Content uploaded by Narender Ramnani
Author content
All content in this area was uploaded by Narender Ramnani on Sep 28, 2014
Content may be subject to copyright.
184 |MARCH 2004 |VOLUME 5 www.nature.com/reviews/neuro
REVIEWS
Although the importance of the prefrontal cortex
(PFC) for higher-order cognitive functions is largely
undisputed, it is unclear how (and whether) functions
are divided within this region. Cytoarchitectonic sub-
divisions are thought to correspond well with func-
tional boundaries, although there is little agreement
about the functions of specific subregions (for review,
see REF.1). Human neuropsychological studies, lesion
and electrophysiological studies in the monkey and,
more recently, human functional neuroimaging studies
have largely failed to map specific cognitive functions
onto anatomical or cytoarchitectonic subdivisions of
the PFC. For example,the dorsolateral frontal cortex
(BRODMANN AREA (BA) 9/46) has been implicated in many
cognitive functions, including holding spatial informa-
tion ‘on-line’2–4,monitoring and manipulation within
working memory5,6,response selection7,the implementa-
tion of strategies to facilitate memory 8,the organization
of material before encoding9and the verification and
evaluation of representations that have been retrieved
from long-term memory10,11.The mid-ventrolateral
frontal cortex (BA 47) has been specifically implicated in
a similarly wide range of cognitive processes, including
the selection, comparison and judgement of stimuli
held in short-term and long-term memory6,holding
non-spatial information ‘online’2,12,task switching13,
reversal learning14,stimulus selection15,the specification
of retrieval cues10 and the ‘elaboration encoding’of
information into episodic memory16,17.Finally, the
orbitofrontal cortex has been implicated in processes that
involve the motivational or emotional value of incoming
information, including the representation of primary
(unlearned) reinforcers such as taste, smell and
touch18–20,the representation of learnt relationships
between arbitrary neutral stimuli and rewards or
punishments21,22,and the integration of this informa-
tion to guide response selection, suppression and
decision making23–26.
There is another frontal region for which there are
few, if any, coherent theoretical accounts of function. BA
10, also known as the frontal pole or rostral frontal cor-
tex, comprises the most anterior part of the frontal
lobe (FIG. 1) and, despite much data from functional
neuroimaging studies, has remained resistant to
functional description. This is probably because
the interpretation of functional neuroimaging studies
ANTERIOR PREFRONTAL CORTEX:
INSIGHTS INTO FUNCTION FROM
ANATOMY AND NEUROIMAGING
Narender Ramnani* and Adrian M. Owen‡
The anterior prefrontal cortex (aPFC), or Brodmann area 10, is one of the least well understood
regions of the human brain. Work with non-human primates has provided almost no indications
as to the function of this area. In recent years, investigators have attempted to integrate findings
from functional neuroimaging studies in humans to generate models that might describe the
contribution that this area makes to cognition. In all cases, however, such explanations are either
too tied to a given task to be plausible or too general to be theoretically useful. Here, we use an
account that is consistent with the connectional and cellular anatomy of the aPFC to explain the
key features of existing models within a common theoretical framework. The results indicate a
specific role for this region in integrating the outcomes of two or more separate cognitive
operations in the pursuit of a higher behavioural goal.
*Centre for fMRI of the
Brain, Department of
Clinical Neurology,
University of Oxford, John
Radcliffe Hospital, Headley
Way , Ox fo r d OX3 9DU, UK.
‡MRC Cognition and B
rain Sciences Unit,
15 Chaucer Road,
Cambridge, CB1 3QB, UK.
Correspondence to A.M.O.
e-mail: adrian.owen@
mrc-cbu.cam.ac.uk
doi:10.1038/nrn1343
© 2004Nature Publishing Group
NATURE REVIEWS |NEUROSCIENCE VOLUME 5 |MARCH 2004 |185
REVIEWS
Here,we consider the existing accounts of aPFC
function and evaluate each against a common set of
criteria for relating structure to function. First, any useful
theoretical account should generate testable hypotheses.
In particular, converging evidence from studies of corti-
cal activity using functional imaging and studies that use
an intervention approach (such as lesions, transcranial
magnetic stimulation or neuropsychology) is desirable.
Second, any comprehensive functional theory of a par-
ticular brain region should specify the nature of the
information being processed in that area, where it comes
from and what is being done with it. Third, if localization
of function is the aim, then the data that are used to
relate structure to function must have some functional
and anatomical specificity.This means that the level of
functional description must accommodate the range
of tasks that reliably recruit the aPFC, and the recruit-
ment of that area alone should be the common link
between those tasks. The recent functional neuroimaging
literature is filled with proposals concerning specialization
of function within the PFC, although in most cases these
claims are based on a single observed association between
a specific type of behaviour (or task) and activation in a
particular brain region27.
The anatomy of the aPFC
The Brodmann area that is most commonly associated
with the most anterior region of the PFC is BA 10, which
is conspicuously larger28 and, importantly, comprises a
significantly larger proportion of the cortex in humans
than in other species29.However, the frontal pole and
BA 10 are not synonymous.The most dorsal sector of
the frontal pole includes BA 9 in non-human primates
and probably also in the human brain29.Conversely,
BA 10 extends beyond the frontal pole, particularly
in ‘higher’ primate species28.The complexities of cross-
species homology are highlighted by Brodmann’s analy-
sis of the comparative cytoarchitecture of the primate
cerebral cortex30.He reports that the frontal pole in the
monkey brain (Ceropithicus) is not occupied by BA 10 at
all, but instead by BA 12.In this species, BA 10 occupies a
region of the orbital cortex. Brodmann also argued that
the frontal pole of Ceropithicus shares greater homology
with human BA 11 in terms of cytoarchitecture. Others28
have argued that BA 10 is so large in humans that it
should be divided into three sub-areas. Only one of these
occupies the frontal pole (area 10p), whereas the other
two occupy most of the ventromedial PFC (10m
and 10r). FIGURE 2 shows the relative sizes of human and
macaque prefrontal areas on flattened representations of
the cortex. Later studies have used more reliable method-
ologies to show that the relationship between the frontal
pole and aPFC, while complex,is relatively consistent
across species29,31.An important issue for the localization
of activations in the PFC is the location of the borders
between BA 10 and adjacent anatomical zones, particu-
larly those on the lateral convexity. This issue poses
particular difficulties when considering the functional
neuroimaging literature, because no cytoarchitectonic
information is available in scanned subjects and the cell-
ular characteristics need to be approximated on the basis
relies heavily on investigations in non-human primates.
To our knowledge, there are no studies in which the
activity of frontopolar neurons in monkeys has been
recorded, this area being difficult to access and study
electrophysiologically in the macaque. In addition,essen-
tial neuroanatomical studies of the neuronal connections
and cytoarchitecture of this region in non-human
primates have become available only recently.
ab
cd
e
9/46v
8Av
9/46d
131
13m
11l 12m
47/12r 47/12r
47/121
47/121
47/12m
47/12m
47/12s
47/12s
Iai Iai
131
13a
13b 13b
131
13m
13m
14r
14r
11l 11l
Iapm Iapm
Iam
AON AON
Iam
10p 10p
45
IaI
10o
11m
13b
14r
14c 13a Iam
12o
Iapm Ial
Iapl
G
PrCO
OB
OB
121
32 10m
14r
14c
25
24a
24b
24c 9
11m 11m
14c 14c
10r
32ac
32pl
24
10m
10p
25
14c 14r 11m
9
10
46
47/12
45A 45B
44
9
8B
8Ad 6
4
6
Figure 1 | Brodmann area (BA) 10. a–d| The location of cytoarchitectonic BA 10 (shaded in red)
surface-rendered onto the orbital (a) and medial (b) surface of the human brain, and on the brain
of the macaque monkey (cand d). The size of BA 10 (relative to other prefronal areas) seems to
be significantly larger in the human brain than in the macaque monkey brain. Modified, with
permission, from REF. 28 John Wiley and Sons Inc. (2003). e| The location of BA 10 on the
lateral convexity has been studied by Petrides and Pandya106. On the lateral convexity, area 10
incorporates the most anterior parts of the three frontal gyri. AON, anterior olfactory nucleus;
G, gustatory cortex; OB; olfactory bulb; PrCO; precentral opercular area. Reproduced, with
permission, from REF. 106 Blackwell Publishing (1994).
© 2004Nature Publishing Group
186 |MARCH 2004 |VOLUME 5 www.nature.com/reviews/neuro
REVIEWS
unrelated to the task at hand or the sensory environment,
is often referred to as stimulus-independent thought,
and is most likely to occur when the requirement
to process information from external sources is low.
