Canceling a pending movement is a hallmark of voluntary behavioral control because it allows us to quickly adapt to unattended changes either in the external environment or in our thoughts. The countermanding paradigm allows the study of inhibitory processes of motor acts by requiring the subject to withhold planned movements in response to an infrequent stop-signal. At present the neural processes underlying the inhibitory control of arm movements are mostly unknown. We recorded the activity of single units in the rostral and caudal portion of the dorsal premotor cortex (PMd) of monkeys trained in a countermanding reaching task. We found that among neurons with a movement-preparatory activity, about one-third exhibit a modulation before the behavioral estimate of the time it takes to cancel a planned movement. Hence these neurons exhibit a pattern of activity suggesting that PMd plays a critical role in the brain networks involved in the control of arm movement initiation and suppression.
- countermanding task
- voluntary control
- stop task
- single units
living in a world where events cannot be predicted with certainty, the ability to suppress a pending action after unexpected changes in the environment or in our mind is fundamental. In these instances, volitional inhibition plays a key role in the control of behavior, preventing the prepared movement from occurring. This form of inhibitory control has been studied quantitatively with the countermanding paradigm (Logan 1994). The paradigm probes a subject's ability to withhold a planned movement triggered by a go-signal when an infrequent stop-signal is presented after a variable delay. The behavioral performance of the countermanding task has been modeled by Logan and Cowan (1984), allowing the estimation of an otherwise unobservable variable: the time it takes to cancel a planned movement, the “stop signal reaction time” (SSRT; Band et al. 2003; Boucher et al. 2007; Logan and Cowan 1984).
Over the past 15 years, neural substrates of movement suppression have been explored by correlating the behavioral performance in the countermanding task with the modulation of neural activity in monkeys (Hanes et al. 1998; Ito et al. 2003; Paré and Hanes 2003; Stuphorn et al. 2000), with functional (f)MRI's BOLD activity in volunteers (Aron and Poldrack 2006; Li et al. 2008), or with localized brain lesions in patients (Aron et al. 2003). In particular, single-unit studies revealed that the frontal eye field (FEF; Hanes et al. 1998) and the superior colliculus (SC; Paré and Hanes 2003) contain neurons with activity patterns sufficient to control saccade cancellation. In both studies eye movement suppression is typically associated with a decrease of activity of movement-related neurons before the end of the SSRT and a simultaneous increase of activity in neurons controlling fixation.
Arm movements, differently from saccades, are those that allow physical interactions with the environment, thus leading to achievement of material outcomes, such as the grasping of food, and not only of emotional rewards. So far little is known about the neural processes underlying the inhibitory control of manual movements.
In humans, evidence from lesion (Aron et al. 2003), neuroimaging (Rubia et al. 2003), and transcranial magnetic stimulation (TMS; Chambers et al. 2006) studies in which subjects were required to execute/inhibit a key-press, show the involvement of right inferior frontal cortex (IFC) in the executive control of motor suppression. Aron and Poldrack (2006) and Li et al. (2008) suggested, respectively, that prefrontal cortex countermands planned movements through the right subthalamic nucleus and the head of the caudate nucleus. These brain regions, in turn, are likely exerting their action influencing the primary motor cortex (M1) and the dorsal premotor area (PMd), i.e., those cortical areas critically involved in limb movement preparation and initiation (Cheney and Fetz 1980; Churchland et al. 2006; Churchland and Shenoy 2007; Evarts 1968; Riehle and Requin 1993; Thach 1975; Weinrich et al. 1984). In line with this hypothesis, Coxon et al. (2006), by applying the TMS on M1 during the execution of a countermanding task, demonstrated that this area plays a key role in movement cancellation. Epicortical EEG recording signals further confirmed this evidence (Swann et al. 2009).
In monkeys, recordings of neural activity during Go/No-Go tasks showed the involvement in movement suppression of both M1 (Miller et al. 1992; Port et al. 2001) and PMd (Kalaska and Crammond 1995). However, in the Go/No-Go paradigms it is a potential movement and not an ongoing response that has to be halted. To date the only study in which activity of single neurons was recorded during a manual version of the countermanding task is that of Scangos and Stuphorn (2010). They recorded from the supplementary motor area (SMA) and pre-SMA of monkeys, and they found that the activity of these regions does not control arm movement initiation but might contribute to movement suppression. The latter conclusion has been confirmed by Chen et al. (2010), who showed that local field potential (LFP) power spectra obtained from data recorded over SMA display changes in the low-frequency range (10–50 Hz) early enough to suggest that this region is causally involved in movement inhibition. However, it must be stressed that the percentage of neurons causally involved in movement suppression found by Scangos and Stuphorn (2010) was rather small, i.e., only 8 neurons out of 335 (2.4%). Even though the presence of a recording bias or other factors that might have influenced the total number of identified neurons could not be excluded, it is also plausible to hypothesize that SMA and pre-SMA are not the main actors in canceling a movement after the appearance of an imperative stop-signal. This interpretation is not in contrast with the finding of Chen et al. (2010) because changes in LFPs could be caused not by the local activity but by inputs coming from other brain regions (Logothetis 2003; Mattia et al. 2010).
In the present study, we have reinvestigated the neural correlates of volitional cancellation of a pending arm movement by recording the responses of area PMd neurons in two monkeys performing a countermanding reaching task. For the first time, we report the existence in PMd of reaching-related neurons showing a modulation of activity related to the suppression of programmed arm movements.
Two adult male rhesus macaques (Macaca mulatta; monkey S and monkey L) weighing 7–8 kg were used. Before the training started, a head holding device and a scleral eye coil (Robinson 1963) were implanted under aseptic surgical conditions. Antibiotics and analgesics were administered postoperatively. In each monkey, at the end of the training period, again under general anesthesia, a recording cylinder (18 mm in diameter) was implanted stereotaxically in the left frontal lobe in order to allow recordings over the arm representation of the PMd (Paxinos et al. 2000). The location of the neural recordings was confirmed by structural MRI on monkey S and visual inspections of the anatomic landmarks, such as the central (CS) and the arcuate sulcus (AS), on monkey L after the dura was opened. Both animals were killed at the end of the experimental procedures. In the present report we deal only with recordings obtained from PMd (Fig. 1A).
