Journal of Experimental Psychology: General © 2017 The Author(s)

2018, Vol. 147, No. 2, 243–256 0096-3445/18/$12.00 http://dx.doi.org/10.1037/xge0000356

Negative Emotional Content Disrupts the Coherence of Episodic Memories

James A. Bisby Aidan J. Horner

Institute of Cognitive Neuroscience and Institute University of York
of Neurology, University College London

Daniel Bush and Neil Burgess

Institute of Cognitive Neuroscience and Institute of Neurology, University College London

Events are thought to be stored in episodic memory as coherent representations, in which the constituent
elements are bound together so that a cue can trigger reexperience of all elements via pattern completion.
Negative emotional content can strongly influence memory, but opposing theories predict strengthening
or weakening of memory coherence. Across a series of experiments, participants imagined a number of
person-location-object events with half of the events including a negative element (e.g., an injured
person), and memory was tested across all within event associations. We show that the presence of a
negative element reduces memory for associations between event elements, including between neutral
elements encoded after a negative element. The presence of a negative element reduces the coherence
with which a multimodal event is remembered. Our results, supported by a computational model, suggest
that coherent retrieval from neutral events is supported by pattern completion, but that negative content
weakens associative encoding which impairs this process. Our findings have important implications for
understanding the way traumatic events are encoded and support therapeutic strategies aimed at restoring
associations between negative content and its surrounding context.

Keywords: episodic memory, emotion, hippocampus, pattern completion, memory coherence

Supplemental materials: http://dx.doi.org/10.1037/xge0000356.supp

Episodic memories typically comprise complex events that in- to mind (Tulving, 1983). For this holistic episodic retrieval to
clude multiple elements such as the people we meet, the objects we occur, the elements that form an event must be bound together,
interact with, and the locations in which those encounters take allowing for their subsequent reinstatement. The hippocampus
place. Their retrieval is characterized by a rich recollective expe- plays an essential role as a convergence zone, binding together the
rience in which all of the event’s constituent elements are brought separate elements of an event (Cohen & Eichenbaum, 1993;
Damasio, 1989; Davachi, 2006; Eichenbaum, Yonelinas, & Ran-
ganath, 2007; O’Keefe & Nadel, 1978). Presentation of a partial
cue will lead to the reinstatement of all event elements through a
This article was published Online First September 14, 2017.
James A. Bisby, Institute of Cognitive Neuroscience, Institute of
process of pattern completion in the hippocampus (Marr, 1971;
Neurology, University College London; Aidan J. Horner, Department of McClelland, McNaughton, & O’Reilly, 1995; Norman & O’Reilly,
Psychology, University of York; Daniel Bush and Neil Burgess, Insti- 2003). While negative emotion clearly impacts memory for an
tute of Cognitive Neuroscience, Institute of Neurology, University event (Bradley, Greenwald, Petry, & Lang, 1992; Brown & Kulik,
College London. 1977; Cahill et al., 1996; Cahill & McGaugh, 1995; Christianson,
This work was supported by a Wellcome Trust Fellowship (202805/Z/ 1992), the way in which it affects the binding of event elements
16/Z) and European Research Council grant (ERC-2015-AdG, 694779) into coherent memory representations remains controversial.
awarded to Neil Burgess. Aidan J. Horner is supported by a grant from the
According to a “general facilitation” account, strong emotional
Wellcome Trust (204277/Z/16/Z). Some of the data in this article have
been presented at the International Meeting of the Psychonomics Society
content will strengthen all aspects of memory for the event, en-
(2016). hancing its availability at retrieval (McGaugh, 2003; Rubin, Boals,
This article has been published under the terms of the Creative Com- & Berntsen, 2008; Talarico, LaBar, & Rubin, 2004). In contrast, a
mons Attribution License (http://creativecommons.org/licenses/by/3.0/), ‘dual representation’ account argues that negative emotion will
which permits unrestricted use, distribution, and reproduction in any me- affect different aspects of memory in opposing ways (Brewin,
dium, provided the original author and source are credited. Copyright for Gregory, Lipton, & Burgess, 2010; Jacobs & Nadel, 1998): poten-
this article is retained by the author(s). Author(s) grant(s) the American tially enhancing memory for the negative content itself but weak-
Psychological Association the exclusive right to publish the article and
ening associations between the negative content and its context
identify itself as the original publisher.
Correspondence concerning this article should be addressed to James A. (i.e., other neutral aspects of the event). These two views therefore
Bisby and Neil Burgess, UCL Institute of Cognitive Neuroscience, 17 make competing predictions, one arguing for strengthened mem-
Queen Square, London WC1N 3AZ, UK. E-mail: [email protected] or ory for the whole negative event, and the other for more frag-
[email protected] mented memory due to impaired associations between elements.

243
244 BISBY, HORNER, BUSH, AND BURGESS

When exposed to a negative experience, participants often re- idation processes. Recognition memory was tested by cueing with
port greater vividness, accuracy and confidence during recall (Ca- a single element and asking whether the cue was old or new
hill et al., 1996; Kensinger & Corkin, 2003), and show a recollec- (recognition was only tested in Experiment 1). Associative memory
tion advantage for the negative details (Rimmele, Davachi, Petrov, for each pair of elements was assessed using a six-alternative forced
Dougal, & Phelps, 2011; Sharot, Delgado, & Phelps, 2004; Sharot choice with participants required to select the element that had been
& Yonelinas, 2008). These findings are consistent with a general originally paired with the cue (Figure 1a). To explore potential neural
facilitation account, and might reflect boosting of hippocampal mechanisms that might underpin the pattern of behavioral data, we
encoding or consolidation via fear-related processing in the constructed a simple computational model of associative learning
amygdala (McGaugh, 2004). However, the mnemonic advantage (Marr, 1971).
seems specific to the negative items themselves, whereas memory
for the associations between items or an item and its context is Experiment 1
often impaired when negative items are present (Bisby & Burgess,
2014; Kensinger & Schacter, 2006; Madan, Caplan, Lau, & Fuji-
Method
wara, 2012; Mather, 2007; Mather & Knight, 2008; Touryan,
Marian, & Shimamura, 2007). According to the dual representa- Participants. A total of 17 participants (7 males) with a mean
tion account, negative emotion down-modulates hippocampal pro- age of 23.36 years (SD ⫽ 3.88) were recruited from the university
cessing, disrupting associative/relational binding, while amygdalar student population. A power analysis based on effect sizes reported
up-modulation facilitates encoding of the negative content of an in previous studies (Bisby et al., 2016; Bisby & Burgess, 2014)
event (Bisby, Horner, Horlyck, & Burgess, 2016; Jacobs & Nadel, provided an approximate sample size required for each experiment
1998). (Experiment 1, N ⫽ 18; Experiments 2 and 3, N ⫽ 23; power ⫽
Recent studies have shown that our performance over multiple 0.80, ␣ ⫽ .05). All experiments were approved by the University
trials in retrieving different elements from the same event is College London Ethics Committee and participants provided writ-
statistically related, providing behavioral evidence that episodic ten informed consent prior to taking part in the study. Following
memory reflects “coherent representations” supported by pattern test, participants were debriefed and paid.
completion (Horner & Burgess, 2013, 2014). In these studies, Materials. Stimuli included a total of 216 images consisting
events involving triplets (person, object and location) were either of 72 from each category of locations (e.g., an office), people and
encoded simultaneously or built up over a number of encoding everyday objects (e.g., a telephone). Images of people included 36
trials in which all interelement associations were learned. A sub- neutral and 36 negative pictures (e.g., an injured person). All
sequent memory test of all associations quantified the holistic images of people were acquired from the International Affective
nature of episodic recollection in terms of the statistical depen- Picture System (Lang, Bradley, & Cuthbert, 2005) and the Nencki
dency between retrievals from the same event—that is, when Affective Picture System (Marchewka, Zurawski, Jednoróg, &
participants were successful in one cued retrieval from an event, Grabowska, 2013). We attempted to control for potential differ-
they were more likely to be successful in other cued retrievals ences across stimuli by matching negative and neutral images of
from the same event. This highlights the importance of binding people. That is, we selected images that only displayed the head
together the elements of an event to form a coherent memory and shoulders of a person (see Figure 1a). Images of locations were
representation that can support holistic retrieval. In a further study, acquired from the Internet and objects taken from a database of
statistical dependency was found to be related to hippocampal images used in previous research (Horner & Henson, 2008).
activity and incidental reinstatement of all event elements in neo- Procedure. Participants performed a single session involving
cortical regions (Horner, Bisby, Bush, Lin, & Burgess, 2015). encoding and test. At encoding, participants were presented with
We aimed to investigate the effects of negative emotion on the 36 ‘events’ (18 neutral and 18 negative) with each event including
binding of different event elements into coherent mnemonic rep- a location, person and object (Figure 1a). For negative events, only
resentations. Following previous studies (Horner et al., 2015; images of people were negative in valence, while the location and
Horner & Burgess, 2013, 2014), we assessed the coherence of object were both neutral. Events were randomly generated across
retrievals of within-event associations from multielement events participants. During a single encoding trial, all three elements from
involving neutral and negative elements. If negative emotion im- an event (location, person, and object) were simultaneously pre-
pairs associative binding, it should reduce the coherence of mem- sented on the screen. The three images were presented in a trian-
ories for negative events as evidenced by reduced statistical de- gular configuration with one image presented in the middle of the
pendency between retrievals from the same event. If negative screen above the center and the remaining two images presented to
emotion strengthens all aspects of memory, including associations the left and right of center in the lower half of the screen. Place-
between elements, we should see increased coherence. ment of each element within events was randomized across trials.
We performed three experiments in which participants were Each event was displayed for 6 s and was followed by fixation
required to encode events comprising a person, location and ob- cross for a further 2 s. During each trial, participants were in-
ject, each presented as images on screen. For half of the events, the structed to vividly imagine an event involving all three elements
person was a negative image, such as an injured individual. These presented on screen and to be a creative as possible.
events were either presented simultaneously as triplets (Experi- At test, participants completed a recognition and associative
ment 1) or sequentially as overlapping pairs (Experiments 2 and 3) memory test for each single element and its paired associates
with participants instructed to imagine the elements interacting. (Figure 1a). That is, every association was tested in both directions
Experiment 3 involved a 24hr delay between study and test to resulting in a total of 216 tests trials (plus a further 216 new cue
allow detection of any effects that might be supported by consol- trials). Each trial started with the presentation of a cue image,
 EMOTION AND MEMORY COHERENCE 245

