| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
1Department of Physiology, Hadassah Medical School, Hebrew University, Jerusalem; 2The Interdisciplinary Center for Neural Computation, Hebrew University, Jerusalem; and 3Gonda Brain Research Center, Bar-Ilan University, Ramat-Gan, Israel
Submitted 18 December 2006; accepted in final form 5 March 2007
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMETHODSRESULTSDISCUSSIONAPPENDIXGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONAPPENDIXGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
). Reaching mainly involves movements of proximal joints, the shoulder and elbow, whereas grasping entails shaping the hand by manipulating distal joints, the fingers. Distinct brain regions have been implicated in the control of reach and grasp. In the premotor cortex (PM), inactivation (Fogassi et al. 2001; Kurata and Hoffman 1994), anatomical (He et al. 1993; Tanné-Gariepy et al. 2002), and neurophysiological studies (Luppino and Rizzolatti 2000; Wise et al. 1997) have related the caudal part of dorsal PM (PMd) to proximal arm movements and the control of reaching, whereas the rostral part of ventral PM (PMv) has been implicated in distal hand movements and the control of grasping.
Neurophysiological studies of PM have employed a combination of intracortical microstimulation (ICMS) (Godschalk et al. 1995; Stoney et al. 1968), informal mapping techniques (Gentilucci et al. 1988; Graziano et al. 1997), and single-unit recordings (Kurata and Tanji 1986; Rizzolatti et al. 1988) to study the functional organization of PMd and PMv. Generally, a good match was observed between single neuron activity, ICMS, and other mapping results: neurons recorded at sites mapped as proximal modulated their activity with reach direction (Crammond and Kalaska 1996; Gentilucci et al. 1988), and neurons recorded at sites in which ICMS evoked finger movements modulated their activity with grasp type (Murata et al. 1997; Rizzolatti et al. 1988).
Several issues suggest that the neural representations of reaching and grasping in PM are not completely separate. First, recent studies have shown that within PMd, there are sites related to distal movements (Dum and Strick 2005; He et al. 1993; Raos et al. 2004) and that neural activity in the PMv may be modulated by reach direction (Schwartz et al. 2004). Second, there are spatial differences between single neuron recordings and ICMS: whereas the former are local, the latter affect larger areas (Tehovnik et al. 2006). Third, most experiments to date have studied reach (Caminiti et al. 1991; Shen and Alexander 1997) and grasp (Hepp-Reymond et al. 1994; Murata et al. 1997) using separate experimental paradigms.
Here we studied the functional organization of PM in intact behaving monkeys as both reach and grasp were varied systematically. We found that during prehension, proximal muscles were modulated by reach direction more than distal muscles, and distal muscles were modulated by grasp type more than proximal muscles. We expected that neurons recorded in sites from which ICMS evoked proximal movements would be reach-related and neurons recorded in distal sites would be grasp-related. However, we found that the proportion and properties of single-units related to reaching and to grasping were the same regardless of the responses elicited by threshold ICMS through the same electrode at the recording point.
|
METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONAPPENDIXGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
Two monkeys (Macaca fascicularis, females, D and J, 2.5 and 3.2 kg, respectively) were trained to perform unconstrained prehension movements. All surgical and animal handling procedures were according to the National Institutes of Health Guide for the Care and Use of Laboratory Animals (1996), complied with Israeli law, approved by the Ethics Committee of the Hebrew University, and supervised by a veterinarian.
During recording sessions, the monkey was seated in a primate chair with its head held and left arm restrained. A touch pad with three buttons arranged in a row (each 1.3 x 1.3 cm, with tricolor LEDs) was located 12 cm in front of the monkey's right mid-clavicular line at chest level, and the monkey rested its right hand on the central button. This resting position permitted equal amplitude movements (6.5/7.5 cm, monkey D/J, the difference due to differences in the monkeys’ sizes) in six directions, equally spaced in the horizontal plane (a "center 
out" arrangement). In each trial, an object was presented in one of the six locations, at the same height as the touch pad. During each session, two of the three objects shown in Fig. 1A were used, for a total of 12 different task conditions. During prerecording sessions, we observed that when different objects are presented in the same location, horizontally orientated objects result in considerably less variation of forearm orientation than vertically oriented objects. We therefore presented objects horizontally, requiring a similar orientation of the shoulder, elbow, and wrist joints in a given location. Different finger configurations were essential for a correct grasp of each object. As shown in RESULTS, extrinsic digit muscles were indeed modulated much more strongly by grasp type than shoulder, elbow, and wrist muscles. The precision grip object, for instance, consisted of a groove 8 mm wide, into which only one of the monkey's fingers (diameter, 6 mm) could fit; in this groove, two plates were installed, each connected to a micro-switch. For a correct grasp, both micro-switches had to be depressed simultaneously, possible only by inserting one finger in front of the plates, another behind, and pinching them together. Other objects were designed with other electro-mechanical constraints and gave rise to other grip types (see Fig. 1A for more details). Grip force was not controlled.
