J Neurophysiol 95: 3596-3616, 2006.
First published February 22, 2006; doi:10.1152/jn.01126.2005
0022-3077/06 $8.00
Differential Involvement of Neurons in the Dorsal and Ventral Premotor Cortex During Processing of Visual Signals for Action Planning
Eiji Hoshi and
Jun Tanji
Tamagawa University Research Institute, Tokyo; and Department of Physiology, Tohoku University School of Medicine, Sendai, Japan
Submitted 25 October 2005;
accepted in final form 19 February 2006
 |
ABSTRACT
|
|---|
We examined neuronal activity in the dorsal and ventral premotor cortex (PMd and PMv, respectively) to explore the role of each motor area in processing visual signals for action planning. We recorded neuronal activity while monkeys performed a behavioral task during which two visual instruction cues were given successively with an intervening delay. One cue instructed the location of the target to be reached, and the other indicated which arm was to be used. We found that the properties of neuronal activity in the PMd and PMv differed in many respects. After the first cue was given, PMv neuron response mostly reflected the spatial position of the visual cue. In contrast, PMd neuron response also reflected what the visual cue instructed, such as which arm to be used or which target to be reached. After the second cue was given, PMv neurons initially responded to the cue's visuospatial features and later reflected what the two visual cues instructed, progressively increasing information about the target location. In contrast, the activity of the majority of PMd neurons responded to the second cue with activity reflecting a combination of information supplied by the first and second cues. Such activity, already reflecting a forthcoming action, appeared with short latencies (<400 ms) and persisted throughout the delay period. In addition, both the PMv and PMd showed bilateral representation on visuospatial information and motor-target or effector information. These results further elucidate the functional specialization of the PMd and PMv during the processing of visual information for action planning.
|
INTRODUCTION
|
|---|
In primates, the premotor cortex is located on the lateral surface of the frontal cortex corresponding to Brodmann's area 6 (Brodmann 1909
). Studies have precisely examined the anatomical and functional organization of macaques, which have a premotor cortex located in the caudal bank of the arcuate sulcus and in the convex region rostral to the primary motor cortex (Brodmann's area 4). The premotor cortex was first characterized functionally by its involvement in visually guided motor behavior. Premotor neurons respond to the appearance of visual signals (Godschalk et al. 1981; Kubota and Hamada 1978; Rizzolatti et al. 1981) and discharge during the preparation and execution of movements under visual guidance (Godschalk et al. 1985; Mushiake et al. 1991; Weinrich and Wise 1982). In line with these observations, lesions of the premotor cortex cause deficits in the visual guidance of motor behavior (Halsband and Passingham 1985; Moll and Kuypers 1977).
Many reports have indicated that the premotor cortex is composed of two sections: ventral (PMv) and dorsal (PMd) areas. Previous studies have described differences in neuronal activity between these two areas. PMd neurons are particularly active during a preparatory motor-set period (Weinrich and Wise 1982; Wise 1985) and in relation to visuomotor-association tasks (Kurata and Wise 1988; Mitz et al. 1991). In contrast, PMv neurons respond to somatosensory stimuli applied to either the face or the arm and to visual stimuli corresponding to peripersonal stimuli (Fogassi et al. 1996; Graziano et al. 1997). Subsequent studies revealed further properties of PMv neurons: they are selective for the three-dimensional shape of objects to be grasped (Murata et al. 1997), the direction or movement trajectory in visual/extrinsic space (Kakei et al. 2001; Mushiake et al. 1997; Ochiai et al. 2005; Schwartz et al. 2004), attention to visuospatial stimuli (Boussaoud and Wise 1993a,b), decision-making based on somatosensory signals (Romo et al. 2004), and activity that mirrors the actions of viewed subjects (Rizzolatti and Craighero 2004). Lesion studies have supported these findings by describing different effects resulting from ablation of either the PMv or PMd. Monkeys with PMd lesions exhibit difficulty in executing conditional visuomotor association tasks (Halsband and Passingham 1985; Kurata and Hoffman 1994; Petrides 1986), and monkeys with PMv lesions exhibit deficits in visually guided grasping movements (Fogassi et al. 2001; Rizzolatti et al. 1983), orienting responses (Rizzolatti et al. 1983; Schieber 2000), and remapping from visual to motor space (Kurata and Hoshi 1999). Cortico–cortical connections linking the parietal and premotor cortex have also revealed a difference in major input sources from the parietal cortex to the PMv and PMd: the PMd receives its major input from areas medial to the intraparietal sulcus, whereas the PMv receives major input from areas ventral and lateral to the intraparietal sulcus and its lateral surface (Caminiti et al. 1996; Johnson et al. 1996; Kurata 1991; Luppino et al. 1999; Matelli et al. 1998; Rizzolatti and Luppino 2001; Tanne-Gariepy et al. 2002).
Only a few studies have directly compared PMd and PMv neuronal response properties by recording neurons from both areas in individual subjects. These studies found that set-related activity mainly occurred in the PMd, whereas movement-related activity occurred in both the PMd and PMv (Kurata 1993). A different study that used a behavioral task to dissociate attention to visual space from the intention to move an arm reported that the PMd and PMv tended to be involved in representing intention and attention, respectively (Boussaoud and Wise 1993a,b). Our comparison of motor set-related activity revealed that PMv neurons mainly reflected visual target-location, whereas PMd neurons reflected both target location and arm use (Hoshi and Tanji 2002, 2004d).
We found in a previous study that neurons in the dorsal and ventral parts of the dorsolateral prefrontal cortex (DLPFC) exhibit different response properties. During action planning based on two sets of visual information, neuronal activity in the dorsal DLPFC (DLPFCd, dorsal to the principal sulcus) reflected specific motor instructions for arm use and target location and combined them to plan the action, whereas neuronal activity in the ventral DLPFC (DLPFCv, ventral to the principal sulcus) preferentially represented the cue's visuospatial features (Hoshi and Tanji 2004a). Cortico–cortical connections preferentially link the DLPFCd and DLPFCv with the PMd and PMv, respectively (Barbas and Pandya 1987; Luppino et al. 2003; Petrides and Pandya 1999, 2002; Wang et al. 2002). These findings made it of great interest to compare the PMd and PMv neuron response properties in the same subjects performing the same behavioral task. This paper discusses how PMd and PMv neurons exhibited distinct response properties when subjects received visual signals for two sets of instructions specifying components of forthcoming actions and combined them for action planning. A preliminary report of this study has already appeared as an abstract (Hoshi and Tanji 2004c).
|
METHODS
|
|---|
We studied two male monkeys (Macaca fuscata, 8 kg) that were cared for according to National Institutes of Health guidelines. Our previous studies used the same two monkeys (Hoshi and Tanji 2000, 2002, 2004a,b; Hoshi et al. 2005). Our previous report (Hoshi et al. 2005) described in detail the experimental setup and methods used for data recording, animal surgery, and histological processes.
