INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The muscle activity pattern required to produce a desired whole-arm reaching movement or a net isometric force at the hand varies with the posture of the arm. This results from posture-dependent changes in muscle contractile properties and in their pulling angles and moment arms at the joint(s) across which they act (Buchanan et al. 1986; Buneo et al. 1997; Flanders and Soechting 1990; Hoffman and Strick 1999; Karst and Hasan 1991a,b; Van Zuylen et al. 1988; Wadman et al. 1980; Winters and Woo 1990) as well as from posture-dependent variations of global mechanical properties of the limb (Flash and Mussa-Ivaldi 1990; Gordon et al. 1994; Mussa-Ivaldi et al. 1985; Shadmehr and Mussa-Ivaldi 1994).

The nature of the representation of arm motor output variables in the primary motor cortex (MI) is the subject of continuing lively debate (Ajemian et al. 2000; Ashe 1997; Cabel et al. 2001; Johnson et al. 2001; Kakei et al. 1999; Moran and Schwartz 1999a,b; Scott and Kalaska 1997; Scott et al. 2001; Taira et al. 1996; Todorov 2000). If MI single-cell activity covaries explicitly and exclusively with global extrinsic or hand-centered parameters of motor output, all of the putative transformations necessary to generate muscle activation signals must occur elsewhere. Alternatively, if MI single-cell activity was modulated by arm posture, it could be implicated in the transformation of the representation of motor output variables from external to internal coordinate frameworks. Adding to the debate is evidence that the representation can change with time during motor tasks (Johnson et al. 2001; Shen and Alexander 1997a,b; Zhang et al. 1997).

Signals covarying closely with extrinsic parameters of arm movement can be extracted from MI discharge at both single-cell and population levels (Caminiti et al. 1990, 1991; Georgopoulos et al. 1983, 1988; Johnson et al. 2001; Kalaska et al. 1989; Moran and Schwartz 1999a,b; Schwartz 1992; Schwartz et al. 1988). However, other studies have shown that the activity of wrist-related MI cells is altered by changes in the posture of the wrist during movement (Kakei et al. 1999) and during isometric-force tasks (Fromm 1983).

Similarly, MI single-cell discharge is also modulated by arm geometry during whole-arm reaching movements along parallel hand paths in different locations in a three-dimensional (3D) space (Caminiti et al. 1990, 1991) and while making reaching movements along the same two-dimensional (2D) planar hand paths but with the arm in two different orientations (Scott and Kalaska 1997; Scott et al. 1997). Both studies reported arm geometry-dependent changes in directional tuning of single cells. However, unlike Caminiti et al. (1990, 1991), Scott and Kalaska (1997) found no evidence of a systematic shift in the distribution of directional preferences of the cells between the two arm orientations. This discrepancy was attributed to the differing geometries of the tasks in the two studies (Scott and Kalaska 1997). Specifically, rotation of the arm posture about an axis parallel to the plane of motion of the hand might not result in a net shift in the overall distribution of observed 2D cell directional tuning curves, even if the global tuning functions of single cells are coupled to and rotated with changes in arm geometry (Scott and Kalaska 1997). One can predict that a systematic rotation of MI directional tuning functions would be observed in a 2D task that involved rotation of arm postures about an axis perpendicular to the plane of motor output at the hand. It is also important to assess to what degree this rotation is seen prior to and after the onset of overt motor output, i.e., prior to and after the arrival of peripheral reafferent signals.

Therefore we studied MI single-cell activity while monkeys exerted isometric output forces at the hand in eight constant spatial directions in the horizontal plane against a rigid force transducer that was positioned in nine different spatial locations in a horizontal plane. This required rotations of arm posture about the vertical axis through the shoulder joint. The isometric-force task also avoided confounds introduced by the complex dynamics of whole-arm movement. As predicted, directional tuning rotated systematically with arm posture at both single-cell and population levels. Preliminary results have been reported previously during static force generation (Kalaska et al. 1997, 1998; Sergio and Kalaska 1997a,b). Here we present evidence that similar effects can be seen during the force-ramp epoch and during the behavioral reaction time prior to the onset of force generation. We also present further evidence that this effect reflects mechanical anisotropies in the mechanical properties of the arm.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Task apparatus

Two juvenile male rhesus monkeys (Macacca mulatta, 5 and 6 kg) exerted isometric forces with their whole arm against a rigid manipulandum that they held in their hand. The manipulandum was a 20-mm ball on the end of a 65-mm vertical rod attached to a 6 df force transducer (Gamma F3/T10 system, Assurance Technologies). The transducer was housed inside a box that could be clamped into one of nine different locations in front of the animal (Fig. 1A, squares). The central location was at the midline, 20 cm in front of the monkeys' sternum, approximately level to the zyphoid process. The other eight locations were equally spaced every 45° on a circle of 8 cm radius, starting at 0° to the right (Fig. 1A).

View larger version (43K): [in this window] [in a new window]   Fig. 1. A: experimental setup and task apparatus used in the study. Circles on the monitor illustrate the locations of the peripheral isometric force targets. In the task, only 1 target circle was visible at any one time. Squares in front of animal represent locations on a horizontal plane at which the force transducer was placed. Monkeys grasped the handle of the force transducer and generated force at the hand to displace the cursor on the monitor. The location of the monitor and target locations remained invariant for the different hand locations. B: static shoulder and elbow angles for each hand location in the workspace. Joint angles follow the convention used by Soechting et al. (1989). Zero degrees for each orientation are as follows: shoulder pitchupper arm vertical, positive pitch is a forward rotation; shoulder rollforearm in a parasagittal plane, positive roll is an outward rotation; shoulder yawupper arm in a parasagittal plane, positive yaw is an outward rotation; elbow pitchlower arm fully flexed, positive pitch is elbow extension (angle between upper and lower arms). The oblique grid is the best-fit plane of the 8 points in each figure.

The manipulandum rod passed freely through a small hole in a thin metal plate that was attached by hinges to the top of the transducer housing on the edge furthest away from the monkey. The free end of the plate closest to the monkey sat on a micro-switch. If the monkey rested its hand on the plate or otherwise applied forces to it, the switch would close and halt the task. In this way, all forces generated at the hand were applied only to the manipulandum.

