- Email: [email protected]
The Prefrontal Cortex
Neuron, Vol. 22, 15–17, January, 1999, Copyright 1999 by Cell Press
The Prefrontal Cortex: Complex Neural Properties for Complex Behavior Earl K. Miller* Department of Brain and Cognitive Sciences and The Center for Learning and Memory Massachusetts Institute of Technology Cambridge, Massachusetts 02139
Primates have a remarkably diverse and flexible repertoire of sophisticated behaviors. Their actions are not limited to reflexive or fixed reactions to external events. Rather, they can formulate and implement complex plans to achieve often far-removed goals. The brain region most closely associated with this ability is the prefrontal (PF) cortex, which is significantly larger in primates than in other animals. The PF cortex is a collection of cortical areas in the most anterior portion of the frontal lobes (Figure 1). It has long been associated with high-level, “executive” processes needed for voluntary goal-directed behavior. Its damage in humans does not produce a single, characteristic deficit. Rather, it results in disturbances in a variety of functions, including attention, memory, response selection, planning, and inhibitory control. Studies of the neural basis of PF function in monkeys have focused primarily on active short-term, or working, memory. They have revealed that when a delay is imposed between a visual stimulus and a response based on it, many PF neurons show sustained stimulus-related activity (Goldman-Rakic, 1994; Fuster, 1995). Because sensory inputs are often fleeting, this short-term maintenance may be fundamental to many cognitive functions. However, complex behavior requires more than temporary storage. Relevant sensory inputs need to be selected and integrated with other information common to the goal at hand. Also needed are the executive mechanisms that determine, for example, which stimuli are relevant and should be selected. Much less is known about the neural basis of these higher functions. Here, I briefly review some recent studies that are beginning to provide some insight into the role of the PF cortex in selection, integration, and learning associations. Selection In many views of cognition, the type of executive control thought to be mediated by the PF cortex is synonymous with attentional selection, that is, the ability to voluntarily focus awareness on certain sensory inputs, thoughts, or actions. Selection is necessary because higher-order cognitive functions are severely limited in capacity. Indeed, at a given moment we are aware of only a small fraction of available sensory information. Given these capacity limitations, the ability to ignore distractions and select behaviorally relevant information is critical. Evidence that the lateral PF cortex is involved in selection processes comes from a number of studies, including a recent study by Rainer et al. (1998b). Monkeys were first shown a cue object. On subsequent trials, they had to find that object (the target) in a display of * E-mail: [email protected].
Minireview
three objects and remember its location. After a delay, another display of three objects appeared and the monkeys released a lever if the target was in the same location as in the previous display. The activity of most PF neurons reflected the attributes of the target only; the other, irrelevant nontarget objects were almost completely filtered out and had little influence on neural activity. Interestingly, neuronal selectivity for the target occurred very rapidly, as early as 140 ms after display onset. By contrast, target selection is not reflected in the visual cortex until later, for example at about 175 ms in the inferior temporal cortex (Chelazzi et al., 1993). Earlier target selection in the PF cortex suggests that the visual cortex may “inherit” target information from the PF cortex. That is, that the PF cortex may be a source of the top-down signals that mediate attentional selection in the visual cortex (Desimone and Duncan, 1995). This role is also suggested by the observation that humans with PF damage fail to show attentionrelated modulation of visual cortical evoked potentials (Knight, 1997). Complex behavior, however, depends on more than selecting sensory information. To benefit from past experience, we must be able to select (recall) stored knowledge. Watanabe (1996) demonstrated this ability in PF
Figure 1. The Lateral Prefrontal PF Cortex of the Macaque Monkey This figure illustrates how diverse inputs to the PF cortex may be combined to produce the complex properties of its neurons. For simplicity, only some of the PF subregions and a fraction of their inputs and interconnections are illustrated. Cytoarchitectonic areas are numbered and shown in their approximate location. Orange arrows show inputs from dorsal visual cortical areas involved in visuospatial processing. Green arrows show inputs from ventral visual cortical areas involved in object recognition; blue arrows show inputs from auditory cortex. Black arrows show intrinsic prefrontal connections. The studies discussed in this review primarily involve the lateral prefrontal cortex (areas 12, 46, 45, and 8). The ventrolateral PF cortex includes areas 12 and 45, while the dorsolateral PF cortex includes areas 46 and 8 and ventral area 9. Abbreviations: PS, principal sulcus; AS, arcuate sulcus.
