HIPPOCAMPUS 13:164 –174 (2003)

Regional Analysis of Hippocampal Activation During

Memory Encoding and Retrieval: fMRI Study
Michael D. Greicius,1* Ben Krasnow,1
Jesse M. Boyett-Anderson,1 Stephan Eliez,1
Alan F. Schatzberg,1 Allan L. Reiss,1,2,3 and
Vinod Menon1,2,3

1
Department of Psychiatry & Behavioral Sciences,
Stanford University School of Medicine,
Stanford, California
2
Program in Neurosciences, Stanford University School
of Medicine, Stanford, California
3
Stanford Brain Research Center, Stanford University
School of Medicine, Stanford, California

ABSTRACT: Investigators have recently begun to examine the differen- the effects of susceptibility-induced signal loss on hip-
tial role of subregions of the hippocampus in episodic memory. Two pocampal activation and suggests that this artifact has
distinct models have gained prominence in the field. One model, outlined significantly biased the interpretation of earlier fMRI
by Moser and Moser (Hippocampus 1998;8:608 – 619), based mainly on studies. Hippocampus 2003;13:164 –174.
animal studies, has proposed that episodic memory is subserved by the © 2003 Wiley-Liss, Inc.
posterior two-thirds of the hippocampus alone. A second model, derived
by Lepage et al. (Hippocampus 1998;8:313–322) from their review of 52 KEY WORDS: hippocampus, functional imaging,
PET studies, has suggested that the anterior hippocampus is activated by memory, susceptibility artifact
memory encoding while the posterior hippocampus is activated by mem-
ory retrieval. Functional magnetic resonance imaging (fMRI) studies have
tended to show limited activation in the anteriormost regions of the
hippocampus, providing support for the Moser and Moser model. A
potential confounding factor in these fMRI studies, however, is that INTRODUCTION
susceptibility artifact may differentially reduce signal in the anterior
versus the posterior hippocampus. In the present study, we examined
activation differences between hippocampal subregions during encoding The hippocampus has been the focus of memory re-
and retrieval of words and interpreted our findings within the context of search since the seminal neuropsychological studies of
these two models. We also examined the extent to which susceptibility patient H.M., who suffered amnesia after a bilateral me-
artifact affects the analysis and interpretation of hippocampal activation
by demonstrating its differential effect on the anterior versus the posterior
dial temporal lobe (MTL) resection (Scoville and Milner,
hippocampus. Both voxel-by-voxel and region-of-interest analyses were 1957). More recently, investigators have begun to ex-
conducted, allowing us to quantify differences between the anterior and plore functional differences between different subregions
posterior aspects of the hippocampus. We detected significant hippocam- of the hippocampus. The elucidation of functional spe-
pal activation in both the encoding and retrieval conditions. Our data do
cialization within the hippocampus would be critical to a
not provide evidence for regional anatomic differences in activation
between encoding and retrieval. The data do suggest that, even after more thorough understanding of the neuroanatomic ba-
accounting for susceptibility artifact, both encoding and retrieval of ver- sis of memory. In addition, this information could have
bal stimuli activate the middle and posterior hippocampus more strongly significant clinical implications for disorders involving
than the anterior hippocampus. Finally, this study is the first to quantify hippocampal dysfunction (Small et al., 2000). Two sep-
arate models describing functional specialization be-
tween subregions of the hippocampus have gained prom-
Grant sponsor: National Institutes of Health; Grant number: MH50604; inence in the field.
Grant number: MH01142; Grant number: MH19938; Grant number: Moser and Moser (1998) developed a model, based
HD40761; Grant number: HD31715; Grant sponsor: The Kendall Fund;
Grant sponsor: The Sinclair Fund. mainly on animal studies, proposing that the anterior
*Correspondence to: Michael D. Greicius, Department of Psychiatry and third of the hippocampus is functionally distinct from
Behavioral Sciences, Stanford University School of Medicine, Stanford, CA the posterior two-thirds. This hypothesis is derived pri-
94305-5719. E-mail: [email protected] marily from observations that the afferent and efferent
Accepted for publication 12 March 2002
DOI 10.1002/hipo.10064
connections of the anterior (ventral) third of the hip-
pocampus are largely distinct from the connections of the
more posterior (dorsal) region. In particular, the anterior

© 2003 WILEY-LISS, INC.

