HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Abstract Full Text (PDF) All Versions of this Article: 91/3/1260    most recent 00168.2003v1 Alert me when this article is cited Alert me if a correction is posted Citation Map Services Email this article to a friend Similar articles in this journal Similar articles in ISI Web of Science Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via ISI Web of Science (4) Citing Articles via Google Scholar Google Scholar Articles by Ohki, Y. Articles by Watanabe, K. Search for Related Content PubMed PubMed Citation Articles by Ohki, Y. Articles by Watanabe, K.

Dependence of Reactive Responses in Human Bimanual Finger Movements on Sensory Feedback and Auditory Cues

¹Department of Physiology, Kyorin University School of Medicine, Shinkawa, Mitaka, Tokyo 181-8611; and ²Department of Sports Science, Faculty of Education, Shinshu University, Nagano 380-8544, Japan

Submitted 24 February 2003; accepted in final form 29 October 2003

	ABSTRACT

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION ACKNOWLEDGMENTS REFERENCES

 
We examined the effects of repetitive experience of feedback events and cue signals on adaptation of bimanual reactive response, both of which can be used to predict load forces on handheld objects. Normal human subjects used bimanual index fingers to hold two plates mounted on separate torque motors. Sensory-driven reactive forces from the fingers were measured during concurrent loading of both plates or isolated loading of the left plate. After repeated experiences of single load conditions, right and left fingers increased normal forces in a coordinated manner during concurrent loading, while the left finger responded almost exclusively during isolated loading of the left plate. However, after switching load conditions from concurrent to isolated load, or vice versa, several trials were needed to adapt to the new condition, as was clearly observed in the dynamic phase of responses from the right finger. Adaptation depends on the number of feedback experiences, and specifically on prediction error, showing intermediate predictions between the two conditions. Thus influences of the previous experiences show similar properties to those observed in self-generated movement, which were reported previously. In addition, auditory cues informing subjects of coming load conditions also partially adapted responses to these conditions. The neuronal center for reactive responses is therefore influenced by both previous experience and cue signals, inducing the appropriate responses for the predicted sensory inputs.

	INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION ACKNOWLEDGMENTS REFERENCES

 
Sensorimotor control of the hand during object manipulation is characterized by feedforward, predictive control policies based on internal models that reflect both the behaviors of our motor system and the relevant properties of external objects (Johansson 1996; Johansson and Cole 1994; Wing 1996; Wolpert and Flanagan 2001). When we lift a handheld object, for example, internal models predict the consequences of self-generated actions, in addition to friction between the object and individual fingertips, and modify grip forces accordingly (Burstedt et al. 1999; Edin et al. 1992; Johansson and Westling 1984). This predictive control is essential, given the inevitable neuromechanical delays limiting the utility of closed-loop feedback control (Hogan et al. 1987; Johansson and Cole 1994).

However, sensory-driven feedback control is also essential during manipulation, to correct error during movements. An extreme example occurs when we manipulate "active" objects that may threaten grasp stability because of load changes with unpredictable timing. In such cases, normal subjects produce sensory-driven reactive grip responses to restrain the grasped object (Johansson et al. 1992a,b; Macefield 1996). However, even in these cases, human subjects try to predict the behavior of the object to compensate for delayed feedback effects. The simplest expression of such predictions is increased background grip force when a rapid load increase is expected at some time in the near future (Cole and Johansson 1993; Johansson and Westling 1988; Johansson et al. 1992a,c; Winstein et al. 2000). In addition, the reactive grip responses themselves are rescaled using early cues about the rate of load-force changes (Cole and Abbs 1988; Johansson et al. 1992a; Winstein et al. 2000) and friction (Birznieks et al. 1998; Cole and Johansson 1993). Furthermore, reactive grip response can be induced by sensory cues from different digits on either the same or the contralateral hand (Ohki and Johansson 1999), and gain from interdigital response is scaled by subject predictions about simultaneity of load forces to handheld objects (Ohki et al. 2002). The neuronal center for reactive grip response, which processes the sensory inputs to induce reactive movements, should therefore be influenced by feedforward predictive control to allow effective feedback control. The internal model generating motor commands may therefore also affect the center.

In self-generated bimanual movements, the forward internal model is proposed to be adapted by changing the weighting given to possible motor control modules based on sensory errors (Witney et al. 2000; Wolpert and Flanagan 2001). Subjects can thus make intermediate predictions between possible internal models, and adaptation sometimes takes more than three feedback events, i.e., appropriate feedback inputs to both hands in association with three manipulations of an object. If selection of an internal model also influences sensory-driven reactive movements, then adaptation of the reactive movement should show similar properties to those observed in self-generated movements. However, adaptation of reactive movement currently seems to be more transient, being completed within the first trial after changing the properties of an object, although effects of global predictions have also been reported (Birznieks et al. 1998; Winstein et al. 2000).

The present study was designed to examine in detail the effect of previous experiences on current reactive movements, using the bimanual task developed by Ohki et al. (2002). In this task, subject predictions about load conditions to handheld objects can be controlled. We compare our results with those obtained for self-generated bimanual movements by Witney et al. (2000) and examine the hypothesis that the forward internal model could affect the neuronal center for sensory-driven reactive movements.

The present study also examined the effect of cues, which help subjects to predict load conditions on objects without sensory feedback. We hypothesized that humans can use such predictions free from feedback events, for example, when we manipulate familiar objects, in which we can predict the properties of the objects from visual information before even touching them.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION ACKNOWLEDGMENTS REFERENCES

 
The present study was conducted in accordance with the principles provided in the Declaration of Helsinki. With approval of the local ethical committee, experiments were performed on seven healthy right-handed human volunteers (3 females, 4 males) between 28 and 42 years old. Subjects were naïve to the research aims of the study and provided written informed consent. Subjects were seated in a chair with upper arms parallel to the trunk and forearms extended anteriorly in the horizontal plane. The forearms up to the palms were supported by a tabletop, and the hands were in intermediate pronation (palms down). Subjects used the tips of the left and right index fingers positioned side-by-side to restrain an instrumented apparatus (see APPARATUS; Fig. 1A; see also Fig. 1 in Ohki and Johansson 1999). Fingers were slightly flexed and the plane of the grasp plates approximately intersected the centers of the metacarpophalangeal joints. In such a posture, movements of the apparatus caused minimal passive normal force changes. Subjects were blindfolded during the experiments.