Neuron 16
neurons. He found that when monkeys could predict that a specific reward would appear (e.g., raisins, cabbage, etc.), the activity of many PF neurons reflected that expected reward. This ability to “prospectively” code predicted events provides information about expected consequences and is thus critical for choosing among response alternatives. Integration Sensory processing is highly fragmented. Even within a modality, different stimulus attributes may be separately processed. The primate cortical visual system, for example, is thought to analyze the form and color information needed to identify a stimulus (i.e., what it is) largely separately from information about stimulus location (i.e., where it is). An area involved in complex behavior, however, needs to have access to diverse information. Even simple actions almost invariably require satisfying multiple, diverse constraints. When I search for my coffee cup, for example, I have in mind not only what it looks like but also where it is likely to be. So, somewhere and somehow, diverse information such as what and where needs to come together. Given its role in organizing complex behavior and its extensive connections, the PF cortex seems a likely region where integration of diverse information might be evident, particularly when integration is needed for behavior. Early evidence for a role of PF cortical neurons in integrative functions was provided by Fuster and colleagues (Fuster et al., 1982). They trained monkeys to switch between a what memory task (delayed match to sample) and a where memory task (spatial delayed response). They found that cells selectively activated during one or the other task were intermixed throughout the lateral PF cortex. This colocalization of what and where cells suggested an infrastructure for integrating these attributes. Integration could be mediated by a population of neurons whose activity reflects both an object’s identity and its location. Given that task demands have a major influence on which sensory inputs are processed by PF neurons (Rainer et al., 1998b), this might be especially evident during tasks that require integration of these attributes. Rainer et al. (1998a) trained monkeys on such a task. One of five objects appeared in one of twenty-five locations spanning 208 of central vision. After a delay, a test object appeared. If it was identical to the first object and in the same location, monkeys needed to release a lever. Thus, they had to simultaneously remember both what and where. About half of the PF neurons engaged by this task showed activity that reflected both attributes. They were highly object selective yet had discrete receptive fields (on average about 98) that were distributed throughout a wide portion of the visual field. Thus, they could simultaneously convey an object’s identity and its precise location, just what the task demanded. Other studies have also found that many PF neurons can process both object identity and location. Rao et al. (1997), for example, trained monkeys to remember first an object and then its location. Again, about half of the PF neurons engaged by this task processed both what and where information. This, of course, does not necessarily mean that the PF cortex is functionally uniform. Different PF regions
have unique patterns of connections with the rest of the brain, and this may have an impact on processing (e.g., Figure 1). Wilson et al. (1993) trained monkeys on a conditional association task. Each of two objects was uniquely associated with one of two saccades. The monkey made a saccade to the right following one object and a saccade to the left following the other object. More ventrolateral PF (area 12) neurons were activated selectively by this object–spatial associative task than by a “pure” spatial task, in which a spot appeared on the left or right and monkeys made a saccade to its remembered location. This could reflect a greater ventrolateral involvement in what processing or in processing the what–where (object–saccade) associations. Also, PF cells specialized for processing information about faces seem to be highly localized to the ventrolateral PF cortex (O Scalaidhe et al., 1997), much as they are highly localized within the inferior temporal cortex, a visual cortical area important for object recognition. Of course, biases in where certain types of information are processed (or in how they are processed; see Owen et al., 1996) do not preclude integration. Interconnections between different PF regions (Figure 1) could result in a population of PF neurons with multimodal properties. There are, for example, auditory inputs to the lateral PF cortex (Figure 1), and many of its neurons respond to both visual and auditory stimulation (Watanabe, 1992). It is also important to note that the separation between posterior sensory processing systems is often relative, not absolute. Visual cortical areas thought to be relatively specialized for processing either object or spatial information also have neurons selective for, or modulated