The roles of prefrontal brain regions in components of working memory: effects of memory load and individual differences.

Abstract

Using an event-related functional MRI design, we explored the relative roles of dorsal and ventral prefrontal cortex (PFC) regions during specific components (Encoding, Delay, Response) of a working memory task under different memory-load conditions. In a group analysis, effects of increased memory load were observed only in dorsal PFC in the encoding period. Activity was lateralized to the right hemisphere in the high but not the low memory-load condition. Individual analyses revealed variability in activation patterns across subjects. Regression analyses indicated that one source of variability was subjects' memory retrieval rate. It was observed that dorsal PFC plays a differentially greater role in information retrieval for slower subjects, possibly because of inefficient retrieval processes or a reduced quality of mnemonic representations. This study supports the idea that dorsal and ventral PFC play different roles in component processes of working memory.

Full text

Page 1

Proc. Natl. Acad. Sci. USA
Vol. 96, pp. 6558–6563, May 1999
Psychology
The roles of prefrontal brain regions in components of working
memory: Effects of memory load and individual differences
BART RYPMA* AND MARK D’ESPOSITO
Department of Neurology, University of Pennsylvania, Philadelphia, PA 19104-4283
Communicated by Edward E. Smith, University of Michigan, Ann Arbor, MI, March 23, 1999 (received for review October 2, 1998)
ABSTRACT
we explored the relative roles of dorsal and ventral prefrontal
cortex (PFC) regions during specific components (Encoding,
Delay, Response) of a working memory task under different
memory-loadconditions.Inagroupanalysis,effectsofincreased
memory load were observed only in dorsal PFC in the encoding
period. Activity was lateralized to the right hemisphere in the
high but not the low memory-load condition. Individual analyses
revealed variability in activation patterns across subjects. Re-
gression analyses indicated that one source of variability was
subjects’ memory retrieval rate. It was observed that dorsal PFC
plays a differentially greater role in information retrieval for
slower subjects, possibly because of inefficient retrieval pro-
cesses or a reduced quality of mnemonic representations. This
study supports the idea that dorsal and ventral PFC play
different roles in component processes of working memory.
Using an event-related functional MRI design,
Working memory (WM), the cognitive system that allows indi-
vidualstomaintainandmanipulateinformationheldtemporarily
in mind, may be divided into separate processes such as those
required for the retention of information and those required for
allocating attention and coordinating information that is being
temporarily maintained (1, 2). Evidence has accumulated to
support the notion that PFC may be organized to support
different WM processes. Imaging studies have observed that
ventral PFC is engaged by simple rehearsal (or maintenance)
processes (3, 4). Studies that use more complex WM tasks have
shown PFC activations that often occur bilaterally and in regions
dorsal to those found in studies of simple retention (Brodmann’s
areas 9 and 46) (5). One theory has proposed that PFC is
functionally divided in a dorsal/ventral fashion according to the
type of cognitive operation that must be performed on informa-
tionheldinWM(6).Empiricalstudieshavesupportedthenotion
thatventralPFCmediatesmaintenanceprocesseswhereasdorsal
PFC is recruited when additional processing of information, such
as manipulation or monitoring, is necessary (7, 8).
Studies that use more complex tasks, such as n-back and
dual-tasks (e.g., refs. 9–11, 5), have not clearly distinguished the
roleofPFCinthesedifferentWMprocessesbecauseinformation
processing demands increase concurrently with maintenance
demands. In fact, one study that used a task with no overt
requirements to manipulate information held in WM suggested