View larger version (37K): [in this window] [in a new window]   FIG. 1. Schematic illustration of the apparatus and constructions of test series. A: schematic illustration of the manipulandum, which comprised two horizontally oriented flat circular grasp plates, each connected to a separate servoregulated torque motor via a stiff beam. Normal and tangential load forces on both grasp plates were measured. B: load force change during 2 consecutive trials. Force change during the 2nd trial is partially shown. Note that time before load increase differed between short- and long-interval test series. When a beep was given, it occurred at the start of a trial (see onset of second trial). C: schematic illustrations for constructions of the link-predictive and unlink-predictive series. Both test series comprised linked (L) and unlinked (U) trials, in which concurrent load to both plates or isolated load to the left plate was delivered, respectively. The range of sequential unlinked trials in a test series is indicated by n, whereas the corresponding number of sequential linked trials in the link-predictive series is represented by m. In the unlink-predictive series, beeps notified subjects of the start of sequential unlinked trials. Similarly, in the link-predictive series, subjects were cued on starts of the linked trials by beeps, although the cue was followed by unlinked trials in 5 of 25 batches (i.e., m = 0). For each test series, two kinds of series were utilized depending on the duration of the trial: i.e., short- (3.5-4.5 s) or long-interval (6.5-7.5 s) series.

The apparatus was designed along similar principles to the apparatus used by Ohki et al. (2002) (Fig. 1A), and comprised two horizontally oriented flat circular grasp plates (30-mm diam, center-to-center distance 50 mm) covered with suede. Each plate was connected via 10-cm-long rigid beams to separate servoregulated torque motors (SGM-A5BWG32, Yaskawa) that could generate loading forces away from the subject at the grasp plate (0-5 N, 0-20 Hz bandwidth, noise < 0.05 N). A laboratory computer controlled both force motors. The apparatus provided no sound cues. Strain gauges were connected to each beam, to provide independent recordings of load force tangential to the grasp plate and the normal force produced by each finger perpendicular to the grasp plate.

Load trials

Each trial consisted of a waiting phase, load phase (0.5 s), hold phase (1.0 s) and resting phase (0.5 s; Fig. 1B). Depending on the duration of the waiting phase, test series were categorized as either short interval (1.5-2.5 s) or long interval (4.5-5.5 s). To ensure that subjects kept their fingers in contact with the grasp plate between trials (i.e., during waiting and resting phases), a 0.2-N baseline load was applied. After the waiting phase, load was stepped from baseline to 0.8 N, then ramped to 3 N at a rate of 4 N/s. The initial step load force served to trigger a distinct normal force response, and the following ramp phase of the load guaranteed a substantial response amplitude (Johansson et al. 1992a).

Test series

In any one test series, subjects experienced two types of trials, linked and unlinked, each of which displayed different load conditions. In linked trials (L), the two grasp plates were loaded concurrently, whereas only the left plate was loaded in unlinked trials (U). In one set of experiments, subjects participated in four different test series, comprising short-interval link-predictive, long-interval link-predictive, short-interval unlink-predictive, and long-interval unlink-predictive test series. Each series consisted of 200 load trials (25 batches of 8 trials). To prevent fatigue, short rest periods were given after every 40 trials.

LINK-PREDICTIVE SERIES. A beep (1000 Hz, 100 ms in duration) was given at the start of every batch (i.e., 8 trials). Each batch started with a succession of one to four linked trials, with the remaining trials in a batch unlinked. Subjects were therefore able to safely predict the start of sequential linked trials by the beep (link-predictive). Conversely, subjects could not precisely predict the start of unlinked trials. However, once the trial was changed to unlinked in the middle of a batch, subjects were able to predict the succession of unlinked trials until the next beep.

It should be noted here that 5 of the 25 batches started with an unlinked trial, with no linked trial delivered after the beep (see trials after the 3rd beep, Fig. 1C, top). This type of batch was introduced to clarify the effects of cue, and the 5 batches occurred in random positions within each series. However, even in such cases, the 8 unlinked trials after the beep were treated as U1-U8, as the beep clearly induced some adaptation of response to the linked trial. A total of 50 linked and 150 unlinked trials were included in the test series.

Although this series was designed to observe the progression of adaptation to unlinked trials, subjects also experienced sequential linked trials in the same series (L1-Lm). However, adaptation to linked trials was slightly different in nature from that observed in the other test series (see UNLINK-PREDICTIVE SERIES), as the starts of sequential trials were cued with beeps. We thus compared responses in linked trials between link-predictive and unlink-predictive series, to show the effects of cues on adaptation.

UNLINK-PREDICTIVE SERIES. Linked and unlinked trials were identical to those in the link-predictive series. Construction of this test series resembled the "three link condition" used by Witney et al. (2000). That is, 3 successive linked trials (L1-L3) occurred in each batch of 8 trials, starting with one of the first 4 trials in the batch. The last linked trial was therefore followed by 2 to 5 unlinked trials in single batches. However, when the first trial in the next batch was unlinked, subjects perceived that the unlinked trials were continuing without notifying the start of the next batch (see start of 2nd batch, Fig. 1C, bottom). The test series was named unlink-predictive, as subjects were cued on the starts of sequential unlinked trials by beeps, although the starts of sequential linked trials occurred unexpectedly. Thus in this series, we could observe adaptation of reactive responses to linked trials with concurrent loading following unexpected changes in load condition from isolated load to the left plate. A total of 75 linked trials and 125 unlinked trials were included in the test series.