A cursor on a monitor positioned at eye level 60 cm in front of the monkey gave continuous visual feedback about the current force generated by the monkey at its hand in the x-y (horizontal) plane (Fig. 1A). The x axis was aligned to the 0-180° (right-left) direction in front of the monkey, and the y axis to the 90-270° direction. Force data were sampled and stored every 5 ms.

The spatial position of the limb joints was recorded (OptoTrak 3020, Northern Digital) in one animal at each of the nine transducer locations to calculate limb segment orientations using standard rigid body reconstruction techniques (Fig. 1B) (see Soechting and Flanders 1989 for a description of the coordinate frame).

Behavioral task

To start each trial, a circle representing a central force target appeared at the center of the monitor. The diameter of the circle corresponded to a force range of 0.20 N. A constant programmed bias in the display of cursor position required the monkey to exert a force of +0.3 N at 90° along the y axis away from its body to position the cursor at the center of the central target (Fig. 2). The monkey kept the cursor within the central target for a variable time period (2,000 ± 500 ms, Gaussian distribution). At the end of the central hold time, the central target disappeared and one of eight peripheral force targets arrayed in a circle around the central target appeared (Fig. 1A, circles; diameter: 0.28 N). The separation of the centers of the central and peripheral targets corresponded to a force change of 1.5 N. The monkey generated the 1.5 N force ramp in the indicated direction in the horizontal plane to move the cursor into the peripheral target and held it there for 2,000 ms to receive a liquid reward (Fig. 2). Note how the animals generated a smooth force ramp and did not relax the bias force before directing a force toward the peripheral target. Peripheral target directions were spaced at 45° intervals, starting from 0° to the right and proceeding counterclockwise (Fig. 1A). A complete data file comprised five successful replications of the isometric ramp-and-hold force outputs to the eight peripheral targets, presented in a randomized-block sequence, while the transducer was clamped at one hand location.

View larger version (10K): [in this window] [in a new window]   Fig. 2. Isometric force profile (solid line) and single-cell discharge (horizontal row of vertical lines) for 1 trial showing epochs analyzed in this study. Second and third vertical lines, the onset and end of dynamic force time. First vertical line, the appearance of the peripheral target on the monitor. The time scale is aligned to the onset of the isometric force ramp aimed at the peripheral target.

The monkey performed sequential files of trials with the manipulandum clamped in different workspace locations. Locations were tested in a pseudo-random sequence for a total of nine files (360 successful trials) per complete data set. In this way, the animal viewed the same display and produced output forces at the hand in the same eight constant directions in hand-centered spatial coordinates while holding the hand at up to nine different spatial locations.

The monkeys also received continual visual feedback about the forces generated at the hand in the vertical (z) axis. A rectangular box was drawn about the cursor (Fig. 1A) and moved with it on the monitor as the monkey generated forces in the horizontal plane. The vertical position of this box relative to the cursor was continuously adjusted as a function of the measured vertical forces. A small constant vertical offset in the display of the box required the monkey to apply a 0.3-N downward force in the z axis to center the cursor in the box. This allowed the monkeys to rest their hand gently on the manipulandum while performing the task. The vertical length of the box was scaled as a function of the acceptable range of vertical forces (±0.26 N for monkey A, ±0.40 N for monkey B) about the constant vertical offset. If the z axis forces exceeded that range, the box shifted to a point where the cursor was no longer inside it, and the trial was terminated. In this way, the net output force vectors generated by the monkeys at the hand were confined to a narrow vertical range about the horizontal plane. This extra behavioral control prevented the monkeys from developing a strategy of systematically varying the z axis component of the net output force vectors at the hand as a function of target force direction or hand location in the workspace, which would confound the interpretation of cell activity.

Data collection

The animals were trained to >80% success rates at all nine locations. They were then prepared for recording by implantation of a recording cylinder over MI using standard aseptic surgical techniques (Kalaska et al. 1989).

Conventional techniques were used to record the activity of single cells in MI (Kalaska et al. 1989). During each recording session, a microelectrode was advanced through the cortex while the animal performed the task at the central workspace location. When a cell was isolated, its task-related responses were tested initially by performing a few trials at several hand locations. The passive responses of cells were studied by manipulating the arm joints, brushing the skin and palpating muscles. The arm was monitored for signs of movements or muscle contractions during low-threshold intracortical microstimulation (ICMS) of the cortex. If a cell responded to active or passive movements of the contralateral shoulder and/or elbow but not to more distal joints and displayed directional tuning in at least one hand location, it was then subjected to further study in the task.

Attempts were made to record cells from all cortical layers. However, because the task required data collection for long time periods to test across all hand locations, there was a bias toward collecting complete data sets more frequently from large cells in intermediate layers.

The time needed to collect a complete data set also raises the possibility that changes in a cell's activity across data files might be due to temporal variations in cell responsiveness independent of the experimentally manipulated parameter of hand location (Kalaska et al. 1989; Scott and Kalaska 1997). This potential confound was handled in two ways. First, files were collected from different locations in a pseudorandom sequence for each cell to minimize the risk of confounding systematic changes in activity due to hand location with systematic temporal changes. Second, a repeated data file was periodically recorded from the first hand location tested. If it appeared to be comparable to the data from the original file at that location, the cell's activity was considered stable. If changes were evident, data files were also replicated from other hand locations until consistent results were obtained. If repeatable results were never achieved, all data files for that cell were deleted. This rarely happened.