 _______________________________________ REGIONAL ANALYSIS OF HIPPOCAMPAL ACTIVATION 165

third has robust efferent connections to the rostral hypothalamus and most other studies, may not have provided adequate spatial
and amygdala (Canteras and Swanson, 1992; Risold and Swanson, resolution (10 mm in-plane) to examine subregions within the
1996). Furthermore, studies of spatial learning in rats (E.I. Moser hippocampus, a structure that extends ⬃40 mm along its ante-
et al., 1993; M-B. Moser et al., 1995) and monkeys (Colombo et rior-posterior axis (Duvernoy, 1998).
al., 1998) have suggested that the anterior third of the hippocam- Thus, a great deal of uncertainty remains regarding func-
pus may not be required for visuospatial memory. Moser and tional distinctions within the hippocampus proper. One poten-
Moser then adapted this model to the human hippocampus, citing tial source of discrepant findings between fMRI studies as well
several functional magnetic resonance imaging (fMRI) studies that as between fMRI and PET studies pertains to susceptibility
detected predominantly posterior hippocampal activation during artifact (Veltman et al., 2000). This artifact, found only in
verbal (Fernandez et al., 1998) and visual (Stern et al., 1996; Rom- fMRI, results from abrupt changes in magnetic susceptibility
bouts et al., 1997) memory tasks. In sum, the Moser and Moser that occur across tissue interfaces such as the border between
model proposes that the anterior third of the hippocampus is not air-filled sinuses and brain parenchyma or between bone and
integral to episodic memory. brain parenchyma. Brain regions closest to such borders are
The second model, based on a meta-analysis of 52 PET studies by especially susceptible to loss of blood oxygen level-dependent
Lepage et al. (1998), holds that the anterior hippocampus is activated (BOLD) signal due to this artifact. Ojemann et al. (1997) dem-
by encoding and the posterior hippocampus by retrieval. Lepage’s onstrated that signal attenuation was most prominent in the
hippocampus in encoding and retrieval (HIPER) model was derived inferior frontal and inferolateral temporal regions. On the basis
by comparing the activation foci of 22 encoding studies and 32 re- of these findings, Schacter and Wagner (1999) suggested that
trieval studies on a sagittal hippocampal slice from the Talairach and reduced anterior hippocampal activation in fMRI studies was
Tournoux atlas (1988). These investigators concluded that the phys- not due to susceptibility artifact. It is important to note, how-
iological basis for such a distinction was unclear and encouraged fur- ever, that Ojemann et al. (1997) specified that signal loss was
ther studies that could prospectively test the model. The HIPER “relative rather than absolute” and decreased with distance away
model was refuted by Schacter and Wagner (1999) in their review of from air– brain or bone– brain borders, so that other temporal
the functional imaging literature. In this updated meta-analysis, lobe regions may still show signal attenuation, albeit less prom-
positron emission tomography (PET) studies tended to show anterior inent. Ojemann et al. (1997) also cautioned that “it remains
and posterior activation with encoding and predominantly posterior unknown exactly how much a local BOLD activation signal
activation with retrieval, whereas fMRI studies showed a predomi- would be affected by a superimposed macroscopic signal loss.”
nance of posterior MTL activation for encoding. However, there were Lipschutz et al. (2000) demonstrated that “sensitivity to BOLD
insufficient fMRI data on hippocampal activation during retrieval to effects is directly proportional to signal intensity,” suggesting
detect a distinct anatomic pattern. Schacter and Wagner (1999) con- that BOLD signal detection can be reduced due to susceptibil-
cluded that the limited anterior activation in the MRI studies was due ity artifact, even in the absence of complete signal dropout.
to differences in task design between the PET and fMRI studies. In Two recent studies have extended the work of Ojemann et al.
particular, they posited that tasks involving relational processing such to show that signal loss due to susceptibility artifact can have
as word pairs (Dolan and Fletcher, 1999) or pictures of people com- significant effects on MTL activation. Devlin et al. (2000) com-
bined with pictures of housing (Henke et al., 1997) might account for pared PET and fMRI directly, using a nearly identical semantic
the anterior hippocampal activation seen with encoding in some PET task, and found that anteromedial temporal activation was de-
studies. A second criticism leveled at the HIPER model was that most tected with PET, but not with fMRI. Cordes et al. (2000)
studies examined either encoding or retrieval, while few used a within- showed susceptibility-induced signal loss in the parahippocam-
subjects design to compare encoding and retrieval directly. pal and amygdala regions while a subject mentally rehearsed a
Schacter et al. (1999) subsequently carried out a PET study gymnastics routine. Because the hippocampus rises superiorly
to test the HIPER model prospectively. Efforts were made to from anterior to posterior, one would expect greater suscepti-
control for potential confounding factors by using a within- bility-induced signal loss in the anterior (inferior) relative to the
subjects design and nonrelational stimuli. The analysis was not posterior (superior) hippocampus. It is plausible therefore that
limited to the hippocampus but showed predominantly poste- the relative lack of anterior hippocampal activation and the
rior MTL activation, including posterior hippocampus, para- preponderance of posterior hippocampal activation in fMRI
hippocampal gyrus, and fusiform gyrus for both encoding and studies of memory is due to this pervasive artifact. No study to
retrieval. Direct comparisons of encoding and retrieval also date, however, has directly examined the effects of susceptibility
were carried out. The contrast comparing encoding versus re- artifact on hippocampal activation.
trieval showed posterior MTL activation, while the opposite In this study, we investigated regional differences in fMRI acti-
comparison, retrieval versus encoding, showed no MTL activa- vation of the hippocampus during encoding and retrieval. In par-
tion. Although this study attempted to clarify the issue of ana- ticular, data were examined within the context of the Moser and
tomic localization of MTL activation associated with episodic Moser and Lepage models discussed above. This study includes
memory, the design did not include a comparison of activation several methodological improvements in comparison with previ-
between anatomically specified hippocampal subregions. Fur- ous studies. A within-subjects design was used to examine encod-
thermore, the group-averaged, voxel-by-voxel analysis em- ing and retrieval of nonrelational stimuli. The same control con-
ployed might eliminate MTL activation occurring in voxels that dition was used across the encoding and retrieval experiments, to
only a few subjects activated. In addition, the use of PET, in this facilitate a direct comparison. In addition to the standard voxel-
166 GREICIUS ET AL.

by-voxel analysis, we report a more specific, quantitative region- After encoding, subjects performed a distractor task that lasted
of-interest (ROI) analysis focusing on statistical comparisons be- 5–10 min before beginning the retrieval task. The retrieval task
tween hippocampal subregions. We have previously used a similar consisted of six retrieval epochs alternating with six control epochs.
approach to investigate regional differences in novelty, memory, In the retrieval epochs, subjects again saw a single word for 2.5 s
and spatial processing within MTL structures (Menon et al., and were asked to press one of two buttons, depending on whether
2000). To assess Lepage’s HIPER model, in which PET activations the word had been seen in the encoding task or not. Sixteen “new”
of encoding mapped to the anterior half of the hippocampus and words and 32 “old” words were intermixed across the six retrieval
retrieval activation to the posterior half, we divided our hippocam- epochs. Each of the six retrieval epochs was followed by a control
pal ROIs into anterior and posterior halves. To assess the Moser epoch in which either of two words was presented in a random
and Moser model, positing a functional difference between the order over the eight stimulus periods. The same two words used in
anterior third and the posterior two-thirds of the hippocampus, we the five encoding control epochs were used for all six control ep-
divided our hippocampal ROIs into anterior, middle, and poste- ochs in the retrieval task, so that by the end of the study, these two
rior thirds. Finally, and perhaps most importantly, we include an words had been seen 44 times each. During the control epoch, the
analysis of susceptibility artifact and its effect on BOLD signal subjects were instructed to alternate pressing buttons 1 and 2 with-
detection in the hippocampus. out making a recognition judgment. As in the encoding task, each
stimulus was presented for 2.5 s, followed by a 0.5-s interstimulus
interval; each epoch was preceded by an instruction screen shown
for 4 s, and a 24-s rest period with fixation occurred before the
MATERIALS AND METHODS onset of the first cycle, at the midway point (after the third control
epoch in this case), and after the last control epoch. The task order
for retrieval can be abbreviated as F-R-C-R-C-R-C-F-R-C-R-C-
Subjects
R-C-F, again, where F is fixation, R is retrieval, and C is control.
Fourteen healthy, right-handed subjects (six males and eight fe- All 12 subjects performed the identical retrieval task.
males, aged 18– 48 years) participated in the study after giving in- The first five subjects underwent a slightly modified version of
formed consent. Two subjects were eliminated before fMRI analysis the encoding protocol that differed only in two ways: (1) the mid-
because their behavioral data indicated that they performed at or be- dle fixation epoch was placed after the third control epoch, rather
low chance on either the encoding or the retrieval task. Analysis of than before it (F-E-C-E-C-E-C-F-E-C-E-C-F); and (2) they were
fMRI data was performed on the remaining 12 subjects (five males not shown an instruction screen at the beginning of each epoch,
and seven females, aged 18– 48 years, mean age 26 ⫾ 8). but received verbal instructions at the beginning of the protocol.

