emotional memory: separating content and context
TRANSCRIPT
www.elsevier.com/locate/psychresns
Psychiatry Research: Neuroim
Emotional memory: Separating content and context
Nicholas Medforda,T, Mary L. Phillipsa, Barbara Brierleya, Michael Brammerb,
Edward T. Bullmorec, Anthony S. Davida
aSection of Cognitive Neuropsychiatry, PO Box 68, Department of Psychological Medicine, Institute of Psychiatry and GKT School of Medicine,
King’s College, London SE5 8AF, UKbDepartment of Biostatics and Computing, Institute of Psychiatry, London, UK
cDepartment of Psychiatry, University of Cambridge, Cambridge, UK
Received 3 February 2004; received in revised form 23 September 2004; accepted 26 October 2004
Abstract
It is now well established that emotion enhances episodic memory. However, it remains unclear whether the same neural
processes underlie enhancement of memory for both emotional stimuli and neutral stimuli encoded in an emotive context. We
designed an experiment that specifically attempted to separate these effects and that was validated on 30 participants. We then
used functional magnetic resonance imaging (fMRI) to examine the neural correlates of encoding and retrieval of the two classes
of stimuli in 12 healthy male volunteers. We predicted that aversive emotional context would enhance memory regardless of
content and that activation of anterior cingulate would be inversely related to retrieval of aversive items. Both predictions were
supported. Furthermore we demonstrated apparent asymmetrical lateralisation of activation in the hippocampal/parahippocampal
complex during recognition of words from aversive sentences: more left-sided activation for neutral words from aversive
contexts, and more right-sided activation for aversive content words. These findings, if applicable to the wider population, may
have application in a range of psychiatric disorders where interactions between emotion and cognition are relevant.
D 2005 Elsevier Ireland Ltd. All rights reserved.
Keywords: Emotion; Memory; Amygdala; Hippocampus; Anterior cingulate cortex; Functional MRI
1. Introduction tively valenced assessments of the past, while post-
Interactions between emotion and memory play an
important role in psychiatry. For example, cognitive
theories of depression emphasize the role of nega-
0925-4927/$ - see front matter D 2005 Elsevier Ireland Ltd. All rights re
doi:10.1016/j.pscychresns.2004.10.004
T Corresponding author. Tel.: +44 20 7 848 0138; fax: +44 20 7
848 0572.
E-mail address: [email protected] (N. Medford).
traumatic stress disorder (PTSD) arises as a conse-
quence of an aversive event and the handling of that
event by memory and other cognitive systems. While
there is strong evidence that memory is enhanced by
emotional arousal at the time of encoding (Burke et
al., 1992; Sierra and Berrios, 1999), the specific
neural and cognitive mechanisms remain elusive.
Animal research (Gallagher and Chiba, 1996;
LeDoux, 1998) has implicated the amygdaloid com-
aging 138 (2005) 247–258
served.
N. Medford et al. / Psychiatry Research: Neuroimaging 138 (2005) 247–258248
plex (AC) in the formation of conditioned fear
responses, leading to the idea that this structure may
be involved in the formation of emotional memories
in man.
There is now good evidence for amygdala involve-
ment in human emotional memory. Patients with
amygdala lesions have deficits in recognizing the
emotions of others (Adolphs et al., 1994; Broks et al.,
1998), and may also lose the enhancement of memory
normally conferred by emotion (Phelps and Anderson,
1997). A study examining the memory of people with
Alzheimer’s disease for an emotional event (a
devastating earthquake in Kobe, Japan) found that
impairment of memory for this event was correlated
with the density of amygdala damage (Mori et al.,
1999). A landmark PET study (Cahill et al., 1996) in
which subjects viewed emotionally arousing video
clips found that amygdala activation at encoding was
correlated with subsequent recall of the emotional
material, suggesting that amygdala arousal may
modulate the formation of memories for emotional
material. This idea is supported by studies using
pharmacological manipulation to influence arousal
(McGaugh et al., 1996; O’Carroll et al., 1999), and by
an event-related fMRI study (Canli et al., 2000) in
which, within individual subjects, amygdala activa-
tion at encoding was found to correlate with intensity
of emotional response to stimuli. With regard to a
possible role for the amygdala in retrieval, animal
studies suggest that an intact amygdala is required for
encoding (McGaugh et al., 1996), but not recall, of
emotional material, although elsewhere (Dolan et al.,
2000) it has been suggested that the left amygdala
may have a specific role in retrieval of emotional
memories. A more recent study (Dolcos et al., 2004)
supports the idea that interactions between the
amygdala and the medial temporal lobe (MTL)
memory system underlie the emotional enhancement
of memory, and suggests that more anterior parts of
the MTL have a specific role in encoding emotional
information.
In considering emotional memory, relative contri-
butions of, firstly, emotional material (content) and,
secondly, the context in which this material occurs are
of particular relevance in understanding the role of
emotional memory in psychopathology. Studies of
these effects have yielded conflicting results (Burke et
al., 1992). After an emotional event, while recall of
the event itself is enhanced, memory for surrounding
contextual information is variously found to be either
enhanced or diminished, raising the possibility that
memory for content and context are subserved by
different neural networks. Indeed, work by Kensinger
et al. (2002) showed that the direct enhancement
effect on memory of emotional content is preserved in
normal ageing while that of context is diminished.
Patients with Alzheimer’s disease, presumed to have
diffuse brain disease, showed loss of emotional
enhancement from both content and context.
A small number of functional imaging studies
have attempted to address these issues. Maratos et al.