Behavioral task
The monkeys were trained to perform a target-reach movement using two sets of instructions: one instruction indicated the target location and the other indicated which arm to use when reaching for the target (Fig. 1A). After an intertrial interval of 
3 s, the task commenced when a monkey placed one hand on each touch pad and gazed at a fixation point (FP; 1.2° in diameter) that appeared at the center of the touch-sensitive screen. Their eye movement/position was monitored using an infrared eye-camera system (R-21C-AS, RMS, Hirosaki, Japan). If fixation continued for 1,200 ms, the monkey was given the first instruction (the 1st cue; 400-ms duration), which contained information about either the target location or which arm to use. A small, colored cue superimposed on the central FP indicated the type of instruction, i.e., whether it related to target location or arm use. For monkey 1, a green circle or red square indicated an arm-use instruction, whereas a blue circle or red cross indicated a target-location instruction. For monkey 2, a green square and a blue cross indicated instructions for arm use and target location, respectively. At the same time, a white square (8 x 8°) appeared to the left or right of the FP and indicated laterality of arm use (for arm use-related instructions) or target location (for target-related instructions). If fixation continued for 1,200 ms during the subsequent delay period (1st delay), the second instruction (the 2nd cue; 400 ms) was given to complete the information required for the subsequent action. Thereafter if fixation continued for 1,200 ms during the second delay, squares appeared on each side of the FP (set cue 1,000 ms), instructing the monkey to prepare to reach for the target when the FP disappeared (the GO signal). If the monkey subsequently reached for the target with a reaction time of <1 s, it received a reward of fruit juice 600 ms after touching the screen. Monkey 1 was required to gaze at the FP for 800–1,200 ms before the GO signal appeared. Monkey 2 was also required to gaze at the fixation point until the end of the second delay but not required to do so after the onset of the set cue. The order of appearance of the target and arm instructions was alternated in a block of 20 trials, and laterality was randomized within each block. A series of five 250-Hz tones after a reward signaled reversal of instruction orders.
View larger version (21K):
[in this window]
[in a new window]
|
FIG. 1. Temporal sequence of behavioral events and temporal changes in neuronal activity in the dorsal and ventral premotor cortex (PMd and PMv). A: temporal sequence of behavioral events. Top: trial in which two instructions were given, i.e., which arm to use ("arm") and which target to reach ("target"), in that order. Bottom: trial in which the two instructions were given in the reverse order. B and C: temporal changes in neuronal activity during the behavioral task. Thick line, fraction of neurons that exhibited a significant increase or decrease in activity compared with the control period. Thin solid and dotted lines, fractions of neurons that exhibited a significant increase or decrease in activity compared with the control period, respectively. Gray areas, task periods during which visual cues were presented (from left to right: the 1st and 2nd cues). The numbers at the top are the actual numbers of neurons that exhibited a significant change in activity during each behavioral period. B: data for the PMd. C: data for the PMv. Note that an individual neuron could exhibit an increase in activity during 1 sequence of 1st and 2nd cues and a decrease in activity during another sequence, i.e., the sum of the fractions could exceed the total number of neurons.
|
|
Data analysis
Definition of Task-Related Neurons and the 10 Task Periods.
We sampled all neurons for which activity was recorded during at least four blocks of trials (i.e., 80 trials). To apply a task-related definition to neuronal activity, we divided the behavioral task into the following six phases: control (200–700 ms after attaining fixation); prefirst cue (the 500-ms period before the appearance of the 1st cue); first cue and delay (100 ms after onset of the 1st cue to the onset of the 2nd cue); second cue and delay (100 ms after onset of the 2nd cue to the onset of the set cue); set cue (onset of the set cue to the appearance of the GO signal); and movement (the 500-ms period around the time at which movement began). We defined a neuron as "task-related" if distribution of the discharge rate (spikes per second) differed significantly in at least one of eight trial types [ANOVA, P < 0.05, repeated over 8 types of trials with 8 sequences of 1st and 2nd cue presentation (as shown in 8 panels in Fig. 2)]. For statistical analysis and display, data for the five task events were aligned separately to onsets of first and second cues, the set cue, the GO signal, and the time at which the screen was touched. These data were analyzed separately before being merged at the midpoint of the first and second delays and at the set cue phase (i.e., 600 ms after cue offset and 600 ms before the onset of the 2nd cue or the set cue phase, and 600 ms after the set cue onset and 600 ms before the GO signal).
View larger version (41K):
[in this window]
[in a new window]
|
FIG. 2. An example of PMv neuronal activity presented with raster displays and plots of spike density functions (SDFs). Gray areas (from left to right) represent when the first, second, and set cues were presented. Tic marks on the abscissa are at 400-ms intervals. First and second instructions are shown on top of each panel (RA, right arm; LA, left arm; RT, right target; LT, left target). SDFs (Gaussian kernel,  = 20 ms, mean ± SE) appear below each raster display. Raster plots and SDFs were aligned to the onset of the first and second instructions, and the onset of the set cue, and were merged at the midpoint of each delay period (600 ms after the disappearance of the cue and 600 ms before the onset of the second or set cue). Ordinate represents the instantaneous firing rate, the degree of which is indicated on the ordinate. This neuron mainly reflected the left position of the white square appearing in instruction cues, soon after the appearance of the first and second cues. In addition, in the latter part of the second delay period, the activity started reflecting the future reach target location (left) regardless of arm use.
|
|
Our subsequent statistical analysis of neuronal activity properties involved dividing the entire task into one control period (200–700 ms after attaining fixation) and 10 task periods defined as follows: precue (the 500-ms period before onset of the 1st cue); first cue (100–500 ms after onset of the 1st cue); early first delay (500–1,000 ms after onset of the 1st cue); late first delay (the last 500 ms before onset of the 2nd cue); second cue (100–500 ms after onset of the 2nd cue); early second delay (500–1,000 ms after onset of the 2nd cue); late second delay (the last 500 ms before onset of the set cue); early set cue (the 500-ms period after onset of the set cue); late set cue (500 ms before the appearance of the GO signal); and movement (the 500-ms period before the screen was touched).
Statistical analysis using interspike intervals
To analyze neuronal activity with high temporal resolution, we first calculated the instantaneous firing rate as the inverse of the interspike interval (inverse-ISI; 1-ms resolution). When a spike occurred, the firing rate was updated and stored in the subsequent 1-ms data points until the next spike occurred, which, in turn, renewed the subsequent data points. Although original inverse-ISI data were calculated with 1-ms resolution, statistical analyses were applied for data re-sampled from the original data set at every 10 ms (i.e., 10-ms bin). We did this because the firing rate generally remained <100 Hz. As the rate of neuronal discharge tended to follow a Poisson distribution, the inverse-ISI data were square root-transformed to stabilize the variance (Zar 1999).
