Proc. Nati. Acad. Sci. USA
Vol. 91, pp. 8690-8694, August 1994
Functional magnetic resonance imaging of human prefrontal cortex
activation during a spatial working memory task
(echo-planar hnagng/mapetic susceptibility)
GREGORY MCCARTHY*t, ANDREW M. BLAMIREt, AINA PUCE*t, ANNA C. NOBRE*t, GILLES BLOCHU§,
FAHMEED HYDER¶, PATRICIA GOLDMAN-RAKICII, AND ROBERT G. SHULMANt
*Neuropsychology Laboratory 116B1, Veterans Affairs Medical Center, West Haven, CT 06516; and Departments of tSurgery (Neurosurgery), tMolecular
Biophysics and Biochemistry, lChemistry, and 'INeurobiology, Yale University School of Medicine, New Haven, CT 06510
Contributed by Robert G. Shulman, April 25, 1994
ingwasused to detect activation in thehuman prefrontal cortex
induced by a spatial working memory task modeled on those
used to elucidate neuronal circuits in nonhuman primates.
Subjects were required tojudge whether the location occupied
by the current stimulus had been occupied previously over a
sequence of 14 or 15 stimuli presented in various locations.
Control tasks were similar in all essential respects, except that
the subject's task was to detect when one of the stimuli
presented was colored red (color detection) or when a dot
briefly appeared within the stimulus (dot detection). In all
tasks, two to three target events occurred randomly. The MR
signal increased in an area of the middle frontal gyrus corre-
sponding to Brodmann's area 46 in all eight subjects perform-
ing the spatial working memory task. Right hemisphere acti-
vation was greater and more consistent than left. The MR
signal change occurred within 6-9 sec of task onset and
declined within a similar period after task completion. An
increase in MR signal was also noted in the control tasks, but
the magnitude of change was less than that recorded in the
working memory task. These differences were replicated when
testing was repeated in five of the original subjects. The
localization of spatial working memory function in humans to
a circumscribed area of the middle frontal gyrus supports the
compartmentalization of working memory functions in the
humanprefrontal cortex and the localization ofspatial memory
processes to comparable areas in humans and nonhuman
Hih-speed magnetic resonance (MR) imag-
Working memory (1) provides temporary storage of infor-
mation for cognitive capacities including comprehension and
reasoning. The neural substrates for working memory have
been elucidated in studies ofnonhuman primates (2). Spatial
working memory tasks requiring the brief storage of spatial
information throughout the visual field engage neurons in the
principal sulcus in the mid-dorsolateral cortex of rhesus
monkeys (3, 4). Individual neurons in this region increase
their discharge for particular target locations as monkeys
recall the position ofa previous stimulus (memory fields) (5).
The same region is innervated by cortico-cortical projections
from the posterior parietal cortex (6), where spatial percep-
tions are consolidated (7). These studies suggest that areas of
the dorsolateral prefrontal cortex, by virtue oftheir extrinsic
connections and physiological properties, are specialized for
processing of spatial information in working memory.
Positron emission tomography (PET) has recently been
used to study working memory in humans. Petrides et al. (8)
found that object working memory using pictures activated a
region of the right middle frontal gyrus, including Brod-
mann's area 9 and 46 and that these regions were activated
bilaterally by a verbal working memory task (9). On the other
hand, Paulescu et al. (10) found more posterior frontal
activation in Broca's area (area 44) and parietal cortex (area
40) in a verbal working memory task. Jonides et al. (11)
reported activation in inferior frontal cortex (Brodmann's
area47) and posterior parietal cortex during a spatial working
memory task. The reasons for the differences among human
studies and the lack of correspondence of some studies with
nonhuman primate research are not wellunderstood but may,
in part, be due to the influence of verbal strategies.
The present study reinvestigates the role ofthe dorsolateral
prefrontal cortex in spatial working memory using the
method of functional magnetic resonance (MR) imaging
(fMRI). The fMRI method has been shown by us (12) and
others (13) to be a sensitive indicatorofneuralfunction, while
providing good anatomical and temporal resolution. We have
used a design that taxes spatial working memory capacity
while discouraging verbal strategies. Preliminary reports of
these data have been presented in abstract form (14, 15).