(2001) examined memory for emotional context by
testing recognition memory for neutral words pre-
sented in emotional sentences. However, there was
no test of recall for words whose content was
intrinsically emotive, and thus no comparison of
the neural correlates of emotional content and
context recognition memory. Erk et al. (2003)
examined the effect of emotional context at encoding
on subsequent memory performance, but scanning
data were only obtained during encoding. One recent
study (Smith et al., 2004) examined contextual
influences on both encoding and recognition memory
by the use of pictorial stimuli presented on either
emotional or neutral backgrounds. However, this
method of associating stimulus with context may not
be analogous to real-life situations, where contextual
information is likely to have more semantic rele-
vance than an arbitrary association. In this study, we
aimed to examine the neural correlates of memory
for both content and context, encoded in such a way
that they were semantically linked.
In addition to studying temporal lobe structures
involved in emotion and memory, we wished to
examine the role of anterior cingulate cortex (ACC) in
encoding and subsequent recognition of emotional
material. It has been suggested (Hamner et al., 1999)
that the ACC interprets and contextualises emotional
information, thus brationalisingQ and dampening the
emotional response, this being consistent with a study
(Bremner et al., 1999) in which war veterans with
post-traumatic stress disorder (PTSD) showed less
ACC activation in response to combat-related stimuli
than veterans without PTSD. Thus this area has
particular significance for understanding the role of
context in emotional memory.
Table 2
Examples of emotional and neutral sentence pairs
Emotional sentence Neutral sentence
He stood on the balcony
and watched the riot.
He stood on the balcony
and watched the tide.
He would abuse the
children at every party.
He would amuse the children
at every party.
There was a scream in the hall. There was a carpet in the hall.
The parcel contained a bomb. The parcel contained a bowl.
Each pair differs by one word. In the above examples, briotQ andbtideQ are matched to within one standard deviation on measures of
written frequency, imageability, concreteness and familiarity.
Similarly, babuseQ and bamuseQ, bscreamQ and bcarpetQ, bbombQand bbowlQ are all matched pairs. For clarity of explanation,
embedded words are shown underlined, but were not underlined
when presented to subjects.
N. Medford et al. / Psychiatry Research: Neuroimaging 138 (2005) 247–258 249
2. Methods
This study employed a modified version of a test of
verbal emotional memory designed by the authors
(NM and BB) and utilized in previous studies of
emotional memory following temporal lobectomy
(Brierley et al., 2004) and Alzheimer’s disease (Ken-
singer et al., 2002). Forty-two emotionally aversive
btargetQ words were selected from a list rated on
measures of arousal and valence (Bradley and Lang,
1999). Each word was used to create a sentence. A
psycholinguistic database (Coltheart, 1981) was then
used to generate a further set of btargetQ words that
were matched to the emotional words for length, and
for frequency, imageability, concreteness and familiar-
ity (Table 1), but that were affectively neutral. From
this, a second, neutral set of 42 sentences was created,
identical to the emotional sentences except for the
neutral–emotional word swap (see Table 2 for
examples). These were arranged into two sets,
designated bXQ and bYQ sets, each containing 21
emotional and 21 neutral sentences.
Sentences were printed on cards in plain text, and
in an initial experiment conducted outside the scanner,
30 normal volunteers (17 male, 13 female, age 17–43,
mean age 27.8) performed three tasks. Firstly, encod-
ing, in which subjects were asked to silently read each
sentence. Half of the subjects were shown the bXQ setand the other half viewed the bYQ. A three-alternative
forced-choice recognition test followed after a 5-min
delay. For each item, three words were presented, of
Table 1
Psycholinguistic variables: comparisons between emotional and
neutral target words on measures of frequency (FREQ), concrete-
ness (CONC), imageability (IMAG), and familiarity (FAM)
Word type N Mean SD df t Sig
(2-tailed)
FREQ Emotional 42 31.12 38.73 81.7 �0.85 0.933
Neutral 42 31.86 40.98
CONC Emotional 40 456.55 109.11 79.9 �1.52 0.133
Neutral 42 493.31 110.10
IMAG Emotional 41 524.24 62.18 75.7 0.68 0.502
Neutral 42 513.33 83.73
FAM Emotional 42 505.79 49.58 81.0 �1.24 0.219
Neutral 42 520.00 55.34
There are no significant differences between the groups (SD=stan-
dard deviation, df=degrees of freedom).
which one had been seen before in the sentences at
encoding, the other two being distracters. Subjects
were asked to indicate which word had been seen
previously. Half of the correct responses were btargetQwords, and half were neutral bembeddedQ words. Theincorrect distracters were words matched to the
correct responses for emotional valence, word length
and to within one standard deviation for frequency
and familiarity.
Twenty volunteers independently rated all target
and embedded words used in the study for emotional
valence and arousal. This confirmed that words
designated as being emotive were indeed signifi-
cantly different from neutral words on these measures
(data available on request). In addition to these
ratings of individual words, 10 different volunteers
rated each set of sentences for arousal and valence.
Arousal was rated from 1 to 7 where 1=dno reactionTand 7=dstrongest reaction imaginableT, while valence
was rated from 1 to 7 on a scale where 1=dveryunpleasantT, 4=dneutralT, and 7=dvery pleasantT. Forthe dXT set, aversive sentences were significantly
different from neutral sentences on both measures:
For arousal, the mean score for the 21 aversive
sentences was 3.3, SD 0.84, while for the 21 neutral
sentences, the mean arousal score was 2.1, SD 0.4
(t=�5.94, Pb0.001, two-tailed test of significance).
For valence, the mean for aversive sentences was 2.7,
SD 0.56, and for neutral sentences 4.3, SD 0.4
(t=10.81, Pb0.001). For the dYT set, the mean arousal
score for aversive sentences was 3.1, SD 0.69, and
for neutral 2.0, SD 0.28 (t=�6.94, Pb0.001), while
N. Medford et al. / Psychiatry Research: Neuroimaging 138 (2005) 247–258250
mean valence for aversive sentences was 2.95, SD
0.35, and for neutral sentences 4.1, SD 0.31 (t=11.81,
Pb0.001). It is of note that for each set, the mean
valence for neutral sentences was approximately 4,
the neutral point on our valence scale, confirming
that these sentences did not have any significant
emotional valence, either positive or negative. (Rat-
ings from individual volunteers, and for individual
sentences, are available on request.)