Christoff and Gabrieli44 propose that such unsupervised
mental activity results in ‘mind wandering’and could
underlie the changes observed in the aPFC in studies in
which the experimental requirements are insufficiently
demanding to prevent such activity.This hypothesis
concurs with the observation in functional neuroimaging
experiments that anterior prefrontal regions, including
BA 10, are disproportionately activated during the less
demanding of two conditions, usually the baseline
or rest condition27,45.Although ‘mind wandering’ is
difficult to investigate,the idea that the aPFC is involved
in the explicit processing of internal mental states and
events generates testable hypotheses. In particular, tasks
that require the generation and monitoring of internally
generated responses should disproportionately activate
the aPFC, and patients with damage to this region
should have particular difficulty with problems of
this type. An example is the TOWER OF LONDON TEST of plan-
ning46,in which possible solutions must be internally
generated and then accepted or rejected, before making
a move. This test activates the aPFC, particularly when
it emphasizes internalized, rather than explicit, prob-
lem-solving strategies47.Neurosurgical patients with
damage to the frontal lobe are impaired at this
task46,48,49,although no study has looked specifically at
patients with damage limited to the aPFC. This model
also makes predictions about different types of mem-
ory task; in particular, recall tasks (‘What is it that you
remember?’), which emphasize self-generation, should
involve aPFC more than recognition tasks (‘Do you
remember this?’). Christoff and Gabrieli44 re-evaluated
the episodic memory literature and found a higher
incidence of aPFC activation in episodic recall tasks
than in tasks that required only recognition. Again,
although no studies have looked at patients with
damage limited to aPFC, frontal-lobe damage, in
general, disproportionately impairs recall relative to
recognition memory6,50.
Whereas the notion that the aPFC is crucial for the
explicit processing of internal mental states and events
generates testable hypotheses, our second criterion is
less clearly met. The model does not specify where the
‘internally generated’ information comes from that is
processed by aPFC (spontaneous thoughts, problem
solutions and so on) and the processes that are carried
out on that information (the model’s functional speci-
ficity) are not sufficiently constrained to be falsifiable
(‘internally generated information is evaluated’44).With
respect to anatomical specificity,it is also not clear
whether activity observed in the aPFC can be reliably
differentiated from that observed in the mid-dorsolateral
frontal cortex (BA 9 and 46). For example44,frontopolar
activity has been observed in 13 out of 15 studies involv-
ing episodic memory tasks that require ‘evaluation of
self-generated material’, but 11 of these cases also
showed mid-dorsolateral activity.Similarly, in studies
that involve problem solving, significant activity in the
of gross morphology1.Studies agree that the cytoarchi-
tectonic zone that occupies the anterior polar region of
the frontal lobe is BA 10 (area 10p (REF.28)). These studies
also indicate that area 10 extends into the most anterior
portions of the frontal gyri. So, a conservative estimate of
the posterior border of area 10p on the lateral convexity
would be the most anterior coronal plane in which the
three frontal gyri are present. In this article, we refer to
aPFC as the region within this boundary that is occupied
by area 10p defined according to Ongur et al.28.
One important distinguishing feature of the aPFC,
even in comparison with other areas of supramodal
(prefrontal) cortex, is that the number of dendritic spines
per cell and the spine density are higher than in other
comparable areas of the cortex, but the density of cell
bodies is markedly lower32.This indicates that the com-
putational properties of aPFC are more likely than those
of comparable areas to involve the integration of inputs.
The primate cerebral cortex is hierarchically organized
(BOX 1),with information becoming increasingly abstract
as it is processed at higher points in the cortical hierarchy.
At the apex of each information processing stream is a
supramodal area (regions within the PFC and anterior
temporal cortex) in which information is represented at
its most abstract level33.The aPFC is unique in this
respect: it seems not to be interconnected with ‘down-
stream’areas in the way that other prefrontal areas are.
Therefore, it is the only prefrontal region that is predom-
inantly (and possibly exclusively) interconnected with
supramodal cortex in the PFC34–38,anterior temporal
cortex39,40 and cingulate cortex37,38,41,42.In addition, its
projections are broadly reciprocal43.
Current perspectives on aPFC function
Processing of internal states. Christoff and Gabrieli44
suggested that human area 10 might be specialized “for
the explicit processing of internal mental states and
events — or introspective evaluation of one’s own
thoughts and feelings.” Spontaneous mental activity,
BRODMANN AREA
(BA). Korbinian Brodmann
(1868–1918) was an anatomist
who divided the cerebral cortex
into numbered subdivisions on
the basis of cell arrangements,
types and staining properties
(for example, the dorsolateral
prefrontal cortex contains
subdivisions, including BA 46,
BA 9 and others). Modern
derivatives of his maps are
commonly used as the reference
system for discussion of brain-
imaging findings.
TOWER OF LONDON TEST
A widely used
neuropsychological test of
planning and problem solving.
Participants move a set of three
balls between three rods (or
‘pockets’) to match a separate
goal arrangement.
86
45 12l
12o
12r
11l
11m
10o
46
12m
13l
13l
13m
13m
13b
14r
14r 14c
13a 14c
Iam
Iam
AON
Iapm
Iapm
25
10m
10p
10r 10m
11m
11l
47/12r
45
47/12l
47/12m
47/12s
Iai
PrCO
32
24b
13b
24c
24
32ac
932pl
25
9
46 8
24
Corpus
callosum
Iai
Ial
Ial
G
G
PrCO
*
13a
Right
Human Monkey
Figure 2 | The right prefrontal cortex of in human (left) and macaque (right) brains
represented as flattened surfaces. These illustrate the considerably larger proportion of the
surface area of prefrontal cortex occupied by area 10 in humans relative to macaque monkeys.
AON, anterior olfactory nucleus; G, gustatory cortex; PrCO; precentral opercular area. Modified,
with permission, from REF. 28 John Wiley and Sons, Inc. (2003).
© 2004Nature Publishing Group
NATURE REVIEWS |NEUROSCIENCE VOLUME 5 |MARCH 2004 |187
REVIEWS
both the nature of the task and the retrieval cue.
Therefore, in experimental studies of episodic memory
retrieval, the attentional set and the corresponding brain
state will be established by the task instructions and
maintained throughout the retrieval period. In addition,
patients with damage to this area should be impaired
regardless of the mechanism of retrieval (recall or recog-
nition) and irrespective of whether retrieval cues are
previously studied or non-studied items. Subsequent
studies have provided mixed support for these hypo-
theses. Rugg et al.11 showed that right prefrontal activity
during episodic memory retrieval was greater during a
cued recall task than during a recognition memory task.
This result concurs with neuropsychological findings
that patients with frontal-lobe damage are more
impaired at recall than recognition6.The hypothesis
that aPFC activity is insensitive to whether the retrieval
cue is a studied or unstudied item has also been ques-
tioned11,56.Activity in the right aPFC was shown, in
general, to increase with the proportion of studied rel-
ative to non-studied items presented at retrieval
(although these findings have been challenged57).
However, the hypothesis that the neural signature of
retrieval mode should be sustained activity in the aPFC
throughout retrieval has received some support.
Velanova et al.58 have shown temporally extended
activity in the aPFC during controlled retrieval. They
concluded that the aPFC is involved in forming and
maintaining an attentional set, or task mode, that
extends over several retrieval attempts.
With respect to our second criterion, the retrieval
mode hypothesis has been computationally well-
specified. For example,LePage et al.59 defined retrieval
mode as “A neurocognitive set, or state, in which one
mentally holds in the background of focal attention a
segment of one’s personal past, treats incoming and
on-line information as retrieval cues for particular events
in the past, refrains from task-irrelevant processing and
becomes consciously aware of the contents of successful
ecphory [the cue-triggered retrieval of an episodic mem-
ory] should it occur as a successful event.” However,
where the information comes from, and what the aPFC
does to that information, is vague beyond this general
operational description. Regarding the third criterion,
the retrieval mode hypothesis might be functionally too
specific. Many tasks that do not require retrieval mode,
or indeed any episodic retrieval, activate this region.For
example, one review60 found that the aPFC was routinely
activated during episodic memory retrieval, working
memory and ‘miscellaneous tasks’, leading the authors
to conclude that “the critical processing demands
[aPFC] subserves are most likely shared by a set of tasks
extending beyond episodic retrieval.”
Regarding anatomical specificity, retrieval-related
activity has been observed in many brain regions other
than PFC. For example,in experiments designed specifi-
cally to investigate retrieval mode, robust activity in
multiple sites outside the aPFC was reported, including
the anterior cingulate cortex, the frontal operculum/
ventrolateral frontal cortex and a region of the right
dorsolateral frontal cortex59.
mid-dorsolateral frontal cortex is reported as often as
activity in BA 10 (REF.44).So,although the most anterior
regions of the frontal lobe might be involved in the
evaluation of internally generated information, they are
probably not unique in this respect.
Memory retr ieval models. A number of authors have sug-
gested that the aPFC is involved in aspects of memory
retrieval, including retrieval mode, success monitoring
(or retrieval verification) and source memory.Although
none of these ideas constitutes a functional model as such
(they do not incorporate data on the aPFC outside the
process under consideration), they are derived from a
considerable data set and therefore require consideration.