In the test series, subjects also experienced sequential unlinked trials (U1-Un), with starts cued by beeps. As with linked trials in the previous test series, effects of cue on adaptation to unlinked trials were analyzed by comparing responses between link- and unlink-predictive test series.

In both unlink- and link-predictive series, the number of trials was smallest for the eighth unlinked trial (U8, n = 5). Subjects let the apparatus slip in 0.005% of trials and, in such cases, the batch was repeated from the beginning, and the first trial after restart was excluded from analysis. Similarly, first trials after the short rest periods were excluded from analysis.

Data collection and analysis

Data were collected and analyzed using a laboratory computer system (SC/ZOOM; Physiology Section, IMB, University of Umeå). Force signals were sampled at 400 Hz. Force rates were obtained as a function of time using symmetrical numerical time differentiation within a time window corresponding to ±5 data samples. That is, averages of 5 points before and 5 points after the point were obtained for each sampling point, and the slope between the two averages was calculated.

The following measurements from each finger were taken from every trial. Preload normal force was the normal force present at the onset of load increase. Grasp response onset latency was the time interval from onset of load force increase to onset of reactive normal force increase, as assessed from force rate signals. Although we identified one or more peaks in the normal force rate profile (Fig. 2) (cf. Johansson et al. 1992a; Ohki and Johansson 1999), first peak amplitude and first peak latency relative to onset of load force were measured. A normal force response was considered present when the maximum normal force rate exceeded 2.5 N/s. Responses of weaker strengths could not be reliably detected in single trial records. In such cases in which weaker responses occurred, none of the above response parameters were measured, but peak rates were set to zero in statistical analyses. To quantify synchronization of early responses by the two fingers, the correlation coefficient of dynamic responses was calculated in a single trial with normal force rates of left versus right fingers during 50-300 ms (i.e., 100 points, sampled at 2.5-ms intervals) after load onset (cf. Shadmehr and Mussa-Ivaldi 1994). Static normal force response was measured as the difference between preload normal force and normal force at 100 ms before the end of the hold phase.

View larger version (42K): [in this window] [in a new window]   FIG. 2. A composite figure of averaged normal force rates and normal and load forces from a single subject in sequential linked (L1-L4) and unlinked (U1-U8) trials in a link-predictive series. Dashed arrows on the top indicate possible flows from linked to unlinked trials in the test series. Relative time scales from load onset are shown at the bottom for each trial number. Note that only the first 2 trials (e.g., U1-U2 for unlinked behavior) and the last trial (e.g., U8) were shown in the sequential experiences of one type of load condition. A: short-interval series. B: long-interval series. White arrows indicate responses of the nonloaded right finger in the unlinked trials.

Statistical methods

Numerical values of normal forces were input into statistical software (STATISTICA, Statsoft, Tulsa, OK). Unless otherwise stated, statistical reports were made using repeated measures ANOVA. One type of ANOVA was performed primarily to assess the influence of adaptation after unexpected changes in load condition. Factors comprised trial number (L1-L3 for linked trials or U1-U8 for unlinked trials) and interval length (2 levels: short or long interval). The other type of ANOVA was performed to analyze cue effects. Factors comprised cue (2 levels: with or without cue), trial number (see trial number above), and interval length (2 levels, see interval length above). In addition, planned comparisons were performed to analyze specific effects, as described in the RESULTS. Values of P < 0.05 were considered statistically significant and population estimates are given in the text in the form of subject mean ± SD (n = 7). For each subject, and for each experimental condition, values for all trials were averaged and subject SD (n = 7) was calculated for these "average trials." Average trials were also used in ANOVA analyses. Prior to calculating subject mean, correlation coefficients were z-transformed to normalize distributions. In graphs, SE is indicated unilaterally (n = 7).

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION ACKNOWLEDGMENTS REFERENCES

 
Response adaptation during sequential unlinked trials

For self-generated bimanual movements, Witney et al. (2000) used unlinked trials to assess development of the forward model, as any change in grip force seen in the nonloaded right hand reflects purely predictive components and thus is a consequence of internal modeling. Thus we first examined adaptive changes in reactive response during sequential unlinked trials.

The framed responses in Fig. 2 were performed by a representative subject during sequential unlinked trials; i.e., during link-predictive series with short (Fig. 2A) and long (Fig. 2B) intertrial intervals. Loading of the apparatus reliably triggered normal force responses at the loaded left finger (Cole and Abbs 1988; Johansson et al. 1992a,c; Johansson and Westling 1988). Nonloaded right fingers occasionally displayed a transient force increase when left fingers were loaded in isolation (white arrows) (cf. Ohki and Johansson 1999). These responses at loaded and nonloaded fingers were observed in all subjects and in all test series. However, when normal subjects experience one type of load condition throughout most of the trials, normal force responses become adapted to that condition (Ohki et al. 2002). Adapted responses were also observed in the present task after experiencing sequential unlinked trials. Transient force increases at the nonloaded right finger decreased gradually from U1 to U8 (see U8 in Fig. 2, A and B).

Figure 3, A and B, shows response onset latencies in sequential unlinked trials (U1-U8) during the link-predictive series. Note that latencies in nonloaded right fingers are presented only for U1-U5, as some subjects showed no detectable response in U6-U8. Normal force response from left fingers loaded in isolation started at 72.5 ± 7.0 ms after load onset. Latency of nonloaded right fingers was 93.4 ± 10.0 ms, significantly longer than that of loaded left fingers [F(1,6) = 47.7, P < 0.01; main effect by finger, for factors: finger x trial number (U1-U5) x interval length] (Ohki and Johansson 1999). Changes in response onset latencies during sequential unlinked trials demonstrated no adaptive changes with changing interval length, which was 5.6 ms longer in long-interval series than in short-interval series [data pooled from two fingers; F(1,6) = 13.1, P < 0.05; main effect by interval length].

