movement. For the picking up movements, the main effects for the practice blocks were shown for the shoulder (F5,65 = 3.75, < 0.005), elbow (F5,65 = 3.05, < 0.05), and upper limb (F5,65 = 5.45, < 0.001) integrated muscle torques, but not for the trial types (Fig. 11a–d). There were no statistically significant interactions between the trial types and practice blocks. Ryan’s multiple comparisons for the practice blocks revealed significant decreases in the integrated muscle torques for the shoulder joint from block 1 to blocks 4 and 6, and for the upper limb from block 1 to blocks 4 and 6; the elbow integrated muscle torque did not decrease. For the total movements, the shoulder integrated muscle torque showed the main effects for the practice blocks (F5,65 = 2.53, < 0.05). Ryan’s multiple comparisons for the practice blocks revealed that the shoulder integrated muscle torques showed a tendency to decrease as learning progressed, although this decrease was not statistically significant. These results indicate that the shoulder, elbow, and upper limb integrated muscle torques that are required for picking up a ball decrease with learning, irrespective of the outcome of the trials.

Relationship between NIRS and motion data

Analysis of the NIRS data revealed that the event-related oxy-Hb responses to each trial showed significant decreases in cortical activation over the sensorimotor cortex as learning progressed. On the other hand, analysis of the upper limb’s motion data revealed that the integrated muscle torques required for picking up a ball and the dispersions in the movement patterns of catching a ball decreased as learning progressed. To examine the relationships between the decreases in the oxy-Hb responses and those in the dispersion or integrated muscle torques during learning, we calculated the Pearson’s product–moment correlation coefficients between the changes in the magnitudes of the oxy-Hb responses and the changes in the motion parameters (mean values averaged in 1 and 2 blocks minus mean values averaged in 5 and 6 blocks). The magnitudes of the oxy-Hb responses at the significantly activated measurement channels in the first two blocks that showed significant decreases (CH4, CH9, and CH12) were tested with the dispersions and integrated muscle torques that showed significant decreases as a result of the one-way repeated-measures ANOVAs (refer to the Sect. “Results”). We found significant positive correlations between the decrease in the oxy-Hb responses at CH4 and the decrease in the integrated muscle torque during the picking up movements for the shoulder joint (r = 0.579, P < 0.05, n = 14) and for the upper limb (r = 0.534, P < 0.05, n = 14) (Fig. 12), and during the total movements for the shoulder joint (r = 0.627, P < 0.05, n = 14). We also found significant positive correlations between the decrease in the oxy-Hb responses at CH12 and the decrease in the integrated muscle torques during the picking up movements for the shoulder joint (r = 0.566, P < 0.05, n = 14) and for the upper limb (r = 0.545, P < 0.05, n = 14), and during the total movement for the shoulder joint (r = 0.582, P < 0.05, n = 14). There was no significant correlation between the changes in the oxy-Hb responses and those in the dispersions.
https://static-content.springer.com/image/art%3A10.1007%2Fs00221-008-1518-2/MediaObjects/221_2008_1518_Fig12_HTML.gif
Fig. 12

Correlation between the oxy-Hb differences in CH4 and the upper limb integrated muscle torque differences during the picking up movements. The black line denotes the least square regression line (< 0.05). Each circle represents individual data

Discussion

Simultaneous measurement of event-related cortical activation to each trial by NIRS and upper limb movements by the 3D motion capture system during learning of a multi-joint discrete motor task