These various positions will be considered collectively
and our assessment of each will be necessarily brief.
Tulving51 proposed that a prerequisite for successful
episodic retrieval is that the person doing the remem-
bering is in the appropriate cognitive state, which he
termed ‘retrieval mode’. Retrieval mode is a tonically
maintained state or ‘set’, which is entered into when
episodic memory retrieval is necessary. Importantly, a
stimulus event will be treated as an episodic retrieval cue
only when an individual is in this state. On the basis of a
series of positron emission tomography (PET) imaging
studies52–55,Tulving and colleagues51 proposed that
BA 10 in the right hemisphere is primarily responsible
for retrieval mode. The retrieval mode hypothesis gener-
ates clear predictions: if the aPFC is activated by episodic
memory retrieval, this activity should be invariant to
Box 1 | Information processing in hierarchical networks
Heirarchical organization seems to be a general principle of networks in the primate
brain. Primary cortical areas lie at the foundations of these networks and are
interconnected with supramodal areas of the prefrontal and anterior temporal cortex,
through intermediate connections in unimodal and heteromodal areas32.The visual
and motor systems are well understood examples. In the case of the visual system,
information relayed to area V1 (primary visual cortex) cascades through a series of
areas that elaborate and represent increasingly abstract aspects of visual stimuli (such
as motion-sensitive and colour-sensitive areas95). These relay information to higher-
order visual areas that represent visual information in terms of objects in temporal
lobe circuitry (the ventral visual stream or ‘what’ pathway96–98).Finally,this
information is relayed further to supramodal areas in the ventral prefrontal cortex
(PFC), where it has been argued that object identity is represented in terms of its
context99.Activity in Brodmann area (BA) 12/47, for example, is evoked when there are
breaches of expectation during tasks that are concerned with visual attention100.The
representation of rewards in the PFC is a special case of this general rule. Neurons in
the orbitofrontal cortex can represent not just an object, but also its reward value and
the degree to which it is expected in a given context92,101.In the case of the motor
system, it has been argued that information cascades downwards from supramodal
prefrontal areas to the primary motor cortex through a series of intermediate premotor
areas. This cascade operates to convert relatively abstract goals in prefrontal areas into
motor plans in the premotor system, and finally the more concrete representations of
activity of specific motor units in the primar y motor cortex33.The mid-portion of the
sulcus principalis in the macaque monkey brain (BA 9/46) is the region of PFC that
sends the strongest projections to the premotor system102.Cells in this area become
active when the cognitive parameters of actions are manipulated (for example, the
instruction to select an action from a number of alternatives7,103). The human
homologue of this area is thought to lie in the mid-portion of the middle frontal
gyrus34,and is activated by similar tasks in functional imaging studies7.
© 2004Nature Publishing Group
188 |MARCH 2004 |VOLUME 5 www.nature.com/reviews/neuro
REVIEWS
tested explicitly using PET. Four prospective memory
tasks were given, each involving a different action (for
example, respond in the direction indicated by an
arrow). In certain trials (for example, when two arrows
were the same colour) the volunteers were required to
perform an additional act (such as pressing a buzzer),
and remembering to do this was assumed to make
demands on prospective memory.Regional blood flow
increased bilaterally in the frontal pole during the task,
indicating that this region is crucial for the maintenance
of an intention. Because prospective memory has high
functional specificity, this model makes clear predic-
tions about the sorts of problems that should be
encountered by patients with damage to the aPFC.
Burgess et al.71 described five frontal-lobe patients with
some damage to aPFC. In all cases, their cognitive
deficits included a failure to create and carry out inten-
tions. The model also generates testable hypotheses
about other types of task that should activate this area.
For example, in one study volunteers heard a list of
nouns and had to remember to tap a finger whenever
they heard a previously learned target, a requirement
that clearly involves prospective memory72.Frontopolar
cortex was activated bilaterally. Unfortunately, there are
also a number of studies that use prospective memory,
but do not activate the aPFC. For example,in one such
study,volunteers were asked to watch a series of num-
bers and to press a button whenever a pre-specified
number appeared73,a requirement that is functionally
equivalent to that of the tasks used in REF.70 andREF.72.
This model does reasonably well with respect to our
second criterion: the information stream in a typical
prospective memory task has been well specified by
Burgess et al.60.Situations that require prospective
memory clearly involve a delay (from a few seconds to
hours or days), an intended act (‘to meet with John’), an
ongoing task or set of tasks that occur during this delay
period and are typically unrelated to the intended act or
the retrieval context (thereby preventing continuous
rehearsal) and a trigger to self-initiate the planned activ-
ity.According to Burgess et al.70,the aPFC is involved in
maintaining the intention, while additional processes
required for the realization of that intention are carried
out elsewhere in the brain.
Like the source memory account, a prospective
memory view of aPFC is cognitively well-defined and
cannot account for studies that have reported activity in
this region during tasks that do not require prospective
memory (for review, see REF. 44). With respect to
anatomical specificity,the prospective memory model is
also limited. In both of the imaging studies described
above70,72,frontopolar changes were accompanied by
activity in the mid-dorsolateral frontal cortex, and in
neither case was it possible to differentiate functionally
between these two regions.
Branching and reallocation of attention. Koechlin
et al.74,75 have suggested that the frontopolar cortex medi-
ates ‘cognitive branching’, or “The human ability to hold
in mind goals while exploring and processing secondary
goals, a process generally required in planning and
Theories of episodic memory retrieval often relate
controlled retrieval processing to the recollection of con-
textual details surrounding a previous encounter with a
stimulus58.Ranganath et al.61 used functional magnetic
resonance imaging (fMRI) to look at encoding and
retrieval of pictures of objects. By having volunteers
decide whether the identified pictures were bigger or
smaller at retrieval than at encoding, the authors could
ascertain whether activated regions were modulated by
the specificity of the information to be retrieved. A
region in the left aPFC was activated during retrieval,
and activity increased as the retrieval of more perceptu-
ally detailed information was required. On this basis,
they argued that the frontopolar cortex is important for
source memory — memory not for the item, but for the
context within which it was encoded.
A source memory view of aPFC function generates
clear,testable hypotheses. Tasks that require source
memory will activate aPFC whereas those that don’t will
not, and source memory tasks should be disproportion-
ately impaired in patients with frontal-lobe damage.
Although some studies of context (or source) memory
have found more BA 10 activation for source than for
item memory62–66,others have not16,67.The neuropsycho-
logical literature is similarly equivocal; some studies find
that patients with frontal-lobe damage are dispropor-
tionately impaired at tests of source memory68,but
others find no significant deficit69.
With respect to the second of our criteria, there is no
consensus regarding the nature of the information being
processed by aPFC during source memory tasks. For
example, Ranganath et al.62 suggested that the aPFC
“implements monitoring or evaluative processes that are
engaged when one attempts to retrieve information from
memory,” and “was especially engaged during the evalua-
tion of specific memory characteristics.”On the other
hand,Velanova et al.58 conclude that frontopolar cortex
“may contribute to gating processes in the cognitive
domain, helping to maintain cognitive set during
retrieval, particularly when a retrieval task requires that
one constrain retrieval to a specific past context.” The
source memory account has functional specificity,but
it cannot account for the repeated finding of activity in
this region during tasks that seem not to require the
retrieval of contextual memory (for review, see REF.44).
With respect to anatomical specificity, results are again
mixed, although we know of no study where the aPFC
has been activated in the absence of any other frontal
region during a source memory task58,62.These findings
indicate that models of episodic memory retrieval,
including source memory,might be relevant to under-
standing the role of the aPFC in certain cognitive tasks,
but are incomplete as a theoretical framework for
understanding its functions.
Prospective memory. Although it is less a comprehensive
model than an inferred association between function
and anatomy, the aPFC has been proposed to be crucial
for prospective memory70.Prospective memory allows
an intended act to be carried out after a delay
(‘Remember to meet John at 5pm’). This framework was
© 2004Nature Publishing Group
NATURE REVIEWS |NEUROSCIENCE VOLUME 5 |MARCH 2004 |189
REVIEWS
bilaterally,whereas Braver and Bongiolatti76 found that
the aPFC was co-activated with several other regions,
including the mid-dorsolateral frontal cortex.
Some aspects of the cognitive branching hypothesis
are reminiscent of models of multi-tasking or ‘multiple
sub-goal scheduling’ that have also implicated anterior
regions of the PFC71.Multi-tasking or multiple-task
situations typically require the allocation of attentional
resources in the face of competing demands. Pollmann
and colleagues78,79 have suggested that the frontopolar
cortex is involved in controlling ‘attentional weight shift-
ing’. The core feature of this view is that the most salient
stimuli are often not the most relevant for a task and
intentional selection is often required. Pollmann
et al.78 reported activity in the left lateral frontopolar
cortex in trials where the relevant dimension changed
relative to those where it stayed the same,and concluded
that this region helps to control attentional reallocation
after dimensional changes. In other studies,left fronto-
polar activity was associated with the deployment of
attentional resources away from currently attended
visual dimensions or spatial locations to a new dimension
or location (for review, see REF. 79). However,such changes
were observed only when subjects were not in a top-down
controlled state of selective attention. So,activation of the
left lateral frontopolar cortex seemed to be specific to
stimulus-driven attentional shifts.
