INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Several cortical motor areas have been identified on the medial wall of the hemisphere in monkeys (for review, see Strick et al. 1998). These include a group of premotor regions located in the cingulate sulcus collectively termed the cingulate motor areas (CMA) The rostral cingulate motor area (CMAr) is located within the dorsal and ventral bank of the cingulate sulcus primarily rostral to the genu of the arcuate sulcus. Two additional cingulate motor areas are located caudal to CMAr on the dorsal (CMAd) and ventral (CMAv) banks of the cingulate sulcus. Each area is confined to separate cytoarchitectonic fields and has direct projections to the spinal cord (Biber et al. 1978; Dum and Strick 1991a, 1996; Galea and Darian-Smith 1994; He et al. 1993, 1995; Hutchins et al. 1988; Keizer and Kuypers 1989; Macpherson et al. 1982; Morecraft et al. 1997; Murray and Coulter 1981; Nudo and Masterton 1990; Toyoshima and Sakai 1982). Each area is also reciprocally connected to the premotor cortex (Barbas and Pandya 1987; Deacon 1992; Ghosh and Gattera 1995; Godschalk et al. 1984; Kunzle 1978; Kurata 1991; Matelli et al. 1986; Morecraft and Van Hoesen 1993) and primary motor cortex (Dum and Strick 1991a; Leichnetz 1986; Morecraft et al. 1997; Morecraft and Van Hoesen 1992; Nimchinsky et al. 1996).

Physiological evidence indicates a functional linkage between the anatomical connections of CMA with the motor system and limb-movement generation. Studies using positron emission tomography in humans indicate an increase in regional cerebral blood flow at foci in cingulate cortex related to pointing (Grafton et al. 1996; Lacquaniti et al. 1997), reaching, and grasping (Grafton et al. 1996), finger movements (Larsson et al. 1996; Paus et al. 1993; Wessel et al. 1995, 1997), and motor-sequence learning (Grafton et al. 1998). Similar results were found using functional magnetic resonance imaging during visually triggered or self-paced finger movements (Deiber et al. 1999) and the exertion of finger force (Dettmers et al. 1995). Using 2-deoxyglucose as a metabolic marker, overall activity of CMA in monkeys increased during performance of a remembered sequence of reaching movements (Picard and Strick 1997). Finally, electrical microstimulation at specific foci within cingulate cortex evokes movements (Godschalk et al. 1995; Luppino et al. 1991; Mitz and Godschalk 1989; Mitz and Wise 1987).

Although these observations are consistent with CMA involvement in the generation of limb movements, the fundamental question of how these cortical areas process visuomotor information can only be addressed by studying the activity of single neurons in awake animals. However, there are few published reports of single neuron activity in CMA during visually guided limb movements. Niki and Watanabe (1976) found cingulate neurons in the vicinity of CMAr that were active during a delayed-response task. In studies that targeted CMA, responses to pressing a key by flexing the digits (Shima et al. 1991) or the production of grip and lifting forces (Cadoret and Smith 1997) have also been described. Furthermore, neural activity in CMA has also been studied in the context of bimanual coordination (Kermadi et al. 2000) and reward probability (Shima and Tanji 1998). Although these studies have demonstrated the presence of neural activity during motor behavior and explored some of the necessary and sufficient conditions for observing it, no study has specifically investigated how different sensorimotor variables are represented in CMA.

To further elucidate the role of CMA in visuomotor control, we studied the activity of single neurons in CMAd and CMAv during a task requiring two-dimensional proximal arm movements. The task contained discrete cue, delay, and movement periods and tested different directions of movement to investigate how the activity of single neurons encode these task parameters during each epoch of the task. We also tested the hypothesis that CMAd and CMAv have unique functions in motor control by directly comparing activity in these areas to the activity of neurons in the supplementary motor areas (SMA) under identical visuomotor conditions in the same monkeys. We found directional neural activity associated with the visual stimulus, movement preparation, or movement execution in both CMAd and CMAv in different combinations within single neurons. There were some significant differences in movement-related activity across areas, suggesting that SMA has a more prominent, albeit not exclusive, role in movement initiation, whereas CMAd and CMAv have a more central role in movement execution. However, most aspects of neural activity were indistinguishable across all three motor areas, indicating that CMAd, CMAv, and SMA utilize many of the same fundamental neurophysiological mechanisms for controlling movements. A preliminary account of these results has been reported (Backus et al. 2001).

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Surgical and behavioral protocols were approved by the Institutional Animal Care and Use Committee and complied with United States Public Health Service policy on the humane care and use of laboratory animals.

Behavioral methods

Two female pigtail monkeys (Macaca nemestrina, 5.0 and 6.2 kg) were used in this study. The monkeys sat in a primate chair with their right hand grasping the top of a joystick that was moveable in two dimensions in the horizontal plane. Indirect visual feedback of joystick position was provided by a small (3 mm) white spot on a computer monitor. Output from two precision potentiometers coupled to the joystick provided a record of arm position. A neutral arm position with the forearm horizontal, shoulder adducted, and the elbow at ~90° corresponded to a cursor position at the center of the monitor. During the course of training and experimentation, the monkeys did not have free access to food in their home cage. Instead, they obtained food to satiety each day while in the primate chair by performing a task wherein each correct trial was rewarded by ~0.5 ml of monkey chow blended with water and sugar. Training and experimental sessions usually lasted for 2-4 h. When the monkeys ceased to initiate trials, they were returned to the home cage and given fruit and ad libitum water. Body weight, intake of food, and other indicators of health were closely monitored throughout the training and experimentation period.

Two interconnected computers controlled the visual stimuli, rewarded the monkeys for correct behavior, and collected behavioral and neural data at 1-ms intervals. Visual stimuli were light-gray squares (1.6 ×1.6 cm) presented on a 37 × 28.5-cm high-resolution computer monitor 29 cm from the monkey's eyes. A solid-state video camera was used to monitor the monkeys' gross behavior during experimental sessions. The monkeys were trained to perform a visually guided instructed delay task by moving the cursor with the joystick using combined elbow and shoulder movements. Figure 1 shows a diagram of the visual display during performance of one correct trial and one trial of behavioral data from each monkey. A centrally located visual target was illuminated and the monkey initiated a trial by aligning the cursor over it. Immediately after "capturing" this center target, either four (monkey M) or eight (monkey A) peripheral targets equally spaced around and 6 cm away from the center target simultaneously appeared. During the ensuing preinstruction control period, the monkey was required to hold the cursor over the center target for 1-2 s. An instruction cue was then presented by brightening one of the peripheral targets for 0.5 s and then returning it to its previous level of illumination. The monkey was required to continue holding the cursor over the center target during the instruction cue and afterward during a 1- to 2-s postinstruction delay period. The center target was then turned off, serving as a GO signal that directed the monkey to move the cursor to the previously cued peripheral target within 1 s and hold it there for 0.5-1 s. Movement of the cursor from the center target to a peripheral target required moving the joystick ~3.4 cm. Targets were selected as the instruction cue in a pseudorandom order.

