- Email: [email protected]
Neurons that Program What to Do and in What Order
Previews 177
Toni, N., Buchs, P.A., Nikonenko, I., Bron, C.R., and Muller, D. (1999). Nature 402, 421–425. Woolley, C.S., Gould, E., Frankfurt, M., and McEwen, B.S. (1990). J. Neurosci. 10, 4035–4039.
Neurons that Program What to Do and in What Order In this issue of Neuron, Lu et al. (2002) describe the activity of single neurons in the SEF of monkeys, an oculomotor area of the frontal lobe, during the performance of stereotyped sequences of saccades. The monkey had to look at one of two identical stimuli, but the only way to choose the “correct” stimulus was to learn and remember its position in each presentation of the sequence. SEF neurons could do it. The most common paradigm to study neuronal correlates of behaviors more complicated than reflexes is to present a combination of stimuli, specify a correct motor response among several possible ones, and reinforce the subject when that correct response is produced according to some arbitrary rule. A delay is usually imposed between stimulus presentation and motor response to allow the investigator to attend to the neuronal events that accompany the subject’s choice. More and more often in such tasks (e.g., Shadlen and Newsome, 2001), saccadic eye movements are used as ideal motor responses because they are simple and natural movements. As there are no two ways to turn our eyes 10⬚ right (to indicate a decision), there is little ambiguity in the firing of a neuron always active before such a saccade. Furthermore, saccades are parts of an intelligent motor repertoire. Yarbus (1967) has shown that, in humans, the sequence of saccades to examine a scene reflects the particular information that the subject is seeking. Norton and Stark (1971) have also shown that, to examine an image, humans tend to repeat a stereotyped, personal scan-path. Ordinarily, the stimuli used in such studies are distinguishable from each other by some feature (shape, size, color, movement, duration, oddity, etc) or by their position. However, none of these attributes was helpful in the experiment of Lu et al. because there were only two identical stimuli (dots) in each presentation, and their relative position by itself did not matter. In each display, either dot could be the target, then the other was a distractor. As each trial consisted of five stereotyped presentations (repeats allowed) forming a “hyperset,” the only way for the monkey to be correct (and rewarded) was to know in advance the sequence of correct responses for each hyperset. Learning proceeded by trial and error. The training that made this experiment successful is a remarkable achievement for which the investigators (and perhaps the monkeys) should be complimented. The choice of the SEF as the structure to be investigated is justified for several reasons. First, the SEF is
clearly an oculomotor area. When SEF cells become active, it is regularly in advance of an eye movement in a specific direction, and electrical stimulation at such sites—as tested in the present study—easily evokes such eye movements. Whereas, obviously, eye movements are the output of this region, the input is less obvious. Second, the SEF is contiguous to the supplementary motor area (SMA), a region known to be implicated in the performance of sequential motor tasks (Tanji and Shima, 1994). Third, confirming this view in humans, Gaymard et al. (1990) found that SEF lesions impair the performance of sequences of memory saccades. More recently, Tobler and Mu¨ri (2002) found that transcranial magnetic stimulation over the SEF can disrupt the sequence of saccades. Finally, SEF can exhibit rapid changes of activity with associative learning (Chen and Wise, 1995). And one important dimension for the SEF seems to be spatial location; it is more important than any other stimulus attribute (Tremblay et al., 2002). Lu et al. describe three types of SEF cells. D cells (D for direction) are the most trivial because they conform to what almost all studies have said (see above): that the most important dimension for SEF neurons is the direction of the impending eye movement. D neurons are likely to be the SEF output neurons. It is known that this output directly reaches parts of the saccade brain stem generator. C neurons (C for target/distractor combination) are more intriguing. Their activation seems to correspond to a condition where it is specified not only the target to choose, but also the distractor to avoid. There is some evident similarity between this situation and the performance of antisaccades in conflict with prosaccades (Schlag-Rey et al., 1997). S neurons (S for sequence) are most interesting candidates for memory repositories, elemental components in a sequential performance. It is difficult to explain the activation of such neurons in terms of known physiological mechanisms. However, because of their existence, one can start dreaming of recording the correlate of a musical note, not simply because it has a given sound frequency, but because it occurs at a particular place in a theme. Like all landmark studies, this one raises a number of issues. First, although well trained, the monkeys certainly made errors