Behavioral and neurophysiological correlates of episodic coding, proactive interference, and list length effects in a running span verbal working memory task.

Abstract

Updating refers to (1) discarding items from, (2) repositioning items in, and (3) adding items to a running working memory span. Our behavioral and fMRI experiments varied three factors: trial length, proactive interference (PI), and group integrity. Group integrity reflected whether the grouping of items at the encoding stage was violated at discarding. Behavioral results were consistent with the idea that updating processes have a relatively short refractory period and may not fatigue, and they revealed that episodic information about group context is encoded automatically in working memory stimulus representations. The fMRI results did not show evidence that updating requirements in a task recruit executive control processes other than those supporting performance on nonupdating trials. They did reveal an item-accumulation effect, in which signal increased monotonically with the number of items presented during the trial, despite the insensitivity of behavioral measures to this factor. Behavioral and fMRI correlates of PI extended previous results and rejected an alternative explanation of PI effects in working memory.

Full text

Page 1

Copyright 2001 Psychonomic Society, Inc.10
Cognitive, Affective, & Behavioral Neuroscience
2001, 1 (1), 10-21
The concept of “updating” in working memory re-
search refers to the operations needed to modify the con-
tents of working memory in response to the appearance of
newer, superceding information in the environment. In
the laboratory, updating has often been operationalized
with variants of the running memory span task, in which
participants are presented a serial presentation of items
and must steadily update the contents of working memory
in order to retain then most recently presented items (e.g.,
“remember the four most recent items you have seen”;
Kiss, Pisio, Francois, & Schopflocher, 1998; Morris &
Jones, 1990; Salmon et al., 1996). Such a task requires at
least three discrete processing steps: (1) discarding items
from, (2) repositioning items in, and (3) adding items to
working memory. (By “repositioning” we mean changing
the ordinal position of each item; for example, changing
the item previously in the 2nd position of a 4-item list
to the 1stposition, the item previously in the 3rd position
to the 2nd position, and so on.) These steps are also re-
quired in more complex working memory tasks, such as
the n-back task, the processing components of which are
considered by Jonides et al. (1997) and Postle, Stern,
Rosen, & Corkin. (2000). Discarding and repositioning
operations engage executive control processes not re-
quired within individual trials of simple tests of short-
term memory span or duration, such as digit span and de-
layed-recognition tasks,respectively (D’Esposito & Postle,
1999, 2000). Indeed, tasks requiring updating of items
held in working memory have often been used to assess
the contributions of executive functioning to working
memory (Kiss et al., 1998; Lehto, 1996; Morris & Jones,
1990; Salmon et al., 1996). This report presents the re-
sults of experiments investigating the behavioral and
neural correlates of manipulating factors related to these
three processes, as well as to proactive interference (PI),
in a running span task.
In Experiment1, we performed two behavioral studies
in which three factors were varied. The factor of trial length
controlled the number of items presented (and thus, the
This research was supported by the American Federation for Aging
Research and by NIH Grants NS01762 and AG13483. B.R.P. received
support from NIH Grant AG00255 to VirginiaM.-Y. Lee (University of
Pennsylvania). The authors thank Geoff Aguirre for consultation on
fMRI analyses, Hugh Garavan, Rik Henson, Dylan Jones, Ivan Kiss,
Sharon Thompson-Schill, and Eric Zarahn for helpful discussions of
this work, and three anonymous reviewers for their careful critiques.
Correspondence should be addressed to B. R. Postle, Department of Psy-
chology, University of Wisconsin, Madison, WI 53706-1696(e-mail:
postle@facstaff.wise.edu).
Behavioral and neurophysiological
correlates of episodic coding, proactive
interference, and list length effects in a running
span verbal working memory task
BRADLEY R. POSTLE
University of Wisconsin, Madison, Wisconsin
JEFFREY S. BERGER and JEREMY H. GOLDSTEIN
University of Pennsylvania Medical Center, Philadelphia, Pennsylvania
and
CLAYTON E. CURTIS and MARK D’ESPOSITO
University of California, Berkeley, California
Updating refers to (1)discarding items from, (2)repositioning items in, and (3)adding items to a run-
ning working memory span. Our behavioral and fMRI experiments varied three factors: trial length,
proactive interference (PI), and group integrity. Group integrity reflected whether the grouping of items
at the encoding stage was violated at discarding. Behavioral results were consistent with the idea that
updating processes have a relatively short refractory period and may not fatigue, and they revealed that
episodic information about group context is encoded automatically in working memory stimulus rep-
resentations. The fMRI results did not show evidence that updating requirements in a task recruit ex-
ecutive control processes other than those supporting performance on nonupdating trials. They did re-
veal an item-accumulation effect, in which signal increased monotonically with the number of items
presented during the trial, despite the insensitivity of behavioral measures to this factor. Behavioral and
fMRI correlates of PI extended previous results and rejected an alternative explanation of PI effects in
working memory.

Page 2

UPDATING IN WORKING MEMORY 11
amount of updating required) in each trial. The factor of
PI reflected whether or not an invalid probe stimulus (i.e.,
a probe that did not match any item in the final memory
set) matched an item that had been presented in that trial,
but prior to the final four items of the trial. The factor of
group integrity reflected the fact that our experimental
design varied the grouping of items by presenting mem-
oranda in groups of one, two, or three items per stimulus
presentation epoch. For example, the letter string bcdf
might be grouped as b and cdf or as bc and df, depend-
ing on how the items were presented. Results from a
pilot study suggested that breaking up such groups at the
discarding stage (Figure 1) had a deleterious effect on
performance. We believed that the results of this manip-
ulation might have bearing on our understanding of the
retention in working memory of contextual information
specific to the encoding episode. Guided by knowledge
obtained from these behavioral studies, in Experiment 2
we employed functional magnetic resonance imaging
(fMRI) to investigate the neural correlates of these ex-
perimental factors.
EXPERIMENT 1
Behavioral
Trial Length
The results of previous studies suggested that updating
has an all-or-none effect on running span performance.
That is, although immediate serial recall performance is
significantly lower on trials that require updating than on
trials that do not, performance is not further sensitive to
the number of updates required in a trial (Morris & Jones,
1990). Specifically, in Morris and Jones, performance on
6-item, 8-item, and 10-item trials was equivalent. From
these results, it was suggested that the control processes
governingupdating operations had a rapid recovery rate.
Seemingly at odds with this behavioral result was the re-
port that the magnitude of event-related potentials (ERPs)
attributable to updating-related processes increased
monotonically as a function of the number of updates re-
quired in a single trial (Kiss et al., 1998; behavioral data
from this experiment were not suitable for assessment of
possible performance correlates of this ERP effect [I. Kiss,
personal communication]). The principal methodologi-
cal difference between these two studies was that Morris
and Jones employed an immediate serial recall proce-
dure and Kiss et al. a yes/no recognition procedure. In
the present experiment, therefore, we sought to assess
the generality of the step function relating performance
to updating demands (at least across a small range of
trial lengths) on a task that employed a recognition pro-
cedure. We varied trial length (4, 8, or 12 items) in a run-
ning span task that required memory for the 4 most re-
cently presented items.We considered only the latter two
to require updating, because no repositioning or discard-
ing operations were required on 4-item trials. Failure to
replicate the all-or-none effect of updating on behavioral
performance would call into question a fundamental as-
sumption about the refractory rate of updating-related
processes. A replication of this effect, in contrast, would
highlight the divergence between behavioral and ERP
correlates of updating.
PI
Previous studies of delayed item recognition have
characterized behavioral and neurophysiological corre-
lates of PI that manifests itself in “high-overlap” condi-
tions when an item from a previous trial matches the probe
on the subsequent trial (D’Esposito, Postle, Jonides, &
Smith, 1999; Jonides, Marshuetz, Smith, Reuter-Lorenz,
& Koeppe, 2000; Jonides, Smith, Marshuetz, Koeppe, &
Reuter-Lorenz, 1998). This effect is believed to index
the operation of a mechanism that resolves the ambigu-
ity about the validity of the probe on high-overlap trials
(Jonides et al., 2000). The behavioral studies presented
here, in contrast to the earlier delayed item recognition
studies, examined the effects of PI among items from the
same trial. Our design also permitted us to investigate in-
teractions between two putatively discrete (sets of) control
functions that can contribute to working memory per-
formance: updating and the resolution of PI.
Group Integrity
Presenting verbalizable memoranda in groups, as op-
posed to one at a time, has salutary effects on immediate
serial recall (Ryan, 1969). Such grouping effects are be-
lieved to be represented at the level of the phonological
loop (i.e., they are represented by verbal rehearsal;
Hitch, Burgess, Towse, & Culpin, 1996) and may be en-
coded by internal timing signals that represent temporal
order (Burgess & Hitch, 1999). Grouping effects in work-
ing memory, when induced externally, can be viewed as
Figure 1. Schematic illustration of four examples of trials in
the running span task. Each row represents a trial. Grouping to-
gether of lowercase letters represents groups of memoranda pre-
sented simultaneously in a stimulus-presentation epoch. The up-
percase letter bounded by asterisks represents the probe. (a) A
nonupdating trial. (b) A trial requiring updating in which group
integrity is preserved throughout the trial; the probe is valid.
(c) A trial requiring updating in which group integrity is violated
by two stimulus presentation epochs. This is a no-PI trial because
the invalid probe does not match any memoranda presented dur-
ing the trial. (d) A trial requiring updating in which the integrity
of one group (cg) is violated by a subsequent stimulus presenta-
tion epoch ( fzp) but in which group integrity is preserved in the
two groups making up the final memory set ( fzp and x). This is
a PI trial, because a memorandum appearing during the trial
matches the probe (the letter c) but is not a part of the final mem-
ory set.