Figure 1. Memory for events encoded with simultaneously presented elements. (a) In Experiment 1, each
encoding trial included three separate event elements (location, person and object) presented simultaneously and
followed by a 2 s intertrial interval (ITI). At retrieval, a cue image was presented and participants were required
to respond whether the image was old or new. If old, the participant was then presented with six options and
instructed to select the image that had been originally presented with the cue image at encoding (see Methods).
(b) Recognition accuracy for each element type (hits minus misses) was compared between neutral and negative
events (collapsed across first and second presentation during test). (c) Associative memory performance for
neutral and negative events split by the different element pair types for each event (collapsed across testing
direction; chance performance ⫽ 0.17). (d) Dependency in the data was compared to independent and dependent
models across neutral and negative events. Error bars represent standard error; NS ⫽ not significant. ⴱⴱ p ⫽ .001.
ⴱⴱⴱ
p ⬍ .001. See the online article for the color version of this figure.

which could either be a location, person or object. For example, on Analysis of memory coherence. The statistical dependency
one trial the participant could be cued with an image of a person (see supplementary information for more details) between the
and shown six objects. Following interleaved trials from other retrieval of associations from the same event was assessed as in
events, the participant would be cued with the same person image previous reports (Horner et al., 2015; Horner & Burgess, 2013,
and shown six locations to choose from. The six options on the 2014). For each participant, we created 2 ⫻ 2 contingency
alternative forced choice task always comprised (a) the correct tables for the retrieval of two elements when cued by the
association, (b) two foil images taken from events of the same remaining within-event element (ABAC; e.g., cueing with a
emotional category as the correct response, and (c) three foil person to retrieve the associated location and object), as well as
images taken from events of the opposite emotional category as the for retrieving an element when cued by its two associated
correct response. This was done to control for emotion at retrieval elements (BACA; e.g., retrieving a person when cued by the
with all retrieval trials including three neutral and three negative associated location and object). This resulted in six 2 ⫻ 2 tables
options. A further 36 images from each element category were per participant across each of the experimental conditions. To
added as new items (including 18 neutral and 18 negative new examine dependency, we took the proportion of events in which
person images). On presentation of the cue, participants were both associations were either correctly or incorrectly retrieved.
required to respond OLD or NEW via button press. If the cue was We then averaged this measure across contingency tables for
an old element, participants were then presented with six other each condition.
images and participants were instructed to select the one that had We also created independent and dependent models of retrieval
been paired with the cue image at encoding. Participants were from each contingency table (see Table 1 for details on how these
given a maximum of 10 s to make a response. Each trial ended models were calculated). The independent model predicts the
with fixation cross that remained on screen for 2 s. amount of dependency in relation to the participant’s mean level of
246 BISBY, HORNER, BUSH, AND BURGESS

Table 1
Contingency Tables for Independent and Dependent Models Giving the Frequency (Over Events)
of the Four Combinations of Correct or Incorrect Retrievals of Elements B and C When Cued
With Element A

Retrieval of element (B)

Retrieval of
element (C) Correct (PAB) Incorrect (1 ⫺ PAB)

Independent model
Correct (PAC) 兺i⫽1
N
PABPAC 兺i⫽1
N
PAC(1 ⫺ PAB)
Incorrect (1 ⫺ PAC) 兺i⫽1
N
PAB(1 ⫺ PAC) 兺i⫽1
N
(1 ⫺ PAB)(1 ⫺ PAC)
Dependent model
Correct (PAC) 兺i⫽1
N ´i
P ABP´iAC 兺i⫽1
N ´i
P AC(1 ⫺ P´iAB)
Incorrect (1 ⫺ PAC) 兺i⫽1P AB(1 ⫺ P´iAC)
N ´i
兺i⫽1(1 ⫺ P´iAB)(1 ⫺ P´iAC)
N

Note. Dependent model replaces the probability of correctly recalling B when cued with A (across all events;
PAB) with P´iAB ⫽ EiAB (PAB – PG/c) ⫹ PG/c where the episodic factor EAB i
reflects performance on event i
relative to other events (based on retrievals other than B and C cued by A), PG is the probability of guessing,
and c ⫽ 6 is the number of choices in the test trial. PAC and P´AC defined similarly. The dependency model
equates to the independent model if the episodic factors are set to 1.

performance for all associations across events, estimating the for better associative memory performance for object-location
amount of statistical dependency expected if retrieval success for pairs compared to both person-object, t(16) ⫽ 2.29, p ⫽ .06, d ⫽
specific cue-test pairs (e.g., cue location, test person) is indepen- 0.55 and location-person pairs, t(16) ⫽ 1.84, p ⫽ .08, d ⫽ 0.44.
dent of retrieval success for other cue-test pairs (e.g., cue location, Importantly, there was no interaction between emotion and pair-
test object). The dependent model estimates retrieval performance type (F [2, 32] ⫽ 0.01, p ⫽ .99, ␩P2 ⬍ 0.01) suggesting that the
for a given question adjusted by the mean performance over reduction in associative accuracy for negative events was consis-
questions for that event (the episodic factor E). This allows us to tent across all pair types (and not specific to pairs including a
predict the maximal dependency based on a participant’s overall negative person element). Indeed, a direct comparison of associa-
performance, amount of guessing, and overall variance across all tive accuracy for location-object pairs (comprising neutral ele-
events. ments for both neutral and negative events) showed reduced per-
formance for negative compared to neutral events, t(16) ⫽ 4.39,
Results p ⬍ .001, d ⫽ 1.06. This reduction in associative memory for
negative events was also seen for person-object, t(16) ⫽ 3.51,
Recognition memory performance. As each element from an
p ⬍ .01, d ⫽ 0.85 and location-person pairs, t(16) ⫽ 3.03, p ⬍
event was presented twice at test (e.g., location served as a cue for
.01, d ⫽ 0.73. In summary, there was a general reduction in
both object and person associative test trials), recognition perfor-
associative memory performance for negative events and this
mance was assessed using a 2 ⫻ 2 ⫻ 3 repeated measures ANOVA
reduction was evident across all pairs that formed part of the
(neutral vs. negative, first vs. second presentation, location vs. person
vs. object cue; Figure 1b). Recognition memory performance was negative event.
high across all conditions (see Table S1 for a full breakdown of hits Memory coherence. Coherence was assessed by constructing
and misses across conditions). Analysis of recognition memory per- contingency tables for retrieving two elements when cued with the
formance (hits minus false alarms) showed no significant main effects third element, and retrieving one element when cued by the other
of emotion (F [1, 16] ⫽ 2.79, p ⫽ .12, ␩P2 ⫽ 0.15), presentation two elements across separate retrieval trials. We then calculated
(F [1, 16] ⫽ 1.07, p ⫽ .32, ␩P2 ⫽ 0.06) or cue type (F [2, 32] ⫽ dependency (D) in the data by taking the proportion of events
0.52, p ⫽ .60, ␩P2 ⫽ 0.03) and no interactions of Emotion ⫻ Cue where elements were both correctly or incorrectly retrieved (see
Type (F [2, 32] ⫽ 0.65, p ⫽ .53, ␩P2 ⫽ 0.04), Emotion ⫻ Methods and supplementary information for details). This depen-
Presentation (F [1, 16] ⫽ 0.77, p ⫽ .39, ␩P2 ⫽ 0.05), Presenta- dency (D) was compared to the amount of dependency predicted if
tion ⫻ Cue-Type (F [2, 32] ⫽ 0.78, p ⫽ .47, ␩P2 ⫽ 0.05) or retrievals from the same event were completely independent (Di)
Emotion ⫻ Presentation ⫻ Cue-Type (F [2, 32] ⫽ 0.19, p ⫽ .83, or dependent (Dd; see Methods and supplementary information for
␩P2 ⫽ 0.01). details on how the models were constructed). We then compared
Associative memory performance. As each event consisted the dependency in the data with both independent and dependent
of three separate associations between elements (location-object; models separately for neutral and negative events (Figure 1d).
person-location; object-person), we analyzed associative memory Performing a 2 ⫻ 3 ANOVA (emotion: neutral or negative, de-
performance across these pairs (collapsed across test direction) pendency-measure: D, Di, Dd), we found a significant Emotion ⫻
using a 2 ⫻ 3 ANOVA (emotion; pair-type; Figure 1c). We saw a Dependency-Measure interaction (F [2, 32] ⫽ 3.72, p ⬍ .05, ␩P2 ⫽
significant main effect of emotion reflecting better associative 0.20). Analyzing neutral events separately, we saw evidence of
accuracy for neutral events compared to negative events (F [1, greater dependency in the data compared to the Independent model
16] ⫽ 17.10, p ⫽ .001, ␩P2 ⫽ 0.52). We also saw a main effect of (D ⬎ Di, t[16] ⫽ 4.07, p ⫽ .001, d ⫽ 0.99) but no difference from
pair type (F [2, 32] ⫽ 3.95, p ⬍ .05, ␩P2 ⫽ 0.20) with a tendency the Dependent model, t(16) ⫽ 1.46, p ⫽ .16, d ⫽ 0.35.
 EMOTION AND MEMORY COHERENCE 247