This model is easily testable, although as an account
of aPFC function, it is perhaps too functionally specific;
although it is true that the aPFC is activated during some
tasks that require the relocation of attention, this does
not account for its role in most cases where this region
has been activated44.To our knowledge, this model has
not been tested neuropsychologically, although atten-
tional shifting deficits have been reported in patients
with frontal-lobe damage80,81.However, these deficits are
functionally specific and do not correlate well with other
deficits in these patients, again indicating that attentional
shifting is one aspect of, rather than an explanation for,
frontal-lobe function. With respect to information flow,
the model is well-specified conceptually, although
whether the aPFC is involved in the monitoring of events
that make the reallocation of attention necessary or in
the initiation of these attentional shifts is unclear.
Moreover, how this region interacts with other cortical
and sub-cortical regions to control attentional realloca-
tion is not specified. Finally, although the aPFC has been
activated during several tasks that require the realloca-
tion of attention, this region has not been shown to be
unique in that respect78.
Relational integration. An alternative hypothesis is that
the aPFC is required for the explicit representation and
manipulation of relational knowledge82and is essential
for ‘relational integration’; that is, the simultaneous
consideration of multiple relations83 (between objects or
thoughts). The relational complexity of a reasoning task
is assumed to increase with the number of relations that
must be simultaneously considered to infer the required
conclusion84.In one fMRI study,a problem-solving
task based on RAVEN’S PROGRESSIVE MATRICES85,86 was used to
reasoning.” This hypothesis was based on fMRI data that
showed selective bilateral activity in aPFC when volun-
teers were required to keep in mind a main goal (a
working memory task) while performing concurrent
sub-goals (dual-task performance). Similar data led
Braver and Bongiolatti76 to describe the role of fronto-
polar cortex in cognitive branching as “subserving
processes related to the management and monitoring of
sub-goals while maintaining information in working
memory”. In that study, the crucial experimental condi-
tion required subjects to perform semantic classification
as a sub-goal task while concurrently maintaining infor-
mation in working memory and then combining the
results of both sources of information to generate an
appropriate response. So,according to this view, branch-
ing successively allocates processing resources between
concurrent tasks, keeping relevant information in work-
ing memory to allow a return to a main task following
the completion of a secondary task.
On the face of it, the branching hypothesis is readily
testable, although the coordination and management of
sub-goals is a requirement of many behavioural tasks
including planning, problem solving, decision making
and attentional set-shifting. In fact, almost all complex
cognitive activities can be said to involve some form of
branching, and even ostensibly simpler processes such
as episodic memory retrieval can, and have, been recast
in these terms76.In that sense, the functional specificity
of the model is unclear because, whereas the cognitive
processes in branching are clearly defined, they also
occur in a wide variety of different tasks. However, many
of these tasks do reliably and robustly activate aPFC. For
example, the Tower of London planning task, which
requires volunteers to integrate a number of sub-goals
(moving balls around on pegs) with a main goal (reach-
ing the problem solution), activates aPFC47,77 and the
task is sensitive to frontal-lobe damage46,48.
Exactly what information is processed and where it
comes from is not fully specified in the branching
hypothesis. Koechlin et al.74,75 focus on integration of
task-switching and working memory processes, but as
this is thought to occur when subjects are required to
hold an ongoing task temporarily to complete an inter-
mediate task, storage is also implied.Where the informa-
tion that is integrated comes from and how it is combined
and utilized is unclear.Braver and Bongiolatti76 are
similarly vague about the flow of information during
cognitive branching, suggesting that aPFC activation
“may reflect a specialized representational code that is
used to actively maintain information in tasks that
require sub-goal processing,” and/or might be “recruited
to provide storage of this information in a form that is
more protected from interference,” and/or “may be criti-
cally involved in the actual integration of the results of
sub-goal processing with the information that had been
actively maintained prior to the sub-goal task.”
Relatively few studies, and none outside functional
imaging, to our knowledge, have investigated branching
specifically.Even so, the results are mixed with respect to
anatomical specificity.For example, Koechlin et al.74
found that cognitive branching activated only the aPFC
© 2004Nature Publishing Group
190 |MARCH 2004 |VOLUME 5 www.nature.com/reviews/neuro
REVIEWS
models have attempted to accommodate identical data
sets into ostensibly unrelated theoretical frameworks.
For example, in their discussion of the role of the aPFC
in branching and sub-goal processing, Braver et al.
76
reconceptualize episodic retrieval in terms of monitor-
ing sub-goals in working memory. Conversely,Christoff
and Gabrieli
44
,in their discussion of the importance of
the aPFC in the processing of internal states, suggest that
many episodic retrieval tasks can be thought of as
requiring the evaluation of internally generated infor-
mation. Finally, in their discussion of the role of BA 10
in relational integration, Kroger et al.
82
suggest that
many episodic memory tasks involve “binding new
information with existing knowledge into a meaningful
and useful representation … potentially forming a com-
plex relational structure.” Given the prevailing emphasis
on ‘tasks’ in defining the functions of the aPFC, it is per-
haps unsurprising that there is considerable disparity
between the various models. On the other hand, if one
considers the type of information processing that aPFC
might perform, a common thread emerges. The account
that follows combines the core features of many of the
models described above, but focuses on the common
processes that are implied rather than the specific tasks
that are employed. In doing so, it accommodates rele-
vant findings from anatomical studies as well as much
of the existing functional neuroimaging literature. We
consider how it meets the three criteria defined earlier,
and discuss the hypotheses that this account generates
and how these hypotheses can be tested.
We propose that the aPFC is engaged when problems
involve more than one discrete cognitive process: that is,
when the application of one cognitive operation (such as
a rule) on its own is not sufficient to solve the problem
as a whole, and the integration of the results of two or
more separate cognitive operations is required to fulfill
the higher behavioural goal. Multiple,related cognitive
operations can only be performed successfully if they are
coordinated, and we speculate that the coordination
of information processing and information transfer
between multiple operations across supramodal cortex is
an important aspect of aPFC function. One new
and crucial aspect of this view is that it is consistent with
the cellular and connectional properties of neurons in the
primate aPFC. Earlier, we discussed evidence that the
aPFC is interconnected exclusively with supramodal cor-
tex. This connectional architecture indicates two general
levels of information processing. First, when information
is transmitted into supramodal cortex from areas outside
it, it is represented at a more abstract level.Second, these
representations form the inputs into the aPFC, where
they are processed further.We also note that the cellular
properties of neurons in the aPFC are better suited than
those of other areas to integrate their inputs32,and we
speculate that this might allow the aPFC to integrate
information from locations across supramodal cortex.
This account is also broadly consistent with much of
the functional neuroimaging literature. As we have
described, activity in aPFC is ubiquitous in many types
of functional imaging study, making it difficult to spec-
ify the processes that activate this area on a task-by-task
compare problems that required analysis of the relations
between multiple objects in two dimensions (for exam-
ple, two aspects of shape), relations in one dimension
(such as shape) or relations in zero dimensions (simple
matching-to-sample). Comparing the first two condi-
tions revealed frontopolar activity, which was taken to
indicate a role for this region in the consideration of
multidimensional relations83.In a similar experiment
using Raven’s progressive matrices, Kroger et al.82 sought
to decouple the effects of relational complexity from
those of other factors that increase task difficulty. They
found that reasoning problems that vary in the number
of explicit relations that jointly determine the solution
to a problem recruit a cortical network that includes the
aPFC. In addition, the change in activity in this region as
relational complexity increased was distinguishable
from activity changes resulting from increases in
(non-relational) task difficulty.
Relational complexity can be easily quantified, yielding
testable hypotheses about expected activity in imaging
studies and impairments in patients. Patients with frontal-
lobe damage are significantly impaired in their ability
to solve matrix problems that require the integration of
multiple dimensions87,88.However, because this view is
overly specific, it cannot account for many of the tasks
that activate aPFC but that do not involve relational
integration44,nor for patients’ deficits in such tasks. In
addition, what information is integrated, where it comes
from and what processes are carried out on it are not
well specified. Finally, with respect to anatomical speci-
ficity, both of the studies described above found that
increases in activity in the aPFC were accompanied by
similar changes in the mid-dorsolateral frontal
cortex.So, although the aPFC seems to be involved in
tests that tap relational complexity, it is neither uniquely
nor solely involved in such tasks.