by, the other attribute (e.g., Sereno and Maunsell, 1998). In any case, PF neurons’ multimodal properties or, more to the point, their ability to acquire them through experience, may be key to their role in guiding behavior. This is discussed in the next section. Associative Learning, Rules, Context, and Cognitive Control The complexity of primate behavior is partially attributable to the fact that primates can acquire new goals and manners of achieving them. Not surprisingly, the PF cortex is thought to be central to this ability. Its executive role in brain function has been hypothesized to result from the acquisition and representation of “rules” that guide goal-directed behavior. In fact, based on the results of animal studies, Passingham (1993) and Wise et al. (1996) argue that rule learning is a cardinal PF function, and the variety of deficits following PF damage arises from its loss. Rule learning depends on forming arbitrary associations between disparate, but behaviorally related, information. We learn that “red” means “stop,” for example. The PF cortex seems well positioned to play a role in associative learning. It is interconnected with all sensory systems, with the motor system, and with limbic structures involved in long-term memory and affect; it is truly “association cortex.” Neurophysiological studies have shown that the activity of lateral PF neurons does reflect learned associations. For example, Asaad et al. (1998) recorded PF activity while monkeys learned to associate particular movements with particular visual stimuli. They learned to saccade to the right after one object was
Minireview 17
presented and saccade to the left after another object was presented. Once this was acquired, monkeys had to then learn the reverse, that is, to saccade leftward after presentation of the object that had previously been associated with a rightward saccade and vice versa. Many lateral PF neurons were selectively activated by particular combinations of objects and saccade directions. For example, a given cell might only be strongly activated when object “A” instructed “saccade left” and not when object “B” instructed the same saccade or when object “A” instructed another saccade. Further, this property developed during learning. At first, activity related to the direction of the saccade appeared only just before the saccade at the end of the trial. As monkeys learned the associations, however, saccade-related activity appeared progressively earlier on each trial, merging with activity related to the stimulus presentation, evident at the start of the trial. Bichot et al. (1996) found that when monkeys were trained to saccade to a target of a particular color, neurons in the frontal eye fields (a PF region whose neurons convey spatial information about forthcoming volitional saccades) became selective for that color. Watanabe (1992) trained monkeys on association tasks in which visual and auditory stimuli signaled, on alternate blocks of trials, that a reward would or would not be delivered. He found that many PF neurons responded to both visual and auditory stimuli and also reflected whether the stimulus was currently associated with reward. He also found that PF activity could reflect a cued location and the type of reward expected at that location (e.g., a raisin or a piece of cabbage) (Watanabe, 1996). Thus, PF activity can reflect stimuli, associated actions, and expected consequences. Cohen and colleagues have suggested how the ability of PF neurons to reflect conjunctions of behaviorally related information might result in mechanisms for executive control. They posit that control emanates from a PF representation of context (Cohen and Servan-Schreiber, 1992). Context is the constellation of task-relevant information needed to guide a given behavior. A simple example of context is from the Stroop test in which subjects must either read a word or report the color in which the word is printed. In this case, the PF activity is thought to convey, among other things, which attribute is relevant. Context-related activity is hypothesized to feed back to posterior cortical systems to enhance processing of task-relevant information. This is similar to the biased competition model of visual attention (Desimone and Duncan, 1995). According to biased competition, neurons in the visual cortex that represent different stimuli are mutually inhibitory. Top-down signals from the PF cortex are excitatory and represent the to-beattended item. These signals increase activity of neurons that process the relevant information and, by virtue of the mutual inhibition, suppress activity of neurons processing irrelevant information. Context representations may act in a similar fashion. However, rather than convey only visual attributes, context representations are thought to be multimodal and include information about stimuli, actions, etc., that have become associated through experience. Thus, they can bias motor as well as sensory processing and also allow appropriate actions to be selected and executed.