Activity was recorded from 16 proximal-arm muscles in monkey A in separate recording sessions. Muscles were implanted percutaneously with pairs of Teflon-insulated 50-µm single-stranded stainless steel wires. Implantations were verified by passing current through the wires to evoke focal muscular contractions (<1.0 mA, 30 Hz, 300-ms train). Multi-unit electromyographic (EMG) activity was amplified, band-pass filtered (100-3,000 Hz), half-wave rectified, integrated (5-ms time bins), and digitized on-line at 200 Hz. The muscles studied included the biceps brachii (2 sets), brachialis (1), anterior deltoid (1), medial deltoid (2), posterior deltoid (2), dorsoepitrochlearis (1), infraspinatus (2), latissimus dorsi (2), pectoralis (2), subscapularis (2), supraspinatus (1), teres major (2), rostral trapezius (2), caudal trapezius (2), triceps longus (2), and triceps medialis (1). These recordings were done to assess the general effects of arm posture on EMG activity in this task to provide a benchmark against which to compare cell activity. They were not designed as a definitive biomechanical study of the properties of each muscle.

Near the end of recordings in each cylinder, small electrolytic lesions were made (5-10 µA, 5 s) in selected penetrations. At the conclusion of the experiment, the monkeys were deeply anesthetized with barbiturates and perfused with buffered saline and formalin. Pins were inserted into the cortex at known grid coordinates to delimit the area from which cell recordings were made, and the cortex was sectioned to permit localization of the marked penetrations.

Data analysis

BEHAVIORAL DATA. Successful performance of a trial involved three phases of force production (Fig. 2). First, exertion of a small static bias force in the horizontal plane directed away from the monkey positioned the cursor in the central target. Second, following peripheral target presentation, dynamic horizontal force ramps in different directions displaced the cursor into the target windows. Third, static horizontal forces in different directions kept the cursor at the peripheral targets. A number of sequential behavioral epochs were defined in each trial (Fig. 2). Center hold time (CHT) began when the animal placed the cursor in the central target and ended on peripheral target appearance. A mean (±SD) baseline force vector was calculated from the 100 force samples collected during the last 500 ms of the CHT. Reaction time (RT) was the interval between target appearance and the time at which there was a significant change in force applied to the transducer. This was defined as the 5-ms sample interval at which the horizontal force vector length deviated 2 SDs from the baseline force and remained so for three more consecutive 5-ms sample intervals. This marked the end of RT and the beginning of the dynamic-force time (DFT), which ended when the cursor first reached a stable force level in the peripheral target window (Fig. 2). This was defined as the time at which the rate of horizontal force change (the 1st derivative of the force level) was within 2 SDs of the mean rate of force change calculated during the last 1,500 ms of the trial. Peripheral target hold time (THT, Fig. 2) was the time from the end of the DFT to the end of the trial.

NEURAL DATA. Cell activity was subjected to a number of different analyses. First, an unbalanced repeated-measures ANOVA (program 5V, BMDP Statistical Software) determined if force direction or hand location had a significant main effect on the mean cell discharge rate during each behavioral epoch. Cells showing a significant statistical interaction between force direction and hand location were identified as evidence that the covariation of cell discharge level with force direction was modulated as a function of the location of the hand in the workspace.

(1)

(2)

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Data base

The task-related activity was recorded from 179 cells during 151 penetrations in MI of three hemispheres from two juvenile monkeys (108 and 32 cells from the left and right hemispheres of monkey A and 39 cells from the right hemisphere of monkey B). Of these cells, 61 were successfully studied in all nine hand locations (48 and 13 in monkeys A and B, respectively). In an additional 35 cells (25 and 10 from monkeys A and B), data were collected from only five to eight hand locations (the center plus at least the 4 cardinal or diagonal hand locations). The remaining 83 cells were studied at fewer than five hand locations and will not be described here. Most of the cells were recorded from caudal MI (Fig. 3) (Crammond and Kalaska 1996, 2000).

View larger version (10K): [in this window] [in a new window]   Fig. 3. Drawing of the cortical surface of both monkeys (right hemisphere only for monkey B) showing the electrode penetration sites. P, posterior; A, anterior; CS, central sulcus; AS, arcuate sulcus; PS, precentral sulcus.

Isometric forces

The mean behavioral reaction time across all trials for both animals was 240 ms, and the mean dynamic force time was 313 ms.

The mean force profiles generated at the hand were similar at all nine hand locations (Fig. 4) despite a large variation in the limb segment orientation angles (Fig. 1B). The nine force profiles for each of the eight force directions start and end at nearly the same location in force space and are confined to a narrow range in the vertical axis (Fig. 4A). Note the lack of initial hooks in the force profiles, indicating that the monkeys did not release the bias force before generating a force in the target direction. Few data sets showed a main effect of hand location on the x, y, or z components of the force outputs during CHT, RT, and THT (Table 1). Some 14-19% of the force traces showed a main effect of target direction during RT. In principle, the RT is a static force period that ends with the onset of the force ramps. The directional effects seen here are likely a threshold phenomenon, reflecting the small directional force-vector deviations necessary to detect the end of the RT epoch (see METHODS, Fig. 2). In contrast, during the THT epoch, while the animals maintained static forces in the peripheral targets, all data sets showed a main effect of target direction on the x and y force components, whereas only one showed a main effect on the z component (Table 1). Across epochs, few force data showed a location-direction interaction with one exception17 sets showed an interaction during THT for the y axis force component only (Table 1). Also noteworthy is the scarcity of significant effects on the z axis force components in all three static epochs.

View larger version (54K): [in this window] [in a new window]   Fig. 4. A: mean force trajectories for monkey A during dynamic-force time (DFT). The 9 lines in each direction are the mean of all trials at each hand location for that force direction. B: result of functional ANOVA (fANOVA) analysis for 1 cell in 1 force direction across all hand locations. The thick horizontal line indicates the critical F value for significance at the 0.05 level. Thick dashed line: x-axis component of force, solid line: y-axis component, thin dashed line: z-axis component.

These ANOVAs used the mean force outputs averaged over periods of hundreds of ms. Such an analysis is inappropriate for the DFT epoch, during which the forces change rapidly. A modified ANOVA intended for time-series data, fANOVA, was applied to the DFT data (see METHODS). A total of 1,368 tests were performed (57 data sets, 8 target directions, 3 force vector components). Figure 4B displays the F value as a function of time for one data set in one target direction and indicates no significant change across hand locations. For monkey A, only five data sets showed a significant difference in the x component of the force output for target direction 270°. These deviations lasted for periods of from 3 to 10 of the 20 force-profile segments in different sets. This typically represents a 48- to 160-ms-long deviation in force output for this one target direction. There were no other significant differences for any other data set or in any other target direction for this animal (P < 0.05), nor for any force data in monkey B. Because only 5/1368 tests displayed a significant difference (0.36%), the animals did not appear to display any systematic differences in their force outputs at the different hand locations during DFT.

