1

Accepted for Publication in NeuroImage, 2005

Hippocampal Activations During Encoding and Retrieval in a Verbal Working MemoryParadigm

Katherine H. Karlsgodt1, David Shirinyan1, Theo G.M. van Erp1, Mark S. Cohen2,3,4,5, Tyrone D.Cannon1,2

1. Department of Psychology, UCLA, Los Angeles, CA 90095, USA2. Department of Psychiatry and Biobehavioral Sciences, UCLA, Los Angeles, CA 90095, USA

3. Department of Neurology, UCLA, Los Angeles, CA 90095, USA4. Division of Brain Mapping, UCLA, Los Angeles, CA 90095, USA

5. Department of Biomedical Physics, UCLA, Los Angeles, CA 90095, USA

Corresponding Author:Katherine H. Karlsgodt

Department of Psychology, UCLA1285 Franz Hall

Los Angeles, CA 90095-1563, USA(310) 794-9673 voice(310) 794-9740 Fax

karlsgodt@psych.ucla.edu email

2

Abstract

Though the hippocampus has been associated with encoding and retrieval processes in episodic

memory, the precise nature of its involvement in working memory has yet to be determined. This

functional magnetic resonance imaging (fMRI) study employed a verbal working memory

paradigm that allows for the within-subject comparison of functional activations during

encoding, maintenance, and retrieval. In each trial, participants were shown 5 target words and,

after an 8 second delay, a series of probe words. Probe words consisted of target matches,

phonetically or semantically related foils, or foils unrelated to the target words. Both the left and

right hippocampi showed higher mean activation amplitudes during encoding than maintenance.

In contrast, the right dorsolateral prefrontal cortex (DLPFC) showed greater activation during

maintenance than encoding. Both hippocampal and DLPFC regions were more active during

retrieval than maintenance. Furthermore, an analysis of retrieval activation separated by probe

type showed a trend toward greater bilateral hippocampal activation for probes related (both

semantically and phonetically) to the target than for unrelated probes and still greater activation

for target matches. This pattern suggests that there may be roles for the hippocampus and

DLPFC in working memory that change as function of information processing stage.

Additionally, the trend towards increased involvement of the hippocampus with the increase in

relatedness of the probe stimuli to the information maintained is interpreted to be consistent with

the role of the hippocampus in recollection-based retrieval in long-term memory and may

indicate that this role extends to working memory processes.

3

Introduction

Several lines of evidence support the involvement of the hippocampus in long-term memory

(LTM), including reports of memory impairments following hippocampal damage in both

humans and animals (Cave and Squire, 1992; Scoville and Milner, 1957), and hippocampal

activation in human neuroimaging studies of episodic memory (e.g. Eldridge et al, 2000). More

specifically, the hippocampus has been associated with encoding and retrieval of items in LTM

(Lepage et al, 1998; Schacter et al, 1999). The involvement with encoding has been proposed to

be in part due to the binding together of separate stimuli into a combined representation that

includes information about the relationship between the stimuli (Cohen et al, 1999; Davachi,

2004). Hippocampal activation during retrieval has been frequently discussed as it relates to

recognition memory, in particular to recollection of specific aspects of an episode (Manns et al,

2003; Squire et al, 2004).

Traditionally, this hippocampal association with LTM was thought to be at the exclusion of any

involvement in working memory (WM). Indeed, there is evidence that hippocampal damage may

leave basic WM functioning intact (Cave and Squire, 1992; Overman et al, 1990; Sidman et al,

1968; Wickelgren, 1968), and that WM and LTM can be independently impaired (Janowsky et

al, 1989; Warrington and Baddeley, 1974). However, it has also become clear that under normal

circumstances WM and LTM do not function in isolation. Rather, LTM and WM functions

interact in some way, as information stored in LTM can be applied to a current problem-solving

task and influence performance on WM tasks (Forde and Humphreys, 2002; Gathercole and

Baddeley, 1989; Hodges et al, 1994; Hulme et al, 1991). Many cognitive models of WM have

4

been adapted to incorporate an intersection with LTM processes (e.g. Baddeley, 2000; Cowan,

2000; Ericsson and Delaney, 2000).

Additional evidence for an overlap between WM and LTM comes from investigations of the

neuroanatomical structures implicated in each type of processing. In particular, functional

investigations of hippocampal activity have suggested that structures associated with LTM may

also be implicated in WM. Animals with hippocampal lesions can show alterations in WM

performance (Zola-Morgan and Squire, 1986) and hippocampal activity has been demonstrated

during WM tasks in behaving laboratory animals (Friedman and Goldman-Rakic, 1988; Sakurai,

1994). Furthermore, functional magnetic resonance imaging (fMRI) studies have shown

hippocampal activations during WM in humans (Cabeza et al, 2002; Davachi and Wagner, 2002;

Kato et al, 1998; Monk et al, 2002; Ranganath and D'Esposito, 2001). A question of great

importance is how to reconcile demonstrations of WM-LTM interactions and hippocampal

activations during WM with findings of preserved working memory performance in amnesiacs.