To estimate whether neuronal activity reflected information contained in the first or second cue, or their combination, we used one-way ANOVA. We examined how well each of the following equations accounted for neuronal activity
 | (1) |
 | (2) |
 | (3) |
In Eqs. 1–3, the firing rate index is for inverse-ISI data that were sampled every 10 ms; 
0 is the intercept and a, b, and c are coefficients. Categorical factors for the first and second cues are the four instructions provided by the cues (right arm, right target, left arm, and left target). Categorical factors for the combination of the first and second cues are the four possible combinations of arm use and target location provided by the first and second cues. First, we calculated the probability (P value) that the coefficient in each equation equaled zero. We calculated P values for each 10-ms time point (i.e., for each bin) by creating an algorithm that was executed using commercial software (MATLAB 6.5, MathWorks, Natick, MA). The threshold used for statistical significance was set at 
= 0.01. Then, we calculated the sum of squares (SS) among groups and divided this value by the total SS to obtain the SS ratio; these SS values were obtained using ANOVA tables. The SS ratio was analyzed for each 10-ms data bin. The larger the SS ratio, the better the neuronal activity reflected the equation factor (i.e., the 1st cue, the 2nd cue, or the combination of the 2 cues). Based on the analysis of probability and the SS ratio, we classified the neurons into four categories according to whether instantaneous activity was best and significantly represented by the first cue, second cue, combined information from both cues, or whether none of the regression coefficients significantly differed from zero. This classification was used for data in every 10-ms bin. We also applied multi-way ANOVA to the inverse-ISI data to categorize neuronal activity with high temporal resolution. The next section details this analysis.
|
RESULTS
|
|---|
Neuronal database
We recorded neuronal activity in the postarcuate premotor areas of monkeys performing the behavioral task described in METHODS. Success rates recorded during the behavioral task exceeded 96% for both monkeys. We delineated the PMd from the PMv using the spur of the arcuate sulcus; the area dorsal to the spur was defined as the PMd and the area ventral to the spur was defined as the PMv. We found a total of 1,016 task-related neurons in the PMd (n = 505 in monkey 1; n = 511 in monkey 2) and 358 task-related neurons in the PMv (n = 204 in monkey 1; n = 154 in monkey 2). Because previous research has examined properties of neuron activity in the PMd and PMv during the motor set period (Hoshi and Tanji 2002), we focused on neuronal activity relevant to reception of visual cues and to subsequent processing to achieve action planning. For this report, we constructed a database using neurons that exhibited significantly changed activity in at least one of the first seven task periods (i.e., from precue to late 2nd delay; see METHODS for task period definitions) compared with the control period (paired t-test, = 0.05, corrected for 8 trial types but not corrected for 7 task periods). Consequently, this report examines the response properties of 825 PMd and 223 PMv neurons. Figure 1, B and C, shows the fraction of neurons in the database that changed activity during each task period. The neurons in both areas changed their activity by >35% during these task periods (represented by thick lines); the fraction peaked when the second cue was presented. These results suggest that both the PMd and PMv were actively involved in the earlier part of the behavioral task when the animal was required to detect visuospatial information (i.e., the position of the white square), to retrieve a specific motor instruction from the visual cue, and to combine the arm-use and target-location instructions to plan the future action.
Activity in anticipation of a cue's appearance
Before the appearance of the first cue (i.e., during the precue period), 218 (26%) PMd and 36 (16%) PMv neurons exhibited activity significantly changed compared with the control period (Fig. 1, B and C). It is possible that anticipatory activity reflected specific expectations for the appearance of the first instruction because instructions for arm use and target location were presented in a fixed order within each block of 20 trials. For example, the neuron illustrated in Fig. 3A showed greater activity before the appearance of the arm instruction than the target instruction. To investigate this possibility, we applied a two-sample t-test for activity within the precue period (factor: order of instructions). Of the 218 PMd neurons that exhibited anticipatory activity, 28 neurons (12%) exhibited activity that significantly differed depending on the forthcoming first instruction (2-sample t-test, = 0.01). Selective neurons had a median difference of 5.1 spikes per second (range: 2.6–12.6 spikes/s). In contrast, of the 36 PMv neurons that exhibited anticipatory activity, only 1 neuron (2%) exhibited a significant difference in activity (2-sample t-test, = 0.01). Although the difference in distribution between the PMd and PMv was not significant (P = 0.0922, Fisher's exact test for count data) due to the small PMv sample size, results suggest that specific expectations of a forthcoming cue tended to be represented more frequently in the PMd than in the PMv.
View larger version (38K):
[in this window]
[in a new window]
|
FIG. 3. Two examples of PMd neuronal activity. Display formats are as in Fig. 2. A: this neuron showed vigorous activity if the first cue, but not the second cue, instructed use of the right arm. The activity preceding the appearance of the first cue was greater before the appearance of the arm instruction (left column) than the target instruction (right column). B: this neuron showed vigorous activity if the first cue, but not the second cue, instructed reach to the right target.
|
|
Differences in neuronal activity in the PMd and PMv during the first cue and first delay periods
After the first cue appeared, >35% of neurons in the PMd and PMv exhibited activity significantly changed compared with the control period (Fig. 1, B and C). We found three classes of neuronal activity in the PMd and PMv. The first class of activity reflected only the visuospatial feature of the first cue (i.e., the position of the white square). In the example shown in Fig. 2, the PMv neuron showed visual responses if a white square appeared to the left of the fixation point, regardless of whether the cue instructed the arm use or reach target. The second class of activity responded to the instruction for the arm use. In the example shown in Fig. 3A, the PMd neuron responded preferentially to the right arm instruction. Finally, the third class of activity responded to the instruction for the target location. In the example shown in Fig. 3B, the PMd neuron responded preferentially to the right target instruction.
To analyze how PMd and PMv neuron activity represented information provided by the first cue, we examined neuronal activity using two-way ANOVA ( = 0.01); the two categorical factors (i.e., independent variables) included the position of the white square (position: left or right) and the instruction type (instruction: arm use or target location). The square-rooted activity of each 10-ms bin obtained by inversing the ISI (see METHODS) was used as a dependent variable. We applied this analysis to every 10-ms bin of activity for all neurons in the database (n = 825 in the PMd; n = 223 in the PMv).
Figure 4, A and B, presents the fraction of neurons that exhibited significant selectivity to the cue position (i.e., the position of the white square; position < 0.01 or instruction*position < 0.01). During the first-cue period and the following delay period, we found position selectivity continuously in 30% of PMd neurons (Fig. 4A, —), and in 20% of PMv neurons (Fig. 4B, —). Although the PMv neuron illustrated in Fig. 2 did not show vigorous activity during the delay period, we found that a group of neurons in PMv showed sustained delay period activity. These results suggest that the selectivity for the spatial location of the visual cue (right or left) was reflected well in both areas during the cue and delay periods. For these cue-position selective neurons, we studied how many neurons were selective only for the position of the white square and not for the type of instruction (2-way ANOVA: position < 0.01, instruction 0.01, position*instruction 0.01). We considered these neurons to represent only a cue's visuospatial information rather than its motor instruction. The · · · in Fig. 4, A and B, illustrates results. In the PMv (Fig. 4B), 81% of cue-position selective neurons, on average, were judged as visuospatial, whereas in the PMd, only 54% were visuospatial (Fig. 4A).