Towards a common theoretical framework
In this review, we have considered a number of models
that have attempted to describe the functions of the
aPFC, largely on the basis of functional neuroimaging
studies. In the introduction, we specified three criteria
against which models of function should be judged.
These were that they should (1) generate testable
hypotheses, (2) specify where information comes from
and how it is processed, and (3) be supported by data in
a manner that is anatomically and functionally specific.
With respect to the latter two criteria,many of these
models fail to acknowledge a subtle but crucial distinc-
tion between ascribing a task-specific ‘function’ to an
area (for example, source memory) and describing the
information processing that is performed by that region.
In terms of the functional architecture of the brain, ‘task’
and ‘process’ are rarely synonymous (several tasks can
commonly recruit a given process for their successful
completion). Although it can be argued that a specific
kind of information processing is the function of an
area, models that are based on tasks or even classes of
task will inevitably have limited explanatory value out-
side the immediate cognitive domain of those tasks.
This problem is illustrated by considering how different
RAVEN’S PROGRESSIVE
MATRICES
A non-verbal test of inductive
reasoning in which participants
are required to discern the
relationship between complex
shapes, usually in more than one
dimension.
© 2004Nature Publishing Group
NATURE REVIEWS |NEUROSCIENCE VOLUME 5 |MARCH 2004 |191
REVIEWS
There are also some parallels with models of memory.
We suggest that source memory tasks activate aPFC
because they require the simultaneous resolution of
information about an item and about the context in
which it was encoded. Successful performance is not
possible with the resolution of only one or the other, as
they are both required. Prospective memory tasks might
activate this area for similar reasons: one rule (‘meet
John’) has to be applied only in the context of another
(‘when it is 5pm’), and the processing streams that
mediate these two aspects of behaviour undoubtedly
proceed independently, but have to be brought together
to fulfill the overall behavioural goal (remembering to
meet John at 5pm). Concepts such as ‘retrieval mode’
and ‘success monitoring’ are also easily recast in these
terms, requiring the integration of at least two distinct
operations (for example, retrieval itself and a compari-
son between the retrieved item and either a stimulus or
another internal representation).
Models that emphasize the importance of ‘branching’
and the reallocation of attentional resources also have
parallels with our account. Branching involves holding
primary goals in mind in working memory while simul-
taneously exploring secondary goals. Although it is not
explicitly stated in these models, branching occurs
because the solution to the primary problem can be
found only when the solution to the secondary one
is found first. The solution to the first depends on the
resolution of the second and is therefore entirely com-
patible with, but only a specific instance of, our account
of aPFC function. Similarly,we take the view that the
reallocation of attentional resources is a secondary effect
of branching rather than an independent explanation
for activity in the aPFC, because holding a primary
problem online while exploring a second inevitably
requires a shift of attention. In one recent study, sus-
tained aPFC activity was observed when volunteers were
required to switch between two tasks within a
single block of trials, but not during single-task blocks93.
The behavioural ‘mixing cost’ associated with the
former condition indicates that the mixed blocks
require the on-going consideration of the two tasks,
whereas the single-task blocks allow attention to be
allocated to one of the tasks rather than the other.Again,
this resonates well with our view that aPFC activity
occurs in situations that require the coordination of
multiple related cognitive operations (in this case, two
sub-tasks). Indeed, the authors conclude that their task
requires that “the stimulus–response mappings for two
different tasks have to be maintained simultaneously”,
that “attention toward the task cue must be maintained
… in order to be sensitive to trials in which the cue
indicates a task switch”, and that “the task-set mappings
have to be maintained in working memory, while
attention is directed toward completing the various
sub-goals”, all of which imply the coordination of
multiple cognitive operations.
‘Relational integration’is perhaps the most closely
related concept to our account,although again we believe
that it represents a specific example of a more general
requirement to combine and coordinate the outputs
basis. However, if we look for commonalities in the way
that information is processed across these studies, it
becomes clear that aPFC is almost always activated
when the solutions of two or more discrete cognitive
operations need to be integrated in the pursuit of a
more general behavioural goal.
Perhaps the clearest examples come from studies that
have investigated the cognitive control of action. Actions
are often executed under the guidance of either a sensory
cue or a rule (for example, a particular sequence of finger
movements, or a sequence of timings). Several studies
have manipulated both these factors so that timings and
finger sequences are either under the guidance of a
learned rule or under the guidance of a sensory cue89,90.
In those experiments, the aPFC was activated when
rules about timing and finger sequences were applied
simultaneously to the same actions (so, the sensory cues
are present, but not required).When one or the other (or
neither) was rule-based, the aPFC was not significantly
active. In fact,Ramnani and Passingham89 showed that
when one rule was overlearned (sequence) and the other
was in the process of being learned (timing), there was a
parametric increase in activity in this area. This indicates
that the aPFC was engaged in the simultaneous repre-
sentation of multiple rules to solve a motor control
problem. In this example,the rules that are integrated to
fulfill the higher goal are both from the motor domain.
However, there are also examples from different
domains. Rogers et al.91 reported aPFC activation when
a decision had to be made by resolving a choice between
two independent probability judgements. Our account
suggests that the aPFC should also be active when the
rules that comprise a complex behavioural goal are
themselves from different domains. Ramnani and
Miall92 found that the aPFC was specifically activated
when subjects were required to apply two unrelated
rules simultaneously: which action to perform on the
basis of stimulus shape, and whether to expect a reward
for correct performance on the basis of stimulus colour.
In this case, the combination of the outcomes of a cogni-
tive operation relating a stimulus attribute to an action
and an unrelated operation relating an independent
stimulus attribute to reward predicts activity in aPFC.
Despite the lack of coherence between other models of
aPFC function, there are some parallels between
specific models and our proposal. For example,Christoff
and Gabrieli44 proposed that the aPFC is specialized for
the explicit processing of ‘internal’ information (includ-
ing our own thoughts and feelings). By ‘internal’ they
mean representations that are stimulus independent, and
in this sense their account resonates with our view that
the aPFC processes abstract information without refer-
ence to lower-order information (such as sensory input).
However, unlike Christoff and Gabrieli44,we would not
predict that the consideration of internally generated
information would always activate aPFC,particularly if a
process or set of processes did not involve the simultane-
ous consideration of the outcomes of several independent
cognitive operations. In fact, there are clear examples
in the literature where the consideration of internally
generated information does not activate aPFC89,90.
© 2004Nature Publishing Group
192 |MARCH 2004 |VOLUME 5 www.nature.com/reviews/neuro
REVIEWS
This account also seems to accommodate some of the
core features of most of the models described above that
otherwise present rather disparate views. One of our cri-
teria was that models should specify the nature of the
information processing that takes place in a given region.
We have suggested that the aPFC receives its inputs from
other areas of supramodal cortex where information
from lower-order areas is abstracted. This abstract infor-
mation forms the input into the aPFC. Another criterion
is related to anatomical and functional specificity. We
have made a clear case in favour of the view that the aPFC
processes information by integrating the outcomes of
two or more separate cognitive operations in the pursuit
of a higher behavioural goal. But our primary criterion
was that any reasonable model of BA 10 function should
generate testable hypotheses. Our account does not make
task-specific predictions, but instead makes predictions
related to how the aPFC might process information
across various tasks. Specifically, it suggests that activity in
the aPFC will be evoked when the solution to an overall
problem can be arrived at only by the simultaneous
consideration of multiple sub-problems. In essence, this
predicts that specific computational resources in the
aPFC will be devoted to the integration of sub-problem
solutions, over and above the processing that is required
for the individual solutions themselves. This account can
from multiple cognitive operations. In particular,
relational integration requires the simultaneous con-
sideration of multiple relations between objects or
thoughts83,Raven’s matrices being the typical example.
Such tasks can be solved only by considering multiple
dimensions of a problem and abstracting a solution
from their simultaneous consideration. As the literature
review above demonstrates, there are many examples
where aPFC has been activated but no ‘relational integra-
tion’is required.We would argue, however, that in almost
all cases, more than one cognitive operation has to be
completed and the results combined to solve the prob-
lem. Moreover, several current models,such as relational
integration and branching, do not accommodate one
another except at this more general level of description.
For example, branching does not involve the simultane-
ous consideration of relations, by definition, as one
operation is held in check while another is completed.
However, if one accepts the suggestion that aPFC is gen-
erally involved in integrating the outcomes of separate
cognitive operations, then both relational integration
and branching can easily be accommodated, as can the
results of many studies outside these two models.
The main advantage of our account is that the level
of explanation is neither specifically tied to a task or class
of tasks, nor too general to be empirically testable.
Box 2 | Interactions in functional neuroimaging experiments: a hypothesis
We ha ve argued that the anterior prefrontal cortex
(aPFC) is engaged when successful task completion
requires problem solving in two or more domains
and during successful coordination of two or more
cognitive operations. Crucially,it is the interaction
of these solutions that allows task completion. So, it
is not sufficient to show that activity in the aPFC
occurs in the general context of problem solving.