View larger version (19K): [in this window] [in a new window]   Fig. 1. Behavioral paradigm. Top left: cartoon showing monkey's typical posture at the beginning of each trial. Top right: sequence of visual stimuli presented on the monkey's monitor during different times in 1 trial. All targets were light gray; the instruction cue was an increase in the illumination of 1 target for 0.5 s. Note that 4 peripheral targets were used with monkey M (not shown). Middle: behavioral data from 1 trial by monkey M. There were no oculomotor requirements during the task. Timing of central target, peripheral targets, peripheral cue, and reward delivery are shown as gray horizontal bars. Numbered arrows and horizontal bracket indicate correspondence between timing of task events and illustrations of monitor appearance above. Hlimb, horizontal arm position; Vlimb, vertical arm position, |Vellimb|, absolute arm velocity; Heye, horizontal eye position; Veye, vertical eye position. Bottom: behavioral data from 1 trial by monkey A. This monkey was required to simultaneously fixate the center target while holding the cursor over it during presentation of the instruction cue and ensuing delay period. All other aspects of the task were identical for both monkeys.

Eye-position coordinates were obtained with a high-speed infrared pupil tracking system (ISCAN). There were no specific oculomotor requirements during the task with monkey M. Analysis of this monkey's oculomotor behavior, however, showed a recurring pattern of saccadic eye movements (Fig. 1, middle). After moving the cursor over the central target, the monkey usually made one or more saccadic eye movements between the peripheral targets. After presentation of the instruction cue, the monkey usually made a saccade to the central target and fixated it until its disappearance (GO signal). After the GO signal, a saccade was usually made to the correct peripheral target around the same time the monkey began moving the cursor toward the target by moving the joystick. To better control oculomotor behavior, the second monkey studied (monkey A) was required to simultaneously fixate the central target and hold the cursor over it during the entire precue, cue, and postcue periods (Fig. 1, bottom). After the GO signal, the fixation requirement was terminated, and a saccade to the cued peripheral target was usually made as the monkey moved the cursor over it.

Surgical procedures

After an initial training period, each monkey was prepared for chronic single-neuron recording in an aseptic surgical procedure under anesthesia initially induced by a mixture of ketamine and xylazine followed by isoflourane. With the head held in a primate stereotaxic instrument, a craniotomy was trephined in the skull, and a recording chamber with a removable cap was placed over the craniotomy. The craniotomy was centered 3 mm from the vertex of the skull and 20 mm anterior to the interaural line, allowing access to the medial wall of the left hemisphere where SMA was located and to both the dorsal and ventral banks of the cingulate sulcus where CMAd and CMAv were located. Stainless steel screws were secured to the skull, and the entire implant system consisting of screws, recording chamber, and receptacles for the monkey chair's head-holder was bonded together with dental acrylic.

During a 2-wk postoperative recovery period, the monkeys were given food and water ad libitum, analgesics, and extra fruit. Prophylactic antibiotics were given preoperatively and continued postoperatively for 10 days.

Recording procedures

During experimental sessions, the monkey sat in the primate chair with its head held stationary by the restraining device fixed to its skull. Neural activity was recorded from the left hemisphere with microelectrodes made from glass-coated platinum/iridium wire (tip exposures: 10-30 µm) advanced through the intact dura with a hydraulic microdrive (MO-95, Narishige) mounted on the recording chamber. The minimum penetration spacing across the cortical surface was 0.5 mm. Action potentials from single neurons were sorted on-line by a template-matching system (Alpha Omega Engineering). We randomly switched between SMA, CMAd, and CMAv across recording sessions as experimentation with each monkey progressed.

After sampling neural activity during task performance, the responses to somatosensory and proprioceptive stimuli were assessed outside the behavioral paradigm by manual manipulation of the leg, arm, face, and trunk while listening to audio feedback from the recording amplifier. Stimuli consisted of passive joint rotation, muscle palpation, tendon taps, and cutaneous stimulation. Neural activity in response to active limb and eye movements was also assessed.

Microstimulation through the recording electrode was used to help identify the "arm" regions of SMA, CMAd, and CMAv. Stimulating currents were <100 µA. Stimulation consisted of 100-ms trains of 333-Hz biphasic (negative-positive) shocks, with each negative and positive phase having a duration of 0.2 ms and separated by 0.1 ms (0.5 ms per shock). Microstimulation was applied outside the context of the task while the monkey was calm and relaxed. The threshold to elicit movements from a cortical site was defined as the magnitude of negative-going current necessary to elicit movements on ~50% of trials. Microstimulation was also carried out systematically at 0.5-mm intervals along most penetrations, irrespective of the locations of task-related activity. This was done as the microelectrode was being withdrawn from each track after single-neuron recording was completed.

In separate sessions that followed the period of single-neuron recording, electromyographic (EMG) activity was recorded differentially from Teflon-insulated stainless steel wires inserted percutaneously. The EMG activity was amplified, filtered (0.1- to 2-kHz passband), rectified, and processed by a sample-and-hold integrator (Bak Electronics). EMG activity was recorded from the following muscles: triceps brachii (long), pectoralis major, deltoideus (acromiodeltoideus), deltoideus (cleidodeltoideus), deltoideus (spinodeltoideus), dorsoepitrochlearis, infraspinatus, latissimus dorsi, supraspinatus, teres major, trapezius (caudal portion), trapezius (cranial portion), rhomboideus (pars dorsi), spinalis cervicis, spinalis dorsi, extensor carpi radialis, flexor carpi radialis, flexor carpi ulnaris, palmaris longus, biceps brachii (long), biceps brachii (short), brachialis, brachioradialis, atlantoscapularis posterior, flexor digitorum profundus, extensor carpi ulnaris, extensor digitorum communis.

Data analysis

Rasters and spike-density functions of neural activity aligned to various task and behavioral events were both visually assessed and subjected to statistical analyses. Movement onsets, offsets, and peak velocities were computed using movement velocity signals obtained by digital differentiation of horizontal and vertical joystick position signals.

We use the terminology of Weinrich and Wise (1982) to describe the basic patterns of task-related neural activity. Phasic increases in activity after the onset of the instruction cue were defined as signal activity, tonic activity after the instruction cue that persisted until the GO signal was defined as set activity, and increases in activity prior to and/or during the monkey's arm movement was defined as movement activity.

The onset and offset times of signal, set and movement activity were found using an adaptation of the burst detection algorithm originally described by Legendy and Salcman (1985) and later modified by Hanes et al. (1995). This algorithm iteratively evaluated the probability that the number of action potentials within successive time intervals could have occurred by chance by comparing the actual number of spikes to the number of spikes predicted by a Poisson distribution. Correct trials were grouped according to peripheral target direction and the neural activity from these trials was aligned to either the onset of the instruction cue or the beginning of limb movement. The spike trains from each group of trials were then collapsed into one spike train for each group and scanned for epochs where the number of action potentials were increasingly improbable compared with the mean firing rate during a control period (see following text). Putative responses found in the collapsed spike train (P  0.01) were verified across the individual trials by testing for a significant increase in the mean discharge rate during the putative response period compared with the mean discharge rate during the control period using a one-tailed paired t-test (P  0.05). Only increases in discharge rate that fulfilled both statistical criteria were accepted as neural responses. If two or more target directions yielded significant responses, the target yielding the largest response was chosen as the optimal target and the onset and offset time of that response was taken as the neuron's characteristic response time. Neural responses characterized by a decrease in discharge frequency were not analyzed.