These analyses confirmed that behavioral control of the forces generated by the monkeys, in particular in the vertical axis, had resulted in largely invariant force outputs in the horizontal plane at each location and had successfully eliminated nearly all location-related biases in the vertical component of the force output. Any systematic hand location-dependent variation in cell activity could not be readily explained by systematic changes in the net force output at the hand.

EMG activity: general observations

Muscle activity varied systematically with force direction and hand location (Fig. 5). In particular, muscle activity often displayed two major effects during force generation at different locationslarge changes in the overall level of activity and changes in the force-direction tuning function of the muscle, including changes in its depth (dynamic range) and directionality (PD). For example, changes in EMG activity of the right infraspinatus muscle for isometric forces at different directions while the hand was at the 0° location (Fig. 5A) and the 180° location (Fig. 5B) resulted in a shift in PD between the two hand locations (Fig. 5C). The PD shifts across all locations followed an arc-like pattern about the mid-sagittal axis whereby the PD tended to rotate clockwise (CW) relative to that seen at the central location for positions on the right of the midline and counterclockwise (CCW) for positions on the left.

View larger version (22K): [in this window] [in a new window]   Fig. 5. A: electromyographic (EMG) activity of the right infraspinatus muscle during isometric force production while the hand is at the rightward (0o) hand location. Each histogram illustrates the average activity from 5 trials in that force direction. Data are oriented to the onset of DFT (time 0, 1st solid vertical line). The 2nd vertical line denotes the average end of DFT and beginning of target hold time (THT) for the 5 trials at each force direction. B. EMG activity from the same muscle while the hand is held at the leftward (180o) hand location. C: polar-plot representations of the response of the same muscle at all nine hand locations. The position of each polar plot corresponds to the relative location of the hand on the planar work surface. The polar plots for the hand locations at 0o (A) and 180o (B) are reproduced on the right and left, respectively, and the upper-most polar plot corresponds to the hand location furthest away from the monkey's body (90o). The radius of the circle in the polar plot represents the mean muscle activity during the center hold (CHT) epoch. The length of each of the 8 axes in the polar plot represents the mean activity over the 5 trials of force production in each direction during the peripheral target hold (THT) epoch. The heavy arrow corresponds to the preferred direction of the muscle during THT. Note the rotation in the directional tuning of the muscle during THT and the change in mean muscle amplitude during CHT for the different hand locations.

The effect of force direction and hand location were as strong during dynamic force generation as during static force maintenance. All muscle sets showed a significant main effect of both hand location and force direction and a significant interaction effect between location and direction during both DFT and THT (Table 2; Wald test, P < 0.01). During RT, in contrast, all sets showed a main effect of location, but somewhat fewer showed direction and interaction effects. Even prior to target appearance (CHT), 26 (96%) muscle sets showed a main effect of hand location due to systematic changes in tonic EMG activity while generating a constant output force of +0.3 N along the +y axis (Fig. 5C, circles).

Neuronal activity: general observations

The activity displayed by single MI cells (Figs. 6 and 7) showed many parallels to that seen for muscles. Nearly all cells showed a significant main effect of force direction and hand location on discharge level during all three trial epochs (Table 2). Significant interactions between force direction and hand location were less prominent during RT (77%) than during DFT and THT (98%) (Table 2).

View larger version (29K): [in this window] [in a new window]   Fig. 6. Discharge pattern of a primary motor cortex cell during isometric force production while the hand was held at the rightward (A), central (B), and leftward (C) locations. The 8 rasters in each panel illustrate cell activity during 5 trials for 1 force direction, and each raster location corresponds to the direction of force production away from the starting central force target. Data are oriented to the onset of DFT denoted by a solid vertical line at time 0. For each trial, the heavy tick mark to the left of DFT onset shows the time of target onset while the heavy tick mark to the right shows the time at which the final static level of force within the peripheral force target was attained (start of THT). The 8 rasters surround a polar plot depiction of cell activity. The radius of the circle in the polar plot represents the mean cell discharge rate during the center hold (CHT) epoch. The length of each axis represents the mean discharge rate over the 5 trials in that direction during the peripheral target hold (THT) epoch. The heavy arrow corresponds to the preferred direction of the cell during THT.

View larger version (28K): [in this window] [in a new window]   Fig. 7. Polar plot representations of the cell shown in Fig. 6 at all 9 hand locations. Display format is the same as for the EMG polar plots in Fig. 6.

Changes in directional tuning of muscle activity with hand location

One cause of an interaction effect between force direction and hand location is a change in a muscle's directional tuning with arm posture (Scott and Kalaska 1997). The distribution of PD changes was calculated at each hand location for all instances in which a muscle was significantly directionally tuned at both that location and the central location (Table 3A, Fig. 8). There were significant differences (ANOVA, P < .01) in the distributions of PD shifts across the eight peripheral hand locations during DFT and THT but not during RT (Table 3A). A different question is whether the PD shift distribution at each location was directionally biased, that is, significantly skewed away from 0° mean PD shift. No directional bias was found at any location during RT (paired t-test, P > 0.01), eliminating the possibility that the nonsignificant ANOVA during RT resulted from a significant but uniform shift across all hand locations. In contrast, there was a significant CW shift of PDs at the 0° location and CCW shift at the 180° hand location during DFT, and significant CW rotations at both the 0 and 45° locations during THT (P < 0.01, Table 3A, Fig. 8).

View larger version (27K): [in this window] [in a new window]   Fig. 8. Histograms (10o bins) displaying the distribution of shifts in directional tuning between central and peripheral hand locations for each muscle during RT (A) and THT (B). To be included, a muscle had to be significantly directionally tuned (bootstrap test) during the epoch at the central location and the particular peripheral location. The position of the histogram corresponds to the peripheral hand location. , nonsignificant directional shifts at a given peripheral hand location (bootstrap test). , muscles showing a significant shift in their preferred direction (PD) relative to the central location at a given peripheral hand location. , mean PD shift for the total distribution shown. CCW, counterclockwise; CW, clockwise.