For negative events, we again saw greater dependency in the data coded events consisting of a location, person and object. However,
compared to the Independent model (D ⬎ Di, t(16) ⫽ 4.20, p ⫽ .001, rather than simultaneously presenting all three items, participants
d ⫽ 1.02) but this was less than that in the Dependent model (D ⬍ Dd, viewed all event components as paired associates across three
t(16) ⫽ 4.77, p ⬍ .001, d ⫽ 1.16). A direct comparison between encoding blocks. The number of events at encoding was also
neutral and negative events showed a greater decrease in dependency increased to 72 (36 neutral and 36 negative events). For encoding,
relative to the dependent model (D-Dd; t[16] ⫽ 4.47, p ⬍ .001, d ⫽ we presented events as paired associates across three blocks using
1.08; no difference between neutral and negative events on the in- two different encoding orders. For example, half of the events (18
crease in dependency relative to the independent model; D-Di; neutral and 18 negative) were encoded using the order location-
t(16) ⫽ 0.48, p ⫽ .64, d ⫽ 0.11). Memory coherence was therefore object, person-location, and object-person (person-last; Figure 2a).
reduced for negative relative to neutral events. The other half of the events was presented using the order object-
person, person-location, and location-object (person-first). Using
Experiment 2 these two orders allowed us to manipulate whether the first en-
coded pair of a negative event consisted of two neutral elements or
Experiment 1 showed that the presence of a negative element
a negative and neutral element. All paired associates were ran-
at encoding reduced subsequent memory for all within-event
domized within each encoding block. Each paired associate was
associations. Both neutral and negative events showed evidence
presented for 6 s and was followed by a 2 s fixation inter trial
of dependency when compared to the Independent model. How-
interval. Participants were required to imagine the two items
ever, dependency for negative events was lower than that pre-
interacting in a meaningful way, and the overlapping nature of
dicted by the Dependent model, supported by an interaction
pairs across blocks was not mentioned. After encoding, partici-
between emotion and the dependency-measure, suggesting that
pants completed the memory test (Figure 2a). The test was the
negative elements at encoding reduce memory coherence. Pre-
same as described in Experiment 1 except that the recognition
vious studies have shown that events are not only stored as
component was omitted (resulting in a total of 432 associative test
coherent representations when all elements are presented simul-
trials). Therefore, on each trial participants were cued with one of
taneously (Horner & Burgess, 2013) but also when presented
the previously seen images and given a six alternative forced
sequentially as overlapping paired associates (Horner et al., 2015;
choice to try and remember the paired associate.
Horner & Burgess, 2014; Milivojevic, Vicente-Grabovetsky, &
Doeller, 2015; Schlichting & Preston, 2015). We next examined
how the presence of negative items might alter memory Results
coherence when events were built up over a sequence of over-
Associative memory. We analyzed associative memory per-
lapping encoding trails, to separate the effects of associated
formance for neutral and negative events for the three pair-types
negative elements from the effect of negative elements on the
(location-object, person-location, object-person) under the two en-
screen during encoding. In addition, we assessed whether co-
coding orders (person-last; person-first) in a 2 ⫻ 2 ⫻ 3 ANOVA
herence was further affected by negative pairs being encoded
(emotion, encoding order, pair-type; Figure 2b). Memory for as-
either early or late during the sequence of pairs forming an
sociations was generally worse for events that included a negative
event.
element (F [1, 25] ⫽ 60.07, p ⬍ .001, ␩P2 ⫽ 0.71). Performance
Participants were required to learn events comprising three
also varied across pair-type (F [1, 50] ⫽ 16.89, p ⫽ .001, ␩P2 ⫽
elements (location, person and object) presented sequentially as
0.40) with better memory for the associations presented first com-
overlapping paired associates over a series of encoding trials
pared to the second presented pair, which was always person-
(e.g., A-B, B-C, A-C) interleaved with trials from other events.
location (location-object in the person-last encoding order, t(25) ⫽
Participants thus associated all within-event elements with each
5.16, p ⬍ .001, d ⫽ 1.01; person-object in person-first encoding
other despite never seeing all three at once (Horner & Burgess,
order, t(25) ⫽ 4.99, p ⬍ .001, d ⫽ 0.98). Performance across
2014). In addition, we manipulated the order in which the
pair-type also varied more for negative events (Emotion ⫻ Pair-
events were encoded. For half of the events, the location-object
Type interaction, F (1.62, 40.44) ⫽ 12.30, p ⬍ .001, ␩P2 ⫽ 0.33;
pair was the first encoded (we refer to this condition as person-
Greenhouse-Giesser corrected). This interaction reflected worse
last). For the other half of the events, the location-object pair
memory for associations from negative events that involved a
was the last pair encoded (we refer to this condition as person-
person element relative to location-object pairs. That is, there
first). As the negative element of an event was always the
was a greater memory reduction for person-location relative to
person element, negative events were therefore encoded with
location-object pairs (t[25] ⫽ 4.77. p ⬍ .001, d ⫽ 0.94) and
either the first or last study trial comprising a “pure-neutral”
object-person pairs relative to object-location pairs, t(25) ⫽ 3.00,
pair.
p ⬍ .01, 0.59 for negative compared to neutral event, a difference
that was greater in the person-last encoding order (Order ⫻ Pair-
Method
Type, F [1.33, 33.21] ⫽ 13.86, p ⬍ .001, ␩P2 ⫽ 0.36).
Experiment 2 was identical to Experiment 1 with the following We were interested to see whether negative items might influ-
changes: ence memory for overlapping neutral items from the same event
Participants. A total of 26 participants (10 males) with a even when encoded in absence of the negative element (i.e.,
mean age of 24.16 years (SD ⫽ 3.16) were recruited from the location-object pairs). While there was no difference in memory
university student population. performance for the location-object pairs between neutral and
Procedure. The materials used in Experiment 2 were exactly negative events when studied under the person-last encoding order
the same as Experiment 1. As in Experiment 1, participants en- (when location-object is encoded before pairs involving negative
248 BISBY, HORNER, BUSH, AND BURGESS

Figure 2. Memory for events encoded as overlapping pairs. (a) For Experiment 2, each event was encoded over
three separate blocks (i.e., separated by encoding trials for other events). Events were either encoded with the
location-object pair presented on the first encoding trial (person-last encoding order) or as the final encoding trial
(person first encoding order). Associative memory was testing in a similar way to Experiment 1. (b) Associative
memory performance for each encoded pair across neutral and negative events split by the person-last encoding
order (upper panel, location-object studied first) and person-first encoding order (lower panel, object-person
studied first). Note that the person image was always the negative element within negative events.
(c) Dependency for the data and the independent and dependent models, across neutral and negative events split
by person-last and person-first encoding orders. Note the overall decrease in dependency from neutral to negative
events. Error bars represent standard error; NS ⫽ not significant. ⴱⴱ p ⱕ .01. ⴱⴱⴱ p ⬍ .001. See the online article
for the color version of this figure.

elements, t(25) ⫽ 0.99, p ⫽ .33, d ⫽ 0.19), we found reduced supplementary information for details). A 2 ⫻ 2 ⫻ 3 repeated
associative accuracy for location-object pairs that were part of a measures ANOVA (emotion; encoding-order; dependency-mea-
negative events when studied during the person-first encoding sure: D, Di, Dd) showed a significant three-way interaction (F [2,
order (when location-object is encoded last, t(25) ⫽ 2.64, p ⫽ .01, 50] ⫽ 3.43, p ⬍ .05, ␩P2 ⫽ 0.12). To further analyze this interac-
d ⫽ 0.52). Further, the difference in location-object accuracy from tion, we performed separate 2 ⫻ 3 ANOVAs (encoding-order,
the person-last order to the person-first order showed a numeri- dependency-measure) for neutral and negative events. For neutral
cally greater reduction for negative compared to neutral events, events, we saw a significant main effect of dependency-measure
though this was not statistically significant, t(25) ⫽ 1.72, p ⫽ .09, (F [2, 50] ⫽ 24.46, p ⬍ .001, ␩P2 ⫽ 0.50) but no main effect of
d ⫽ 0.34. encoding order (F [1, 19] ⫽ .001, p ⫽ .98, ␩P2 ⬍ 0.01) or Order ⫻
In summary, associative memory was consistently reduced by Dependency-Measure interaction (F [2, 38] ⫽ 0.70, p ⫽ .50, ␩P2 ⬍
the presence of a negative element. We also found better associa- 0.01). Further analysis showed greater dependency (Figure 2c) for
tive accuracy for the initial pair from an event across both neutral neutral events compared to the Independent model (D ⬎ Di,
and negative events, possibly due to event-related primacy effects. t[25] ⫽ 5.98, p ⬍ .001, d ⫽ 1.17) but no difference when
Interestingly, further analysis revealed that memory for the neutral compared to the Dependent model, t(25) ⫽ 1.56, p ⫽ .13, d ⫽
pairs from a negative event was reduced when those items had 0.31.
been previously paired with a negative element (but not when they For negative events, this analysis showed a significant interac-
were subsequently paired with negative elements). tion of Encoding-Order ⫻ Dependency-Measure (F [2, 50] ⫽ 5.40,
Memory coherence analysis. Dependency was calculated in p ⬍ .01, ␩P2 ⫽ 0.18) and main effects of dependency-measure F [2,
the data (D) and compared to the dependencies Di and Dd as 50] ⫽ 30.76, p ⬍ .001, ␩P2 ⫽ 0.55; main effect of encoding-order
predicted by Independent and Dependent models (see Methods and F [1, 25] ⫽ 1.09, p ⫽ .31, ␩P2 ⫽ 0.04). Events encoded under the
 EMOTION AND MEMORY COHERENCE 249