Animal care, housing, and surgical procedures were in conformity with European (Directive 86/609/ECC) and Italian (D.L. 116/92) laws on the use of nonhuman primates in scientific research and were approved (no. 58/2005-B) by the Italian Ministry of Health.
Apparatus and electrophysiological recordings.
Animals were placed in a darkened, sound-attenuated chamber and seated in a primate chair, with their head fixed in front of a 21-inch PC monitor (CRT noninterlaced, refresh rate 85 Hz, 800 × 600 resolution, 32-bit color depth; monitor-eye distance 21 cm) equipped with a touch screen (MicroTouch, sampling rate 200 Hz) for touch position monitoring. Touch screen sensitivity was set to the maximum value in order to detect minimal changes in the touch position. A noncommercial software package, CORTEX (www.cortex.salk.edu), was used to control stimuli presentation and behavioral responses and to collect neural (1,000 Hz) and eye movement (200 Hz) data. During the task, eye movements were monitored by a magnetic search coil technique (Fuchs and Robinson 1966; Remmel Labs, Ashland, MA). Saccades were detected off-line with velocity threshold criteria (30°/s). Eye reaction times were measured as the interval from target appearance to the beginning of the saccade.
Neural activity of single units was recorded extracellularly with a seven-channel multielectrode system (Thomas Recording, Giessen, Germany). Electrodes were quartz-insulated platinum-tungsten fibers (80-μm diameter, 0.8- to 2.5-MΩ impedance) and were inserted transdurally, one at a time, with microdrives (Thomas Recording). Electrical signals were amplified and filtered, and single units were isolated online exploiting a dual time amplitude window discriminator (BAK Electronics, Mount Airy, MD).
Stimuli, task, and neuron selection.
Visual stimuli consisted of red circles (2.43 cd/m2) with a diameter of 7.6° (2.8 cm) on a dark background of uniform luminance (<0.01 cd/m2). The presentations of the stimuli were synchronized with the monitor refresh rate (85 Hz). Monkeys were required to use the arm (right arm) contralateral to the recording hemisphere. The other arm was physically constrained.
After one or two cells were isolated at each electrode, we qualitatively determined whether neurons exhibited a preparatory activity correlated to reaching movements with an instructed delay task (Johnson et al. 1996). The delay task allowed us to qualitatively select neurons showing preparatory activity. Whenever we had at least four isolated neurons with preparatory activity across all electrodes, a reaching version of the countermanding task was administered (see Fig. 1B; Mirabella et al. 2006, 2008, 2009). This consisted of one block of 480 trials, where no-stop trials (67%) were randomly intermixed with stop trials (33%). All trials began with the appearance of a stimulus at the center of the display (Fig. 1B). Monkeys were required to touch it with their finger(s) within 2 s, and hold it for 500–800 ms. In the no-stop trials the central stimulus went off and, simultaneously, a target appeared (go-signal) randomly at one of two possible opposite positions, 21.8° (8 cm) from the central stimulus, virtually arranged in a circle at 45° of interval between the potential positions. For each trial, targets were presented at the preferred location (corresponding to the positions that better modulated most of the isolated neurons) or at the opposite location. To get the juice reward, animals were required to reach and hold the target for 300 ms. Stop trials differed from the no-stop trials because at a random delay (stop signal delay, SSD) during the reaction time (RT), the central stimulus reappeared. In these instances, monkeys had to inhibit the pending movements, holding the central position for an additional interval of 650–850 ms (450–550 ms for monkey L) after stop appearance, until reward delivery. To discourage monkeys from adopting the strategy of slowing down RT to maximize the number of correct responses to stop-signals, we set a maximum time for response, called upper RT (600 ms for monkey L; 750 ms for monkey S). An auditory feedback was given for correct responses. A time-out of 800 ms was given after each error.
The probability of inhibiting a movement critically depends on the length of the SSD. Stopping becomes increasingly more difficult as the SSD is lengthened. Logan and Cowan (1984) developed the horse race model to explain these results. The model assumes that the behavioral outcome of the task is the result of a race (Fig. 2A) between two stochastically independent processes: a go process triggered by the go stimulus and a stop process triggered by the stop-signal. If the stop process wins, participants will inhibit their response (success). On the other hand, when the go process wins, participants will respond (failure). Recently, the assumption of independence of the two processes has been challenged. Boucher et al. (2007) proposed an interactive race model, in which the go and stop processes are independent for much of their latencies but interact near the end of the race, when the stop process tries to interrupt the go process. However, even in the new formulation the model reliably describes the performance in the countermanding task and allows the estimation of the SSRT.
In the two animals, we used two different procedures for setting the SSDs. In most sessions (all 33 sessions for monkey L and 9 of 24 sessions for monkey S) we used a fixed-SSD procedure (Band et al. 2003). On the basis of the average RT measured at the beginning of each session, we computed four progressively longer SSDs so that monkeys were able to successfully inhibit a movement in ∼85%, ∼65%, ∼35%, and ∼15% of the stop trials. The SSDs were set independently for each of the two movement directions, to compensate for possible differences of RT. Whenever, after some trials, we realized that the performance did not satisfy the above-defined criteria, the SSDs were adjusted and the session was restarted until a good control of the behavior was obtained for at least one of the two directions of movement. In monkey L the SSDs ranged from 129.4 ms (11 units of refresh rate) to 341 ms (29 units of refresh rate), with a mean value of 219.7 ± 4.16 ms (variance will always be reported with the standard error). In monkey S the SSDs ranged from 117.6 ms (10 units of refresh rate) to 471 ms (40 units of refresh rate), with a mean of 341.1 ± 14.5 ms.