|
Following training, water-filled glass beads were glued onto the skull surface over the left hemisphere under aseptic conditions (medetomidine hydrochloride (Domitor) and ketamine anesthesia). A localizing MRI scan (Biospec Bruker 4.7 T animal system, fast spin echo sequence; effective echo time, 80 ms; repetition time, 2.5 s; 0.5 x 0.5 x 2 mm resolution) was then performed. Chamber (22 x 44 mm) implantation (halothane anesthesia, induced by Domitor and ketamine) was guided by the relative positions of the beads and cortical landmarks. Analgesia [pentazocine (Talwin) and carprofen (Rymadil)] and antibiotics (ceftriaxone) were administered peri-operatively. The dura mater was left intact. Sulci locations relative to chamber were then determined by another MRI scan.
Data acquisition and preprocessing
During each recording session, 
16 glass-coated tungsten micro-electrodes were employed (impedance 0.2–2 M
at 1 kHz). Electrodes were arranged in two circular guide tubes that were lowered down to 
1 mm above the dura mater (8 electrodes in each guide tube; inter-electrode spacing within tube 300 µm; Double MT, Alpha-Omega Engineering, Nazareth, Israel). During each session, one guide tube was aimed toward PMd and another toward PMv. The border between primary motor cortex (M1) and PM was established based on stimulations made during separate mapping sessions (Fig. 4A): in M1 movements were typically evoked at low currents (40 µA) and visual responses during mapping were rare (for details of stimulation parameters and mapping protocol, see Cortical mapping). The border between PMd and PMv was defined as the line extending caudally from the arcuate spur (dotted lines in Fig. 4, C and D). Each electrode was lowered through the dura mater independently (EPS 1.31, Alpha-Omega Eng.) until spiking activity was encountered, inserted an additional distance into the cortex (median, 0.71 mm), and left in the same site during the entire recording and mapping session. The signal from each electrode was amplified (10,000), band-pass filtered (1–10,000 Hz), and sampled at 25 kHz (Alpha-Map 5.4, Alpha-Omega Eng.).
|
|
|
50-Hz line influences were removed by cycle-triggered averaging: the signal following AC polarity change in a main was averaged over many cycles and subtracted from the original signal in an adaptive manner. Second, spikes were detected by computing a modified second derivative (7 samples backward and 11 forward), accentuating "spiky" signal features (bimodal, skewed, and sharp). Segments that crossed a threshold (4.5 SDs from the mean derivative) were identified. Within each segment, the occurrence time of the putative spike was defined as the time of the maximal derivative. If a sharper spike was not encountered within 1.2 ms, then 64 samples, starting 10 before the peak, were extracted. We also detected spikes based on amplitudes (
5 times the SD of the 300- to 6,000-Hz band-passed trace): spikes detected using the two methods were the same in 95% of the cases, and results were independent of the detection method. Third, extracted spike waveforms were linearly de-trended and aligned so each started at the point of maximal fit with two library principal components accounting, on average, for 93% of the waveform variance (Abeles and Goldstein 1977). Finally, waveforms were projected onto the principal component basis to arrive at two coefficients. These coefficients were subjected to manual spike-sorting in the environment of Alpha-Sort (4.0, Alpha-Omega Eng.). Eye movements were recorded using an infra-red beam system tracking movements of one eye (Oculometer, Dr. Bouis, Karlsruhe, Germany). The horizontal and vertical signals from this system were sampled at 400 Hz and low-pass filtered (40 Hz). Behavioral events (LEDs, switches, lights, and so on) were sampled at 6 kHz. The workspace and monkey's movements were monitored using three infra-red cameras and recorded on VHS tapes.
During some sessions intra-muscular electromyograms (EMGs) were recorded in parallel to neurons. Wire pairs were inserted into shoulder (acromion deltoid, latissimus dorsi, pectoralis major), elbow (biceps brachii, brachialis, triceps brachii), wrist (extensor carpii radialis longus, extensor carpii ulnaris, palmaris longus), and extrinsic digit (extensor digitorum communis, flexor digitorum profundus, flexor digitorum sublimis) muscles. Wire positions were verified by stimulation (90-ms trains of 0.2-ms biphasic pulses at 330 Hz) at currents of 50–500 µA. The signals from these wires were amplified (30,000), band-pass filtered (30–3,000 Hz), and sampled at 6.25 kHz. Root mean square (RMS) values of EMG were computed by raising signals to the second power, applying a low-pass filter (100 Hz), and taking the square root. Each muscle was recorded during 2–10 separate sessions for a total of 61 recordings; each recording was 184–527 (median, 414) trials long. Different recordings of the same muscle were consistent (correlation-coefficient between mean activity in each of the 12 task conditions, 0.86 ± 0.05, mean ± SE; 164 pairs of same-muscle recordings).