Signal activity and the time of its occurrence was found by aligning neural activity to the onset of the instruction cue, computing the mean discharge rate during a 1-s preinstruction control period, and scanning for a response beginning at the onset of the cue until the earliest GO signal using the algorithm described in the preceding text. Movement activity and the time of its occurrence was found by aligning activity to the beginning of the limb movement, computing the mean discharge rate during the 500 ms before the GO signal, and scanning for responses from 200 ms before the onset of movement until the earliest movement end time. Because of its distinctively tonic nature, set activity was uniformly defined as any significant elevation in mean firing rate 750 ms before the GO signal compared with the preinstruction control period using a one-tailed paired t-test (P  0.05). If this test indicated the presence of set activity, its onset time was estimated by finding the time of the first action potential after the onset of the cue where successive action potentials from that point forward was consistently higher than what would be predicted by a Poisson distribution compared with the preinstruction control period. The criterion distinguishing between signal and set activity was that signal activity had a statistically significant decrease in firing rate before the GO signal, whereas set activity had a continuously elevated mean firing rate before the GO signal.

The mean discharge rates during signal, set, and movement response epochs were computed on a trial-by-trial basis and subjected to further analysis. Responses during the signal and movement epochs were taken as the mean discharge rate between the response onset and offset time that was found for the optimal target direction and calculating the mean discharge rate in that epoch for all target directions. When necessary, we limited the epoch used to compute movement activity to the movement end time in trials where the monkey captured the target and ceased moving before the neuron's movement response ended so that movement responses did not include activity that occurred after the monkey stopped moving. Set activity was always computed as the mean discharge rate 750 ms before the GO signal.

A one-way ANOVA (P  0.05) was used to determine whether neural responses exhibited a significant directional bias. Neural activity optimal direction (), defined as the hypothetical target direction yielding the largest response, was estimated for activities with a significant ANOVA by calculating the circular mean angle of the target vectors weighted by the neural responses (Batschelet 1981). A bootstrapping procedure was used to determine the statistical certainty of  (Fisher 1993). The set of target directions used to test each neuron was randomly assigned a discharge rate taken from the original set of responses (with replacement) and the mean vector length was calculated. The length of the neuron's actual mean vector was then compared with the distribution of 10,000 bootstrap mean vectors. The calculated  was classified as significant if <500 bootstrap mean vectors exceeded the length of the neuron's actual mean vector (~P < 0.05). Multiple activities within the same neuron were analyzed separately, with their optimal direction designated _signal when signal activity was analyzed, _set when set activity was analyzed, and as _move when movement activity was analyzed.

Neurons exhibiting activity with a significant directional bias in the ANOVA were also fit to a sinusoidal function of the form
where f() was discharge frequency,  was target direction,  estimates the neuron's optimal direction, v is an index of the neuron's tuning with respect to direction, B estimates its median firing rate, and R estimates its peak response magnitude (Batschelet 1981 p. 173). The parameter v (limited to the interval 60°  v  60°) allows the function to accommodate more narrowly or broadly tuned directional variation than a simple cosine function. Note that the equation is reduced to a simple cosine function when v = 0. The parameter estimates were obtained using the Gauss-Newton method by the NLINFIT function in MATLAB (Mathworks). The statistical certainty of the overall fit was assessed using a variance ratio (F) test. Tuning width () was defined as the distance between consecutive intersections of the function where y equals the midpoint between the minimum and maximum of the function.

The circular-circular correlation (r_aa) between optimal directions (e.g., _set vs. _move) was calculated using the formula of Fisher and Lee (Fisher 1993 p. 151). We used the convention of measuring angles counterclockwise from the direction directly rightward (0°).

Histology

After the final experimental session, the monkeys were anesthetized with ketamine and fiducial marks were made by inserting pins into the brain at known microdrive coordinates. The monkeys were then deeply anesthetized with pentobarbital and perfused transcardially with normal saline followed by 10% neutral formalin. The surface of the brain was photographed in situ with the dura reflected and pins in place. The pins were then removed and each brain was blocked, frozen, cut into 40-µm coronal sections, and stained with cresyl violet. The locations of recorded neurons were reconstructed using the fiducial marks, gliosis from electrode tracks, and depths where electrophysiological transitions were noted as the electrode was advanced through the cortical folds.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Task performance

We compared neural activity in the CMAd, CMAv, and SMA under identical task conditions. Because recordings took place over a period of several months in both monkeys, we attempted to minimize the possibility of differences across motor areas due to gradual changes in behavioral performance by randomly switching between motor areas between recording sessions as experimentation with each monkey progressed. On completion of recording, we analyzed each monkey's behavior and found similar performance profiles while recording in each area. Percent correct performance (all initiated trials that were rewarded) in monkey M were similar for CMAd, CMAv, and SMA recordings (88, 89, and 88%) as were the mean and SD of the reaction times (203 ± 26, 204 ± 25, and 199 ± 23 ms) and movement durations (347 ± 75, 363 ± 73, and 356 ± 69 ms). Monkey A exhibited a similar percent correct performance during CMAd, CMAv, and SMA recordings (88, 87, and 89%), reaction times (254 ± 106, 243 ± 88, and 245 ± 89 ms), and movement durations (448 ± 117, 445 ± 107, and 447 ± 113 ms).

Localization of recorded neurons

Task-related activity from 180 neurons in the CMAd, 47 neurons in the CMAv, and 209 neurons in SMA are the subject of this report. Figure 2 shows the entry points of electrode penetrations on the dorsal surface of each monkey's brain. The different symbols indicate to which motor area neurons recorded at those locations were ultimately assigned. Most of the anteroposterior extent of CMAd, CMAv, and SMA was explored. The pre-SMA and CMAr were not sampled. The locations of CMAd, CMAv, and SMA were in close agreement with previous anatomical (Dum and Strick 1991a, 1996; He et al. 1995; Hutchins et al. 1988; Luppino et al. 1990; Morecraft and Van Hoesen 1992, 1993; Morecraft et al. 1997; Wang et al. 2001), microstimulation (Luppino et al. 1991; Mitz and Wise 1987), and recording (Cadoret and Smith 1995, 1997; Shima et al. 1991) studies. All three motor areas were characterized by robust task-related activity and responses to somatosensory stimuli or passive movement of the contralateral arm. Muscle contractions or arm movements were often evoked with intracortical microstimulation using 15-100 µA of current. SMA recordings were located in the medial portion of area 6 at depths of 1-5 mm from the dorsal surface and extending rostrocaudally from just rostral to the genu of the arcuate sulcus to ~1-1.5 cm anterior to the central sulcus, placing them in the subdivision of SMA termed SMA proper (Luppino et al. 1991; Matsuzaka et al. 1992). CMAd and CMAv recordings were located in the cingulate sulcus at approximately the same rostrocaudal level as SMA (Dum and Strick 1993). We ultimately sampled fewer CMAv neurons because CMAd was encountered first, and the monkeys were often satisfied after the many CMAd recording trials and stopped working soon after the electrode tip reached CMAv.