MATERIALS AND METHODS
Subjects. Eight subjects (four females) participated in the
main experiment. Five subjects were retested to obtain data
on reliability. Each session lasted 2.5-3 hr. The experimental
protocol was approved by the Yale Human Investigations
Committee, and all subjects provided informed consent.
Working Memory Task. Three behavioral tasks employed
identical visual stimuli presented against a dark background
with a central fixation cross. In the spatial working memory
task ("Location"), subjects were presented with white ir-
regular stimuli that appeared at different locations. Each run
consisted of 14-15 stimuli presented at a rate of one per 1.5
sec. The stimuli for each run were randomly chosen from 20
different designs and 20 spatial locations. The stimuli were
chosen not to suggest nameable objects and were white,
except for two to three per run that were colored red. The
pattern oflocations was irregular, so that subjects would not
attach verbal labels to spatial positions. The subject raised an
index finger whenever a shape appeared in a location previ-
ously occupied during that same run. There weretwo to three
such target events per run. Neither shape nor color was
Control Tasks. Similar stimuli, locations, and timing were
used in the control tasks. In the "Color" target detection
Abbreviations: EPI, echo-planar imaging; MR, magnetic resonance;
fMRI, functional MR imaging; PET, positron emission tomography;
ROI, region of interest; T1, longitudinal relaxation time; TR, repe-
tition time; TE, echo time.
§Present address: Service Hospitalier Frederic Joliot, Department
Pharmacologie et Physiologie, 4, Place Du Gal LeClerc, 91406
The publication costs of this article were defrayed in part by page charge
payment. This article must therefore beherebymarked "advertisement"
in accordance with 18 U.S.C. §1734 solely to indicate this fact.
Proc. Natl. Acad. Sci. USA 91 (1994)
task, subjects made a finger response whenever a red shape
appeared. Spatial location was irrelevant. In the "Dot" target
detection task, subjects responded whenever a small dot
briefly appeared within a shape. The dot's onset was random,
thus requiring subjects to attend to a shape during its entire
1500-ms duration. Both control tasks were sensory-guided,
but the Dot task required constant attention to the stimuli and
was more difficult than the Color control task. In an addi-
tional "Baseline" control task, subjects were instructed to
relax with eyes open during image acquisition.
The Location and control tasks were randomly grouped
and repeated three to six times within each session. Each task
occurred with equal frequency over the course ofthe imaging
session. Control tasks occurred in close temporal proximity
to the primary spatial working memory task. The Dot task
was added in the replication study.
The experimental tasks were controlled by a laptop com-
puter using the MEL software system (Psychology Software
Tools, Pittsburgh). Stimuli were displayed on a liquid crystal
display panel (Sharp Instruments) and projected onto a
screen at the end of the patient gurney. The subjects viewed
the screen with prism glasses. To insure stable performance,
subjects practiced the Location task during the 1- to 1.5-hr
Imaging Sequence. All images were obtained at 2.1 T on a
Bruker Biospec spectrometer using a linear bird-cage reso-
nator. Scout MR images were acquired in sagittal planes
about the midline using a four-plane multislice inversion
recovery sequence to give good gray-white matter contrast
[slice thickness = 5 mm, separation = 7 mm, repetition time
(TR) = 2100 ms, echo time (TE) = 17 ms, inversion time =
760 ins, matrix = 1282]. Scout images were acquired in the
coronal plane about a point 4 cm anterior to the anterior
commissure using the above sequence. This slice was se-
lected to include Brodmann's area 46 and 9 within the middle
frontal gyrus (16).
Localized shimming ofthe slice was done (17) to maximize
the magnetic field homogeneity and hence the functional
signal changes. To establish the location of major vessels, a
single-slice time-of-flight angiogram was acquired (slice
thickness = 5 mm, TR = 70 ms, TE = 17 ms, flip angle = 45°).
With these parameters, the angiogram depicted mainly ves-
sels with high flow rates.
Coronal functional MR images were acquired using the
echo-planar imaging (EPI) sequence (18). The first study on
each subject was performed with an asymmetric spin-echo
version of EPI which generated a spin-echo from the slice
followed by a gradient echo delay (TE) of 50 ms for evolution
of functional signal changes (19). Images in the second
session were acquired using a gradient-echo version of EPI
(TE = 50 ms and added cerebrospinal fluid suppression using
an adiabatic inversion pulse and inversion time of 2040 ms).