Stimuli Stimulus Presentation

The stimuli for the encoding condition were 40 unique visually The tasks were programmed using Psyscope (http://poppy.psy-
presented nouns. In the retrieval condition, the stimuli were 32 of .cmu.edu/psyscope) on a Macintosh (Sunnyvale, CA) notebook
the same words plus 16 new words. The stimuli for the control computer. Onset of scanning and task were synchronized using a
condition consisted of the same 2 nouns alternating repeatedly. TTL pulse delivered to the scanner timing microprocessor board
from a CMU Button Box microprocessor connected to the Macin-
Design and Procedure tosh with a serial cable. Stimuli were presented visually at the
center of a screen using a custom-built magnet compatible projec-
The encoding task consisted of epochs of eight words each pre-
tion system (Resonance Technology, CA).
sented independently for 2.5 s, with a 0.5-s interstimulus interval.
Each of the five encoding epochs was followed by a control epoch
fMRI Acquisition
in which either of two words was presented in a random order over
the eight stimulus periods. The same two words were used for all Images were acquired on a 3-tesla (T) GE Signa scanner using a
five control epochs. Each epoch was preceded by an instruction standard GE whole head coil. The scanner runs on an LX platform,
screen shown for 4 s. In addition, a 24-s rest period with fixation with gradients in Mini-CRM configuration (35 mT/m, SR 200
occurred before the onset of the first cycle, at the midway point mT/m/s), and has a Magnex 3-T 80-cm magnet. A custom-built head
(after the third encoding epoch), and after the last control epoch. holder was used to prevent head movement; 28 axial slices (4 mm
The task order can be abbreviated as F-E-C-E-C-E-F-C-E-C-E- thick, 0.5 mm skip) parallel to the ACPC line and covering the whole
C-F, where F is fixation, E is encoding, and C is control. To brain were imaged with a temporal resolution of 2 s, using a T2*-
enhance encoding (Craik et al., 1994), subjects were instructed to weighted gradient echo spiral pulse sequence (TR ⫽ 2,000 ms, TE ⫽
press one of two buttons to make a man-made/not man-made 30 ms, flip angle ⫽ 89° and 1 interleave) (Glover and Lai, 1998). The
semantic discrimination for each word in the encoding epoch. The field of view was 200 ⫻ 200 mm2, and the matrix size was 64 ⫻ 64,
words were equally divided between the two semantic categories. giving an in-plane spatial resolution of 3.125 mm. To reduce blurring
The subjects were also instructed to remember the words as they and signal loss arising from field inhomogeneities, a shimming proto-
would be asked to identify them later. During the control epoch, col was used before acquiring functional MRI scans (Spielman et al.,
they were instructed to alternate pressing buttons 1 and 2 without 1998). The protocol uses a short spiral acquisition to obtain the field
making a semantic discrimination. maps and downloads the resistive shims automatically. To aid in lo-
 _______________________________________ REGIONAL ANALYSIS OF HIPPOCAMPAL ACTIVATION 167

calization of functional data, a high-resolution T1-weighted spoiled dition compared with the control condition was determined, using
grass gradient recalled (SPGR) 3D MRI sequence with the following multivariate regression analysis with correction for temporal autocor-
parameters was used: TR ⫽ 35 ms; TE ⫽ 6 ms; flip angle ⫽ 45°; relations in the fMRI data (Friston et al., 1995). Confounding effects
24-cm field of view; 124 slices in the sagittal plane; 256 ⫻ 192 matrix; of fluctuations in global mean were removed by proportional scaling,
acquired resolution ⫽ 1.5 ⫻ 0.9 ⫻ 1.2 mm. The images were recon- and low-frequency noise was removed with a high-pass filter (0.5
structed as a 124 ⫻ 256 ⫻ 256 matrix with a 1.5 ⫻ 0.9 ⫻ 0.9-mm cpm). A regressor waveform for each condition, convolved with a 6-s
spatial resolution. Structural and functional images were acquired in delay Poisson function accounting for delay and dispersion in the
the same scan session. hemodynamic response, was used to compute voxel-wise t-statistics,
which were then normalized to z-scores to provide a statistical measure
Region of Interest of activation independent of sample size. Using random-effects anal-
ysis (Holmes and Friston, 1998; Friston et al., 1999a), brain activation
Our hippocampal ROI was based on the protocol outlined by
for each of the experimental conditions, contrasted with the control
Kates et al. (1997). Each subject’s structural MRI was first normal-
condition, was determined: (1) encoding, (2) retrieval. These activa-
ized in standard Talairach space. Using the normalized structural
tion images were then directly compared. Given the hypothesis-driven
image allowed for a more reliable comparison with the normalized
nature of our study, we used a threshold of z ⬎ 1.67 (P ⬍ 0.05)
functional data minimizing, for example, the effects of changes in
without spatial correction (i.e., no minimal cluster size) to identify
head position between the structural and functional scans. The
significantly activated voxels. The right and left hippocampal ROIs
hippocampus was delineated on coronal slices strictly perpendicu-
from the group brain were used as a mask on the group-averaged
lar to the anterior commissure–posterior commissure (AC–PC)
activation image, generating a map of activation within the hippocam-
axis. The anterior slice of the hippocampus is most easily defined
pus only. This masked activation image was then superimposed on the
by the superior shift of the temporal horn of the lateral ventricles to
group-averaged structural image.
the point at which the temporal horn turns and points superome-
For the HIPER model, we calculated the percentage of voxels
dially and is positioned so that the amygdala is located superior and
activated (P ⬍ 0.05 without spatial correction) in each individual’s
the hippocampus inferior to it. Posteriorly, the hippocampus fuses
four hippocampal ROIs. While both the percentage of voxels ac-
with the fornix and is measured until it is no longer visible. The
tivated (Gabrieli et al., 1997) and the absolute number of voxels
medial border is defined by the ambient cistern. The inferior bor-
activated (Small et al., 2001) have been used as measures in hip-
der is marked by the collateral white matter, the subiculum, and
pocampal activation studies, a statistical comparison of the two
the parahippocampal gyrus. The lateral border is marked by the
ROI techniques (Constable et al., 1998; Fig. 6, p 297) has shown
temporal horn of the lateral ventricles and more superiorly by the
that measuring the percentage of voxels activated is the more ro-
white matter. To test the HIPER model, each individual’s hip-
bust approach. These data were then entered into an analysis of
pocampal ROIs were then split into anterior and posterior hip-
variance (ANOVA) to examine the effects of three factors on hip-
pocampal ROIs. This was done by dividing the full ROI at its
pocampal activation: the three-way ANOVA examined task (en-
midpoint along the anterior-posterior axis. To test the Moser and
coding/retrieval) ⫻ anterior-posterior (A/P) location ⫻ hemi-
Moser model, each full hippocampal ROI was divided into thirds
sphere (left/right) interactions. A similar analysis was performed
along the anterior-posterior axis. Thus, in the ROI analysis of the
with the hippocampal ROIs divided into thirds (anterior, middle,
HIPER model each individual subject had four hippocampal
and posterior), to examine the Moser and Moser model.
ROIs: left anterior, right anterior, left posterior, and right poste-
rior. For the Moser and Moser model analysis, each individual had
six hippocampal ROIs: left anterior, left middle, left posterior, Artifact Quantification
right anterior, right middle, and right posterior.
To quantify the amount of signal loss due to susceptibility arti-
In addition to the individual ROIs, we used the same protocol to
fact, we first calculated the average voxel intensity for each individ-
draw bilateral hippocampal ROIs on a normalized brain averaged
ual’s four ROIs from a T2* image averaged over the fixation epochs
from all 12 subjects. The left hippocampal ROI had an anterior-
in the retrieval paradigm. We chose the fixation epochs to ensure
posterior extent of 32 mm beginning at a Talairach y-coordinate of
that differences in voxel intensity were not due to activation during
⫺8 and ending at ⫺40 with a midpoint of ⫺24. The right hip-
the experimental task, but rather to baseline differences in T2*
pocampal ROI had an anterior-posterior extent of 34 mm, begin-
signal. These average voxel intensities were used to examine A/P
ning at ⫺6 and extending to ⫺40 with a midpoint of ⫺23.
location by Hemisphere interactions among all 12 subjects using a
two-way ANOVA. Then, for each subject, we graphed the inten-
Data Analysis
sity of every hippocampal voxel against its subsequent activation
fMRI data from each subject were analyzed using Statistical Para- t-score taken over the averaged retrieval epochs (Fig. 1). Each sub-
metric Mapping (SPM99) (http://www.fil.ion.bpmf.ac.uk/spm). Be- ject had an individual voxel intensity threshold (ranging across the
fore statistical analysis, images were corrected for movement using group from 432 to 652, arbitrary scale), below which there was no
least-squares minimization without higher-order corrections for spin activation but above which all voxels were equally likely to be
history, normalized to stereotactic Talairach coordinates (Talairach activated (i.e., there was no correlation between voxel value and
and Tournoux, 1988), resampled every 2 mm using sinc interpolation t-score). We refer to this threshold as intensity threshold for detec-
and smoothed with a 4-mm Gaussian kernel to eliminate spatial noise. tion of activation (ITDA). The ITDA results from a threshold
For each subject, voxel-wise activation during each experimental con- masking procedure applied by SPM to remove “non-brain” voxels.
168 GREICIUS ET AL.