a role for dorsal PFC in WM maintenance. Rypma et al. (12)
observed dorsal PFC activation in a WM task in which subjects
were required to maintain one, three, or six letters for 5 seconds.
When subjects were required to maintain three letters in WM,
relative to one letter, activation in frontal regions was limited to
left ventral PFC (Brodmann’s area 44). When subjects were
required to maintain six letters, relative to one letter, additional
activation of dorsal PFC was observed, similar to that observed
during tasks in which successful performance required the ma-
nipulation of information held in WM (e.g., refs. 7 and 8).
WMtasksliketheoneusedbyRypmaetal.(12)involveseveral
componentsforencoding,retention,andretrievalofinformation.
The blocked design used in the Rypma et al. study did not permit
observation of the temporal dynamics of brain activity associated
with these different components. Thus, the recruitment of dorsal
PFC they observed in the high memory-load condition (com-
pared with the low memory-load condition) may have occurred
during any or all of these task periods. The aim of the current
study was to explore the relative roles of dorsal and ventral PFC
regionsduringspecificcomponentsofaWMtaskunderdifferent
memory-load conditions. We accomplished this by using an
event-related functional MRI (fMRI) (13) design that permits
discrimination among functional changes associated with tempo-
rally separated trial components. Specifically, this method al-
lowed us to isolate the task periods (i.e., Encoding, Delay, or
Retrieval) in which dorsal PFC and ventral PFC exhibit load-
dependent effects, that is, where cortical activity is greater with
higher memory loads.
Another limitation of blocked neuroimaging studies is that
inferences are usually based mainly on aggregate data in task
performanceandfMRIactivation.Thisapproachhasbeenuseful
for understanding the role of various cortical regions in WM, but
does not consider the possibility that inferences may be drawn
about the neural correlates of individual differences in WM task
performance. The present study also examined possible relation-
ships between individual differences in performance and indi-
vidual differences in cortical activity.
METHODS
Subjects. Six right-handed subjects (age ? 21–30 yr; 3 men)
were recruited from the medical and undergraduate campuses of
theUniversityofPennsylvania.Subjectswereexcludediftheyhad
any medical, neurological, or psychiatric illness or if they were
taking any type of prescription medication. All subjects gave
informed consent.
MRI Technique. Imaging was carried out on a 1.5T SIGNA
scanner (GE Medical Systems) equipped with a fast gradient
system for echoplanar imaging. A standard radiofrequency head
coil was used with foam padding to comfortably restrict head
motion. High-resolution sagittal and axial T1-weighted images
were obtained from every subject. A gradient echo, echoplanar
sequence (repetition time ? 2,000 ms, echo time ? 50 ms) was
used to acquire data sensitive to the blood-oxygen-level-
dependent (BOLD) signal. Resolution was 3.75 mm ? 3.75 mm
in-planeand5mmbetweenplanes(21axialsliceswereacquired).
Twenty seconds of gradient and radiofrequency pulses preceded
the actual data acquisition to allow tissue to reach steady-state
magnetization.
The publication costs of this article were defrayed in part by page charge
payment. This article must therefore be hereby marked ‘‘advertisement’’ in
accordance with 18 U.S.C. §1734 solely to indicate this fact.
PNAS is available online at www.pnas.org.
Abbreviations: PFC, prefrontal cortex; WM, working memory; fMRI,
functional magnetic resonance imaging; ITI, intertrial interval; ROI,
region of interest; RT, reaction time; BOLD, blood-oxygen-level
dependent.
*To whom reprint requests should be addressed at: Department of
Neurology, 3 West Gates, Hospital of the University of Pennsylvania,
3400 Spruce Street, Philadelphia, PA 19104-4283. e-mail:
rypma@mail.med.upenn.edu.
6558