Conclusions Recent studies have shown that, consistent with their putative role in the highest level of cognitive function, PF neurons have complex response properties that are highly dependent on, and shaped by, task demands. They selectively process and integrate information needed for a common behavioral goal. Thus, its extensive connections and the ability of its neurons to be modified by experience may allow the PF cortex to play a role in knitting together behaviorally relevant associations, a process needed for acquisition of “rules” that guide goal-directed behavior. This may result in a representation of context, a template of a previously successful configuration of sensory and response-related information that biases processing in other brain regions in favor of task-relevant information. Such complicated and malleable activity would be expected for a region so closely linked with the complexity and flexibility that are the hallmarks of primate behavior. Selected Reading Asaad, W.F., Rainer, G., and Miller, E.K. (1998). Neuron 21, 1399– 1407. Bichot, N.P., Schall, J.D., and Thompson, K.G. (1996). Nature 381, 697–699. Chelazzi, L., Miller, E.K., Duncan, J., and Desimone, R. (1993). Nature 363, 345–347. Cohen, J.D., and Servan-Schreiber, D. (1992). Psychol. Rev. 99, 45–77. Desimone, R., and Duncan, J. (1995). Annu. Rev. Neurosci. 18, 193–222. Fuster, J.M. (1995). Memory in the Cerebral Cortex (Cambridge, MA: MIT Press). Fuster, J.M., Bauer, R.H., and Jervey, J.P. (1982). Exp. Neurol. 77, 679–694. Goldman-Rakic, P.S. (1994). In Motor and Cognitive Function of the Prefrontal Cortex, A.M. Thierry, J. Glowinski, P.S. Goldman-Rakic, and Y. Christen, eds. (Berlin: Springer-Verlag), pp. 112. Knight, R.T. (1997). J. Cogn. Neurosci. 9, 75–91. O Scalaidhe, S.P., Wilson, F.A.W., and Goldman-Rakic, P.S. (1997). Science 278, 1135–1138. Owen, A.M., Evans, A.C., and Petrides, M. (1996). Cereb. Cortex 6, 31–38. Passingham, R. (1993). The Frontal Lobes and Voluntary Action (Oxford: Oxford University Press). Rainer, G., Asaad, W.F., and Miller, E.K. (1998a). Proc. Natl. Acad. Sci. USA 95, 15008–15013. Rainer, G., Asaad, W.F., and Miller, E.K. (1998b). Nature 393, 577–579. Rao, S.C., Rainer, G., and Miller, E.K. (1997). Science 276, 821–824. Sereno, A.B., and Maunsell, J.H.R. (1998). Nature 395, 500–503. Watanabe, M. (1992). Exp. Brain Res. 89, 233–247. Watanabe, M. (1996). Nature 382, 629–632. Wilson, F.A.W., O Scalaidhe, S.P., and Goldman-Rakic, P.S. (1993). Science 260, 1955–1958. Wise, S.P., Murray, E.A., and Gerfen, C.R. (1996). Crit. Rev. Neurobiol. 10, 317–356.
The Prefrontal Cortex: Complex Neural Properties for Complex Behavior Earl K. Miller* Department of Brain and Cognitive Sciences and The Center for Learning and Memory Massachusetts Institute of Technology Cambridge, Massachusetts 02139
Primates have a remarkably diverse and flexible repertoire of sophisticated behaviors. Their actions are not limited to reflexive or fixed reactions to external events. Rather, they can formulate and implement complex plans to achieve often far-removed goals. The brain region most closely associated with this ability is the prefrontal (PF) cortex, which is significantly larger in primates than in other animals. The PF cortex is a collection of cortical areas in the most anterior portion of the frontal lobes (Figure 1). It has long been associated with high-level, “executive” processes needed for voluntary goal-directed behavior. Its damage in humans does not produce a single, characteristic deficit. Rather, it results in disturbances in a variety of functions, including attention, memory, response selection, planning, and inhibitory control. Studies of the neural basis of PF function in monkeys have focused primarily on active short-term, or working, memory. They have revealed that when a delay is imposed between a visual stimulus and a response based on it, many PF neurons show sustained stimulus-related activity (Goldman-Rakic, 1994; Fuster, 1995). Because sensory inputs are often fleeting, this short-term maintenance may be fundamental to many cognitive functions. However, complex behavior requires more than temporary storage. Relevant sensory inputs need to be selected and integrated with other information common to the goal at hand. Also needed are the executive mechanisms that determine, for example, which stimuli are relevant and should be selected. Much less is known about the neural basis of these higher functions. Here, I briefly review some recent studies that are beginning to provide some insight into the role of the PF cortex in selection, integration, and learning associations. Selection In many views of cognition, the type of executive control thought to be mediated by the PF cortex is synonymous with attentional selection, that is, the ability to voluntarily focus awareness on certain sensory inputs, thoughts, or actions. Selection is necessary because higher-order cognitive functions are severely limited in capacity. Indeed, at a given moment we are aware of only a small fraction of available sensory information. Given these capacity limitations, the ability to ignore distractions and select behaviorally relevant information is critical. Evidence that the lateral PF cortex is involved in selection processes comes from a number of studies, including a recent study by Rainer et al. (1998b). Monkeys were first shown a cue object. On subsequent trials, they had to find that object (the target) in a display of * E-mail: [email protected].