Although it has been demonstrated that amnesic patients show intact performance on basic

working memory tasks such as the digit span (Cave and Squire, 1992; Sidman et al., 1968;

Wickelgren, 1968), there is a paucity of data on whether more subtle aspects of WM are affected,

such as the longer WM span for real words than non-words observed by Hulme in healthy

patients (Hulme et al., 1991), or influences of semantic information on memory span. If the

hippocampus is found to be involved in these more complex aspects of WM but not necessary

for more basic processes, such a result would help to reconcile the finding of hippocampal

activations during WM with the data showing that subjects without a hippocampus can still

perform basic WM tasks.

5

The discrepancies in the neuropsychological and animal findings over whether WM is impaired

after hippocampal damage invite further exploration of hippocampal activations during WM

using fMRI in healthy, non-lesioned subjects. Since the hippocampus has been implicated in

episodic binding during encoding and recollective retrieval functions in LTM, the most

parsimonious prediction is that it’s role in WM may also involve these processes.

Existing functional imaging studies of WM have not allowed for clear parsing of the different

process stages (i.e. encoding, maintenance, retrieval); rather, these studies have collapsed

analyses across stages (e.g. encoding+maintenance or retrieval+baseline) (Cabeza et al., 2002;

Davachi and Wagner, 2002), focused the analysis only on a subset of the stages (Monk et al.,

2002), or incorporated additional factors such as novelty (Ranganath and D'Esposito, 2001) or

scene complexity (Park et al, 2003). However, tasks without such secondary factors have

demonstrated hippocampal activations in time periods that include encoding (Cabeza et al., 2002;

Davachi and Wagner, 2002; Monk et al., 2002) and retrieval (Cabeza et al., 2002). Further

evidence for hippocampal involvement during encoding and retrieval comes from single-unit

recording studies in which hippocampal cells fired for encoding and retrieval but not

maintenance (Deadwyler et al, 1996; Hampson and Deadwyler, 2003). However, human fMRI

studies clearly differentiating each stage of activation are needed to resolve this issue.

Assuming the hippocampus has a role in WM, a second issue is whether the hippocampus is

performing a function of a similar nature across these phases of WM and LTM. During encoding

in WM the hippocampus may be functioning much as it would in LTM, by binding the

6

information and experience into an episode, even if in the context of the task the episode only

needs to be maintained for a brief period of time (e.g. Postle, 2003). Similarly, the hippocampal

role in WM retrieval may be associated with its involvement in the recollective aspect of

retrieval in LTM (Davachi, 2004; Eldridge et al., 2000; Manns et al., 2003; Nemanic et al, 2004).

It’s role during retrieval in WM may be to recollect specific items out of the material being

maintained, and recognize whether they match the probe. This issue can be addressed by

incorporating different levels of recollection into the framework of a WM task.

This study uses verbal stimuli (non-novel, familiar words) to investigate the phasic role of the

hippocampus in WM using a within-subjects design. Foil types presented during retrieval varied

in the degree of relatedness of the probe to the target. This manipulation was designed to assess

the nature of the information being accessed during the retrieval phase and whether it contained

an episodic-like trace.

Materials and Methods

Participants: Thirteen healthy adult subjects (7 female, 6 male; age (mean +/- SD) = 24.08+/-

3.48) participated in this study. All subjects gave written informed consent before their

participation on a protocol approved by the Internal Review Board of the University of

California, Los Angeles.

Cognitive paradigm: The cognitive task is adapted from a paradigm used by Jonides et al (1998).

In each trial subjects were shown a set of 5 unrelated target words displayed in pink capital

letters; each word was displayed for 1 second with a 1 second fixation between words. After an 8

7

second delay, subjects were shown a series of 8-10 probe words displayed in yellow lowercase

letters. Probes were presented for 1.5 seconds and were separated by either a 500ms or 2500ms

fixation (presentation timing of probes was “jittered”). Each trial was followed by 18.5-20.5

seconds of rest (see Figure 1). The probe words were of four types: target matches, unrelated

foils, semantically related foils (synonyms of a target word), and phonetically related foils

(differing from a target word by one phoneme). All words were 3-8 letters long and had 1-2

syllables. Subjects responded to probe words via handheld button box by pressing “1” if the

word had been in the target list and “2” if it had not. The experiment was divided into two 6.5

minute runs of 7 trials each, resulting in a total of approximately 33 of each probe type. The task

was run on a PC laptop using E-Prime presentation software (Psychology Software Tools) and

displayed to the subjects using goggles (Resonance Technologies, Inc, Northridge CA).

Functional Imaging Parameters: All scans were acquired on a 3T Siemens Allegra scanner at the

Ahmanson-Lovelace Brain Mapping Center at UCLA. The first part of the imaging session

consisted of a 3-plane localizer, a brief field mapping/shimming sequence, and sagittal scan to

aid in slice alignment. A co-planar T2 weighted image with 1.5mm in-plane resolution was taken

for anatomical registration. This scan was acquired using a set of high-resolution EPI localizers

(TR/TE 5000/33ms, 25 3-mm slices with 1mm gap, 128x128 matrix, 200mm FOV) in the same

AC-PC aligned plane as the functional images. The high-resolution scans had readout bandwidth

along the phase encoding direction identical to that in the functional scans so that any B0-related

distortions would be matched. Functional slices were matched to those in the T2 weighted image,

and utilized an echo planar (EPI) sequence, acquiring 25 3-mm slices using a TR/TE of

8

2500/45ms, and flip angle of 80 degrees. Each run consisted of 158 scans and lasted

approximately 6.5 minutes with each subject performing 2 runs of the task.