In 15 experimental sessions of monkey S, the length of the SSDs was dynamically changed with a staircase procedure (Band et al. 2003; Mirabella et al. 2008, 2009; Osman et al. 1986, 1990). The SSD duration varied from one stop trial to the next according to the behavioral performance: if the monkey succeeded in withholding the response, the SSD increased by 5 refresh rates (or 58.8 ms); if it failed, the SSD decreased by the same amount of time. We used two independent staircases, one for each movement direction, to compensate for eventual differences in RT. Both staircases started from a SSD of 246.9 ms (21 refresh rates), which preliminary data obtained in monkey S suggested were appropriate for quickly obtaining the desired performance (50% success). This procedure provides different SSDs for each sessions; however, to maintain a statistical power similar to that we had for the fixed-SSD procedure, we further analyzed the neural responses only when a SSD was presented at least 20 times and when at least 5 of these trials were correctly suppressed.
Since in each session the target could appear at two possible locations, from each countermanding block we obtained either two inhibition functions (that is, the relationship between the probability of stop-failure trial occurrence as a function of the SSDs), one for each target direction, or two possible outcomes of the staircase procedures. To derive reliable parameter estimates for each inhibition function, the data were fit with a Weibull cumulative distribution [W(t), where t is time after target presentation; Hanes et al. 1998]. Overall, the Weibull function fits had a mean r2 of 0.8 (±0.02) and the χ2-test was always nonsignificant (P > 0.05). For each inhibition function, we estimated the SSRT with the two methods described in detail by Mirabella et al. (2006), based on two different assumptions. The first method assumes that SSRT is a random variable. Under this hypothesis the SSRT is estimated by the difference between the mean RT of no-stop trials and the mean of the inhibition function (method of the mean; Logan and Cowan 1984; Hanes and Schall 1995). The mean of the inhibition function corresponds to the SSD at which P(failure) = 0.5. We evaluated numerically the integral, using the fitted W(t) and a trapezoidal rule with bins of 1 ms (Hanes et al. 1998): (1) The second method assumes that the SSRT is a constant and that go process durations are roughly the same for no-stop and stop trials (integration method; Band et al. 2003; Logan and Cowan 1984). With this method, the SSRT is obtained for each given SSD by subtracting the finishing time of the stop process from the starting time (the SSD value). The finishing time of the stop process is calculated by integrating the no-stop trial RT distribution from the onset of the go-signal until the integral equals the corresponding observed proportion of stop-failure trials (Logan 1994). Then the SSRT is calculated as the mean value of the SSRT computed at each SSD.
Whenever the staircase procedure was employed, the SSRT was computed with two procedures (described in detail in Mirabella et al. 2009), both based on the use of the integration method. The two procedures differed in the method used to obtain the starting time. In the first procedure, for each session, using the midrun estimate method (Levitt 1971; Wetherill and Levitt 1965; Wetherill 1966), we worked out the starting time as the delay that better corresponds to the time needed for the subject to withhold a response ∼50% of the time (“representative” SSD). In the second procedure, for each session, we took as starting times the length of those SSDs that were presented at least 20 times. For each SSD selected a value of the SSRT was computed; then the behavioral estimate of cancellation time in a given session was obtained by averaging all the SSRTs computed at each SSD.
In summary, whatever the method used for setting the SSDs (fixed or staircase), we obtained, for each recording session, two estimates of the SSRT for each direction of movement/target appearance.
Neuronal data analysis.
Whenever not otherwise specified, for each neuron with a significant preparatory activity (see results) we analyzed neural data for the movement direction for which we had the best behavioral performance. That is, as far as the fixed-SSD procedure is concerned, we selected the movement direction for which inhibition function was as close as possible to the one desired, i.e., the one for which the monkeys failed to successfully cancel a movement in ∼15% (shortest SSD), ∼35% (2nd SSD), ∼65% (3rd SSD), or ∼85% (longest SSD) of the stop trials. For the staircase procedure we considered the movement direction for which the P(failure) was closest to 0.5. In both cases, only data from one direction of movement have been used to assess for countermanding related modulations.
To visualize the neural data, rasters of neuronal discharge and spike density functions were aligned on the time of the go-signal. Spike density functions were obtained by convolving spike trains with a Gaussian kernel function (kernel width 13 ms).
To detect countermanding related activity in our sample of neurons, following the line of reasoning of Hanes et al. (1998) and of Paré and Hanes (2003), we contrasted the activity during stop-success trials with the activity recorded during those no-stop trials in which the reaching movement initiation would have been canceled if the stop-signal had been presented at the same SSD. These are the trials in which, given the length of the SSRT, the go-process was slower than the stop-process if the stop-signal had occurred. This subset of no-stop trials, which we will refer to as latency-matched no-stop trials, is given by those reaching movements with RTs greater than the sum of the SSD and the SSRT calculated from the same data (e.g., for the longest SSD, dark region of the no-stop trial RT distribution in Fig. 2B). To quantify the time course of the neuronal activation during stop-success trials and latency-matched no-stop trials, we calculated a differential spike density function (Hanes et al. 1998) by subtracting the absolute values of the average spike density functions (aligned on the time of the go-signal) associated with each type of trials. We defined the time at which significant differential activation began (the neural cancellation time) as the instant when the differential spike density function exceeded at least by 2 SD the mean value of the differential activity recorded during the 300-ms period preceding the go-signal provided that the difference remained above this threshold for at least 60 ms. The reference period was subdivided into 12 bins (each lasting 25 ms), neural activity was calculated for each bin in absolute value, and finally the mean and SD values across bins were worked out.
Behavioral estimate of reaching arm movement cancellation.