In this procedure, all voxels below a given T2* signal intensity are
assigned a t-score of 0 in the subsequent statistical analysis. The
threshold masking function is determined using each individual’s
mean T2* signal intensity (rather than a fixed T2* intensity across
all individuals), accounting for the variance in the ITDA values
across individuals. It should be noted that threshold masking is a
widely used procedure to remove “non-brain” voxels, employed
not only by SPM, but by alternative fMRI software programs, such
as analysis of functional neuroimages (AFNI), as well.
We then compared the percentage of voxels exceeding the
ITDA in each of the four ROIs, using another two-way
ANOVA examining A/P location by hemisphere interactions.
Finally, to determine the effect of signal loss on the percentage
of voxels activated, we performed the ROI analyses again, this
time restricting them to voxels that exceeded the ITDA calcu-
lated for each individual.

FIGURE 1. Graphic representation of hippocampal voxel inten-

sities from a single subject during rest on the x-axis against the t-score RESULTS
of each voxel during the retrieval task on the y-axis. Voxels in the
anterior hippocampus are shown in blue. Voxels in the posterior
hippocampus are shown in red. Below the intensity threshold to de- Behavioral Task Performance
tect activation (ITDA) of 501 (shown with arrow and vertical bar), no
signal modulation is detected because these voxels are removed from Behavioral data demonstrated that 12 of the 14 subjects per-
the statistical analysis. In this subject (and in 8 of 11 others), 100% of
formed both the encoding and retrieval tasks correctly (i.e., better
the posterior hippocampal voxels were above the ITDA, while only
1022/1221(83.7%) of the anterior hippocampal voxels were above than chance). One subject responded correctly to only 45% of the
the ITDA in this subject.

FIGURE 2. Group activation during memory encoding (n ⴝ 12, P < 0.05) is shown within a hippocampal region of interest (ROI).
Coronal slices from a Talairach y-coordinate of 0 to y ⴝ – 45 are shown. Images are arranged in anterior to posterior order, from upper left to
bottom right. The midpoint of the hippocampal ROI is at y ⴝ –24. The right side of the image corresponds to the right side of the brain.
 _______________________________________ REGIONAL ANALYSIS OF HIPPOCAMPAL ACTIVATION 169

FIGURE 3. Group activation during memory retrieval (n ⴝ 12, P < 0.05). See Fig. 2 for further details.

encoding judgments. A separate subject responded correctly to

only 42% of the retrieval judgments. Both subjects were excluded
from fMRI analysis, as we could not be certain that they were
performing the encoding and retrieval tasks correctly. The remain-
ing 12 subjects had accuracy rates on the encoding semantic judg-
ment (man-made or not man-made) ranging from 87.5% to
100%, with a mean of 95% and an SD of 4.1%. In the retrieval
task, accuracy rates on the “new” versus “old” judgment (combin-
ing correctly identified old words and correctly rejected foils)
ranged from 79.2% to 91.6% correct, with a mean of 85.4% and
an SD of 4.7%. There were no significant differences in the encod-
ing or retrieval task performances between the five subjects who
underwent the slightly modified encoding protocol (see Materials
and Methods) and the remaining seven subjects.

Hippocampal Activation
Figures 2 and 3 show group hippocampal activation for two
comparisons: encoding versus the control condition and retrieval
versus the control condition. During encoding, activation was seen
bilaterally both anterior and posterior to the midpoint of the hip- FIGURE 4. T1-weighted structural image for a single subject is
pocampus (y ⫽ –23), although the bulk of the activated voxels shown with the hippocampal ROI superimposed on it. Voxels with
appears to be at or just posterior to the midpoint (Fig. 2). During significant susceptibility-induced signal loss (T2* signals that were
retrieval, activation was observed mainly on the right side, both below the intensity threshold to detect activation [ITDA]) are shown
anterior and posterior to the midpoint with a slight posterior pre- in red. Note that in this subject (and in 8 of 11 other subjects), such
voxels were found only in the anterior half of the hippocampus.
dominance (Fig. 3). Within the anterior hippocampus, such voxels were located inferiorly,
We then performed two additional group comparisons: encod- closest to the skull base and sinuses, further suggesting that this signal
ing versus retrieval and retrieval versus encoding. The encoding loss is caused by susceptibility artifact.
170 GREICIUS ET AL.

TABLE 1.

Results of Three-Way ANOVA With Factors Task (Encoding, Retrieval), A/P Location (Anterior Half, Posterior Half), and
Hemisphere (Left, Right)*

Uncorrected ROIs Corrected ROIs

Effect DF F P DF F P

Task 1,11 0.05 0.831 1,11 0.03 0.873

A/P location 1,11 2.02 0.183 1,11 1.54 0.240
Hemisphere 1,11 2.42 0.148 1,11 2.09 0.176
Task ⫻ A/P location 1,11 0.1 0.758 1,11 0.03 0.856
Task ⫻ hemisphere 1,11 0.54 0.479 1,11 0.46 0.512
A/P location ⫻ hemisphere 1,11 1.39 0.263 1,11 1.63 0.228
Task ⫻ A/P location ⫻ hemisphere 1,11 0.01 0.911 1,11 0.03 0.876

ANOVA, analysis of variance; A/P, anterior-posterior; ROI, region of interest.