Cortical mapping
During each session, immediately following neural recordings and without moving the electrodes, two standard methods were used to characterize each electrode's recording site: threshold ICMS and sensory-motor mappings (SMM). In the SMM, we assessed recording site properties by listening to the modulation of the multi-unit rustle of each electrode during proprioceptive input (passive movements of individual joints, palpation of muscles), tactile input (stroking of skin, stroking of facial skin while eyes were covered), visual input (3 dimensional objects, LEDs, hand movements of the experimentalist; at reaching distance and beyond), and active movements (see Gentilucci et al. 1988; Graziano et al. 1997; Kakei et al. 2001; Schwartz et al. 2004 for similar SMM protocols). The latter were examined using a Kluver board variant: a Perspex plate with nine holes in three rows, 6.5 cm apart. Each hole contained a small food pellet and was covered by an object requiring a different grasp (ball, cork, plate, string, and so on). The SMM in each site was summarized by two properties: response modality (motor, somatosensory, and/or visual) and organ (elbow, eye, face, finger, multi-joint, mouth, pinna, shoulder, and wrist were observed).
In the ICMS we noted, for each electrode's recording site separately, the movement elicited by the lowest possible current during at least half of the stimulation trials (0.2-ms biphasic pulses at 330 Hz for 90 ms at currents of 5–90 µA) (Godschalk et al. 1995; see Gentilucci et al. 1988; Kakei et al. 2001 for similar ICMS protocols employed in PM). Movement was summarized by two properties: movement type (flexion, extension, adduction, abduction, opposition, and so on) and organ (same as SMM organs in the preceding text). We classified evoked forelimb movements as follows: proximal: movements around the shoulder and elbow flexion/extension; wrist: forearm supination/pronation and wrist movements; and distal: finger movements. Only results from proximal and distal sites are reported in detail in this paper. Although at suprathreshold currents we sometimes observed complex multi-joint movements (Godschalk et al. 1995; Graziano et al. 2002), lower currents always yielded single-joint movements with the exception of distal movements that combined several fingers in 2/3 of the cases. We did not observe transitions from proximal to distal (or vice versa) movements as current was changed.
Neural database and data analyses
One baseline (control) and six task epochs were defined for analyses, all 400 ms long (Fig. 1B, bars). During the Control epoch, starting 100 ms after Ready, the monkey did not know in which direction it would have to reach and what type of grip it would have to use. During the Signal epoch, starting 50 ms after Cue On, the identity and location of the target objects was briefly visible yet no movement was required. During Delay epochs, starting 450 ms after Cue On, there was no visual cue and no movement was required. These exact same conditions were maintained throughout the PreGo epoch (starting 400 ms prior to the Go Signal). The Movement epoch started 150 ms before the hand left the touch pad, and the Hold epoch started 100 ms after a Correct Grasp. Note that the Delay2 and PreGo epochs overlap on average by 100 ms, and the Movement epoch includes both reaction and movement time (as in Weinrich and Wise 1982).
Units fulfilling the following criteria were analyzed. 1) Anatomy. Only units recorded at PM sites were considered. 2) Mapping. Only units recorded at sites classified as proximal or distal by ICMS and/or SMM were included. 3) Isolation. Quality of single-unit isolation was determined by the homogeneity of spike waveforms, separation of the projections of spike waveforms onto principal components during spike-sorting and clear refractory periods in ISI histograms. Only well-isolated units were considered. 4) Number of trials. Each unit had to be recorded for at least five trials per task condition and exhibit stationary activity. This was determined by visual inspection of mean firing rates and raster plots of individual trials. 5) Firing rate. Only units with mean firing rates 1 spike/s during at least one task epoch were used. 6) Task dependency. Spike counts for all task conditions together, during the Control and during each task epoch, were compared using a two-sample, two-tailed t-test. Only units with P values <0.01/6 during at least one of the six epochs were included.