View larger version (19K): [in this window] [in a new window]   Fig. 2. Dorsal views of left frontal lobes of monkeys M and A showing the location of electrode entry points that yielded single-neuron data analyzed in this report. Inset: magnified view of area studied. , penetrations where CMAd neurons were recorded. +, penetrations where CMAv neurons were recorded. ×, penetrations where SMA neurons were recorded. , penetrations where task-related activity was not recorded. , rostral-caudal location of the histological section shown in Fig. 3.

Because of their close proximity, a combination of physiological and anatomical criteria were used to determine whether individual neurons were located in CMAd, CMAv, or SMA along each electrode penetration. In addition to the surface maps shown in Fig. 2, the depths where neural activity was recorded and the somatotopy of sensory responses and electrically evoked movements were plotted in the coronal plane and compared with the histologically prepared tissue sections. Visually identified electrode tracks and the location of fiducial marks from the pins inserted before perfusion were used to align the physiological maps with the histological sections to reconstruct the location of recorded neurons. Figure 3 shows the reconstruction of recording sites in the coronal plane 4 mm caudal from the center of the recording chamber in monkey M. CMAd is confined to the subdivision of area 6 located in the dorsal bank of the cingulate sulcus (area 6c) and CMAv is confined to the ventral bank of the cingulate sulcus corresponding to area 23c (Dum and Strick 1991a; Matelli et al. 1991; Vogt 1993; Vogt et al. 1987). Because the cytoarchitecture in the fundus of the cingulate sulcus was distorted due to folding, we were unable to precisely locate the border between areas 6a and 23c. We placed the dividing line between CMAd and CMAv slightly ventral to the sulcus because our microstimulation and somatosensory mapping indicated an abrupt transition in somatotopy there, and it coincides more closely with the boundary of a zone receiving prefrontal projections (Barbas and Pandya 1987; Bates and Goldman-Rakic 1993; Lu et al. 1994; McGuire et al. 1991; Morecraft and Van Hoesen 1993).

View larger version (49K): [in this window] [in a new window]   Fig. 3. Reconstruction of electrode recording sites. Left: photograph of histological section from the left frontal lobe of monkey M at the anterior-posterior level denoted by the dashed line in Fig. 2. Right: reconstruction of electrode recording sites. , location of recorded neurons.

In this study, we analyzed only neurons with task-related responses to proximal arm movements according to the following physiological criteria: activity was modulated during whole arm movements outside the context of the task, activity was modulated during passive elbow and/or shoulder movements, or palpation of the muscles or cutaneous stimulation around those joints, there was evidence of movement or signs of muscle contractions around the elbow and/or shoulder region in response to microstimulation (<100 µA) through the recording electrode at the site of the neuron, and physiological data from other recordings at the same or closely neighboring sites confirmed the neuron had been located within a local region representing the elbow and/or shoulder. When the consensus of these criteria indicated that the neuron was related to proximal arm movements, it was selected for further analysis.

Response characteristics

The instructed-delay task we used can be divided into three distinct periods: an instruction period, a postinstruction delay period, and a movement period. During the instruction period, a visual cue provided information about the spatial location of a target that was to be the goal of a subsequent motor action. The monkey had to remember this information during the postinstruction delay period, during which a motor plan may be formulated. Finally, a GO signal marked the beginning of a movement period during which neural activity directed an appropriate motor action. Neural activity within each task period (signal, set, and movement activity) were assessed to determine the extent to which target, preparatory, and movement variables were represented in the discharge of single neurons in CMAd, CMAv, and SMA.

All three types of task-related activities were found in different combinations in all three motor areas. Figure 4 shows one example each of signal, set, and movement activity in CMAd, CMAv, and SMA. The neural activity in the left and middle columns is aligned to the onset of the instruction cue, and the activity in the right column is aligned to the beginning of movement. The activity of each of the nine different neurons shown in this figure was only from trials with the target direction that gave the strongest response for that neuron. Signal activity (Fig. 4, left) was characterized by an increase in firing rate after the onset of the instruction cue followed by a decrease in activity before the onset of the GO signal. Set activity (Fig. 4, middle) was usually either an abrupt increase in firing rate time locked to the onset of the cue that persisted or continued to increase during the delay period, or a gradual rise in firing rate during the delay interval with an onset time that varied between neurons. None of the neurons in the middle column of Fig. 4 were deemed to have signal activity because their response persisted or continued to increase until the GO signal without significantly decreasing. Movement activity (Fig. 4, right) was characterized by an abrupt increase in firing rate associated with the movement.

View larger version (38K): [in this window] [in a new window]   Fig. 4. Examples of signal, set, and movement activity in the dorsal and ventral cingulate motor areas (CMAd and CMAv) and supplementary motor area (SMA). Rasters and spike density functions of neural activity aligned on the onset of the cue (left and middle) or movement (right) from neurons located in CMAd (top), CMAv (middle), and SMA (bottom). Each vertical tick represents the occurrence of one action potential, and each row of ticks represents the neuronal activity recorded during 1 trial. Spike-density functions were obtained by collapsing the spike trains of all trials into a single spike train, counting the action potentials in consecutive 1-ms time bins, dividing by the number of trials, and then convolving with a Gaussian function having a SD of 20 ms. , the onset and offset times of neural responses (set activity persists until the GO signal and thus only has an onset). Hlimb, horizontal limb position. Vlimb, vertical limb position. |Vellimb|, absolute limb velocity. , center target offset (GO signal).

The proportion of neurons with signal, set, and/or movement activity were similar in each of the three motor areas studied. The top left of Fig. 5 shows the percentage of neurons in CMAd, CMAv, and SMA with each type of activity independent of whether they exhibited any other activity. Of the 180 CMAd neurons, 51 (28%) exhibited signal activity, 83 (46%) exhibited set activity, and 155 (86%) exhibited movement activity. Of the 47 neurons recorded in CMAv, 14 (30%) exhibited signal activity, 22 (47%) exhibited set activity, and 37 (79%) exhibited movement activity. Of the 209 SMA neurons recorded, 61 (29%), 127 (61%), and 178 (85%) exhibited signal, set and movement activity, respectively. The frequency of signal, set and movement activities were not significantly different across motor areas (²_[4] = 2.94, P > 0.5).

View larger version (41K): [in this window] [in a new window]   Fig. 5. Proportions of neurons with signal, set, and movement activity in the CMAd, CMAv, and SMA. Bar chart shows the percentages of total neurons in CMAd, CMAv, and SMA with signal, set, or movement activity. Pie charts show the proportion of neurons with each possible combination of activity in each motor area. Notice that CMAd and CMAv have a larger proportion of neurons with only movement activity compared with SMA.