In both cases lipid signals from the scalp were suppressed.
Slice thickness was 10 mm, and nominal in-plane resolution
was 6 x 3 mm with a matrix of 642.
One run of each task consisted of a time series of 32
echo-planar images (TR = 3000 ms), which included three
intervals: pretask (images 1-12), task (images 13-22), and
posttask (images 23-32). Four additional scans were acquired
before the time series to achieve steady-state transverse
Artifact Detection. Before further analysis, the center of
mass of each image within each run was calculated. Devia-
tions in this measure and examinations of runs in a cine loop
were used to identify movement and other artifacts. With
these procedures, 15 of172 runs were eliminated from further
Data Analysis. A multistep analysis procedure was devel-
oped based upon our prior studies (12, 19). (i) A f-test image
comparing the pretask and task intervals (images 1-12 vs.
14-22) was calculated for each run. The task interval was
offset by 6 sec (i.e., two images) from the actual duration of
the task (images 12-20), as our prior studies had shown a
delay of %6 sec in the MR activation response. (ii) For each
subject, individual t-test images were combined into a mean
t-test image for each task. These mean f-test images retained
only t values above a specified criterion and were superim-
posed upon corresponding anatomical images. The t thresh-
old criteria varied between 3.0 and 4.0 across subjects due to
different signal/noise levels, but the same criterion was
always used for comparison oftasks within each subject. (iii)
An AND f-test image was computed for each subject in which
the f-test images for each task were combined such that only
voxels exceeding a specified t value (t = 1.5, P < 0.05) in all
runs were retained. (iv) Clusters of activated voxels were
identified in the thresholded mean f-test images for each
subject and task. These regions of interest (ROIs) were
interrogated for each image in each run, such that an average
time course of activation could be determined. The time
course ofactivation was scaled torepresent percentage signal
change (or AS) relative to the mean ofthe pretask images. (v)
A grand mean t-test image was calculated across all subjects.
All images were first aligned to a representative subject
whose anatomy approximated that ofthe corresponding level
in the Talairach atlas (16). A "blind" observer determined
scaling and translation factors separately for the x and y
dimensions to align each individual's anatomical image to the
standard. A grand-mean f-test image was then computed for
each task by averaging each individual's t-test images in this
common dimensional space. (vi) Anatomical ROIs were
computed in which the gray matter bounding the middle
frontal gyri (encompassing area 46) was outlined and used for
image interrogation in each run, as described in step iv above.
Fig. 1 represents data from subject AAH. In Fig. 1A, the
mean f-test image for Location (thresholded at t -
represents voxels in the echo-planar image in which signal
intensity increased significantly during performance of the
Location task relative to the pretask interval. The AND t
image (data not shown) showed the same ROI. The activation
occurred in the right middle frontal gyrus. No other area in
the slice was activated above pretask levels. Note that the
anatomical resolution of the activation effect is less precise
than that of the Ti-weighted image because the echo-planar
voxels are larger.
Fig. 1C presents the average time course of activation for
subjectAAH for the single activated cluster (Fig. 1A) forboth
Location and Color. Note that change in signal intensity
exceeds pretask noise levels within 3-6 sec after onset for the
Location task and reaches a peak activation increase of
=2.6% by 9 sec. The activation declines immediately after
the task cessation but requires =15 sec to reach pretask
levels. The time course for Color shows a similar profile, but
the activation level reached is approximately half that of
Location (P < 0.0001).
An anatomically based ROI was defined by outlining the
gray matter for the right middle frontal gyrus. This was done
to determine whether using the ROI based on the Location t
image was underestimating the strength of activation for
Color. The anatomical ROI and activation time courses are
shown in Fig. 1B and D, respectively. Similar time courses
were observed for both Location and Color. Greater activa-
tion was obtained in Location (P < 0.0001).
Fig. 2 presents data for subject APE's first session in which
the mean t-test image for Location (thresholded at t . 3.00)
shows two clusters ofactivated voxels, one cluster in each of
the right and left middle frontal gyri. Significant activation of
=5% occurred in Location. Smaller activation was also
Neurobiology: McCarthyet al.