Minireview
three objects and remember its location. After a delay, another display of three objects appeared and the monkeys released a lever if the target was in the same location as in the previous display. The activity of most PF neurons reflected the attributes of the target only; the other, irrelevant nontarget objects were almost completely filtered out and had little influence on neural activity. Interestingly, neuronal selectivity for the target occurred very rapidly, as early as 140 ms after display onset. By contrast, target selection is not reflected in the visual cortex until later, for example at about 175 ms in the inferior temporal cortex (Chelazzi et al., 1993). Earlier target selection in the PF cortex suggests that the visual cortex may “inherit” target information from the PF cortex. That is, that the PF cortex may be a source of the top-down signals that mediate attentional selection in the visual cortex (Desimone and Duncan, 1995). This role is also suggested by the observation that humans with PF damage fail to show attentionrelated modulation of visual cortical evoked potentials (Knight, 1997). Complex behavior, however, depends on more than selecting sensory information. To benefit from past experience, we must be able to select (recall) stored knowledge. Watanabe (1996) demonstrated this ability in PF
Figure 1. The Lateral Prefrontal PF Cortex of the Macaque Monkey This figure illustrates how diverse inputs to the PF cortex may be combined to produce the complex properties of its neurons. For simplicity, only some of the PF subregions and a fraction of their inputs and interconnections are illustrated. Cytoarchitectonic areas are numbered and shown in their approximate location. Orange arrows show inputs from dorsal visual cortical areas involved in visuospatial processing. Green arrows show inputs from ventral visual cortical areas involved in object recognition; blue arrows show inputs from auditory cortex. Black arrows show intrinsic prefrontal connections. The studies discussed in this review primarily involve the lateral prefrontal cortex (areas 12, 46, 45, and 8). The ventrolateral PF cortex includes areas 12 and 45, while the dorsolateral PF cortex includes areas 46 and 8 and ventral area 9. Abbreviations: PS, principal sulcus; AS, arcuate sulcus.
Neuron 16
neurons. He found that when monkeys could predict that a specific reward would appear (e.g., raisins, cabbage, etc.), the activity of many PF neurons reflected that expected reward. This ability to “prospectively” code predicted events provides information about expected consequences and is thus critical for choosing among response alternatives. Integration Sensory processing is highly fragmented. Even within a modality, different stimulus attributes may be separately processed. The primate cortical visual system, for example, is thought to analyze the form and color information needed to identify a stimulus (i.e., what it is) largely separately from information about stimulus location (i.e., where it is). An area involved in complex behavior, however, needs to have access to diverse information. Even simple actions almost invariably require satisfying multiple, diverse constraints. When I search for my coffee cup, for example, I have in mind not only what it looks like but also where it is likely to be. So, somewhere and somehow, diverse information such as what and where needs to come together. Given its role in organizing complex behavior and its extensive connections, the PF cortex seems a likely region where integration of diverse information might be evident, particularly when integration is needed for behavior. Early evidence for a role of PF cortical neurons in integrative functions was provided by Fuster and colleagues (Fuster et al., 1982). They trained monkeys to switch between a what memory task (delayed match to sample) and a where memory task (spatial delayed response). They found that cells selectively activated during one or the other task were intermixed throughout the lateral PF cortex. This colocalization of what and where cells suggested an infrastructure for integrating these attributes. Integration could be mediated by a population of neurons whose activity reflects both an object’s identity and its location. Given that task demands have a major influence on which sensory inputs are processed by PF neurons (Rainer et al., 1998b), this might be especially evident during tasks that require integration of these attributes. Rainer et al. (1998a) trained monkeys on such a task. One of five objects appeared in one of twenty-five locations spanning 208 of central vision. After a delay, a test object appeared. If it was identical to the first object and in the same location, monkeys needed to release a lever. Thus, they had to simultaneously remember both what and where. About half of the PF neurons engaged by this task showed activity that reflected both attributes. They were highly object selective yet had discrete receptive fields (on average about 98) that were distributed throughout a wide portion of the visual field. Thus, they could simultaneously convey an object’s identity and its precise location, just what the task demanded. Other studies have also found that many PF neurons can process both object identity and location. Rao et al. (1997), for example, trained monkeys to remember first an object and then its location. Again, about half of the PF neurons engaged by this task processed both what and where information. This, of course, does not necessarily mean that the PF cortex is functionally uniform. Different PF regions
