REVIEWS

AN ADAPTIVE CODING MODEL

OF NEURAL FUNCTION IN
PREFRONTAL CORTEX
John Duncan
The prefrontal cortex has a vital role in effective, organized behaviour. Both functional
neuroimaging in humans and electrophysiology in awake monkeys indicate that a fundamental
principle of prefrontal function might be adaptive neural coding — in large regions of the prefrontal
cortex, neurons adapt their properties to carry specifically information that is relevant to current
concerns, producing a dense, distributed representation of related inputs, actions, rewards and
other information. A model based on such adaptive coding integrates the role of the prefrontal
cortex in working memory, attention and control. Adaptive coding points to new perspectives on
several basic questions, including mapping of cognitive to neurophysiological functions, the
influences of task content and difficulty, and the nature of frontal lobe specializations.

SPEEDED RESPONSE CHOICE There is much evidence that the prefrontal cortex of disorganization in many different forms of behaviour
Tasks in which simple stimuli, makes a vital contribution to effective, organized (BOX 1). On the other hand, more general accounts —
such as lights or tones, call for behaviour. Patients with prefrontal lesions can show a including concepts such as EXECUTIVE FUNCTION9, temporal
speeded keypress or other
responses.
broad loosening in the structure of thought and action: structuring of behaviour2, control by cognitive context10
the normal picture, a coherent sequence of actions and or goal–subgoal selection11 — can be hard to apply in
EPISODIC MEMORY mental activities that allow the achievement of some detail to any specific problem.
The recollection of events in an selected goal, is distorted, sometimes bizarrely, by the Essentially, there are two approaches to understand-
autobiographical context.
omission of crucial components and by the intrusion ing broad deficits of the sort associated with prefrontal
EXECUTIVE FUNCTION
of irrelevant or interfering material1,2. According to the lesions. The first is to assume that damage to combina-
High-level processes that are circumstances, the patient might seem mentally passive tions of functionally specialized frontal regions under-
proposed to organize and or inert, or disinhibited and distracted. In formal test- lies the range of deficits seen after typical lesions.
control cognitive function. ing, the result is quantitative impairment in a broad Certainly, this is a plausible view, given the size of pre-
variety of tasks, including picture description1, SPEEDED frontal cortex, and its diversity in terms of cytoarchitec-
3 4 5
RESPONSE CHOICE , EPISODIC MEMORY , maze learning , prob- ture and patterns of connectivity with other brain
lem solving6 and many others. Similarly, functional structures12. The second approach is to assume that, in
imaging studies show activation of the prefrontal cortex some frontal regions at least, functions are sufficiently
in many different task contexts. A broad role in effec- broad to contribute to many different cognitive tasks.
tive cognition is also shown by prefrontal activation in This view, in turn, is given plausibility by the extensive
conventional tests of ‘general intelligence’7 (FIG. 1). interconnections between one frontal region and
Although of unquestioned importance, prefrontal another, indicating substantial sharing and exchange of
MRC Cognition and functions are particularly difficult to characterize and information12.
Brain Sciences Unit, understand. On the one hand, accounts based on highly In this review, the question is considered in the light
15 Chaucer Road,
Cambridge CB2 2EF, UK.
specific deficits — such as an impairment of ‘delayed of evidence from functional neuroimaging in humans
e-mail: john.duncan@ response’ after frontal lesions in the monkey8 — seem and single-unit electrophysiology in the behaving mon-
mrc-cbu.cam.ac.uk too restricted to apply convincingly to the broad problem key. From imaging results, there is a strong indication

820 | NOVEMBER 2001 | VOLUME 2 www.nature.com/reviews/neuro

© 2001 Macmillan Magazines Ltd
 REVIEWS

that many different types of cognitive demand produce a

similar broad patterns of prefrontal recruitment.
Electrophysiological data explain this result, in part, by
showing substantial adaptability of function even at the
level of the single neuron13,14. These conclusions are cap-
tured in an outline model of the role of the prefrontal b LHEC DFIM TQNK HJMQ
cortex in working memory, attention and control — the
c
adaptive coding model (see also REF. 15). The model has Shapes
important implications for the nature of prefrontal spe-
cializations, and for the future design of imaging, single-
unit and lesion studies. It also explains why it is so diffi-
cult to characterize frontal functions in terms of specific
cognitive operations; essentially, prefrontal cortex
adjusts its function to match the requirements of the
particular task undertaken.
Letters

Evidence from functional imaging

In early positron emission tomography (PET) studies
of human brain function, it seemed that distinct, focal
activations of the frontal lobe were produced by such
cognitive demands as word generation16, DIVIDED VISUAL
17 18
ATTENTION and response suppression . However, as evi-
Figure 1 | ‘General intelligence’ reflects prefrontal
dence from PET and functional magnetic resonance function. If any battery of cognitive tasks is administered to a
imaging (fMRI) has accumulated, a different picture has large group of people, the resulting matrix of correlations is
begun to emerge. Certainly, there is some differentiation universally positive — to some extent at least, a person doing
between frontal activations associated with different well on one task is also likely to do well on others97. What
mental activities. Dorsomedial activity, for example, has common factor in different tasks is reflected by this result?
Using multivariate analysis, it can be found which single tasks
been repeatedly and specifically associated with the pro-
are the best measures of a broad ability to perform well —
cessing of social materials19,20, whereas retrieval from empirically, some of the best such measures are novel problem
episodic memory is reliably associated with activations solving tasks, which have been used as tests of ‘general
towards the frontal pole21,22. Accompanying these intelligence’. a,b | Spatial and verbal examples of problem
results, however, is increasing evidence of commonali- solving tasks. Here, the task is to find the set of shapes (a) or
ties in the patterns of frontal activity associated with the letter string (b) that does not belong with the others. c | As
shown by positron emission tomography (PET), lateral
many quite different cognitive demands.
prefrontal recruitment is a characteristic property of such tasks.
Such commonalities have been shown clearly in a The suggestion is that general intelligence is in large part a
recent systematic review21. For this exercise, we chose reflection of prefrontal function. Answers: a | Item 3
studies in the literature that had, as far as possible, (asymmetrical); b | Item 3 (different alphabetical progression).
manipulated a single demand in the context of an oth- Adapted with permission from REF. 7 © 2000 American
erwise identical task. Five demands were considered. Association for the Advancement of Science.
First, a common theme in accounts of prefrontal func-
tion is the suppression of strong but inappropriate this final category, we selected four studies in which
response tendencies. When subjects are asked to name perceptual difficulty was manipulated; for example,
the ink colour of a printed word, for example, there is a object recognition with varying degrees of stimulus
strong tendency to read the word itself if this spells the degradation.
name of a different colour23. For our first category, we Combined results from all twenty selected studies
combined data from six studies that contrasted tasks appear in FIG. 2. Only activations within the prefrontal
with and without such strong, inappropriate response cortex are plotted, on views of the lateral (top row) and
tendencies. A second common theme is the role of the medial (middle row) surfaces of each hemisphere, and
prefrontal cortex in the early, subjectively attentional on views of the whole brain from above (bottom left)
phase compared with the later, more automatic phase and below (bottom right). Each point is a focus of peak
of task performance. For our second category, we com- activation for a direct contrast between high (strong
bined data from five studies that compared early and response suppression, early learning phase, long work-
later phases of a single task. A third theme in the liter- ing memory list, long working memory delay, high per-
ature is the role of the frontal lobe in working mem- ceptual difficulty) and low demand, with different
ory; for our third and fourth categories, we combined colours distinguishing the five demand types.
data from two studies that manipulated the number of In anatomical terms, the results provide striking
items to be retained in a simple working memory list, evidence for regional specialization within the pre-
and, separately, data from three studies that manipu- frontal cortex. On the medial surface, activations are
DIVIDED VISUAL ATTENTION
A requirement to process two or
lated the length of the working memory delay. Finally, almost entirely restricted to the region immediately
more simultaneous stimuli in a we chose to address a cognitive demand that is less dorsal to the corpus callosum, in and around the dorsal
visual display. conventionally associated with prefrontal function. For anterior cingulate. On the lateral surface, activations