The effect of hand location was next evaluated for only the cases of significant PD shifts for single muscles between the central and peripheral locations as assessed by bootstrapping (see METHODS). Significant single-muscle PD shifts were common at several hand locations in the DFT and THT epochs but infrequent during RT (Table 3B, Fig. 8, ). As was the case for the full sample, significant differences in the distributions of significant PD shifts were seen during DFT and THT but not RT (ANOVA, Table 3B). Significant CW shifts occurred at 0 and 45° and CCW shifts at 180 and 225° during both DFT and THT (t-test, Table 3B, Fig. 8).

These quantitative tests confirmed the general impression that hand location had a significant effect on the force direction-related tuning of muscles, and that this resulted in an arc-like rotation of PDs about the mid-sagittal axis. They also revealed that the effects were equally strong during DFT and THT but less prominent prior to the onset of force generation (RT).

Changes in directional tuning of cells with hand location

Similar systematic patterns of directional tuning changes at different times during the trial were also found for many MI cells (Table 3, A and B, Fig. 9). Quantitative analysis of all instances of significant directional tuning at both the central location and individual peripheral locations revealed a significant difference in the distributions of PD shifts across the eight peripheral hand locations during DFT and THT but not RT (ANOVA, P < 0.01, Table 3A). A paired t-test revealed significant CW- and CCW-biased directional shifts at the rightward and leftward hand locations, respectively, especially during DFT and THT (Table 3A). The hand locations along the mid-sagittal plane (90 and 270°) did not show a significantly biased directional shift in any trial epoch.

View larger version (27K): [in this window] [in a new window]   Fig. 9. Histograms displaying the distribution of shifts in directional tuning between central and peripheral hand locations for each cell during RT (A) and THT (B). Display format is the same as in Fig. 8. , mean magnitude of PD shift for all cells in each histogram.

The percentage of single cells displaying a significant PD change at a given hand location relative to that at the center increased from RT to DFT and THT (Table 3B, Fig. 9, ). An ANOVA on only those cells revealed a significant difference in distributions across locations in all trial epochs, and paired t-tests showed systematic CW shifts to the right of the midline and CCW shifts to the left of the midline in all trial epochs (Table 3B, Fig. 9B, ).

Quantitative analysis indicated a similarity in the directional shifts of muscles and cells. A two-way unbalanced ANOVA (main factors: hand location and sourcecells or muscles) revealed a significant main effect (P < 0.01) for hand location but not for source during both DFT and THT.

Closer scrutiny of Fig. 9 revealed other trends. First, during THT, the frequency of significant single-cell PD shifts was fairly constant at all eight peripheral hand locations (Table 3B; Fig. 9B, ). However, for locations to the right or left of midline, the distribution was strongly biased toward CW and CCW changes, respectively. In contrast, they were more symmetrically distributed at the two extremes of the distributions of PD shifts for the 90 and 270° locations. Second, the distributions of all PD shifts were usually unimodal with a single pronounced mode during THT (Fig. 9B) and DFT (data not shown). This was also the case for the PD shift distributions at 90 and 270° during RT (Fig. 9A). However, the distributions at the lateral hand locations usually showed two pronounced peaks during RT, one spanning 0^o and one more eccentrically located (Fig. 9A) where the few instances of significant PD shifts in that epoch were concentrated. The eccentric peaks during RT corresponded to the single major peak at the same locations during THT and DFT. This suggested the transient presence of two functionally different sub-populations of directionally tuned cells during RT, one that showed a directional shift prior to force onset at a given hand location and one that did not. During (DFT) and after (THT) the isometric force ramps had been generated, a greater proportion of the cells showed posture-dependent PD changes.

The activity during the early part of the RT epoch is a continuation of the CHT tonic discharge before the cells begin to respond to the target information. However, this is not likely to explain the weaker influence of arm posture on directional tuning during RT compared with later epochs. Several analyses were repeated using only the activity during the last 100 ms of the RT. The number of cells showing a main effect of location remained essentially constant, 89/96 rather than 92/96 (Table 2). The number of cells that were directionally tuned increased slightly, from 80% on average across hand locations (Table 3A) to 85%. Most importantly, there was little effect on the directional analysis. There was still no significant difference in the PD shift distributions of the whole sample (ANOVA, cf. Table 3A) nor a significant directional bias in the PD shifts (paired t-test, cf. Table 3A). Only 17-32% of the cells showed significant PD shifts between the central and peripheral locations (cf. Table 3B), there was a significant difference in the PD shift distributions of those cells across locations (ANOVA) and significant CW biases in the PD shift distributions at 0 and 315° (paired t-test).

One final observed effect of hand location was a loss of directional tuning. During RT, for instance, 82% of the cells were directionally tuned at the central location. Of those cells, 11-25% were not directionally tuned at one or another of the peripheral locations. This loss of directional tuning was less prominent during THT, when 96% of the cells were tuned at the central location and only 1-13% of those cells were not tuned at different peripheral locations.

These analyses were all based on the mean discharge rate during different epochs. Peak discharge rates were also determined on a trial-by-trial basis using a 50-ms sliding window incremented in 10-ms intervals from 200 ms before target onset to the end of the DFT. Similar overall trends were found for the effects of hand location on peak rates as for epoch means. For instance, 84% of the cells showed a significant main effect of hand location as opposed to 96-98% seen for the mean discharge in RT or DFT epochs (Table 2). This difference in incidence is likely due to the greater inter-trial variability of 50-ms peak rates than epoch means. Because there were no substantive differences, they will not be described further.

Regression analysis of muscle and cell activity with hand location

A linear regression was performed on muscle activity in each epoch to calculate a best-fit planar function for the grand mean of activity averaged across all force directions at each location against its x-y coordinates (Fig. 10A, see METHODS, Eq. 2). The incidence of significant nonhorizontal planar changes in grand mean activity level with hand location increased systematically from CHT to DFT/THT, and those planar functions accounted for 66-98% (mean: 80-82%) of the variance in the data across epochs (Figs. 10A and 11, ; Table 4).