View larger version (40K): [in this window] [in a new window]   FIG. 3. Adaptive changes during sequential unlinked trials after unexpected switch in load condition. Parameters are response onset latency (A and B), preload normal force (C and D), amplitude of first force rate peak (E and F), and correlation coefficient of the dynamic responses of the 2 fingers (G and H). Data were obtained from link-predictive series, except for those indicated by gray lines in E-H. Note that latencies in nonloaded right finger are not shown in U6-U8, as no response was detected in some subjects.

Contrasting with the relatively minor adaptive change for preload normal force, amplitude of response, particularly at the nonloaded right finger, displayed clear effects of adaptation during sequential unlinked trials. When load conditions suddenly changed from concurrent loading to isolated loading during the link-predictive series, response of the nonloaded right finger was relatively strong in the first trial (black solid line, representing changes in first force rate peak of the response, Fig. 3, E and F). During subsequent unlinked trials, response weakened by 9.0 N/s [F(7,42) = 38.8, P < 0.01; main effect by trial number] and was virtually absent by U8. Latency of the peak was 119.8 ± 2.1 ms in U1 and 109.4 ± 3.4 ms in U5. Close inspection revealed that the most prominent reduction in peak amplitude (by 6.4 N/s) occurred after experiencing one unlinked trial, i.e., between U1 and U2. However, response in U2 was still stronger than that in U8 (P < 0.005, Tukey's honestly significant difference test). The first force rate peak of the left finger loaded in isolation appeared at 121.3 ± 15.1 ms after load onset. Peak amplitude of the finger remained fairly constant during sequential unlinked trials (black broken lines, Fig. 3, E and F). The loaded left finger thus responded with minor response from the nonloaded right finger in the final stage of adaptation [F(7,42) = 4.0, P < 0.01; interaction between finger and trial number, for factors: finger x trial number x interval length]. Neither finger displayed any significant difference in peak amplitude with changing interval length.

Adaptive change during unlinked trials was also observed in the correlation coefficient for dynamic responses of the two fingers (black lines, Fig. 3, G and H), which were high in U1 and declined in subsequent trials [F(7,42) = 18.0, P < 0.01; main effect by trial number]. No correlation was observed between the correlation coefficient and interval length.

No adaptive change in the amplitude of response during hold phase was apparent during sequential unlinked trials. For example, amplitude of static response displayed no main effect by trial number during unlinked trials.

In contrast to the link-predictive series, subjects received advance notice of the starts of sequential unlinked trials in the unlink-predictive series (cf. Fig. 5). Thus by comparing responses during unlinked trials between link- and unlink-predictive series, we examined the effects of auditory cueing; i.e., subject predictions independent from feedback experiences. In U1, after switching from linked to unlinked trials, the first force rate peak of the nonloaded right finger tended to be weaker when subjects were able to predict the new load condition based on an auditory cue (gray solid lines from unlink-predictive series, Fig. 3, E and F) than when they were unable to make such a prediction (black solid lines from link-predictive series). Amplitude was influenced by the cue [F(1,6) = 8.1, P < 0.05; main effect by cue], but the interaction of cue and trial number was also significant [F(7,42) = 4,8, P < 0.01]. Indeed, subsequent trials displayed a smaller decline in amplitude with cue (gray solid lines) than without (black solid lines), and the effects of cueing were reversed between U1 and the remaining trials [planned comparison; F(1,6) = 20.2, P < 0.01]. Note that after sequential unlinked trials preceded by the cue during the unlink-predictive series (gray solid lines), subjects were not notified when a linked trial would be imposed (cf. Fig. 1C). Change in amplitude thus suggested that subjects gradually shifted their predictions to a linked trial with concurrent loading during sequential unlinked trials in the unlink-predictive series, irrespective of feedback experiences. The same reversal of effect was also observed in the correlation coefficient between dynamic responses [planned comparison; F(1,6) = 23.8, P < 0.01; Fig. 3, G and H].

View larger version (40K): [in this window] [in a new window]   FIG. 5. A composite figure of averaged normal force rates and normal and load forces from a single subject in sequential unlinked (U1-U8) and linked (L1-L3) trials in unlink-predictive series. Dashed arrows on the top indicate possible flows from unlinked to linked trials in the test series. A: short-interval series. B: long-interval series. White arrows indicate responses of nonloaded right fingers in unlinked trials.

View larger version (26K): [in this window] [in a new window]   FIG. 4. Effects of preceding linked trials during sequential unlinked trials. First force rate peaks in sequential unlinked trials are shown for nonloaded right fingers and were obtained in link-predictive test series. Values are separately presented with lines in different styles depending on the number of experienced linked trials after the beep, which preceded unlinked trials.

Although grip force modulation in linked trials does not represent a pure predictive component, Witney et al. (2000) observed adaptive change of grip force modulation during linked trials. Adaptive changes were therefore also examined during sequential linked trials (Fig. 5). In linked trials during the unlink-predictive series, normal force response of the left and right fingers started at 70.8 ± 5.5 and 77.6 ± 6.7 ms after the onset of concurrent loading, respectively (Fig. 6, A and B). Only latency of the right finger was shortened by 4.1 ms during sequential trials [F(2,12) = 27.5, P < 0.01; main effect by trial number]. Latency was also influenced by interval length for both left [F(1,6) = 25.5, P < 0.01; main effect by interval length, 3.8 ms difference] and right [F(1,6) = 23.3, P < 0.01; 3.9 ms] fingers. Adaptive changes in preload normal force were again relatively minor. Significant effect was observed only in the left finger [F(2,12) = 13.4, P < 0.01; main effect by trial number, 0.25 N decrease; Fig. 6, C and D]. Adaptive decreases in the left finger were similarly observed irrespective of interval length, although main effect by interval length was observed [F(1,6) = 6.4, P < 0.05].