person-last order (pure-neutral pair presented first) showed evi- Method

dence of increased dependency (Figure 2c) compared to the Inde-
Experiment 3 was identical to Experiment 2 with the following
pendent model (D ⬎ Di, t[25] ⫽ 4.77, p ⫽ .001. d ⫽ 0.94) but also
changes:
less dependency than the Dependent model (D ⬍ Dd, t[25] ⫽ 2.85,
Participants. A total of 27 participants (11 males) with a
p ⬍ .01, d ⫽ 0.56). Negative events encoded using the person-first
mean age of 23 years (SD ⫽ 2.56) were recruited from the
order (pure-neutral pair presented last) showed no evidence of
university student population.
dependency, with no difference between the data and Independent
Procedure. The procedure was exactly the same as that used
model, t[25] ⫽ 1.51, p ⫽ .14, d ⫽ 0.30 and significantly less
for Experiment 2 with the only difference being that test took place
dependency than the Dependent model (D ⬍ Di, t[25] ⫽ 4.41, p ⬍ 24 hours after encoding.
.001, d ⫽ 0.86).
To directly compare dependency for negative events between Results
the two encoding orders, we calculated dependency relative to the
Independent model (D-Di). Importantly, this analysis showed Associative memory performance. We examined associative
memory accuracy for the encoded pairs across neutral and negative
greater dependency for the person-last encoding order, t(25) ⫽
events split by encoding order (Figure 3a and supplementary
2.40, p ⬍ .05, d ⫽ 0.47, while a similar comparison for neutral
information for details). Performing a 2 ⫻ 2 ⫻ 3 ANOVA (emo-
events showed no significant difference between encoding orders,
tion, encoding order, pair-type), we saw a similar pattern of per-
t(25) ⫽ 0.50, p ⫽ .62, d ⫽ 0.10; a difference reflected in the
formance to Experiment 2 (Figure 3a) with worse associative
Emotion ⫻ Order ⫻ Dependency-Measure interaction. Similarly,
accuracy for negative events (F [1, 26] ⫽ 26.49, p ⬍ .001; ␩P2 ⫽
we also performed a direct comparison on the dependency increase
0.51). Again, we found that performance varied across pair-type
relative to the independent model (D-Di) between neutral and (F [2, 52] ⫽ 12.15, p ⬍ .001, ␩P2 ⫽ 0.32) and this was particularly
negative events. Supporting the reduction in dependency for neg- evident for negative events (Emotion ⫻ Pair-Type, F [2, 52] ⫽
ative events, this analysis found significantly greater dependency 5.99, p ⬍ .01; ␩P2 ⫽ 0.19) with worse memory for pairs that
relative to the independent model for neutral compared to negative included the person element compared to location-object for neg-
events when studied under the person-first encoding order, t(25) ⫽ ative compared to neutral events (person-location, t[26] ⫽ 2.38,
2.93, p ⬍ .01, d ⫽ 0.57. We saw no difference between neutral and p ⬍ .05, d ⫽ 0.46; object-person, t(26) ⫽ 3.57, p ⫽ .001, d ⫽
negative studied under the person-last encoding order, t(25) ⫽ 0.69). Performance differences between pair-type were also
0.54, p ⫽ .60, d ⫽ 0.12. greater during the person-last encoding order (Order ⫻ Pair-Type,
These results demonstrate that negative elements impair within- F [2, 52] ⫽ 22.98, p ⬍ .001, ␩P2 ⫽ 0.47) with better memory for
event dependency, and that this effect was particularly sensitive to location-object pairs compared to person-location, t(26) ⫽ 8.43,
the order in which elements were encoded. When the first two p ⬍ .001, d ⫽ 1.62 and object-person, t(26) ⫽ 6.03, p ⬍ .001, d ⫽
pairs within an event included a negative element (person-object; 1.16. For the person-first encoding order, object-person accuracy
location-person, person-first encoding order), dependency did not was greater than person-location accuracy, t(26) ⫽ 2.62, p ⬍ .05,
differ from the independent model. In contrast, when the negative d ⫽ 0.50; all other p’s ⬎ 0.11. Similar to Experiment 2, associative
element was presented during the final two within-event encoding accuracy was worse for pairs from a negative event compared to
trails (i.e., the first trial included two neutral items, location-object, neutral events (person-location, t[26] ⫽ 4.16, p ⬍ .001, d ⫽ 0.80;
person-last encoding order), dependency was greater than the Inde- object-person, t(26) ⫽ 4.30, p ⬍ .001, d ⫽ 0.83). However, there
pendent model, although still less than predicted by the Dependent was no difference between neutral and negative events in accuracy
model (similar to the pattern of dependency seen for negative events for location-object pairs (the pair involving two neutral elements in
in Experiment 1). Interestingly, we also saw a reduction in both neutral and negative events, t(26) ⫽ 0.88, p ⫽ .39, d ⫽ 0.17).
associative memory for neutral pairs from negative events when Memory coherence analysis. We again calculated depen-
the previously encoded pairs from the same event included a dency in the data (D) and compared it to the dependencies Di and
negative item (i.e., person-first encoding order, Figure 2b). Dd as predicted by Independent and Dependent models (see Meth-
ods and supplementary information for details) by performing a
2 ⫻ 2 ⫻ 3 repeated measures ANOVA (emotion; encoding-order;
Experiment 3 dependency-measure). Consistent with Experiment 2, we found a
three-way interaction (F [2, 52] ⫽ 5.56, p ⬍ .01, ␩P2 ⫽ 0.18) and
As we have previously reported, the presence of negative stimuli therefore performed separate 2 ⫻ 3 ANOVAs on neutral and
during encoding can disrupt associative memory, possibly via negative events.
hippocampal down-modulation, while memory for the negative Analysis of neutral events (see Figure 3b) showed a significant
content can be facilitated, possibly via amygdalar up-modulation main effect of dependency-measure (F [2, 52] ⫽ 41.44, p ⬍ .001,
(Bisby et al., 2016). We attempted to gain further insight into the ␩P2 ⫽ 0.61) and an effect of encoding order that approached
results of Experiment 2 by adding a 24-hour delay between study significance (F [1, 26] ⫽ 4.15, p ⫽ .052, ␩P2 ⫽ 0.14; Order ⫻
and test to see whether postencoding processes such as consolida- Dependency-Measure interaction p ⬎ .7). Further analysis showed
tion might affect associative binding and memory coherence. Emo- greater dependency when compared to the Independent model
tion is thought to modulate memory consolidation (McGaugh, (D ⬎ Di, t[26] ⫽ 4.56, p ⬍ .001, d ⫽ 0.88) but also less
2000; Roozendaal & McGaugh, 2011), so we wanted to ascertain dependency than the Dependent model (D ⬍ Dd, t[26] ⫽ 4.93, p ⬍
whether any effect of delay on associative memory or coherence .001, d ⫽ 0.95; collapsed across encoding order). Importantly,
was modulated by the presence of a negative element. there was no difference in dependency between encoding orders
250 BISBY, HORNER, BUSH, AND BURGESS

Figure 3. Memory for events following a 24hr delay. a, Associative memory performance across each pair type
at encoding across neutral and negative events split by the person-last encoding order (location-object pair
studied first) and person-first encoding order (object-person pair studied first). b, Dependency results for neutral
and negative events following a 24-hour delay between encoding and test, split by the person-last and
person-first encoding orders. Error bars represent standard error; NS ⫽ not significant. ⴱⴱ p ⬍ .01. ⴱⴱⴱ p ⬍ .001.