Although the neurons shown in Fig. 4 exhibited only one type of activity, most neurons exhibited different combinations of these pure types. The pie charts in Fig. 5 show the proportions of each possible combination of activity in our sample population of CMAd, CMAv, and SMA neurons. The proportions of each activity combination were not significantly different across motor areas (²_[12] = 19.28, P > 0.08). However, visual inspection of these charts suggests a higher proportion of neurons with only movement activity in CMAd and CMAv compared with SMA. Conversely, there appears to be larger proportion of neurons in SMA with both set and movement activity. Configuring the contingency table into frequency of neurons with only movement activity versus other combinations reveals a significant difference across motor areas (²_[2] = 7.63, P < 0.03), indicating that neurons with purely movement activity are more prevalent in CMAd and CMAv than in SMA.

Time course and magnitude of responses

RESPONSE LATENCY. The distribution of onset times for each activity in each motor area are shown in Fig. 6. The histograms in the Fig. 6, left, shows the onset latency of signal activity in CMAd, CMAv, and SMA relative to the time when the cue first appeared. Onset latencies ranged between 45-630 ms (mean: 204 ± 119 ms) in CMAd, 47-390 ms (mean: 196 ± 103 ms), in CMAv, and 60-587 (mean: 224 ± 140 ms) in SMA. A Kruskal-Wallis test failed to indicate a difference in mean latency across motor areas (²_[2] = 0.13, P > 0.9).

View larger version (20K): [in this window] [in a new window]   Fig. 6. Distributions of onset latencies for signal, set, and movement activity of the population of CMAd, CMAv, and SMA neurons sampled. Histogram binwidth is 50 ms for signal and set activity, 100 ms for movement activity. , mean. Notice that the overall distributions of latencies are similar across motor areas even for movement activity where the mean onsets are statistically different.

RESPONSE MAGNITUDE. Neural response magnitudes were on average largest in SMA and smallest in CMAd, although the response magnitude of individual neurons within a motor area varied widely. Figure 7 shows the signal, set, and movement activity magnitudes in CMAd, CMAv, and SMA. This response parameter was measured by taking the mean spike rate during the response epoch for each trial and averaging across all trials in the neuron's optimal target direction. The mean signal response in CMAd (23 ± 14), CMAv (30 ± 13), and SMA (31 ± 21) were similar, although a Kruskal-Wallis test indicated they were significantly different (²_[2] = 6.7, P < 0.04). The mean set response in CMAd (19 ± 14), CMAv (22 ± 14), and SMA (26 ± 18) were significantly different across areas (Kruskal-Wallis ²_[2] = 9.0, P < 0.02), and the difference in mean movement response in CMAd (34 ± 22), CMAv (39 ± 23), and SMA (48 ± 30) was highly significant (Kruskal-Wallis ²_[2] = 24.0, P < 0.0001). Control period activity in CMAd (11 ± 10 spikes/s), CMAv (13 ± 11 spikes/s), and SMA (12 ± 11 spikes/s) was not significantly different across areas (Kruskal-Wallis ²_[2] = 0.97, P > 0.6). Thus the strength of signal and set activity was slightly higher in SMA, whereas movement activity was clearly strongest in SMA and weakest in CMAd.

View larger version (18K): [in this window] [in a new window]   Fig. 7. Response magnitudes of signal, set, and movement activity of the population of CMAd, CMAv, and SMA neurons sampled. Scales in top left histogram also applies to all other histograms. Bin width is 5 spikes/s. , mean. Notice that the mean response magnitude of all activities was strongest in SMA and weakest in CMAd, although there was substantial overlap in the distributions across areas.

RESPONSE DURATION. The average duration of signal activity was similar in all three motor areas. However, movement activity lasted longer in CMAd and CMAv than in SMA. Figure 8 shows the distribution of signal and movement activity durations in CMAd, CMAv, and SMA. The mean duration of signal activity in CMAd, CMAv, and SMA was 477 ± 307, 604 ± 326, and 451 ± 269 ms, respectively, with a Kruskal-Wallis test failing to indicate a significant difference across motor areas (²_[2] = 2.88, P > 0.2). In contrast, movement activity lasted on average longer in CMAd and CMAv than SMA. The mean duration of movement activity was 370 ± 244, 433 ± 294, and 291 ± 201 ms, respectively, with a Kruskal-Wallis test indicating a highly significant difference across motor areas (²_[2] = 15.2, P < 0.0005). An unplanned comparison between means indicated that movement activity duration in SMA was significantly shorter than in CMAd and CMAv, which were not significantly different from each other. Thus signal activity persisted for ~0.5 s in all three motor areas, roughly the same duration as the visual cue. In contrast, the duration of movement activity was shorter in SMA than in CMAd and CMAv, although there was substantial overlap in their distributions.

View larger version (17K): [in this window] [in a new window]   Fig. 8. Distributions of signal and movement activity response durations of the population of CMAd, CMAv, and SMA neurons sampled. Histogram binwidth is 50 ms for signal activity and 100 ms for movement activity. , mean.

POPULATION ACTIVITY. To reveal the overall neural activity in each motor area during task performance, averaged spike density functions were compiled from all neurons in each cortical area. Population activity was qualitatively similar in CMAd, CMAv, and SMA, although there were minor variations that reflect the quantitative differences already described. Figure 9 shows the average signal, set, and movement activity in each motor area from the subsample of neurons that exhibited significant activity in each of these response epochs. These spike-density functions were compiled by first computing the spike density functions for each neuron with the cue direction that evoked the largest response and then averaging them together. Neural activity in all three motor areas consisted of a phasic increase in firing rate ~10 spikes/s that began ~200 ms after the cue, a gradual increase in activity during the delay period, followed by another burst of activity that began just before the beginning of the monkey's movement.

View larger version (10K): [in this window] [in a new window]   Fig. 9. Response profiles of activities in the CMAd, CMAv, and SMA. Each plot shows the population spike density functions of the average signal, set, and movement activity from all significantly active neurons using trials where the instruction cue was located in the direction eliciting the largest response. The average response from each motor area are depicted as different color lines within the same plot. Top: activity aligned to the onset of the instruction cue showing signal activity. Middle: activity aligned to the GO signal showing set activity. Bottom: activity aligned to the beginning of movement showing movement activity. Notice that all 3 motor areas exhibited similar response profiles.