View larger version (31K): [in this window] [in a new window]   FIG. 6. Adaptive changes during sequential linked trials after unexpected switch in load condition. Parameters are the same as in Fig. 3. Data were obtained from unlink-predictive series, except for those indicated by gray lines in E-H.

The effects of auditory cues were examined by comparing responses during sequential linked trials with (gray lines from link-predictive series, Fig. 6, E-H) and without (black lines from unlink-predictive series) cues. An interaction was identified between cue and trial number in first force rate peaks over sequential linked trials [F(2,12) = 4.6, P < 0.05, data from L1-L3 analyzed; Fig. 6, E and F]. The effect of cue in L2-L3 was reversed from that in L1 during the short-interval test series [planned comparison; F(1,6) = 17.9, P < 0.01], and disappeared during the long-interval test series. In addition, the correlation coefficient between dynamic responses was affected differently by the cue in L1 and subsequent trials [interaction; F(2,12) = 4.2, P < 0.05].

The effects of preceding unlinked trials on response were also analyzed during sequential linked trials in the unlink-predictive series. First force rate peak in the linked trial was significantly influenced by the number of experienced unlinked trials before the switch in load condition [F(5,30) = 4.1, P < 0.01; main effect by experienced unlinked trials, for factors: experienced unlinked trials (3-8) x interval length x trial number]. This was largely attributable to the strong response after the eighth unlinked trial (U8), with no influence on first force rate peak observed if U8 was omitted from analysis [experienced unlinked trials (3-7)].

Adaptation to linked and unlinked trial depends on prediction error

Regarding adaptation with self-generated movement, the magnitude of anticipatory grip modulation changes depending on prediction error in the previous trial. We thus examined whether this was also the case for sensory-driven reactive movement.

Speed of adaptation was calculated in accordance with the model developed by Witney et al. (2001). In this model, subjects are assumed to alter future responses in proportion to prediction error on the current response. Thus the prediction for concurrent loading in the n^th trial can be determined as p_n = p_n-1 + (1 - p_n-1)g_L when the load condition was concurrent in the previous trial, and p_n = p_n-1 - p_n-1g_U when an unlinked trial with isolated load is experienced (g_L and g_U, proportionality gains for linked and unlinked trials, respectively). Response amplitude in the n^th trial can be assumed to be R_n = R_U + p_n(R_L - R_U) (R_U and R_L, responses when p = 0 and p = 1, respectively), and thus proportionality gains can be determined by response amplitudes in sequential trials based on the following equations: R_n - R_n-1 = g_LR_L - g_LR_n-1 (linked trial) and R_n - R_n-1 = g_UR_U - g_UR_n-1 (unlinked trial).

Figure 7A shows a scatter plot of first force rate peaks for the right finger from single unlinked trials in the link-predictive series. In the graph, the abscissa and ordinate represent R_n-1 and R_n - R_n-1, respectively. As can be expected from the model described above, amplitude change from U1 to U2 displayed correlations with response amplitude in U1 for both short- (black circle) and long-interval series (gray circle; r -0.80, P < 0.01). Significant correlations were observed until n = 4 (r -0.52, P < 0.01), and g_U determined from the slope was about 0.8. Thus, although most adaptive change occurred between U1 and U2 (see above), adaptation was in effect for at least three consecutive trials.

View larger version (29K): [in this window] [in a new window]   FIG. 7. Prediction error and response adaptation. First force rate peaks were obtained in single trials during adaptation after unexpected changes in load condition. A and B: values from unlinked trials in link-predictive series and from linked trials in unlink-predictive series, respectively. First peak amplitude changes from the previous trial (Rn - Rn-1) are plotted against first force rate peaks in the previous trial (Rn-1). Values when n = 2 (U2 or L2) or 3 (U3 or L3) during short- and long-interval series are shown with different symbols, and regression lines are shown separately with different lines. Proportionality gains, gLs and gUs, were determined from the slopes of regression lines.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION ACKNOWLEDGMENTS REFERENCES

 
The present study examined the effects of repetitive experiences of feedback events on the adaptation of bimanual reactive response. We hypothesized that, if the internal model, with which the brain generates motor commands, could modulate the neuronal center engaging in reactive responses, adaptation of the reactive response would be controlled in a manner resembling that observed for voluntary movement (Witney et al. 2000, 2001). We therefore utilized a bimanual task similar to that used by Ohki et al. (2002). During the task, subjects performed bimanual manipulation of two objects. The objects were loaded under computer control, allowing instant changes in load condition, that is, simultaneity of load forces to the objects, on a trial-to-trial basis. Changes in load condition were either preceded by an auditory cue or occurred unexpectedly without cue. Subjects were required to maintain the object in the original position and thus needed to produce a rapid increase in normal force when object load changed.

The task in the present study was comparable to the bimanual task that has been used to examine predictive mechanisms for self-generated movements (Witney et al. 2000), although our apparatus differed from the one used in that study in two major ways. First, our apparatus was not designed to exert force on one plate by movement of the other, as seen in systems used to examine self-generated movements (Blakemore et al. 1998; Witney et al. 2000). However, concurrent load to both plates provides a good reproduction of situations in daily life, in which a single object is held in both hands and pulled by a third party. Indeed, even in self-generated movement, the crucial part is not necessarily to replicate sensory feedback caused by the other hand, but to reproduce sensory feedback consistent with a context (Witney et al. 2001). Second, movements of our plates were rotational, so they were turning in different directions, and this certainly cannot be reconciled with the behavior of a single object. However, movement of the plates during test series was 2 mm at most, as can be seen in the previous study (Fig. 2 of Ohki and Johansson 1999). Given the 10-cm beams connecting plates to servomotors, the distance between the two plates did not change considerably. Indeed, our subjects reported that they could not feel any difference in directions of movements for the two plates. We therefore performed direct comparison of the predictive mechanisms influencing self-generated and sensory-driven reactive movements.