for neutral events relative to the independent model (i.e., in D-Di, a 24-hour delay) suggesting that a 24 hour delay might reduce
t[26] ⫽ 0.52, p ⫽ .61, d ⫽ 0.10). dependency. Importantly, a 24-hour delay did not interact with
Analysis of negative events (see Figure 3b) showed a significant emotion on dependency (p ⬎ .12).
Encoding-Order ⫻ Dependency-Measure interaction (F [2, 52] ⫽
7.45, p ⫽ .001, ␩P2 ⫽ 0.22) and a main effect of dependency-
A Computational Model
measure (F [2, 52] ⫽ 39.98, p ⬍ .001, ␩P2 ⫽ 0.61; main effect of
encoding order p ⬎ .5). For events encoded using the person-last To explore potential neural mechanisms that might underlie the
order (pure-neutral pair first), we saw significantly greater depen- pattern of results observed in Experiment 2, we constructed a
dency compared to the Independent model (D ⬎ Di, t[26] ⫽ 4.26, simple computational model of associative memory (Marr, 1971).
p ⬍ .001, d ⫽ 0.82) and less dependency than the Dependent We have previously shown that dependency for neutral events can
model (D ⬍ Dd, t[26] ⫽ 2.96, p ⬍ .01). In contrast, negative emerge through a process of pattern completion within a model of
events encoded under the person-first order showed less depen- hippocampal function (Horner et al., 2015). To examine the mech-
dency than the Dependent model (D ⬍ Dd, t[26] ⫽ 6.20, p ⬍ .001, anisms that might support the pattern of results found here, we
d ⫽ 1.19) and no difference when compared to the independent implemented a similar model.
model, t(26) ⫽ 0.31, p ⫽ .76, d ⫽ 0.06. A direct comparison of As in Experiment 2, events were formed by encoding overlap-
dependency relative to the independent model (D-Di) between ping pairwise associations between separate neurons coding for
encoding orders showed significantly greater dependency during individual elements within a fully recurrent attractor network.
the person-last encoding order, t(26) ⫽ 2.85, p ⬍ .01, d ⫽ 0.55. Encoding order and the associative structure of events were iden-
Similar to Experiment 2, there was more dependency relative to tical to Experiment 2. To account for overall behavioral perfor-
the independent model for neutral events studies under the person- mance, we assumed that the successful encoding of any given asso-
first encoding order when compared to negative events (D-Di; ciation was probabilistic. In addition, to model down-modulation of
t[26] ⫽ 3.07, p ⬍ .01, d ⫽ 0.59; no difference between neutral and hippocampal synaptic plasticity by negative emotion, we assumed
negative events under the person-last encoding order, t(26) ⫽ 0.20, that the strength of successfully encoded pairwise associations was
p ⫽ .85, d ⫽ 0.04). lower when a negative element was either presented or incidentally
We performed a further analysis to examine whether the 24- retrieved during encoding. Importantly, this does not prevent success-
hour delay further contributed to effects of emotion on dependency fully encoded negative associations from being recalled, but does
(i.e., a direct comparison between Experiments 1 and 2). We reduce pattern completion (and thus both performance and coherence)
therefore assessed the decrease in dependency relative to the in negative events during retrieval. At retrieval, a single “cue”
Dependent model for neutral and negative events across Experi- neuron was activated along with six “target” neurons that received
ments 2 and 3 (D-Dd). A 2 ⫻ 2 ⫻ 2 mixed ANOVA (Experi- partial activation to model the six-alternative forced choice task.
ment ⫻ Emotion ⫻ Order) with experiment added as a between Additional activity reflects inputs from recurrent synaptic currents,
participants factor showed a trend of a main effect of experiment with retrieval of any element (excluding the cue) determined by a
(F [1, 51] ⫽ 3.58, p ⫽ .06, ␩P2 ⫽ 0.07; all other p’s ⬎ 0.43) with firing rate threshold. Overall performance was submitted to the
a greater decrease in dependency in Experiment 3 (i.e., following statistical dependency analysis in the same way as behavioral data.
 EMOTION AND MEMORY COHERENCE 251

Method sources for a period of tret ⫽ 1,000 ms, while neurons that
represent the six forced choice target elements receive a constant
We simulate a simple network of N rate-coded neurons that are current of Iext ⫽ 5nA. Additional activity is generated by the
fully recurrently connected except for self-connections. The total recurrent synaptic currents, and the learning rate is set to zero (k ⫽
input current to each neuron Itotal is a combination of external and 0) to prevent further encoding. Behavioral performance is com-
recurrent synaptic currents Iext and Isyn, respectively. Recurrent puted by setting a firing rate threshold of ␪ret ⫽ 5 Hz for ‘retrieval’
synaptic currents Isyn are equal to the product of synaptic weights of any element (excluding the cue). Statistical dependency is then
wij and firing rates of connected neurons (Equation 1). computed as described above. The retrieval order for each pairwise
Itotal ⫽ Iext ⫹ Isyn association for the neutral and negative event conditions was
identical to Experiment 2 with the negative element presented
Isyn ⫽ 兺Nj⫽1 wijrj (1) during the last encoding trial (negative-last, corresponding to
person-last in the behavioral experiments) or during the first en-
The firing rate ri of each neuron is dictated by a threshold-linear
coding trial (negative-first, corresponding to person-first). Twenty-
activation function that converts the total input current Itotal into a
six simulations were performed, each containing 36 neutral and 36
firing rate output, with a threshold of ␪T ⫽ 5nA and limited to a
negative events.
peak firing rate of rmax ⫽ 10Hz (see Equation 2, where [x]⫹ ⫽ 0
for x ⱕ 0 and [x]⫹ ⫽ x for x ⬎ 0). All firing rates ri and synaptic
connections wij within the network are initially set to zero. Results

ri ⫽ 再[Ir
total ⫺ ␪T]⫹
max
for Itotal ⱕ 15nA
for Itotal ⬎ 15nA
(2)
In accordance with behavioral data, associative accuracy (Figure
4a) and dependency (Figure 4b) for negative events were both
reduced in comparison to neutral events (irrespective of encoding
Each element of an event is represented by a unique neuron.
order). Recurrent excitation passing through “strong” synaptic
During encoding, neurons that represent the stimuli being pre-
connections (i.e., those formed for neutral pairs) from the cue
sented in each trial receive a fixed amplitude synaptic current of
element via the nontarget element is sufficient to “retrieve” the
Iext ⫽ 15nA from an external source for a period of tenc ⫽ 1000
target element, even when the direct association between cue and
ms. During this period, synaptic weights develop according to a
target elements has not been formed. This increases both overall
Hebbian learning rule, that is, proportional to the product of pre-
performance, as associations can be recalled even when they were
and post- synaptic firing rates and a learning rate k. In addition, we
not successfully encoded; and dependency, as the likelihood of
impose a postsynaptic firing rate threshold of ␪p ⫽ 7.5 Hz, anal-
retrieving all three associations when only two have been success-
ogous to the BCM learning rule, below which no synaptic weight
fully encoded is increased. Conversely, recurrent excitation pass-
change takes place (Equation 3; Bienenstock, Cooper, & Munro,
ing through “weak” synaptic connections (e.g., those formed for
1982); and a hard limit of wmax ⫽ 1 on all synaptic weights.
negative pairs) from the cue element via the nontarget element is
⌬wij ⫽ kri[r j ⫺ ␪p]⫹ (3) not sufficient to ‘retrieve’ the target element in the absence of a
direct association between cue and target elements.
To model variation in extracellular and intracellular conditions
Similar to Experiment 2, negative events where the neutral pair
during encoding, we assume that learning is probabilistic, such that
was presented first (the order in which the negative element or
there is a probability penc ⫽ 0.65 that the learning rate will take a
person was presented last) showed increased associative accuracy
positive value k ⫽ klearn and a probability 1 ⫺ penc that it will take
(Figure 4a) and dependency (Figure 4b) compared to negative
a value of zero on any given encoding trial. The positive learning
events where the neutral pair was presented last (the order in which
rate klearn is a product of the learning rate for associations from
the negative element or person was presented first). These differ-
neutral events klearn ⫽ 1.6 ⫻ 10⫺4 and a modulation factor m. This
ences can be accounted for by the model, due to incidental reac-
modulation factor generally takes a value of m ⫽ 1, but is reduced
tivation of negative elements at encoding. When the neutral pair
to a value of m ⫽ 0.6 when any of the neurons that encode a
from a negative event is presented first, a strong association
negative element fire above a threshold rate ␪neg ⫽ 1Hz (Equation
between the neutral elements can be formed (see Figure 4c) and
4, where H represents the Heaviside function). Importantly, activ-
drive subsequent pattern completion. When the neutral pair from a
ity in neurons that encode a negative element can be generated
negative event is presented last, however, there is a strong possi-
either by external input (i.e., when an association including a
bility that one or both of the neutral elements will have been
negative element is being encoded), or by recurrent excitation (i.e.,
associated with a negative element in previous learning trials. This
when a negative element is incidentally retrieved by association
leads to the reactivation of that negative element, and a subsequent
with one of the elements being encoded). The encoding order and
reduction in the strength of the association formed between neutral
resulting associative structures for the neutral and negative events
elements that impairs pattern completion.
are identical to Experiment 2.
Hence, the model is able to reproduce each key feature of the
p(k ⫽ klearn) ⫽ penc experimental data provided that a single criterion is satisfied: the
p(k ⫽ 0) ⫽ (1 ⫺ penc) strength of learned connections between pairs of items that either
klearn ⫽ mkneut (4) include or cause the incidental retrieval of a negative element
m ⫽ 1 ⫺ 0.4H共 兺 H(ri,neg ⫺ ␪neg)兲 during encoding should be weaker than those between pairs of
neutral items, such that they are sufficient to allow retrieval of that
During retrieval, neurons that represent the cued element receive association but not support pattern completion (i.e., retrieval of the
a fixed amplitude synaptic current Iext ⫽ 15nA from external target element via reactivation of the third, nontarget element). The
252 BISBY, HORNER, BUSH, AND BURGESS