NATURE REVIEWS | NEUROSCIENCE VOLUME 2 | NOVEMBER 2001 | 8 2 1

© 2001 Macmillan Magazines Ltd
REVIEWS

Many monkey studies have examined the properties

Box 1 | The delayed response task
of single prefrontal neurons during a wide range of sim-
Study of prefrontal function in the monkey has been profoundly influenced by the ple and complex tasks, including perceptual categoriza-
early finding of deficits in a ‘delayed response’ task after prefrontal lesions48. In this tion, working memory for objects and spatial locations,
task, a cue at the start of a trial indicates a target location; after a few seconds’ delay, rule learning and switching, CROSS-MODAL INTEGRATION and
the monkey must reach or make an eye movement to that location to receive a reward. many other forms of decision making. Commonly,
Because of the need to retain the target location across a brief delay, the deficit has recordings are made in a large region of the lateral sur-
been interpreted as one of ‘working memory’8,9. In this context, variants of the delayed face, spanning the principal sulcus and much of the cor-
response task have been used in highly productive investigations of the tex above and below (FIG. 3a). In detail, each of these
neurophysiology30,36, neuropharmacology91, development92 and many other aspects of studies is valuable in analysing the prefrontal response
prefrontal function. Delay itself, however, is not the only important factor in
to a specific task demand. Taken as a whole, the work
determining task deficits — among other influential factors are interference from
casts light on questions of functional specialization and
distracting sensory inputs50 and competition from the response made on the previous
adaptation.
trial92. Lesions of the type that impair delayed responses also impair other complex
response choices that involve no element of working memory delay93. Undoubtedly,
A first remarkable fact is that, whatever arbitrary
deficits in delay tasks are just one example of a more general cognitive impairment task a monkey has been trained to carry out, a substan-
after prefrontal lesions. tial proportion of lateral prefrontal neurons will be
found to show selective responses to some aspect of
are more scattered, but strong clusters are again visible. that task’s events. In early studies, this result was
A more dorsal cluster in each hemisphere lies in and obscured by an unknown degree of sampling bias; neu-
around the posterior part of the inferior frontal sulcus rons that seemed to be task related were investigated in
(IFS). A more ventral cluster, plotted in the figure as a detail, whereas neurons that did not were abandoned
set of points just anterior to the Sylvian fissure, in fact without thorough testing. However, even with random
extends into the brain along the surface of the frontal sampling, the same result has been found in a number
operculum, becoming continuous with further activa- of tasks, including object–saccade association24, rule
tions within the anterior insula. Although further scat- switching25, spatial delayed responses26, sound–colour
tered points are seen elsewhere, much of the dorso- matching27 and visual same–different comparisons28. If
lateral surface is entirely free of demand-related task-related activity is defined simply as a significant
activations (see brain view from above). Finally, on the
whole orbital surface, only occasional points are seen,
indicating little response to demands of this sort (see
IFS IFS
view from below).
In terms of differing demands, however, there is no
evidence for regional distinctions. Instead, all five
demands show much the same picture of co-recruit-
ment in the dorsal anterior cingulate, the dorsolateral
region around the IFS, and the ventral region extend-
ing into the frontal operculum. Indeed, this overall
SF SF
pattern can be seen even in individual studies, CC
although it is clearer when data from multiple studies
are combined.
Of the two possibilities distinguished earlier, these
data support broad functions in selected frontal regions.
Although this pattern of activation is especially clear in
the above analysis, activations within the same three
frontal regions — dorsal anterior cingulate, peri-IFS,
and ventral/opercular — appear repeatedly in the imag-
ing literature, in studies of perception, response choice,
memory retrieval, language processing, problem solving
and many other cognitive domains21. So, imaging data
lead to a first important discovery — a specific set of
frontal regions that are commonly co-recruited in
response to a diverse range of cognitive challenges.

Evidence from single-unit physiology Figure 2 | Prefrontal activations associated with five
Of course, the imaging method has several restrictions. different cognitive demands. Green, response conflict;
Might different cognitive demands, for example, be pink, task novelty; yellow, number of elements in working
associated with distinct patterns of frontal recruitment memory; red, working memory delay; blue, perceptual
difficulty. Lateral (top) and medial (middle) views of each
on a scale that is too fine for imaging to resolve? hemisphere, together with whole brain views from above
CROSS-MODAL INTEGRATION
A requirement to combine
Assuming that different primate species are comparable, (bottom left) and below (bottom right). CC, corpus callosum;
information from different more detailed information comes from single-unit IFS, inferior frontal sulcus; SF, Sylvian fissure. Adapted with
sensory modalities. physiology in the behaving monkey. permission from REF. 21 © 2000 Elsevier Science.

822 | NOVEMBER 2001 | VOLUME 2 www.nature.com/reviews/neuro

© 2001 Macmillan Magazines Ltd
 REVIEWS

a b Second, these selective responses are of many differ-

C2
Principal sulcus ent kinds, capturing different aspects of task events.
Some responses reflect the occurrence, identity or loca-
C1 C3 tion of stimuli in visual24,29–32 or other27,33 modalities.
Some are motor related, occurring at the time of
9 8
manual34,35 or saccadic24 responses. Many carry informa-
46 tion across delays, indicating a role in working mem-
45 Cats
12 ory30,36. There is information concerning higher-level
10 aspects of task rules, such as whether a response should
2-class be made to matching or mismatching stimuli28, or
boundary
whether the task is to match the identity of successive
stimuli, associate a visual shape with a specific saccade,
Arcuate or simply remember where a previous stimulus was pre-
sulcus Dogs
sented25. Cells of the lateral frontal cortex can also carry
information on reward state — for example, which
reward is available in the current trial block37 — or
D1 D3
respond at the time rewards are given38.
Third, a given type of neural selectivity — such as a
3-class 3-class
boundary D2 boundary response to a given stimulus or delay event — is typi-
cally widely distributed across the lateral frontal surface,
c with neurons of different types within any given task
Baseline Sample Delay Choice closely intermingled. To a first approximation (see
13
below), the usual result is that neurons of all types
examined in any given study are distributed throughout
the recording zone, on dorsolateral, ventrolateral and
Dog 100%
Dog 80%
sometimes even orbital surfaces28,31,38–41.
10 Dog 60% A recent experiment by Freedman et al.42 is worth
considering in detail. For this experiment, morphing
software was used to create stimuli that fell into two gen-
Firing rate (Hz)