Page 2

Behavioral Task. To start each trial, two or six letters were
presented simultaneously in pseudo-random order for 4 seconds
followed by a 12-second unfilled delay. A probe letter then
appeared for 2 seconds during which the subject pressed a button
with their right thumb if the probe item was part of the memory
set or with their left thumb if the probe item was not part of the
memory set (Fig. 1). After these behavioral events there was a
16-second intertrial interval (ITI). The total time from trial onset
to trial offset was 34 seconds. This design allowed us to examine
neural activity associated with stimulus-encoding, 4 seconds and
8 seconds into the delay period, and response. Subjects viewed a
backlit projection screen from within the magnet bore through a
mirror mounted on the head coil. Stimulus presentation and
reaction time (RT) recording were handled by a Power Macin-
tosh computer.
All subjects completed 10 runs of 8 trials each. A total of 136
gradient-echo echoplanar images in time were obtained per slice
in each 272-second run. Thus, a total of 1,360 observations were
obtained for each voxel in the brain for each subject, giving us
considerable power to detect effects within subjects.
Data Analysis. Off-line data processing was performed on
Sun Microsystems (Mountain View, CA) Ultra workstations.
Afterimagereconstructionandbeforemotioncorrection,data
were sinc interpolated in time to correct for the fMRI acqui-
sition sequence because hemodynamic responses were to be
compared across slices that were obtained at different points
intheacquisitionsequence.Ifleftuncorrected,thiswouldhave
introduced considerable variability and bias (a phase advance)
into the hemodynamic responses. The data were motion-
correctedbyusingaslicewisemotion-compensationmethodto
remove spatially coherent signal changes with a partial corre-
lationmethod(13)andbyapplyingasix-parameter,rigid-body,
least squares realignment routine (14).
Thedetailsoftheevent-relatedfMRIanalysisusedinthisstudy
are presented elsewhere (13). Briefly, fMRI signal changes that
occurred during particular temporal periods of the behavioral
trials were modeled with covariates comprised of shifted, BOLD
impulse-response functions, the fMRI response resulting from a
brief pulse of neural activity (15). Changes in BOLD signal
associatedwiththeEncoding,Delay,andResponseperiodsofthe
behavioral task were tested with covariates that modeled the
expected BOLD signal response in the event of an increase in
neural activity (relative to the ITI) occurring in each of the task
periods (Fig. 2).
Because fMRI data are temporally autocorrelated under the
nullhypothesis(13),thedatawereanalyzedbyusingthemodified
general linear model for serially correlated error terms (16). A
time-domain representation of the expected 1/f power structure
(13)andafilterthatremovesfrequenciesabove0.244wereplaced
within the K matrix. This filter was also applied to the fMRI time
series to remove artifacts at the Nyquist frequency (0.25 Hz).
Low-frequency (sine and cosine) confounds up to 0.025 Hz and
trial-effect covariates were included in our model to account for
frequencycomponentsandmeansignalchange,respectively,that
were associated with each trial. To examine activity in specific
regionsofPFC,dorsalPFCregionsofinterest(ROIs)weredrawn
to include middle and superior frontal gyri (corresponding to
Brodmann’s areas 9 and 46) according to the Talairach and
Tournoux (17) atlas on standard T1 axial slices in the axial plane.
A similar procedure was used to draw ventral PFC ROIs to
includeinferiorfrontalgyruscorrespondingtoBrodmann’sareas
44,45,and47.TheseROIswerethennormalizedtoeachsubject’s
T1 axial images by using a 12-parameter affine transformation
(14) with a nonlinear deformation routine (18).
Relationships with each task period and the ITI were assessed
by contrasts (yielding t statistics with ?1,195 df) involving the
parameter estimates that corresponded to the covariates that
modeled each task period (see Fig. 1). Note that three covariates
modeled each 4 seconds of the Delay period. Given estimates of
the temporal smoothness of the hemodynamic response (13), the
covariate modeling the first 4 seconds of the Delay period would
be contaminated by hemodynamic activity from the Encoding
period. Thus, only the second 4-second interval (designated as
Delay period 1) and third 4-second interval of the Delay period
(designated as Delay period 2) are considered in the analyses.
Delay periods 1 and 2 were analyzed both separately and to-
gether. The mapwise-corrected false positive rate was controlled
at??0.05byBonferronicorrectionforthenumberofvoxelsper
map (?15,000 voxels; t ? 4.5) or per ROI (?400 voxels; t ? 3.7).
We assessed cortical activation in each subject’s dorsal and
ventralPFCintheEncoding,Delay,andResponseperiodsforthe
two memory-load conditions (i.e., two vs. six letters) by using two
FIG. 1.Trial sequence.
FIG. 2.
activity (Top) associated with both encoding and the response com-
ponents, with no increase above baseline during the delay. Such neural
activity change leads to a profile of fMRI signal change (Middle). The
model covariates (shifted impulse-response functions) scaled by their
resulting least-squares coefficients are shown at Bottom (gray lines,
delay covariates; black lines, encoding and response covariates). The
covariates modeling the delay make only a small contribution to
variance explanation in A. In B, there is some neural activity increase
relative to baseline during the delay. It can be seen that in B, the delay
covariates explain more variance in the fMRI signal than in A.
Individual analysis logic. A depicts a brief period of neural
Psychology: Rypma and D’Esposito Proc. Natl. Acad. Sci. USA 96 (1999)6559