View larger version (35K):
[in this window]
[in a new window]
|
FIG. 4. Time course of the cue-position selectivity during the 1st-cue and delay periods. A and B: bin-by-bin plots of position selectivity expressed as the fraction of all neurons in the database (n = 825 in PMd, n = 223 in PMv).  , when the cues appeared. —, fraction of neurons that were position-selective (2-way ANOVA, position < 0.01 or position*instruction < 0.01), calculated successively for each 10-ms bin. · · · , fraction of neurons that were position-selective only (not instruction-selective; 2-way ANOVA, position < 0.01, instruction 0.01, and position*instruction 0.01). A and B are data for neurons in the PMd and PMv, respectively. The tick marks on the horizontal axis represent 400-ms intervals.  , onset latency, i.e., the time at which the fraction of selective neurons first exceeded 10% of the total population. The time below the indicates the actual latency (latency for cue-position selectivity, followed by the latency for visuospatial selectivity in the parenthesis). C and D: time course of instruction selectivity during the 1st-cue and delay periods. C and D: bin-by-bin plot of instruction selectivity expressed as the fraction of all neurons in the database. —, fraction of neurons that exhibited arm- or target-instruction selectivity, calculated successively for each 10-ms bin (2-way ANOVA, instruction < 0.01 or position*instruction < 0.01). · · · , fraction of neurons that exhibited greater activity for the arm instruction. , onset latency, i.e., the time at which the fraction of neurons selective for the instruction 1st exceeded 10% of the total population. The time below the indicates the actual latency.
|
|
Subsequently, we analyzed selectivity for instruction (arm use or target location). Figure 4, C and D, shows the fraction of neurons in each 10-ms bin that appeared to be significantly selective for the type of instruction (2-way ANOVA: instruction < 0.01 or position*instruction < 0.01). More than 20% of neurons in the PMd were instruction-selective. Of these, an average of 50% preferred arm instruction (represented by the dotted line in Fig. 4C) and 50% preferred target instruction (represented by the difference between · · · and — in Fig. 4C). In contrast, we only rarely found instruction selectivity in the PMv (Fig. 4D), supporting our judgment that an average of 81% of position-selective PMv neurons represented only visuospatial information (Fig. 4B). A 
2 test ( = 0.01) confirmed that the PMd exhibited more frequent target instruction-selective activity (detected in 136 bins of 160 10-ms bins during the cue and delay periods). Arm-selective activity was also more frequent in the PMd (148 bins). In 
5% of PMd neurons, instruction selectivity appeared even before the onset of the first cue (anticipatory activity, Fig. 4C; see the above section).
We analyzed the timing of the onset of activity changes in response to the first cue. We defined the onset of cue-selective activity as the time when the fraction of selective neurons first exceeded 10% of the total neuron population (i.e., all neurons in the data base, n = 825 in PMd and n = 223 in PMv). The onset of position selectivity was 100 ms in the PMv and 140 ms in the PMd (Fig. 4, A and B, —). The onset of visuospatial selectivity (i.e., selective only for the position, and not for the instruction) was 110 ms in the PMv and 150 ms in the PMd (Fig. 4, A and B, · · · ). Therefore visuospatial selectivity appeared 40 ms earlier in the PMv than in the PMd. The onset of instruction selectivity appeared at 200 ms in the PMd (Fig. 4C), 50 ms later than the onset of visuospatial selectivity (150 ms) in the same area. The onset of instruction selectivity in the PMv appeared at 1,540 ms (Fig. 4D).
These results suggest a difference in neuronal response properties between the PMd and PMv. PMv neurons mainly represented a cue's visuospatial information, whereas PMd neurons amply retrieved instruction contents along with the visuospatial information, suggesting that PMd neurons represented more processed information than PMv neurons. Data on response latencies support this view because the latency of the visuospatial activity was 40 ms longer in the PMd than in the PMv, and the latency of instruction selectivity in the PMd was 50 ms longer than that of the visuospatial selectivity.
Laterality of responses during the first cue and first delay periods
Because we made the task symmetrical for right–left laterality, it was possible to analyze the preferred side of the visuospace or instruction by recording neuronal activity in a single hemisphere. We studied side preference for neurons judged to be selective for visuospatial or instruction information. Figure 5, A and B, shows the results for visuospatial activity (i.e., neuronal activity selective only for the cue position, and not for the instruction). In the PMd (Fig. 5A), neuronal activity that preferred the contralateral side developed earlier than ipsilateral-preferring activity; response latency was 140 ms for contralateral-preferring activity and 170 ms for ipsilateral-preferring one. In this analysis, the threshold of the response latency was halved into 5% of the total population (i.e., all neurons in the data base, n = 825 in PMd and n = 223 in PMv) because an original group of selective neurons was subdivided into right- and left-selective groups. Fractions of neurons preferring either side reached similar levels at 350 ms and this trend continued throughout the delay period. In the PMv (Fig. 5B), contralateral-preferring visuospatial activity had a latency of 100 ms and ipsilateral one had a latency of 150 ms. Fractions of ipsilateral- and contralateral-preferring activity did not differ after 1,000 ms from when the first cue appeared. These results suggest that visuospatial information from the contralateral side arrived earlier in both the PMd and PMv and that the representation of space eventually became bilateral. We used the same method to study the laterality preference of instruction-selective neurons in the PMd (Fig. 5C). Contralateral- and ipsilateral-preferring activity had response latencies of 220 and 250 ms, respectively. Neurons preferring the contralateral target or arm tended to dominate. For arm-selective activity, 59% on average showed greater activity for the contralateral arm-use instruction. For target-selective activity, 54% on average showed greater activity for the contralateral target instruction. However, because >40% of neurons preferred ipsilateral arm use or target location, PMd neurons in one hemisphere could be viewed as representing instructions for bilateral arm use and target location.