Page 3

12 POSTLE, BERGER, GOLDSTEIN, CURTIS, AND D’ESPOSITO
an influence of episode-specific information about context
at encoding. We investigated the nature of this episodic
code in verbal working memory by exploring what hap-
pens when the integrity of a group defined by temporal
grouping of items at encoding is subsequently violated at
the discarding component of the updating task.
Method
Participants. We tested a total of 25 healthy undergraduate vol-
unteers in two experiments (Experiments 1A and 1B). Twelve, re-
cruited and tested at the University of Pennsylvania (5 males, 7 fe-
males; mean age = 21.0 years), participated in Experiment 1A, and
15 (4 males, 11 females; mean age = 19.0 years), recruited and
tested at the University of Wisconsin—Madison, participated in
Experiment 1B. Informed consent was obtained from each partici-
pant prior to testing.
Procedure. Stimuli were drawn randomly for each trial from the
set of 21 consonant letters. No items repeated as memoranda on a
trial. Each stimulus presentation event (hereafter referred to as an
epoch) presented one, two, or three lowercase letters for 500 msec.
Items in two- and three-letter groups were presented simultane-
ously, in a horizontal row. Group size within a trial varied pseudo-
randomly and was constrained by the group integrity factor (see
below). Epoch onset asynchrony (EOA) was 4 sec. Trials presented
a total of 4, 8, or 12 stimuli, followed by a probe stimulus. (Within
each of these levels of the factor of trial length—defined by num-
ber of items—were two temporal lengths. See Table 1.) The partic-
ipants were instructed to maintain a memory of the 4 most recently
presented items and to press the “yes” (right) or the “no” (left) but-
ton to indicate whether or not the probe matched an item in the
memory set (Figure 1). They were also instructed to respond as
quickly and accurately as possible, but not to sacrifice accuracy for
the sake of speed. Valid and invalid probes were equiprobable. Each
participant was tested on 32 trials of each length, which occurred in
a randomly determined order.
PI was manipulated on 8- and 12-item trials featuring invalid
probes (32 trials per participant) by the presence or absence of a
lure item that matched the probe but appeared among the first 4
items on 8-item trials or among the first 8 items on 12-item trials
(Figure 1). Item-specific PI across adjacent trials was not studied.
Group integrity was manipulated on all 8- and 12-item trials (i.e.,
on all trials that required updating). On one half of all such trials,
the final stimulus presentation epoch required the breakup of a
group (Figure 1). Such violations of group integrity could also
occur unpredictably in portions of the trial preceding the final stim-
ulus presentation epoch, so that these events would not be predic-
tive of the end of the trial.1
A comprehensive accounting of the exact numbers of each trial
type is provided in Table 1.
Results: Experiment 1A
The data revealed an effect of length consistent with
the hypothesized all-or-none effect of updating on perfor-
mance and revealed a robust PI effect. There was no evi-
dence of a reliable effect of group integrity, assessed across
invalid probes. The planned analysis of group integrity col-
lapsed across all 8- and 12-item trials was not performed,
due to an interaction of PI with group integrity in the re-
action time (RT) data, which signaled the possibility of a
speed–accuracy tradeoff in performance.
Trial length. An analysis of variance (ANOVA) as-
sessing accuracy data from the 12 participants yielded a
main effect of length (Length 4, M = 97.7%, SE = 0.6;
Length 8, M = 90.0%, SE = 1.8; Length 12, M = 88.5%,
SE = 2.2) [F(2,22)= 13.9, p , .0001]. The planned com-
parison confirmed that this effect was carried by the pre-
dicted all-or-none effect of updating on running span
performance: Performance at Length4 was significantly
better than performance at Lengths8 and 12 [t(11)= 4.3,
p , .001] (Figure 2). Although mean RT values in-
creased monotonically with trial length (Length 4, M =
1,538.5 msec, SE = 92.9; Length 8, M = 1,599.3 msec,
SE= 97.0; Length12, M= 1,642.3msec, SE= 161.5), this
effect was not reliable [F(2,22) = 0.76, n.s.] (Figure 2).
PI and group integrity. These data assessed accu-
racy and RT data from updating trials with invalid probes
[accuracy: PI/violated, M = 79.2%, SE = 5.8; no-PI/
violated, M = 96.9%, SE = 1.6; PI/preserved, M = 86.5%,
SE = 2.9; no-PI/preserved, M = 99.0, SE = 1.0; RT: PI/
violated, M = 1,898.0msec, SE = 128.1; no-PI/violated,
Table 1
Itemization by Type of the 96 Trials Administered to
Each Participant in Experiments 1A and 1B
Trial Length
84 12
Number of Stimulus
Presentation Epochs
2
Number of Stimulus
Presentation Epochs
4
Number of Stimulus
Presentation Epochs
6 Group Integrity357
Valid-Probe Trials
8
n.a.
Preserved
Violated
84
4
4
4
4
4
4
4 n.a.
Invalid-Probe Trials
Preserved
PI
No-PI
Violated
PI
No-PI
Note—Trial length was determined by the number of items presented as memoranda.
n.a. = not applicable.
n.a.
8
n.a.
8
2
2
2
2
2
2
2
2
n.a.
n.a.
n.a.
n.a.
2
2
2
2
2
2
2
2