have unique patterns of connections with the rest of the brain, and this may have an impact on processing (e.g., Figure 1). Wilson et al. (1993) trained monkeys on a conditional association task. Each of two objects was uniquely associated with one of two saccades. The monkey made a saccade to the right following one object and a saccade to the left following the other object. More ventrolateral PF (area 12) neurons were activated selectively by this object–spatial associative task than by a “pure” spatial task, in which a spot appeared on the left or right and monkeys made a saccade to its remembered location. This could reflect a greater ventrolateral involvement in what processing or in processing the what–where (object–saccade) associations. Also, PF cells specialized for processing information about faces seem to be highly localized to the ventrolateral PF cortex (O Scalaidhe et al., 1997), much as they are highly localized within the inferior temporal cortex, a visual cortical area important for object recognition. Of course, biases in where certain types of information are processed (or in how they are processed; see Owen et al., 1996) do not preclude integration. Interconnections between different PF regions (Figure 1) could result in a population of PF neurons with multimodal properties. There are, for example, auditory inputs to the lateral PF cortex (Figure 1), and many of its neurons respond to both visual and auditory stimulation (Watanabe, 1992). It is also important to note that the separation between posterior sensory processing systems is often relative, not absolute. Visual cortical areas thought to be relatively specialized for processing either object or spatial information also have neurons selective for, or modulated by, the other attribute (e.g., Sereno and Maunsell, 1998). In any case, PF neurons’ multimodal properties or, more to the point, their ability to acquire them through experience, may be key to their role in guiding behavior. This is discussed in the next section. Associative Learning, Rules, Context, and Cognitive Control The complexity of primate behavior is partially attributable to the fact that primates can acquire new goals and manners of achieving them. Not surprisingly, the PF cortex is thought to be central to this ability. Its executive role in brain function has been hypothesized to result from the acquisition and representation of “rules” that guide goal-directed behavior. In fact, based on the results of animal studies, Passingham (1993) and Wise et al. (1996) argue that rule learning is a cardinal PF function, and the variety of deficits following PF damage arises from its loss. Rule learning depends on forming arbitrary associations between disparate, but behaviorally related, information. We learn that “red” means “stop,” for example. The PF cortex seems well positioned to play a role in associative learning. It is interconnected with all sensory systems, with the motor system, and with limbic structures involved in long-term memory and affect; it is truly “association cortex.” Neurophysiological studies have shown that the activity of lateral PF neurons does reflect learned associations. For example, Asaad et al. (1998) recorded PF activity while monkeys learned to associate particular movements with particular visual stimuli. They learned to saccade to the right after one object was
Minireview 17
presented and saccade to the left after another object was presented. Once this was acquired, monkeys had to then learn the reverse, that is, to saccade leftward after presentation of the object that had previously been associated with a rightward saccade and vice versa. Many lateral PF neurons were selectively activated by particular combinations of objects and saccade directions. For example, a given cell might only be strongly activated when object “A” instructed “saccade left” and not when object “B” instructed the same saccade or when object “A” instructed another saccade. Further, this property developed during learning. At first, activity related to the direction of the saccade appeared only just before the saccade at the end of the trial. As monkeys learned the associations, however, saccade-related activity appeared progressively earlier on each trial, merging with activity related to the stimulus presentation, evident at the start of the trial. Bichot et al. (1996) found that when monkeys were trained to saccade to a target of a particular color, neurons in the frontal eye fields (a PF region whose neurons convey spatial information about forthcoming volitional saccades) became selective for that color. Watanabe (1992) trained monkeys on association tasks in which visual and auditory stimuli signaled, on alternate blocks of trials, that a reward would or would not be delivered. He found that many PF neurons responded to both visual and auditory stimuli and also reflected whether the stimulus was currently associated with reward. He also found that PF activity could reflect a cued location and the type of reward expected at that location (e.g., a raisin or a piece of cabbage) (Watanabe, 1996). Thus, PF activity can reflect stimuli, associated actions, and expected consequences. Cohen and colleagues have suggested how the ability of PF neurons to reflect conjunctions of behaviorally related information might result in mechanisms for executive control. They posit that control emanates from a PF representation of context (Cohen and Servan-Schreiber, 1992). Context is the constellation of task-relevant information needed to guide a given behavior. A simple example of context is from the Stroop test in which subjects must either read a word or report the color in which the word is printed. In this case, the PF activity is thought to convey, among other things, which attribute is relevant. Context-related activity is hypothesized to feed back to posterior cortical systems to enhance processing of task-relevant information. This is similar to the biased competition model of visual attention (Desimone and Duncan, 1995). According to biased competition, neurons in the visual cortex that represent different stimuli are mutually inhibitory. Top-down signals from the PF cortex are excitatory and represent the to-beattended item. These signals increase activity of neurons that process the relevant information and, by virtue of the mutual inhibition, suppress activity of neurons processing irrelevant information. Context representations may act in a similar fashion. However, rather than convey only visual attributes, context representations are thought to be multimodal and include information about stimuli, actions, etc., that have become associated through experience. Thus, they can bias motor as well as sensory processing and also allow appropriate actions to be selected and executed.