Figure 4. Model of associative learning and simulated results. (a) Performance and (b) Coherence (statistical
dependency between retrievals from the same event) for each event type and encoding order (whether the
negative element or person was presented last or first). (c) During encoding, negative events are constructed from
three pairwise associations, one of which contains two neutral elements. Associative learning is reduced when
the negative element (always a person) is active, so associations involving the negative element are weaker than
those involving neutral events, reducing pattern completion, performance and dependency. If the neutral pair is
presented after one or more negative-neutral pairs (the Negative-First Order), reactivation of the negative
element can occur via the associations already learned to it, weakening the association formed between the
neutral elements, further reducing pattern completion, performance and dependency.

specific implementation and parameter values used here were Here we provide evidence that, while neutral events are bound
chosen to ensure that this was the case, but any associative mem- together as coherent representations in memory, negative emotion
ory model that incorporated this property would produce similar can disrupt associative binding between event elements and
results. This supports the hypothesis that each of the key experi- weaken the coherence of those events. Across three experiments,
mental results can be accounted for by a single mechanism—a the presence of a negative element within an event reduced both
reduction in the strength of associations formed between pairs of associative memory performance and the amount of statistical
items that include a negative element, or cause a negative element dependency between within-event elements. As demonstrated by
to be incidentally retrieved. our computational model, reduced coherence of negative events
can be accounted for by weakened associative binding in the
General Discussion presence of negative elements, or the reactivation of their memory
A defining property of episodic memory is that events consist- traces, which impairs pattern completion and thus the coherence of
ing of multiple elements are stored as coherent representations, so the event as experienced in holistic memory retrieval.
that episodic retrieval corresponds to holistic reexperience of all In accordance with previous findings (Horner et al., 2015;
types of element (whether each is correctly remembered or not; Horner & Burgess, 2013, 2014), retrieval of neutral events showed
Horner et al., 2015; Tulving, 1983). It has been argued that coherence (i.e., retrieval success for associations from the same
negative emotion will strengthen memory, resulting in a ‘general event were statistically related), supporting the idea that episodic
facilitation’ effect (Cahill & McGaugh, 1998; Rubin et al., 2008; memories are unitary and retrieved holistically (Tulving, 1983).
Talarico et al., 2004). However, others have proposed that differ- This coherence was present irrespective of whether events were
ent aspects of memory will be affected in opposing ways (Brewin encoded simultaneously or as overlapping pairs. It should be noted
et al., 2010; Jacobs & Nadel, 1998) so that memory for the that some of the coherence seen when events are encoded simul-
negative elements of an event may be strengthened but memory for taneously could be driven by variations in encoding strength
associations between elements or between elements and their between events (e.g., reflecting attention), but this is unlikely to be
context may be weakened (Bisby et al., 2016; Madan et al., 2012). the case for sequentially presented events.
 EMOTION AND MEMORY COHERENCE 253

Negative events showed reduced associative accuracy and less in our model. In contrast, a negative element on the first encoding
coherence compared to neutral events. These reductions were trial (person-first encoding order) would mean no strong associa-
observed when events were presented simultaneously or as a tion would be present on subsequent encoding or retrieval trials.
sequence of overlapping pairs. Successful encoding of all within- When a negative element was presented on the initial encoding
event pairs supports a coherent associative structure and enables trials of the event (as in the person-first encoding order), formation
pattern completion of all event elements at retrieval irrespective of of these negative associations can disrupt encoding of subsequent
which is the cue or the target, increasing coherence. We assume overlapping pairs. Subsequent presentation of neutral elements
that the presence of a negative element at encoding could reduce associated to the negative element may result in activation of the
the formation of associations with other elements presented in negative element which would disrupt the formation of new asso-
conjunction (Bisby & Burgess, 2014; Kensinger & Schacter, 2006; ciations, even in the absence of the negative element. This would
Madan et al., 2012; Mather & Knight, 2008; Rimmele et al., 2011; explain the reduced associative accuracy for neutral pairs from
Touryan et al., 2007), decreasing associative memory performance negative events during the person-first encoding order. The for-
and coherence. Our computational model verified this assumption mation of associations/relations between the emotional properties
as a potentially valid explanation of our data. Given the important and items could be supported by the amygdala, consistent with an
role of the hippocampus in associative/relational binding (Davachi, emotional binding hypothesis (Bisby et al., 2016; Yonelinas &
2006; Eichenbaum et al., 2007; O’Keefe & Nadel, 1978) and Ritchey, 2015).
within-event dependency (Horner et al., 2015), our results are con- This is the first study to examine episodic memory coherence
sistent with reports that negative emotion might down-modulate hip- following a prolonged delay. Interestingly, coherence for neutral
pocampal processing to impair associative memory formation (Bisby events was slightly (though not significantly) reduced after 24
et al., 2016). hours, suggesting that some of the important associations required
Dependency for negative events showed a greater reduction to form, and maintain, coherent representations might be weak-
when the initial studied pair included a negative element, high- ened or lost over time. The pattern of associative memory and
lighting the significance of the order in which events were studied. coherence results for negative events were replicated across delays
A decrease in our measure of dependency reflects a weaker rela- with a similar pattern observed whether memory was tested im-
tionship between retrieval success for different associations from mediately or following a 24 hour delay. It is well established that
the same event (i.e., the success of retrieving one element from an negative emotion can influence memory during both encoding
event did not predict retrieval of other elements from the same (Dolcos, LaBar, & Cabeza, 2004; Kensinger & Corkin, 2004) and
event). Thus, reduced dependency when the initial encoded pair consolidation (LaBar & Cabeza, 2006; McGaugh, 2004). How-
included a negative element highlights a potential lack of pattern ever, many of these reports have demonstrated enhanced memory
completion to support holistic retrieval of all event elements. for individual emotional items or the subjective feelings attached
Negative events were therefore encoded as less coherent represen- to them (Cahill & McGaugh, 1995; Sharot et al., 2004). In contrast,
tations than neutral events, and this reduction was accentuated by our findings support the view that the disruptive nature of negative
encoding order. emotion on associative binding can occur during encoding (or
Interestingly, it has been shown that participants are worse at retrieval), but perhaps does not affect consolidation.
learning novel associations between a negative item and its screen While we provide clear evidence that associative binding was
location when the negative item has previously been encoded impaired by negative emotion, we did not expect that memory for
(Nashiro, Sakaki, Huffman, & Mather, 2013). Further, the amygdala the individual elements themselves should be reduced, indeed
is thought to prevent memory updating when negative items are later memory for individual items is often enhanced by negative emo-
reencountered and that novel learning requires its inhibition via tional content (Bisby & Burgess, 2014; Cahill & McGaugh, 1995;
orbitofrontal areas (Sakaki, Niki, & Mather, 2011). The combina- Sharot et al., 2004; Sharot & Yonelinas, 2008; Yonelinas &
tion of these findings and our own demonstrate how encoded Ritchey, 2015). Although we only directly assessed item recogni-
negative items can further disrupt encoding on subsequent learning tion in Experiment 1, our results show comparable memory per-
trials whether the negative item is later presented or when items formance across neutral and negative conditions. This is consistent
that have previously been associated with a negative item are with the view that the negative emotion specifically impairs asso-
presented. ciations, possibly through disrupting hippocampal function, while
For neutral events, pattern completion could facilitate memory memory for individual items could be supported by modality-
performance and coherence even when two elements are only specific neocortical regions (Aggleton & Brown, 1999; Eichen-
weakly associated, retrieval being boosted by activity passing baum et al., 2007; Marr, 1971; McClelland et al., 1995) and their
though indirect connections via the third element. This would affective properties associated to them via the amygdala (LeDoux,
strengthen weak associations and support inference and integration 2000; Paz, Pelletier, Bauer, & Paré, 2006; Wang et al., 2014).
of overlapping information (Milivojevic et al., 2015; Zeithamova, The precise mechanisms that might support the up- and down-
Dominick, & Preston, 2012). These pattern completion processes modulation of distinct memory representations during a negative
can also account for differences in coherence across encoding event are unclear. However, an arousal-biased competition model
orders for negative events. Comparable accuracy for location- (Mather & Sutherland, 2011) proposes that observed memory
object pairs across neutral and negative events when presented on enhancements and impairments are generated via competition of
the first encoding trial (person-last encoding order) suggest a limited mental resources for encoded information. Within this
strong association was formed. This strong association could pro- account, arousal is thought to bias processing toward high priority
vide a basis for pattern completion during subsequent encoding representations at a cost of low priority representations, changes
trials or during test, aiding performance and coherence, as shown which could be supported by complex interactions between gluta-
254 BISBY, HORNER, BUSH, AND BURGESS