eral categories: ‘cats’ and ‘dogs’ (FIG. 3b). Three species of

cat and three breeds of dog were used as prototypes; the
7 software produced linear blends between all possible
prototype pairs (FIG. 3b, double-headed arrows), allow-
ing stimuli to vary continuously, either between ‘cat’ and
‘dog’ (for example, blends between prototypes C1 and
D1), or between two prototypes within the same cate-
4
Cat 100% gory (for example, C1 and C2, or D1 and D2). For the
Cat 80% main experiment, these stimuli were used to train mon-
Cat 60% keys in a successive cat–dog categorization task. On each
trial, two stimuli — ‘sample’ and ‘choice’ — were pre-
sented for 600 ms each, separated by an empty interval
1 of 1,000 ms. Monkeys responded to indicate whether
0 500 1000 1500 2000 2500
Time (ms)
the sample and choice came from the same or different
categories, ignoring within-category differences.
Figure 3 | Neural responses in macaque prefrontal cortex. a | Lateral view of macaque After training in this task, many prefrontal neurons
prefrontal cortex, with numbers indicating approximate cytoarchitectonic subdivisions
(Brodmann areas). b | Cat–dog stimuli from REF. 42. Morphing software created blends between
gave responses that were themselves categorical. For such
all prototype pairs (double-headed arrows), including within-category (for example, C1–C2) and neurons, activity differentiated between cats and dogs,
between-category (for example, C1–D1) cases. c | Mean response of one neuron to six blends. even those close together across the category boundary,
Vertical lines indicate (left to right) sample onset, sample offset, choice onset. This neuron but much less between morphs within a category. An
responded well to all dogs but poorly to all cats, irrespective of proximity to the category example is shown in FIG. 3c; for this neuron, strong
boundary. Adapted with permission from REF. 42 © 2001 American Association for the responses were given to all dogs, even those made up of
Advancement of Science.
only 60% dog and 40% cat, whereas weak responses were
given to all cats, even those made up of only 60% cat and
change from baseline in any trial epoch, the proportion 40% dog. Overall, significantly different responses to cats
of task-related cells in such studies can be close to 100% and dogs were seen in more than 20% of all neurons
(REFS 24,25), and even more specific forms of selectivity encountered throughout a broad, largely ventrolateral
are frequent. In a two-choice object–saccade association area of prefrontal cortex.
task, for example, Asaad et al.24 found 80% of all cells to It is hardly likely that, outside the context of this
be selective for object identity, saccade direction or particular task, more than 20% of prefrontal units act
both; similarly, in a sound–colour matching task, Fuster as cat–dog categorizers. Already, therefore, the results
et al.27 found 29% of all cells to be tone selective. point to an important conclusion: in all probability, the

NATURE REVIEWS | NEUROSCIENCE VOLUME 2 | NOVEMBER 2001 | 8 2 3

© 2001 Macmillan Magazines Ltd
REVIEWS

Importantly, individual neurons are strongly influenced

Box 2 | Integrated competition in visual attention
by task context, with a selective focus on information of
Physiological models of visual selective attention provide a concrete example of the relevance to current behaviour.
problem of processing coherence. Representations of an object’s different properties —
its colour, shape, motion, location in the environment and so on — are distributed The adaptive coding model
across multiple, partially specialized regions of extrastriate cortex94. But cognitive An outline model makes the above conclusions more
experiments show that visual objects are attended as wholes: directing attention to an concrete. This adaptive coding model relates closely to
object makes its multiple properties concurrently available to awareness95. According to three ideas that are prominent in accounts of pre-
the integrated competition hypothesis54–56, objects compete simultaneously for frontal function. The first is a concern with working
representation in multiple extrastriate systems. As an object gains dominance in any one memory, in large part based on the deficits in delayed
system, its representation elsewhere is supported. The net result is a tendency for
response tasks that follow prefrontal ablations in the
multiple systems to converge, giving dominance to the same selected object with all of its
monkey8,48. The second is a concern with selective
properties and implications for action. In part, the prefrontal cortex might have a
attention, reflecting distractibility and the processing
guiding role in this process56, giving initial prominence or support to objects of current
behavioural significance.
of irrelevant inputs in frontal patients49,50. The third is a
concern with cognitive control, which derives from the
disorganization and fragmentation of behaviour in
properties of prefrontal neurons are strongly tuned by patients with frontal lesions1. The adaptive coding
the particular task that the monkey has learned. To con- model owes much to previous accounts based on each
firm this directly, Freedman et al.42 went on to train one of these ideas (REFS 10,51,52 and, in particular, REFS 14,53).
monkey in a new task that was based on the same stim- In this model, working memory, selective attention
ulus set. Now the cat–dog distinction (FIG. 3b, 2-class and control are simply three different perspectives on
boundary) was irrelevant; instead the animal had to sort the same underlying processing function.
stimuli into three new categories, each based on one The central idea is that, throughout much of pre-
cat–dog pair (FIG. 3b, 3-class boundaries). After training frontal cortex — certainly including much of the lateral
on this new task, cat–dog information was now lost surface — the response properties of single neurons are
from neural responses. Instead, these respected the new highly adaptable. Any given cell has the potential to be
category distinctions relevant to the new task that the driven by many different kinds of input — perhaps
monkey had learned. through the dense interconnections that exist within the
Other experiments show a similar focus on infor- prefrontal cortex12. In a particular task context, many
mation of specific relevance to current behaviour. In a cells become tuned to code information that is specifi-
study by Rao et al.39, monkeys performed a combined cally relevant to this task. In this sense, the prefrontal
‘what–where’ working memory task. In different cortex acts as a global workspace or working memory53
phases of each trial, monkeys retained either target onto which can be written those facts that are needed in
identity or target location. Over a wide region of lateral a current mental program. It is exactly this adaptability
prefrontal cortex, many single neurons carried both that is reflected in the large proportions of frontal neu-
identity and location information. Importantly, when rons that are found to code events in whatever arbitrary
the task required a switch from identity to location, task a monkey carries out. The same adaptability is
this switch was reflected in the responses of individual reflected in the imaging finding that the same overall
neurons, identity information being discarded and patterns of prefrontal recruitment are associated with
location information taken up. Studies of visual selec- widely different cognitive demands.
tive attention also indicate selective prefrontal repre- Such adaptability in itself implies selective atten-
sentation of attended or task-relevant objects43–45 and tion, or a selective emphasis on relevant inputs and the
object properties32 in a visual display. filtering out of irrelevant inputs. If a given neuron can
It is worth emphasizing that most recording studies code different information in different task contexts, it
in the monkey have focused on the middle and poste- follows that, in any particular context, there is a selec-
rior parts of the dorsolateral and ventrolateral surfaces. tive removal of inputs that might drive the cell, but are
Less is known, for example, about neural properties in currently unnecessary. Again, the loss of currently
the anterior cingulate, although the picture that is irrelevant stimulus distinctions is seen directly in the
beginning to emerge is again one of multiple, intermin- single-unit data39,42,44,45.
gled response types, with activity related to stimuli, Although more speculative, a further idea relates to
responses, delays, rewards and other aspects of task the familiar view that frontal systems in some sense con-
events38,46,47. Indeed, when lateral prefrontal and ante- trol or shape task-relevant processing elsewhere in the
rior cingulate responses have been directly compared, brain10,52,53. The proposal is that, to achieve processing
the data have indicated highly similar response proper- coherence, multiple brain systems share a strong ten-
ties in these regions38. dency to converge to represent similar or related infor-
At least for the lateral surface, the electrophysiologi- mation54,55 (BOX 2). Because of this, a highly selective
cal results lead to much the same conclusions as does focus on task-relevant information in the prefrontal
functional neuroimaging. Throughout this region, there cortex supports the processing of related information in
is conspicuous neural activity in many different tasks. other systems, including those concerned with the
Such activity produces a dense, distributed description description of sensory inputs, the generation of motor
of inputs, outputs, rewards and other relevant events. commands, the representation of long-term or semantic