*ANOVA was run with uncorrected ROIs and then repeated with ROIs corrected for susceptibility-induced signal loss.

versus retrieval comparison activated only 29 hippocampal voxels ROI Analysis

(of 909 total voxels in the bilateral hippocampi) compared with
300 voxels which were activated when the encoding and control HIPER model
conditions were compared. These 29 voxels were spread diffusely
across the anterior and posterior regions of the hippocampus, The HIPER model was tested by dividing each hippocampal
mainly in the left hemisphere. The retrieval versus encoding com- ROI in half. For each of the four ROIs (left anterior, right anterior,
parison activated only six voxels compared with the 106 voxels, left posterior, right posterior), we calculated the percentage of ac-
which were activated when the retrieval and control conditions tivated voxels. These percentages were then used as the dependent
were compared. These six voxels were near the midpoint of the variable in a three-way ANOVA that examined task ⫻ A/P loca-
hippocampus in the right hemisphere. tion ⫻ hemisphere interactions. There were no main effects or
interactions (Table 1). When the analysis was limited to artifact-
Artifact Quantification free voxels whose intensity exceeded the ITDA, there were still no
main effects nor any interactions (Table 1).
Figure 4 illustrates the effect of susceptibility-induced voxel
intensity attenuation on BOLD signal detection in the hip-
Moser and Moser model
pocampus. To quantify differences in voxel intensity between
hippocampal subregions, we performed a two-way ANOVA The Moser and Moser model was tested by dividing each hip-
examining the effects of A/P location and hemisphere on the pocampal ROI into thirds. We then performed a three-way
average voxel intensity (during fixation) of each of four ROIs ANOVA (task ⫻ A/P location ⫻ hemisphere) using three A/P
(left anterior, right anterior, left posterior, right posterior). This locations (anterior, middle, and posterior). There were no main
analysis showed a significant main effect of A/P location (F effects or interactions among the three factors (Table 2). When the
(1,11) ⫽ 28.89, P ⫽ 0.0002)—the anterior ROIs had signifi- analysis was limited to artifact-free voxels whose intensity exceeded
cantly lower average voxel intensities than the posterior ROIs. the ITDA, the ANOVA still did not yield any significant main
The main effect of hemisphere showed a trend towards signifi- effects or interactions (Table 2).
cance (F (1,11) ⫽ 4.73, P ⫽ 0.052), in which the left hip- Based on our voxel-by-voxel analysis, which demonstrated a
pocampus had lower average voxel intensities than the right. preponderance of activation in the middle third of the hip-
There was no significant interaction between A/P location and pocampus, moderate activation in the posterior third, and low-
hemisphere (F (1,11) ⫽ 0.39, P ⫽ 0.55). est activation in the anterior third, we hypothesized that acti-
We then examined the effect of A/P location and hemisphere vation in the posterior third might be diluting a significant
on our ability to detect BOLD activation. We performed a difference in activation between the anterior and middle thirds
similar two-way ANOVA (A/P location ⫻ hemisphere), using when all three A/P locations were combined in the ANOVA.
the percentage of voxels above the ITDA as the dependent We therefore performed a post-hoc three-way ANOVA (task ⫻
variable (see Materials and Methods; see also Fig. 1). There was A/P location ⫻ hemisphere) comparing only the anterior and
a main effect of A/P location (F (1,11) ⫽ 10.31, P ⫽ 0.008). middle thirds. This analysis showed a significant main effect for
The posterior ROIs had significantly more voxels exceeding the A/P location, where the middle third had 21% of voxels acti-
ITDA than were found for the anterior ROIs. We did not detect vated and the anterior third had 16% of voxels activated (F
a main effect of hemisphere (F (1,11) ⫽ 1.95, P ⫽ 0.19) or an (1,11) ⫽ 7.73, P ⫽ 0.018, ␩2 ⫽ 0.89). There were no other
interaction between A/P location and hemisphere (F (1,11) ⫽ main effects and no interactions (Table 3). When the ANOVA
2.05, P ⫽ 0.18). was limited to artifact-free voxels whose intensity exceeded the
 _______________________________________ REGIONAL ANALYSIS OF HIPPOCAMPAL ACTIVATION 171

TABLE 2.

Results of Three-Way ANOVA With Factors Task (Encoding, Retrieval), A/P Location (Anterior Third, Middle Third, and
Posterior Third), and Hemisphere (Left, Right)

Uncorrected ROIs Corrected ROIs

Effect DF F P DF F P

Task 1,11 0.03 0.86 1,11 0.01 0.91

A/P location 2,22 2.06 0.15 2,22 1.44 0.26
Hemisphere 1,11 1.84 0.2 1,11 1.6 0.23
Task ⫻ A/P location 2,22 0.79 0.47 2,22 0.77 0.48
Task ⫻ hemisphere 1,11 0.46 0.51 1,11 0.44 0.52
A/P location ⫻ hemisphere 2,22 2.29 0.12 2,22 2.4 0.11
Task ⫻ A/P location ⫻ hemisphere 2,22 0.33 0.72 2,22 0.26 0.77

ANOVA, analysis of variance; A/P, anterior-posterior; ROI, region of interest.

*The ANOVA was run with uncorrected ROIs and then repeated with ROIs corrected for susceptibility-induced signal loss.

ITDA, the percentage of activated voxels in the anterior third model proposed by Lepage and colleagues. This model was sup-
increased slightly to 17%, while the percentage of activated ported by neither voxel-by-voxel analysis nor ROI analysis (per-
voxels in the middle third remained essentially unchanged at formed both with and without a correction for susceptibility arti-
21.2%. The main effect of A/P location was reduced to a non- fact). The findings offer minimal support for the Moser and Moser
significant trend (F (1,11) ⫽ 4.67, P ⫽ 0.054, ␩2 ⫽ 0.82). model in that the episodic memory tasks activated the posterior
There were no other main effects or interactions (Table 3). two-thirds of the hippocampus to a slightly greater extent than the
anterior third. Finally, and perhaps most importantly, our data
suggest that susceptibility artifact has played a significant, con-
founding role in the interpretation of previous hippocampal acti-
DISCUSSION vation studies.
In testing the HIPER model, we followed guidelines, suggested
Our study demonstrates that activation occurs across the full by Schacter et al (1999), for minimizing potential confounds in
anterior-posterior extent of the hippocampus (defined in this study tasks designed to compare encoding and retrieval. In particular, we
as the hippocampal formation distinct from entorhinal and para- used a within-subjects design, nonrelational stimuli, and an anal-
hippocampal cortices) for both encoding and retrieval tasks. The ysis that included direct comparisons of encoding and retrieval.
results do not support an anterior-posterior activation differential Our results showed greater anterior hippocampal activation during
for encoding versus retrieval and thus, fail to support the HIPER both encoding and retrieval than was found in the Schacter PET

TABLE 3.