To control in our data the validity of the stochastic independence of go and stop processes, we checked how well the race model predicted the RTs of stop-failure trials, that is, the RTs of those reaching movements that could not be canceled even though a stop-signal was presented (Logan and Cowan 1984; Mirabella et al. 2006, 2008). In stop-failure trials, reaching movements were produced because the go process won the race against the stop process. Therefore, considering the distribution of the RTs of the no-stop trials, the responses that would not be stopped despite the presentation of the stop-signal should be those corresponding to reaching movements with RTs shorter than the SSD plus the SSRT (e.g., for the longest SSD, light region of the no-stop trial RT distribution in Fig. 2B). Three predictions should be satisfied (Logan and Cowan 1984; Logan 1994). First, the mean RT in stop-failure trials should never be longer than the mean RT in no-stop trials. Second, the mean RT in stop-failure trials should linearly increase with increasing SSDs. Third, the mean RT in the stop-failure trials at each SSD should be equal to those predicted from the race model. Figure 3 shows that in an example session these predictions were satisfied. Figure 3A shows that the cumulative RT distribution for no-stop trials (mean 285.8 ± 3.1 ms) is shifted to the right with respect to the cumulative RTs distribution of stop-failure trials (mean 268.4 ± 4.2 ms), namely, the latter are faster than the former (Kolmogorov-Smirnov test; P < 0.0005). From the same data set, Figure 3B shows that the second and third predictions of the race model are also satisfied. In fact, RTs in the stop-failure trials increase as a function of the length of the SSDs and they are not significantly different from those predicted by the race model (paired t-test; all P > 0.05), with the known exception of the shortest SSD (Logan 1994). All these predictions were largely satisfied across all sessions in both monkeys. The RTs in stop-failure trials were significantly shorter than the RTs in no-stop trials (see Table 1; Kolmogorov-Smirnov test, all P < 0.05) in 45 of 54 cases (or 83.3%). The other two assumptions were tested in those sessions in which the fixed-SSD procedure was employed. Linear regression analysis showed that in all occurrences but 2 (40/42 or 95.2%) the mean RTs in stop-failure trials increase with increasing SSD (mean slope 0.56 ± 0.08). The violations observed for the shortest SSD are consistent with previous observations, and they are attributed either to the very few stop-failure trials occurring at the shortest SSD (Logan and Cowan 1984; Logan 1994; Mirabella et al. 2006) or to self-generated movements produced after the initial movement was inhibited (Boucher et al. 2007). Finally, in 125 of 154 cases (or 81.1%) the observed mean RTs in the stop-failure trials at each SSD were equal to those predicted (t-test, all P > 0.05).
Figure 4A plots the inhibition function, and the corresponding W(t), for one representative session of monkey L. Figure 4B shows the average inhibition function across all sessions separately for the two monkeys. To obtain the latter, data from single sessions were combined by averaging for each single SSD the probability of generating a movement [P(failure)] even though a stop-signal was presented. These results demonstrate the reliability of the behavioral control. In the staircase sessions the goodness of the behavioral control was further demonstrated by the fact that the average P(failure) was close to 0.5 (0.48 ± 0.3, Table 1; see also Band et al. 2003).
Table 1 summarizes all relevant parameters describing the behavioral performance of each monkey for all sessions used for the analysis of the neural activity, separately for the fixed-SSD procedure and the staircase procedure. With the fixed-SSD procedure, the SSRT estimated with the integration method did not significantly differ from that obtained assuming that the SSRT is a random variable (paired t-test; monkey L: df = 32, t = 0.97, P = 0.34; monkey S: df = 8, t = −1.1, P = 0.29). Therefore, we averaged them (monkey L: average SSRT = 137.7 ms; monkey S: average SSRT = 160.6 ms). In monkey S, two other estimates of the SSRT were obtained from the analysis of the staircase sessions. Again, the two estimates of the SSRT were not significantly different (paired t-test, df = 14, t = 0.001, P = 0.99); thus we averaged them (average SSRT = 147.5 ms). These values of SSRT for reaching arm movements are very similar to those recently reported (∼140 ms) by Scangos and Stuphorn (2010) for arm movement in monkeys.
Classification of neural activity.
A total of 163 individual neurons were recorded from the left PMd areas of the two monkeys (93 and 70 neurons from monkey L and S, respectively).
As a first step, we assessed the number of cells exhibiting a reaching-related activity in the selected direction (on the basis of the good behavioral control; see methods). To this end, for each recorded cell we compared with an analysis of variance (1-way ANOVA) the firing rates during no-stop trials in three epochs: 1) the period of 400 ms during the holding time preceding the appearance of the target; 2) the RT epoch; and 3) the movement time (MT) epoch, defined as the time window between movement onset and the time when the target was touched. A cell was classified as reaching related if it showed a main effect at the ANOVA and if post hoc analysis (Tukey-Kramer test; P < 0.05) revealed that the firing rate during the RT and/or the MT epoch differed from the discharge in the 400-ms epoch before target onset. The results of this analysis showed that 154 neurons (94.5%) had a main effect (P < 0.05), namely, they were modulated during the task. Of major relevance, post hoc tests revealed that 22 neurons (14.3%) significantly changed their discharge exclusively during the RT epoch; 19 neurons (12.3%) were significantly modulated exclusively during the MT epoch; and finally 113 neurons (73.4%) showed a significant change of the firing rate during both the RT and MT epochs. Overall 135 neurons (87.6%) were modulated during motor preparation, i.e., during the RT epoch. These neurons were those selected for further analyses in this report because they showed a modulation before the start of the movement and could potentially be involved in its generation. Thus the selected neurons are those whose discharge was modulated during the preparation of the arm movement, and therefore they are the best candidates to show a modulation of their firing rate according to the fact of whether a movement should be executed or not.
Cancellation signals for reaching movements in PMd.
To determine whether and how PMd neurons, with significant activity during the RT, were involved in inhibiting a planned arm movement, we compared the activity of the 135 neurons in those trials in which reaching movements were executed (no-stop trials) versus trials in which they were successfully inhibited (stop-success trials).
To influence the behavior, a reaching-related cell must change its discharge when a reaching movement is executed with respect to when it is inhibited. Moreover, to be causally involved in movement suppression, the divergence in neural activity should take place before the behavioral estimate of the end of the cancellation process, i.e., the SSRT (Hanes et al. 1998; Parè and Hanes 2003). To this purpose, we analyzed data by aligning neural activity to target presentation (go-signal) and, since the stop-signal appears at different SSDs, we analyzed the modulation of neural activity separately for each SSD.