View larger version (43K):
[in this window]
[in a new window]
|
FIG. 5. Time course of ipsi- or contralateral selectivity during the 1st-cue and delay periods. A: bin-by-bin plot of laterality selectivity for visuospatial PMd neurons (selective only for the cue position but not for the instruction) as the fraction of all neurons in the database. The gray line represents the total fraction of visuospatial neurons, reproduced from the dotted line in Fig. 4A. Please note that the scale of the y axis is different from that in Fig. 4A. The solid black line represents the fraction of neurons that exhibited greater activity for the contralateral white-square position. The dotted line represents the fraction of neurons that exhibited greater activity for the ipsilateral white-square position. The triangles at the bottom of the panel indicate the response latency (i.e., when the fraction exceeded 5% of the total) for neurons the activity of which was greater for the contralateral (black triangle) or ipsilateral (gray triangle) side. The actual latencies are indicated in parentheses using the format (latency for the contralateral, latency for the ipsilateral). B: bin-by-bin plot of laterality selectivity for visuospatial PMv neurons displayed using the same format as in A. C: bin-by-bin plot of laterality selectivity for instruction-selective PMd neurons as a fraction of all the neurons in the database. The gray line represents the total fraction of instruction-selective neurons, reproduced from the solid line in Fig. 4C. The black line represents the fraction of neurons that exhibited greater activity for the contralateral instruction (arm-use or target-location). The dotted line represents the fraction of neurons that exhibited greater activity for the ipsilateral instruction.
|
|
Location of neurons responding to the first cue
Figure 6A shows a cortical surface map indicating the recording site covering both the PMd and PMv where we sampled neuronal activity with electrode penetrations spaced with 1 mm (gray area). A thick dotted line indicates the border between the PM and the primary motor cortex (MI), operationally defined according to the criteria established previously (Fogassi et al. 2001; Kurata 1993). In this map, we plotted the locations of all task-related neurons by classifying their activity (at the end of the 1st cue period) into one of three categories: visuospatial-selective (2-way ANOVA, position < 0.01, instruction 0.01 and position*instruction 0.01), instruction-selective (instruction < 0.01 or position*instruction < 0.01), and nonselective (instruction 0.01, position 0.01, and position*instruction 0.01). In the PMd, we found both visuospatial- and instruction-selective neurons predominantly in the rostrocaudal extent. Interestingly, quantities of these neurons decreased abruptly at penetrations 3–4 mm more rostral to the genu of the arcuate sulcus. In the PMv, we found visuospatial neurons predominantly in the caudal bank and its lip region of the arcuate sulcus ventral to the spur. Seventy percent of the visuospatial PMv neurons were recorded >1,000 µm (1,893 ± 1,244, mean ± SD; range, 606,600 µm) below the depth where we first encountered neuronal activity. The neuron shown in Fig. 2 was found at the depth of 2.4 mm.
View larger version (33K):
[in this window]
[in a new window]
|
FIG. 6. Anatomical location of visuospatial and instruction neurons. A: location of each neuron is rendered on the cortical surface. This figure shows data for the end of the first cue period. For the purpose of display, random noise of < ±0.4 mm was added to the original data, except for the data crossing the PMd–PMv border. The 2 left hemispheres from the 2 monkeys are superimposed. Red circles, instruction-selective neurons (2-way ANOVA, instruction < 0.01 or position*instruction < 0.01). Blue squares, visuospatial neurons (2-way ANOVA, position < 0.01, instruction 0.01, and position*instruction 0.01). Bar, a nonselective neuron (2-way ANOVA, position 0.01, instruction 0.01, and position*instruction 0.01). The gray area in the cortical surface indicates the recording site where neurons were sampled with electrodes inserted at 1-mm intervals. B and C: rostrocaudal distribution of instruction-selective or visuospatial neurons in the PMd. The number of neurons obtained in each 1-mm width of the PMd in the mediolateral plane is plotted in the rostrocaudal direction. Red line, instruction-selective neurons. Blue line, visuospatial neurons. Black dotted line, instruction-selective neurons with greater activity for the arm-use cue. Gray solid line, instruction-selective neurons with greater activity for the target-location cue. B shows the data at the end of the 1st cue period, whereas C shows the data at the end of the 1st delay period.
|
|
We performed a more precise analysis of the rostrocaudal distribution of the two classes of PMd neurons by measuring the distance in the rostral direction relative to the genu of the arcuate sulcus and counting the number of classified neurons within each 1-mm width of cortex positioned rostrocaudally. Figure 6, B and C, shows results obtained at two different times after the cue (Fig. 6B represents the end of the 1st cue period and Fig. 6C represents the end of the 1st delay period). We found visuospatial neurons (represented by blue trace) at both times, predominantly in the rostrocaudal extent of the PMd. In contrast, we found instruction-selective neurons (represented by red trace) more often rostral to the genu of the arcuate sulcus. Of these, neurons preferring arm instruction (represented by the dotted line) and target instruction (represented by the gray line) overlapped in their distribution.
Differences in PMd and PMv neuronal activity during the second cue and second delay periods
After the appearance of the second cue, >50% of neurons in both the PMd and PMv exhibited changes in activity compared with the control period (Fig. 1, B and C). We found that PMd neurons primarily reflected a specific combination of the two instructions, rather than the second cue itself. For example, the PMd neuron shown in Fig. 7A reflected left-arm use in the forthcoming action. On the other hand, the PMd neuron shown in Fig. 7B reflected right-target reach. In contrast, PMv neurons tended to reflect the visuospatial property of the second cue. For instance, the PMv neuron shown in Fig. 2 responded if the second cue contained a white square to the left of the fixation point. We also found that information for the forthcoming action gradually developed in the PMv during the second delay period and that the activity preferentially reflected the location of reach target. For example, the PMv neuron shown in Fig. 2 started reflecting the left target reach during the late phase of the second delay period.
View larger version (46K):
[in this window]
[in a new window]
|
FIG. 7. Two examples of PMd neuronal activity. Display formats are as in Fig. 2. A: this neuron showed greater activity if a white square of the 1st cue appeared to the left of the fixation point. After the appearance of the 2nd cue, however, the same neuron reflected left arm use of the future movement. B: this neuron showed brief visual responses if the 1st cue indicated right target reach. During the 1st delay period, the activity was greater if the 1st cue indicate the use of the left arm. After the appearance of the 2nd cue, this neuron showed greater activity if the animal was required to reach the right target.
|
|
We applied regression analysis to neuronal activity in 10-ms bins (see Eqs. 1–3) and were able to classify each neuron into one of four categories based on whether its activity was significant and most selective for the first cue, the second cue, a combination of cues, or was nonselective. We repeatedly calculated the fraction of neurons assigned to each of the four categories for successive 10-ms bins.
Figure 8A shows a bin-by-bin plot of the PMd neuron fractions that exhibited significantly modulated and most selective activity for the first cue (black trace), second cue (blue trace), and a combination of cues (red trace). After the first cue appeared, the PMd neuronal activity that was selective for the first cue (including both the visuospatial and instruction selectivity) developed quickly and became dominant; the first cue-selective activity was most frequently found from 130 ms after the first cue presentation to 200 ms after the second cue presentation (the black tick marks at the top of the panel; 2 test, = 0.01). After the second cue appeared, the fraction of first cue-selective neurons decreased rapidly. The decrease or disappearance of the first-cue selective activity, after the appearance of the second cue, is apparent in the examples of PMd neurons illustrated in Fig. 3, A and B, and also in Fig. 7, A and B. By contrast, the fraction of neurons that was selective for the second cue (blue) or combination of cues (red) quickly increased. This property is illustrated for PMd cells shown in Fig. 7, A and B. In these examples, activity began to reflect the forthcoming action, already while the second cue was still presented.