View larger version (49K): [in this window] [in a new window]   Fig. 10. A: mean activity of the right anterior deltoid of monkey A during THT at all hand locations () and its best-fit regression plane. B: mean activity of 1 left MI cell at all hand locations during THT and its best-fit regression plane.

View larger version (37K): [in this window] [in a new window]   Fig. 11. Range of variance of mean activity of muscles (left, n = 27) and cells (right, n = 61) accounted for by a planar regression with x-y hand location. Cells and muscles having both significant () and nonsignificant planar fits () are shown.

Comparable results were observed for the 61 cells recorded at all nine hand locations. The incidence of significant nonhorizontal planar functions for grand mean cell activity increased modestly from CHT to THT (Table 4). The significant planar functions accounted for between 63 and 98% (mean 81-84%) of the variance in the data across epochs (Figs. 10B and 11, ; Table 4). Many of the 35 cells with data collected from only five to eight hand locations also showed significant planar regressions, but they were not included in Table 4 or Fig. 11 (see METHODS). The mean slopes for cells with significant nonhorizontal planes were between 0.06 and 0.09 imp · s¹ · mm¹ (range: 0.02-0.25) across epochs.

The planar regression analysis was repeated using the muscle or cell activity for each of the eight force directions separately. This had only a modest effect on results. For example, the percentage of cells that displayed a significant planar regression on hand location during THT ranged from 46 to 74% for different force directions, compared with 64% when the grand mean activity across all eight force directions at a given location was used.

Distribution of regression plane orientations

The orientation of each significantly nonhorizontal regression plane can be quantified by calculating the normal vector to the best-fit plane. For example, the best-fit plane for the muscle in Fig. 10A had a normal vector pointing toward the animal and to the right, whereas that for the cell in Fig. 10B was oriented toward the monkey and to the left.

Across the sample population, more planes were tilted toward the animal, as indicated in Fig. 12 by the greater number of normal vectors intersecting the spherical surface in front of the frontal plane axis (P < 0.05, ² test). Those cells had positive regression coefficients along the y axis, indicating greater activity for hand locations further away from the animal (Fig. 11). Second, most cell regression-plane normal vectors were distributed in the lower right and upper left quadrants of the sphere, forming an elliptical distribution whose major axis deviated away from the body midline toward the arm being used (Fig. 12). The normal vectors for the cells collected when the monkeys performed the task with their left arm (Fig. 12, ) were mirror-reflected about the mid-sagittal axis on the figure. As originally collected, those best-fit plane normal vectors were mainly in the upper right and lower left quadrants, opposite to that seen for the right arm (Fig. 12, ). These trends were seen in all epochs and suggest that there was a mirror-image change about the body midline for the spatial orientation of the best-fit regression planes for cells in the two hemispheres. The best-fit planes of muscles showed similar trends (Fig. 12,  and ).

View larger version (41K): [in this window] [in a new window]   Fig. 12. Projection of vectors normal to best-fit planes for individual cells ( and ) and muscles ( and ) onto the surface of a sphere. Left MI cells and right arm muscles:  and ; right MI cells and left arm muscles:  and . For graphic purposes, the data collected when the monkey used its left arm have been mirror-reflected about the mid-sagittal axis.

The difference in grand mean cell activity between the central and each peripheral hand location was calculated for each cell in each epoch. A paired t-test (P < 0.01) tested the null hypothesis that the distributions of these differences were centered on zero at each hand location separately. Significant overall increases in the mean discharge level relative to that at the center were found for the three outer hand locations during CHT (45, 90, and 135°), and for four hand locations during THT (45, 90, 135, and 180°). No significant decreases were found. There were no systematic changes in overall cell discharge with hand location for the RT and DFT epochs.

Effect of hand location on dynamic range of cell activity

An interaction effect between direction and location could also be due to location-dependent changes in the depth of a cell's force-direction tuning curve. To test this possibility, we calculated dynamic range (DR) ratios (see METHODS, Eq. 1) for the whole cell sample during RT, DFT, and THT. There was no systematic difference in the distributions of DR ratios as a function of hand location in any epoch (ANOVA, P > 0.01). In addition, a paired t-test performed at each peripheral hand location found no systematic nonzero biases in the distributions of DR values (P > 0.01, RT, DFT, THT) between peripheral and central locations.

A planar regression of DR ratios with x-y hand location was done on a cell-by-cell basis to further test for any systematic bias. As with the planar regressions on mean cell activity, only the 61 cells that were recorded in all nine hand locations were used for this analysis. A moderate number of cells had a hand location-dependent change in dynamic range that could be fit by a significantly nonhorizontal plane in the RT (20%), DFT (30%), and THT (18%) epochs.

The dynamic range also provides a measure of the relative strength of effect of force direction and hand location on cell activity levels. We calculated the distribution of direction-dependent DRs across all force directions at each hand location separately, and the location-dependent DRs across all hand locations for each force direction separately. The median value of the force direction-dependent DR ranged from 23.1 (DFT) to 13.1 imp/s (THT). In contrast, the median hand location-dependent DRs ranged from 22.8 (DFT) to 16.2 imp/s (THT). Values during the RT were intermediate.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

At each hand location, MI cell activity was broadly tuned for net isometric force direction at the hand. This has been described previously for whole-arm isometric force generation (Georgopoulos et al. 1992; Taira et al. 1996) and while defending a given arm posture against loads applied to the limb in different directions (Kalaska and Hyde 1985; Kalaska et al. 1989). Broad tuning is also characteristic of EMG activity during similar movement and isometric-force tasks (Buchanan et al. 1986; Buneo et al. 1997; Flanders and Soechting 1990; Hoffman and Strick 1999; Kakei et al. 1999; Kalaska et al. 1989).