824 | NOVEMBER 2001 | VOLUME 2 www.nature.com/reviews/neuro

© 2001 Macmillan Magazines Ltd
 REVIEWS

a Object task Of course, many central questions are left open by this
o outline scheme. One of these is how task relevance itself is
l determined as the prefrontal cortex chooses which infor-
n2 mation to represent. In simple situations, this process can
o
l
be thought of in terms of rewards and their use58; more
n1 broadly, however, detailed world knowledge must be used
to establish how facts do or do not bear on the pursuit of
o particular goals — the central conceptual problem that is
l addressed in artificial intelligence systems for effective
n3
action planning in complex, real-world domains59 (BOX 3).
b Location task A second unresolved question concerns the mechanism
for achieving processing coherence; in particular, the
o
mechanism underlying frontal control of the processing
l
o n2 focus in sensory and other systems60.
However, for some questions, the implications of
l the model are already fairly strong. In the next section,
n1
this point is illustrated with a discussion of regional
o prefrontal specializations.
l
n3 Adaptation and specialization
Figure 4 | The adaptive coding model. a,b | Selectivity of
Much attention has been given in both human and
three model neurons (n1–n3) for object (o) and location (l) is monkey studies to the question of prefrontal regional
indicated by schematic tuning curves; a sharper curve specialization. In imaging work, certainly, the common
reflects greater selectivity for the indicated dimension. aim is to attach specific cognitive interpretations —
Although neurons vary in relative selectivity for object and control of retrieval from episodic or working memory,
location, this variation is also modulated by task context; context setting, error monitoring and so on — to the
object selectivity becomes sharper in an object task (a) and
function of specific frontal regions. On the one hand,
location selectivity becomes sharper in a location task (b).
the adaptive coding model suggests clear limits to this
endeavour. As we have seen (FIG. 4), neurons carrying
knowledge, and the assessment of motivational signifi- very different types of information are closely inter-
cance. A good example is the suggested role of the pre- mingled in the prefrontal cortex, and, even at the level
frontal cortex in the control of visual attention56 (BOX 2): of the single neuron, adaptability is the converse of spe-
in multiple extrastriate systems, objects in the visual cialization. On the other hand, adaptability does not
input compete for representation, and the proposal is imply equipotentiality: neurons undoubtedly differ in
that frontal emphasis on a task-relevant object supports their relative potential for coding the many different
dominance of that object throughout the processing types of information that the prefrontal cortex can rep-
network. Subjectively, a selective prefrontal focus on resent (FIG. 4, compare n1 with n2). One obvious possi-
task-relevant information, with its accompanying domi- bility is that regional specializations within the pre-
nant representation in sensory, motor, memory, motiva- frontal cortex might be statistical rather than absolute
tional and other systems, would correspond to the state (REF. 61; for a more general proposal concerning statisti-
of controlled, active attention to this information, or, cal specializations in brain function, see REF. 62). Across
equivalently, to controlled, active maintenance in work- the prefrontal cortex, there might be a broad distribu-
ing memory. In this way, the prefrontal cortex carries tion of cells with the potential to carry any specific type
out a central function in configuring a flexible cognitive of information; for example, location information in
system to address specific, current concerns. working memory. For different kinds of information,
In FIG. 4, this general view of prefrontal representa- such as location and object, these distributions will
tion is illustrated for the case of working memory for overlap, even down to the level of the single neuron
location and object information30,39,57. Throughout (FIG. 4); but these overlapping distributions might have
much of the lateral prefrontal cortex, cells with variable different shapes and, in particular, different peaks or
potential for coding location and object information regions of maximal sensitivity.
are closely intermingled39,40. When object information This view is highly consistent with single-unit data. In
must be retained in working memory (upper panel), one important study, Ó Scalaidhe et al.63 measured face
object tuning across the population is enhanced, and object selectivity throughout a large region of the
whereas location tuning is weakened. When location monkey’s lateral and orbital frontal cortex. If analysis was
information must be retained, this situation is reversed. restricted only to the most highly selective cells, a strong
As a whole, the population produces a distributed rep- concentration was found on the ventrolateral surface, the
resentation, which selectively favours information of region that receives direct afferents from inferotemporal
current task relevance. As described earlier, exactly this cortex. However, as selection criteria were relaxed,
picture of adaptable location–object representation has increasing numbers of face- and object-selective cells
been shown experimentally across a large region of were found to be distributed elsewhere throughout the
dorsolateral and ventrolateral frontal cortex39. recording region. These data directly show a broad

NATURE REVIEWS | NEUROSCIENCE VOLUME 2 | NOVEMBER 2001 | 8 2 5

© 2001 Macmillan Magazines Ltd
REVIEWS

Box 3 | Coherence and relevance in complex behaviour

The problem of determining task relevance is well illustrated by Luria’s1 insightful analyses of impairment in
frontal patients.
“These patients are unable to systematically analyse the conditions of a problem and to select the important
connections within it … The selective system of operations that normally successively leads to the solution of the
problem disintegrates and is replaced by a series of isolated, fragmentary connections, not subservient to the general
plan and without a clearly defined hierarchical structure.”
Good examples come in the solution of arithmetic problems: “Complex problems, such as ‘A son is 5 years old; in
15 years his father will be twice as old as he. How old is his father now?’ are completely beyond the grasp of such
patients. Without listening to the conditions, they at once begin to make such calculations as 15 × 5 = 75 or 3 × 15 =
45”. Such cases illustrate the potential for irrelevant operations in constructing a sequence of behaviour. The
problem is to focus on just those aspects of the situation that, when adequately recognized, have a bearing on the
intended goal and allow it to be reached.
Much work in symbolic artificial intelligence has been devoted to complex action planning of this sort (for
example, REF. 59). As this work makes clear, the problem is one of effective use of semantic memory — it is knowledge
of the world that dictates how multiple facts and actions bear on the intended goal, and so which paths to pursue and
which to avoid. According to current conceptions, semantic memory is in large part a temporal lobe function96. The
implication is that frontal–temporal interaction must have a basic role in the process of selecting relevant facts and
actions for inclusion in a current task representation.