Image Processing: Image analysis was performed using FSL tools (FMRIB's Software Library,

www.fmrib.ox.ac.uk/fsl; Smith et al, 2004). EPI data were corrected to combat potential motion

artifacts by utilizing a 3D co-registration (6 parameter rigid-body) to register each BOLD image

in the time series to the middle data point in the time series. Data were then registered, first the

EPI to the subject’s individual T2-weighted structural image, then the T2 to the MNI-152

standard space template (Jenkinson et al, 2002; Jenkinson and Smith, 2001). Data were spatially

smoothed using a 5mm (FWHM) Gaussian kernel. Individual subject analyses were carried out

using FEAT (FMRI Expert Analysis Tool) Version 5.1. Intensity normalization (“grand-mean

scaling”) was performed so that the runs being carried into the second level analysis would have

the same overall mean intensity and comparisons in higher-level analyses would be valid. Thus,

each individual run (each 4D data set) was scaled by multiplying the whole dataset by a single

scaling factor such that all (non-background) voxels would have the same mean intensity of

10,000. This is equivalent to demeaning the data across run datasets, and does not bias the final

analysis. The data were high-pass filtered to remove low frequency artifacts using a non-linear

high-pass temporal filter with a cut off of 50s, in which a Gaussian-weighted straight line was

locally fit using a least squares fit (LSF) method.

Time-series statistical analysis was carried out using FILM (FMRIB's Improved Linear Model)

with local autocorrelation correction (Woolrich et al, 2001). The hemodynamic response

function (HRF) was described using a gamma variate function with a mean lag of 6s, which was

9

convolved with the modeled elements. Two analyses were carried out, Analysis 1, in which

encoding, retrieval, and maintenance were modeled separately, and Analysis 2, in which the

retrieval activation was decomposed by individually modeling each probe type. In each analysis,

the general linear model was applied to the fMRI time series using FILM (FMRIB’s Improved

Linear Model). A univariate General Linear Model was applied on a voxel-by-voxel basis so that

each voxel’s time course was individually fit to the resulting model, with local autocorrelation

correction that was applied within tissue type to improve the estimation of temporal smoothness

(Smith et al, 2004; Woolrich et al., 2001). Goodness-of-fit of each voxel to the models was

estimated, and the resulting parameter estimates indicated the degree to which the change in

fMRI signal could be explained by each model.

The results of this first level analysis were fed into a second level analysis, in which a model was

made for each subject such that for each subject the two runs of the task were combined. This is

to ensure that in the group analysis the degrees of freedom are appropriately considered as being

based on the number of subjects rather than the number of runs. The third level analysis was

carried out using FLAME (FMRIB's Local Analysis of Mixed Effects) (Behrens et al, 2003;

Smith et al., 2004). Each subject’s combined data from the second level analysis was entered and

all subjects were modeled together as one group. This third level design matrix was individually

applied to each of the contrasts from the first level analysis. For Analysis 1, encoding,

maintenance, and retrieval were each modeled against baseline. In Analysis 2, encoding,

maintenance, and each of the 4 probe types were modeled against baseline. In this two-stage

higher level analysis first Bayesian modeling was used to make an initial estimate of the

activation, then to complete the final estimate a Markov Chain Monte Carlo (MCMC) based

10

analysis was performed at all voxels determined to be near threshold in the first stage. Resulting

Z-statistic images were thresholded using clusters determined by Z>2.3 and a (corrected) cluster

significance threshold of P=0.01 (Forman et al, 1995; Friston et al, 1994; Worsley et al, 1992).

Cluster p-values were determined using a spatial smoothness estimation implemented in Feat

(Jenkinson, 2001a) based on the methods of Forman (1995) and Friston (1994)

Region of Interest Definition: Regions of Interest (ROI’s) were created in the standard space.

First, the regions in question (e.g. the hippocampus and DLPFC) were defined anatomically.

Then, the t-statistic images from the functional contrasts of all conditions vs. baseline (e.g.

encoding - baseline, maintenance - baseline, and retrieval - baseline) were filtered using an

uncorrected t12>3.055 threshold of the t-statistic image (2-tailed, p<0.01). The union of the

thresholded active voxels was used as a functional ROI. All conditions were allowed to

contribute to the formation of the ROI, so that when comparing between conditions, the levels of

activation would not be artificially skewed towards any one of the individual contrasts. The

standard space ROI’s were warped into each subject’s individual space and the motion corrected,

smoothed and high-pass filtered timeseries data was probed for percent signal change from

baseline using Featquery (FMRIB Software Library).

Results

Behavioral data: Behavioral measures of accuracy and reaction time were acquired for all

subjects, although due to a button-box failure data from one of the subjects was unusable.