Figure 5 shows the activity, aligned to target presentation, for two example neurons recorded during the same session (monkey S; arm movements directed toward a right target, SSD of ∼300 ms). For both neurons, the activity during stop-success trials is compared with the activity recorded during latency-matched no-stop trials (see methods). Figure 5 also reports the traces of the horizontal component of eye movements during stop-success and latency-matched no-stop trials. In no-stop trials, the monkey always moves the eyes toward the target after its appearance; in addition, during stop trials it moves the gaze back to the center of the screen as soon as the stop-signal appears. For both neurons, the activity of no-stop trials starts to increase ∼150–200 ms after the go-signal and peaks at ∼100 ms before the average time of movement onset (Fig. 5; M_on). During successful stop trials the activity of both neurons initially resembles that of no-stop trials. However, after the stop-signal appearance, for one neuron, hereafter called the “type A” neuron, the activity significantly decreases after stop-signal presentation with respect to that recorded during latency-matched no-stop trials (Fig. 5A) while for the other, hereafter called the “type B” neuron, the activity significantly increases (Fig. 5B). The differential spike density functions (Fig. 5, bottom) indicate that a significant divergence (see methods) occurs before the end of the SSRT for both the type A (neural cancellation time: −74 ms) and type B (neural cancellation time: −63 ms) neurons. This divergence is well before (∼200 ms) the average time of movement onset (M_on, Fig. 5).
The estimate of the neural cancellation time of the 388 computable SSDs (namely, those SSDs presented at least 20 times in the recording block of trials and with at least 5 trials correctly executed; see Table 2) is presented in Fig. 6A for the 263 with a significant differential activation (see methods). In 153 of 263 (58.2%) SSDs the cancellation time preceded the end of SSRT by 48.6 (±2.4 SE) ms on average. The number is still consistent (88/263; 33.5%) when considering only neural cancellation times shorter than the value obtained by subtracting from the SSRT the estimated average delay (50 ms) needed for neural activity in PMd to influence arm muscle activity (Lemon et al. 1986; McKiernan et al. 1998; Morrow and Miller 2002; Tokuno and Nambu 2000).
To give an account of the countermanding modulation in terms of number of cells we used the following procedure. We assessed the number of neurons in which at least 60% of their SSDs1 showed a significant countermanding-related modulation (i.e., a cancellation time < SSRT). We found that 44 of 135 (32.6%) neurons showed such a modulation. When considering the 50-ms efferent delay, the number of neurons with a countermanding behavior becomes 34 of 135 (25%), still a consistent population.
As stated above, we found two different types of countermanding modulation: type A and type B. To evaluate the frequency of these two neural behaviors, for each SSD with a significant countermanding-related modulation we computed a normalized index of the discharge rate (IDR; Fig. 6B): (2) where ST and NoST represent the activity during stop-success trials and latency-matched no-stop trials, respectively, in a 50-ms window centered on the end of the SSRT. The index can take a negative value up to −1, when the cell discharged only during stop-success trials, corresponding to a neural modulation similar to that reported in Fig. 5B (type B), or a positive value up to +1, corresponding to the absence of activity during stop-success trials at the end of SSRT, corresponding to a neural modulation similar to that reported in Fig. 5A (type A). As shown in Fig. 6B, the number of SSDs with positive IDRs was higher than the number of SSDs with negative IDRs, 94 vs. 59, respectively (χ2-test, P < 0.005). Of the 44 neurons exhibiting countermanding-related modulation, 26 had positive IDRs and 18 had negative IDRs. Interestingly, a neuron exhibiting a countermanding modulation always showed the same type of response for each SSD analyzed. In addition, a one-way ANOVA (4 levels: cancellation time at SSD1, SSD2, SSD3, SSD4) revealed that cancellation time did not change as a function of SSDs length (F[1,3] = 1.6, P = 0.23). We controlled for differences in the neural cancellation time of the two classes of neurons. The activity started to diverge on average 46 ± 3.3 ms and 52.6 ± 3.1 ms before the end of the SSRT in type A and type B neurons, respectively. Statistical analysis showed that in the two classes of neurons the activity in stop-success trials diverges from that of latency-matched no-stop trials at the same time (t-test: df = 151, t = 1.4, P = 0.17).
Since we had a restricted subset of neurons with a movement-preparatory activity for which the behavioral control was good in both directions of movement (76/135 or 56%), we analyzed the neural activity of those cells to shed light on whether 1) neurons countermand a movement in both directions or just in one direction and 2) countermanding behavior differs between neurons that have a preferred direction versus those that do not have it.
First of all, for each of these neurons, we considered those SSDs that were computable (see methods) in both directions. Overall these SSDs were 181 of 388; of those 99 showed a countermanding-related modulation. Forty-five of 99 SSDs showed a countermanding behavior in both directions of movements, while 54 of 99 did not. The frequency of the two types of SSD was not different (χ2 = 0.37). Importantly, a neuron showing a countermanding modulation in one or in both movement directions did so at all its SSDs. Therefore we found two groups of neurons: one that countermanded in both movement directions and the other that countermanded just in one direction.
In principle it could be postulated that neurons having a directional tuning might have a different modulation even for movement suppression. Visual inspection revealed that often the neuronal discharge during the no-stop trials was higher in one direction than in the opposite direction. We quantitatively assessed the number of cells exhibiting a preferred direction, comparing with a t-test the firing rates of no-stop trials during the RT epoch. We found that the majority of cells were directionally tuned (59/76 or 77.6%). Among cells with a preferred direction, 35 of 59 SSDs showed a countermanding behavior in both direction of movements and 24/59 did not; their frequency was not significantly different (χ2 = 0.26). The same was true for non-directionally tuned neurons (10/17 SSDs had a countermanding modulation in both movement directions and 9 just in one direction; χ2 = 0.82). Thus directional tuning does not seem to affect the countermanding behavior of PMd neurons. However, we are aware that our task is not ideal for tackling the relationship between directional tuning and countermanding modulation, given that we did not test the neurons in more than two directions. Further studies need to be performed to clarify this issue.