distribution of neurons carrying some face and object the conclusion will favour strong regional specializa-
information, but with a focused, ventrolateral peak. In a tion. However, with greater demand or power, the full
study that compared location and object information in distribution might be made visible. In this case, the con-
working memory, Rainer et al.40 directly confirmed the clusion will favour overlapping frontal recruitment for
presence of overlapping regions of selectivity for these different types of demand.
two types of information, but found a higher frequency This pattern is clearly evident in imaging studies of
of location-selective cells in the posterior part of the working memory. There is some evidence that pre-
recording area, approaching the arcuate sulcus. In a suc- frontal recruitment is stronger in the left hemisphere if
cessive same–different matching task, with trial-by-trial materials are verbal70. However, such specialization is far
cues indicating whether the monkey should respond to from absolute; although weaker, activation is also seen
matches or mismatches, Wallis et al.28 showed similar in homologous right hemisphere regions, themselves
rule (match versus mismatch) and object selectivity strongly activated when materials are non-verbal70.
across dorsolateral, ventrolateral and orbital surfaces, but Results such as these are strongly consistent with a sta-
again with statistical differences (including higher inci- tistical view of hemispheric specialization, cells that
dence of object selectivity on the ventrolateral surface) code verbal information being most common on the
from one region to another. left, but also present to some extent on the right.
Relative rather than absolute specialization between More generally, activation in several regions of frontal
frontal regions is also the conclusion most consistent and posterior cortex becomes stronger with increased
with lesion data. In human lesion studies, evidence for working memory demand, leading to increasing overlap
different patterns of deficit associated with different pre- of activation associated with different memory materi-
frontal regions is actually remarkably scant, being all but als71. Along related lines, it has been proposed that
restricted to a few cases of hemispheric differences64, and episodic memory retrieval is associated specifically with
differential deficits after orbital and lateral lesions65,66. right frontal recruitment when demands are low; how-
Potentially, monkey studies should be clearer, as lesion ever, with higher demands, a fuller pattern of bilateral
locations are better controlled; but here too, full-scale recruitment might emerge72.
double dissociations are the exception rather than the Such ideas might also help to explain evidence for
rule. For example, despite some evidence of stronger frontal adaptability after damage. As early as three days
spatial deficits after dorsolateral lesions, and stronger after a stroke affecting the left inferior frontal gyrus, a
object deficits after ventrolateral lesions67,68, the broad verbal task that usually recruits this region can instead
picture is of some deficit on both types of task after produce homologous activation on the right73,74. Such
either type of lesion67–69. adaptability is easily comprehensible if the full distribu-
The view that regional specializations are statistical tion of relevant cells extends to both hemispheres,
rather than absolute has strong implications for the but with a particular focus on the left. When the left
interpretation of imaging data. According to this view, hemisphere is intact, it might be the main focus seen in
results will depend jointly on the level of demand an imaging study. However, when it is damaged,
imposed by the task — and so the strength of neural task demands will not be satisfied without stronger
recruitment — and on the statistical power of the recruitment of further relevant cells on the right.
experiment. With low demand or low statistical power, Consistent with this interpretation, some control sub-
the active region observed might be restricted only to jects show weak right-sided recruitment in addition to
the peak of the underlying neuronal distribution. Here, dominant activity on the left74.

826 | NOVEMBER 2001 | VOLUME 2 www.nature.com/reviews/neuro

© 2001 Macmillan Magazines Ltd
 REVIEWS

Of course, this general view permits some cases of itself. In monkey experiments, for example, we need to
relatively strong dissociation. In the monkey lesion know the relative roles of long-term learning versus
literature, some of the clearest double dissociations short-term task context in determining neural proper-
come from comparisons between large lesions of the ties. Do experiments show high proportions of task-
orbital and lateral surfaces75,76. In the imaging litera- related neurons because the same task has been
ture, as we have seen, simple cognitive demands pro- learned over many months, or because the monkey is
duce a pattern of strong lateral but only occasional carrying out this particular task at the moment the
orbital recruitment (FIG. 2); instead, orbital activations recordings are made? To answer such questions, stud-
might be associated with the processing of emo- ies need to be made of how prefrontal responses
tional77 and motivational78,79 materials. Single-unit develop through successive stages of training85, and
data are less clear: although reward-related activity is how they alter with a switch between tasks25 or atten-
certainly one prominent property of orbitofrontal tional sets44,45. Certainly, more data are needed to sup-
neurons 80–82, some direct comparisons between port the most basic prediction of the adaptive coding
orbital and lateral recordings have shown similar model — that single frontal neurons will vary widely
kinds of reward-related and other responses 28,82. in the information that they code from one task context
Further direct comparisons of this sort are needed; to another.
meanwhile, the data as a whole support relatively The statistical view of prefrontal regional special-
strong differentiation between lateral and orbital ization has strong implications for how such special-
functions. izations should be studied. In imaging, comparisons
One particularly significant question raised by the between tasks might profitably be made at various lev-
imaging data concerns comparative function in lat- els of demand; as we have seen, the strong possibility
eral, opercular and anterior cingulate regions. Do all is of focused peak activities at low demand, evolving
three of these show similar adaptive coding properties into a pattern of largely overlapping activity at higher
or, despite their common co-recruitment, do they demand. In electrophysiology, statistical specializa-
make essentially different contributions to cognitive tions can best be seen by a direct quantitative compar-
function? Hints of such differences already exist in the ison of population properties in different prefrontal
imaging literature83,84, and are likely to be developed in regions28,63.
future work. More broadly, quantitative comparisons might also
be crucial in comparing prefrontal and other
Future directions regions86,87. Although this review emphasizes adapt-
The lack of specificity in many models of prefrontal ability in prefrontal responses, such adaptability is
function reflects the breadth of cognitive deficits asso- also, to some extent, a property of many — perhaps
ciated with frontal lesions. Adaptability helps to most — other brain representations. Selective empha-
explain this; in different contexts, the prefrontal cortex sis of task-relevant information, for example, is seen in
might assist in many different cognitive operations. attentional modulations of neural response through-
The adaptive coding model, for example, cannot gen- out the visual system56, at least as early as the primary
erate detailed predictions about exact error types to be visual cortex88. What then is the special contribution of
expected after prefrontal damage. At the physiological the prefrontal representation, leading to such a key
level, however, it is more specific, with fairly direct im- role in overall behavioural coherence? Several possibil-
plications for how future questions can be formulated ities immediately come to mind. In early visual areas,
and addressed. for example, suppression of irrelevant information is
As we have seen, imaging studies show much the predominantly local89, being strongest when relevant
same patterns of prefrontal recruitment in association and irrelevant inputs fall within one cell’s receptive
with many different kinds of cognitive demand. When field. In the prefrontal cortex, such suppression could
this pattern is seen in any individual study, it would be more global, extending to inputs in different hemi-
seem to be inappropriate to interpret this in terms of fields, or perhaps even in different sensory modalities.
that particular study’s demand. Instead, the results Other possibilities are that the prefrontal representa-
probably reflect an adaptation of general prefrontal tion of relevant information is particularly stable in
mechanisms to the specific task context. Similarly, any the face of irrelevant, distracting inputs90; or that it
single-unit study is likely to find large proportions of can adapt more quickly as task context changes.
lateral prefrontal units showing task-related responses. Formulating and testing such hypotheses will be a par-
Again, the interpretation should not be in terms of ticularly important step in developing the adaptive
specific frontal involvement in these particular cogni- coding idea.
tive operations; rather, these data should be viewed as Interestingly, such approaches point to a view of
an adaptation of neural properties to this particular the prefrontal cortex not so much as the seat of partic-
task’s requirements. On the whole, the adaptive cod- ular cognitive operations, but as a resource that gives
ing model argues against simply extending these such operations greater focus, power or flexibility. In
common findings to an ever-expanding list of further any given experiment, the challenge is then to separate
task domains. the general from the specific — to deduce general
Instead, the model emphasizes other research ques- principles of prefrontal function from its contribution
tions. One, as we have seen, is the nature of adaptability to one particular sample of effective behaviour.