However, this subject performed normally on the out-of-scanner training session. Accuracy did

not significantly vary by probe type, (p=.191). The average overall accuracy was 95.2±.08%

11

correct. Reaction times were 813.8±77.5ms for phonetic foils, 821.92±80.7ms for semantic foils,

834.3±91.9ms for target matches and 717.2±81.9ms for unrelated foils. A within-subjects

repeated measures ANOVA showed a significant difference in reaction time by probe type

(p<.001). Post-hoc decomposition of the effect showed that while phonetic foils, semantic foils,

and target matches did not significantly differ from each other, unrelated probes had a

significantly shorter reaction time than the other three probe types (P<.01 for all three

comparisons).

fMRI Analysis: Whole brain analyses showed that a variety of regions commonly associated with

verbal processing and WM (i.e. the parietal lobe and inferior frontal regions) were active in

comparison to baseline during the encoding, maintenance, and retrieval phases of the task (see

Table 1, Figure 2). However, the primary investigation will focus on the findings in the

hippocampus, which was further probed through ROI analyses.

ROI Analysis: Bilateral hippocampal regions of interest were defined (see Figure 3). A within-

subjects repeated measures ANOVA showed a significant effect of task phase across both ROI’s

(p<.001) (see Figure 4a). A post-hoc test differentiating the activation by hemisphere showed no

effect of hemisphere (p=.851) and no phase x hemisphere interaction (p=.178). The specificity

of hippocampal activity to encoding and retrieval was reinforced by the effect size calculation in

all hippocampal voxels, where for encoding r=.479 and retrieval r=.699 but for maintenance

r=.064.

12

To further examine the hippocampal activity during retrieval a separate analysis broken down by

probe type was performed (see Figure 4b). A repeated measures within-subjects ANOVA

examining the effects of probe types (“unrelated probes” as the least related, “semantic foils” and

“phonetic foils” combined into a mid-level group, and “target matches” being the most related)

was performed. Bilateral hippocampal activation in each hemisphere was compared to baseline,

and in a linear contrast there was a trend (p=.084) towards an effect of relatedness with unrelated

probes showing less activation related probes which showed less than target matches.

A second ROI in the right dorsolateral prefrontal cortex (DLPFC) was assessed. A within-

subjects repeated measures ANOVA showed activity compared to baseline in this region to be

greater in maintenance than in encoding, and greater still in retrieval (p<.001) (see Figure 5a).

When the retrieval period was further dissociated by probe type there was a significant effect of

type (p=.008). Post hoc paired t-tests indicate that both targets (p=.025) and Related foils

(p=.004) showed greater activation than Unrelated probes, though they did not significantly

differ from each other (p=.987).

Additionally there was a significant interaction between region and task phase (P<.01) with

hippocampal activation in comparison to baseline decreasing from encoding to maintenance,

DLPFC activation increasing from encoding to maintenance, and both increasing during

retrieval.

13

Discussion

The design of this task allowed for the assessment of hippocampal activation during encoding,

maintenance, and retrieval stages of verbal WM within-subjects. The principal findings are that

hippocampal activity during this verbal WM task is specifically associated with the encoding and

retrieval of information, but not with its maintenance. In addition, hippocampal activation shows

a tendency to vary according to whether the probe stimulus is related to the initially seen target.

Such a pattern suggests specific involvement in an episodic component of short term or working

memory, a possibility that is discussed further below.

Hippocampal activation during encoding

The hippocampal activation during WM encoding is similar to findings of hippocampal

involvement in LTM encoding (Davachi and Wagner, 2002; Lepage et al., 1998; Nyberg et al,

1996; Schacter et al., 1999), and thus may support the idea that the hippocampus has similar

roles in WM and LTM. Postle (2003) has proposed that episodic-type information about context

may be routinely stored as part of short-term memory traces, and that encoding of episodic

information is not limited only to LTM. This finding may extend that idea further, into a WM

context, and indicate that the context encoding in both LTM and WM relies on similar

mechanisms. In this sense, the WM-LTM overlap may resemble Baddeley’s “episodic buffer”,

an intermediate storage system controlled and accessed by the central executive that plays a role

in temporary episodic binding of information in WM and in moving information in and out of

episodic LTM (Baddeley, 2000), a function which could be supported by the hippocampus.

Although it is possible for WM and LTM to be dissociated in neurological patients and in

animals with lesions, there is cognitive evidence for an interaction between the systems during

14

normal functioning (Forde and Humphreys, 2002; Gathercole and Baddeley, 1989; Hodges et al.,

1994; Hulme et al., 1991). If the WM trace can include information about context, it may be the

basis of some of the findings in which information associated with LTM has an influence on

WM performance. Additionally, if the hippocampus is active during both types of context

encoding, it may provide a physiological substrate for this intersection of information held in

WM and LTM.

In addition to the behavioral evidence, functional data support our finding of hippocampal

activation during encoding. There are animal data showing hippocampal activity specifically

during the encoding of a delay-non-match-to-sample task (Deadwyler et al., 1996; Hampson and

Deadwyler, 2003); evidence of neuronal firing during encoding lends support to the current

neuroimaging findings showing increased BOLD activity during the same period. Although the

fMRI data in humans has not been conclusive, some working memory studies have shown

hippocampal activation during time periods that include encoding (Cabeza et al., 2002; Davachi

and Wagner, 2002; Monk et al., 2002). Other studies, such as that of Ranganath and D’Esposito

(2001) have shown a different pattern, with activation during maintenance rather than encoding.

However, if the activation in the current study is, as hypothesized, due to an episodic binding

function in WM, it is possible that the Ranganath stimuli were not sufficient to induce this effect.