We made a total of 617 PM penetrations, 531 during recording sessions, of which 378 were in sites classified as proximal or distal. A total of 724 units, recorded from 326 of these sites, passed the preceding criteria (Table 1, 1st 3 columns, lists the sites, units, and their classification). Although the number of units varied when changing parameters used for inclusion criteria 4–6, results were not sensitive to specific parameter values. These units were recorded during 70–582 (median, 257) trials.
|
We used a two-way ANOVA, with direction and object as factors, to determine relation of neural activity to task parameters (effects were considered significant at P < 0.01 levels). This analysis assumes that spike counts distribute normally: because spike counts are nonnegative and distribute as a Poisson counting process, for some units a normal distribution was not a good approximation (Bera-Jarque test of normality: 76/724 units with P < 0.01). We therefore repeated analyses using a nonparametric ANOVA (2-way Kruskal-Wallis test); results were almost identical for the two ANOVA tests. Because vector summation was used to estimate PDs and ANOVA to determine significance of directional tuning, we compared the vector summation PD estimate with a discrete estimate, the direction in which a unit fired maximally: the mean absolute difference between the two estimates was 30 ± 0.9° (SE) (724 units). Thus for the neural data used in this study, an ANOVA is a reasonable method for estimating the significance of tuning. The same results were obtained when estimating significance using a resampling test (Crammond and Kalaska 1996).
To estimate effect sizes we computed eta-squared, defined as 
2 = 
effect2/total2 (Fisher 1925). Effect variance is defined as the variance of the mean discharge in each relevant task condition (e.g., task2 is the variance of 12 numbers, and dir2 is the variance of 6 numbers). 2 is a unit-less measure of the fraction of the total variance associated with an effect. If the signal-to-noise ratio (SNR) is defined as effect2/(total2 – effect2) then 2 equals SNR/(1 + SNR) and is thus a monotonic function of the SNR. 2 is bounded between 0 and 1 and accounts for linear and nonlinear effects. Because 2 is additive for different effects, the total variance associated with the prehension task parameters is equal to the sum of direction, object, and interaction effect sizes: task2 = dir2 + obj2 + int2. To compare variance associated with direction with that not associated with direction alone, we employed a "reach-grasp index": RGI = (dir2 – obj2 – int2)/task2. This index equals 1 when there is only a direction effect and –1 when there is no direction effect.
We used multiple linear regression, model Fu ,t = b0 + b1·X t + b2·Y t + 
t, to test for relations between neural activity and eye positions. In this model, F is the spike count of unit u during the relevant epoch of trial t, X and Y are the horizontal and vertical eye positions during the same time, and is an error term. All values were standardized prior to regressing (the mean subtracted and divided by the SD) for each of the 12 task conditions separately to remove possible task-related effects.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONAPPENDIXGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
We measured the activity of shoulder, elbow, wrist, and extrinsic digit muscles during the prehension task (Fig. 1) as monkeys reached in different directions and grasped various objects. Muscles were generally quiescent before the Go Signal and became active just before Movement Onset as the monkey's hand released the touch pad (Fig. 2). To quantify the temporal relations between reach and grasp movement components and muscle activity, we identified, for each muscle, the time of peak activity. Activity of all muscles peaked during movement, and activation of proximal (shoulder and elbow) and distal (extrinsic digit) muscles overlapped (Fig. 3A). However, activity of proximal muscles peaked 125 ms after Movement Onset, whereas activity of distal muscles peaked 221 ms after Movement Onset (medians; Mann-Whitney U test: P < 0.001). Similar results were obtained when the time of median (instead of peak) activity was considered. Thus although activation times of proximal and distal muscles were not completely distinct, activity of proximal muscles tended to peak before the activity of more distal muscles.
To quantify the modulation of muscle activity by reach and grasp during the Movement epoch, we employed a reach-grasp index (RGI) that equaled 1 when all task-related variance was attributed to reach direction and –1 when all task-related variance was related grasp type (see METHODS). For instance, the biceps brachii (an elbow flexor and forearm supinator; Fig. 2A) had an RGI of 0.9, reflecting a dominant influence of reach over grasp, whereas the FDS (an extrinsic digit flexor; Fig. 2B) had an RGI of 0.05, reflecting a more balanced effect of reach and grasp. Proximal muscles had consistently higher RGIs than distal muscles (medians: 0.85 and 0.15, respectively; Mann-Whitney U test: P << 0.001). In fact, the distributions of RGIs of proximal and distal muscles were completely distinct (Fig. 3B). Wrist muscles had high RGIs, similar to the proximal muscles (median: 0.94; U test: P = 0.11). This is in accordance with the experimental design where all grasp types at a given location required the same wrist orientation. Thus in the prehension task, proximal and wrist muscles showed larger effects of reach direction than did distal muscles, while distal muscles showed larger effects of grasp type.
Proximal and distal sites aggregate in different regions of premotor cortex
We mapped the PM using two methods: ICMS and SMM. During ICMS, we identified the weakest (threshold) current that evoked perceptible movement by stimulating a given site, and during SMM, we listened to the multi-unit rustle during passive and active manipulations (see METHODS). We made a total of 617 electrode penetrations in PM. Movements were observed following ICMS in 288 of these sites (47%; Fig. 4B. The mean threshold for obtaining a response was 58.7 ± 1.4 (SE) µA. SMM yielded clear classification in 549 of the sites (89%). In 68 sites (11%), no movement was evoked and no specific organ could be determined. There was a good correspondence between the two methods: when both yielded positive results, the ICMS organ was usually the same as the SMM organ (283/288 sites, 98%).