The present study demonstrated that NIRS was successfully applied to detect the event-related oxy-Hb responses of each kendama task trial over the left sensorimotor cortex, which was contralateral to the arm used in the task. Furthermore, the study revealed that the magnitudes of the cortical responses significantly decreased as learning progressed. In most of the previous PET or fMRI studies that have used block designs with long periods of repetitions of motor tasks and rest, accumulated effects of the brain activation associated with multiple trials have been measured. In contrast, the present study specified the amount of the brain activation in relation to a single event of performing a discrete movement. Thus, this approach allowed us to study one-to-one correspondence between movement changes and cortical activation changes in the course of motor skill acquisition. In fact, analysis of the upper limb’s movements demonstrated that the integrated muscle torque required for picking up a ball and the dispersion of movement patterns of catching a ball significantly decreased as learning progressed. Furthermore, we found significant positive correlations between the changes in the magnitudes of the oxy-Hb responses to each trial and those in the integrated muscle torques of the upper limb during learning, while the changes in the magnitudes of the oxy-Hb responses and those in the dispersion did not show significant correlations. Using a successfully combined NIRS and 3D motion capture system, we characterized the time evolution of both of the cortical activation and the upper limb movement during the course of motor learning.

Movement changes during learning

The kendama task is a goal-directed movement that is executed within several hundred ms. The task requires accurate control of the trajectory of the ball, which is determined by the initial movement of the upper limb when picking up the ball with the string. Since this action is ballistic and rapid, the movement should be largely executed by a mechanism of feedforward control. On the other hand, catching the ball with the cup of the kendama also requires accurate eye-hand coordination. Since this action varies with the proceeded trajectories of the ball and hand, the movement should be mainly executed by a mechanism of feedback control.

The analysis of the upper limb’s motion data revealed two features of the significant changes that occur over the course of learning. One is that the upper limb integrated muscle torque required for picking up a ball shows a significant decrease across practice blocks, irrespective of the outcome of the trials. This result implies that the participants learned that a motor action with less muscle torque production allowed them to perform the task efficiently and may mainly reflect the changes in the mechanism of feedforward control. The other is that the dispersion in the upper limb’s movement patterns of catching a ball showed a significant decrease. In particular, the upper limb dispersions in the unsuccessful trials decreased significantly across practice blocks, whereas those in the successful trials did not change. These results were very similar to those for the dispersion of ball trajectories. Therefore, the decreases in the upper limb dispersions would be associated with those in the ball dispersions that were observed in the unsuccessful trials during learning.

The abovementioned motion analysis revealed the nature of the learning process of a multi-joint movement task. The shoulder joint, which is the most proximal joint, showed a notable decrease in the integrated muscle torques involved in picking up a ball, irrespective of the outcome of the trials. On the other hand, the elbow and wrist joints, which are more distal-joints, showed significantly large variability in terms of the kinematic patterns of the catching movement induced by adjusting to the divergent ball trajectories in the unsuccessful trials in the early leaning periods. Since proximal segments have higher inertia and more massive musculature than distal segments, the mechanical influence of the proximal segment motion on the distal segments is much higher than the influence of the distal segment motion on the proximal segments. Recent studies suggest that the upper limb individual joints play different roles in controlling a multi-joint movement (Dounskaia 2005; Galloway and Koshland 2002; Galloway et al. 2004; Hirashima et al. 2007). Dounskaia (2005) further proposed that the proximal joint plays a role in production of gross characteristics of the entire limb movement, whereas the distal joint plays a role in regulation of the mechanical influence of the proximal segment motion on the distal segments and attainment of the motion of the end effector required by the task. The notable decrease in the integrated shoulder muscle torque during the learning of the kendama task leads us to infer that the CNS may adopt a strategy that reduces the integrated muscle torque of the proximal joint in order to achieve more efficient movement of the entire limb. The notably large variation in the elbow and wrist joint movements in the unsuccessful trials during the early learning of the kendama task leads us to infer that the CNS may adopt a strategy of regulating the movements of the more distal joints in order to adjust to the large variability in ball trajectories.