Results of Three-way ANOVA With Factors Task (Encoding, Retrieval), A/P Location (Anterior Third, Middle Third), and
Hemisphere (Left, Right)

Uncorrected ROIs Corrected ROIs

Effect DF F P DF F P

Task 1,11 0.01 0.933 1,11 ⬍0.01 0.984

A/P location 1,11 7.73 0.018* 1,11 4.67 0.054
Hemisphere 1,11 0.37 0.554 1,11 0.24 0.634
Task ⫻ A/P location 1,11 2.1 0.175 1,11 1.78 0.21
Task ⫻ hemisphere 1,11 0.17 0.687 1,11 0.16 0.696
A/P location ⫻ hemisphere 1,11 0.01 0.927 1,11 ⬍0.01 0.949
Task ⫻ A/P location ⫻ hemisphere 1,11 0.15 0.704 1,11 0.04 0.704

ANOVA, analysis of variance; A/P, anterior-posterior; ROI, region of interest.

†
The ANOVA was run with uncorrected ROIs and then repeated with ROIs corrected for susceptibility-induced signal loss.
*significant at P ⬍ 0.05.
172 GREICIUS ET AL.

study. One possible explanation for this difference is that the voxels in the left hippocampus. This type of individual variability
Schacter study used visual stimuli instead of verbal stimuli. An- in activation emphasizes the importance of using an ROI analysis
other possibility is that the limited spatial resolution of PET may to complement the voxel-by-voxel analysis (Montaldi et al., 1998).
make fMRI better suited for more precise, anatomical dissections A potential confounding factor in using a block design to inves-
of regional hippocampal activation. Our fMRI study and the tigate retrieval processes involves the inclusion of new words in the
Schacter PET study constitute two significant challenges to the retrieval blocks to maintain some trial-to-trial variability. In an
HIPER model and demonstrate the importance of a priori testing event-related fMRI study, Buckner et al. (2001) recently provided
of hypotheses generated a posteriori from a meta-analysis. evidence of incidental encoding of new words during a recognition
Our data are less conclusive with respect to the Moser and test, raising the concern that block designs of retrieval are subject to
Moser model. We were clearly able to detect activation in the most an incidental encoding confound. This issue was explicitly ad-
anterior aspect of the hippocampus during each task, suggesting dressed by Stark and Squire (2000) in a study examining retrieval
that this part of the hippocampus is integral to episodic encoding of nameable objects compared with retrieval of words. In that
and retrieval of verbal material. However, visual inspection of the study, novel words (used as foils in an old-new recognition para-
voxel-by-voxel T maps (Figs. 2, 3) did suggest a qualitative differ- digm) resulted in hippocampal deactivation below baseline. Novel
ence in the degree of activation between the anterior and middle objects showed less hippocampal deactivation than words (but did
third of the hippocampus. This impression was not borne out by not show hippocampal activation) and thus tended to blunt hip-
the ROI analysis when all three thirds (anterior, middle and pos- pocampal activation in the old versus new contrast. The investiga-
terior) were compared by ANOVA; however a post-hoc ANOVA tors concluded that novelty and incidental encoding effects are
comparing the anterior to the middle third indicated marginally diminished with verbal stimuli owing to the “high preexperimental
significant differences with less activation in the anterior third. familiarity” of words (Stark and Squire, 2000). A study conducted
There are a few caveats to this finding. First, this result may have by Kelley et al. (1998) showed a similar discrepancy between the
been subject to the bias of multiple comparisons, because we com- incidental encoding effects of novel words compared with novel
pared two factors of an ANOVA directly despite no difference in objects during passive viewing. In this regard, it is worth noting
the initial three-way comparison. We justified this post-hoc com- that the paper by Buckner et al. (2001) showed differential activa-
parison based on visual inspection of the voxel-by-voxel analysis, tion during incidental encoding in a number of regions but not in
which suggested a difference in the amount of activation between the hippocampus or MTL. Thus, while hippocampal activation to
the anterior and middle thirds. Second, the size of the difference incidental encoding during a retrieval block remains a theoretical
between activation in the anterior and middle thirds was modest concern, it has not been demonstrated in either of the two fMRI
(16% vs 21%, P ⫽ 0.018, ␩2 ⫽ 0.89). Finally, this difference was studies that have addressed the possibility (Buckner et al, 2001;
reduced to a statistical trend when we corrected our ROI analysis Stark and Squire 2000). We concur, therefore, with Stark and
for susceptibility artifact. If our data support the Moser and Moser Squire (2000) that the relative gain in retrieval activation offered
model, they suggest the difference between the anteriormost part by the block design compared with an event-related design (Fris-
of the hippocampus and the posterior aspects is a matter of degree ton et al., 1999b) offsets the potential confound of incidental
rather than kind. Moser and Moser considered this as a possibility encoding, particularly as pertains to verbal stimuli and the hip-
in stating “We do not know whether the differentiation of the pocampus.
hippocampus along the longitudinal axis is graded or discontinu- A second potential confound in our encoding task is that the
ous”(p. 613). They favored a discontinuous model of hippocampal experimental epochs involve both an intentional encoding compo-
differentiation but our data are more supportive of a graded model. nent and a semantic component (required for the man-made ver-
We also hasten to point out that the bulk of the Moser and Moser sus not man-made judgment). To facilitate a comparison between
review pertains to visuospatial memory and we cannot generalize the encoding and retrieval tasks, we decided to use an identical
beyond verbal memory from our results. baseline across the tasks. The baseline epochs did not require se-
One finding we did not anticipate arose from the voxel-by-voxel mantic processing; thus, subtracting it from our encoding epochs
analysis, which showed almost no activation in the left hippocam- did not account for the semantic processing. Of the several event-
pus during retrieval in the block design analysis. This asymmetry related studies comparing subsequently remembered versus forgot-
was not detected in any of the ANOVA performed in our ROI ten words, one has shown (Otten et al., 2001) and one has sug-
analysis where we would have expected a task ⫻ hemisphere inter- gested (Kirchhoff et al., 2000) that successful encoding, distinct
action (Tables 1–3). The discrepancy between the voxel-by-voxel from semantic processing, activates the hippocampus. The con-
and ROI analyses was likely attributable to the fact that the ROI verse—that semantic processing distinct from encoding, activates
analysis examines activation of individual voxels on a subject-by- the hippocampus— has not been demonstrated. The question of
subject basis without regard to whether each subject of a group whether semantic processing alone activates the hippocampus
activated exactly the same voxel as other subjects in the group. In could be addressed in an event-related study in which the semantic
contrast, the voxel-by-voxel analysis is more stringent in that it processing versus nonsemantic processing contrast was restricted
highlights voxels that were activated in common across all or most to words that were subsequently forgotten. To our knowledge,
of the subjects. Thus, the most likely explanation is that in each such an analysis has not been reported. Based on the literature and
subject the left hippocampus was as involved in retrieval of verbal given that our analysis was limited to the hippocampus, we hold
material as the right hippocampus, but that across subjects there that the hippocampal activation in our encoding task is mainly a
happened to be greater anatomic variability in location of activated reflection of encoding, though we cannot dismiss the possibility
 _______________________________________ REGIONAL ANALYSIS OF HIPPOCAMPAL ACTIVATION 173