2.1. Functional MRI experiments
Twelve different volunteers (right-handed males,
age range 22–34 years, mean 27.8) then underwent
three 5-min functional imaging data acquisitions
while performing equivalent tasks in the scanner. In
the encoding task, subjects silently read a set of 42
sentences plus eight bfillersQ, a total of 50 sentences,
each sentence being projected onto a screen for 6 s.
No sentence was repeated. After reading each
sentence, subjects pressed a button (this was simply
to confirm they had read it). bFillersQ–extra sentences
(rated as affectively neutral by the same individuals
who rated the test sentences) that were projected but
not included in the scoring–were used to simplify the
timing of sentence projection and to ensure that each
sentence was onscreen for the same length of time.
bFillersQ were equally distributed across aversive and
neutral blocks (see below). Two forced-choice recog-
nition tasks followed. For ease of presentation, two-
choice, rather than three-choice, recognition memory
tests were used. Pairs of words (targets plus matched
distracters) were projected onto the screen and
subjects pressed a button to indicate which word
had been in the sentences at encoding. In one task, the
correct responses were btargetQ words, and in the
other, the correct responses were neutral bembeddedQwords. In total, therefore, there were three (one
encoding and two recognition) 5-min tasks. Within
each task, 50 items were presented according to
block-design methodology. This involves presenting
alternating blocks of different classes of stimuli
(aversive and neutral, in this case). Thus, brain
activation in response to each class of stimuli can be
measured separately, allowing emotional vs. neutral
comparisons within each task. This enabled identi-
fication of brain areas selectively activated by the
aversive items in each task.
For the encoding task, the block design depended
on the fact that both bXQ and bYQ sentence sets
contained equal numbers of emotional and neutral
sentences. To control for possible order effects, each
set was arranged into groups (blocks) of emotional
and neutral, prior to onscreen projection. Subjects
viewed alternating blocks of aversive and neutral
sentences (five sentences per block, each block
lasting 30 s). The within-task order of these blocks
was randomized across subjects, although they were
always presented such that emotional blocks alter-
nated with neutral blocks. The eight bfillerQ stimuli
were equally distributed across emotional and neutral
blocks to give a total of 50 sentences for each
subject. For the target and embedded word recog-
nition tasks, 50 word pairs per task were presented in
alternating blocks so that in one block the correct
responses were words from emotional sentences, and
in the next, correct responses were from neutral
sentences. In the embedded word task, one subject’s
data were excluded due to technical failure. Thus
n=12 for target word recognition testing, n=11 for
embedded word recognition. In all three tasks,
stimuli (sentences or word pairs) were presented
every 6 s. To control for different reading speeds
between subjects, stimuli disappeared from the
screen once the subject pressed a response button.
For example, in the encoding task, if a subject
pressed the button after 3 s, the sentence would then
disappear and the screen would be blank for 3 s
before the next sentence appeared.
2.2. Image acquisition
Gradient echo echoplanar images were acquired
on a GE Signa 1.5 T Neurovascular system (General
Electric, Milwaukee, WI, USA) at the Maudsley
Hospital, London. One hundred T2*-weighted
images depicting BOLD (blood oxygenation level
dependent) contrast (Ogawa et al., 1990) were
acquired over 5 min (for each task) at each of 14
near-axial non-contiguous 5-mm thick planes parallel
to the intercommissural (AC-PC) line: TE 40 ms, TR
3 s, in-plane resolution 5 mm, and interslice gap 0.5
mm. This EPI dataset provided complete coverage of
the temporal lobes and almost complete coverage of
the frontal, occipital and parietal lobes (Simmons et
al., 1999).
Table 3
Recognition memory scores for the initial experiment with 30
subjects outside the scanner
Recognition memory comparison MeanFSD t P
Emotional target words 16.63F2.20 5.13 b0.001
Neutral target words 13.10F3.17
Embedded words (E) 13.63F2.61 3.60 0.001
Embedded words (N) 11.67F2.50
E=embedded words from emotional sentences, N=embedded words
from neutral sentences. Maximum possible score on each recog
nition memory subtest is 21. All significance tests two-tailed
Degrees of freedom=29.
N. Medford et al. / Psychiatry Research: Neuroimaging 138 (2005) 247–258 251
2.3. Statistical analysis
Following motion correction (Bullmore et al.,
1999), periodic change in T2*-weighted signal
intensity at the (fundamental) experimentally deter-
mined frequency of alternation between A and B
conditions (=1/60 Hz in all experimental conditions)
was estimated by an iterated least squares fit of a
sinusoidal regression model to the fMRI time series
observed at each voxel. A standardized test statistic,
the standardized power (SP), was obtained for each
voxel (Bullmore et al., 1996). Parametric maps
representing SP observed at each intracerebral voxel
were constructed. To sample the distribution of SP
under the null hypothesis that observed values of SP
were not determined by experimental design (with
few assumptions), the 99 images observed in each
anatomical plane were randomly permuted and SP
was estimated exactly as above in each permuted
time series. This process was repeated 10 times at
each voxel, resulting in 10 permuted parametric maps
of SP at each plane for each subject. Thus, this
analysis is based on a random effects model, fully
described elsewhere (Brammer et al., 1997). The
observed and randomized SP maps were transformed
into standard space (Talairach and Tournoux, 1988)
and smoothed by a 2D Gaussian filter with full width
at half maximum=11 mm. This procedure is
described in detail elsewhere (Brammer et al., 1997;
Bullmore et al., 1996) but essentially involves
normalization onto a standard template (an average
of 10 IR images already transformed into standard
Talairach space). The median observed SP at each
intracerebral voxel in standard space was tested
against a critical value of the null distribution of
median SPs constructed from the permuted SP maps
(Brammer et al., 1997). For a test at any desired P-
value, the critical value is extracted from the random-
ization distribution such that 1/P of the random-
izations exceed that value. In the observed data,
voxels with SPs exceeding this critical value had a
probability under the null hypothesis less than or
equal to the chosen P-value. The particular P-value
was determined by setting the expected number of
false-positive voxels (EPI), so that P=EPI/total search
volume. For this study the EPI was set at 50 voxels
so that the threshold P-value was 50/20,000 (approx-
imate search volume) or 0.0025.