Directionality

The strength of task-related activity usually depended on the direction of the instruction cue and subsequent movement. Examples of directional signal, set, and movement activity are illustrated in Fig. 10. Figure 10, top, shows the directional signal activity of an SMA neuron using four cue directions. This neuron responded with a phasic increase in activity associated with the onset of the cue when it was located either above (90°) or to the right (0°) of the central target. A one-way ANOVA indicated a highly significant effect of target direction [F(3,16) = 13.64, P < 0.0002], and the weighted mean vector of signal activity indicated an optimal direction (_signal) of 54° (P < 0.001), about half-way between the two targets that elicited the largest responses. Figure 10, middle, shows directional set activity of a CMAd neuron. The discharge rate of this neuron gradually increased during the delay period primarily when the instruction cue was located down and to the right (315°), corresponding to an impending joystick movement backward and rightward. The one-way ANOVA indicated a highly significant effect of target direction [F(7,20) = 11.9, P < 0.00001], and _set was 321° (P < 0.00001), very close to the direction with the largest response. Figure 10, bottom, shows directional movement activity of a CMAv neuron. This neuron exhibited a robust response for several different directions, with the strongest response for leftward and downward targets (225°) corresponding to leftward and backward movements of the joystick. As expected, the one-way ANOVA indicated a highly significant effect of direction [F(7,32) = 20.5, P < 0.00001], and its _move was 220° (P < 0.00001).

View larger version (18K): [in this window] [in a new window]   Fig. 10. Directionally tuned signal, set, and movement activity. Left: spike density functions showing the responses of 3 different neurons from each target location tested. Trials for each cue location were pseudorandomly intermixed. Signal activity was aligned to the onset of the cue, set activity was aligned to the GO signal, and movement activity was aligned to the beginning of joystick movement. Notice that each neuron exhibited the strongest response with instruction cues in a particular direction. Right: polar plots of response magnitudes and estimated optimal direction of neural activity. Response magnitudes are plotted as the mean spike rates ± SE. Signal activity was taken as the mean spike rate over the 147-ms epoch starting 157 ms after the instruction cue appeared. Set activity was taken as the mean spike rate 750 ms before the GO signal. Movement activity was taken as the mean spike rate over a 480-ms epoch starting 71 ms before the beginning of joystick movement. Arrows show the mean vector of the target angles weighted by the response magnitude at each target; the direction of the arrows indicates the optimal direction of neural activity and its amplitude indicates its overall directionality scaled to the maximum response magnitude. Each neuron was recorded during different experimental sessions.

The optimal directions of signal, set, and movement activity were uniformly distributed in all three motor areas. Approximately three-quarters of all recorded neurons exhibited a significant directional bias of its signal, set, or movement activity in the one-way ANOVA, and a significant optimal direction () was calculated for most (94%) of them. Figure 11 shows the distribution of _signal, _set, and _move in each motor area. Interestingly, the proportion of activities that were significantly directional were not the same across activity types. Only about one-fourth of the signal activity (20% CMAd, 29% CMAv, and 30% SMA) was significantly directional. In contrast, roughly half of set activity (52% CMAd, 41% CMAv, and 62% SMA), and about three-fourths of the movement activity (70% CMAd, 62% CMAv, and 78% CMAv) was significantly directional. Rayleigh tests for each set of _signal, _set, and _move in each motor area failed to reject the null hypothesis of uniformly distributed directions.

View larger version (18K): [in this window] [in a new window]   Fig. 11. Distribution of signal (signal), set (set), and movement (move) activity optimal directions. Angle histograms show the number of neurons with optimal directions within group intervals of 22.5°. These data were from activities that exhibited both a significant direction effect in the 1-way ANOVA and significant  assessed using a bootstrap procedure. A Rayleigh test on each distribution failed to indicate a significant deviation from uniformity.

In many cases, a single neuron contributed more than one datum in Fig. 11 because it had more than one type of activity with an optimal direction calculated. When within-neuron activities were compared, we found a strong tendency for their optimal directions to be similar. Figure 12 shows scatterplots of _set versus _move from neurons with both activities in CMAd, CMAv, and SMA. In all three motor areas, the circular-circular correlation between _set and _move was highly significant (CMAd: r_aa = 0.41, n = 33, P < 10¹⁰; CMAv: r_aa = 0.76, n = 8, P < 0.001; SMA: r_aa = 0.25, n = 57, P < 10⁸). The relatively smaller number of neurons with signal activity did not allow a similar analysis of _signal. However, taking all neurons with signal activity regardless of motor area and plotting _signal versus either _set and _move revealed a similar trend (r_aa = 0.12, n = 29, P < 0.02). Thus there was a tendency for different activities within the same neuron to have similar optimal directions.

View larger version (10K): [in this window] [in a new window]   Fig. 12. Relationship between the optimal directions of set (set) and movement (move) activity within single neurons. Set activity was calculated as the mean spike rate 750 ms before the go signal; movement activity was calculated as the mean spike rate during each neuron's unique movement response epoch (see METHODS). , linear regression of set on move. - - -, unity slope to emphasize overall similarity between set and move. Length and placement of plot frame indicate range of data along each axis.

To compare the directional tuning profiles of neurons from each motor area, responses were fit to a cosine function that included a parameter of tuning width (see METHODS). Figure 13, left, shows narrow, intermediate, and broadly tuned directional activity from three different neurons. Tuning width () was calculated as the distance between points where y equals the midpoint between the minimum and maximum of the fitted function. Values of  were constrained between 93 and 267° to prevent undesirable secondary peaks and troughs in the function ( = 180° in a standard cosine). Over half (61%) of the activities with a significant  resulted in a significant fit to the modified cosine function. As expected, the peak of each fit closely matched the  derived from the mean weighted vector method (mean difference = 3.4°). The distributions of  for neurons in CMAd, CMAv, and SMA are shown in Fig. 12, right. Because there were fewer data points, we combined  from signal, set, and movement activity in each motor area into a single histogram. In all three motor areas the distribution of  was skewed toward narrow (<180°) tuning widths (median ~120°), about two-thirds of them in the first quartile. A Kruskal-Wallis test indicated no difference across motor areas (²_[2] = 0.85, P > 0.6). Thus the directional tuning profiles of neural activity in CMAd, CMAv, and SMA are generally broad but somewhat narrower than a simple cosine with no difference across motor areas.

View larger version (15K): [in this window] [in a new window]   Fig. 13. Analysis of directional tuning width. Left: directional movement activity of 3 neurons with narrow (top), intermediate (middle), and broad (bottom) directional tuning. Spike rates were fit to a modified cosine function that contained a parameter of tuning width (see METHODS). Response magnitudes for each target direction are plotted as the mean spike rates ± SE. All 3 fits were highly significant [top: F(79,76) = 11.8, P < 109; middle: F(39,36) = 14.8, P < 1012; bottom: F(96,93) = 4.73, P < 1012]. Tuning width () was measured by calculating the distance between consecutive points in the function where it crosses a line at the function's half-height (corresponding to the fit parameter B). Right: distribution of . Signal, set, and movement activity of each motor areas are combined into single histograms.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The present study has two principal findings. The first is that both the CMAd and CMAv contain signal, set, and movement activity with directional tuning similar to what is found in all other known cortical motor areas. This finding supports the hypothesis formulated from anatomical and imaging studies that CMAd and CMAv directly participate in the transformation of sensory input into motor commands during visually guided reaching. The second finding is that although there were some quantitative differences in the movement activity of CMAd and CMAv compared with SMA, the composition of activities and their response profiles were remarkably similar in all three motor areas. The small differences in neural activity suggest that each area may have distinct priorities with regard to solving different problems in motor control. However, the striking parallel in their neural activity indicates that CMAd, CMAv, and SMA utilize many of the same fundamental neurophysiological mechanisms to solve them.