Page 4

UPDATING IN WORKING MEMORY13
M = 1,454.0 msec, SE = 85.2; PI/preserved, M =
1,689.2msec, SE = 105.3; no-PI/preserved, M = 1,435.7,
SE = 99.0) (Figure 3). The two-factor ANOVA assessing
accuracy revealed a main effect of PI [F(1,11) = 19.18,
p ,.001], no effect of group integrity [F(1,11)= 1.9, p =
.19], and no interaction [F(1,11) = 0.53, n.s.] (Figure 3,
top).2The two-factor ANOVA assessing RT data from
these same trials in the same participants revealed a
main effect of PI [F(1,11)= 12.87, p ,.0005], no effect of
group integrity [F(1,11)= 2.9, p = .11], and an interaction
[F(1,11) = 7.12, p , .05] (Figure 3, bottom).
Because the two-factor ANOVAs described in the pre-
vious paragraph were restricted to updating trials with
invalid probes, we had planned to increase our power to
detect effects of group integrity by performing paired
t tests on data from all 8- and 12-item updating trials
(i.e., including trials with valid and invalid probes). The
appropriateness of collapsing across levels of PI was ques-
tionable, however, due to a possible speed–accuracy
tradeoff in performance: Although there was no PI 3
group integrity interaction in the accuracy data, these
two factors did interact in the RT data.
Thus, in order to investigate the group integrity effect
in an unequivocal way, we performed asecond experiment
(Experiment1B), in which no trials featured within-trial
PI. Experiment 1B was identical in design to Experi-
ment 1A in all other respects.
Results: Experiment 1B
The accuracy results replicated the trial-length effect
from Experiment 1A, whereas the effect of trial length
on RT was, unexpectedly, in the opposite direction from
Experiment1A. A reliable effect of group integrity in Ex-
periment 1B was evident in both the accuracy data and
the RT data.
Trial length. The accuracy results from these 15 par-
ticipants (Length 4, M = 92.1%, SE = 1.7; Length8, M =
85.4%, SE = 1.6; Length 12, M = 87.5%, SE = 2.7)
replicated those from Experiment1A, with a main effect
of length [F(2,28) = 4.83, p , .05] and performance at
Length 4 significantly better than performance at Lengths8
and 12 [t(14) = 3.2, p , .01] (Figure 2). The RT results
(Length 4, M = 1,342.1 msec, SE = 70.0; Length 8, M =
1,282.9msec, SE = 64.2; Length 12, M = 1,237.9msec,
Figure2. Trial-length data of Experiments 1A and 1B. Top: Accuracy data illustrating the
all-or-none effect of updating on performance (on this and on all subsequent graphs, bars
represent SEs). Bottom: RT data show a numerical, but nonsignificant, increasing trend in
Experiment 1A and show the opposite trend in Experiment 1B. This difference may be ex-
plained in part by the absence of within-trial PI in Experiment 1B.

Page 5

14 POSTLE, BERGER, GOLDSTEIN, CURTIS, AND D’ESPOSITO
SE = 41.7), in contrast, differed markedly from those of
Experiment1A, decreasing monotonically with trial length
[F(2,28) = 5.14, p = .01] (Figure 2).
Group integrity. The group integrity data revealed
significantly lower accuracy on trials violating group in-
tegrity (violated, M = 84.8%, SE = 2.3; preserved, M =
89.4%, SE= 2.0) [t(14)= 2.2, p ,.05] (Figure 4, top) and
longer RTs on these trials (violated, M = 1,263.2 msec,
SE = 49.7; preserved, M = 1,200.5 msec, SE = 51.8)
[t(14) = 3.0, p , .01] (Figure 4, bottom).
Discussion
Two experiments failed to demonstrate a monotonic
effect of list length on accuracy, a result that is consistent
with the suggestion that, across a limited number of trial
lengths, the effect on performance of updating is all or
none, not cumulative (Morris & Jones, 1990). These re-
sults indicate that this property generalizes across testing
procedures. This is consistent with the idea that updating-
related processes have a refractory period that is shorter
than 4,000 msec (Morris & Jones, 1990)3and that they
do not decline in efficiency with repeated use. Of course,
what appears to be a step function when plotted across
only three data points may be only a nonlinear beginning
segment of a function that may contain a monotonically
decreasing segment. RTs associated with another puta-
tive executive process—attention switching among rep-
resentations in working memory (Garavan, 1998)—in-
creased monotonically as list length increased from 2 to
24 items (H. Garavan, personal communication, July
2000). Thus, additional studies of the running span task
that feature longer trials will be required to more com-
pletely characterize the effect of trial length on updating-
related processes. Nonetheless, the fMRI study presented
as Experiment 2 was designed to assess whether the he-
modynamic correlate of list length would echo the step
function suggested by our accuracy data, whether it would
increase monotonically with list length (as does the ERP
correlate of list length; Kiss et al., 1998) or whether it
would demonstrate a novelrelationship with trial length.
Interpretation of the RT data is complicated by the re-
versal in sign, between Experiments 1A and 1B, of the
slope of the line characterizing RT as a function of list
length (Figure 2). This may be explained in part by the
absence of intratrial PI in Experiment 1B. Perhaps the
unpredictable presence of intratrial PI in Experiment1A,
which could occur only at longer length trials, induced a
strategy of greater caution on longer trials. It also bears
restating that, although the participants were instructed to
respond as quickly as possible, they were also instructed
not to sacrifice accuracy in doing so. Thus, the RT data
might provide a less reliable index of updating-related
processes than the accuracy data.
The robust effect of (intratrial) PI on accuracy and RTs,
mirroring those from previous studies of tasks that gen-
erated PI across trials (D’Esposito, Postle, Jonides, &
Smith, 1999; Jonides et al., 2000; Jonides et al., 1998), in-
dicatedthat we could reasonably predict an fMRI corre-
late of PI in the left inferior prefrontal cortex (PFC), mir-
roring neuroimaging results from those previous studies.
A novel finding from Experiment 1 was the group in-
tegrity effect. The sensitivity of updating performance to
violations of group integrity suggests that episodic in-
formation about group context at encoding, although ir-
relevant to task demands (and, indeed, deleterious to per-
formance), is encoded automatically as a feature of
stimulus representations in working memory. We will
consider further the implications of this result in the
General Discussion.
EXPERIMENT 2
fMRI
Our fMRI method featured an event-related technique
that permitted us to assess fMRI signal associated with
each stimulus presentation epoch, uncontaminated by vari-
ance in the signal arising from epochs preceding or follow-
ing the epoch in question (Postle, Zarahn, & D’Esposito,
2000; Zarahn, Aguirre, & D’Esposito, 1997b). We could
thus examine directly the evolution of the fMRI signal
across the seven or eight stimulus presentation epochs as-
sociatedwith a 12-item trial and the probe-evoked signal
from PI versus no-PI trials. In Experiment2, we addressed
several basic questions about updating that could not be
addressed with behavioral studies. Among these, we in-
vestigated whether the processes supporting updating
can be dissociated neurally from nonupdating processes
Figure 3. PI 3 group integrity data of Experiment 1A. Top:
Accuracy data from 8- and 12-item invalid-probe trials, illus-
trating a main effect of PI, no effect of group integrity, and no
interaction. Bottom: RT data from the same trials, illustrating
a main effect of PI, no effect of group integrity, and an inter-
action.