NATURE REVIEWS | NEUROSCIENCE VOLUME 2 | NOVEMBER 2001 | 8 2 7

© 2001 Macmillan Magazines Ltd
REVIEWS

1. Luria, A. R. Higher Cortical Functions in Man (Tavistock, 29. Fuster, J. M., Bauer, R. H. & Jervey, J. P. Functional 55. Duncan, J., Humphreys, G. & Ward, R. Competitive brain
London, 1966). interactions between inferotemporal and prefrontal cortex in activity in visual attention. Curr. Opin. Neurobiol. 7, 255–261
A rich, indispensable, clinical description of the a cognitive task. Brain Res. 330, 299–307 (1985). (1997).
loosened structure of thought and behaviour that can 30. Funahashi, S., Bruce, C. J. & Goldman-Rakic, P. S. 56. Desimone, R. & Duncan, J. Neural mechanisms of selective
follow frontal lobe lesions. Mnemonic coding of visual space in the monkey’s visual attention. Annu. Rev. Neurosci. 18, 193–222 (1995).
2. Fuster, J. M. The Prefrontal Cortex: Anatomy, Physiology, dorsolateral prefrontal cortex. J. Neurophysiol. 61, 331–349 57. Wilson, F. A. W., Ó Scalaidhe, S. P. & Goldman-Rakic, P. S.
and Neuropsychology of the Frontal Lobe 2nd edn (Raven, (1989). Dissociation of object and spatial processing domains in
New York, 1989). 31. Watanabe, M. Prefrontal unit activity during delayed primate prefrontal cortex. Science 260, 1955–1958 (1993).
3. Drewe, E. A. Go–no go learning after frontal lobe lesions in conditional go/no-go discrimination in the monkey. I. 58. Braver, T. S. & Cohen, J. D. in Control of Cognitive
humans. Cortex 11, 8–16 (1975). Relation to the stimulus. Brain Res. 382, 1–14 (1986). Processes: Attention and Performance XVIII (eds Monsell,
4. Shimamura, A. P. in The Cognitive Neurosciences (ed. 32. Sakagami, M. & Niki, H. Encoding of behavioral significance S. & Driver, J.) 713–737 (MIT Press, Cambridge,
Gazzaniga, M. S.) 803–813 (MIT Press, Cambridge, of visual stimuli by primate prefrontal neurons: relation to Massachusetts, 2000).
Massachusetts, 1995). relevant task conditions. Exp. Brain Res. 97, 423–436 59. Newell, A. Unified Theories of Cognition (Harvard Univ.
5. Milner, B. Visually-guided maze learning in man: effects of (1994). Press, Cambridge, Massachusetts, 1990).
bilateral hippocampal, bilateral frontal and unilateral cerebral 33. Romo, R., Brody, C. D., Hernández, A. & Lemus, L. 60. Tomita, H., Ohbayashi, M., Nakahara, K., Hasegawa, I. &
lesions. Neuropsychologia 3, 317–338 (1965). Neuronal correlates of parametric working memory in the Miyashita, Y. Top–down signal from prefrontal cortex in
6. Shallice, T. in The Neuropsychology of Cognitive Function prefrontal cortex. Nature 399, 470–473 (1999). executive control of memory retrieval. Nature 401, 699–703
(eds Broadbent, D. E. & Weiskrantz, L.) 199–209 (The Royal 34. Watanabe, M. Prefrontal unit activity during delayed (1999).
Society, London, 1982). conditional go/no-go discrimination in the monkey. II. 61. Haxby, J. V., Petit, L., Ungerleider, L. G. & Courtney, S. M.
7. Duncan, J. et al. A neural basis for general intelligence. Relation to go and no-go responses. Brain Res. 382, 15–27 Distinguishing the functional roles of multiple regions in
Science 289, 457–460 (2000). (1986). distributed neural systems for visual working memory.
8. Goldman-Rakic, P. Topography of cognition: parallel 35. Di Pellegrino, G. & Wise, S. P. Visuospatial versus Neuroimage 11, 380–391 (2000).
distributed networks in primate association cortex. Annu. visuomotor activity in the premotor and prefrontal cortex of Strong fMRI evidence that regional specializations in
Rev. Neurosci. 11, 137–156 (1988). a primate. J. Neurosci. 13, 1227–1243 (1993). a working memory network are relative rather than
9. Baddeley, A. D. Working Memory (Oxford Univ. Press, 36. Fuster, J. M. & Alexander, G. E. Neuron activity related to absolute.
Oxford, UK, 1986). short-term memory. Science 173, 652–654 (1971). 62. Requin, J., Riehle, A. & Seal, J. in Attention and
10. Cohen, J. D. & Servan-Schreiber, D. Context, cortex, and 37. Watanabe, M. Reward expectancy in primate prefrontal Performance XIV (eds Meyer, D. E. & Kornblum, S.)
dopamine: a connectionist approach to behavior and neurons. Nature 382, 629–632 (1996). 745–769 (MIT Press, Cambridge, Massachusetts, 1993).
biology in schizophrenia. Psychol. Rev. 99, 45–77 (1992). A demonstration of reward-related activity in neurons 63. Ó Scalaidhe, P., Wilson, F. A. W. & Goldman-Rakic, P. S.
11. Duncan, J., Emslie, H., Williams, P., Johnson, R. & Freer, C. of the lateral prefrontal cortex. Face-selective neurons during passive viewing and working
Intelligence and the frontal lobe: the organization of goal- 38. Niki, H. & Watanabe, M. Prefrontal and cingulate unit activity memory performance of rhesus monkeys: evidence for
directed behavior. Cogn. Psychol. 30, 257–303 (1996). during timing behavior in the monkey. Brain Res. 171, intrinsic specialization of neuronal coding. Cereb. Cortex 9,
12. Pandya, D. N. & Yeterian, E. H. Comparison of prefrontal 213–224 (1979). 459–475 (1999).
architecture and connections. Phil. Trans. R. Soc. Lond. B One of the few direct comparisons of neuronal An important demonstration that regional specificity
351, 1423–1432 (1996). properties in dorsolateral prefrontal cortex and within prefrontal cortex depends on the criterion for
13. Miller, E. K. The prefrontal cortex and cognitive control. anterior cingulate, revealing highly similar response cell classification; although the most highly face-
Nature Rev. Neurosci. 1, 59–65 (2000). types in these two regions. selective cells are clustered on the ventrolateral
14. Miller, E. K. & Cohen, J. D. An integrative theory of prefrontal 39. Rao, S. C., Rainer, G. & Miller, E. K. Integration of what and surface, more weakly selective cells are more broadly
function. Annu. Rev. Neurosci. 24, 167–202 (2001). where in the primate prefrontal cortex. Science 276, distributed.
An authoritative recent review and theoretical 821–824 (1997). 64. Milner, B. Interhemispheric differences in the localization of
synthesis of physiological and cognitive work on 40. Rainer, G., Asaad, W. F. & Miller, E. K. Memory fields of psychological processes in man. Br. Med. Bull. 27,
prefrontal function. neurons in the primate prefrontal cortex. Proc. Natl Acad. 272–277 (1971).
15. Duncan, J. & Miller, E. K. in Principles of Frontal Lobe Sci. USA 95, 15008–15013 (1998). 65. Bechara, A., Damasio, H., Tranel, D. & Anderson, S. W.
Function (eds Stuss, D. T. & Knight, R. T.) (Oxford Univ. 41. White, I. M. & Wise, S. P. Rule-dependent neuronal activity Dissociation of working memory from decision making
Press, Oxford, UK, in the press). in the prefrontal cortex. Exp. Brain Res. 126, 315–335 (1999). within the human prefrontal cortex. J. Neurosci. 18,
16. Petersen, S. E., Fox, P. T., Posner, M. I., Mintun, M. & 42. Freedman, D. J., Riesenhuber, M., Poggio, T. & Miller, E. K. 428–437 (1998).