Characteristics of CMAd and CMAv activity

We used an instructed delay task to dissociate different neural processes associated with visuomotor behavior. The monkeys were required to hold the cursor at the center target without moving it during the instruction cue, and one monkey was additionally required to fixate the center target with its eyes. During this instruction period, a significant minority (~1/4) of CMAd and CMAv neurons responded to the onset of the instruction cue with relatively short (~200 ms) latencies, suggesting that both cingulate areas are involved in the earliest stages of visuomotor processing. Interestingly, signal activity was less often directional than set or movement activity. The paucity of directional signal activity may be due to the relatively small angle (23°) our visual targets subtended. Placing visual targets at more peripheral locations could have stimulated the response fields of many neurons with signal activity more optimally.

A larger proportion (~1/2) of CMAd and CMAv neurons exhibited sustained or increasing activity during the delay period. We used the term set activity to describe this type of response because it appeared qualitatively similar to responses described as set activity in the premotor cortex (PMC) on the lateral surface of the hemisphere during instructed delay tasks similar to the one used here (e.g., Weinrich and Wise 1982). In the present study, the onset times of set activity were variable across neurons, not significantly different across motor areas, and may have a bimodal distribution. The bimodal distribution may represent two distinct subtypes of set activity. Alternatively, weak signal activity below the threshold of detectability may have resulted in some erroneously early onsets. Nevertheless, the presence of set activity indicates that CMAd and CMAv may participate in several processes that occur simultaneously during the delay period. These include remembering the location of the visual target, maintaining spatial attention, preparing to make a saccade to the target, and/or preparing to make an arm movement. In PMC, set activity is believed to largely represent neural processes related to the preparation of movement (Kurata and Wise 1988; Wise and Mauritz 1985). However, some neural activity in PMC (di Pellegrino and Wise 1993) and prefrontal cortex (Funahashi et al. 1989) carry attentional and/or mnemonic information, and both areas project to various subdivisions of CMA (Barbas and Pandya 1987; Deacon 1992; Ghosh and Gattera 1995; Godschalk et al. 1984; Kunzle 1978; Kurata 1991; Matelli et al. 1986; Morecraft and Van Hoesen 1993). We think the majority of set activity in CMAd and CMAv represents neurophysiological processes related to movement preparation rather than spatial memory, attention, or preparing to make a saccade because this activity was not present in the vast majority of neurons that were also tested with an oculomotor version of our instructed-delay task, even though the task imposed identical visual, memory and attentional demands (unpublished observations).

The vast majority (>3/4) of neurons in both CMAd and CMAv responded in conjunction with the monkey's arm movement. The majority of these responses (~3/4) began before the monkey's arm began to move and were maximal for a particular movement direction. This category of activity may represent the coding of movement parameters that are subsequently directed to the motor apparatus via direct projections to the primary motor cortex or spinal cord.

In many cases, signal, set, and movement activity was strongest for a particular direction. Directional activity was broadly tuned with all directions uniformly represented across the neuronal population in both CMAd and CMAv. This finding is similar to what has been found in M1 (e.g., Georgopoulos et al. 1982), parietal area 5 (e.g., Kalaska et al. 1983), and PMC (e.g., Caminiti et al. 1990). Interestingly, directional activity in CMAd and CMAv was on average more sharply tuned than the simple cosine that has traditionally been used to model the directionality of neural activity in motor areas. Recent studies, however, indicate that tuning curves of M1 neurons are more sharply tuned than previously thought (Amirikian et al. 2000). Thus both CMAd and CMAv encode this important task parameter in a manner similar to other cortical motor areas.

The gradually increasing proportions of signal, set, and movement activity is consistent with the close proximity of CMAd and CMAv to the motor apparatus. Nevertheless, the plentiful supply of all three types of activity suggest that both premotor areas actively participate in all stages of visuomotor processing during visually guided reaching. In addition, the presence of all different possible combinations of activity within individual neurons and broad directional tuning curves is consistent with the emerging view that visuomotor transformations are distributed across a large array of neurons where single elements can be involved in many different information-processing functions but make only fragmentary contributions (Hinton et al. 1986).

CMAd and CMAv compared with SMA

An obvious question is whether CMAd and CMAv subserve specific aspects of motor control and how they compare to the adjacent SMA. Although these three medial premotor areas are interconnected and generally have similar patterns of input-output connectivity, there are  several anatomical differences between them that could mediate different functions. CMAd, CMAv, and SMA each occupy cytoarchitectonically distinct cortical fields (Dum and Strick 1991a; Matelli et al. 1991; Vogt 1993; Vogt et al. 1987) and have distinct patterns of neurotransmitter receptors (Zilles et al. 1996). There are also some noticeable differences in their afferent and efferent connectivity. For example, both CMAd (Holsapple et al. 1991) and SMA (Rouiller et al. 1999; Schell and Strick 1984; Wiesendanger and Wiesendanger 1985) receive the major component of their thalamic input from the nucleus ventralis lateralis pars oralis, which receives inputs primarily from the pallidum (DeVito and Anderson 1982; Kim et al. 1976). In contrast, CMAv receives a significant component of its input from the mediodorsal and intralaminar nuclei and ventralis lateralis pars caudalis (Holsapple et al. 1991; Yeterian and Pandya 1988), which receives inputs primarily from the deep cerebellar nuclei (Asanuma et al. 1983a,b; Kalil 1981; Percheron 1997; Stanton 1980). In addition, CMAv receives input from prefrontal and rostral premotor cortices not shared by CMAd and SMA (Barbas and Pandya 1987; Bates and Goldman-Rakic 1993; Lu et al. 1994; McGuire et al. 1991; Morecraft and Van Hoesen 1993), CMAv and SMA receive input from rostral and ventral PMC not shared by CMAd (Barbas and Pandya 1987; Luppino et al. 1993; Matelli et al. 1986; Morecraft and Van Hoesen 1993; Tokuno and Inase 1994), and SMA and CMAd receive inputs from primary somatosensory cortex and parts of area 5 not shared by CMAv (Battaglia Mayer et al. 1998; Jones and Powell 1970). Finally, SMA corticospinal neurons have a more extensive distribution in the spinal cord compared with CMAd neurons that terminate primarily in the dorsolateral portion of the intermediate zone and CMAv neurons that terminate primarily in the dorsomedial region (Dum and Strick 1996). These anatomical differences suggest that each motor area may have evolved independently for the purpose of solving different problems in motor control.