in the sequential task, and one would like to know what the cell activity was before wrong moves: was it in line with what the monkey should have done or with what it did? Second, are there observable differences during the learning of a new hyperset and the execution of a familiar one, as shown in pre-SMA (Nakamura et al., 1998)? Along the same lines, would it not be worthwhile to try to reproduce the observation of neuronal activation up to the moment when the monkey has found the correct responses by trial and error, and the immediate disappearance of this activation thereafter, when the solution has been found (in anterior cingulate cortex, Procyk et al., 2000)? Third, is there in SEF any evidence for numerical representation for moves in a sequential task, as now prominently found in parietal cortex (Sawamura et al., 2002)? None has been found yet, but this question may call for a different paradigm. Fourth, and most importantly, what is common between the various findings regarding the role of SEF neurons? Do they converge toward a common
Neuron 178
interpretation in terms of internally generated, nonreflexive commands of inquisitive gazing? Finally, this study exemplifies one of the great and still unchallenged merits of single-unit recording. Since neurons active during particular phases of sequential tasks are packed together, imaging techniques cannot miss to reveal a global activation during the whole execution of a sequential task. But these techniques would not yet be able to reveal the specific contribution of individual cells in the performance of sequences. John D. Schlag Department of Neurobiology School of Medicine CHS- UCLA Los Angeles, California 90095
Selected Reading Chen, L.L., and Wise, S.P. (1995). J. Neurophysiol. 73, 1101–1121. Gaymard, B., Pierrot-Deseilligny, C., and Rivaud, S. (1990). Ann. Neurol. 28, 622–626. Lu, X., Matsuzawa, M., and Hikosaka, O. (2002). Neuron 34, this issue, 317–325. Nakamura, K., Sakai, K., and Hikosaka, O. (1998). J. Neurophysiol. 80, 2671–2687. Norton, D., and Stark, L. (1971). Science 171, 308–311. Procyk, E., Tanaka, Y.L., and Joseph, J.P. (2000). Nat. Neurosci. 3, 502–508. Sawamura, H., Shima, K., and Tanji, J. (2002). Nature 415, 918–922. Schlag-Rey, M., Amador, N., Sanchez, H., and Schlag, J. (1997). Nature 390, 398–401. Shadlen, M.N., and Newsome, W.T. (2001). J. Neurophysiol. 86, 1916–1936. Tanji, J., and Shima, K. (1994). Nature 408, 857–860. Tobler, P.N., and Mu¨ri, R. (2002). Neuroreport. 13, 253–255. Tremblay, L., Gettner, S.N., and Olson, C.R. (2002). J. Neurophysiol. 87, 333–350. Yarbus, A.L. (1967). Eye Movements and Vision (New York: Plenum Press).
Toni, N., Buchs, P.A., Nikonenko, I., Bron, C.R., and Muller, D. (1999). Nature 402, 421–425. Woolley, C.S., Gould, E., Frankfurt, M., and McEwen, B.S. (1990). J. Neurosci. 10, 4035–4039.
Neurons that Program What to Do and in What Order In this issue of Neuron, Lu et al. (2002) describe the activity of single neurons in the SEF of monkeys, an oculomotor area of the frontal lobe, during the performance of stereotyped sequences of saccades. The monkey had to look at one of two identical stimuli, but the only way to choose the “correct” stimulus was to learn and remember its position in each presentation of the sequence. SEF neurons could do it. The most common paradigm to study neuronal correlates of behaviors more complicated than reflexes is to present a combination of stimuli, specify a correct motor response among several possible ones, and reinforce the subject when that correct response is produced according to some arbitrary rule. A delay is usually imposed between stimulus presentation and motor response to allow the investigator to attend to the neuronal events that accompany the subject’s choice. More and more often in such tasks (e.g., Shadlen and Newsome, 2001), saccadic eye movements are used as ideal motor responses because they are simple and natural movements. As there are no two ways to turn our eyes 10⬚ right (to indicate a decision), there is little ambiguity in the firing of a neuron always active before such a saccade. Furthermore, saccades are parts of an intelligent motor repertoire. Yarbus (1967) has shown that, in humans, the sequence of saccades to examine a scene reflects the particular information that the subject is seeking. Norton and Stark (1971) have also shown that, to examine an image, humans tend to repeat a stereotyped, personal scan-path. Ordinarily, the stimuli used in such studies are distinguishable from each other by some feature (shape, size, color, movement, duration, oddity, etc) or by their position. However, none of these attributes was helpful in the experiment of Lu et al. because there were only two identical stimuli (dots) in each presentation, and their relative position by itself did not matter. In each display, either dot could be the target, then the other was a distractor. As each trial consisted of five stereotyped presentations (repeats allowed) forming a “hyperset,” the only way for the monkey to be correct (and rewarded) was to know in advance the sequence of correct responses for each hyperset. Learning proceeded by trial and error. The training that made this experiment successful is a remarkable achievement for which the investigators (and perhaps the monkeys) should be complimented. The choice of the SEF as the structure to be investigated is justified for several reasons. First, the SEF is