To estimate the differences in mean SP between the
two recognition memory experimental conditions
(target and embedded), we fitted repeated measures
analysis of variance (ANOVA) models at each voxel
of the observed SP maps in standard space. Differ-
ences in mean SP between the two conditions were
tested for significance only at voxels generically
activated by one or both of the conditions considered
independently, thereby substantially reducing the
search volume or number of tests conducted.
Other methodological considerations: Each subject
completed the National Adult Reading Test (NART),
the Beck Depression Inventory (BDI), and the Beck
Anxiety Inventory (BAI). Subjects’ verbal IQ, as
predicted by the NART, ranged from 105 to 126
(mean 114.3). Subjects had no history of neurological
or psychiatric illness, were on no medications, and all
scored below clinical cut-offs on the BDI and BAI.
Subjects were paid o20 for their participation.
3. Results
3.1. Behavioural data
The initial experiment conducted outside the
scanner (results summarised in Table 3) showed that
memory for emotional target words was significantly
superior to matched neutral words (Pb0.001), evi-
dence of emotional memory enhancement. In addition,
memory for neutral words that were bembeddedQin emotional sentences–either before or after the
aversive word–was also enhanced to a significant
albeit lesser extent (P=0.001). These significant
findings were essentially unchanged in further experi-
ments in which the interval between encoding and
-
.
N. Medford et al. / Psychiatry Research: Neuroimaging 138 (2005) 247–258252
recognition testing (here 5 min) was varied from 30
min to 24 h (data available on request).
In the subsequent fMRI study, recognition memory
for words from emotional sentences was significantly
greater than that from neutral sentences [33.6 (SD 4.2)
vs. 31.0 (SD 4.3); t=2.58, P=0.027]. There were
trends for emotional target vs. neutral words [17.8
(SD 2.59) vs. 16.5 (SD 2.28), P=0.1], and neutral
words embedded in aversive sentences vs. neutral
words in neutral sentences to be better remembered
[15.8 (2.4) vs. 14.4 (2.5), P=0.08], but the contrasts
failed to reach statistical significance.
Neuroimaging findings are summarised in Table 4.
Encoding of aversive sentences caused significant
additional activation over matched neutral sentences
in the left anterior cingulate gyrus and left precuneus.
Table 4
Number
of voxels
Talairach
co-ordinates
Side Brain area
(approx. BA)
P
Areas activated by encoding of emotional sentences
17 �11 33 31 L Anterior cingulate
gyrus (32)ER, Content0.000009
6 �7 �52 31 L Precuneus (7) 0.00004
Areas activated by recall of target emotional words
84 15 �52 �13 R Cerebellum 0.000009
53 �4 39 �13 L Medial prefrontal
cortex (11)ER, Content0.000007
42 15 46 �2 R Anterior cingulate
gyrus (32)
0.000028
18 �47 17 4 L Inferior frontal
gyrus (47)Content0.00017
73 �7 �43 37 L Posterior cingulate
gyrus (31)Content0.000007
50 �4 �46 31 L Precuneus (7) 0.000007
78 �40 �60 26 L Middle temporal
gyrus (39/21)Content0.000007
23 �40 13 �2 L Anterior insula
(47)ER, Content0.00002
28 �40 13 �13 L Superior temporal
gyrus (38)
0.000007
18 25 �43 �2 R Parahippocampal
gyrus (30)
0.00002
12 36 �10 �7 R Amygdaloid complex 0.00002
8 40 �10 �13 R HippocampusER, Content 0.0002
Areas activated by recall of embedded words from emotional
sentences
5 �26 �33 �13 L Parahippocampal
gyrus (36)
0.0007
5 �26 �36 42 L Inferior
parietal lobule (40)
0.0012
Recognition of aversive target words—compared with
neutral target words—activated anterior and posterior
cingulate gyrus, inferior frontal and areas of medial
prefrontal cortex, precuneus, anterior insula, right
amygdaloid complex, right hippocampus, right para-
hippocampal gyrus, and right cerebellum.
Embedded word recognition: Recognition of
embedded words from emotional sentences—com-
pared with those from neutral sentences—activated
areas of left parahippocampal and inferior parietal
lobule. Specific P-values for individual activations
are shown in Table 4.
Since automatic encoding of new distracter words
occurs during recognition testing, it is necessary to
carry out further contrasts to refine the activation
maps corresponding to retrieval alone. Comparing
retrieval of emotional targets with encoding of emo-
tional sentences showed significantly greater activa-
tion in left medial frontal lobe, left insula, right
hippocampus, and right anterior cingulate cortex
during retrieval. Retrieval of emotional target words
confounds emotional content (the word itself) and
context (the meaning of the sentence in which the
word was encoded). Hence, we contrasted target
recognition (content plus context) with embedded
word recognition (context only) to refine the activa-
tion corresponding to emotional retrieval. This
showed significantly greater activation for target
emotional word recognition in left anterior and
posterior cingulate cortex, left middle temporal gyrus,
left medial frontal cortex, left anterior insula and right
hippocampus (Fig. 1). To explore further the robust-
ness of the medial temporal lobe laterality effects
observed for the two recognition memory tasks, we
also looked specifically at the area of the left
parahippocampal gyrus activated during the embed-
ded word recognition task, extracting the mean SP for
this area from each subject’s data and comparing it
with the mean SP from the same area during the target
recognition task (see Fig. 2). When a single outlying
value was excluded, a paired t-test showed signifi-
cantly greater SP in the embedded word condition for
this left parahippocampal region (t=2.06, df=10,
P=0.034, one-tailed test).