mate and norepinephrine (Mather, Clewett, Sakaki, & Harley, here, might be fragmented rather than simply strengthened relative
2016). While this highlights a potential mechanism that could to neutral memories. It is important to note that providing more
contribute to our pattern of results, this competitive mechanism information about the event may strengthen memory for the neg-
would require extension to explain the reduction in associative ative content by increasing the number of retrieval cues (Pearson,
memory for pairs of neutral items encoded after related negative 2012; Pearson, Ross, & Webster, 2012), as might ongoing reacti-
associations (Experiments 2 and 3), the absence of an inverse vation or rumination concerning the event (Berntsen, Staugaard, &
relationship between memory for negative and neutral items in Sørensen, 2013; Rubin et al., 2008), facilitating subsequent re-
Experiment 1, and the reduction in associative memory seen be- trieval via sensory cues (Kleim, Ehring, & Ehlers, 2012). How-
tween pairs of negative items in previous studies (Bisby et al., ever, a dual representation account adds to this understanding by
2016; Bisby & Burgess, 2014). Further studies will be required to proposing how altered associative processing at encoding can also
fully elucidate the complex mechanisms supporting impaired as- contribute to memory disruptions.
sociative memory for negative events. Our findings have potential therapeutic implications. They high-
The salience of emotional items is likely to attract greater light the importance of the formation of associations between the
processing due to their attentional capture and distinctiveness negative element of an event and the surrounding neutral elements
(Talmi, 2013), but we do not think that these attributes can fully or context. This suggests that an important process in recovery of
account for observed reductions in associative memory. Studies
healthy memory function could be the formation of strong asso-
demonstrate similar reductions in associative memory for emo-
ciations between the negative content and the neutral context of the
tional words, even when word stimuli are well-matched across
event. However, our findings also indicate that the continued
numerous dimensions (Madan et al., 2012) and this pattern of
presence of negative emotion might hinder the formation of new
behavior is mirrored when using emotional pictures (Bisby &
associations. Trauma-focused psychological therapies such as im-
Burgess, 2014). Item memory was unaffected by emotion suggest-
agery rescripting (Hackmann, 1998) and eye movement desensi-
ing that attentional processing between emotional categories did
tization and reprocessing (Shapiro, 2001; EMDR) aim at revisiting
not contribute to the pattern of associate memory and dependency.
Further evidence against an attentional explanation can be drawn the negative information while ameliorating its negative emotional
from Experiment 2, in which associative memory was reduced for impact. These techniques often require patients to elaborate on the
location-object pairs when each item had previously been paired negative imagery, thus associating them to appropriate neutral con-
with a negative item (which was no longer on the screen to distract textual information (such as when and where it happened). The
attention). In a series of experiments, we have also shown that attenuation of symptoms might be understood in terms of the mech-
when participants encode neutral and negative item-context pair- anisms of episodic memory formation studied here and the strength-
ings on background contexts that predict wither safety or threat of ening of weak connections to reestablish coherent episodic memories,
shock, the threatening contexts have little effect on item memory as predicted by a dual representation framework (Brewin et al., 2010).
but impair the association between the neutral item and its context In conclusion, we provide new evidence that negative events
(Bisby & Burgess, 2014). Further, when participants encode can disrupt associative binding and the coherence of single
negative-negative item pairs (both of which should capture atten- representations in memory. While neutral events were consis-
tion), associative memory is still disrupted compared to neutral- tently found to be bound together as single representations in
neutral item pairs (Bisby & Burgess, 2014; Bisby et al., 2016). memory, negative emotion disrupts associative binding between
Further experiments could attempt to dissociate salience from event elements, resulting in a weakened associative structure.
emotion within the current experimental design by replacing emo- The presence of a negative element consistently resulted in a
tional items with nonemotional salient items (e.g., “oddballs”; decrease in statistical dependency between within-event ele-
Strange, Hurlemann, & Dolan, 2003) to see whether they also ments. Overall, our data and computational model demonstrate
affect associative memory in a similar way. Taken together, while that reduced coherence in memory for negative events can be
attention and distinctiveness surely play an important role in accounted for by weakened associative binding in the presence
emotional memory alterations, our results suggest that they cannot of negative elements, disrupting pattern completion processes
fully account for the pattern of our results. that support holistic retrieval. These findings highlight the
Our findings have important clinical implications for the way in importance of associative/relational memory mechanisms in
which negative events might contribute to memory disturbances, contributing to memory disturbances in PTSD and their treat-
as seen in disorders such as posttraumatic stress disorder (Brewin, ment by therapeutic interventions.
2003; PTSD). The debate between “general facilitation” and “dual
representation” accounts of the effect of emotion on memory
extends to a debate concerning whether a traumatic event will lead Code Availability
to a generally strengthened memory or a fragmented memory
The code for the computation model used in this article is freely
comprising some very strong elements and some absent or weak
elements (cf. Brewin et al., 2010; Rubin et al., 2008). Here, we available on FigShare (http://dx.doi.org/10.6084/m9.figshare.5240
demonstrate that experiencing mildly negative events in healthy 752).
volunteers reduced the coherence of episodic memories. This
fragmented associative structure for negative events reduces the Data Availability
likelihood that a partial cue would trigger holistic retrieval via
pattern completion. Thus our findings support the view that trau- The data presented in this article are freely available on Fig-
matic memories, like the mildly negative emotional memories used Share (http://dx.doi.org/10.6084/m9.figshare.5240749).
 EMOTION AND MEMORY COHERENCE 255

References Horner, A. J., Bisby, J. A., Bush, D., Lin, W. J., & Burgess, N. (2015).
Evidence for holistic episodic recollection via hippocampal pattern
Aggleton, J. P., & Brown, M. W. (1999). Episodic memory, amnesia, and completion. Nature Communications, 6, 7462. http://dx.doi.org/10.1038/
the hippocampal-anterior thalamic axis. The Behavioral and Brain Sci- ncomms8462
ences, 22, 425– 444. http://dx.doi.org/10.1017/S0140525X99002034 Horner, A. J., & Burgess, N. (2013). The associative structure of memory
Berntsen, D., Staugaard, S. R., & Sørensen, L. M. T. (2013). Why am I for multi-element events. Journal of Experimental Psychology: General,
remembering this now? Predicting the occurrence of involuntary (spon- 142, 1370 –1383. http://dx.doi.org/10.1037/a0033626
taneous) episodic memories. Journal of Experimental Psychology: Gen- Horner, A. J., & Burgess, N. (2014). Pattern completion in multielement
eral, 142, 426 – 444. event engrams. Current Biology, 24, 988 –992. http://dx.doi.org/10
Bienenstock, E. L., Cooper, L. N., & Munro, P. W. (1982). Theory for the .1016/j.cub.2014.03.012
development of neuron selectivity: Orientation specificity and binocular Horner, A. J., & Henson, R. N. (2008). Priming response learning and
interaction in visual cortex. The Journal of Neuroscience, 2, 32– 48. repetition suppression. Neuropsychologia, 46, 1979 –1991.
Bisby, J. A., & Burgess, N. (2014). Negative affect impairs associative Jacobs, W. J., & Nadel, L. (1998). Neurobiology of reconstructed memory.
memory but not item memory. Learning & Memory, 21, 760 –766. Psychology, Public Policy, and Law, 4, 1110 –1134. http://dx.doi.org/10
http://dx.doi.org/10.1101/lm.032409.113 .1037/1076-8971.4.4.1110
Bisby, J. A., Horner, A. J., Hørlyck, L. D., & Burgess, N. (2016). Opposing Kensinger, E. A., & Corkin, S. (2003). Effect of negative emotional content
effects of negative emotion on amygdalar and hippocampal memory for on working memory and long-term memory. Emotion, 3, 378 –393.
items and associations. Social Cognitive and Affective Neuroscience, 11, http://dx.doi.org/10.1037/1528-3542.3.4.378
981–990. http://dx.doi.org/10.1093/scan/nsw028 Kensinger, E. A., & Corkin, S. (2004). Two routes to emotional memory:
Bradley, M. M., Greenwald, M. K., Petry, M. C., & Lang, P. J. (1992). Distinct neural processes for valence and arousal. PNAS: Proceedings of
Remembering pictures: Pleasure and arousal in memory. Journal of the National Academy of Sciences of the United States of America, 101,
Experimental Psychology: Learning, Memory, and Cognition, 18, 379 – 3310 –3315. http://dx.doi.org/10.1073/pnas.0306408101
390. http://dx.doi.org/10.1037/0278-7393.18.2.379 Kensinger, E. A., & Schacter, D. L. (2006). Amygdala activity is associ-
Brewin, C. (2003). Posttraumatic stress disorder: Malady or myth? New ated with the successful encoding of item, but not source, information
Haven, CT: Yale University Press. for positive and negative stimuli. The Journal of Neuroscience, 26,
Brewin, C. R., Gregory, J. D., Lipton, M., & Burgess, N. (2010). Intrusive 2564 –2570. http://dx.doi.org/10.1523/JNEUROSCI.5241-05.2006
images in psychological disorders: Characteristics, neural mechanisms, Kleim, B., Ehring, T., & Ehlers, A. (2012). Perceptual processing advan-
and treatment implications. Psychological Review, 117, 210 –232. http:// tages for trauma-related visual cues in post-traumatic stress disorder.
dx.doi.org/10.1037/a0018113 Psychological Medicine, 42, 173–181. http://dx.doi.org/10.1017/
Brown, R., & Kulik, J. (1977). Flashbulb memories. Cognition, 5, 73–99. S0033291711001048
http://dx.doi.org/10.1016/0010-0277(77)90018-X LaBar, K. S., & Cabeza, R. (2006). Cognitive neuroscience of emotional
Cahill, L., Haier, R. J., Fallon, J., Alkire, M. T., Tang, C., Keator, D., . . . memory. Nature Reviews Neuroscience, 7, 54 – 64. http://dx.doi.org/10
McGaugh, J. L. (1996). Amygdala activity at encoding correlated with .1038/nrn1825
long-term, free recall of emotional information. Proceedings of the Lang, P., Bradley, M. M., & Cuthbert, B. N. (2005). International affective
National Academy of Sciences of the United States of America, 93, picture system (IAPS): Affective ratings of pictures and instruction
8016 – 8021. http://dx.doi.org/10.1073/pnas.93.15.8016 manual. Tech. Rep. No. A-6. University of Florida, Gainsville, FL.
Cahill, L., & McGaugh, J. L. (1995). A novel demonstration of enhanced LeDoux, J. E. (2000). Emotion circuits in the brain. Annual Review of
memory associated with emotional arousal. Consciousness and Cogni- Neuroscience, 23, 155–184. http://dx.doi.org/10.1146/annurev.neuro.23
tion, 4, 410 – 421. http://dx.doi.org/10.1006/ccog.1995.1048 .1.155
Cahill, L., & McGaugh, J. L. (1998). Mechanisms of emotional arousal and Madan, C. R., Caplan, J. B., Lau, C. S. M., & Fujiwara, E. (2012).
lasting declarative memory. Trends in Neurosciences, 21, 294 –299. Emotional arousal does not enhance association-memory. Journal of
http://dx.doi.org/10.1016/S0166-2236(97)01214-9 Memory and Language, 66, 695–716. http://dx.doi.org/10.1016/j.jml
Christianson, S. A. (1992). Emotional stress and eyewitness memory: A .2012.04.001
critical review. Psychological Bulletin, 112, 284 –309. http://dx.doi.org/ Marchewka, A., Zurawski, L., Jednoróg, K., & Grabowska, A. (2014). The
10.1037/0033-2909.112.2.284 Nencki Affective Picture System (NAPS): Introduction to a novel,
Cohen, N. J., & Eichenbaum, H. (1993). Memory, amnesia and the hip- standardized, wide-range, high-quality, realistic picture database. Behav-
pocampal system. Cambridge, MA: MIT Press. ior Research Methods, 46, 596 – 610. http://dx.doi.org/10.3758/s13428-
Damasio, A. R. (1989). The brain binds entities and events by multire- 013-0379-1
gional activation from convergence zones. Neural Computation, 1, 123– Marr, D. (1971). Simple memory: A theory for archicortex. Philosophical
132. http://dx.doi.org/10.1162/neco.1989.1.1.123 Transactions of the Royal Society of London Series B, Biological Sci-
Davachi, L. (2006). Item, context and relational episodic encoding in ences, 262, 23– 81. http://dx.doi.org/10.1098/rstb.1971.0078
humans. Current Opinion in Neurobiology, 16, 693–700. http://dx.doi Mather, M. (2007). Emotional Arousal and Memory Binding: An Object-
.org/10.1016/j.conb.2006.10.012 Based Framework. Perspectives on Psychological Science, 2, 33–52.
Dolcos, F., LaBar, K. S., & Cabeza, R. (2004). Interaction between the http://dx.doi.org/10.1111/j.1745-6916.2007.00028.x
amygdala and the medial temporal lobe memory system predicts better Mather, M., Clewett, D., Sakaki, M., & Harley, C. W. (2016). Norepi-
memory for emotional events. Neuron, 42, 855– 863. http://dx.doi.org/ nephrine ignites local hotspots of neuronal excitation: How arousal
10.1016/S0896-6273(04)00289-2 amplifies selectivity in perception and memory. Behavioral and Brain
Eichenbaum, H., Yonelinas, A. P., & Ranganath, C. (2007). The medial Sciences, 39, e200. http://dx.doi.org/10.1017/S0140525X15000667
temporal lobe and recognition memory. Annual Review of Neuroscience, Mather, M., & Knight, M. (2008). The emotional harbinger effect: Poor
30, 123–152. http://dx.doi.org/10.1146/annurev.neuro.30.051606 context memory for cues that previously predicted something arousing.
.094328 Emotion, 8, 850 – 860. http://dx.doi.org/10.1037/a0014087
Hackmann, A. (1998). Working with images in clinical psychology. In Mather, M., & Sutherland, M. R. (2011). Arousal-Biased Competition in
Comprehensive clinical psychology (Vol. 6, pp. 301–318). New York, Perception and Memory. Perspectives on Psychological Science, 6,
NY: Elsevier. http://dx.doi.org/10.1016/B0080-4270(73)00196-6 114 –133. http://dx.doi.org/10.1177/1745691611400234
256 BISBY, HORNER, BUSH, AND BURGESS