Finally, since we recorded from both the rostral and caudal portion of PMd, and it is known that these two regions have different proportions of reaching-related neurons and signal/motor-related activities (Hoshi and Tanji 2006; Johnson et al. 1996), we explored 1) whether type A and type B countermanding neurons have a different distribution along the rostrocaudal dimension and 2) whether neurons with a “motor” prevalent activity display a different modulation during the countermanding task with respect to those with a “visual” prevalent activity. We found that there were no evident clusters or gradients of cell properties in the tangential cortical domain explored (Fig. 1A). As far as the second argument is concerned, for each neuron exhibiting a significant countermanding modulation (44/135), following the logic of Ray et al. (2009), we computed a visual-movement index (VMI) as follows: VMI = (MA − VA)/(MA + VA), where VA is visual activity and MA is movement activity. Since the two animals had a different average RT (see Table 1), the time windows for the computation of the mean firing rates were different for the two monkeys. The time window of VA was 50–200 ms and 50–170 after go-signal onset for monkey S and monkey L, respectively. The time window of MA was −100 to +50 ms and −70 to +50 ms after movement onset for monkey S and monkey L, respectively. Neurons with negative values of VMI represent cells for which visually evoked response is prevalent (VP = visual prevalent cells), while neurons with positive values of VMI represent cells for with prevalent arm movement-related activity (MP = movement prevalent cells). First of all, we looked at whether VP and MP had a different distribution among the population of countermanding cells. Twenty-five of 44 were MP neurons (∼57%), while 19 of 44 were VP neurons. The difference is not significative (χ2-test, P = 0.36). On average the VMI was 0.45 ± 0.06 and −0.37 ± 0.06 for MP and VP neurons, respectively [t-test, t(42) = 9.54, P < 0.001]. Second, we compared the cancellation time of all MP and VP neurons measured at each computable SSD. On average, the cancellation time preceded the SSRT by 46.1 ± 3.4 ms and by 51.5 ± 3.8 ms for the MP and for the VP neurons, respectively. There was not a significant difference [either by a parametric test, the t-test, t(114) = 1.02, P = 0.31, or a nonparametric test, the Kolmogorov-Smirnov test, P = 0.11]. In conclusion, both neuronal types play a similar role as far as the production and the suppression of an arm reaching movement are concerned, in contrast the findings of Ray et al. (2009) in FEF, further suggesting that oculomotor centers have a different functional organization with respect to brain areas controlling arm movements.
Interpretational issues about the countermanding modulation of PMd neurons.
We have interpreted the modulations of PMd neurons as they were related to the production or the cancellation of pending reaching arm movements. However, at least in principle, it is possible that PMd neuronal activity could be related to other processes. In fact, PMd activity may have been linked to eye movements/gaze position (Boussaoud et al. 1993, 1998; Fujii et al. 2000; Pesaran et al. 2006, 2010) or to the visual presentation of the stop-signal.
First of all, we assessed the relationship between saccadic and arm movements during the task. As expected (Carey 2000), the eyes on average reacted to the target presentation faster than the arm (Kolmogorov-Smirnov test, all P < 0.001). However, the saccadic RT of monkey L was longer than that of monkey S (325.7 ± 5.8 and 194.9 ± 7 ms, respectively; Kolmogorov-Smirnov test, P < 0.001). As a consequence, the difference between the RTs of arm and eye was bigger for monkey S (mean difference 303.2 ± 7.3 ms) than for monkey L (mean difference 35.8 ± 3.3 ms; t-test, df = 43, t = −34.3, P < 0.0001). The two animals employed different ocular strategies during the countermanding task. In no-stop trials, monkey S quickly moved the eyes toward the peripheral target, while the arm movements were procrastinated (Fig. 5); conversely, monkey L made saccades to the target just before executing the arm movements (Fig. 7). These differences probably account for the very different ocular behavior displayed during stop trials by the two animals. During stop-success trials, monkey L did not move the gaze from the central position (Fig. 7, bottom). Conversely, monkey S first made a saccade toward the peripheral target, as for no-stop trials, and after the presentation of the stop-signal it moved the eyes back on it (Fig. 5). Nevertheless, despite the very different patterns of eye movements displayed by the two monkeys during SSRT, the neural modulation during the countermanding task was very similar, as shown for the example neurons of Figs. 5 and 7. In both monkeys, type A/B neurons decreased/increased their discharge during stop-success trials after the stop-signal presentation and before the end of the SSRT.
The oculomotor strategy of monkey S allows us to further tackle the issue of the possible relation between the neural modulation and eye movements. The peak of activity shown during successful stop trials of the neuron shown in Fig. 5B might seem to be related to the eye movement following stop-signal appearance (or to a visual-related activation). We excluded the possibility for this neuron, and for all of the other neurons showing both movement-preparatory activity and countermanding modulation in monkey S, by comparing the neural activity elicited by eye movements with similar vectors occurring during different phases of the task. Figure 8 shows, for the same neuron shown in Fig. 5B, the activity during stop-success trials compared with the activity obtained in the stop-failure trials for the two positions in which the target could appear. During stop-success trials, the monkey exhibits a pattern of eye movements qualitatively similar for both target positions. When the target appears to the left (Fig. 8A), the monkey performs first a leftward eye movement and, after appearance of the stop signal, makes a rightward saccade. Exactly the opposite eye movement sequence takes place when the target is presented to the right. Thus the peak of neural activity observed during the SSRT cannot be related either 1) to the immediately following eye movement, because a similar saccade does not elicit a similar neural modulation (e.g., compare the activity during stop-success trials after rightward eye movements in Fig. 8, A and B), or 2) to a visual response, since it does not appear in stop-failure trials even though the stop-signal was presented exactly at the same time as in success-stop trials. All type B neurons, with a pattern of activity similar to that shown by the neuron of Fig. 5B, were analyzed in this way, and in all cases we have been able to exclude that their neural modulation could be linked to eye movement generation and/or sensory stimulation.