Page 3

measures. The first was the spatial extent of activation as mea-
sured by the numbers of suprathreshold voxels that occurred
within each ROI during each task period in each memory-load
condition. The second measure was activation magnitude, as
measured by the “regional mean t,” i.e., the t value across voxels
(nonthresholded) for the spatially averaged time series, within
each ROI, during each task period, in each memory-load con-
dition. Laterality was measured by assessing differences in extent
and magnitude of activation between each hemispheric ROI
during each task period and memory-load condition. We tested
differentialhemisphericinvolvementindorsalandventralPFCto
investigate Rypma et al.’s (12) report of greater right- than
left-hemisphere dorsal involvement in high memory-load condi-
tions. The hypothesis of greater dorsal and ventral PFC involve-
ment with increased memory load across subjects was tested by
using Fisher’s sign tests.
Derivation of an Impulse-Response Function. Our rationale
for deriving an impulse-response function is explained elsewhere
(15). An impulse-response function was derived from primary
sensorimotor cortex in each subject in the following manner.
Before performing the WM task described above, each subject
performedasimpleRTtaskinwhichacentralwhitefixationcross
changed briefly (500 msec) to a circle every 16 seconds, cueing
subjects to make a bilateral button press. Twenty such events
occurred during the 320-second scan (160 images). All scanning
parameters were identical to those used for the WM experiment.
RESULTS
Behavioral Performance
Performance accuracy was not different between the two-letter
(90.2%, SD ? 5.0) and the six-letter conditions, (89.0%, SD ?
10.4), t(5) ? 1. Mean median RTs, shown in Fig. 3A, were
significantly faster in the two-item condition (837.2 ms, SD ?
222.3) as compared with the six-item condition (1061.7 ms, SD ?
383.0), t(5) ? 3.3, P ? 0.02. Examination of individual subjects’
RT slopes (the least-squares estimated coefficient of a linear
regression of RT on memory load interpolated between the two-
and six-letter memory-load conditions; see Fig. 3B) varied con-
siderably between subjects. The RT slopes ranged from 9.8 msec
(subject DK) to 111.0 msec (subject EH).
Imaging Data
Mapwiseanalysisofvoxelsthatweresuprathresholdforacontrast
of the coefficients for each task period, across memory-load
conditions (the Encoding period, each of the two Delay periods
separately and combined, and the Response period) vs. the ITI,
revealed activity across a number of brain regions. In all six
subjects, activity correlated with the Encoding period was ob-
served in bilateral dorsal and ventral PFC, bilateral premotor
cortex,anteriorandposteriorcingulate,bilateralmiddletemporal
gyrus, bilateral inferior parietal lobule, and left superior occipital
gyrus. In the Delay periods, activity was observed in bilateral
dorsal and ventral PFC, lateral premotor, and parietal cortex. In
theResponseperiod,activitywasobservedinbilateraldorsaland
ventral PFC, lateral premotor cortex, anterior and posterior
cingulate, bilateral middle temporal, and bilateral inferior pari-
etal lobules.
Comparison of Activation Between Memory-Load Condi-
tions During Task Periods in PFC ROIs. Dorsal PFC. In the
Encoding period, all six subjects showed a greater extent of
activation in the six-letter than two-letter memory-load con-
dition across the two hemispheres (Fisher’s sign test P ? 0.02;
Fig. 4). There was a trend toward a difference in signal
magnitude between memory-load conditions only in the right
hemisphere (five of the six subjects, P ? 0.09). All six subjects
showed a more right-lateralized extent of activation in the
six-letter condition compared with the ITI (P ? 0.02). No such
effect occurred in the two-letter memory-load condition.
In the Delay periods, activation was quite variable across
subjects. In the first and second Delay periods analyzed individ-
ually, no significant difference in the extent or magnitude of
activation was found between memory-load conditions. A similar
finding was observed when the two Delay periods were analyzed
together. Tests of laterality were also not significant.
In the Response period, there were trends toward greater
extent (P ? 0.09) and magnitude of activation (P ? 0.09) in the
two-letterthanthesix-letterconditionacrosshemispheres.There
was a trend toward more right-lateralized extent and magnitude
of activation in the six-letter condition compared with the ITI
(P ? 0.09). A similar effect occurred in the two-letter load
condition (P ? 0.09).
Ventral PFC. In the Encoding period, no significant difference
in the extent or magnitude of activation was found between
memory-load conditions. In the two-letter condition compared
with the ITI, there were trends toward greater activation extent
(P ? 0.09) and magnitude (P ? 0.09) in the left hemisphere. No
reliable differences between hemispheres were observed in the
six-letter condition.
In the Delay periods, activation was quite variable across
subjects. In the first and second Delay periods analyzed individ-
ually, no significant difference in the extent or magnitude of
FIG. 3.
load conditions. (B) Intercept-normalized RTs show individual differ-
ences in slopes.
(A) Behavioral results in two-letter and six-letter memory-
6560 Psychology: Rypma and D’Esposito Proc. Natl. Acad. Sci. USA 96 (1999)