The anterior cingulate cortex, which has reciprocal
connections with the amygdala and the hippocampus,
emerged as crucial in the encoding of emotional
content. Previous neuroimaging research (Bremner et
Fig. 1. Regional brain activation during recognition of aversive target words. Key areas of activation are labeled (amyg=amygdala,
hippo=hippocampus, ACC=anterior cingulate cortex, ant ins=anterior insula, cx=cortex). (a) Shows an axial slice at the Talairach level z=�13,
(b) is a slice at level z=�2.
N. Medford et al. / Psychiatry Research: Neuroimaging 138 (2005) 247–258 253
al., 1999; Hamner et al., 1999) has suggested that this
region modulates or inhibits activation in limbic
structures. To test for this inverse relationship in our
subjects, the mean SP scores (see above) for each
subject were extracted from the area of this region
activated in the encoding condition. These scores
were significantly negatively correlated with perform-
ance on the emotional target word recognition task—
the greater the degree of anterior cingulate activation,
the fewer emotional target words correctly recognized
(Fig. 3). When all 12 subjects were considered
together, Kendall’s tau=�0.46, P=0.044. When one
outlying subject (see Fig. 2) was excluded, tau=�0.53,
P=0.027. There was also a weak, non-significant
negative correlation between the same SP scores and
the number of neutral target words correctly recog-
nized (tau=�0.39, P=0.083, all 12 subjects included
in analysis).
4. Discussion
The behavioural data support the idea that
emotionally salient verbal material is better remem-
bered than affectively neutral, but otherwise similar,
material. This effect extends to context words as
well as words that are emotionally salient. This
does not necessarily imply, however, that this
contextual bemotional memory effectQ would extend
to other contextual information in different settings
(e.g. incidental visuospatial material in emotional
scenes).
The neuroimaging data suggest that mechanisms
for emotional retrieval differ according to the type
of information to be recalled. The right amygdala
complex and right hippocampus were activated by
the recognition of target emotional words, but not
by the encoding or embedded recognition condi-
tions. In the embedded word recognition condition,
activation was present in the left parahippocampal
gyrus when the words to be retrieved came from
emotional contexts. Thus areas involved in retrieval
are activated more by emotional than neutral
content. Of particular interest is the finding that
hippocampal/parahippocampal activations were on
the right in the target word recognition task and on
the left for the embedded recognition task. Directly
comparing the target and embedded word tasks
showed that the right hippocampus was one of the
areas significantly more activated during the former
condition, while comparing each subject’s SP for
the left parahippocampal gyrus across the two
conditions showed a significant difference (with
more activation in the embedded word condition)
when a single outlying value was excluded from the
analysis. Thus, these laterality differences appear
highly robust when subjected to further levels of
analysis. This implies a laterality difference in the
recall of material that is emotive in and of itself, as
compared with material that is emotive only by
Fig. 3. Scatterplot showing inverse relationship between ACC
activation at time of encoding and subsequent score for recognition
of target emotional words. Outlying value in brackets.
Fig. 2. Scatterplot showing mean standardized power (SP) in left
parahippocampal cortex for each subject during target (TARG) and
embedded (EMB) word recognition tasks. Outlying value in
brackets.
N. Medford et al. / Psychiatry Research: Neuroimaging 138 (2005) 247–258254
association (in other words, content as compared
with context).
There are two possible ways of interpreting the
activation of the right amygdala seen in the target
recognition condition. One interpretation is that the
amygdala is involved in recall of emotional material.
This runs counter to the view (discussed in Section 1)
that the amygdala acts at the time of encoding to
modulate memory storage, but is not involved in the
subsequent process of recall. The second interpreta-
tion is that the amygdala activation is due to the
viewing of pairs of emotional words, from which
subjects had to choose their responses. In other words,
it reflects the response to emotional material rather
than the process of recognition per se. This latter
interpretation sits more comfortably with other work
in this field, and at the time of writing it is the one we
favour.
A study of recognition memory of words from
emotive contexts (Maratos et al., 2001) found left
amygdala and hippocampus were selectively activated
by recognition of neutral words from aversive emo-
tional sentences, broadly in line with our findings,
although the methodology did not involve content vs.
context, or encoding vs. recognition comparisons. Erk
et al. (2003) examined the effect of context on
subsequent memory performance by presenting neu-
tral words paired with unrelated images, which could
be affectively positive, negative, or neutral, and then
assessing free recall of the presented words. The recall
phase was performed outside the scanner, so that the
neuroimaging findings relate only to the encoding
phase, but the authors were able to identify a network
of right-sided MTL structures in which the degree of
activation was apparently predictive of subsequent
memory performance for words presented in pos-
itively valenced contexts. Interestingly, they found an
emotional enhancement of memory for words pre-
sented in positive, but not negative, contexts, and
suggest that this may be because the impact of the
negatively valenced pictures is such that encoding of
N. Medford et al. / Psychiatry Research: Neuroimaging 138 (2005) 247–258 255
associated, but semantically unrelated, stimuli is
adversely affected. This is likely to be the case, and
such a phenomenon is probably in part responsible for
the conflicting results found in previous studies of
emotional memory effects (Burke et al., 1992). A
striking, if extreme, example of this is the bweaponfocusQ effect described in victims or witnesses of
violent crime (Loftus and Burns, 1982), where
attention is narrowed to the weapon, and memory
for peripheral but crucial information (e.g. the dress
and appearance of the person holding the weapon)
may thus be unreliable. It is likely that negative
words, as used in our study, have less impact in this
sense than the intensely aversive images used by Erk
et al. (and employed in many other studies of
emotional processing), and this may also explain
why, in contrast to other studies using emotional
images, we did not find any amygdala activation
during the encoding phase. Another difference
between our study and that of Erk et al. that is
relevant here is that our study uses semantically
related stimuli—in a sentence where a single word is
aversive, the context depends on understanding the
meaning of the whole sentence. A recent commentary
(Lewis and Critchley, 2003) outlines a bsemantic-
network approachQ to the study of emotion–memory
interactions, and it is likely that stimuli which are
indeed semantically related are better suited to study-
ing such networks. This is not to dismiss the
usefulness of other means of establishing context, as
in real life contextual information may be highly
variable in terms of its relevance to an emotional
event—it may be directly relevant, entirely coinci-
dental, or anywhere between these poles. These
considerations are also relevant to a more recent
study (Smith et al., 2004), where contextual memory
effects were probed using pictures of objects paired
with neutral or emotional picture backgrounds. As in
our study, both encoding and recognition were studied
in the scanner, and the authors identified left para-
hippocampal gyrus as having a specific role in
contextual recall. Our data support this finding,
although the foregoing caveats about the possible
significance of different methods of establishing
context must also be borne in mind.