Nor would it be sufficient to demonstrate such
activity in the presence of two unrelated problems.
A stringent test of our hypothesis would be that we
expect activity in the aPFC to reflect an interaction
between two main effects of the same task. Although
this criterion might not by itself be sufficient for
testing our hypothesis, we consider it to be necessary
and more stringent than the ones previously
mentioned.
In a typical factorial design experiment, the absence and presence of two main effects are independently manipulated (see
table). Interactions reflect the information processing resources that are specifically devoted to the combining operations,
over and above the main effects themselves. So, when all the rules that are required to solve a problem are available to
subjects in a functional neuroimaging experiment, the ‘super-additive’ activity should exceed the sum of activities found
when subjects solve one problem at a time (see figure). This might, for example, be achieved by manipulating the
availability of information required to complete the task independently of sensory guidance, such that it can be completed
using only abstract information in one condition (A+ B+). Some of the activity in this condition will be due to the
additional processing demands of integrating the solutions to A and B, and this can be statistically partitioned by taking
into account the main effects. So, Interaction effect = ([A– B–] + [A+ B+]) – ([A– B+] + [A+ B–]) (see table)
The anatomical component of our account emphasizes the combined use of inputs from supramodal cortex, so our
proposal does not predict interaction effects in aPFC in all studies that test for interactions, but only those in which
cognitive operations are combined where abstract information is represented. Note,for example, that Ramnani et al.104
tested for activity related to the coordination of two over-practised and automatic motor tasks in a factorial design (finger
movement and arm movement), but no interaction effects were seen in the prefrontal cortex. The figure shows a schematic
diagram of expected changes in activity in voxels that do (fourth column) and do not (third column) demonstrate super-
additive activity (interactions) in factorial experimental designs in functional neuroimaging exper iments.
Main effect B Main effect B
Main effect A A– B– A– B+
Main effect A A+ B– A+ B+
Task A
Task B
Interaction
Activity
Execute
task A Execute
task B Execute task A
and task B together
‘Super-additive’
activity expected in
aPFC
Increased, but not
‘super-additive’
activity
© 2004Nature Publishing Group
NATURE REVIEWS |NEUROSCIENCE VOLUME 5 |MARCH 2004 |193
REVIEWS
suggest that a better understanding of the aPFC might be
gained by investigating information flow through the
networks of which it is a part, particularly in respect of
the information flow from extra-supramodal to intra-
supramodal areas during the processing of component
problems, and from supramodal areas to anterior
prefrontal areas when solutions from this first processing
level are integrated to find a solution to the overall
problem.
In summary,although recent studies have demon-
strated that the aPFC is activated during many ‘high-level’
cognitive tasks, an adequate functional explanation for
this ubiquitous involvement has remained elusive. We
suggest that the coordination of information processing
and information transfer between multiple cognitive
operations within supramodal cortex is an important
aspect of aPFC function. This explanation concurs with
much of the functional neuroimaging literature, accom-
modating many of the key features of existing models
into a common theoretical framework. Crucially, this
framework is also entirely consistent with the connec-
tional and cellular anatomy of the aPFC, which is the only
prefrontal region that is predominantly interconnected
with supramodal cortex.
be most easily realized in terms of the main effects and the
interaction term of a factorial experimental design, where
the solubility of each secondary problem can be indepen-
dently manipulated as a main effect. Interactions allow
us to examine how the activity related to one main effect
is modulated by the context of another where specific
information processing resources are dedicated to the
modulation, over and above the processing of the main
effects themselves. Experimental designs in functional
neuroimaging experiments can statistically partition
brain activity specific to the interactions between two or
more main effects from the activity related to the main
effects per se (BOX 2). In one of the conditions, where both
problems become solvable, some of the activation can be
ascribed to processing over and above the main effects.
Our account predicts that this portion, the interaction
effect, dedicated to the integrative processing of the two
main effects, will be localized to the aPFC. Our hypothesis
also specifically predicts, on the basis of the anatomical
connections of the aPFC, that the main effects that yield
this interaction should be tasks in which processing of
abstract information is demanded. Recent evidence
shows support for this view90,92,94 but further studies
are needed to test this possibility more robustly. We also
1. Brett, M., Johnsrude, I. S. & Owen, A. M. The problem of
functional localization in the human brain. Nature Rev.
Neurosci. 3, 243–249 (2002).
2. Goldman-Rakic, P. S. Working memory dysfunction in
schizophrenia. J. Neuropsychiatry Clin. Neurosci. 6,
348–357 (1994).
3. Jonides, J. et al. Spatial working memory in humans as
revealed by PET. Nature 363, 623–625 (1993).
4. Courtney, S. M., Petit, L., Haxby, J. V. & Ungerleider, L. G.
The role of prefrontal cortex in working memory: examining
the contents of consciousness. Philos. Trans. R. Soc. Lond. B
353, 1819–1828 (1998).
5. Owen, A. M. The functional organization of working
memory processes within human lateral frontal cortex: the
contribution of functional neuroimaging. Eur. J. Neurosci. 9,
1329–1339 (1997).
6. Petrides, M. Frontal lobes and behaviour. Curr. Opin.
Neurobiol. 4, 207–211 (1994).
7. Rowe, J. B., Toni, I., Josephs, O., Frackowiak, R. S. &
Passingham, R. E. The prefrontal cortex: response
selection or maintenance within working memory? Science
288, 1656–1660 (2000).
8. Bor, D., Duncan, J., Wiseman, R. J. & Owen, A. M.
Encoding strategies dissociate prefrontal activity
from working memory demand. Neuron 37, 361–367
(2003).
9. Fletcher, P. C., Shallice, T. & Dolan, R. J. The functional
roles of prefrontal cortex in episodic memory. I. Encoding.
Brain 121, 1239–1248 (1998).
10. Dobbins, I. G., Foley, H., Schacter, D. L. & Wagner, A. D.
Executive control during episodic retrieval: multiple
prefrontal processes subserve source memory. Neuron 35,
989–996 (2002).
11. Rugg, M. D. et al.Neural correlates of memory retrieval
during recognition memory and cued recall. Neuroimage 8,
262–273 (1998).
12. Courtney, S. M., Ungerleider, L. G., Keil, K. & Haxby, J. V.
Transient and sustained activity in a distributed neural
system for human working memory. Nature 386, 608–611
(1997).
13. Dove, A., Pollmann, S., Schubert, T., Wiggins, C. J. & von
Cramon, D. Y. Prefrontal cortex activation in task switching:
an event-related fMRI study. Brain Res. Cogn. Brain Res. 9,
103–109 (2000).
14. Cools, R., Clark, L., Owen, A. M. & Robbins, T. W. Defining
the neural mechanisms of probabilistic reversal learning
using event-related functional magnetic resonance imaging.
J. Neurosci. 22, 4563–4567 (2002).
15. Rushworth, M. F., Nixon, P. D., Eacott, M. J. & Passingham,
R. E. Ventral prefrontal cortex is not essential for working
memory. J. Neurosci.17, 4829–4838 (1997).
16. Henson, R. N., Shallice, T. & Dolan, R. J. Right prefrontal
cortex and episodic memory retrieval: a functional MRI test
of the monitoring hypothesis. Brain 122, 1367–1381 (1999).
17. Wagner, A. D.et al.Building memories: remembering and
forgetting of verbal experiences as predicted by brain
activity. Science281, 1188–1191 (1998).
18. Tataranni, P. A. et al.Neuroanatomical correlates of hunger
and satiation in humans using positron emission tomography.
Proc. Natl Acad. Sci. USA 96, 4569–4574 (1999).
19. Gottfried, J. A., O’Doherty, J. & Dolan, R. J. Appetitive and
aversive olfactory learning in humans studied using event-
related functional magnetic resonance imaging. J. Neurosci.
22, 10829–10837 (2002).
20. Rolls, E. T. The orbitofrontal cortex and reward. Cereb.
Cortex 10, 284–294 (2000).
21. Rolls, E. T., Critchley, H. D., Browning, A. & Hernadi, I. The
neurophysiology of taste and olfaction in primates, and
umami flavor. Ann. NY Acad. Sci. 855, 426–437 (1998).
22. Tremblay, L. & Schultz, W. Relative reward preference in
primate orbitofrontal cortex. Nature 398, 704–708 (1999).
23. Roberts, A. C. & Wallis, J. D. Inhibitory control and affective
processing in the prefrontal cortex: neuropsychological
studies in the common marmoset. Cereb. Cortex 10,
252–262 (2000).
24. Montague, P. R. & Berns, G. S. Neural economics and the
biological substrates of valuation. Neuron 36, 265–284
(2002).
25. Elliott, R., Friston, K. J. & Dolan, R. J. Dissociable neural
responses in human reward systems. J. Neurosci. 20,
6159–6165 (2000).
26. Dias, R., Robbins, T. W. & Roberts, A. C. Dissociation in
prefrontal cortex of affective and attentional shifts. Nature
380, 69–72 (1996).