In that study, the encoding material was a single face presented for 1s, in contrast to the 5 words

presented over 10s used in our study. It may be that with the longer time period and greater

quantity of material more extensive binding was induced and that this was reflected in the BOLD

signal. The finding of hippocampal activity during maintenance by Ranganath and D’Esposito

may have been a function of the novelty of their stimuli, as they reported maintenance activation

15

during the novel but not the familiar condition, just as the present study did not find activity

during maintenance when using familiar stimuli. In further support of the WM binding

hypothesis, another fMRI study (Mitchell et al, 2000) found higher levels of hippocampal

activation during a WM task in the condition associated with a greater feature binding demand.

However, this analysis did not subdivide the task by phase, so is inconclusive as to when such

binding may be occurring. Our results support the idea that this type of episodic-like binding is

occurring during the encoding period of WM tasks.

Hippocampal activation during retrieval

In addition to hippocampal activation during WM encoding, we also demonstrated activation

during WM retrieval, a finding that has been supported by animal literature (Deadwyler et al.,

1996; Hampson and Deadwyler, 2003) and a portion of the human WM fMRI data (Cabeza et

al., 2002). This result further supports the idea that the hippocampus may be performing similar

functions during both WM and LTM, as hippocampal activation has been found during LTM

retrieval (Eldridge et al., 2000; Ranganath et al, 2004; Small et al, 2001). Such retrieval

activation has been proposed to be due to the reactivation of representations bound together

during encoding (Nyberg et al., 1996; Squire and Zola-Morgan, 1991). If the encoding activation

found in this study was indeed due to episodic binding, then the retrieval activation may be a

result of the reactivation of the trace, just as in LTM. One caveat is that as there are multiple

probes for each trial, the retrieval phase contains an additional ongoing maintenance aspect.

However, as the activation during the pure maintenance phase was significantly lower than both

encoding and retrieval in comparison to baseline, it is less likely that the retrieval activation is

due to maintenance loads than that it is more specifically related to retrieval related processes.

16

The finding that hippocampal activation tended to increase as the resemblance of the probe to the

original target increased is interesting both in regards to hippocampal involvement in recollection

(Eldridge et al., 2000; Nemanic et al., 2004; Ranganath et al., 2004) and to its proposed role in an

episodic binding component of WM. During our task, determining that an unrelated foil differs

from the target only necessitates recollection of an imprecise representation of the original target,

possibly even using a familiarity-based trace. However, to respond to a related word, it is

necessary to recall a more specific version of the target in order to ensure that it indeed differs

along the dimension the two words share. And finally, to identify a target match it is necessary to

specifically and accurately recollect the target word to guarantee that they are exactly the same.

In this way the different probe types may result in a gradient of the level of episodic recollection

needed to respond to them, and it may be this increase in recollection that is reflected in the

change in hippocampal activation. Another possibility is that the related and target words serve

as cues, and that the more closely related cue results in a more detailed recollection of the

encoding episode. Future studies investigating this with greater power are needed further inform

this issue.

One alternative explanation for the finding of retrieval-based hippocampal activation is that it

reflects incidental simultaneous encoding of the information into LTM that is unrelated to WM

processes (Buckner et al, 2001). Other studies finding hippocampal activation during WM

maintenance have made a similar argument, that the maintenance period in WM might co-occur

with encoding into episodic memory (Davachi and Wagner, 2002; Ranganath and D'Esposito,

2001). However, the pattern of results in our study, in which the retrieval activation tends to

17

change with probe type, makes it less likely that the activity is due only to encoding. In this task,

in the absence of the memory of the target there is no differentiation between probe types, as

their “relatedness” is defined in terms of the target, not in any quality of the individual words or

of the relationship of the probes to each other. Thus, although it is possible that some of the

activation seen is due to LTM encoding processes, these results indicate that even within the

confines of a WM task, the hippocampus may be sensitive to a trace containing information

about the relatedness of the probes and targets and is responding accordingly.

Hippocampal and DLPFC Activation

Our study showed the hippocampus to be active during verbal encoding and retrieval, but less

active during maintenance. In contrast, the DLPFC was more active during maintenance than

during encoding of verbal information. Previous work has demonstrated that the hippocampus

and prefrontal regions are highly interconnected (Goldman-Rakic et al, 1984), and the

dorsolateral prefrontal cortex has unique cellular mechanisms that can support WM maintenance

(Goldman-Rakic, 1995). It is thus feasible for the hippocampus to be involved with the

encoding, but not the maintenance, of information being held in WM, with the DLPFC taking

over and becoming more active during maintenance. Reasons for the high activation of both

regions during retrieval may be due to the hippocampus’ role in recollecting the initial stimuli

and to the DLPFC’s role in both the ongoing maintenance of the targets throughout the retrieval

phase, and in directing selection of the target from the memory set that is being actively

maintained.