   The most salient finding of this study is that changes in arm posture had a powerful effect on MI cell activity, including changes in both discharge level and directional tuning, when monkeys used their whole arm to generate identical net isometric force ramps in different directions at the hand while holding the hand in different workspace locations. Cells typically showed planar changes in activity level as a function of hand location (i.e., arm posture) in a horizontal workspace. Changes in directional tuning varied systematically with hand location, resulting in a rotation of tuning at the single-cell and population levels about an approximately body-centered reference point. The cell response modulations could not be explained by systematic changes in output forces at different hand locations. Corresponding effects were observed for task-related muscle activity.

The second salient observation in this study was that arm posture-related effects on cell activity could be seen prior to (RT) and during (DFT) the generation of dynamic force output ramps of constant spatial directionality, as well as during the steady-state maintenance of isometric forces (THT). Furthermore, the strength of the posture-related effect was weaker during the RT epoch than later in the trial. This was especially evident for directional tuning. During the RT epoch, fewer cells showed a significant interaction between hand location and force direction (Table 2), overall directional tuning changes were smaller (Table 3A, Fig. 9), and fewer cells showed significant changes in directional tuning at different hand locations (Table 3B, Fig. 9).

Another salient finding was of directional and spatial anisotropy in the activity patterns. Arm location-dependent planar changes in overall activity tended to be inclined toward the body along an axis oriented away from the body midline toward the arm. The distribution of axes normal to the best-fit planes showed a mirror-image reversal for data collected in opposite hemispheres while the monkey used the arm contralateral to the recording site.

Effects of arm posture on directionality of cell activity in MI and PMd

DIRECTIONALITY. Caminiti et al. (1990, 1991) reported that the directional tuning of cell activity in MI and PMd rotated systematically about the vertical axis by an amount similar to the net rotation in shoulder angle about the same axis while reaching in the different horizontally distributed workspaces. Kakei et al. (1999) likewise reported that many MI cells showed systematic rotations of their directional tuning about the long axis of the forearm, when monkeys made isolated wrist movements in eight directions while holding the wrist in pronated, middle, or supinated postures. In contrast, Scott and Kalaska (Scott and Kalaska 1997; Scott et al. 1997) reported single-cell PD changes when monkeys made reaching movements along the same hand paths while holding the arm in two different postures but no evidence of a systematic rotation of the directional tuning of the entire sample population. They suggested that this was due to the geometry of their task in which the arm rotated about an axis parallel to the plane of movement. No net rotation of tuning functions would occur at the population level because the spatial tuning functions of different cells rotated into and out of the plane of action of the task (Scott and Kalaska 1997). In contrast, in the tasks used by Caminiti et al. (1990, 1991) and Kakei et al. (1999), the axis of rotation of the limb was perpendicular to at least one of the planes of motion.

DIRECTIONAL CHANGES IN PRECENTRAL CORTEX: ARM POSTURE OR GAZE DIRECTION? Boussaoud et al. (Boussaoud and Bremmer 1999; Boussaoud et al. 1998; Jouffrais et al. 1999) have reported that the strength and directionality of arm movement-related cell activity in PMd is modulated by gaze direction and suggested that the directional shifts reported by Caminiti et al. (1990, 1991) might be due in part or entirely to the gaze shifts that occurred while the monkey made arm movements in the three adjacent target cubes. However, Scott and Kalaska (Scott and Kalaska 1997; Scott et al. 1997) found that single-cell tuning in MI and PMd often changed while monkeys made arm movements to targets in the same spatial locations using two different arm postures. In the present study, rotations of cell PDs occurred with hand location in the workspace that were of the same magnitude as that in Caminiti et al. (1990, 1991) at both single-cell and population levels even though the desired force targets and visual feedback of net force output were presented in the same positions on a fixed computer monitor, independent of hand location. Therefore global gaze shifts are not necessary to produce changes in single-cell tuning coupled to changes in arm posture during both reaching movements and isometric force generation. Arm posture changes alone are sufficient. This is consistent with modeling studies that proprioceptive feedback about arm posture can provide sufficient information to resolve the visuomotor transformations between extrinsic and intrinsic movement parameters (Baraduc et al. 2001; Burnod et al. 1999).

Changes in response intensity

Many MI cells showed a systematic change in the level of task-related activity with arm posture that frequently took the form of a planar function of hand location. This was seen for both the overall grand mean of task-related activity across all directions of force output and for the activity related with each direction of force output. A smaller number of cells also showed a systematic planar change in the dynamic range of force direction-related changes in activity in different hand locations. Changes in the level of task-related activity independent of any directional changes represent another mechanism by which MI can contribute to the transformations required to compensate for posture-related changes in the coupling between extrinsic spatial and intrinsic movement parameters (Baraduc et al. 2001; Pouget and Sejnowski 1997; Salinas and Abbott 1995; Zipser and Andersen 1988).

Temporal gradient of posture-related discharge modulation

The effects of arm posture on MI activity, especially its directionality, were less prominent at the single-cell and population levels during RT than during DFT and THT. In particular, both Kakei et al. (1999) and the present study found evidence of the transient expression of a signal by a subset of cells in MI that appears to be coupled more closely to the extrinsic spatial attributes of the task than to the mechanics of its execution. A similar feature of MI activity has been found during the RT in many other studies (Georgopoulos et al. 1989, 1992; Scott and Kalaska 1996; Shen and Alexander 1997a,b; Taira et al. 1996; Zhang et al. 1997). The changes in the apparent coupling of MI activity to motor output parameters may also reflect the neuronal mechanisms that resolve the transformation from extrinsic to intrinsic reference frames (Johnson et al. 2001).

These findings are also consistent with evidence that the earliest part of the motor output signal is only an approximation of the exact motor command needed to successfully perform an action (Desmurget and Grafton 2000; Ghilardi et al. 1995; Gordon et al. 1994; Karst and Hasan 1991a,b; Massey et al. 1991; Soechting and Flanders 1989; Smyrnis et al. 2000). The existence of these early motor output errors has implications for the computational architecture of the motor system. It suggests that the motor system either does not attempt or is not capable of generating the perfect solution to the putative inverse kinematic and inverse dynamic transformations necessary to determine the correct muscle activity prior to initiation of the motor output. This could result from various sources of errors such as in estimation of initial arm position, target location, or limb dynamics. Peripheral feedback during generation of the isometric forces could help correct for these initial errors and cause the increased prominence of arm posture-related effects on cell activity seen between the RT and later trial epochs. Feedback signals could come from proprioceptors in the arm as well as visual inputs about cursor motion on the screen.