Comparisons of predictive mechanisms for self-generated and reactive movements

As reported previously (Ohki et al. 2002), we observed that not only preload normal force, but also normal force response became adapted to a load condition when subjects repeatedly experienced that condition during sequential trials. However, response immediately after unexpected change of load condition was clearly influenced by previous condition, and at least several trials were required for the subject to display complete adaptation to the new condition, as seen with self-generated movement (Witney et al. 2000). Such adaptive changes were predominantly observed in responses of the right finger when loading differed dramatically between the two conditions. Adaptive changes were clear in the early dynamic part of the response, both when the finger was not loaded and when concurrent loading was present.

Adaptation revealed a number of properties similar to those observed with self-generated movement. First, striking similarities in adaptation were observed during short- and long-interval series. Specifically, adaptation depends on feedback events (i.e., cutaneous inputs from the finger carrying information regarding stability of the finger-object interface), but not on time after an event (Witney et al. 2000 for self-generated movement). Second, feedback experiences influence future responses in a manner whereby prediction error determines modification of the response, as revealed after analysis of adaptive speed in self-generated movement (cf. Witney et al. 2001). In reactive responses, speed of adaptation to concurrent load (g_L 0.4) differs from that to isolated load (g_U 0.8), although the former value is somewhat smaller than the corresponding value for self-generated movement (g_L = 0.51-0.55). Differences in g_L might be explained by the fact that g_L was determined from reactive responses in linked trials, which represent a mixture of sensory-driven and predictive components. Finally, as seen in Fig. 7, subjects can display intermediate prediction between the two basic response patterns on a single trial basis. The only difference from the results seen with self-generated movement was that effects of the number of preceding feedback trials before unexpected change of load condition could not be detected. However, this could be due to the construction of the current test series, in which subjects gradually shifted predictions in preparation for a sudden switch in load condition. The results of the present study therefore fit the hypothesis that selection of an internal model can modulate the neuronal center processing sensory-driven reactive responses.

Previous studies have reported that automatic normal force responses, triggered by inputs from cutaneous receptors in the fingertips, are under predictive control and are scaled using early cues about rate of load force changes (Cole and Abbs 1988; Johansson et al. 1992a; Winstein et al. 2000) and friction (Birznieks et al. 1998; Cole and Johansson 1993). However, such adaptation is known to occur quickly, with initial adjustment taking place soon after the subject touches an object, and thus before loading force and total adjustments are completed within one or two trials. Such quick adaptation is also known to take place for self-generated movements. Parameterization of grip force to load force during self-generated movements is proposed to be under predictive control depending on previous experience, which is described as anticipatory parameter control (Johansson and Cole 1994). For example, several studies have examined how subjects perform when the normal physical properties of an object, such as weight, frictional surfaces, and surface orientation, are changed with each trial (Birznieks et al. 1998; Edin et al. 1992; Gordon et al. 1993; Jenmalm and Johansson 1997; Johansson and Westling 1984). Such studies have demonstrated rapid updating to the new situation within one trial. Similarly, studies of unusual dynamic properties in objects have displayed quick learning of the new dynamics by subjects (Wing and Flanagan 1998). However, if such parameter control is insufficient to allow adaptation to the current context for self-generated movement, several trials are required to switch to the proper control pattern. Such slow adaptation was observed when bimanual coordinated and independent movements must be switched (Witney et al. 2000) or when the object displays unusual weight (Flanagan et al. 2001). The current study therefore shows that such slow adaptation could also take place for sensory-driven reactive responses.

Possible neuronal substrates for adaptive changes

The cerebellum has been considered crucial for forward modeling of movements (Miall et al. 1993). In accordance with the hypothesis that the internal model also affects reactive movement, the cerebellum is known to affect the predictive adaptation of reactive responses. For example, cooling the cerebellum eliminates preparatory responses to resist displacement of handheld objects (Dugas and Smith 1992) and eliminates prediction-dependent modification of scaled wrist responses to pulse and step perturbations (Hore and Villis 1984). In addition, neurons in the cerebellar cortex display preparatory responses that change in parallel with preparatory increases in grip force preceding force-pulse perturbations (Dugas and Smith 1992). Although cerebellar patients are able to use sensorimotor memory about object load to adapt force output based on previous experience, they are significantly less efficient at such activity than healthy controls (Müller and Dichgans 1994).

The present results suggest that the neuronal center for sensory-driven reactive responses also displays predictive adaptation (Ohki et al. 2002; Winstein et al. 2000) in association with selection of an internal model. Automatic motor responses of digits elicited by tactile input are likely to be mediated by a variety of centers that engage different areas in the CNS. These areas would include subcortical centers, such as the spinal cord, and cortical sites (Deuschl et al. 1995; Jenner and Stephens 1982; Johansson et al. 1994; Macefield and Johansson 1994). The influence of subject predictions on response onset latency indicate that initiation of response is also the target of predictive adaptation, which takes place in the subcortical reflex center (cf. Johansson et al. 1994). However, strong adaptive changes are also observed in the early dynamic phase of the reactive response, which most probably involves cortical centers, including the primary motor cortex (Johansson et al. 1994). In accordance with our results, subject preparations are known to influence neural activity in both the primary sensorimotor areas (Evarts and Tanji 1974, 1976) and the premotor cortex (di Pellegrino and Wise 1991; Godschalk et al. 1981; Riehle and Requin 1993). Interestingly, when relatively long preparatory periods are introduced, reaction times increase (cf. Teichner 1954) and set-related activities in the premotor cortex are decreased (Wise and Kurata 1989). These could explain the small differences between responses in short- and long-interval series; responses in the long-interval series displayed consistently longer latencies with weaker preload forces, although similar adaptive changes were observed for both interval lengths. Furthermore, auditory cues induced stronger preparation for linked trials in short-interval series than in long-interval series in the link-predictive series (U1, Fig. 4).