Combination-selective neurons became dominant at 220 ms after the appearance of the second cue when the second cue was still presented (the red tick marks at the top of the panel; 2 test, = 0.01). Simultaneously, the fraction of neurons selective for the second cue decreased rapidly. Selectivity for the combination of cues (red trace) reached a plateau (25% of all PMd neurons) within 400 ms after onset of the second cue. Representation of the first cue, which was dominant during 200 ms after the second cue onset, turned out to represent an action (i.e., the combination of 2 instructions) at 220 ms after the second cue was presented.
Figure 8B shows a bin-by-bin plot of PMv data using the same format as in Fig. 8A. After the appearance of the first cue, the fraction of PMv neurons selective for the first cue increased quickly. The first cue-selective activity was most frequently found from 110 ms after the first cue onset to 120 ms after the second cue onset (the black tick marks at the top of the panel). After the appearance of the second cue, the fraction of first cue-selective neurons decreased rapidly, whereas the fraction of neurons selective for the second cue or combination of the cues rapidly increased. An example of the second-cue response of a PMv neuron is shown in Fig. 2, where responses were apparent if the second cue displayed a white square in the left side. The fraction of PMv neurons selective for the second cue was greater than the fraction of PMd neurons (56 10-ms bins, represented by the blue tick marks in Fig. 8C; 2 test, = 0.01), suggesting that the second cue's visual feature had a relatively greater impact in the PMv than in the PMd. During the second delay period, the fraction of neurons selective for the combination increased gradually, whereas the fraction of neurons selective for the second cue decreased; the PMv neuron illustrated in Fig. 2 started reflecting the left target reach toward the end of the second delay period. The first 10-ms bin when the combination-selective activity became dominant was at 400 ms (the red tick marks at the top of the B), which was 180 ms later compared with the PMd. The 10-ms bins in which the combination-selective activity was found most frequently increased progressively toward the end of the delay period.
Next, we analyzed the timing of the onset of changes in PMd and PMv neuron activity in response to the first and second cues using the data shown in Fig. 8, A and B. We defined "onset of activity" as the time when the fraction of selective neurons first exceeded 10% of the total neuron population. In this study, onset of selectivity for the first cue in the PMv was 110 ms after the appearance of the first cue, and in the PMd, it was 140 ms after the appearance of the first cue. The onset of selectivity for the second cue was 150 ms after the appearance of the second cue in both areas, and the onset of combination selectivity was 170 ms after the appearance of the second cue in the PMv and 140 ms after the appearance of the second cue in the PMd. These data reveal that while information about the cue reached the PMv earlier (110 ms in the PMv vs. 140 ms in the PMd after the 1st cue), information about action developed earlier in the PMd (170 ms in the PMv vs. 140 ms in the PMd after the appearance of the 2nd cue).
To examine the selectivity for the second cue position and for the combination of two cues, we quantitatively compared neuronal activity in the PMd and PMv. We applied the analysis to neurons selective for the first cue, second cue, or combination of two cues at the end of the second cue period (n = 386 in PMd and n = 67 in PMv; see Fig. 8, A and B). To evaluate position selectivity of the second cue, we applied two-way ANOVA with two factors (the position of the white square in the 1st and 2nd cues) on the square-root transformed neuronal activity at the end of the second cue period. From the ANOVA table, we estimated spatial selectivity by calculating the sum of squares (SS) between groups (SS-bg) and dividing this value by the total SS (SS-total) to obtain the SS ratio
 | (4) |
The greater this value was, the greater the selectivity for the second-cue position; Fig. 9A shows results of the comparison of spatial selectivity in the PMd and PMv. The SS ratio for the position of the second cue was significantly greater for the PMv neurons than for the PMd neurons (Kolmogorov–Smirnov test, KS = 0.3101, P < 0.0001), indicating that spatial selectivity was represented more in the PMv than in the PMd.
View larger version (21K):
[in this window]
[in a new window]
|
FIG. 9. Comparison of selectivity of the 2nd cue responses for PMd and PMv neurons. A: comparison of position selectivity for the 2nd cues in PMd and PMv neurons. The cumulative fractions of the SS ratio for the position of the second cue for the PMd (n = 386, · · · ) and PMv (n = 67, —) are shown. B: comparison of 2-cue combination selectivity for PMd and PMv neurons. The cumulative fractions of the SS ratio for the combination, for the PMd (n = 386, · · · ) and PMv (n = 67, —), are shown.
|
|
To evaluate the two-cue combination selectivity, we applied the following linear regression model on activity
 | (5) |
In this formula, the firing rate index is calculated at the end of the second cue period, 0 is the intercept, and 1 and 2 are coefficients. Categorical factors of the second cue (CUE2) are the 4 instructions that could be conveyed by the second cue (right-arm, right-target, left-arm, and left-target). Categorical factors of the two-cue combination (COMBINATION) are the 4 possible combinations of the two instructions given by the first and second cues. We calculated the sum of squares (SS) between groups (SS-bg) and divided this value by the total SS (SS-total) to obtain the SS ratio. We calculated the SS ratio as follows
 | (6) |
The greater this value was, the greater the selectivity for the second-cue position; Fig. 9B shows result of the comparison of combination selectivity in the PMd and PMv. The SS ratio for the two-cue combination was significantly greater for the PMd neurons than for the PMv neurons (Kolmogorov–Smirnov test, KS = 0.3302, P < 0.0001), indicating that combination selectivity was represented more in the PMd than the PMv.
Location of neuronal selectivity after the appearance of the second cue
We mapped the locations of neurons classified as first-cue selective, second-cue selective, combination selective, and nonselective after onset of the second cue, using the same method shown in Fig. 6A. Figure 10, A and B present two maps based on data obtained at the end of the second cue period (Fig. 10A) and at the end of the second delay period (Fig. 10B). In the PMd, we found the three classes of neurons (first-cue selective, black circles; second-cue selective, blue squares; combination selective, red crosses) located broadly in the rostrocaudal extent at the end of the second cue period. At the end of the second delay period (Fig. 10B), we again found the combination-selective neurons located broadly in the PMd. In the PMv, we found the three classes of neurons mainly in the caudal bank and its lip region of the arcuate sulcus ventral to the spur. These neurons were located in areas of the PMd and PMv similar to the map shown in Fig. 6A.