18

Conclusions

In this study we found a role for the hippocampus in WM that appears to be much like its role in

long-term episodic memory, indicating that it may be involved in encoding and retrieval in

general rather than only during LTM tasks. Additionally, the activity during WM may occur in

concert with the activity in the pre-frontal cortex with each contributing uniquely to the distinct

phases of WM. The finding that a region associated with LTM is selectively active during WM

may provide a physiological basis for the cognitive findings of overlap between WM and LTM

processes that should be explored in future research. Other studies have found that hippocampal

activation increased during WM with novel or complex stimuli, however this study shows that

even with non-novel stimuli it is possible to see phasic task related differences in hippocampal

activation in a WM paradigm. Additionally it appears that its previously noted role in

recognition/recollection may be an aspect of its involvement in WM. The observation of

variation of hippocampal activation with the relatedness of probes to targets within the time-

frame of a WM paradigm could potentially indicate that, like LTM, WM is sensitive to episodic

aspects of information being maintained.

19

Acknowledgements:This research was supported by grant MH65079 from the National Institute of Mental Health(NIMH), by grant RR00827 to the FIRST Biomedical Informatics Research Network (BIRN,http://www.nbirn.net), that is funded by the National Center for Research Resources (NCRR) atthe National Institutes of Health (NIH), and by a gift to the UCLA Foundation from Garen andShari Staglin. The authors wish to thank David Glahn, Ph.D., Russell Poldrack, Ph.D., KristinHerzberg, Vikas Rao and Tyler Lesh, as well as the study participants.

20

References

Baddeley, A. 2000. The episodic buffer: a new component of working memory? Trends CognSci. 4, 417-423.

Behrens, T.E., Johansen-Berg, H., Woolrich, M.W., Smith, S.M., Wheeler-Kingshott, C.A.,Boulby, P.A., Barker, G.J., Sillery, E.L., Sheehan, K., Ciccarelli, O., Thompson, A.J.,Brady, J.M.,Matthews, P.M. 2003. Non-invasive mapping of connections between humanthalamus and cortex using diffusion imaging. Nat Neurosci. 6, 750-757.

Buckner, R.L., Wheeler, M.E.,Sheridan, M.A. 2001. Encoding processes during retrieval tasks. JCogn Neurosci. 13, 406-415.

Cabeza, R., Dolcos, F., Graham, R.,Nyberg, L. 2002. Similarities and differences in the neuralcorrelates of episodic memory retrieval and working memory. Neuroimage. 16, 317-330.

Cave, C.B.,Squire, L.R. 1992. Intact verbal and nonverbal short-term memory following damageto the human hippocampus. Hippocampus. 2, 151-163.

Cohen, N.J., Ryan, J., Hunt, C., Romine, L., Wszalek, T.,Nash, C. 1999. Hippocampal systemand declarative (relational) memory: summarizing the data from functional neuroimagingstudies. Hippocampus. 9, 83-98.

Cowan, N., 2000. An embedded-process model of working memory. In: Shah P, (Ed.). Models ofworking memory: Mechanisms of active maintenance and executive control. CambridgeUniversity Press, New York 2000.

Davachi, L. 2004. The ensemble that plays together, stays together. Hippocampus. 14, 1-3.Davachi, L.,Wagner, A.D. 2002. Hippocampal contributions to episodic encoding: insights from

relational and item-based learning. J Neurophysiol. 88, 982-990.Deadwyler, S.A., Bunn, T.,Hampson, R.E. 1996. Hippocampal ensemble activity during spatial

delayed-nonmatch-to-sample performance in rats. J Neurosci. 16, 354-372.Eldridge, L.L., Knowlton, B.J., Furmanski, C.S., Bookheimer, S.Y.,Engel, S.A. 2000.

Remembering episodes: a selective role for the hippocampus during retrieval. NatNeurosci. 3, 1149-1152.

Ericsson, K.A.,Delaney, P., 2000. Long-term working memory as an alternative to capacitymodels of working memory in everyday skilled performance. In: Shah P, (Ed.). Models ofworking memory: Mechanisms of active maintenance and executive control. CambridgeUniversity Press, New York 2000.

Forde, E.M.,Humphreys, G.W. 2002. The role of semantic knowledge in short-term memory.Neurocase. 8, 13-27.

Forman, S.D., Cohen, J.D., Fitzgerald, M., Eddy, M.A., Mintun, M.A.,Noll, D.C. 1995.Improved assessment of significant activation in functional magnetic resonance imaging(fMRI): Use of a cluster-size threshold. Magn Reson Med. 33, 636-647.

Friedman, H.R.,Goldman-Rakic, P.S. 1988. Activation of the hippocampus and dentate gyrus byworking-memory: a 2-deoxyglucose study of behaving rhesus monkeys. J Neurosci. 8,4693-4706.

Friston, K., Worsley, K., Frakowiak, R., Mazziotta, J.C.,Evans, A.C. 1994. Assessing thesignificance of focal activations using their spatial extent. Hum Brain Mapp. 1, 214-220.

Gathercole, S.E.,Baddeley, A. 1989. Evaluation of the role of phonological STM in thedevelopment of vocabulary in children: A longitudinal study. Journal of Memory andLanguage. 28, 200-213.

21

Goldman-Rakic, P.S. 1995. Architecture of the prefrontal cortex and the central executive. AnnN Y Acad Sci. 769, 71-83.

Goldman-Rakic, P.S., Selemon, L.D.,Schwartz, M.L. 1984. Dual pathways connecting thedorsolateral prefrontal cortex with the hippocampal formation and parahippocampalcortex in the rhesus monkey. Neuroscience. 12, 719-743.