The evidence has increasingly suggested that motor skill learning can be mediated by experience-driven changes within the somatosensory (Braun et al. 2000, 2001; Moore et al. 1999) and motor (Gandolfo et al. 2000; Karni et al. 1995) cortices. Different neuronal groups in cortical centers may therefore be involved in bimanual coordinated movement and unimanual movement, causing different reactive responses when sensory input occurs. Indeed, a separate group of neurons in the primary motor cortex is known to be activated during bilateral movements (Aizawa et al. 1990).

Adaptation without feedback experiences

The present study also displayed top-down effects on the adaptation of finger response to a load condition. That is, the auditory cue or speculation by the subject about the potential for a sudden switch in load condition in the near future could modify response. However, these effects seem rather weak in comparison with those resulting from sensory feedback. That is, if the subject experienced a feedback signal contradictory to the cue in the link-predictive series, response in the next trial was adapted, depending on the immediate sensorimotor memory obtained from the previous trial (Fig. 4).

With self-generated movement, visual information is known to affect the anticipatory parameterization of fingertip forces. Particularly when subjects encounter misleading visual size cues about the weight of an object, they continue to influence predictive force control for a number of trials despite somatosensory information (Flanagan et al. 2001). The influence of visual cues thus appears stronger than that of auditory cues. One possible reason for this difference is that visual size cues are frequently used in daily life to gauge the weight of an object, whereas auditory cues were temporarily used in the current experiment. Thus repeated experiences might strengthen the influence of visual cues on the internal model or visual cues might even be involved in the model.

	ACKNOWLEDGMENTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION ACKNOWLEDGMENTS REFERENCES

 
The authors extend thanks to Prof. Hirai for his valuable comments on the manuscript.

GRANTS

This work was supported by Kyorin University.

	FOOTNOTES

 
The costs of publication of this article were defrayed in part by the payment of page charges. The article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

Address for reprint requests: Y. Ohki, Department of Physiology, Kyorin University School of Medicine, Shinkawa, Mitaka, Tokyo 181-8611, Japan (E-mail: ohkiy{at}kyorin-u.ac.jp).

	REFERENCES

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION ACKNOWLEDGMENTS REFERENCES

 
Aizawa H, Mushiake H, Inase M, and Tanji J. An output zone of the monkey primary motor cortex specialized for bilateral hand movement. Exp Brain Res 82: 219-221, 1990.[ISI][Medline]

Birznieks I, Burstedt MK, Edin BB, and Johansson RS. Mechanisms for force adjustments to unpredictable frictional changes at individual digits during two-fingered manipulation. J Neurophysiol 80: 1989-2002, 1998.[Abstract/Free Full Text]

Blakemore SJ, Goodbody SJ, and Wolpert DM. Predicting the consequences of our own actions: the role of sensorimotor context estimation. J Neurosci 18: 7511-7518, 1998.[Abstract/Free Full Text]

Braun C, Heinz U, Schweizer R, Wiech K, Birbaumer N, and Topka H. Dynamic organization of the somatosensory cortex induced by motor activity. Brain 124: 2259-2267, 2001.[Abstract/Free Full Text]

Braun C, Schweizer R, Elbert T, Birbaumer N, and Taub E. Differential activation in somatosensory cortex for different discrimination tasks. J Neurosci 20: 446-450, 2000.[Abstract/Free Full Text]

Burstedt MK, Flanagan JR, and Johansson RS. Control of grasp stability in humans under different frictional conditions during multidigit manipulation. J Neurophysiol 82: 2393-2405, 1999.[Abstract/Free Full Text]

Cole KJ and Abbs JH. Grip force adjustments evoked by load force perturbations of a grasped object. J Neurophysiol 60: 1513-1522, 1988.[Abstract/Free Full Text]

Cole KJ and Johansson RS. Friction at the digit-object interface scales the sensorimotor transformation for grip responses to pulling loads. Exp Brain Res 95: 523-532, 1993.[ISI][Medline]

Deuschl G, Feifel E, Guschlbauer B, and Lucking CH. Hand muscle reflexes following air puff stimulation. Exp Brain Res 105: 138-146, 1995.[ISI][Medline]

di Pellegrino G and Wise SP. A neurophysiological comparison of three distinct regions of the primate frontal lobe. Brain 114: 951-978, 1991.[Abstract/Free Full Text]

Dugas C and Smith AM. Responses of cerebellar Purkinje cells to slip of a hand-held object. J Neurophysiol 67: 483-495, 1992.[Abstract/Free Full Text]

Edin BB, Westling G, and Johansson RS. Independent control of human finger-tip forces at individual digits during precision lifting. J Physiol 450: 547-564, 1992.[Abstract/Free Full Text]

Evarts EV and Tanji J. Gating of motor cortex reflexes by prior instruction. Brain Res 71: 479-494, 1974.[CrossRef][ISI][Medline]

Evarts EV and Tanji J. Reflex and intended responses in motor cortex pyramidal tract neurons of monkey. J Neurophysiol 39: 1069-1080, 1976.[Abstract/Free Full Text]

Flanagan JR, King S, Wolpert DM, and Johansson RS. Sensorimotor prediction and memory in object manipulation. Can J Exp Psychol 55: 87-95, 2001.[ISI][Medline]

Gandolfo F, Li C, Benda BJ, Schioppa CP, and Bizzi E. Cortical correlates of learning in monkeys adapting to a new dynamical environment. Proc Natl Acad Sci USA 97: 2259-2263, 2000.[Abstract/Free Full Text]

Godschalk M, Lemon RN, Nijs HG, and Kuypers HG. Behaviour of neurons in monkey peri-arcuate and precentral cortex before and during visually guided arm and hand movements. Exp Brain Res 44: 113-116, 1981.[ISI][Medline]

Gordon AM, Westling G, Cole KJ, and Johansson RS. Memory representations underlying motor commands used during manipulation of common and novel objects. J Neurophysiol 69: 1789-1796, 1993.[Abstract/Free Full Text]