Of course, the initial output errors described in those studies do not necessarily reflect an absolute performance limit of the motor system. In most cases, the sole behavioral constraint related to endpoint accuracy with no control imposed on initial outputs. Likewise the postural effect in this study may have become more prominent after force output onset as behavioral constraints on force directionality became more critical. It would seem important practically and theoretically to devise experiments to define the limits on the accuracy of initial outputs.

Despite the potential role of feedback on these hand location effects, it is also clear that some cells showed significant directional shifts at different hand locations even prior to force ramp onset (Fig. 9A). This indicates that some MI cells can generate predictive feed-forward signals that can take arm geometry into account prior to motor onset to generate appropriate outgoing motor commands. Human behavioral studies suggest that this capacity of the motor system may result from adaptive "internal inverse models" of the biomechanical properties of the peripheral musculoskeletal system and of the mechanical properties of the environment with which it is interacting (Bhushan and Shadmehr 1999; Conditt et al. 1997; Flanagan and Lolley 2001; Flanagan and Wing 1997; Ghilardi et al. 1995; Jordan 1990; Kawato 1999; Krakhauer et al. 1999; Sabes 2000; Shadmehr and Mussa-Ivaldi 1994; Thoroughman and Shadmehr 1999; Wolpert and Gharamani 2000; Wolpert et al. 1995). Behavioral evidence further suggests that the computational elements of the internal models of task dynamics function in an intrinsic (joint- or muscle-centered) coordinate system (Bhushan and Shadmehr 1999; Krakhauer et al. 1999; Shadmehr and Moussavi 2000; Shadmehr and Mussa-Ivaldi 1994; Thoroughman and Shadmehr 1997). The presence of arm geometry-dependent directional shifts in MI during RT is consistent with this prediction from the behavioral evidence and would imply that MI receives signals from or even contributes to the putative internal inverse model for task dynamics.

Anisotropies of posture-related response modulations in MI

The net isometric force vectors generated at the hand were evenly distributed at 45° intervals around the circle and were of uniform length across all hand locations. Despite this uniformity of motor output, there were a number of nonuniform anisotropies in the patterns of MI activity at both the single-cell and population level that varied as a function of arm posture.

These nonuniformities likely reflect part of a central compensatory mechanism for the anisotropic mechanical properties of the limb. The limb tends to be stiffer along an axis between the hand and shoulder and more compliant along the orthogonal axis (Flash and Mussa-Ivaldi 1990; Mussa-Ivaldi et al. 1985). As a result, the spatial orientation of the stiffness ellipse changes systematically in an arc-like manner centered on the shoulder as the hand sweeps across the horizontal plane, and the degree of eccentricity of the stiffness ellipse decreases as the hand is held closer to the body (Flash and Mussa-Ivaldi 1990; Mussa-Ivaldi et al. 1985). Both of these trends were paralleled in this study by the patterns of anisotropies of MI activity. The PD shifts varied systematically with hand location in the same manner as the spatial orientations of the stiffness ellipses. Overall levels of discharge across force directions and the dynamic ranges of cell activity for different force directions tended to be greater when the arm was extended out away from the body even though the force levels were constant at all hand locations. These changes in activity level tended to vary in a planar fashion with hand location, and the orientations of the planes were nonuniform, tending to cluster along an axis that deviated away from the midline toward the shoulder of the arm being used to generate the forces. Finally, the anisotropic distributions of planes for cells in the two hemispheres were mirror inverted across the body midline, further supporting their relation to the geometry-dependent anisotropies of arm mechanics (Scott and Kalaska 1997). These effects could be seen to differing degrees at all times in the trial.

Limitations on interpretation

The cell sample in this study was drawn primarily from the caudal part of MI forming the bank of the central sulcus where previous studies have found that cells were more strongly modified by external loads and more unconditionally coupled to the execution of reaching movements than cells in more rostral parts of MI (Crammond and Kalaska 1996, 2000; Kalaska et al. 1989). Every attempt was made to record from cells across all cortical layers that met the basic criteria of a relationship to proximal-arm movement and directional tuning. We did not reject cells that showed little modulation of activity at different hand locations. Indeed, many of the cells were initially isolated and studied in an arm-movement task (Sergio and Kalaska 1998) before evaluating their activity in the isometric task. However, because of the requirement to isolate cells for long periods of time while testing them sequentially at different hand locations, there was an inevitable bias toward a greater probability of successful testing of all nine hand locations for the larger cells concentrated in intermediate layers of MI. The properties of cells with only partial data sets, including cells located outside of the large-cell layer, did not appear to be significantly different from those cells with complete data sets. Nevertheless it must be acknowledged that cell populations located in more rostral parts of MI or in the superficial layers of the cortex within the central sulcus may behave differently from the sample described here.

This study shows a number of parallels between some properties of the sampled MI cell activity, muscle activity, and limb mechanics. This does not imply, however, that all MI cells represent movement in a muscle-centered coordinate system or generate signals that exclusively and explicitly represent forces or muscle activity. For instance, whereas prime-mover muscles in this task typically showed ramp-like increases in activation during force generation (Fig. 5) (Sergio and Kalaska 1998), many MI cells showed an initial phasic response component that did not correlate with any initial component of the time profile of force output or EMG activation (Fig. 6) (Sergio and Kalaska 1998). Furthermore the findings in this study do not preclude the existence in MI of representations of higher-order aspects of movement planning independent of causal task dynamics and muscle activity. The findings do, however, provide further compelling support for the hypothesis that the primary motor cortex is implicated in the transformation from a more abstract or global representation to a more causal representation of motor outputs (Scott and Kalaska 1997; Shen and Alexander 1997a,b; Zhang et al. 1997).