Hampson, R.E.,Deadwyler, S.A. 2003. Temporal firing characteristics and the strategic role ofsubicular neurons in short-term memory. Hippocampus. 13, 529-541.

Hodges, J.R., Patterson, K.,Tyler, L.K. 1994. Loss of semantic memory: Implications for themodularity of mind. Cognitive Neuropsychology. 11, 505-542.

Hulme, C., Maughan, S.,Brown, G.D.A. 1991. Memory for familiar and unfamiliar words:Evidence for a long-term memory contribution to short term memory span. Journal ofMemory and Language. 30, 685-701.

Janowsky, J.S., Shimamura, A.P., Kritchevsky, M.,Squire, L.R. 1989. Cognitive impairmentfollowing frontal lobe damage and its relevance to human amnesia. Behav Neurosci. 103,548-560.

Jenkinson, M., Bannister, P., Brady, M.,Smith, S.M. 2002. Improved optimisation for the robustand accurate linear registration and motion correctino of brain images. Neuroimage. 17,825-841.

Jenkinson, M.,Smith, S.M. 2001. A global optimisation method for robust affine registration ofbrain images. Medical Image Analysis. 5, 143-156.

Jenkinson, M. 2001a. Estimation of Smoothness from the Residual Field: FMRIB TechnicalReport TR00MJ3. http://www.fmrib.ox.ac.uk/analysis/techrep/tr00mj3/tr00mj3/

Jonides, J., Schumacher, E.H., Smith, E.E., Koeppe, R.A., Awh, E., Reuter-Lorenz, P.A.,Marshuetz, C.,Willis, C.R. 1998. The role of parietal cortex in verbal working memory. JNeurosci. 18, 5026-5034.

Kato, T., Erhard, P., Takayama, Y., Strupp, J., Le, T.H., Ogawa, S.,Ugurbil, K. 1998. Humanhippocampal long-term sustained response during word memory processing. Neuroreport.9, 1041-1047.

Lepage, M., Habib, R.,Tulving, E. 1998. Hippocampal PET activations of memory encoding andretrieval: the HIPER model. Hippocampus. 8, 313-322.

Manns, J.R., Hopkins, R.O., Reed, J.M., Kitchener, E.G.,Squire, L.R. 2003. Recognitionmemory and the human hippocampus. Neuron. 37, 171-180.

Mitchell, K.J., Johnson, M.K., Raye, C.L.,D'Esposito, M. 2000. fMRI evidence of age-relatedhippocampal dysfunction in feature binding in working memory. Brain Res Cogn BrainRes. 10, 197-206.

Monk, C.S., Zhuang, J., Curtis, W.J., Ofenloch, I.T., Tottenham, N., Nelson, C.A.,Hu, X. 2002.Human hippocampal activation in the delayed matching- and nonmatching-to-samplememory tasks: an event-related functional MRI approach. Behav Neurosci. 116, 716-721.

Nemanic, S., Alvarado, M.C.,Bachevalier, J. 2004. The hippocampal/parahippocampal regionsand recognition memory: insights from visual paired comparison versus object-delayednonmatching in monkeys. J Neurosci. 24, 2013-2026.

Nyberg, L., McIntosh, A.R., Houle, S., Nilsson, L.G.,Tulving, E. 1996. Activation of medialtemporal structures during episodic memory retrieval. Nature. 380, 715-717.

Overman, W.H., Ormsby, G.,Mishkin, M. 1990. Picture recognition vs. picture discriminationlearning in monkeys with medial temporal removals. Exp Brain Res. 79, 18-24.

22

Park, D.C., Welsh, R.C., Marshuetz, C., Gutchess, A.H., Mikels, J., Polk, T.A., Noll,D.C.,Taylor, S.F. 2003. Working memory for complex scenes: age differences in frontaland hippocampal activations. J Cogn Neurosci. 15, 1122-1134.

Postle, B.R. 2003. Context in verbal short-term memory. Mem Cognit. 31, 1198-1207.Ranganath, C.,D'Esposito, M. 2001. Medial temporal lobe activity associated with active

maintenance of novel information. Neuron. 31, 865-873.Ranganath, C., Yonelinas, A.P., Cohen, M.X., Dy, C.J., Tom, S.M.,D'Esposito, M. 2004.

Dissociable correlates of recollection and familiarity within the medial temporal lobes.Neuropsychologia. 42, 2-13.

Sakurai, Y. 1994. Involvement of auditory cortical and hippocampal neurons in auditory workingmemory and reference memory in the rat. J Neurosci. 14, 2606-2623.

Schacter, D.L., Curran, T., Reiman, E.M., Chen, K., Bandy, D.J.,Frost, J.T. 1999. Medialtemporal lobe activation during episodic encoding and retrieval: a PET study.Hippocampus. 9, 575-581.

Scoville, W.B.,Milner, B. 1957. Loss of recent memory after bilateral hippocampal lesions. JNeurochem. 20, 11-21.

Sidman, M., Stoddard, L.T.,Mohr, J.P. 1968. Some additional quantitative observations ofimmediate memory in a patient with bilateral hippocampal lesions. Neuropsychologia. 6,245-254.

Small, S.A., Nava, A.S., Perera, G.M., DeLaPaz, R., Mayeux, R.,Stern, Y. 2001. Circuitmechanisms underlying memory encoding and retrieval in the long axis of thehippocampal formation. Nat Neurosci. 4, 442-449.