8692 Neurobiology: McCarthy et al.
Arrow indicates focus ofactivation. (Note, in this and all subsequent images, the right side ofthe brain appears on the left side ofthe image.) The
color scale indicates increasing t valuesfrom violet to redbeginningat the specifiedthreshold. (B) AnatomicallybasedROIoutliningofgray matter
ofthe right middle frontal gyrus. (C) Average time course ofactivation for the ROI in theright middle frontal gyrus shown inA for Location and
Color tasks. In this and all subsequent plots, the x axis shows the image number in the time series, and the y axis depicts the percentage signal
change between taskand pretask period dividedbythe meanpretask signal (i.e., AS/S). The vertical lines atimages 12and20indicate thebeginning
and end of the task. (D) Average activation time course for the ROI displayed in B. Note the change in scale of the y axis.
SubjectAAH. (A) Mean t-testimage forLocationtasksuperimposeduponalongitudinal relaxationtime (T1)weightedanatomicalimage.
observed in Color (P < 0.016 for left and P < 0.011 for right
hemisphere). No activation was observed in Baseline. Ana-
Location task superimposed upon anatomical image.
Subject APE, first test session. Mean f-test image for
tomical ROIs were traced for each middle frontal gyrus. In
both gyri, significant overall activation was obtained, and
Location showed greater activation than Color (P < 0.003 for
rightandP < 0.001 forleft hemisphere). The remainderofthe
subjects in the first session showed consistent activation in
the middle frontal gyri for Location. Other regions also
showed activation but not consistently across subjects.
To represent consistent findings across subjects, intersub-
ject grand mean t-test images were calculated (see v in the
Data Analysis section of Materials and Methods). A small
cluster of activation is seen in the right middle frontal gyrus
for Location group data (Fig. 3A) similar to that seen in the
individual data of subjects AAH and APE above. A smaller
midline cluster ofactivation is also visible near the cingulate
gyrus. No additional clusters were obtained (t 2 1.5 thresh-
old); however, a small cluster could be seen in the left middle
frontal gyrus when the t threshold was lowered. Fig. 3B
presents the grand mean t-test image for Color. A small
cluster is visible in the right frontal region somewhat ventral
to that seen in Location. As the f-test images were computed
identically for each task and displayed by using the same
Proc. NatL Acad. Sci. USA 91(1994)
Proc. NatL Acad. Sci. USA 91 (1994)
eight subjects studied in the first test session showing activation in
the right middle frontal region and cingulate cortex. These group data
have been superimposed upon a single subject's anatomical T1-
weighted image. (B) Corresponding grandmean f-testimageforColor
task across eight subjects showing minimal activation in the right
middle frontal region.
(A) Grand mean f-test image for Location task across
threshold, the relative degree of activation for each task can
be directly compared in these figures. No activation was seen
in the Baseline condition (data not shown).
Five subjects participated in a replication study several
weeks later. A gradient echo EPI sequence was used with an
inversion recovery pulse to suppress the cerebrospinal fluid
signal. The Dot task was also introduced in this session. Fig.
4 presents the second-session data for subject APE (see Fig.
2). In this session, the contrast in theTi-weightedanatomical
image was different. The shimmed echo-planar image was
asymmetric with some signal loss from the left hemisphere.
The mean t-test image (t> 4.0) for Location is shown in Fig.
4, where a strong activation was found in the right middle
frontal gyrus. Location produced a 3.5% signal increase,
which was significantly greater than the activation produced
by either Dot (P < 0.0035) or Color (P < 0.001).
Grand mean t-test images were computed for the replica-
tion study. Fig. 5 A and B represents the results of Location
and Color, respectively. The same region in the right middle
Location task superimposed upon anatomical image.
Subject APE, second test session. Mean t-test image for
five subjects restudied in the second test session showing activation
in the right middle frontal region. These group data have been
superimposed on a single subject's anatomical Ti-weighted image.
(B) Corresponding grand mean t-test image for Color task across
subjects showing minimal activation in the right middle frontal
region. (C) Dot task grand mean t-test image showing coextensive
activation to Location task A.