The possibility of laterality effects in emotional
function has been extensively explored in both
animals and humans (Heilman, 2000), and a range
of experimental and empirical findings support a
model based on two basic circuits: approach and
withdrawal (see Davidson, 2000, for a review). Our
findings support the idea that aversive material (the
emotional target, or content, words) selectively
activates the right amygdala, while the finding of a
left hemisphere bias for recognition memory of
neutral words from aversive contexts underlines the
idea that contextual processing may be served by
different circuitry from that used for content. All the
emotional stimuli in our study were negatively
valenced, which is more relevant to psychopathology,
so while our findings may have relevance to prior
work on the withdrawal circuit, we cannot comment
here on the approach circuit. The use of positively
valenced content/context stimuli may be a fruitful
avenue for future work.
Despite the tightly controlled parameters in our
experiment, retrieval of emotional words produced
widespread activation. This could reflect the tendency
for emotive stimuli to trigger many associations.
Posterior cingulate gyrus was activated by the emo-
tional component of the target retrieval task, lending
weight to the suggestion (Maddock, 1999) that it has a
key role in the integration of emotion and episodic
memory processes. The anterior insula is known to be
involved in processing emotionally aversive stimuli
(Phillips et al., 1997), and its activation during
emotional target recognition is consistent with this,
although this activation may be due to viewing
emotional word pairs rather than recognition memory.
The anterior cingulate cortex (ACC) was strongly
activated by reading emotional sentences, and by the
recognition of target emotional words. The finding
that ACC activation was negatively correlated with
emotional target recognition provides direct evidence
that this region modulates the impact of emotionally
salient material, thus influencing how it is encoded
and retrieved. Data from neuroimaging studies and
animal work suggest that the ACC can be considered
as having cognitive and affective subdivisions (Bush
et al., 2000). The activation observed in the encoding
phase lies in the cognitive subdivision, while that seen
in the target recognition phase falls within the
affective division, but it is the former that appears to
have a crucial effect on the subsequent recognition of
aversive target words. This area is known to be
activated by tasks that involve competing categories
N. Medford et al. / Psychiatry Research: Neuroimaging 138 (2005) 247–258256
of information (Bush et al., 2000). In the encoding
phase, subjects were merely asked to read each
sentence, not to provide any ratings of bemotionalityQ.It may be that in these circumstances, the affective
content of the aversive sentences is treated as a con-
flicting information stream by the ACC. This would
account for the inverse correlation seen between
activation of this area of the ACC at encoding and
the subsequent recognition memory score for target
emotional words—the greater the ACC activation, the
more the response to affective content is inhibited, and
thus the emotional memory effect is diminished.
However, this interpretation needs to be approached
with some caution, as there is also a weak non-
significant negative correlation between the same
ACC SP values during encoding and the subsequent
recognition scores for neutral emotional words. This
suggests that activity of this area during encoding may
have a more general limiting effect on subsequent
recognition memory, besides any specific modulation
of the bemotional memory effectQ. It may be that a
particular attentional style, associated with activity in
this region of the ACC, affects the encoding and
subsequent memory trace for verbal material. Other
attentional processes may potentiate the emotional
memory effect: while most accounts of emotional
memory emphasize the role of arousal (see Section 1),
it is at least possible that improved recognition
memory for aversive target words need not rely
entirely on arousal-mediated mechanisms. Differences
in valence between matched targets could account for
the observed pattern of recognition memory due to a
bpop-outQ effect (Nakayama and Joseph, 1998): the
aversive targets could be more perceptually salient,
and therefore memorable without invoking a separate
mechanism based on arousal (see also Worthen et al.,
2000). The matching of aversive and neutral words for
psycholinguistic variables provides some safeguard
against such effects. However, even after matching of
individual words, there may still be sentence-level
effects.
In contrast to the activation of the cognitive
division of the ACC seen in encoding, during target
word recognition, the affective division was activated
by the aversive phase. This may reflect the viewing of
aversive word pairs rather than any specific retrieval
process (cf. amygdala activation during this phase, see
above). It is likely that reading individual words
produces a more affectively driven response than does
reading entire sentences, where the emphasis is on the
meaning of the sentence construct, the latter task thus
engaging more cognitive areas. Activations in middle
temporal gyrus probably reflect the use of verbal
stimuli, but they are nonetheless confined to the
emotional phase. Thus, it appears that emotionally
salient material exerts a modulatory effect on areas
involved with higher cognitive functions such as
language and memory, perhaps analogous to the
modulatory effect previously described in sensory
cortex (Morris et al., 1998).
The present study has a number of limitations.