27. Duncan, J. & Owen, A. M. Common regions of the human
frontal lobe recruited by diverse cognitive demands. Trends
Neurosci. 23, 475–483 (2000).
28. Ongur, D., Ferry, A. T. & Price, J. L. Architectonic subdivision
of the human orbital and medial prefrontal cortex. J. Comp.
Neurol. 460, 425–449 (2003).
This paper describes the cytoarchitecture of the
medial and orbital PFC. The results are particularly
striking because they indicate that BA 10 is larger
than all other prefrontal areas in the human brain.
29. Semendeferi, K., Armstrong, E., Schleicher, A., Zilles, K. &
van Hoesen, G. W. Prefrontal cortex in humans and apes: a
comparative study of area 10. Am. J. Phys. Anthropol. 114,
224–241 (2001).
This is the only anatomical study to comprehensively
examine the comparative cytoarchitecture of BA 10 in
the brains of several primate species, including
humans.
30. Brodmann, K. Vergleichende Lokalisationlehre der
Grosshirnrinde in ihren Prinzipien Dargestellt auf Grund des
Zellenbaues (Barth, Leipzig, Germany, 1909).
31. Petrides, M. & Pandya, D. N. Comparative
cytoarchitectonic analysis of the human and the macaque
ventrolateral prefrontal cortex and corticocortical
connection patterns in the monkey. Eur. J. Neurosci. 16,
291–310 (2002).
32. Jacobs, B. et al.Regional dendritic and spine variation in
human cerebral cortex: a quantitative golgi study. Cereb.
Cortex 11, 558–571 (2001).
33. Passingham, R. E. The Frontal Lobes and Voluntary Action
(Oxford Univ. Press, Oxford, 1995).
34. Petrides, M. & Pandya, D. N. Dorsolateral prefrontal cortex:
comparative cytoarchitectonic analysis in the human and
the macaque brain and corticocortical connection patterns.
Eur. J. Neurosci. 11, 1011–1036 (1999).
35. McGuire, P. K., Bates, J. F. & Goldman-Rakic, P. S.
Interhemispheric integration: I. Symmetry and convergence
of the corticocortical connections of the left and the right
principal sulcus (PS) and the left and the right supplementary
motor area (SMA) in the rhesus monkey. Cereb. Cortex1,
390–407 (1991).
36. Barbas, H. & Pandya, D. N. Architecture and intrinsic
connections of the prefrontal cortex in the rhesus monkey.
J. Comp. Neurol. 286, 353–375 (1989).
37. Bachevalier, J., Meunier, M., Lu, M. X. & Ungerleider, L. G.
Thalamic and temporal cortex input to medial prefrontal cortex
in rhesus monkeys. Exp. Brain Res. 115, 430–444 (1997).
38. Andersen, R. A., Asanuma, C. & Cowan, W. M. Callosal and
prefrontal associational projecting cell populations in area 7A
of the macaque monkey: a study using retrogradely
transported fluorescent dyes. J. Comp. Neurol. 232,
443–455 (1985).
39. Moran, M. A., Mufson, E. J. & Mesulam, M. M. Neural inputs
into the temporopolar cortex of the rhesus monkey. J. Comp.
Neurol. 256, 88–103 (1987).
40. Amaral, D. G. & Price, J. L. Amygdalo-cortical projections in
the monkey (Macaca fascicularis). J. Comp. Neurol. 230,
465–496 (1984).
41. Morecraft, R. J. & Van Hoesen, G. W. Frontal granular cortex
input to the cingulate (M3), supplementary (M2) and primary
(M1) motor cortices in the rhesus monkey. J. Comp. Neurol.
337, 669–689 (1993).
42. Arikuni, T., Sako, H. & Murata, A. Ipsilateral connections of
the anterior cingulate cortex with the frontal and medial
temporal cortices in the macaque monkey. Neurosci. Res.
21, 19–39 (1994).
43. Passingham, R. E., Stephan, K. E. & Kotter, R. The
anatomical basis of functional localization in the cortex.
Nature Rev. Neurosci. 3, 606–616 (2002).
© 2004Nature Publishing Group
194 |MARCH 2004 |VOLUME 5 www.nature.com/reviews/neuro
REVIEWS
44. Christoff, K. & Gabrieli, J. D. E. The frontopolar cortex and
human cognition: evidence for a rostrocaudal heirarchical
organisation within the human prefrontal cortex.
Psychobiology 28, 168–186 (2000).
In this comprehensive review of functional
neuroimaging studies of reasoning and episodic
memory, Christoff and Gabrieli set out their
arguments that BA 10 might be specialized for the
explicit processing of internal states.
45. Buckner, R. L., Raichle, M. E., Miezin, F. M. & Petersen, S. E.
Functional anatomic studies of memory retrieval for auditory
words and visual pictures. J. Neurosci. 16, 6219–6235 (1996).
46. Shallice, T. Specific impairments of planning. Philos. Trans. R
Soc. Lond. B 298, 199–209 (1982).
47. Baker, S. C. et al.Neural systems engaged by planning: a
PET study of the Tower of London task. Neuropsychologia
34, 515–526 (1996).
48. Owen, A. M., Downes, J. J., Sahakian, B. J., Polkey, C. E. &
Robbins, T. W. Planning and spatial working memory
following frontal lobe lesions in man. Neuropsychologia 28,
1021–1034 (1990).
49. Owen, A. M., Sahakian, B. J., Semple, J., Polkey, C. E. &
Robbins, T. W. Visuo-spatial short-term recognition memory
and learning after temporal lobe excisions, frontal lobe
excisions or amygdalo-hippocampectomy in man.
Neuropsychologia 33, 1–24 (1995).
50. Lee, A. C., Robbins, T. W. & Owen, A. M. Episodic memory
meets working memory in the frontal lobe: functional
neuroimaging studies of encoding and retrieval. Crit. Rev.
Neurobiol. 14, 165–197 (2000).
51. Tulving, E. Elements of Episodic Memory (Clarendon,
Oxford, 1983).
52. Cabeza, R. et al.Age-related differences in neural activity
during memory encoding and retrieval: a positron emission
tomography study. J. Neurosci. 17, 391–400 (1997).
53. Duzel, E. et al.Task-related and item-related brain
processes of memory retrieval. Proc. Natl Acad. Sci. USA
96, 1794–1799 (1999).
54. Kapur, S. et al.Functional role of the prefrontal cortex in
retrieval of memories: a PET study. Neuroreport 6,
1880–1884 (1995).
55. Nyberg, L. et al.Functional brain maps of retrieval mode and
recovery of episodic information. Neuroreport 7, 249–252
(1995).
56. Rugg, M. D., Schloerscheidt, A. M., Doyle, M. C., Cox, C. J.
& Patching, G. R. Event-related potentials and the
recollection of associative information. Brain Res. Cogn.
Brain Res. 4, 297–304 (1996).
57. Wagner, A. D., Desmond, J. E., Glover, G. H. & Gabrieli, J. D.
Prefrontal cortex and recognition memory. Functional-MRI
evidence for context-dependent retrieval processes. Brain
121, 1985–2002 (1998).
58. Velanova, K. et al.Functional-anatomic correlates of sustained
and transient processing components engaged during
controlled retrieval. J. Neurosci. 23, 8460–8470 (2003).
59. Lepage, M., Ghaffar, O., Nyberg, L. & Tulving, E. Prefrontal
cortex and episodic memory retrieval mode. Proc. Natl
Acad. Sci. USA 97, 506–511 (2000).
60. MacLeod, A. K., Buckner, R. L., Miezin, F. M., Petersen, S. E.
& Raichle, M. E. Right anterior prefrontal cortex activation
during semantic monitoring and working memory.
Neuroimage 7, 41–48 (1998).
61. Ranganath, C., Johnson, M. K. & D’Esposito, M. Left
anterior prefrontal activation increases with demands to
recall specific perceptual information. J. Neurosci. 20,
RC108 (2000).
62. Ranganath, C. & Paller, K. A. Neural correlates of memory
retrieval and evaluation. Brain Res. Cogn. Brain Res. 9,
209–222 (2000).
63. Raye, C. L., Johnson, M. K., Mitchell, K. J., Nolde, S. F. &
D’Esposito, M. fMRI investigations of left and right prefrontal
contribtions to episodic remembering. Psychobiology 28,
197–206 (2000).
64. Ranganath, C. & Paller, K. A. Frontal brain activity during
episodic and semantic retrieval: insights from event-related
potentials. J. Cogn. Neurosci. 11, 598–609 (1999).
65. Ranganath, C. & Paller, K. A. Frontal brain potentials during
recognition are modulated by requirements to retrieve
perceptual detail. Neuron 22, 605–613 (1999).
66. Nolde, S. F., Johnson, M. K. & D’Esposito, M. Left prefrontal
activation during episodic remembering: an event-related
fMRI study. Neuroreport 9, 3509–3514 (1998).