(A) Grand mean f-test image for Location task across the
frontalgyrms was activated in the second session as in the first
for Location. No consistent activation was seen in the
cingulate region. Color showed only weak activation in afew
voxels slightly ventral to the Location cluster. The activation
for Dot (Fig. 5C) was overlapped with Location but had
smaller magnitude. Again, Baseline yielded no activated
Performance of a spatial working memory task (i.e., the
Location task) causedincreasedMR signal in the right middle
frontal gyrus. This area, including Brodmann's area 46, has
recently been remapped applying cytometric criteria and
shown to be located on the middle frontal gyrms in the region
activated in this study (P.G.R., unpublished work). This
regionwas activated within 6-9 sec oftaskonsetand declined
slowly during the posttask interval. Some activation was also
noted in the anterior cingulate and left middle frontal gyrus-
areas interconnected with area 46 (21).
Potential sources of artifact from head movement, large
blood vessels, and eye movements were investigated, and
Neurobiology: McCarthyet al.
Neurobiology: McCarthy et al.
runs contaminated by artifact were discarded. Cerebrospinal
fluid suppression was used in the replication study to elim-
inate the bright cerebrospinal fluid signal outlining sulci that
might result in edge artifacts with small movements. MR
angiograms were acquired because it has been demonstrated
that relatively large vessels on the brain surface can also give
rise to functional effects. Within the resolution of the MR
angiogram (1.5 mm2), no vessels were apparent in the prox-
imity ofthe observed ROIs. Proximity to thefrontal eye fields
(area 8) raised the concern that differential eye movements
among the tasks could result in differential MR signal
changes. A separate study was done with five subjects in
which a blinded observer measured eye movements from an
electrooculogram during three replications each ofLocation,
Color, and Dot tasks. No significant differences were ob-
tained in eye movements between tasks.
The activation observed for the Location task was greater
across subjects than that obtained in the same regions in the
two sensory-guided, target-detection tasks Color and Dot.
Nevertheless, both caused significant activation, which in
some subjects approached in magnitude that ofthe Location
task. The Dot task, in particular, activated a region that, to
the limits of our anatomical resolution, appeared to be
coextensive with that for Location. This result is not sur-
prising, given that sustained attention directed to peripheral
spatial locations was also an essential component ofthe Dot
task. Prefrontal neurons tuned to spatial stimuli are also
recorded in and around area 46 (22). It is also possible that
subjects were remembering the spatial locations of stimuli in
the Color and Dot tasks despite instructions. The unblocked
design may have encouraged this strategy. Additional studies
are needed to separate the psychological factors contributing
to the activation observed.
While activation was often bilateral (e.g., Fig. 2), the
largest and most consistent activation occurred in the right
hemisphere. We note, however, that inconsistent asymme-
tries in the signal strength of the shimmed images may have
biased results in some subjects. Nevertheless, strong right
lateralization in area 46 was also observed by Petrides et al.
(8) in an experiment that required pointing to designs ar-
ranged in spatial locations. This asymmetry presumably
reflects the right hemisphere's role in processing nonverbal,
spatial material. The activation that we observed in midline
areas (cingulate and superior frontal gyri) was also observed
by Petrides et al. (8).
Our results differ from the PET study ofJonides et al. (11),
who did not observe area 46 activation in a spatial working
memory task but rather found inferior frontal activation in
area 47. Petersen et al. (23) using PET, and McCarthy et al.
(12) using fMRI, have shown activation of area 47 in word
generation tasks. We did not observe activation of area 47,
although the inferior aspects of our shimmed echo-planar
images were affected by field inhomogeneities. We also did
not observe consistent activation of area 9 as did Petrides et
al. (8, 9), at least at the level of prefrontal cortex containing
There are limitations to the present study. While the single
slice chosen for study allowed examination ofareas 9, 46, 23,
and often 47, it did not permit a detailed description of the
anatomical extent of activation and did not include more
posterior frontal (area 44) (10) and parietal (area 40) regions
(10, 11), which have been reported to be activated in prior
PET studies. Nevertheless, our results are consistent with
priorPET studies (8, 9) and with a large corpus ofhuman and
monkey research (2, 3) in indicating the special role of
mid-dorsolateral prefrontal cortex and area 46 in working
memory tasks. We note that in a preliminary report (20),
fMRI activation has also been observed in this region during
performance of a nonspatial working memory task. Whether
these different tasks activate different regions of mid-
dorsolateral prefrontal cortex remains to be determined. The
fMRI technique has demonstrated sensitivity to the cognitive
challenges reflected in working memory performance and
should be of value in further elucidating the functional
network responsible for this basic psychological process.