Hypotheses concerning the functional specializations of premotor cortical areas have emphasized the role of SMA in programming internally generated and sequential movements, in contrast to a more central role of PMC in mediating sensory-guided movements (Goldberg 1985; Mushiake et al. 1991; Passingham 1987; Roland 1987; Roland et al. 1980a; Tanji 1994). However, evidence for or against specializations among the cingulate premotor areas has been sparse and inconclusive. Some neurons in the rostral cingulate motor area (CMAr) appeared more sensitive to the probability of receiving a reward than neurons in CMAd and CMAv (Shima and Tanji 1998). Measuring metabolic activity with 2-deoxyglucose, Picard and Strick (1997) found a stronger activation in CMAd compared with CMAv and SMA during a task that required the recall of particular movement sequences and hypothesized that CMAd may be functionally specialized for remembered sequences of movements. However, Shima and colleagues (1991) failed to find substantial differences in the single-neuron activity between CMAd, CMAv and CMAr during self-paced and visually triggered key press tasks. Furthermore, Cadoret and Smith (1997) reported no differences between SMA and CMAv single neuron activity during an isometric prehension task.

The hypothesis that CMAd, CMAv, and SMA subserve different motor functions during visually guided arm movements was tested by comparing single-neuron activity in each area under identical visuomotor conditions. This hypothesis predicts that differences in the information processing capabilities or priorities of each motor area would be detected in their neural activity when challenged under similar task demands. Our results indicated several clear differences in the way movement activity is represented across motor areas. First, neurons with only movement activity were more prevalent in CMAd and CMAv compared with SMA, whereas neurons with both set and movement activity were more prevalent in SMA. Second, movement activity in SMA began earlier and had a shorter duration compared with CMAd and CMAv. One possible explanation for a later onset and longer duration of movement activity is that more neurons in CMAd and CMAv were responding to somatosensory or proprioceptive stimuli and that much of the "movement activity" was actually somatosensory or proprioceptive feedback reflecting the monkey's physical movements. We do not think this is the case, however, because the proportion of neurons responsive to somatosensory or proprioceptive stimuli outside the context of the task were similar in all three areas. Other comparative studies have also shown that movement activity in SMA generally begins earlier than in M1 (Crutcher and Alexander 1990; Okano and Tanji 1987; Tanji and Kurata 1982) and that movement activity in caudal CMA begins around the same time as M1 (Shima et al. 1991). We think that the earlier onset and shorter duration of movement activity in SMA may indicate a predominant, although not exclusive role in movement initiation, whereas a later onset and longer duration of movement activity in CMAd and CMAv may indicate a more influential role in the guidance of movement.

Notwithstanding the differences in movement activity described in the preceding text, our results are inconsistent with the concept of a strict functional segregation across motor areas in several important ways. First, the overall profile of neural activity in all three motor areas was very similar (see Fig. 9), with very similar proportions of signal, set, and movement activity in each area. Second, the three motor areas were not significantly different in most of the response parameters we analyzed. The average onset latencies of signal and set activities were the same as were the average durations of signal activity. Similar proportions of signal, set, and movement activity were directionally tuned in all three motor areas, with uniform distributions of optimal directions and similar tuning profiles. Furthermore, the optimal directions of multiple activities within single neurons was usually in register in all three motor areas. Even among the response parameters that were different, the substantial overlap in their distributions (see Figs. 6-8) was arguably more striking than the differences in their means. This general similarity of neural activity in CMAd, CMAv, and SMA is consistent with their anatomical connectivity. All three motor areas are reciprocally connected to each other (Luppino et al. 1990, 1993; McGuire et al. 1991; Morecraft and Van Hoesen 1992, 1993, 1998; Wang et al. 2001). Each area also projects to M1 (Dum and Strick 1991a,b; Ghosh et al. 1987; Godschalk et al. 1984; He et al. 1993, 1995; Kunzle 1978; Leichnetz 1986; Luppino et al. 1993; Morecraft and Van Hoesen 1992; Morecraft et al. 1997; Muakkassa and Strick 1979; Nimchinsky et al. 1996; Pandya and Kuypers 1969; Pandya and Vignolo 1971; Tokuno and Tanji 1993; Tokuno et al. 1997) red nucleus (Humphrey et al. 1984), and spinal cord (Biber et al. 1978; Dum and Strick 1991a, 1996; Galea and Darian-Smith 1994; He et al. 1993, 1995; Hutchins et al. 1988; Keizer and Kuypers 1989; Macpherson et al. 1982; Morecraft et al. 1997; Murray and Coulter 1981; Nudo and Masterton 1990; Toyoshima and Sakai 1982). Of course, the similarities in the basic response properties of CMAd, CMAv, and SMA neurons during our relatively simple task does not prove that these areas have identical motor functions. Instead, we think these similarities indicate that all three motor areas share many of the same basic information processing mechanisms during the generation of visually guided limb movements.

Hypotheses of motor system organization

Premotor cortical areas have been viewed as hierarchically superior to M1 in part because it was thought that they were more remote from the spinal cord circuits that generate motor output. It was also thought that their hierarchical superiority reflected their specialized role in processing higher-order aspects of motor control (Eccles 1982; Goldberg 1985; Orgogozo and Larsen 1979; Roland et al. 1980b). A noteworthy example is the finding that neural activity in M1 is similar during both visually and internally guided motor tasks, whereas neural activity in PMC appears more responsive during visually guided limb movements, and neural activity in SMA appears more responsive during internally guided movements (Mushiake et al. 1991).

Recent studies, however, suggest that premotor areas have more in common with primary motor cortex than once thought. In addition to having reciprocal connections among them, each premotor area projects directly to M1 and the spinal cord (Dum and Strick 1991b). Furthermore, numerous studies have documented neural activity in other motor cortical areas similar to what we report here. For example, neurons with signal responses have been found in SMA (Tanji and Kurata 1981), PMC (Weinrich and Wise 1982), and M1 (Kwan et al. 1981). Set activity has also been reported in SMA (Tanji et al. 1980), PMC (Weinrich and Wise 1982), and M1 (Tanji and Evarts 1976). Similarly, neurons in SMA (Brinkman and Porter 1979; Tanji and Kurata 1979), PMC (Tanji and Kurata 1979), and M1 (Evarts 1966) discharge in conjunction with movements. Comparison of our results with those of similar studies in other motor areas indicates little or no difference in many quantitative measures such as the proportion of neurons with different types of activity and their response profiles, latencies and directional characteristics. Thus all cortical motor areas are activated during visually guided limb movements in a qualitatively similarly manner.

Different motor cortical areas certainly do have some functional differences, but these differences are a matter of degree. Thus all motor cortical areas are engaged during motor behavior, but each one is activated in a manner that is commensurate to its information-processing abilities and the particular task demands. This arrangement provides flexibility in the way information flows through the network of interconnected cortical areas, with those having a particular network architecture taking priority under different behavioral circumstances. Many of the apparent similarities among cortical areas merely reflect the utilization of similar neurophysiological strategies to solve different motor control problems. Furthermore, representations of all types of movements in all motor areas is not surprising if one considers individual neurons as part of a distributed neural network where every area potentially contributes to diverse motor actions. De-emphasizing the hierarchical concept of motor organization does not diminish the importance of one structure at the expense of another but instead expands the set of structures that participate in the complex computations that produce behavioral flexibility in mammals. A better understanding of information processing in the motor system will require further neurophysiological studies on the properties and functional specializations of different cortical regions during different motor tasks as well as the development of computer models that predict the patterns of single neuron activity that are found.