We did not attempt to map the entire PM cortex. Instead, we focused on regions in which proximal (shoulder or elbow) or distal (finger) forelimb representations were found. Two attributes of the resulting maps are noteworthy (Fig. 4, C and D). First, in each monkey, there was one region consisting mainly of shoulder and elbow sites (henceforth designated as proximal sites) and a separate region in which finger sites (henceforth designated as distal sites) were most common. Second, in both monkeys, proximal sites were interspersed within the primarily distal region, whereas the opposite was very rare, and multi-joint and wrist sites were scattered among both proximal and distal regions.
To formally quantify these observations, we divided recording sites, in each monkey separately, into two regions using a K-means algorithm (K = 2). Note that this partition is based only on the relative locations of the sites. The borders between the resulting regions are shown by dashed lines in Fig. 4, C and D, and the distributions of sites within each region by the pie charts. In both monkeys, most proximal sites were located in the dorsal regions (75 and 77%, monkeys D and J, respectively), and most distal sites were in the ventral regions (96 and 100%). In both monkeys, the dorsal regions roughly corresponded to the PMd. In one monkey, the region with the predominantly distal sites corresponded to the rostral PMv (Fig. 4D), and in the other monkey, to a region intermediate between the PMd and the PMv (Fig. 4C). In both monkeys, there was a gross anatomical separation between the bulk of proximal and distal sites.
Single-units in proximal and distal sites are active similarly during prehension
We recorded the activity of 724 task-dependent PM units, 551 (76%) from sites classified as proximal and 173 (24%) from distal sites. These two sets of units form the neural database for all analyses. As may be expected, some single-units displayed activity that was consistent with recording site properties. These included reach-related activity of units recorded in proximal sites (Fig. 5A) and grasp-related activity of units recorded in distal sites (Fig. 5B). However, other units displayed activity that was unexpected in terms of the mapping responses of the corresponding recording sites. For instance, the unit the activity of which is shown in Fig. 6A was recorded in a proximal site but exhibited grasp-related activity in addition to reach-related activity. Still other units had discharge patterns that were completely inconsistent with the mapping results; for example, strictly reach-related activity of a unit recorded in a distal site (Fig. 6B).
|
|
|
2 test of independence: P = 0.77; Fig. 8A, left). The fractions of units in proximal and distal sites with an object effect were equal (29% in both sets of units; 2 test: P = 0.99; Fig. 8A, right). Finally, the fractions of units in proximal and distal sites that exhibited interaction effects were similar one to another (25 and 32%; 2 test: P = 0.14). Thus the probability of observing single-unit activity related to task parameters did not correlate with recording site classification.
|
dir2, among units with a significant direction effect were roughly the same for units in proximal and distal sites (means, 0.19 and 0.21, respectively; Mann-Whitney U test: P = 0.5; Fig. 8B, left), and so were the effect sizes for object, obj2 (0.07 and 0.09 for units in proximal and distal sites, respectively; U test: P = 0.08; Fig. 8B, right). The effect sizes for interaction, int2, were also similar for units in proximal and distal sites (0.09 and 0.11; U test: P = 0.27). Moreover, the distributions of preferred directions (PDs) of units in proximal and distal sites were both uniform (Rayleigh test: P = 0.11 and P = 0.34; Fig. 8C). Both distributions of preferred objects were similar to the distributions expected based on the numbers of units with significant object effects that were tested with each object (2 test: P = 0.68 and P = 0.59, units in proximal and distal sites; Fig. 8D). In short, encoding of reach and grasp parameters by units in proximal sites during the Movement epoch appeared to be similar to encoding of those parameters by units in distal sites in all tested aspects. Reach and grasp encoding by units in proximal and distal sites: spatial and temporal considerations
As reported in the preceding section, similar properties were observed for single-units recorded in proximal and distal forelimb representations as determined by ICMS and/or SMM (Table 1, 1st row). The same results were obtained when considering only units recorded at sites in which ICMS was positive (Table 1, 2nd row) or when considering only units recorded at sites in which ICMS was negative but SMM yielded clear classification (Table 1, 3rd row). Moreover, similar results were observed when parsing sites according to anatomical criteria such as dorsal versus ventral regions, caudal versus rostral parts of the PM, or PMd versus PMv, for each monkey separately and for the two monkeys together (Table 1). Specifically, the properties of units recorded in distal sites in the PMd and in the PMv were the same, and the properties of units recorded in proximal sites within the primarily distal regions were the same as for the rest of the units.