that semantic processing also contributed to hippocampal activa- was replicated, this study further obscured the central issue of
tion. susceptibility artifact which increases rather than decreases with
Our data strongly suggest that susceptibility artifact has been a increasing magnetic field strength (Abduljalil and Robitaille,
potential confound in fMRI studies of regional hippocampal acti- 1999).
vation. We attempted to minimize this artifact by using a specially In conclusion, our data suggest that the hippocampus functions
designed shim technique (Spielman et al., 1998) and a gradient as a rather homogeneous unit during encoding and retrieval of
echo spiral pulse sequence rather than a traditional echo planar episodic verbal memories. The relatively decreased anterior activa-
sequence (G. Glover, personal communication). Given the relative tion, apparent in our voxel-by-voxel analysis, was less impressive in
paucity of fMRI studies showing activation in the anterior hip- the ROI analyses where we were able to account for some of the
pocampus, the presence of any activation in this region suggests we effects of susceptibility artifact. At the most, our data show a small
have succeeded in reducing some artifactual signal loss. Nonethe- difference in degree, not kind, between activation in the anterior-
less, the averaged resting T2* voxel intensity in our anterior hip- most aspect of the hippocampus compared with the posterior as-
pocampal ROIs was significantly less than in our posterior hip- pects. Our study does not rule out the possibility that encoding and
pocampal ROIs. More importantly, voxel intensity decreases were retrieval of visuospatial material preferentially activate the poste-
substantial enough to impair the detectability of BOLD signal rior hippocampus. This possibility is consistent with the Moser
changes. We found that a greater percentage of voxels in the ante- and Moser model, the fMRI study by Stern et al. (1996), and the
rior hippocampal ROIs fell below the ITDA where BOLD effects PET study by Schacter et al. (1999), all of which focused on visual
could not be detected. After adjusting ROIs for signal loss, the memory and showed predominantly posterior hippocampal acti-
apparent functional distinction between the anterior and middle vation. Such a distinction between verbal and visual episodic mem-
third of the hippocampus was reduced to a nonsignificant trend. ory merits further investigation. Finally, regardless of the model to
Finally, it should be noted that this adjustment only accounts for be tested, we propose that future fMRI studies of the hippocampus
signal loss severe enough to place voxels below the ITDA. Lips- should include some estimate of susceptibility artifact. Recent
chutz et al. (2001) showed that sensitivity to BOLD effects is work by Devlin et al. (2000) and Cordes et al. (2000) has shown
proportional to signal intensity, so that even if, for example, an that susceptibility artifact is not limited to the temporal pole or
anterior hippocampal voxel survives the ITDA, it would prove lateral temporal regions and our study confirms that it can signif-
more difficult to activate than a posterior hippocampal voxel with icantly impact regional analyses of hippocampal function.
a higher baseline T2* signal. Thus, it is possible that the residual
difference, after adjusting for voxels below the ITDA, between the
anterior third of the hippocampus and the more posterior regions
is still due to susceptibility artifact. REFERENCES
We suspect that many previous fMRI studies could not accu-
rately assess activation differences between hippocampal subre-
gions owing to the effects of susceptibility artifact. Other authors Abduljalil A, Robitaille P. 1999. Macroscopic susceptibility in ultrahigh
field MRI. J Comput Assist Tomogr 23:832– 841.
have mentioned the potential role of susceptibility artifact in their Buckner RL, Wheeler ME, Sheridan MA. 2001. Encoding processes dur-
hippocampal studies, but to our knowledge none have attempted ing retrieval tasks. J Cogn Neurosci 13:406 – 415.
to quantify it. In their review of the fMRI literature, Schacter and Canteras N, Swanson L. 1992. Projections of the ventral subiculum to the
Wagner (1999) displayed data from their own lab (p.21, Fig. 3) amygdala, septum, and hypothalamus—a PHA-L anterograde tract-
showing the anterior MTL to have a signal-to-noise ratio (SNR) of tracing study in the rat. J Comp Neurol 324:180 –194.
Colombo M, Fernandez T, Nakamura K, Gross CG. 1998. Functional
60 compared with ⬎90 in the posterior MTL. They suggested that differentiation along the anterior-posterior axis of the hippocampus in
this apparently substantial difference was not due to susceptibility monkeys. J Neurophysiol 80:1002–1005.
artifact because regions such as the temporal pole that “demon- Constable RT, Skudlarski P, Mencl E, Pugh KR, Fulbright RK, Lacadie
strate considerable susceptibility-induced signal loss” had SNRs C, Shaywitz SE, Shaywitz BA. 1998. Quantifying and comparing re-
closer to 20. On the contrary, we suggest that their data comple- gion-of-interest activation patterns in functional brain MR imaging:
methodology considerations. Magn Reson Imaging 16:289 –300.
ment the findings of Ojemann et al. (1997), who described a Cordes D, Turski P, Sorenson J. 2000. Compensation of susceptibility-
gradient of susceptibility artifact moving away from the skull base induced signal loss in echo-planar imaging for functional applications.
and sinuses. It follows that the temporal pole would have the most Magn Reson Imaging 18:1055–1068.
artifact, the anterior MTL a moderate amount, and the posterior Craik FIM, Moscovitch M, McDowd JM. 1994. Contributions of surface
MTL the least, a pattern that matches the SNRs reported by and conceptual information to performance on implicit and explicit
memory tasks. J Exp Psychol Learn Mem Cogn 20:864 – 875.
Schacter and Wagner (1999). A laudable but incomplete attempt
Devlin J, Russell R, Davis M, Price C, Wilson J, Moss H, Matthews P,
at accounting for susceptibility artifact was described in a review by Tyler L. 2000. Susceptibility-induced loss of signal: comparing PET
Stern and Hasselmo (1999). Stern’s landmark 1996 experiment, and fMRI on a semantic task. Neuroimage 11:589 – 600.
which demonstrated posterior hippocampal activation to picture Dolan RJ, Fletcher PF. 1999. Encoding and retrieval in human medial
encoding, was performed on a 1.5-T scanner (Stern et al., 1996). temporal lobes: an empirical investigation using functional magnetic
This experiment was repeated by the same laboratory in 1998 resonance imaging. Hippocampus 9:25–34.
Duvernoy H. 1998. The Human Hippocampus. Berlin: Springer-Verlag.
while attempting to increase SNR by adding within-subject signal Fernandez GH, Weyerts M, Schrader-Bolsche I, Tendolkar HGO, Smid
averaging and scanning with a 3-T magnet (Kirchhoff et al., 1998). M, Tempelmann C, Hinrichs H, Scheich H, Elger CE, Mangun GR,
Although the primary finding of posterior hippocampal activation Heinze HJ. 1998. Successful verbal encoding into episodic memory
174 GREICIUS ET AL.