Page 6

UPDATING IN WORKING MEMORY15
(i.e., stimulus encoding and maintenance) also engaged
by our task. An affirmative result would provide evidence
consistent with models positing fundamental dissocia-
bility of updating-specific from nonupdating processes.
Neural dissociability is a sufficient, although not neces-
sary, feature of the relationship between two independent
mental processes.4Other examples of hypothesized ex-
tramnemonic executive control processes are those re-
quired to manipulate items in working memory. For ex-
ample, the processes recruited to alphabetize randomly
ordered letter stimuli in working memory are supported
by PFC voxels that do not support maintenance of those
same items (Postle, Berger, & D’Esposito, 1999). And if
updating and nonupdating processes are neurally dissocia-
ble, are updating-related processes supported by a rela-
tively circumscribed region of PFC, as are alphabetization-
related processes (D’Esposito, Postle, Ballard, & Lease,
1999; Postle et al., 1999)? The fMRI experiment was
also expected to provide additional information about two
of the factors that we investigated in depth in Experi-
ment 1: trial length and P1.
Trial Length
Experiment 1 demonstrated the generalizability across
testing procedures of the all-or-none effect of updating on
performance and highlighted the divergent results of stud-
ies of performance (Morris & Jones, 1990) and studies of
ERP measures of updating (Kiss et al., 1998), the latter
being positively correlated with the number of updates in
a trial. Would fMRI data, like ERPs, reveal a process that
corresponds to the accumulation of items withina trial and
that is not detectable by behavioral methods? To address
this question, we obtained the appropriate behavioral and
physiological measures in the same experimental session.
PI
Recently, it has been proposed that the neuroimaging
effects obtained in high-overlap conditions of the de-
layed item recognition task (D’Esposito, Postle, Jonides,
& Smith, 1999; Jonides et al., 2000; Jonides et al., 1998)
may not reflect the operation of a mechanism that re-
solves PI (Jonides et al., 2000) but rather the fact that
high-overlap conditions effectively increase memory
load (Bunge, Matsumoto, Desmond, Glover, & Gabrieli,
2000). That is, a span of four items maintained under
high-overlap conditions may be treated by the nervous
system as a span of five or more items because of intru-
sion into the memory set by proactively interfering items.
Although Experiment2 did not vary memory load, it did
vary trial length, permitting us to assess these two alter-
natives by examining whether the probe-specific fMRI
sensitivity to trial length differed qualitatively (as would
be predicted by Jonides et al., 2000) or quantitatively (as
would be predicted by Bunge et al., 2000) from probe-
specific fMRI sensitivity to trials featuring PI. Specifi-
cally, the memory-load hypothesis (Bunge et al., 2000)
would predict that the distribution and activity patterns
of voxels sensitive to trial length would resemble those of
voxels sensitive to high-overlap conditions (which we are
referring to as the PI condition). The resolution-of-PI hy-
pothesis (Jonides et al., 2000), in contrast, would predict
that PI effects would be focused within left Brodmann’s
area (BA) 45, whereas trial-length effects would be more
broadly distributed, particularly in dorsolateral PFC
(Rypma & D’Esposito, 1999).
Group Integrity
Although the results of Experiment 1A indicated that
a speed–accuracy tradeoff may have masked a possible
interaction of the factors of PI and group integrity, we
were interested in testing the two mutually exclusive ex-
planations of the PI effect (see previous paragraph).
Thus, we chose to keep the PI manipulation in our fMRI
study (Experiment 2) at the expense of generating inter-
pretable group integrity data. Thus, although the behav-
ioral procedure for Experiment 2 was closely modeled
after that used in Experiment 1A, we do not report the
fMRI correlates of group integrity.
Power Considerations
One way in which fMRI and behavioral studies differ
from one another is the markedly higher costs (financial,
time, human resources) of the former. This unavoidable in-
trusion of an extrascientific factor into the experimental
design process typically induces investigators to include
fewer participants in an fMRI study than in a behavioral
study. Our approach on this project was to first perform a
Figure 4. Group integrity data in Experiment 1B (an experi-
ment that featured no within-trial PI). Top: Accuracy data from
all 8- and 12-item trials, revealing significantly lower scores on
trials violating group integrity. Bottom: RT data from the same
trials, illustrating a significantly longer RTs on trials violating
group integrity.