On the grounds of our experimental evidence we strongly believe that our results cannot be explained on the basis of gaze-related and/or saccade-related modulations. Instead, our findings indicate the presence of a subpopulation of PMd reaching-related neurons that displays a modulation of activity that is potentially able to control the production and the suppression of arm movements.
Neural signals for reaching movement inhibition in PMd.
The main goal of the present study was to explore the contribution of single neurons of PMd cortex in inhibiting a planned reaching arm movement. Thus, among recorded neurons, we selected those modulated during the preparation of the movement, and we found that a substantial percentage of these neurons exhibit, after stop-signal presentation, a pattern of activity able to influence the production or the cancellation of reaching arm movements.
Historically, single-unit studies have shown that PMd is involved in several aspects of arm movement control. PMd neurons, also thanks to the direct access to the spinal cord (Dum and Strick 1996), have a role in the preparation of movements (Churchland et al. 2006; Crammond and Kalaska 2000; Johnson et al. 1996; Weinrich and Wise 1982), in learning associations between sensory stimuli and motor responses (Di Pellegrino and Wise 1993; Wise et al. 1983), in online correction of arm movements (Georgopoulos et al. 1983), and in the representation of potential actions (Cisek and Kalaska 2005). To our knowledge, there is just one study showing that neural activity of PMd neurons changes when a movement is suppressed with respect to when it is executed in a Go/No-Go paradigm (Kalaska and Crammond 1995). However, in the Go/No-Go task the signal for inhibiting the movement is presented before the go-signal, while in the countermanding task the stop follows the go-signal. Hence in the Go/No-Go task it is a potential movement and not an ongoing response that has to be canceled.
In our sample more than one-third of neurons involved in movement preparation exhibited a countermanding modulation. In these cells the discharge changed when a reaching movement was executed with respect to when it was inhibited, and this change preceded the end of the behavioral estimate of movement cancellation (the SSRT). We identified two types of cells showing this neuronal pattern. In the most common class of neurons, type A, the activity during stop-success trials decreases before the end of the SSRT with respect to that recorded during no-stop trials. In type B neurons, movement suppression is associated with a temporary increase of activity with respect to the activity recorded during no-stop trials.
The behavior of the two classes of neurons we observed resembles, at a first glance, that of movement and fixation neurons in the FEF (Hanes et al. 1998) and in the SC (Paré and Hanes 2003). However, while the parallel between type A and movement neurons might be supported, that for type B and fixation neurons cannot. In fact, fixation neurons are tonically active during fixation periods while they drastically reduce their activity just before the execution of a saccade (Munoz and Wurtz 1993). During stop-success trials, fixation neurons in FEF and SC increase their discharge after stop signal presentation (Hanes et al. 1998; Paré and Hanes 2003). This increment counteracts the decrease of the discharge occurring after the presentation of the go-signal, allowing fixation cells to reestablish the level of activity typical of fixation periods, in agreement with a system based on a finely controlled gating mechanism (Munoz and Wurtz 1993). In contrast, type B neurons do not display a tonic discharge when the arm is maintained still, and after stop-signal presentation they increase their activity faster than in those trials where a movement has to be produced. In addition, fixation neurons have an important role during saccade generation since they control the discharge of omnipause neurons (OPNs) in the nucleus raphe interpositus. In fact, to generate a saccade the tonic inhibition of OPNs on the “burst neurons” in the brain stem needs to be removed (Bergeron and Guitton 2002; Munoz and Wurtz 1993).
However, it is important to remark, as further suggested by our findings, that the functional organization for saccade control in the oculomotor centers does not have a correspondent in the neural structures controlling arm movements. The overall organization of arm movement control is much more complicated. In principle, it would be possible to speculate that inhibitory interneurons of PMd-M1 could prevent movement execution during the planning of an action by suppressing the activity of corticospinal movement neurons. In this frame the action would start when cortical inhibition would be removed. However, recent evidence (Kaufman et al. 2010; Merchant et al. 2008) shows that inhibitory interneurons in PMd and M1 increase, not decrease, their discharge at the time of movement generation. Therefore, these neurons do not seem to participate in movement control as fixation neurons. In addition, interneurons in PMd are more active both during the preparatory phases of a reaching movement and around movement onset than putative pyramidal neurons (Kaufman et al. 2010). PMd is able not only to influence the neural activity of interneurons in the spinal cord (Dum and Strick 2002; Prut and Fetz 1999) but also to excite or inhibit M1 (Ghosh and Porter 1988; Tokuno and Nambu 2000). Overall, these results strongly suggest that the control of arm movements is organized very differently with respect to that of eye movements. Possibly type A neurons could correspond to PMd projection neurons, directed, e.g., to M1 or to spinal cord interneurons (Dum and Strick 2002), while type B neurons could correspond to PMd inhibitory interneurons, actively controlling the discharge of type A neurons. Unfortunately, we have no further argument to support this idea, since we could not classify neurons on the basis of the recorded waveforms (Kaufman et al. 2010; Mitchell et al. 2007). This topic will be the object of future research together with the description of the neural modulation in M1 during a countermanding task.