Using multiple regression analysis, we tested the influence of gaze angle, reaction time, and/or movement time on neural activity (APPENDIX). We found that single-unit activity during the Movement epoch was only weakly modulated by these parameters and to a similar extent as during the Control epoch. Thus similarities between units in proximal and distal sites observed during the Movement epoch are unlikely to result from modulations by the tested parameters.
Single-unit activity was not limited to the Movement epoch (Fig. 7A). Most units modulated their activity by reach direction (668/724 units, 92%) or grasp type (441 units, 61%) during at least one of the six task epochs (excluding Control, Fig. 1B). Some units exhibited activity that was confined to specific periods, such as the Delay epochs (Fig. 9), whereas others were modulated during several epochs. In the latter case, tuning properties were usually retained. For instance, during the Movement epoch, the RGIs of the units illustrated in Figs. 5 and 6 were 0.8, –0.56, –0.65, and 0.65, and during the delay, the corresponding values were 0.79, 0.09, –0.23, and 0.99. Over the entire sample of units, the mean between-epoch RGI difference was 0.032 ± 0.0086 (SE) (724 units). Thus RGIs of a given unit were quite consistent throughout the trial. Moreover, tuning specifics were consistent. For instance, 184 units (25%) had significant directional modulation that persisted from Signal to Movement epochs, and the mean between-epoch correlation-coefficient was 0.75± 0.02 (184 units; computed between mean spike counts in 6 directions for each pair of epochs), with an inter-epoch PD difference of 38 ± 2.3° (mean ± SE).
|
|
Differences not directly related to the encoding of reach and grasp might distinguish units with properties consistent with the recording site (for instance, reach-related units recorded in proximal sites) from units with inconsistent properties (for instance, reach-related units recorded in distal sites). To test this possibility, we defined consistent/inconsistent units according to RGI values of all 724 units (consistent unit: unit in proximal site with RGI value in the upper 3rd of all RGIs during the Movement epoch or unit in distal site with RGI in the lower 3rd; inconsistent unit: unit in distal site with RGI value in the upper 3rd or unit in a proximal site with RGI in the lower 3rd; 249 consistent, 234 inconsistent). Because none of the units that were recorded together with muscles expressed postspike facilitation effects (114 spike-triggered averages, estimated using 1,000 triggers/average, 30 ms before and after the spike) (Fetz and Cheney 1980), we could not test the tendency of consistent units to have direct output (corticomotoneuronal) connections. We examined other features that included firing rates during the Control epoch (spike/s), trial-to-trial variability (of firing rates), spike waveform peak-to-peak amplitude (µV), ICMS thresholds (µA), and recording depth (mm). However, no significant differences between consistent versus inconsistent units were observed for any of the tested features.
Even if all properties of individual neurons were identical for proximal and distal sites, it would still be possible for groups of neurons to be organized differently with respect to PDs and preferred objects in proximal and in distal sites. To test this possibility we estimated PD differences (PDDs) among pairs of directionally tuned units in proximal and distal sites separately. During the Movement epoch, the mean PDDs among pairs of directionally tuned units recorded in the same proximal site, that is, by the same electrode, was 62.7 ± 3.4° (SE) (235 pairs), smaller than the mean PDDs for all possible pairs of directionally tuned units in proximal sites, which was 89.2 ± 0.24° (SE) (47,278 pairs; U test: P << 0.001). This property was observed for proximal sites during all epochs (Fig. 11A, left) but was weaker or absent among distal sites (Fig. 11A, right) and was maintained even when differences in the number of proximal and distal sites were accounted for.
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONAPPENDIXGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
Mixing of reach and grasp representations
Although psychophysical studies showed that during prehension, reaching and grasping are interdependent and processed in parallel (Jeannerod 1984; Soechting and Flanders 1993), the prevailing view from both neurophysiological and anatomical studies is that separate brain regions and pathways process reaching and grasping and that proximal and distal sites tend to aggregate in anatomically distinct regions, PMd and PMv (Luppino and Rizzolatti 2000; Tanné-Gariepy et al. 2002; Wise et al. 1997). The latter tendency was supported by our observations (Fig. 4). However, there have been several reports of anatomical mixing of muscle fields, both in M1 (finger muscles; Rathelot and Strick 2006; Sato and Tanji 1989) and in the PM (proximal and distal muscles; Dum and Strick 2005; Godschalk et al. 1995; Raos et al. 2004). This was also confirmed in the present study.