Page 4

activation was found between memory-load conditions. A similar
finding was observed when the two Delay periods were analyzed
together. In the first Delay period, there was a trend toward
greater extent of left-hemisphere activation in the six-letter
conditioncomparedwiththeITI(P?0.09)whichwassignificant
in the second Delay period (P ? 0.02). No laterality effect was
found in signal magnitude. No reliable differences between
hemispheres were observed in the two-letter condition.
IntheResponseperiod,allsixsubjectsshowedagreaterextent
of activation in the two-letter than in the six-letter memory-load
condition across the two hemispheres (P ? 0.02). There was a
trend toward greater signal magnitude in the two-letter than in
the six-letter condition (P ? 0.09). There was no evidence of
lateralized activity during the Response period.
In the analyses of task components above, we observed con-
siderable intersubject variability in fMRI signal magnitude and
activation extent. Individual analyses of subjects’ activation pat-
terns and trial-averaged time series suggested a relationship
between the patterns of PFC activity observed during the differ-
ent task periods and task performance. Fig. 5 shows PFC regions
of cortical activity and the associated zero-anchored trial-
averagedtime-seriesofthesubjectwiththesmallestRTslopeand
high accuracy (subject DK) and the subject with the highest RT
slopeandlowaccuracy(subjectEH).EHshowedagreaterspatial
extent,greaterchangeinsignal,andslowerincreaseinsignalthan
DK across task components. We used regression analyses to test
the relationship between memory retrieval rate (independent of
memory-loadcondition)anddorsalandventralPFCactivationin
each task period across all subjects.
IndividualDifferencesinPFCActivity.Totesttherelationship
between PFC activation and task performance in the group, we
determined, for each subject, the extent of activation in each task
period,independentofmemory-load,bydeterminingthenumber
of voxels that were suprathreshold for a contrast involving the
sum of coefficients of the task period covariates in the two-letter
and the six-letter memory-load conditions in each PFC ROI.
Next, we performed a linear regression of subjects’ RT slope and
numberofsuprathresholdvoxelsateachtaskperiod.Thelogicof
these analyses was that the relationship between memory-
FIG. 4.
standard errors in two-letter and six-letter conditions; ? indicates
significant load-dependent effect (i.e., greater activity in the six-letter
than in the two-letter condition).
Regional mean number of suprathreshold voxels and
FIG. 5.
load conditions) for the subjects with highest RT rate and high accuracy
and lowest RT rate and low accuracy and associated time series in the
six-letter (solid lines) and two-letter (dashed lines) conditions.
Examples of activation in dorsal PFC regions (across memory
Psychology: Rypma and D’EspositoProc. Natl. Acad. Sci. USA 96 (1999)6561