Raichle, M. E. Positron emission tomographic studies of the Categorical representation of visual stimuli in the primate 66. Stuss, D. T. et al. Wisconsin Card Sorting Test performance
cortical anatomy of single word processing. Nature 331, prefrontal cortex. Science 291, 312–316 (2001). in patients with focal frontal and posterior brain damage:
585–589 (1988). A key experiment showing that neurons in a wide effects of lesion location and test structure on separable
17. Corbetta, M., Miezin, F. M., Dobmeyer, S., Shulman, G. L. & region of the lateral prefrontal cortex adapt to task cognitive processes. Neuropsychologia 38, 388–402 (2000).
Petersen, S. E. Selective and divided attention during visual context, coding just those stimulus distinctions or A comparison of cognitive deficits after lesions in
discriminations of shape, color, and speed: functional categorizations of current behavioural significance. different regions of prefrontal cortex, showing
anatomy by positron emission tomography. J. Neurosci. 11, 43. Schall, J. D., Hanes, D. P., Thompson, K. G. & King, D. J. comparative preservation of function in orbitofrontal
2383–2402 (1991). Saccade target selection in frontal eye field of macaque. I. patients.
18. Pardo, J. V., Pardo, P. J., Janer, K. W. & Raichle, M. E. The Visual and premovement activation. J. Neurosci. 15, 67. Passingham, R. Delayed matching after selective prefrontal
anterior cingulate cortex mediates processing selection in 6905–6918 (1995). lesions in monkeys (Macaca mulatta). Brain Res. 92,
the Stroop attentional conflict paradigm. Proc. Natl Acad. 44. Rainer, G., Asaad, W. F. & Miller, E. K. Selective 89–102 (1975).
Sci. USA 87, 256–259 (1990). representation of relevant information by neurons in the 68. Mishkin, M. & Manning, F. J. Nonspatial memory after
19. Goel, V., Grafman, J., Sadato, N. & Hallett, M. Modeling primate prefrontal cortex. Nature 393, 577–579 (1998). selective prefrontal lesions in monkeys. Brain Res. 143,
other minds. Neuroreport 6, 1741–1746 (1995). 45. Everling, S., Tinsley, C. J., Gaffan, D. & Duncan, J. Neural 313–323 (1978).
20. Fletcher, P. C. et al. Other minds in the brain: a functional activity in a focused attention task in monkey prefrontal 69. Petrides, M. Impairments on nonspatial self-ordered and
imaging study of ‘theory of mind’ in story comprehension. cortex. Soc. Neurosci. Abstr. 30, 2227 (2000). externally ordered working memory tasks after lesions of
Cognition 57, 109–128 (1995). 46. Shima, K. & Tanji, J. Role for cingulate motor area cells in the mid-dorsal part of the lateral frontal cortex of the
21. Duncan, J. & Owen, A. M. Common regions of the human voluntary movement selection based on reward. Science monkey. J. Neurosci. 15, 359–375 (1995).
frontal lobe recruited by diverse cognitive demands. Trends 282, 1335–1338 (1998). 70. Postle, B. R. & D’Esposito, M. Evaluating models of the
Neurosci. 23, 475–483 (2000). 47. Procyk, E., Tanaka, Y. L. & Joseph, J. P. Anterior cingulate topographical organization of working memory function in
A systematic analysis of published studies showing activity during routine and non-routine sequential behaviors frontal cortex with event-related fMRI. Psychobiology 28,
common patterns of prefrontalrecruitment for a in macaques. Nature Neurosci. 3, 502–508 (2000). 146–155 (2000).
broad range of different cognitive demands. 48. Jacobsen, C. E. Functions of the frontal association area in 71. Nystrom, L. E. et al. Working memory for letters, shapes,
22. Wagner, A. D., Desmond, J. E., Glover, G. H. & Gabrieli, J. primates. Arch. Neurol. Psych. 33, 558–569 (1935). and locations: fMRI evidence against stimulus-based
D. E. Prefrontal cortex and recognition memory. Functional- 49. Chao, L. L. & Knight, R. T. Contribution of human prefrontal regional organization in human prefrontal cortex.
MRI evidence for context-dependent retrieval processes. cortex to delay performance. J. Cogn. Neurosci. 10, Neuroimage 11, 424–446 (2000).
Brain 121, 1985–2002 (1998). 167–177 (1998). 72. Nolde, S. F., Johnson, M. K. & Raye, C. L. The role of
23. Stroop, J. R. Studies of interference in serial verbal 50. Malmo, R. R. Interference factors in delayed response in prefrontal cortex during tests of episodic memory. Trends
reactions. J. Exp. Psychol. 18, 643–662 (1935). monkeys after removal of frontal lobes. J. Neurophysiol. 5, Cogn. Sci. 2, 399–406 (1998).
24. Asaad W. F., Rainer, G. & Miller, E. K. Neural activity in the 295–308 (1942). 73. Thulborn, K. R., Carpenter, P. A. & Just, M. A. Plasticity of
primate prefrontal cortex during associative learning. 51. Engle, R. W., Kane, M. J. & Tuholski, S. W. in Models of language-related brain function during recovery from
Neuron 21, 1399–1407 (1998). Working Memory: Mechanisms of Active Maintenance and stroke. Stroke 30, 749–754 (1999).
25. Asaad, W. F., Rainer, G. & Miller, E. K. Task-specific neural Executive Control (eds Miyake, A. & Shah, P.) 102–134 74. Rosen, H. J. et al. Neural correlates of recovery from
activity in the primate prefrontal cortex. J. Neurophysiol. 84, (Cambridge Univ. Press, Cambridge, UK, 1999). aphasia after damage to left inferior frontal cortex.
451–459 (2000). 52. Norman, D. A. & Shallice, T. Attention to Action: Willed and Neurology 55, 1883–1894 (2000).
26. Funahashi, S. & Inoue, M. Neuronal interactions related to Automatic Control of Behavior. Report No. 8006, Univ. 75. Butter, C. M. Perseveration in extinction and in discrimination
working memory processes in the primate prefrontal cortex California, Cent. Hum. Inf. Process. (1980). reversal tasks following selective frontal ablations in Macaca
revealed by cross-correlation analysis. Cereb. Cortex 10, 53. Dehaene, S., Kerszberg, M. & Changeux, J.-P. A neuronal mulatta. Physiol. Behav. 4, 163–171 (1969).
535–551 (2000). model of a global workspace in effortful cognitive tasks. 76. Dias, R., Robbins, T. W. & Roberts, A. C. Dissociation in
27. Fuster, J. M., Bodner, M. & Kroger, J. K. Cross-modal and Proc. Natl Acad. Sci. USA 95, 14529–14534 (1998). prefrontal cortex of affective and attentional shifts. Nature
cross-temporal association in neurons of frontal cortex. A computational model sharing many central aspects 380, 69–72 (1996).
Nature 405, 347–351 (2000). with the adaptive coding model. 77. Drevets, W. C. Neuroimaging and neuropathological
28. Wallis, J. D., Anderson, K. C. & Miller, E. K. Single neurons 54. Duncan, J. in Attention and Performance XVI (eds Inui, T. & studies of depression: implications for the cognitive-
in prefrontal cortex encode abstract rules. Nature 411, McClelland, J. L.) 549–578 (MIT Press, Cambridge, emotional features of mood disorders. Curr. Opin.
953–956 (2001). Massachusetts, 1996). Neurobiol. 11, 240–249 (2001).