The main result of this study, mixing of single neurons encoding reach and grasp within a given site, is novel. How is this possible, given the large number of studies that tested PM activity in relation to reach and grasp? One possibility is related to learning-induced changes in cortical circuitry during task performance. Alternatively, mixing may not have been demonstrated because of the novel way reach and grasp were combined in the current task. Previous experiments testing directional modulation of PM neurons have typically altered reach direction in an orderly manner without manipulating grasp type (Caminiti et al. 1991; Shen and Alexander 1997), whereas experiments examining grasping used reach-to-grasp movements, varying the grasped object while keeping reach direction fixed (Hepp-Reymond et al. 1994; Murata et al. 1997). The current prehension task systematically varied both reach direction and grasp type, opening the possibility to observe neural activity related to either component.
We focused on PM areas to study the neural control of prehension, but other brain regions are clearly involved. Cerebellar activity has been related to reaching and grasping (van Kan et al. 1994) but not at the same time (Mason et al. 2006). Here we showed that although activity of proximal muscles tended to precede activity of distal muscles (Fig. 3A), the activity of single PM neurons was related to reach and grasp throughout preparation and execution of prehension (Fig. 7A). We did, however, observe a general tendency of PM neurons to become more grasp-related during the Hold period (Fig. 10). Whether these differences are due to anatomical or methodological differences is an issue for future studies.
Behavioral separation between reach and grasp
The current task aimed to differentiate between reach and grasp by systematically varying reach direction and grasp type. As the analysis of muscle activity showed, this was largely successful (Fig. 3B). There are other possible experimental paradigms that may be used for the same purpose. One possibility is to impose time restrictions on the reach and grasp phases, inducing artificial temporal separation. However, even during unconstrained reaching and grasping, the activity of proximal and distal muscles does not peak simultaneously (Fig. 3A). A second venue may be to vary reach alone, with no grasping whatsoever, and then vary grasp alone without reaching. This approach was partially adopted in the current study by requiring reaching from one touch pad to another identically shaped pad (APPENDIX). The main advantage of the current task was to enable characterization of muscle and neural activity during unconstrained reaching and grasping in a systematic manner.
None of the measured muscles displayed activity related only to reach or grasp. Although this may seem surprising, during natural movements reach and grasp are in fact biomechanically dependent. For instance, activity of wrist muscles accompanies digit movements (Schieber 1995), and shoulder movements are accompanied by stabilization of distal joints. Accordingly, activation patterns of proximal and distal muscles are expected to overlap. There were, however, some clear differences between proximal and distal muscles: proximal muscles were related mainly to reach direction while more distal muscles were also related to grasp type (Fig. 3B).
Activity of intrinsic hand muscles (Lemon et al. 1986) was not monitored in this study due to technical constraints. Because the activity of extrinsic digit muscles was modulated by grasp type much more than the activity of proximal muscles, we expect that if monitored, intrinsic muscle activity would be modulated mainly by grasp type.
The task did not call for any changes in wrist orientation when grasping different objects. Although we did not measure hand orientation directly, we videotaped movements from three directions. Examination of these records did not reveal any gross changes in wrist orientation while moving to either object at the same location. Wrist orientation is influenced by activity of wrist muscles, which depends on the movement task. In previous studies, wrist orientation and muscles were modulated by grasp type when grasping different vertically oriented objects at the same location (Brochier et al. 2004). In other behavioral paradigms, the activity of wrist muscles was related to movement direction (Kakei et al. 2001). In the present task, wrist muscles had RGI values close to one and were therefore mainly related to reach direction.
Anatomical and functional considerations
For most analyses, we parsed the sample of neurons into two sets, recorded in proximal and distal sites. There were two reasons for this "functional" rather than an "anatomical" choice. First, there were some differences between the monkeys. Although in both monkeys, the majority of proximal sites were aggregated in similar regions, the caudal PMd (Wise et al. 1997), the predominantly distal regions differed between monkeys, corresponding to the rostral PMv in one monkey and a region between PMd and PMv in another (Luppino and Rizzolatti 2000). This variability is not unique: functional cortical fields are known to vary with respect to sulcal landmarks between animals (Gabernet et al. 1999; Merzenich et al. 1975). Second, it was previously shown that proximal and distal sites are mixed in PM (Dum and Strick 2005; Godschalk et al. 1995). This was also observed in the current study. Thus it was more meaningful to parse sites according to the results of ICMS and/or SMM than by anatomy. Yet parsing sites according to anatomical criteria did not reveal any differences either (Table 1). Nevertheless, it is possible that in other PM sites, not explored in the current study, neurons with different properties do exist.