Page 7

16 POSTLE, BERGER, GOLDSTEIN, CURTIS, AND D’ESPOSITO
series of behavioral studies whose statistical power would
permit adequate characterization of the theoretical phe-
nomena of interest, and then to investigate the neurophys-
iological correlates of these phenomena in a smaller group
of participants. This approach made the assumption that
the psychological processes engaged by the tasks featured
in the behavioral studies would be similarly engaged dur-
ing the fMRI study, because the latter featured the same
behavioral tasks as the former.5This assumption was crit-
ical, because the smaller N in the fMRI experiment was
expected to preclude detection, at a statistically reliable
level, of the behavioral effects corresponding to these psy-
chological processes. Also critical to this approach was the
fact that the fMRI data from each participant were ex-
pected to feature sufficient power to detect, within indi-
vidual participants, two of the neurophysiological effects
of interest (updating and list length). Previous event-
related fMRI studies in our laboratory that employed simi-
lar designs have featured upward of 1,100–1,200 effective
degrees of freedom (Postle, Zarahn, & D’Esposito, 2000).
We planned, therefore, to treat the fMRI results from each
participant as a case study. For data such as these, an N of
between 5 and 10 participants can suffice to assess the reli-
ability of an effect across participants. Our previous experi-
ence with investigations of PI, however, indicated that the
fMRI correlate of this effect was too subtle to investigate
with the case-study approach (D’Esposito, Postle, Jonides,
& Smith, 1999). Nonetheless, the N of 7 in that study suf-
ficed to demonstrate a reliable, topographically specific ef-
fect of PI in the fMRI data.
Method
Participants. Five healthy adults (3 males, 2 females; mean
age = 19.8 years) recruited from the undergraduate and medical
campuses of the University of Pennsylvania, none of whom partic-
ipated in Experiment 1, participated in Experiment 2. Additionally,
2 healthy adults recruited from the undergraduate campus of the
University of California–Berkeley (1 male and 1 female; 23 and 21
years of age, respectively) also participated. Informed consent was
obtained from each participant prior to testing.
Behavioral procedure. Stimulus presentation parameters and
instructions were the same as those in Experiment 1, with the ex-
ception that stimulus exposure duration was increased to 1 sec, to
ensure adequate encoding. Scanning was conducted during eight
behavioral blocks, each 6 min 48 sec long and containing 12 trials.
Trial type was balanced for length, PI status, and number of integrity-
violating versus integrity-preserving epochs within each block.
Trial order was pseudorandomized. Each trial was separated by an
intertrial interval of 12 sec (Berkeley) or 14 sec (Pennsylvania).
Data acquisition. The fMRI scanning was conducted at the Uni-
versity of Pennsylvania with a GE Signa 1.5 T scanner equipped with
a fast gradient system for echoplanar imaging. High-resolution sagit-
tal and axial T1-weighted images were obtained in every participant,
and a gradient echo, echoplanar sequence (TR = 2,000 msec, TE =
50 msec) was used to acquire data sensitive to the blood oxygen level
dependent (BOLD) signal. Whole-brain images were acquired in
twenty-one 5-mm-thick axial slices. The fMRI scanning was con-
ducted at the University of California–Berkeley with a Picker 1.5 T
scanner equipped with a fast gradient system for echoplanar imaging
and employing all the same image acquisition parameters with the
exception of slice number (20) and TE (40 msec). For all participants,
scans of the working memory task were preceded by a scan in which
we derived the hemodynamic response function (HRF) for each par-
ticipant (Aguirre, Zarahn, & D’Esposito, 1998). The HRF, which char-
acterizes the fMRI response resulting from a brief impulse of neural
activity (Boynton, Engel, Glover, & Heeger, 1996), was used to con-
volve independent variables entered into the modified general linear
model (GLM) (Worsley & Friston, 1995) that we used to analyze the
results of the scans of our working memory task.
Data processing. The fMRI time series data were filtered to re-
move high (.0.244 Hz) and low (,0.05 Hz) frequencies and were
adjusted to remove the effects of nuisance covariates (Friston,
Holmes, Poline, Heather, & Frackowiak, 1995). The principle of
the fMRI time series analysis was to model the fMRI signal
changes evoked by each stimulus presentation epoch with a covari-
ate shaped like the HRF (Postle, Zarahn, & D’Esposito, 2000;
Zarahn et al., 1997b). The smoothness of the fMRI response to
neural activity allows evoked responses that arise from temporally
dependent events to be resolved on the order of 4 sec (Zarahn et al.,
1997b). Thus, because different event types in our experiment oc-
curred in a temporally dependent manner, we selected the 4-sec
EOA so that we could estimate the fMRI response evoked by each
stimulus presentation epoch. This approach permitted us, for ex-
ample, to contrast updating- and nonupdating-related activity (see
section on “Updating versus nonupdating activity,” further along in
the Method section) or to contrast probe-related activity on trials
featuring PI and probe-related activity on trials that did not feature
PI. The least-squares solution of the GLM of the fMRI time series
data yielded parameter estimates that were associated with each co-
variate of interest. Differences in fMRI signal (either between con-
ditions or vs. baseline) were tested by computing t statistics result-
ing from linear combinations of the covariates in question.
Controlling Type I error. Our procedure for fMRI analysis is
a “massively univariate” one in which the GLM is solved, indepen-
dently and in parallel, in each voxel in the hypothesis-testing space.
To control the omnibus false-positive rate of each statistical test,
we applied Bonferroni correction, which has been demonstrated to
provide an almost exact correction (to p #.05) for small voxel-wise
false-positive rates in spatially unsmoothed data (Zarahn, Aguirre,
& D’Esposito, 1997a). With these methods, the sensitivity of any
hypothesis-testing contrast was inversely related to the number of
voxels in the hypothesis-testing space. Thus, we maximized the sen-
sitivity of each contrast by restricting each to regions of interest
(ROIs) that were selected a priori (Postle, Zarahn, & D’Esposito,
2000).
ROIs. Our analyses were performed with anatomically defined
ROIs. ROIs for dorsolateral and ventrolateral PFC were created by
first defining them on the “canonical” representation of a brain in
Talairach space that is provided in SPM96b, using the atlas of Ta-
lairach and Tournoux (1988) to confirm our identification of
anatomical landmarks. (The dorsolateral PFC ROI corresponded to
BAs 9 and 46 of the middle and inferior frontal gyri, and the ven-
trolateral PFC ROI to BAs 44, 45, and 47 of inferior frontal gyrus;
we also created a BA 45 ROI.) Next, we transformed these ROIs
from Talairach space into the native space in which each partici-
pant’s data had been acquired by applying the 12-parameter affine
transformation (Friston, Ashburner, et al., 1995) with nonlinear de-
formations (Ashburner & Friston, 1996), routine in SPM96b (ef-
fectively, a “reverse normalization”). Because some individual
anatomical variability is not accounted for in the reverse normal-
ization process, we adjusted the ROIs after transformation to better
correspond to the anatomical images of each participant so that they
would cover perfectly the intended brain regions. Our analyses fo-
cused on PFC because this region is widely accepted as a critical
neural substrate of extramnemonic executive control processes that
are recruited to support working memory performance. PI analyses
focused on BA 45 (D’Esposito, Postle, Jonides, & Smith, 1999).
All ROIs were also broken down by hemisphere for some analyses.
Updating versus nonupdating activity. We determined whether
the processes supporting updating can be dissociated neurally from
nonupdating processes with the contrast [updating 2nonupdating],