We thank Dr. Anthony Adrignolo, Francis Favorini, and Marie
Luby for assistance in data analysis. This work was supported by the
Department of Veterans Affairs, National Institute ofMental Health
Grants MH-05286 and MH-44866, and National Institutes of Health
Baddeley, A. (1992) Science 255, 556-559.
Goldman-Rakic, P. S. (1987) in Handbook ofPhysiology: The
Nervous System, ed. Plum, F. (Am. Physiol. Soc., Bethesda,
MD), Vol. 5, pp. 373-417.
Fuster, J. (1973) J. Neurophysiol. 36, 61-78.
Funahashi, S., Bruce, C. J. & Goldman-Rakic, P. S. (1989) J.
Neurophysiol. 61, 1-19.
Wilson, F. A. W., O'Scalaidhe, S. P. &Golman-Rakic, P. S.
(1993) Science 260, 1955-1958.
Cavada, C. & Goldman-Rakic, P. S. (1989) J. Comp. Neurol.
Mountcastle, V. B., Motter, B. C., Steinmetz, M. A. & Duffy,
C. J. (1984) inDynamicAspects ofNeurological Function, eds.
Edelman, G. M., Gall, W. E. & Cowan, W. M. (Wiley, New
York), pp. 159-193.
Petrides, M., Alivisatos, B., Evans, A. C. & Meyer, E. (1993)
Proc. Nat!. Acad. Sci. USA 90, 873-877.
Petrides, M., Alivisatos, B., Meyer, E. & Evans, A. C. (1993)
Proc. Nat!. Acad. Sci. USA 90, 878-882.
Paulescu, E., Frith, C. D. & Frackowiak, R. S. J. (1993) Na-
ture (London) 362, 342-345.
Smith, E. E.,
Minoshima, S. & Mintun, M. A. (1993) Nature (London) 363,
McCarthy, G., Blamire, A. M., Rothman, D. L., Gruetter, R.
& Shulman, R. G. (1993) Proc. Nat!. Acad. Sci. USA 90,
Ogawa, S., Tank, D. W., Menon, R., Ellermann, J. M., Kim,
S., Merkle, H. & Ugurbil, K. (1992)Proc. Natl. Acad. Sci. USA
McCarthy, G., Blamire, A. M., Nobre, A. C., Puce, A., Hy-
der, F., Bloch, G., Phelps, E., Rothman, D. L., Goldman-
Rakic, P. & Shulman, R. G. (1993) Soc. Neurosci. Abstr. 19,
Blamire, A. M., McCarthy, G., Nobre, A. C., Puce, A., Hy-
der, F., Bloch, G., Phelps, E., Rothman, D. L., Goldman-
Rakic, P. & Shulman, R. G. (1993) Soc. Magn. Reson. Med.
Abstr. 12, 1413.
Talairach, J. & Tournoux, P. (1988) Co-Planar Stereotaxic
Atlas ofthe Human Brain (Georg Thieme, Stuttgart).
Gruetter, R. & Boesch, C. (1992) J. Magn. Reson. 96, 323-334.
Mansfield, P. (1977) J. Phys. C. 10, LSS-L58.
Blamire, A. M., Ogawa, S., Ugurbil, K., Rothman, D., Mc-
Carthy, G., Ellermann, J. M., Hyder, F., Rattner, Z. & Shul-
man, R. G. (1992)Proc. Nat!. Acad. Sci. USA89,11069-11073.
Cohen, J. D., Forman, S. D., Casey, B. J. & Noll, D. C. (1993)
Soc. Magn. Reson. Med. Abstr. 12, 1405.
Bates, J. F. & Goldman-Rakic, P. S. (1993) J. Comp. Neurol.
Funahashi, S., Bruce, C. J. & Goldman-Rakic, P. S. (1993)
Biomed. Res. 14, 85-88.
Petersen, S. E., Fox, P. T., Posner, M. I., Mintun, M. &
Raichle, M. E. (1988) Nature (London) 331, 585-589.
R. A., Awh,
Proc. NatL Acad. Sci. USA 91(1994)