View larger version (55K):
[in this window]
[in a new window]
|
FIG. 10. Anatomical location of neurons selective for the first cue, second cue, or combination of the two cues. The location of each neuron was rendered on the cortical surface. For the purpose of display, random noise of < ±0.4 mm was added to the original data, except for the data crossing the PMd–PMv border. A: data at the end of the second cue period. B: data at the end of the second delay period. The two left hemispheres from the two monkeys are superimposed. Black circles, first cue-selective neurons. Blue squares, second cue-selective neurons. Red crosses, combination-selective neurons. Bar, a nonselective neuron. C and D: rostrocaudal distribution of three classes of neurons in the PMd. The number of neurons obtained in each 1-mm width of the PMd in the mediolateral plane is plotted in the rostrocaudal direction. Black line, the first cue-selective neurons. Blue line, the second cue-selective neurons. Red line, combination-selective neurons. C: data at the end of the second cue period. D: data at the end of the second delay period.
|
|
We analyzed the localization of the three classes of PMd neurons more precisely in the rostrocaudal direction. Figure 10, C and D present results based on the two sets of data (Fig. 10C, end of the second cue period; Fig. 10D, end of the second delay period). At the end of the second cue period, first cue- (black) or second cue- (blue) selective neurons were more likely to be located rostrally. Combination-selective neurons were found broadly in the PMd at both times.
Development of information for arm use and target location
The monkeys were required to integrate two sets of information, i.e., arm use and target location, to plan a future action. The time course of their development in the PMd and PMv was of interest, which we examined by applying a three-way ANOVA to the neurons we judged to be most selective for a combination of the two cues (the red trace shown in Fig. 8, A and B). The three factors included arm use (ARM), target location (TARGET), and the order of the two instructions (ORDER). Based on this analysis, we classified neurons into four categories: selective for arm use (ARM <0.01 or ARM*TARGET <0.01), selective for target location (TARGET <0.01 or ARM*TARGET <0.01), selective for both arm use and target location (ARM <0.01 and TARGET <0.01, or ARM*TARGET <0.01), and nonselective (ARM 0.01, TARGET 0.01, and ARM*TARGET 0.01). We applied this analysis to instantaneous activity during every 10-ms bin (see the Methods section) if it had been classified as combination selective because we wanted to exclude activity primarily involved in detecting the visuospatial signal and in representing partial motor instruction provided by the first or second cue. In the PMd (Fig. 11A), the fractions of neurons selective for target location or arm use increased simultaneously in response to the second cue and remained at approximately equal levels throughout the second-cue and delay periods. In the PMv (Fig. 11B), while arm-use and target-location representations developed simultaneously, information about target location became progressively dominant toward the end of the delay period. Therefore although arm-use and target-location representations were similar and stable in the PMd population, information about target location grew progressively in the PMv during the delay period.
View larger version (57K):
[in this window]
[in a new window]
|
FIG. 11. Time course of neuronal activity representing arm-use and target-location for neurons defined as selective for the combination of the two cues. A and B: bin-by-bin plot of selective activity expressed as the fraction of neurons that were selective for target-location (blue lines; three-way ANOVA, TARGET <0.01 or ARM*TARGET <0.01), arm-use (red lines; three-way ANOVA, ARM <0.01 or ARM*TARGET <0.01), and both target-location and arm-use (thick black lines; three-way ANOVA, ARM <0.01 and TARGET <0.01, or ARM*TARGET <0.01). The gray traces in A and B correspond to the red traces in Fig. 8A and B (i.e., the two-cue combination selective neurons), respectively. A: data for PMd neurons. B: data for PMv neurons. C-F: time course of ipsilateral or contralateral selectivity for arm-use and target-location during the second cue and delay periods. C: bin-by-bin plot of laterality selectivity for arm-use-selective PMd neurons as the fraction of all neurons. The red line represents the fraction of arm-use-selective neurons, reproduced from the red line in A. Note that the scale of the y-axis differs from that in A. The solid line represents the fraction of neurons that exhibited greater activity for contralateral arm-use. The dotted line represents the fraction of neurons that exhibited greater activity for ipsilateral arm-use. The triangles at the bottom of the panel indicate the response latency (i.e., when the fraction exceeded 5% of the total population) for the contralateral-preferring (black triangle) and ipsilateral-preferring (gray triangle) neurons. The actual latencies are indicated in parentheses using the format (latency for the contralateral, latency for the ipsilateral). D: bin-by-bin plot of laterality selectivity for target location-selective PMd neurons displayed using the same format as in B. The blue line, representing the fraction of target-location selectivity, was reproduced from the blue line in A. The solid line represents the fraction of neurons that exhibited greater activity for the contralateral target-location. The dotted line represents the fraction of neurons that exhibited greater activity for the ipsilateral target-location. E: bin-by-bin plot of laterality selectivity for arm use-selective PMv neurons as the fraction of all neurons. The display format is the same as in C. The red line, representing the fraction of arm use-selective PMv neurons, was reproduced from the red line in B. F: laterality selectivity for target location-selective PMv neurons. The display format is the same as in D. The blue line, representing the fraction of target location-selective neurons, was reproduced from the blue line in B.
|
|
Laterality of arm-use and target-location representations
We analyzed side preference (i.e., ipsilateral or contralateral to the recording hemisphere) of information related to arm use or target location by examining whether the left (ipsilateral) or right (contralateral) side led to greater neuron activity related to future action (i.e., combination-selective neurons, see Fig. 8, A and B). Figure 11, C and D present the results for PMd neurons. Arm-selective neurons (Fig. 11, C) exhibited a response latency of 160 ms for contralateral arm-selective activity, versus 200 ms for the ipsilateral arm. This analysis used 5% of the total population as the response latency threshold. Therefore activity preferring the contralateral arm developed 40 ms earlier than the activity preferring the ipsilateral arm. Fractions of neurons preferring contralateral or ipsilateral arms reached similar levels while the second cue was still being presented, and this trend continued throughout the second delay period. During the second-cue and delay periods, on average, 49% of activity preferred the contralateral arm. Response latencies for target-selective activity were 140 ms for the contralateral target and 190 ms for the ipsilateral target (Fig. 11D). The fraction of neurons preferring the contralateral target dominated slightly throughout the delay period; 61% of activity preferred the contralateral target during the second-cue and delay periods.
Figure 11, E and F present results for PMv neurons. Response latencies for the arm-selective neurons (Fig. 11E) were 180 ms for the contralateral arm and 200 ms for the ipsilateral arm. During second-cue and delay periods, on average, 50% of activity preferred the contralateral arm while the remaining 50% preferred the ipsilateral arm. When we examined target-selective activity, we found that response latencies were 180 ms for the contralateral target and 200 ms for the ipsilateral target (Fig. 11F). The fraction of neurons preferring the contralateral target dominated slightly; during the second-cue and delay periods, 60% of activity preferred the contralateral target.