828 | NOVEMBER 2001 | VOLUME 2 www.nature.com/reviews/neuro

© 2001 Macmillan Magazines Ltd
 REVIEWS

78. Elliott, R., Frith, C. D. & Dolan, R. J. Differential neural 86. Chafee, M. W. & Goldman-Rakic, P. S. Matching patterns of 94. Desimone, R. & Ungerleider, L. G. in Handbook of
response to positive and negative feedback in planning and activity in primate prefrontal area 8a and parietal area 7ip Neuropsychology Vol. 2 (eds Boller, F. & Grafman, J.)
guessing tasks. Neuropsychologia 35, 1395–1404 (1997). neurons during a spatial working memory task. 267–299 (Elsevier, Amsterdam, 1989).
79. O’Doherty, J., Kringelbach, M. L., Rolls, E. T., Hornak, J. & J. Neurophysiol. 79, 2919–2940 (1998). 95. Duncan, J. Selective attention and the organization of visual
Andrews, C. Abstract reward and punishment 87. Anderson, K. C. & Miller, E. K. Neural activity in the prefrontal information. J. Exp. Psychol. Gen. 113, 501–517 (1984).
representations in the human orbitofrontal cortex. Nature and posterior parietal cortices during a what-then-where 96. Hodges, J. R., Patterson, K., Oxbury, S. & Funnell, E.
Neurosci. 4, 95–102 (2001). memory task. Soc. Neurosci. Abstr. 30, 975 (2000). Semantic dementia: progressive fluent aphasia with
80. Rolls, E. T. The orbitofrontal cortex. Phil. Trans. R. Soc. 88. Roelfsema, P. R., Lamme, V. A. F. & Spekreijse, H. Object- temporal lobe atrophy. Brain 115, 1783–1806 (1992).
Lond. B 351, 1433–1444 (1996). based attention in the primary visual cortex of the macaque 97. Spearman, C. General intelligence, objectively determined
81. Tremblay, L. & Schultz, W. Relative reward preference in monkey. Nature 395, 376–381 (1998). and measured. Am. J. Psychol. 15, 201–293 (1904).
primate orbitofrontal cortex. Nature 398, 704–708 (1999). 89. Moran, J. & Desimone, R. Selective attention gates visual
82. Hikosaka, K. & Watanabe, M. Delay activity of orbital and processing in the extrastriate cortex. Science 229, 782–784 Acknowledgements
lateral prefrontal neurons of the monkey varying with (1985). I am grateful to E. Miller for his contribution to many of the ideas
different rewards. Cereb. Cortex 10, 263–271 (2000). 90. Miller, E. K., Erickson, C. A. & Desimone, R. Neural presented in this paper.
83. MacDonald, A. W., Cohen, J. D., Stenger, V. A. & Carter, mechanisms of visual working memory in prefrontal cortex
C. S. Dissociating the role of the dorsolateral prefrontal and of the macaque. J. Neurosci. 16, 5154–5167 (1996).
anterior cingulate cortex in cognitive control. Science 280, 91. Williams, G. V. & Goldman-Rakic, P. S. Modulation of Online links
1835–1838 (2000). memory fields by dopamine D1 receptors in prefrontal
84. Owen, A. M. The functional organization of working memory cortex. Nature 376, 572–575 (1995). DATABASES
processes within human lateral frontal cortex: the 92. Diamond, A. & Goldman-Rakic, P. S. Comparison of human The following terms in this article are linked online to:
contribution of functional neuroimaging. Eur. J. Neurosci. 9, infants and rhesus monkeys on Piaget’s A-not-B task: MIT Encyclopedia of Cognitive Sciences:
1329–1339 (1997). evidence for dependence on dorsolateral prefrontal cortex. http://cognet.mit.edu/MITECS/
85. Chen, L. L. & Wise, S. P. Neuronal activity in the Exp. Brain Res. 74, 24–40 (1989). attention | hemispheric specialization | magnetic
supplementary eye field during acquisition of conditional 93. Gaffan D. & Harrison, S. A comparison of the effects of resonance imaging | positron emission tomography |
oculomotor association. J. Neurophysiol. 73, 1101–1121 fornix transection and sulcus principalis ablation upon spatial working memory
(1995). learning by monkeys. Behav. Brain Res. 31, 207–220 (1989). Access to this interactive links box is free online.

NATURE REVIEWS | NEUROSCIENCE VOLUME 2 | NOVEMBER 2001 | 8 2 9

© 2001 Macmillan Magazines Ltd