Cortical activation changes during learning

The magnitudes of the oxy-Hb responses in the left sensorimotor cortex significantly decreased with learning, irrespective of the outcome of the trials. Furthermore, the decrease in the oxy-Hb responses correlated with the decrease in integrated muscle torques of the upper limb for the initial movement required for picking up the ball. These results imply that the decrease in the cortical activation of the left sensorimotor cortex would be primarily attributed to the changes in motor commands that are involved in generating muscle torques to perform the task efficiently. A number of human imaging studies have reported that activation changes in the sensorimotor cortex are associated with performance changes in relation to movement parameters such as movement distance (Shadmehr and Holcomb 1997), movement rate (Seitz et al. 1990), or force production (Dettmers et al. 1995). In addition to this association, our results demonstrated the correlated changes of the brain activation with those of the integrated muscle torque during motor learning.

The decrease in activation of the sensorimotor cortex may be accounted for by changes in sensory input associated with changes in motor output. A previous study using fMRI reported that learning-related activation decreases were observed in the primary motor cortex while performing a visuomotor coordination task (Floyer-Lea and Matthews 2004). Floyer-Lea and Matthews (2004) interpreted that this decrease may be due to increasingly specific afferent input to the primary motor cortex as movement patterns become better defined. In the present study, analysis of the dispersion of multi-joint movements of the upper limb revealed that the upper limb dispersions in the unsuccessful trials for catching movements significantly decreased, while those in the successful trials did not change with learning. We, however, found no significant relationship between the changes in the activation of sensorimotor cortex and the dispersions in the upper limb’s movements during learning. Accordingly, our results were not consistent with the finding that the decrease in the primary motor cortex may be attributed to increasing specific afferent input as the movement patterns become more defined.

Previous studies have proposed that more efficient use of specific neural circuits for an identical task over trials is accompanied by learning-related activation decreases (Petersen et al. 1998; Poldrack 2000; Kelly and Garavan 2005). The efficient use of specific neural circuits has a close relationship with the neural models of “repetitive suppression (RS)” (see review: Grill-Spector et al. 2006). RS has been used to refer to reduction of neural responses following stimulus repetition and is reported to occur in neural firing measured with single-cell recording, local field potentials measured with electroencephalogram/magnetoencephalogram (EEG/EMG), and in hemodynamic changes measured with fMRI. Although the present task did not require simple repetition of identical movements and involved dynamic changes in movements, the activation decrease in the sensorimotor cortex might be partially accounted for by some mechanisms related to RS.

The present study focused on the time evolution of the magnitudes of the cortical activation on a trial-by-trial basis over the sensorimotor cortex during learning of a discrete movement. However, it remains unclear how other regions of the brain are involved in the acquiring of discrete movements. An fMRI study by Schaal et al. (2004) showed that more widely distributed brain regions were involved in the production of discrete movements of a single joint as compared with the activated regions in the production of rhythmic movements. This result suggests that the generation of discrete movement requires cognitive processes such as higher-level planning and execution of movements. A recent fMRI study by Grafton et al. (2008) showed trial-by-trial changes in activation of widely distributed brain regions in relation to improvement of feedforward or feedback control of a visuomotor tracking task with pronation-supination rotation of the forearm. This indicated that separate regions of the brain are involved in the feedforward and feedback control of the compensatory visuomotor tracking. Future NIRS studies are required to investigate activation changes over wider brain regions during learning of multi-joint and ballistic discrete movements, such as the kendama task. Furthermore, it will be interesting to assess the relationship between the changes in brain activation and the changes in the components involved in higher cognitive processing such as motor preparation, planning, and evaluation of the outcome (success/failure).

Conclusion

We simultaneously recorded cortical activation by NIRS and movements by the 3D motion capture system during learning of a multi-joint discrete motor task. We characterized the learning process as the decrease in the cortical activation in the sensorimotor cortex and demonstrated a significant positive correlation between the cortical activation and kinetic movement changes. Our results suggest that the decrease in the cortical activation in the sensorimotor cortex reflects the changes in motor commands for a multi-joint discrete motor task during the course of learning.

Acknowledgments

The authors would like to thank Kayo Asakawa, Hama Watanabe, and Fumitaka Homae for their assistance during the experiments and their helpful comments on this study.

Copyright information

© Springer-Verlag 2008