Localization of arm-use and target-location selectivity in the PMd and PMv
We examined the distribution of neurons selective for arm use and/or target location. Figure 12A presents the selectivity distribution of neuronal activity measured at the end of the second delay. In the PMd, neurons selective only for target location (represented by blue squares; three-way ANOVA: ARM 0.01, TARGET <0.01, and ARM*TARGET 0.01) were more likely to be found rostrally, while neurons only selective for arm use (represented by green diamonds; three-way ANOVA: ARM <0.01, TARGET 0.01, and ARM*TARGET 0.01) were more likely to be found caudally. Interestingly, PMd neurons selective for both arm use and target location (represented by black dots; three-way ANOVA: ARM <0.01 and TARGET <0.01, or ARM*TARGET <0.01) were found broadly in the rostrocaudal extent of the PMd. In the PMv, neurons were mainly selective only for target location; these neurons were predominantly found in the caudal bank and its lip region of the arcuate sulcus ventral to the spur.
View larger version (29K):
[in this window]
[in a new window]
|
FIG. 12. Anatomical location of neurons selective for arm-use, target-location, or both. A: the location of each neuron was rendered on the cortical surface. For the purpose of display, random noise of < ±0.4 mm was added to the original data, except for the data crossing the PMd–PMv border. This figure shows data at the end of the second delay. The two left hemispheres from the two monkeys are superimposed. Blue squares, neurons selective for target-location only (three-way ANOVA, ARM 0.01, TARGET <0.01, and ARM*TARGET 0.01). Green diamonds, neurons selective for arm-use only (three-way ANOVA, ARM <0.01, TARGET 0.01, and ARM*TARGET 0.01). Black dots, neurons selective for both target-location and arm-use (three-way ANOVA, ARM <0.01 and TARGET <0.01, or ARM*TARGET <0.01). The line indicated by * connects points where Mahalanobis distance to the center of the target-location only neurons and to the center of arm-use only neurons was equal. B and C: rostrocaudal distribution of the three classes of PMd neurons after onset of the second cue. The number of neurons obtained in each 1-mm width of the PMd in the mediolateral plane, plotted in the rostrocaudal direction. Blue line, neurons selective for target-location only. Green line, neurons selective for arm-use only. Black line, neurons selective for both target-location and arm-use. B: data at the end of the second cue period. C: data at the end of the second delay period.
|
|
We performed a more precise analysis of the rostrocaudal distribution of the three classes of PMd neurons by measuring the distance of each neuron in the rostral direction relative to the genu of the arcuate sulcus and counted the number of classified neurons within each 1-mm width of the cortex positioned rostrocaudally. Figure 12, B and C show results measured at two different times after cue onset (Fig. 12B, at the end of the second cue period; Fig. 12C, at the end of the second delay period). At both times, we found neurons selective only for target location (blue trace), more often rostral to the genu of the arcuate sulcus. In contrast, we found neurons selective only for arm use (green trace) more often caudal to the genu of the arcuate sulcus. Using these distribution data of the PMd, we measured the Mahalanobis distances from each data point to the center of the arm-use only neurons and target-location only neurons. In Fig. 12A, we draw a line (indicated by *) where the two Mahalanobis distances were equal. Area to the left of the line indicates where the distance is shorter to the center of the target-location only neurons. In contrast, area to the right of the line indicates where the distance is shorter the center of the arm-use only neurons. Thus this analysis successfully classified the PMd into the rostral and caudal part based on the distributions of target-location and arm-use selectivity of the forthcoming action. However, neurons selective for both factors (black trace in Fig. 12, B and C) were distributed widely in the rostrocaudal extent of the PMd. Taken together, these analyses revealed that the PMd rostral region preferentially represented the target location in conjunction with arm use, and that the PMd caudal region preferentially represented arm use in conjunction with target location.
Relationship of response selectivity during the first and second task phases
It was of interest to know how each PM neuron responding to the first cue behaved after the second cue, and how the selectivity of each neuron responsive to the second cue exhibited responses to the first cue. To answer these questions, we examined the relationship of response selectivity of each neuron at two different epochs: 1) at the end of the first cue period (using the data shown in Fig. 4) and 2) at the end of the second delay period (using the data shown in Figs. 8 and 11). The results of this analysis are summarized in Table 1 (for PMd neurons) and Table 2 (for PMv neurons). For neurons in both the PMd and PMv, a great majority of selective activity during the first task phase (i.e., at the end of the first cue period) was not carried over to the second phase (i.e., at the end of the second delay period). In the PMd (Table 1), the visuospatial selectivity or instruction selectivity in the first task phase (n = 347) was replaced by nonselective activity (n = 202, 58%), the combination selective activity (n = 117, 34%), and second cue selective activity (n = 13, 4%) in the second task phase. In the PMv (Table 2), the visuospatial selectivity or instruction selectivity in the first task phase (n = 60) was replaced by nonselective activity (n = 27, 45%), the combination selective activity (n = 26, 43%), or second cue selective activity (n = 6, 1%) in the second task phase.
Looking backward, we examined how the two-cue combination selective neurons in the second phase behaved in the first phase. Out of the 228 combination selective PMd neurons, 111 (49%) neurons were classified as nonselective, 51 (22%) as instruction selective, and 66 (29%) as visuospatial in the first phase (Table 1). In the PMv, out of the 45 combination selective neurons, 19 (42%) neurons were classified as nonselective, 8 (18%) as instruction selective, and 18 (40%) as visuospatial in the first phase (Table 2). We then studied the relationship between the selectivity in the first task phase and selectivity for the arm use or the target location of a forthcoming action in the second task phase. Figure 13, A and B show the results for the PMd and PMv, respectively. In PMd (Fig. 13A), neurons selective for the visuospace or instruction given by the first cue were evenly distributed among the three groups of the two-cue combination selectivity (i.e., arm-use only, target-location only, and both arm-use ant target-location). In contrast, in PMv (Fig. 13B), neurons selective for the visuospatial information given by the first cue (represented with the light gray) showed a tendency to cluster in the group of neurons selective only for the forthcoming target location; this tendency was significantly different compared with the PMd (2 test, P = 0.0008).
View larger version (21K):
[in this window]
[in a new window]
|
FIG. 13. Relationship of response selectivity in the first and second task phases. The distribution of selectivity in the first task phase is displayed in histograms in which the two-cue combination selectivity in the second task phase is divided in three categories in the abscissa. The two-cue combination selectivity was assessed at the end of the second delay period (see Figs. 11 and 12), whereas the selectivity during the first phase was assessed at the end of the first cue period (see Figs. 4 and 6). A is for PMd neurons, and B is for PMv neurons. Original data are summarized in Tables 1 and 2.
|
|
|
DISCUSSION
|
|---|
In this study, we recorded neuronal activity in two separate areas of the premotor cortex (the PMd and PMv) while the two subject monkeys performed a behavioral task during which they were actively engaged in planning a future reaching movement based on two visual instruction signals.
The following four points summarize our main findings. First, PMv neurons predominantly reflected the locations of visuospatial signals given with the first cue, and this visual response property also accompanied the second cue. Second, about half of PMd neuron first-cue responses reflected motor instructions retrieved from the cue (arm or target to be selected). When the second cue was given, PMd neurons began to reflect a combination of the two instructions (one for the arm and the other for the target)
TOP
ABSTRACT
INTRODUCTION