Smith, S.M., Jenkinson, M., Woolrich, M.W., Beckmann, C.F., Behrens, T.E., Johansen-Berg,H., Bannister, P.R., De Luca, M., Drobnjak, I., Flitney, D.E., Niazy, R.K., Saunders, J.,Vickers, J., Zhang, Y., De Stefano, N., Brady, J.M.,Matthews, P.M. 2004. Advances infunctional and structural MR image analysis and implementation as FSL. Neuroimage. 23Suppl 1, S208-219.

Squire, L.R., Stark, C.E.,Clark, R.E. 2004. The medial temporal lobe. Annu Rev Neurosci. 27,279-306.

Squire, L.R.,Zola-Morgan, S. 1991. The medial temporal lobe memory system. Science. 253,1380-1386.

Warrington, E.K.,Baddeley, A.D. 1974. Amnesia and memory for visual location.Neuropsychologia. 12, 257-263.

Wickelgren, W.A. 1968. Sparing of short-term memory in an amnesic patient: Implications forstrength theory of memory. Neuropsychologia. 6, 235-244.

Woolrich, M.W., Ripley, B.D.,Smith, S.M. 2001. Temporal autocorrelation in univariate linearmodelling of fMRI data. Neuroimage. 14, 1370-1386.

Worsley, K.J., Evans, A.C., Marrett, S.,Neelin, P. 1992. A three-dimensional statistical analysisfor CBF activation studies in human brain. Journal of Cerebral Blood Flow andMetabolism. 12, 900-918.

Zola-Morgan, S.,Squire, L.R. 1986. Memory impairment in monkeys following lesions limited tothe hippocampus. Behav Neurosci. 100, 155-160.

23

Figure Legends:Figure 1. Task design and sample trial stimuli

Figure 2. Functional activation at each task phase as resulting from FSL cluster analysis. A. LeftHemisphere activation B. Right Hemisphere activation

Figure 3. Hippocampal ROI placement: A. bilateral view of regions of interest on acoronal slice, B. Right Hippocampal ROI, C. Left Hippocampal ROI.

Figure 4a. Left and Right hippocampal activation by task phaseFigure 4b. Bilateral hippocampal activation at retrieval by probe type

Figure 5a. Right DLPFC activation by task phaseFigure 5b. Right DLPFC activation at retrieval by probe type.

Table 1. Regions reaching significance in cluster analysis (P<.01) in comparison to baseline

Figure 1

24

Figure 1

Figure 2

25

Figure 3

Figure 4a

Figure 4b

26

Figure 5a

Figure 5b

27

Table 1. Regions reaching significance in cluster analysis (p<.01)Hem. BA Talairach MNI Local Max

ZstatEncoding

R 47 34,24,-3 34,25,-2 4.15Inferior frontal gyrusL 47 -41, 24, -3 -41, 25, -2 3.58R 26,11,-4 26,12,-2 4.38PutamenL -26, 10,-4 -26,11,-4 3.98R 6 57,-1,41 58.-3,44 4.47Precentral gyrusL 6 -55,1,41 -56,-1,45 4.96R 2,2,43 2,10,47 4.53Medial frontal gyrusL -8,13,43 -8,14,47 4.87R 17 20,-88,-4 20,-90,-10 4.65Lingual gyrusL 17 -22, -91,-3 -22,-94,-9 4.13

Middle temporal gyrus R 22 45,-26,-3 45,-27,-5 3.67Superior parietal L 7 -30,-62,44 -30,-66,44 3.98

R 19 38,-78,-10 38,-80,-17 4.23Occipital gyrusL 19 -43,-74,-10 -43,-76,-16 4.67

Maintenance R 47 36,28,-10 36,29,-10 3.46Inferior frontal gyrusL 47 -43,5,24 -43,4,26 3.50

DLPFC R 46 47,31,20 47,31,23 3.62Precentral Gyrus L 6 -56,-6,36 -57,-8,39 3.88

R 32 15,12,36 15,10,40 3.94Cingulate gyrusL 24 -8,7,46 -8,5,50 3.46R 40 41,-51,41 41,-55,42 3.55Inferior parietal lobuleL 39/40 -32,-58,38 -32,-62,38 3.38

Retrieval R 45 37,21,-2 37,22,-1 4.53L 47 -31,20,-2 -31,21,-1 4.38

Inferior frontal gyrus

L 44 -52,11,23 -53,10,26 4.90Insula R 13 36,21,1 36,22,2 4.52

L 13 -41,-5,10 -41,-6,11 3.89R 10 31,50,7 31,51,10 3.76L 10 -32,50,0 -32,51,3 4.23

Middle frontal gyrus

L 6 -51,5,44 -52, 3, 48 4.26R 24,-0,6 24,-0,7 4.04Lentiform and putamenL -20,7,0 -20,7,0 3.74R 10,-10,1 10,-10,1 3.55ThalamusL -10,-13,4 -10,-14,4 4.21R 17 20, -97, 4 20, -97, -7 5.36Occipital LobeL 18 -25,-97,4 -25,-94,8 4.94

Medial frontal gyrus L 32 -4, 9, 44 -4, 7, 48 4.92DLPFC R 46 49,31,15 49,31,18 4.99

28