Another possibility to explain the different behavior of type A and type B neurons in the countermanding task is that the decrease of discharge of the type A neurons would correspond to the suppression of agonist muscles of the arm for a given movement, while the increase of type B neurons would correspond to the activation of the antagonist muscles. In this way the activity eventually elicited in the agonist muscles, after the presentation of the go signal, would be suppressed and contrasted at the same time. Results of Kudo and Ohtsuki (1998) provide support for this hypothesis, especially for long SSDs when the agonist muscles are more likely to be activated. In contrast, a recent report did not find evidence for cocontraction of antagonist muscles during movement suppression in a countermanding task (Scangos and Stuphorn 2010). Scangos and Stuphorn (2010) suggest that action inhibition is accomplished by relaxation of the agonist muscle. The discrepancy could be explained by the different arm movements required in the two experiments: in one case subjects were asked to control the elbow movements (Kudo and Ohtsuki 1998), in the other the monkeys have to move a handlebar (Scangos and Stuphorn 2010). However, Toma et al. (1999), using functional magnetic resonance, showed that not only muscle contraction but also muscle relaxation produces a transient increase of activity in the M1 contralateral to the limb used and bilaterally both in the supplementary motor areas and PMd. Thus the peak of activity observed in type B neurons could be associated with the active relaxation of agonist muscles. In line with this hypothesis, it has been shown that suppression of the muscle contraction can occur as a consequence of the discharge of M1 projection neurons likely targeting spinal inhibitory interneurons (Cheney et al. 1985; Lemon et al. 1987). Unfortunately, we cannot argue further on this issue because for technical reasons we have been unable to use data obtained during electromyography of selected muscles. Further studies are needed to clarify this point. However, it is important to underline that the lack of EMG recordings should not impact our findings too much. In fact, if we assume that our monkeys used the arm muscles as the monkeys recorded by Scangos and Stuphorn (2010), we could exploit their observations to interpret the relationship between muscle activity and neural modulation. The average cancellation time for muscles reported by Scangos and Stuphorn (2010) preceded the SSRT by 25 ms. In our sample the average cancellation time for PMd neurons was about 50 ms before the SSRT and therefore in time to drive muscle activity. In addition, we found that neuronal activity of countermanding cells is likely to be dissociated from muscle activity, as the difference between the estimated SSRT and the neural cancellation time does not increase as a function of the SSD's length.
Role of PMd in the brain inhibitory network.
In humans, it has been suggested that the ability to withhold manual motor responses relies critically on the action of a right lateralized fronto-basal-ganglia-thalamic pathway in the motor regions. This network comprises two areas of the frontal cortex, the IFC (Aron et al. 2003, 2007; Chambers et al. 2006) and pre-SMA (Aron et al. 2007; Floden and Stuss 2006; Nachev et al. 2007). Both areas are thought to modulate the cortical neural processes for movement initiation via the hyperdirect route, passing through the subthalamic nucleus (Aron and Poldrack 2006; Aron et al. 2007; van den Wildenberg et al. 2006). Recently, Li et al. (2008) demonstrated that the head of the caudate nucleus plays a key function in movement suppression.
In monkeys during an arm countermanding task, Scangos and Stuphorn (2010) found that the activity of SMA and pre-SMA neurons is not sufficient to control arm movement initiation because the great majority of cells with movement-related activity did not change their activity when a reach was performed with respect to when it was canceled. However, since the discharge of movement-related neurons was reward contingent, Scangos and Stuphorn (2010) concluded that the activity in SMA and pre-SMA represents the motivation for performing a given action, that is, the “urge to act.”
Scangos and Stuphorn (2010) also found a small percentage of neurons (∼2%) that exhibit a countermanding modulation. Those neurons very likely participate in arm movement suppression. The involvement of SMA inhibition of unwanted movements in reaction to the presentation of a stop-signal has been confirmed by Chen et al. (2010), who showed changes of LFP power at low frequencies (10–50 Hz) occurring early enough to be causally involved in movement cancellation. In addition, Chen et al. (2010) demonstrated that SMA plays a key role in proactive control, a form of anticipatory control that, on the basis of known task demands (e.g., presence/absence of stop signal, frequency of stop signals), leads to systematic adjustments of the behavioral performance aimed to enhance the chance of correctly suppressing a movement. However, the low percentage of countermanding neurons found in SMA and pre-SMA areas questions the extent to which those regions are truly involved in the process of suppressing a movement after the appearance of a stop-signal. The LFP modulation observed in SMA by Chen et al. (2010) might in fact not be due to the activity of local neurons but to inputs coming from other brain regions (Logothetis 2003; Mattia et al. 2010) that might provide a source for proactive control.
Even though the exact role of each of these brain regions remains controversial, there is no doubt that their actions have to be exerted through the motor areas. M1 neurons with preparatory activity are a target of SMA output neurons with preparatory activity (Aizawa and Tanji, 1994; Tanji and Kurata, 1985). Somehow neural signals of the motor cortex have to be shaped so that the descending commands to the spinal cord (or to the brain stem) can halt a planned movement. Using TMS, Coxon et al. (2006) demonstrated the involvement of M1 in inhibitory processes. Furthermore, Picton et al. (2007) in humans and Moll and Kuypers (1977) in monkeys pointed out the role of PMd in inhibition, showing that reaching movements become more impulsive and uncontrolled after selective damage to this area.
These studies, however, could not uncover the neural mechanisms underlining the suppression processes. Our study shows, for the first time, the existence of two types of neurons in PMd whose activity significantly changes before reaching arm movements are successfully countermanded in response to a visual stop-signal. Thus we have found that in PMd a substantial proportion of cells produces signals able to participate in the distributed process controlling the execution or the suppression of an arm movement.
This work was supported by MIUR (Grant 2005051741 to S. Ferraina) and by the Italian National Institute of Health (Grant 530/F4/1 to S. Ferraina).
No conflicts of interest, financial or otherwise, are declared by the author(s).
We are grateful to M. Mattia and P. Del Giudice for invaluable comments on a previous version of the manuscript, to R. Caminiti for support throughout this research. and to A. R. Mitz for advice concerning the CORTEX setup. G. Mirabella thanks the Head of the Department of Physiology and Pharmacology of Sapienza University, F. Eusebi, for his support, advice, and encouragement during the preparation of the manuscript.
Present address of P. Pani: Lab Neuro- en Psychopysiologie, K.U. Leuven, Medical School, Campus Gasthuisberg, Herestraat 49, B-3000, Leuven, Belgium.
↵1 We choose 60% of SSDs as the threshold to define a cell as a “countermanding neuron” because this is a very conservative estimate. In fact, when considering the fixed-SSD procedure, 60% means that at least 3 of 4 SSDs must show a significant countermanding-related modulation. On the other hand, when considering the staircase procedure, since each recorded cell had generally 2 analyzable SSDs (see methods), 60% means that all SSDs must show a significant countermanding-related modulation.
- Copyright © 2011 the American Physiological Society