Possible explanations for the lack of correlation between single-unit properties and recording site identity
Single-unit activity was measured during the prehension task while recording site properties were tested using threshold ICMS and SMM. Whereas recording single-unit activity is equivalent to making an observation, presumably with minimal disruption of spontaneous activity, ICMS involves interference with the ongoing neural activity in a nonbiological yet causal manner. Whereas single-units are related to local circuit properties, mapping results represent more distributed properties: SMM involves listening to multi-unit activity, and the number of neurons activated by threshold ICMS depends on the excitability of individual neurons (Stoney et al. 1968). Moreover, ICMS may result in complex activation of large areas (Butovas and Schwarz 2003) and remote targets (Tokuno and Nambu 2000). Despite these differences, a good match between single-unit properties, ICMS, and SMM has been reported repeatedly (Aflalo and Graziano 2006; Asanuma et al. 1968; Gentilucci et al. 1988; Murphy et al. 1978; Strick and Preston 1982). How can the discrepancy between single-unit and recording site properties observed in the current study (Figs. 7, 8, and 10) be accounted for?
One possibility is that some neurons influence output more than others, and the firing properties of these neurons differ for distinct sites. For instance, a directionally tuned neuron in a proximal site might have stronger influence on a proximal muscle than a neuron with the same PD and effect sizes recorded in a distal site (Fig. 12A). This is possible if the neuron in the proximal site has lower excitability threshold or stronger synapses on downstream neurons that in turn synapse on proximal muscles. Then differences between consistent and inconsistent units in terms of ICMS threshold, postspike-facilitation occurrence, and so on are expected. Although none of the latter differences were observed in the current data, this possibility cannot be excluded.
|
; Ben-Shaul et al. 2003). If directionally tuned neurons in a given proximal site have similar PDs, whereas neurons in a distal site have the same firing rates and effect sizes as the proximal site neurons but more diverse PDs, then ICMS in the proximal site may be more prone to yield directional movement by virtue of averaging. In contrast, the similarity of preferred objects among object-tuned neurons in a given distal site may be greater than the similarity among object-tuned neurons in a proximal site (Fig. 12B). Results reported here (Fig. 11) are consistent with this possibility. Concluding remarks
Psychophysical studies showed that reach and grasp are coordinated with one another during prehension, but neurophysiological studies suggested that they are processed separately at the motor cortical level. Complete segregation between two entities makes coordination between them difficult to achieve. Anatomical (Dum and Strick 2005; Huntley and Jones 1991) and physiological (Kwan et al. 1987; Matsumura et al. 1996) studies suggested horizontal connections between motor cortical sites millimeters apart. The current results complement and extend these findings by showing a mixing of neurons encoding reach and grasp between proximal and distal forelimb representations.
What is the functional role of neurons with properties inconsistent with the recording site properties, such as reach-related neurons recorded at distal sites? Such neurons may partake in horizontal connections between distant sites, for instance, with other reach-related neurons at proximal sites. In that case, the former neurons may relay reach-related information from the proximal site to grasp-related neurons at the distal site, facilitating movement coordination. Clearly further research is required to test these ideas.
|
APPENDIX |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONAPPENDIXGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
Because monkeys were not required to maintain fixation during the task, gaze angle (Boussaoud et al. 1998; Mushiake et al. 1997) varied with reach direction during all 43 sessions (1-way MANOVA, P < 0.01). To control for this potentially confounding factor, we tested whether neural activity was related to eye position using regression analysis (METHODS). During the Control epoch, activity of 60/724 units (8%) was modulated by eye position (F test, P < 0.01). During the Movement epoch, the fraction of gaze-related units was similar (81 units, 11%; 2 test: P = 0.06). Mean R2 values of gaze-related units were 0.08 during both epochs. Finally, gaze effects did not correlate with recording site (62/551 and 19/173 gaze-related units in proximal and distal sites, respectively; 11% in both sets; 2 test: P = 0.85). Thus gaze angle varied with task parameters but neural activity during the Control and Movement epochs was related to gaze angle to similar extents. Similar results were obtained for the other epochs.
Another potential confound stems from reaction and movement times (inversely related to speed) which varied with reach direction, grasp type, or both during all sessions (2-way ANOVA, P < 0.01, corrected for multiple comparisons). We tested whether neural activity was modulated by reaction and/or movement time using regression analysis as for gaze angle. During the Movement epoch, the activity of 77/724 units (11%) was modulated by reaction/movement times; the mean R2 was 0.05. Modulation was similar for units in proximal and distal sites (10 and 13%, respectively; 2 test: P = 0.36). During the PreGo epoch similar results were obtained (9 and 6% of units in proximal and distal sites, respectively; 2 test: P = 0.3), but during other epochs, the number of units modulated by reaction/movement times was at chance level. Thus neural activity during the PreGo and Movement epochs was weakly modulated by reaction and movement times, regardless of the recording site.
Dominance of reach-related single-unit activity
Reach-related modulations were larger and more frequent than grasp-related modulations (Figs. 8, A and B, and 10). This could result from a statistical bias: for each unit, six directions but only two objects were sampled. To control for this factor, we diluted reach directions by keeping trials from two randomly selected directions, so diluted data included an equal number of sampled directions and sampled objects. In these data, the dominance of direction over object t