clearly an oculomotor area. When SEF cells become active, it is regularly in advance of an eye movement in a specific direction, and electrical stimulation at such sites—as tested in the present study—easily evokes such eye movements. Whereas, obviously, eye movements are the output of this region, the input is less obvious. Second, the SEF is contiguous to the supplementary motor area (SMA), a region known to be implicated in the performance of sequential motor tasks (Tanji and Shima, 1994). Third, confirming this view in humans, Gaymard et al. (1990) found that SEF lesions impair the performance of sequences of memory saccades. More recently, Tobler and Mu¨ri (2002) found that transcranial magnetic stimulation over the SEF can disrupt the sequence of saccades. Finally, SEF can exhibit rapid changes of activity with associative learning (Chen and Wise, 1995). And one important dimension for the SEF seems to be spatial location; it is more important than any other stimulus attribute (Tremblay et al., 2002). Lu et al. describe three types of SEF cells. D cells (D for direction) are the most trivial because they conform to what almost all studies have said (see above): that the most important dimension for SEF neurons is the direction of the impending eye movement. D neurons are likely to be the SEF output neurons. It is known that this output directly reaches parts of the saccade brain stem generator. C neurons (C for target/distractor combination) are more intriguing. Their activation seems to correspond to a condition where it is specified not only the target to choose, but also the distractor to avoid. There is some evident similarity between this situation and the performance of antisaccades in conflict with prosaccades (Schlag-Rey et al., 1997). S neurons (S for sequence) are most interesting candidates for memory repositories, elemental components in a sequential performance. It is difficult to explain the activation of such neurons in terms of known physiological mechanisms. However, because of their existence, one can start dreaming of recording the correlate of a musical note, not simply because it has a given sound frequency, but because it occurs at a particular place in a theme. Like all landmark studies, this one raises a number of issues. First, although well trained, the monkeys certainly made errors in the sequential task, and one would like to know what the cell activity was before wrong moves: was it in line with what the monkey should have done or with what it did? Second, are there observable differences during the learning of a new hyperset and the execution of a familiar one, as shown in pre-SMA (Nakamura et al., 1998)? Along the same lines, would it not be worthwhile to try to reproduce the observation of neuronal activation up to the moment when the monkey has found the correct responses by trial and error, and the immediate disappearance of this activation thereafter, when the solution has been found (in anterior cingulate cortex, Procyk et al., 2000)? Third, is there in SEF any evidence for numerical representation for moves in a sequential task, as now prominently found in parietal cortex (Sawamura et al., 2002)? None has been found yet, but this question may call for a different paradigm. Fourth, and most importantly, what is common between the various findings regarding the role of SEF neurons? Do they converge toward a common
Neuron 178
interpretation in terms of internally generated, nonreflexive commands of inquisitive gazing? Finally, this study exemplifies one of the great and still unchallenged merits of single-unit recording. Since neurons active during particular phases of sequential tasks are packed together, imaging techniques cannot miss to reveal a global activation during the whole execution of a sequential task. But these techniques would not yet be able to reveal the specific contribution of individual cells in the performance of sequences. John D. Schlag Department of Neurobiology School of Medicine CHS- UCLA Los Angeles, California 90095
Selected Reading Chen, L.L., and Wise, S.P. (1995). J. Neurophysiol. 73, 1101–1121. Gaymard, B., Pierrot-Deseilligny, C., and Rivaud, S. (1990). Ann. Neurol. 28, 622–626. Lu, X., Matsuzawa, M., and Hikosaka, O. (2002). Neuron 34, this issue, 317–325. Nakamura, K., Sakai, K., and Hikosaka, O. (1998). J. Neurophysiol. 80, 2671–2687. Norton, D., and Stark, L. (1971). Science 171, 308–311. Procyk, E., Tanaka, Y.L., and Joseph, J.P. (2000). Nat. Neurosci. 3, 502–508. Sawamura, H., Shima, K., and Tanji, J. (2002). Nature 415, 918–922. Schlag-Rey, M., Amador, N., Sanchez, H., and Schlag, J. (1997). Nature 390, 398–401. Shadlen, M.N., and Newsome, W.T. (2001). J. Neurophysiol. 86, 1916–1936. Tanji, J., and Shima, K. (1994). Nature 408, 857–860. Tobler, P.N., and Mu¨ri, R. (2002). Neuroreport. 13, 253–255. Tremblay, L., Gettner, S.N., and Olson, C.R. (2002). J. Neurophysiol. 87, 333–350. Yarbus, A.L. (1967). Eye Movements and Vision (New York: Plenum Press).