Page 8

UPDATING IN WORKING MEMORY17
where nonupdating epochs were defined as the initial epochs of the
trial that did not require a discarding operation. Any voxels demon-
strating significantly greater updating activity than nonupdating ac-
tivity could be classified, within the context of this experiment, as
specialized for updating. Once voxels had been so classified, we
planned to assess quantitatively the level of nonupdating activity in
updating-speciali zed voxels (with the contrast [nonupdating 2
baseline]), and vice versa. These two contrasts were not mathemat-
ically independent, because of the way in which the interrogated
voxels had been defined. We felt that they would nevertheless pro-
vide interpretable initial indices of the degree of topographical over-
lap of updating- and nonupdating-related activity.
A second type of quantitative analyses, normalized voxel counts,
were employed to assess the spatial extent of nonupdating- and
updating-specialized activity (e.g., Is more cortex in region A than
in region B recruited by updating-specialized operations?). In par-
ticular, these would permit us to assess the relative extent of par-
ticular types of activity in dorsolateral versus ventrolateral PFC, the
former region being associated with a privileged role in many non-
maintenance processes (D’Esposito, Postle, Ballard, & Lease,
1999; Petrides, 1989; Rowe, Toni, Josephs, Frackowiak, & Pass-
ingham, 2000). To perform these analyses, we first normalized the
number of updating-specialized voxels within each ROI of each
participant by dividing the number of suprathreshold voxels by the
number of voxels in the ROI. Then, we assessed mean differences
by ROI with paired t tests. Critical to the validity of this analysis
(which is treated in more detail in Postle & D’Esposito, 2000) was
our ability to define objectively the threshold for determining the
significance of updating-specific activity on a voxel-by-voxel basis
in our unsmoothed data.
The two principal factors of interest. Trial-length effects were
assessed by inspection of trial-averaged time series data from 12-item
6-epoch trials and from 8-item 4-epoch trials (these were the twotrial
types for which the epoch on which updating processes were first
engaged could be determined unambiguously). PI effects were as-
sessed by identifying voxels in BA 45 of the left inferior PFC that
demonstrated invalid probe-evoked activity, as assessed with the
contrast [probeinvalid , PI+ probeinvalid, no-PI], extracting the spatially
averaged time series from these voxels and interrogating this time
series with the orthogonal two-tailed contrast [probeinvalid, PI2
probeinvalid, no-PI]. The result of this contrast was a t value, a noise-
normalized index of the PI effect (Postle, Zarahn, & D’Esposito,
2000). A positive value would indicate greater invalid probe-evoked
activity on trials featuring PI, and a negative value would indicate
greater invalid probe-evoked activity on trials that did not feature
PI. In the event that we found reliably a PI effect in left BA 45 (i.e.,
across participants), the same procedure would be applied to other
PFC areas (right BA 45, and left and right dorsolateral PFC) to as-
sess the anatomical specificity of this effect (D’Esposito, Postle,
Jonides, & Smith, 1999).
Additionally, we planned to assess the memory-load alternative
explanation of the PI effect (Bunge et al., 2000) by also assessing
trial-length effects in each of these four ROIs. This would be im-
plemented by identifying probe-sensitive voxels with the contrast
[probeall2 baseline] and interrogating them with the two-tailed
contrast [probe12-item2 probe4-item]. Note that these analyses mea-
sured an effect that has been attributed to memory scanning
(Rypma & D’Esposito, 1999), an effect that is unrelated to the item-
accumulation effect despite the fact that both relate to the indepen-
dent variable trial length.
Results
Behavior. We did not perform inferential statistical
tests on these data because of the small N of this experi-
ment. Overall mean accuracy was 91.5% (SD = 5.8) cor-
rect. Accuracy on trials that did not require updating
(M = 95.6%, SD = 3.6) was higher than on trials requir-
ing updating (M = 89.4%, SD = 7.4). Mean accuracy by
trial length was 96.7%, 95.6%, and 91.9%, for 4-, 8-, and
12-item trials, respectively. Accuracy on invalid trials
that did not feature PI (M = 94.7%, SD = 9.1) was supe-
rior to that of invalid trials featuring PI (M = 88.0%, SD=
13.0). Mean RTs increased with trial length (4 items,
M= 1,286.9msec, SD= 139.4; 8 items, M = 1,338.3msec,
SD = 241.4; 12 items, M = 1,399.7 msec, SD = 214.5).
Mean RTs were shorter on invalid trials that did not fea-
ture PI (M = 1,420.2 msec, SD = 266.5) than on those
that did feature PI (M = 1,629.6msec, SD= 342.5), yield-
ing a group mean PI effect of 209.3 msec (SD = 118.2).
One participant (W) did not show an appreciable PI ef-
fect (1.7 msec).
fMRI. Voxels displaying sensitivity to nonupdating
portions of the task (i.e., encoding and maintenance), as
well as voxels demonstrating sensitivity to portions of the
task requiring updating, were identified in many brain re-
gions, including PFC, superior frontal areas 8 and 6, an-
terior cingulate cortex, posterior parietal cortex, inferior
temporal cortex, and occipital cortex, a result broadly con-
sistent with a previous PET study (Salmon et al., 1996).In
all these brain regions, updating-related activity was
more extensive than nonupdating-related activity.
Normalized voxel count analyses indicated that
nonupdating-sensitive voxels (identified with the con-
trast [nonupdating 2baseline]) were identified to a com-
parable extent in dorsolateral and ventrolateral PFC, a re-
sult that was consistent with several previous fMRI
studies of working memory for letters (Postle & D’Es-
posito,2000; Rypma & D’Esposito, 1999). The majority
of these voxels also demonstrated sensitivity to updating
portions of the trial, and only a very few voxels demon-
strated greater nonupdating-related activity than updating-
related activity. Updating-sensitive voxels (identified
with the contrast [updating 2 baseline]) were also identi-
fied in dorsolateral and ventrolateral PFC. A group mean
of 43.3% of all updating-sensitive voxels in PFC demon-
strated significantly greater updating activity than nonup-
datingactivity and thus could be classified as specialized
for updating-related processes (see Method). These vox-
els, too, were found in both dorsolateral and ventrolateral
PFC, and the normalized voxel count analysis revealed
no difference in the extent of updating-specialized voxels
in these two ROIs (dorsolateral PFC, M = 14.5%, SE =
2.2; ventrolateral PFC, M = 16.5%, SE = 3.8) [t(6) = 0.8;
n.s.]. Quantitative assessment of comparative intensity of
updating- versus nonupdating-related activity in PFC re-
vealed that nonupdating-sensitive voxels also demon-
strated robust updating-related activity (mean t = 24.5,
SE= 1.6), and updating-sensitive voxels also demonstrated
robust nonupdating-related activity (mean t = 10.2, SE =
1.6). This result confirmed our impression that the dif-
ference between updating- and nonupdating-related ac-
tivity was quantitative, not qualitative, thereby suggesting
that updating and nonupdating processes are not highly
segregated at the neural level.

Page 9

18 POSTLE, BERGER, GOLDSTEIN, CURTIS, AND D’ESPOSITO
Trial length. Examination of the time courses of
updating-sensitive voxels revealed an item-accumulation
effect in each participant: an increase of fMRI signal
with each increase in the number of items presented on
a trial (Figure 5). The slope of the linear regression line
fit to the group data was .037, and the y intercept was
.342.
PI. Each of the 6 participants whose RT data showed
a PI effect also showed a positive fMRI PI effect in left
BA 45, a reliable effect [t(5) = 2.6, p , .05] (Table 2).
The 7th participant (W), whose behavioral PI effect was
negligible (1.7msec), also showed only a negligible pos-
itive fMRI PI effect in left BA 45 [t(1409)= 0.08]. PI ef-
fects did not approach significance in right BA 45, left
dorsolateralPFC, or right dorsolateral PFC of these same
6 participants (Table 2).
The trial-length effect, in contrast with the PI effect,
was stronger in dorsolateral ROIs than in ventrolateral
ROIs and was significant in left dorsolateral PFC [t(5) =
4.4, p , .01] (Table 2). Thus, the PI effect and the trial-
length effect differed qualitatively.
Discussion
Updating engaged broad extents of dorsolateral and
ventrolateral PFC, as well as of other brain regions. This
is consistent with previous PET results with a similar
task (Salmon et al., 1996) and with fMRI results from
the n-back task (Braver et al., 1997; D’Esposito, Aguirre,
Zarahn, & Ballard, 1998; Nystrom et al., 2000; Postle,
Stern, Rosen, & Corkin, 2000; Schumacher etal., 1996),
which is believed to engage updating processes. Updat-
ing processes were associated with greater signal inten-
sity than initial (i.e., nonupdating) encoding and mainte-
nance processes, in both dorsolateral and ventrolateral
PFC. But updating voxels were identified in equivalent
proportions in dorsolateral and ventrolateral PFC. Addi-
tionally, updating voxels, on average, also demonstrated
robust sensitivity to nonupdating portions of the task,
and the converse was also true for voxels identified with
the nonupdating contrast. Thus, we did not find evidence
that the neural substrates of updating-related processes
are anatomically discrete from the strictly mnemonic
portions of the running span task. Although fMRI dis-
sociability is not a necessary feature of two independent
cognitive processes, its demonstration would be strong
evidence for independence. For example, the present re-
sults differ from those of at least two previous neu-
roimaging studies that sought explicitly to dissociate neu-
rallyexecutive control functions contributing to working
memory performance from maintenance functions. In
Table 2
Mean Group PI and Trial-Length Effects (n = 6),
Indexed as t Values, by PFC ROI, from Experiment 2
ROI
BA 45 PFC
Analysis
PI
Trial length
Note—Standard errors of the mean are presented in parentheses.
Left Right
1.29 (1.1)
1.0 (1.0)
Left Dorsolateral
0.96 (0.5)
1.6 (0.4)*
Right Dorsolateral
0.58 (0.8)
1.24 (0.7)
*p , .0.
1.92 (0.7)*
1.17 (0.89)
Figure 5. The fMRI time series data from each participant in Experiment 2, illus-
trating the item-accumulation effect. Data are spatially averaged across PFC updating-
sensitive voxels and are trial-averaged across 12-item 6-epoch trials. Vertical dotted
lines bound the portion of the trial during which updating operations were required.
The probe stimulus appeared at time 28 sec.