Page 5

retrieval rate and regional cortical activity could best be charac-
terized by comparing RT slope to voxels showing a net positive
effect in each task period, that is, voxels that were responsive in
each task period across memory-load conditions.
Table 1 shows the nonnormalized regression coefficients that
characterizetherelationshipbetweenRTslopeandspatialextent
of activation (i.e., number of suprathreshold voxels) in each task
period. The most important finding was in the Response period,
where coefficients for the dorsal PFC ROI showed a positive
relationship between RT slope and spatial extent of activation.
This relationship was not observed in ventral PFC (for a scatter-
plot, see Fig. 6). In dorsal PFC, there was an increase of 1.34
activated voxels per millisecond of increase in RT slope that
accounted for 76% of the variance. Similar analyses involving the
intercept were nonsignificant. Because the small number of
subjectslimitedpowerinthisstudy,weperformednonparametric
tests (which do not require assumptions of underlying normal
distributions) on these data and obtained the same pattern of
significant and nonsignificant results. We also tested the relation-
ship between activation and performance without a statistical
threshold by measuring each subject’s regional mean t value in
dorsalandventralPFCROIsandthenperformedaregressionof
subjects’ RT slope and regional mean t at each task period. The
results of this analysis are similar to those we performed with
suprathreshold voxels (see Table 2). In the Response period, a
strong positive correlation was observed between RT slope and
the mean t value of the dorsal PFC ROI that accounted for 84%
of the variance. Again, this pattern of results also occurred with
nonparametric tests.
DISCUSSION
In this experiment, we identified brain activity that correlated
with each of the components of the task (Encoding, Delay, and
Response). In dorsal PFC, we observed load-dependent memory
effects only during the Encoding period, where greater right-
hemisphere activation than left-hemisphere activation was ob-
served in the high, but not the low, memory-load condition. In
ventral PFC we observed load-equivalent activation.
Rypma et al. (12) observed activation in ventral PFC with both
low (three-letter) and high (six-letter) memory loads and addi-
tional bilateral dorsal PFC only with high memory loads. They
argued that ventral PFC is involved in maintenance of a subca-
pacity WM load. When WM load exceeds short-term memory
capacity (19), dorsal PFC may be additionally recruited to me-
diatestrategicprocessesnecessaryformaintenanceofahighWM
load.Rypmaetal.’sblock-designstudycouldnotdeterminewhen
such processes may have been engaged. Our results suggest that
dorsal PFC may play its principal role during encoding of
information with high memory loads and not necessarily during
the maintenance of information.
We did not observe delay-period load-dependent effects in
PFC as other studies have (20). This variance in results may be
because of differences in the paradigms used in these studies and
the present study. For example, Cohen et al. (20) used an n-back
task in which higher memory-load requirements correlated with
requirements to hold information for longer periods of time and
withmoredistractingstimuli(i.e.,moreinterveningitems)during
theseperiods,whereasourtaskvariedmemoryload(i.e.,amount
of information) alone across a constant time interval. Another
possible explanation for our failure to detect delay-period load
effectsisalackofstatisticalsensitivityinthistaskperiod.Because
load-dependent effects with verbal information have also been
observed in left posterior parietal cortex (Brodmann’s area 40)
(20, 21) and deficits in span performance have been observed
following left parietal lesions (22), we performed analyses of our
data in this region similar to those we performed in PFC. Those
analyses indicated load-dependent effects in signal magnitude
during the delay in each subject, consistent with these prior
studies. This further result ameliorates concerns that the failure
todetectload-dependenteffectswasrelatedtoissuesofstatistical
power.
Our finding of greater dorsal PFC activation during high-load
information encoding may be interpreted as the result of differ-
encesinstimulusdisplaysbetweenthetwoconditions.Thisresult,
FIG. 6.
during the response period in dorsal PFC (upper line, squares; slope ?
1.34, r2? .76) and ventral PFC (lower line, circles; slope ? 0.36, r2?
.18) plotted against reaction time slopes.
A scatterplot of the numbers of suprathreshold voxels
Table 1.
slope and regional suprathreshold voxels across
memory-load conditions
Parameters and statistical tests from regression of RT
Task period
Encoding
Dorsal
Ventral
Delay 1
Dorsal
Ventral
Delay 2
Dorsal
Ventral
Response
Dorsal
Ventral
*Indicates statistical significance.
Regression parameter
Sloper2
tP
0.06
0.07
0.002
0.002
0.08
0.09
0.94
0.93
0.24
?0.41
0.33
0.17
1.40
?0.89
0.24
0.42
0.30
0.10
0.30
0.89
1.30
4.93
0.26
0.02*
1.34
0.36
0.76
0.18
3.53
0.93
0.02*
0.41
Table 2.
slope and regional mean t value across memory-load conditions
Parameters and statistical tests from regression of RT
Task period
Encoding
Dorsal
Ventral
Delay 1
Dorsal
Ventral
Delay 2
Dorsal
Ventral
Response
Dorsal
Ventral
*Indicates statistical significance.
Regression parameter
Sloper2
tP
?0.36
0.11
0.13
0.01
?0.77
0.23
0.48
0.83
?0.23
?0.33
0.05
0.11
?0.47
?0.71
0.66
0.52
0.64
0.50
0.42
0.25
1.68
1.15
0.16
0.31
0.91
0.06
0.84
0.004
4.50
0.13
0.01*
0.91
6562Psychology: Rypma and D’EspositoProc. Natl. Acad. Sci. USA 96 (1999)