engages the posterior hippocampus: a parametrically analyzed func- Moser M-B, Moser EI. 1998. Functional differentiation in the hippocam-
tional magnetic resonance imaging study. J Neurosci 18:1841–1847. pus. Hippocampus 8:608 – 619.
Friston KJ, Holmes K, Worsley JP, Poline C, Frith CD, Frackowiak R. Ojemann JG, Akbudak E, Snyder AZ, McKinstry RC, Raichle ME, Con-
1995. Statistical parametric maps in functional imaging: a general turo TE. 1997. Anatomic localization and quantitative analysis of
linear approach. Hum Brain Mapp 2:189 –210. gradient refocused echo-planar fMRI susceptibility artifacts. Neuro-
Friston KJ, Holmes AP, Price CJ, Buchel C, Worsley KJ. 1999a. Multi- image 6:156 –167.
subject fMRI studies and conjunction analyses. Neuroimage 10:385– Otten LJ, Henson RN, Rugg MD. 2001. Depth of processing effects on
396. neural correlates of memory encoding: relationship between findings
Friston KJ, Zarahn E, Josephs O, Henson RN, Dale AM. 1999b. Stochas- from across- and within-task comparisons. Brain 124:399 – 412.
tic designs in event-related fMRI. Neuroimage 10:607– 619. Risold P, Swanson L. 1996. Structural evidence for functional domains in
Gabrieli JDE, Brewer JB, Desmond JE, Glover GH. 1997. Separate neural the rat hippocampus. Science 272:1484 –1486.
bases of two fundamental memory processes in the human medial
Rombouts S, Machielsen W, Witter M, Barkhof F, Lindeboom J, Schel-
temporal lobe. Science 276:264 –266.
tens P. 1997. Visual association encoding activates the medial tempo-
Glover GH, Lai S. 1998. Self-navigated spiral fMRI: interleaved versus
ral lobe: a functional magnetic resonance imaging study. Hippocam-
single-shot. Magn Reson Med 39:361–368.
Henke K, Buck A, Weber B, Wieser HG. 1997. Human hippocampus pus 7:594 – 601.
establishes associations in memory. Hippocampus 7:249 –256. Schacter DL, Curran T, Reiman EM, Chen K, Bandy DJ, Frost JT. 1999.
Holmes AP, Friston KJ. 1998. Generalisability, random effects and pop- Medial temporal lobe activation during episodic encoding and re-
ulation inference. Neuroimage 7:S754. trieval: a PET study. Hippocampus 9:575–581.
Kates WR, Abrams MT, Kaufmann WE, Breiter SN, Reiss AL. 1997. Schacter DL, Wagner AD. 1999. Medial temporal lobe activations in
Reliability and validity of MRI measurement of the amygdala and fMRI and PET studies of encoding and retrieval. Hippocampus 9:7–
hippocampus in children with fragile ⫻ syndrome. Psychiatry Res 24.
75:31– 48. Scoville WB, Millner B. 1957. Loss of recent memory after bilateral hip-
Kelley WM, Miezin FM, McDermott KB, Buckner RL, Raichle ME, pocampal lesions. J Neurol Neurosurg Psychiatry 20:11–21.
Cohen NJ, Ollinger JM, Akbudak E, Conturo TE, Snyder AZ, Pe- Small SA, Wu EX, Bartsch D, Perera GM, Lacefield CO, DeLaPaz R,
tersen SE. 1998. Hemispheric specialization in human dorsal frontal Mayeux R, Stern Y, Kandel ER. 2000. Imaging physiologic dysfunc-
cortex and medial temporal lobe for verbal and nonverbal memory tion of individual hippocampal subregions in humans and genetically
encoding. Neuron 20:927–936. modified mice. Neuron 28:653– 664.
Kirchhoff BA, Stern CE, Kwong K, Gonzalez R. 1998. A 3 tesla functional Small SA, Nava AS, Perera GM, DeLaPaz R, Mayeux R, Stern Y. 2001.
MRI study of picture encoding. Soc Neurosci Abs 24:266.4. Circuit mechanisms underlying memory encoding and retrieval in the
Kirchhoff BA, Wagner AD, Maril A, Stern CE. 2000. Prefrontal-temporal long axis of the hippocampal formation. Nat Neurosci 4:442– 449.
circuitry for episodic encoding and subsequent memory. J Neurosci Spielman D, Adalsteinsson E, Lim K. 1998. Quantitative assessment of
20:6173– 6180. improved homogeneity using higher-order shims for spectroscopic im-
Lepage M, Habib R, Tulving E. 1998. Hippocampal PET activations of aging of the brain. Magn Reson Med 40:376 –382.
memory encoding and retrieval: the HIPER model. Hippocampus Stark CE, Squire LR. 2000. Functional magnetic resonance imaging
8:313–322. (fMRI) activity in the hippocampal region during recognition mem-
Lipschutz B, Friston KJ, Ashburner R, Turner R, Price C. 2001. Assessing
ory. J Neurosci 20:7776 –7781.
study-specific regional variations in fMRI signal. Neuroimage 13:392–
Stern CE, Corkin S, Gonzalez R, Guimaraes A, Baker J, Jennings P, Carr
398.
C, Sugiura R, Vedantham V, Rosen BR. 1996. The hippocampal
Menon V, White CD, Eliez S, Glover GH, Reiss AL. 2000. Analysis of a
distributed neural system involved in spatial information, novelty, and formation participates in novel picture encoding: evidence from func-
memory processing. Hum Brain Mapp 11:117–129. tional magnetic resonance imaging. Proc Natl Acad Sci U S A 93:
Montaldi D, Mayes A, Barnes A, Pirie H, Hadley D, Patterson J, Wyper 8660 – 8665.
D. 1998. Associative encoding of pictures activates the medial tempo- Stern CE, Hasselmo ME. 1999. Bridging the gap: integrating cellular and
ral lobes. Hum Brain Mapp 6:85–104. functional magnetic resonance imaging studies of the hippocampus.
Moser EI, Moser M-B, Andersen P. 1993. Spatial learning impairment Hippocampus 9:45–53.
parallels the magnitude of dorsal hippocampal lesions, but is hardly Talairach J, Tournoux P. 1988. Co-planar stereotaxic atlas of the human
present following ventral lesions. J Neurosci 13:3916 –3925. brain. Stuttgart: Thieme Verlag.
Moser M-B, Moser EI, Forrest E, Andersen P, Morris RGM. 1995. Spatial Veltman D, Friston KJ, Sanders G, Price C. 2000. Regionally specific
learning with a minislab in the dorsal hippocampus. Proc Natl Acad sensitivity differences in fMRI and PET: where do they come from?
Sci U S A 92:9697–9701. Neuroimage 11:575–588.