Conclusions Recent studies have shown that, consistent with their putative role in the highest level of cognitive function, PF neurons have complex response properties that are highly dependent on, and shaped by, task demands. They selectively process and integrate information needed for a common behavioral goal. Thus, its extensive connections and the ability of its neurons to be modified by experience may allow the PF cortex to play a role in knitting together behaviorally relevant associations, a process needed for acquisition of “rules” that guide goal-directed behavior. This may result in a representation of context, a template of a previously successful configuration of sensory and response-related information that biases processing in other brain regions in favor of task-relevant information. Such complicated and malleable activity would be expected for a region so closely linked with the complexity and flexibility that are the hallmarks of primate behavior. Selected Reading Asaad, W.F., Rainer, G., and Miller, E.K. (1998). Neuron 21, 1399– 1407. Bichot, N.P., Schall, J.D., and Thompson, K.G. (1996). Nature 381, 697–699. Chelazzi, L., Miller, E.K., Duncan, J., and Desimone, R. (1993). Nature 363, 345–347. Cohen, J.D., and Servan-Schreiber, D. (1992). Psychol. Rev. 99, 45–77. Desimone, R., and Duncan, J. (1995). Annu. Rev. Neurosci. 18, 193–222. Fuster, J.M. (1995). Memory in the Cerebral Cortex (Cambridge, MA: MIT Press). Fuster, J.M., Bauer, R.H., and Jervey, J.P. (1982). Exp. Neurol. 77, 679–694. Goldman-Rakic, P.S. (1994). In Motor and Cognitive Function of the Prefrontal Cortex, A.M. Thierry, J. Glowinski, P.S. Goldman-Rakic, and Y. Christen, eds. (Berlin: Springer-Verlag), pp. 112. Knight, R.T. (1997). J. Cogn. Neurosci. 9, 75–91. O Scalaidhe, S.P., Wilson, F.A.W., and Goldman-Rakic, P.S. (1997). Science 278, 1135–1138. Owen, A.M., Evans, A.C., and Petrides, M. (1996). Cereb. Cortex 6, 31–38. Passingham, R. (1993). The Frontal Lobes and Voluntary Action (Oxford: Oxford University Press). Rainer, G., Asaad, W.F., and Miller, E.K. (1998a). Proc. Natl. Acad. Sci. USA 95, 15008–15013. Rainer, G., Asaad, W.F., and Miller, E.K. (1998b). Nature 393, 577–579. Rao, S.C., Rainer, G., and Miller, E.K. (1997). Science 276, 821–824. Sereno, A.B., and Maunsell, J.H.R. (1998). Nature 395, 500–503. Watanabe, M. (1992). Exp. Brain Res. 89, 233–247. Watanabe, M. (1996). Nature 382, 629–632. Wilson, F.A.W., O Scalaidhe, S.P., and Goldman-Rakic, P.S. (1993). Science 260, 1955–1958. Wise, S.P., Murray, E.A., and Gerfen, C.R. (1996). Crit. Rev. Neurobiol. 10, 317–356.