Page 10

UPDATING IN WORKING MEMORY19
one study, alphabetization-specific activity (an opera-
tionalization of the concept of manipulation of items in
working memory) was observed primarily in dorsolateral
PFC and was doubly dissociated from activity that was
sensitive to manipulations of the mnemonic components
of the task (Postle et al., 1999). This result was inter-
preted as evidence that alphabetization-related processes
are fundamentally different from the processes support-
ing delayed item recognition when no manipulation is re-
quired. In a second study, attention switching dissociated
from encoding and maintenance processes in a running
count task, a dissociation interpreted as evidence for cen-
tral executive functioning (Garavan, Ross, Li, & Stein,
2000). The results from the present study, however, do
not rule out the possibility that updating requirements in
a task may simply recruit, to a greater degree, the same
processes that are recruited to support performance on
simpler working memory tasks.
Trial length. The fMRI data from updating epochs re-
vealed an item-accumulation effect: Signal increased
linearly with the number of items presented during the
trial. This result is at variance with behavioral data from
this and other experiments but is consistent with ERP
data (Kiss et al., 1998). We characterize this as an item-
accumulation effect, rather than a PI effect, because there
is no evidence that the accumulation of items interferes
with behavior.6Thus, this effect differs importantly from
“load-sensitivity” effects in which increased working
memory load affects both performance and neuroimag-
ing signal (e.g., Braver et al., 1997; Jonides et al., 1997;
Postle et al., 1999).
Because the longest trials in which we observed this
item-accumulation effect were relatively short (present-
ing a total of 12 items and just six updating operations),
we do not know whether, at longer list lengths, it may sat-
urate in an item-dependent (or a time-dependent) manner
and whether it may begin interfering with task perfor-
mance. The answers to these questions may help deter-
mine whether the item-accumulation effect reflects the
operation of a PI-related process (Kiss et al., 1998), the
operation of previously undiscovered updating-related
processes, or simply the epiphenomenal buildup of a by-
product of updating activity, analogous to the buildup of
lactic acid with extended anaerobic operation of a muscle.
PI. The fMRI correlates of the probe-related intratrial
PI effect were found reliably only in BA 45—and thus
were anatomically specific—and were consistent with
the results obtained with intertrial PI in delayed item
recognition tasks (D’Esposito, Postle, Jonides, & Smith,
1999; Jonides et al., 2000; Jonides et al., 1998). This
suggests that this physiological correlate of the PI effect
may generalize to tasks in which PI is controlled within
trials, as well as across trials. Importantly, the analyses
of the effect of trial length on probe-related activity were
inconsistent with the memory-load alternative explana-
tion of the PI effect (Bunge et al., 2000), because PI and
trial-length effects differed qualitatively from each other,
not just quantitatively. Specifically, only left BA 45
showed a reliable PI effect in the fMRI data, whereas
probe-related activity that was sensitive to trial length
was numerically stronger in the dorsolateral PFC ROIs
and was reliable only in left dorsolateral PFC. This neu-
rophysiological double dissociation of PI and trial-length
effects is inconsistent with the idea that these two effects
reflect the operation of the same underlying mechanisms.
The locus ofthe trial-length effect in the present study—
which may index memory-scanning-related processes—
is consistent with the results of a previous study that ex-
amined the fMRI correlates of memory scanning
directly (Rypma & D’Esposito, 1999).
GENERAL DISCUSSION
The experiments described in this report contain many
novel features. They are the first to explore the role of
group integrity in working memory, the first to collect
suitable high temporal resolution behavioral and physi-
ological measures of updating in the same experiment,
and the first to effect detailed quantitative comparisons of
neuroimaging correlates of updating versus nonupdating
processes in PFC. Thus, it is perhaps not surprising that
they raise more questions than they resolve. The results of
these experiments reveal that working memory encoding
of letter strings automatically incorporates an episodic
contextual code, but the breadth of contextual informa-
tion that can be incorporated in this code and its accessi-
bility to volitional control remain undetermined. Future
research is required to address at least three important
questions about this episodic code. First, do participants
assess the probe against more items on trials that violate
group integrity than on those that do not? That is, is the
effective memory set larger on group integrity-violating
trials than on group integrity-preserving trials (a model
similar to the trial-length explanation of PI; Bunge et al.,
2000)? Second, what, if any, other kinds of contextual
information can be contained in this code? Third, is this
episodic code generated and maintained obligatorily, or
can either its encoding or its maintenance be suppressed
with appropriate training?
These experiments have also confirmed that the phys-
iological item-accumulation effect has no correlate in
accuracy or RT data, but the processes revealed by this
effect remain obscure. These experiments have failed to
reveal evidence that updating-related processes can be
dissociated neurally from nonupdating-related processes,
a result that calls into question the assumption that the
discarding and repositioning operations assumed to be
required by the updating task are fundamentally different
from the encoding- and maintenance-related processes
that are engaged by all, even the simplest, working mem-
ory tasks. Thus, further work is required to resolve whether
or not performance on the updating task reveals the op-