The use of a block-design paradigm does not permit
analysis of specific memory effects for individual
items (see, for example, Dolcos et al., 2004), and
does carry the risk that the data can be contaminated
by bfalse alarmsQ. The technique of alternating task
blocks does not allow computing deactivation asso-
ciated with the task condition, so that areas which
show significantly different activation when neutral
conditions are subtracted from emotional conditions
may include deactivations under the neutral con-
dition, although this need not necessarily imply a
different interpretation. Furthermore, it is possible
that the encoding task was easier than the recog-
nition memory tasks, so that differences in activation
patterns between conditions may in part reflect
differences in task difficulty or differences in
stimulus complexity. In addition, our design did
not permit any separation of the effects of stimulus
valence and arousal, but other work has suggested
that these variables may have dissociable effects
(including effects relating to laterality) on emotional
responses (Canli et al., 1998) and emotional memory
(Canli et al., 2002; Libkuman et al., 2004), while a
study specifically examining memory for emotional
words concluded that both valence and arousal
contribute to emotional enhancement of verbal
memory (Kensinger and Corkin, 2003). Future work,
involving the concurrent acquisition of fMRI and
galvanic skin response data, may be able to address
the relative contributions of arousal and valence to
content–context interactions. The study by Canli et
al. (2002) explored gender differences using fMRI,
and enlarges on earlier work (Cahill et al., 2001) in
suggesting a gender difference in lateralisation of
amygdala function, but our data do not allow us to
N. Medford et al. / Psychiatry Research: Neuroimaging 138 (2005) 247–258 257
comment on gender differences as there were no
female subjects in our fMRI experiment. However, if
our interpretation of the right amygdala activation
seen in our target recognition condition is correct,
this is in line with the finding (Cahill et al., 2001)
that emotional material selectively activates the right
amygdala in males. However, the applicability of the
initial behavioural study, which was conducted with
both male and female participants, to the subsequent
fMRI study (male participants only), may be
compromised by the fact that other studies have
shown that female subjects tend to describe more
intense responses to emotional stimuli (e.g. Canli et
al., 2002).
Interactions between emotion and high-level cog-
nitive processes remain relatively unexplored in
neuroimaging, in which simpler stimuli such as
emotional faces and sounds are frequently used, and
further work is required in this regard. We have shown
the ACC to be critical in emotional memory enhance-
ment of verbal material. In addition our data suggest
that emotional content enhances memory through
increased activation of the hippocampus, and that
right and left medial temporal lobe structures play
different roles with regard to content and context in
the formation of emotional memories. These findings
enlarge on previous work on the neural basis of
emotional memory and its role in both normal and
pathological affect.
References
Adolphs, R., Tranel, D., Damasio, H., Damasio, A., 1994. Impaired
recognition of emotion in facial expressions following bilateral
damage to the human amygdala. Nature 372, 669–672.
Bradley, M.M., Lang, P.J., 1999. Affective norms for English words
(ANEW). The NIMH Center for the Study of Emotion and
Attention, University of Florida, Gainesville, FL.
Brammer, M., Bullmore, E., Simmons, A., Williams, S.C., Grasby,
P.M., Howard, R.J., Woodruff, P.W., Rabe-Hesketh, S., 1997.
Generic brain activation mapping in functional magnetic
resonance imaging: a nonparametric approach. Magnetic Res-
onance Imaging 15, 763–770.
Bremner, J.D., Staib, L.H., Kaloupek, D., Southwick, S., Soufer, R.,
Charney, D., 1999. Neural correlates of exposure to traumatic
pictures and sound in Vietnam combat veterans with and
without posttraumatic stress disorder: a positron emission
tomography study. Biological Psychiatry 45, 808–816.
Brierley, B., Medford, N., Shaw, P., David, A.S., 2004. Emotional
memory and perception in temporal lobectomy patients with
amygdala damage. Journal of Neurology, Neurosurgery and
Psychiatry 75, 593–599.
Broks, P., Young, A.W., Maratos, E.J., Coffey, P.J., Calder, A.J.,
Isaac, C.L., Mayes, A.R., Hodges, J.R., Montaldi, D., Cezayirli,
E., Roberts, N., Hadley, D., 1998. Face processing impairments
after encephalitis: amygdala damage and recognition of fear.
Neuropsychologia 36, 59–70.
Bullmore, E., Brammer, M., Williams, S.C., Janot, N., David, A.,
Mellers, J., Howard, R., Sham, P., 1996. Statistical methods of
estimation and inference for functional MR image analysis.
Magnetic Resonance in Medicine 35, 261–277.
Bullmore, E., Brammer, M., Rabe-Hesketh, S., Curtis, V.A., Morris,
R.G., Williams, S.C., Sharma, T., McGuire, P., 1999. Methods
for diagnosis and treatment of stimulus-correlated motion in
generic brain activation studies using fMRI. Human Brain
Mapping 7, 38–48.
Burke, A., Heuer, F., Reisberg, D., 1992. Remembering emotional
events. Memory & Cognition 20, 277–290.
Bush, G., Luu, P., Posner, M.I., 2000. Cognitive and emotional
influences in anterior cingulate cortex. Trends in Cognitive
Sciences 4, 215–222.
Cahill, L., Haier, R.J., Fallon, J., Alkire, M.T., Tang, C., Keator, D.,
Wu, J., McGaugh, J.L., 1996. Amygdala activity at encoding
correlated with long-term free recall of emotional information.
Proceedings of the National Academy of Sciences of the United
States of America 93, 8016–8021.
Cahill, L., Haier, R.J., White, N.S., Fallon, J., Kilpatrick, L.,
Lawrence, C., Potkin, S.G., Alkire, M.T., 2001. Sex-related
difference in amygdala activity during emotionally influenced
memory storage. Neurobiology of Learning and Memory 75, 19.
Canli, T., Desmond, J.E., Zhao, Z., Glover, G., Gabrieli, J.D., 1998.
Hemispheric asymmetry for emotional stimuli detected with
fMRI. NeuroReport 9, 3233–3239.
Canli, T., Zhao, Z., Brewer, J., Gabrieli, J.D., Cahill, L., 2000.
Event-related activation in the human amygdala associates with
later memory for individual emotional experience. Journal of
Neuroscience 20 (RC99), 1–5.