Hogan N, Bizzi E, Mussa-Ivaldi FA, and Flash T. Controlling multijoint motor behavior. Exerc Sport Sci Rev 15: 153-190, 1987.[ISI][Medline]

Hore J and Villis J. Loss of set in muscle response to limb perturbations during cerebellar dysfunction. J Neurophysiol 51: 1137-1148, 1984.[Abstract/Free Full Text]

Jenmalm P and Johansson RS. Visual and somatosensory information about object shape control manipulative fingertip forces. J Neurosci 17: 4486-4499, 1997.[Abstract/Free Full Text]

Jenner JR and Stephens JA. Cutaneous reflex responses and their central nervous pathways studied in man. J Physiol 333: 405-419, 1982.[Abstract/Free Full Text]

Johansson RS. Sensory control of dexterous manipulation in humans. In: Hand and Brain: the Neurophysiology and Psychology of Hand Movements, edited by AM Wing, P Haggard, and JR Flanagan. San Diego, CA: Academic Press, 1996, p. 381-414.

Johansson RS and Cole KJ. Grasp stability during manipulative actions. Can J Physiol Pharmacol 72: 511-524, 1994.[ISI][Medline]

Johansson RS, Häger C, and Riso R. Somatosensory control of precision grip during unpredictable pulling loads. II. Changes in load force rate. Exp Brain Res 89: 192-203, 1992.[ISI][Medline]

Johansson RS, Häger-Ross C, and Bäckström L. Somatosensory control of precision grip during unpredictable pulling loads. III. Impairments during digital anesthesia. Exp Brain Res 89: 204-213, 1992.[CrossRef][ISI][Medline]

Johansson RS, Lemon RN, and Westling G. Time varying enhancement of human cortical excitability mediated by cutaneous inputs during precision grip. J Physiol 481: 761-775, 1994.[ISI][Medline]

Johansson RS, Riso R, Häger C, and Bäckström L. Somatosensory control of precision grip during unpredictable pulling loads. I. Changes in load force amplitude. Exp Brain Res 89: 181-191, 1992.[CrossRef][ISI][Medline]

Johansson RS and Westling G. Roles of glabrous skin receptors and sensorimotor memory in automatic control of precision grip when lifting rougher or more slippery objects. Exp Brain Res 56: 550-564, 1984.[ISI][Medline]

Johansson RS and Westling G. Programmed and triggered actions to rapid load changes during precision grip. Exp Brain Res 71: 72-86, 1988.[ISI][Medline]

Karni A, Meyer G, Jezzard P, Adams MM, Turner R, and Ungerleider LG. Functional MRI evidence for adult motor cortex plasticity during motor skill learning. Nature 377: 155-158, 1995.[CrossRef][Medline]

Macefield VG, Häger-Ross C, and Johansson RS. Control of grip force during restraint of an object held between finger and thumb: responses of cutaneous afferents from the digits. Exp Brain Res 108: 155-171, 1996.[ISI][Medline]

Macefield VG and Johansson RS. Electrical signs of cortical involvement in the automatic control of grip force. Neuroreport 5: 2229-2232, 1994.[ISI][Medline]

Miall RC, Weir DJ, Wolpert DM, and Stein JF. Is the cerebellum a Smith Predictor? J Mot Behav 25: 203-216, 1993.[ISI][Medline]

Moore CI, Nelson SB, and Sur M. Dynamics of neuronal processing in rat somatosensory cortex. Trends Neurosci 22: 513-520, 1999.[CrossRef][ISI][Medline]

Müller F and Dichgans J. Dyscoordination of pinch and lift forces during grasp in patients with cerebellar lesions. Exp Brain Res 101: 485-492, 1994.[ISI][Medline]

Ohki Y, Edin BB, and Johansson RS. Predictions specify reactive control of individual digits in manipulation. J Neurosci 22: 600-610, 2002.[Abstract/Free Full Text]

Ohki Y and Johansson RS. Sensorimotor interactions between pairs of fingers in bimanual and unimanual manipulative tasks. Exp Brain Res 127: 43-53, 1999.[CrossRef][ISI][Medline]

Riehle A and Requin J. The predictive value for performance speed of preparatory changes in neuronal activity of the monkey motor and premotor cortex. Behav Brain Res 53: 35-49, 1993.[CrossRef][ISI][Medline]

Shadmehr R and Mussa-Ivaldi F. Adaptive representation of dynamics during learning of a motor task. J Neurosci 14: 3208-3224, 1994.[Abstract]

Teichner WH. Recent studies of simple reaction time. Psychol Bull 51: 128-149, 1954.[CrossRef][ISI][Medline]

Wing AM. Anticipatory control of grip force in rapid arm movements. In: Hand and Brain: Neurophysiology and Psychology of Hand Movement. edited by AM Wing, P Haggard, and JR Flanagan, San Diego, CA: Academic Press, 1996, p. 301-324.

Wing AM and Flanagan JR. Anticipating dynamic loads in handling objects. Proceedings of the ASME Dynamic Systems and Control Division. DSC 64: 139-143, 1998.

Winstein CJ, Horak FB, and Fisher BE. Influence of central set on anticipatory and triggered grip-force adjustments. Exp Brain Res 130: 298-308, 2000.[CrossRef][ISI][Medline]

Wise SP and Kurata K. Set-related activity in the premotor cortex of rhesus monkeys: effect of triggering cues and relatively long delay intervals. Somatosens Mot Res 6: 455-476, 1989.[ISI][Medline]

Witney A, Goodbody SJ, and Wolpert DM. Learning and decay of prediction in object manipulation. J Neurophysiol 84: 334-343, 2000.[Abstract/Free Full Text]

Witney AG, Vetter P, and Wolpert DM. The influence of previous experience on predictive motor control. Neuroreport 12: 649-653, 2001.[CrossRef][ISI][Medline]

Wolpert DM and Flanagan JR. Motor prediction. Curr Biol 11: R729-R732, 2001.[CrossRef]