67. Nyberg, L. et al.General and specific brain regions involved
in encoding and retrieval of events: what, where, and when.
Proc. Natl Acad. Sci. USA 93, 11280–11285 (1996).
68. Janowsky, J. S., Shimamura, A. P. & Squire, L. R. Source
memory impairment in patients with frontal lobe lesions.
Neuropsychologia 27, 1043–1056 (1989).
69. Thaiss, L. & Petrides, M. Source versus content memory in
patients with a unilateral frontal cortex or a temporal lobe
excision. Brain 126, 1112–1126 (2003).
70. Burgess, P. W., Quayle, A. & Frith, C. D. Brain regions
involved in prospective memory as determined by positron
emission tomography. Neuropsychologia 39, 545–555
(2001).
71. Burgess, P. W., Veitch, E., de Lacy Costello, A. & Shallice, T.
The cognitive and neuroanatomical correlates of
multitasking. Neuropsychologia 38, 848–863 (2000).
72. Okuda, J. et al.Participation of the prefrontal cortices in
prospective memory: evidence from a PET study in humans.
Neurosci. Lett. 253, 127–130 (1998).
73. Coull, J. T., Frith, C. D., Frackowiak, R. S. & Grasby, P. M.
A fronto-parietal network for rapid visual information
processing: a PET study of sustained attention and working
memory. Neuropsychologia34, 1085–1095 (1996).
74. Koechlin, E., Basso, G., Pietrini, P., Panzer, S. & Grafman, J.
The role of the anterior prefrontal cortex in human cognition.
Nature 399, 148–151 (1999).
We have suggested that the aPFC is required for
integrating the outcomes of two or more separate
cognitive processes. It is necessary to test for super-
additive effects in functional neuroimaging studies to
demonstrate a relationship between such a process
and brain activity. Reference 74 elegantly
demonstrates a highly specific super-additive effect in
the aPFC when subjects held in mind goals while at
the same time they processed secondary goals.
75. Koechlin, E., Corrado, G., Pietrini, P. & Grafman, J.
Dissociating the role of the medial and lateral anterior
prefrontal cortex in human planning. Proc. Natl Acad. Sci.
USA 97, 7651–7656 (2000).
76. Braver, T. S. & Bongiolatti, S. R. The role of frontopolar
cortex in subgoal processing during working memory.
Neuroimage 15, 523–536 (2002).
77. Owen, A. M., Doyon, J., Petrides, M. & Evans, A. C. Planning
and spatial working memory: a positron emission tomography
study in humans. Eur. J. Neurosci. 8, 353–364 (1996).
78. Pollmann, S., Weidner, R., Muller, H. J. & von Cramon, D. Y.
A fronto-posterior network involved in visual dimension
changes. J. Cogn. Neurosci. 12, 480–494 (2000).
79. Pollmann, S. Switching between dimensions, locations, and
responses: the role of the left frontopolar cortex.
Neuroimage 14, S118–124 (2001).
80. Owen, A. M., Roberts, A. C., Polkey, C. E., Sahakian, B. J. &
Robbins, T. W. Extradimensional versus intradimensional set
shifting performance following frontal lobe excisions,
temporal lobe excisions or amygdalohippocampectomy in
man. Neuropsychologia 29, 993–1006 (1991).
81. Owen, A. M. et al.Contrasting mechanisms of impaired
attentional set-shifting in patients with frontal lobe damage
or Parkinson’s disease. Brain116, 1159–1175 (1993).
82. Kroger, J. K. et al.Recruitment of anterior dorsolateral
prefrontal cortex in human reasoning: a parametric study of
relational complexity. Cereb. Cortex 12, 477–485 (2002).
83. Christoff, K. et al.Rostrolateral prefrontal cortex involvement
in relational integration during reasoning. Neuroimage 14,
1136–1149 (2001).
84. Robin, N. & Holyoak, K. J. in The Cognitive Neurosciences
(ed. Gazzaniga, M. S.) 987–997 (MIT Press, Cambridge,
Massachusetts, 1995).
85. Raven, J. C. Standardization of progressive matrices. Br. J.
Med. Psychol. 19, 137–170 (1938, 1941).
86. Carpenter, P. A., Just, M. A. & Shell, P. What one intelligence
test measures: a theoretical account of the processing in the
Raven Progressive Matrices Test. Psychol. Rev. 97,
404–431 (1990).
87. Waltz, J. A. et al.A system for relational reasoning in human
prefrontal cortex. Psychol. Sci. 10, 119–125 (1999).
88. Duncan, J., Burgess, P. & Emslie, H. Fluid intelligence after
frontal lobe lesions. Neuropsychologia 33, 261–268 (1995).
89. Ramnani, N. & Passingham, R.-E. Changes in the human
brain during rhythm learning. J. Cogn. Neurosci. 13, 1–15
(2001).
90. Sakai, K., Ramnani, N. & Passingham, R. E. Learning of
sequences of finger movements and timing: frontal lobe and
action-oriented representation. J. Neurophysiol. 88,
2035–2046 (2002).
91. Rogers, R. D. et al.Choosing between small, likely rewards
and large, unlikely rewards activates inferior and orbital
prefrontal cortex. J. Neurosci. 19, 9029–9038 (1999).
92. Ramnani, N. & Miall, C. Instructed delay activity in the
human prefrontal cortex is modulated by monetary reward
expectation. Cereb. Cortex 13, 318–327 (2003).
93. Braver, T. S., Reynolds, J. R. & Donaldson, D. I. Neural
mechanisms of transient and sustained cognitive control
during task switching. Neuron 39, 713–726 (2003).
94. Ramnani, N. & Passingham, R. E. Changes in the human
brain during rhythm learning. J. Cogn. Neurosci. 13,
952–966 (2001).
95. Zeki, S. & Shipp, S. The functional logic of cortical
connections. Nature 335, 311–317 (1988).
96. Ungerleider, L. G. & Mishkin, M. in Analysis of Visual
Behavior (ed. Mansfield, R. J. W.) 549–586 (MIT Press,
Cambridge, Massachusetts, 1982).
97. Haxby, J. V. et al.Dissociation of object and spatial visual
processing pathways in human extrastriate cortex. Proc.
Natl Acad. Sci. USA 88, 1621–1625 (1991).
98. Ettlinger, G. ‘Object vision’ and ‘spatial vision’: the
neuropsychological evidence for the distinction. Cortex 26,
319–341 (1990).
99. Passingham, R. E. & Toni, I. Contrasting the dorsal and
ventral visual systems: guidance of movement versus
decision making. Neuroimage 14, S125–131 (2001).
100. Nobre, A. C., Coull, J. T., Frith, C. D. & Mesulam, M. M.
Orbitofrontal cortex is activated during breaches of expectation
in tasks of visual attention. Nature Neurosci. 2, 11–12 (1999).
101. Schultz, W., Tremblay, L. & Hollerman, J. R. Reward
processing in primate orbitofrontal cortex and basal ganglia.
Cereb. Cortex 10, 272–284 (2000).
102. Lu, T., Preston J. B. & Strick P. L. Interconnections between
the prefrontal cortex and the premotor areas in the frontal
lobe. J. Comp. Neurol. 341, 375–392 (1994).
103. Hadland, K. A., Rushworth, M. F., Passingham, R. E.,
Jahanshahi, M. & Rothwell, J. C. Interference with
performance of a response selection task that has no
working memory component: an rTMS comparison of the
dorsolateral prefrontal and medial frontal cortex. J. Cogn.
Neurosci. 13, 1097–1108 (2001).
104. Ramnani, N., Toni, I., Passingham, R. E. & Haggard, P. The
cerebellum and parietal cortex play a specific role in
coordination: a PET study. Neuroimage14, 899–911 (2001).
105. Sarkissov, S. A., Filimonoff, I. N., Kononowa, E. P.,
Preopraschenskaja, I. S. & Kukuew, L. A. Atlas of the
Cytoarchitectonics of the Human Cerebral Cortex (Medgiz,
Moscow, 1955).
106. Petrides, M. & Pandya, D. N. in Handbook of
Neuropsychology, Vol. 9 (ed. Grafman, J.) 17–58 (Elsevier,
Amsterdam, 1994).
Acknowledgements
N.R. was funded by a grant from the Medical Research Council to
P. M. Matthews (FMRIB Centre, Oxford). We thank K. Christoff for
helpful discussion during the preparatory stages of this manuscript.
Competing interests statement
The authors declare that they have no competing financial interests.
Online links
FURTHER INFORMATION
Adrian M. Owen’s homepage: http://www.mrc-
cbu.cam.ac.uk/Common/People/People-
pages/Adrian.Owen.shtml
Narender Ramnani’s homepage:
http://www.fmrib.ox.ac.uk/~nramnani/
Access to this interactive links box is free online.
© 2004Nature Publishing Group