Page 11

20POSTLE, BERGER, GOLDSTEIN, CURTIS, AND D’ESPOSITO
eration of executive control functions in working memory
(Kiss et al., 1998; Lehto, 1996; Morris & Jones, 1990;
Salmon et al., 1996). Finally, by rejecting an alternative
explanation, these experiments have strengthened the
case that the neural effect associated with high-overlap
conditions of delayed-recognition tasks is indeed related
to the resolution of PI.
REFERENCES
Aguirre, G. K., Zarahn, E., & D’Esposito, M. (1998). The variability
of human, BOLD hemodynamic responses. NeuroImage, 8, 360-369.
Ashburner, J., & Friston, K. (1996). Fully three-dimensional nonlin-
ear spatial normalization: A new approach. NeuroImage, 3, S111.
Boynton, G. M., Engel, S. A., Glover, G. H., & Heeger, D. J.
(1996). Linear systems analysis of functional magnetic resonance
imaging in human V1. Journal of Neuroscience, 16, 4207-4221.
Braver, T. S., Cohen, J. D., Nystrom, L. E., Jonides, J., Smith, E. E.,
& Noll, D. C. (1997). A parametric study of prefrontal cortex in-
volvement in human working memory. NeuroImage, 5, 49-62.
Bunge, S. A., Matsumoto, A., Desmond, J. E., Glover, G. H., &
Gabrieli, J. D. E. (2000, March). The neural basis of interference
resolution: Manipulations of interference and working memory load
in the item recognition paradigm. Poster presented at the Rotman Re-
search Institute 10th Annual Conference: The frontal lobes.
Burgess, N., & Hitch, G. (1999). Memory for serial order: A network
model of the phonological loop and its timing. Psychological Review,
106, 551-581.
D’Esposito, M., Aguirre, G. K., Zarahn, E., & Ballard, D. (1998).
Functional MRI studies of spatial and non-spatial working memory.
Cognitive Brain Research, 7, 1-13.
D’Esposito, M., & Postle, B. R. (1999). The dependence of span and
delayed-response performance on prefrontal cortex. Neuropsycholo-
gia, 37, 1303-1315.
D’Esposito, M., & Postle, B. R. (2000). Neural correlates of compo-
nent processes of working memory: Evidence from neuropsycholog-
ical and pharmacological studies. In S. Monsell & J. Driver (Eds.),
Control of cognitive processes: Attention and performance XVIII
(pp. 579-602). Cambridge, MA: MIT Press.
D’Esposito, M., Postle, B. R., Ballard, D., & Lease, J. (1999).
Maintenance versus manipulation of information held in working
memory: An event-related fMRI study. Brain & Cognition, 41, 66-
86.
D’Esposito, M., Postle, B.R., Jonides, J., & Smith, E. E. (1999). The
neural substrate and temporal dynamics of interference effects in
working memory as revealed by event-related functional MRI. Pro-
ceedings of the National Academy of Sciences, 96, 7514-7519.
Friston, K. J., Ashburner, J., Frith, C. D., Poline, J.-B., Heather,
J. D., & Frackowiak, R. S. J. (1995). Spatial registration and nor-
malization of images. Human Brain Mapping, 2, 165-189.
Friston, K.J., Holmes, A. P., Poline, J.-B., Heather, J. D., & Frack-
owiak, R.S.J. (1995). Analysis of fMRI time-series revisited. Neuro-
Image, 2, 45-53.
Garavan, H. (1998). Serial attention within working memory. Memory
& Cognition, 26, 263-276.
Garavan, H., Ross, T. J., Li, S.-J., & Stein, E. A. (2000). A paramet-
ric manipulation of central executive functioning. Cerebral Cortex,
10, 585-592.
Hitch, G. J., Burgess, N., Towse, J. N., & Culpin, V. (1996). Tempo-
ral grouping effects in immediate recall: A working memory analysis.
Quarterly Journal of Experimental Psychology, 49A, 116-139.
Jonides, J., Marshuetz, C., Smith, E. E., Reuter-Lorenz, P. A., &
Koeppe, R. A. (2000). Age differences in behavior and PET activa-
tion reveal differences in interference resolution in verbal working
memory. Journal of Cognitive Neuroscience, 12, 188-196.
Jonides, J., Schumacher, E. H., Smith, E. E., Lauber, E. J., A wh, E.,
Minoshima, S., & Koeppe, R. A. (1997). Verbal working memory
load affects regional brain activation as measured by PET. Journal of
Cognitive Neuroscience, 9, 462-475.
Jonides, J., Smith, E. E., Marshuetz, C., Koeppe, R. A., & Reuter-
Lorenz, P. A. (1998). Inhibition in verbal working memory revealed
by brain activation. Proceedings of the National Academy of Sci-
ences, 95, 8410-8413.
Kiss, I., Pisio, C., Francois, A., & Schopflocher, D. (1998). Central
executive function in working memory: Event-related brain potential
studies. Cognitive Brain Research, 6, 235-247.
Lehto, J. (1996). Are executive function tests dependent on working
memory capacity? Quarterly Journal of Experimental Psychology,
49A, 29-50.
Morris, N., & Jones, D. M. (1990). Memory updating in working
memory: The role of the central executive. British Journal of Psy-
chology, 81, 111-121.
Nystrom, L. E., Braver, T. S., Sabb, F. W., Delgado, M. R., Noll,
D. C., & Cohen, J. D. (2000). Working memory for letters, shapes
and locations: fMRI evidence against stimulus-based regional orga-
nization of human prefrontal cortex. NeuroImage, 11, 424-446.
Petrides, M. (1989). Frontal lobes and memory. In F. Boller & J. Graf-
man (Eds.), Handbook of neuropsychology (3rd ed., Vol. 3, pp. 75-
90). New York: Elsevier.
Postle, B.R., Berger, J. S., & D’Esposito, M.(1999). Functional neu-
roanatomical double dissociation of mnemonic and executive control
processes contributing to working memory performance. Proceed-
ings of the National Academy of Sciences, 96, 12959-12964.
Postle, B. R., & D’Esposito, M. (2000). Evaluating models of the
topographical organization of working memory function in frontal
cortex with event-related fMRI. Psychobiology, 28, 132-145.
Postle, B. R., Stern, C. E., Rosen, B. R., & Corkin, S. (2000). An
fMRI investigation of cortical contributions to spatial and nonspatial
visual working memory. NeuroImage, 11, 409-423.
Postle, B. R., Zarahn, E., & D’Esposito, M. (2000). Using event-
related fMRI to assess delay-period activity during performance of
spatial and nonspatial working memory tasks. Brain Research Proto-
cols, 5, 57-66.
Rowe, J. B., Toni, I., Josephs, O., Frackowiak, R.S.J., & Passingham,
R. E. (2000). The prefrontal cortex: Response selection or mainte-
nance within working memory? Science, 288, 1656-1660.
Ryan, J. (1969). Grouping and short-term memory: Different means
and patterns of grouping. Quarterly Journal of Experimental Psy-
chology, 21, 137-147.
Rypma, B., & D’Esposito, M. (1999). The roles of prefrontal brain re-
gions in components of working memory: Effects of memory load
and individual differences. Proceedings of the National Academy of
Sciences, 96, 6558-6563.
Salmon, E., Van der Linden, M., Collette, F., Delfiore, G., Ma-
quet, P., Degueldre, C., Luxen, A., & Franck, G. (1996). Re-
gional brain activity during working memory tasks. Brain, 119, 1617-
1625.
Sarter, M., Berntson, G. G., & Cacioppo, J. T. (1996). Brain imag-
ing and cognitive neuroscience: Toward strong inference in attribut-
ing function to structure. American Psychologist, 51, 13-21.
Schumacher, E. H., Lauber, E., A wh, E., Jonides, J., Smith, E. E., &
Koeppe, R. A. (1996). PET evidence for an amodal verbal working
memory system. NeuroImage, 3, 79-88.
Talairach, J., & Tournoux, P. (1988). Co-planar stereotaxic atlas of
the human brain. New York: Thieme.
Teuber, H.-L. (1955). Physiological psychology. Annual Review of
Psychology, 6, 267-296.
Worsley, K. J., & Friston, K. J. (1995). Analysis of fMRI time-series
revisited—again. NeuroImage, 2, 173-182.
Zarahn, E., Aguirre, G. K., & D’Esposito, M. (1997a). Empirical
analyses of BOLD fMRI statistics: I. Spatially unsmoothed data col-
lected under null-hypothesis conditions. NeuroImage, 5, 179-197.
Zarahn, E., Aguirre, G.K., & D’Esposito, M.(1997b). A trial-based
experimental design for fMRI. NeuroImage, 6, 122-138.
NOTES
1. In pilot testing, one half of all updating trials featured two or more
discards that required violation of group integrity, and one half featured
zero or one such discard. Inspection of the data, however, indicated that

Page 12

UPDATING IN WORKING MEMORY 21
performance was sensitive only to violations of the integrity of groups
containing one or two items that would be in the final span of items
tested by the probe.
2. Note that the apparent ceiling effect could not be avoided, because
pilot testing indicated that maintaining a running span of 5 items was
too difficult.
3. Morris and Jones (1990) did not report EOA.
4. It bears restating that neural dissociability cannot be conclusively
demonstrated by physiological measures (such as fMRI) (Sarter,
Berntson, & Cacioppo, 1996). Doubly dissociable effects of functional
disruption (e.g., a lesion) in two discrete loci are required to draw this
inference conclusively (Teuber, 1955).
5. Note that the design of the behavioral experiments was also influ-
enced by fMRI-related constraints, in that the 4-sec EOA represented
the closest together we could position epochs whose evoked fMRI sig-
nal would be mutually resolvable (Zarahn et al., 1997b).
6. We thank Sharon Thompson-Schill for this observation.
(Manuscript received August 8, 2000;
revision accepted for publication January 2, 2001.)