Canli, T., Desmond, J.E., Zhao, Z., Gabrieli, J.D., 2002. Sex
differences in the neural basis of emotional memories. Proceed-
ings of the National Academy of Sciences of the United States
of America 99, 107879–107894.
Coltheart, M., 1981. The MRC Psycholinguistic Database. Medical
Research Council, UK.
Davidson, R.J., 2000. The functional neuroanatomy of affective
style. In: Lane, R.D., Nadel, L. (Eds.), Cognitive Neuroscience
of Emotion. Oxford University Press, New York.
Dolan, R.J., Lane, R.D., Chua, P., Fletcher, P., 2000. Dissociable
temporal lobe activations during emotional episodic memory
retrieval. NeuroImage 11, 203–209.
Dolcos, F., LaBar, K.S., Cabeza, R., 2004. Interaction between the
amygdala and the medial temporal lobe memory system predicts
better memory for emotional events. Neuron 42, 855–863.
Erk, S., Kiefer, M., Grothe, J., Wunderlich, A.P., Spitzer, M.,
Walter, H., 2003. Emotional context modulates subsequent
memory effect. NeuroImage 18, 439–447.
Gallagher, M., Chiba, A.A., 1996. The amygdala and emotion.
Current Opinion in Neurobiology 6, 221–227.
N. Medford et al. / Psychiatry Research: Neuroimaging 138 (2005) 247–258258
Hamner, M.B., Lorberbaum, J.P., George, M.S., 1999. Potential role
of the anterior cingulate cortex in PTSD: review and hypothesis.
Depression and Anxiety 9, 1–14.
Heilman, K.M., 2000. Emotional experience: a neurological model.
In: Lane, R.D., Nadel, L. (Eds.), Cognitive Neuroscience of
Emotion. Oxford University Press, New York, pp. 328–344.
Kensinger, E.A., Corkin, S., 2003. Memory enhancement for
emotional words: are emotional words more vividly remembered
than neutral words? Memory & Cognition 31, 1169–1180.
Kensinger, E.A., Brierley, B., Medford, N., Growdon, J.H., Corkin,
S., 2002. Effects of normal aging and Alzheimer’s disease on
emotional memory. Emotion 2, 118–134.
LeDoux, J., 1998. The Emotional Brain. Simon and Schuster,
New York.
Lewis, P.A., Critchley, H.D., 2003. Mood-dependent memory.
Trends in Cognitive Sciences 7, 431–433.
Libkuman, T.M., Stabler, C.L., Otani, H., 2004. Arousal, valence,
and memory for detail. Memory 12, 237–247.
Loftus, E.F., Burns, T., 1982. Mental shock can reproduce
retrograde amnesia. Memory & Cognition 10, 318–323.
Maddock, R.J., 1999. The retrosplenial cortex and emotion: new
insights from functional neuroimaging of the human brain.
Trends in Neurosciences 22, 310–316.
Maratos, E.J., Dolan, R.J., Morris, J.S., Henson, R., Rugg, M.,
2001. Neural activity associated with episodic memory for
emotional context. Neuropsychologia 39, 910–920.
McGaugh, J.L., Cahill, L., Roozendaal, B., 1996. Involvement of
the amygdala in memory storage: interaction with other brain
systems. Proceedings of the National Academy of Sciences of
the United States of America 93, 13508–13514.
Mori, E., Ikeda, M., Hirono, N., Kitagaki, H., Imamura, T.,
Shimomura, T., 1999. Amygdalar volume and emotional
memory in Alzheimer’s disease. American Journal of Psychiatry
156, 216–222.
Morris, J.S., Friston, K.J., Dolan, R.J., 1998. Experience-dependent
modulation of tonotopic neural responses in human auditory
cortex. Proceedings of the Royal Society of London. Series B,
Biological Sciences 265, 649–657.
Nakayama, K., Joseph, J.S., 1998. Attention, pattern recognition,
and pop-out in visual search. In: Parasuraman, R. (Ed.), The
Attentive Brain. MIT Press, Cambridge, MA.
O’Carroll, R.E., Drysdale, E., Cahill, L., Shajahan, P., Ebmeier,
K.P., 1999. Stimulation of the noradrenergic system enhances
and blockade reduces memory for emotional material in man.
Psychological Medicine 29, 1083–1088.
Ogawa, S., Lee, T.M., Kay, A.R., Tank, D.W., 1990. Brain magnetic
resonance imaging with contrast dependent blood oxygenation.
Proceedings of the National Academy of Sciences of the United
States of America 87, 8868–8872.
Phelps, E.A., Anderson, A.K., 1997. Emotional memory: what does
the amygdala do? Current Biology 7, R311–R314.
Phillips, M.L., Young, A.W., Senior, C., Brammer, M., Andrew, C.,
Calder, A.J., Bullmore, E., Perrett, D.I., Rowland, D., Williams,
S.C., Gray, J.A., David, A.S., 1997. A specific neural substrate
for perceiving facial expressions of disgust. Nature 389, 495–
498.
Sierra, M., Berrios, G.E., 1999. Flashbulb memories and other
repetitive images: a psychiatric perspective. Comprehensive
Psychiatry 40, 115–125.
Simmons, A., Moore, E., Williams, S.C.R., 1999. Quality control
for functional magnetic resonance imaging using automated data
analysis and Shewart charting. Magnetic Resonance in Medicine
41, 1274–1278.
Smith, A.P.R., Henson, R.N.A., Dolan, R.J., Rugg, M.D., 2004.
fMRI correlates of the episodic retrieval of emotional contexts.
NeuroImage 22, 868–878.
Talairach, J., Tournoux, P., 1988. Co-planar Stereotactic Atlas of the
Human Brain. Thieme, Stuttgart.
Worthen, J.B., Garcia-Rivas, G., Green, C.R., Vidos, R.A., 2000.
Tests of a cognitive resource-allocation account of the bizarre-
ness effect. Journal of General Psychology 127, 117–144.