McClelland, J. L., McNaughton, B. L., & O’Reilly, R. C. (1995). Why Sakaki, M., Niki, K., & Mather, M. (2011). Updating existing emotional
there are complementary learning systems in the hippocampus and memories involves the frontopolar/orbito-frontal cortex in ways that
neocortex: Insights from the successes and failures of connectionist acquiring new emotional memories does not. Journal of Cognitive
models of learning and memory. Psychological Review, 102, 419 – 457. Neuroscience, 23, 3498 –3514. http://dx.doi.org/10.1162/jocn_a_00057
http://dx.doi.org/10.1037/0033-295X.102.3.419 Schlichting, M. L., & Preston, A. R. (2015). Memory integration: Neural
McGaugh, J. L. (2000). Memory—A century of consolidation. Science, mechanisms and implications for behavior. Current Opinion in Behav-
287, 248 –251. http://dx.doi.org/10.1126/science.287.5451.248 ioral Sciences, 1, 1– 8. http://dx.doi.org/10.1016/j.cobeha.2014.07.005
McGaugh, J. L. (2003). Memory and emotion: The making of lasting Shapiro, F. (2001). Eye movement desensitization and reprocessing. Basic
memories. New York, NY: Columbia University Press. principles, protocols and procedures. New York, NY: The Guilford
McGaugh, J. L. (2004). The amygdala modulates the consolidation of Press.
memories of emotionally arousing experiences. Annual Review of Neu- Sharot, T., Delgado, M. R., & Phelps, E. A. (2004). How emotion enhances
roscience, 27, 1–28. http://dx.doi.org/10.1146/annurev.neuro.27.070203 the feeling of remembering. Nature Neuroscience, 7, 1376 –1380. http://
.144157 dx.doi.org/10.1038/nn1353
Milivojevic, B., Vicente-Grabovetsky, A., & Doeller, C. F. (2015). Insight Sharot, T., & Yonelinas, A. P. (2008). Differential time-dependent effects
reconfigures hippocampal-prefrontal memories. Current Biology, 25, of emotion on recollective experience and memory for contextual infor-
821– 830. http://dx.doi.org/10.1016/j.cub.2015.01.033 mation. Cognition, 106, 538 –547. http://dx.doi.org/10.1016/j.cognition
Nashiro, K., Sakaki, M., Huffman, D., & Mather, M. (2013). Both younger .2007.03.002
and older adults have difficulty updating emotional memories. The Strange, B. A., Hurlemann, R., & Dolan, R. J. (2003). An emotion-induced
Journals of Gerontology: Series B: Psychological Sciences and Social retrograde amnesia in humans is amygdala- and beta-adrenergic-
Sciences, 68, 224 –227. http://dx.doi.org/10.1093/geronb/gbs039 dependent. Proceedings of the National Academy of Sciences, USA of
Norman, K. A., & O’Reilly, R. C. (2003). Modeling hippocampal and the United States of America, 100, 13626 –13631. http://dx.doi.org/10
neocortical contributions to recognition memory: A complementary- .1073/pnas.1635116100
learning-systems approach. Psychological Review, 110, 611– 646. http:// Talarico, J. M., LaBar, K. S., & Rubin, D. C. (2004). Emotional intensity
dx.doi.org/10.1037/0033-295X.110.4.611 predicts autobiographical memory experience. Memory & Cognition, 32,
O’Keefe, J., & Nadel, L. (1978). The hippocampus as a cognitive map. 1118 –1132. http://dx.doi.org/10.3758/BF03196886
Oxford, England: Oxford University Press. Talmi, D. (2013). Enhanced emotional memory: Cognitive and neural
Paz, R., Pelletier, J. G., Bauer, E. P., & Paré, D. (2006). Emotional mechanisms. Current Directions in Psychological Science, 22, 430 –
enhancement of memory via amygdala-driven facilitation of rhinal in- 436. http://dx.doi.org/10.1177/0963721413498893
teractions. Nature Neuroscience, 9, 1321–1329. http://dx.doi.org/10 Touryan, S. R., Marian, D. E., & Shimamura, A. P. (2007). Effect of
.1038/nn1771 negative emotional pictures on associative memory for peripheral infor-
Pearson, D. G. (2012). Contextual representations increase analogue trau- mation. Memory, 15, 154 –166. http://dx.doi.org/10.1080/096582106
matic intrusions: Evidence against a dual-representation account of 01151310
peri-traumatic processing. Journal of Behavior Therapy and Experimen- Tulving, E. (1983). Elements of Episodic Memory. Oxford, England: Clar-
tal Psychiatry, 43, 1026 –1031. http://dx.doi.org/10.1016/j.jbtep.2012.04 endon.
.002 Wang, S., Tudusciuc, O., Mamelak, A. N., Ross, I. B., Adolphs, R., &
Pearson, D. G., Ross, F. D. C., & Webster, V. L. (2012). The importance Rutishauser, U. (2014). Neurons in the human amygdala selective for
of context: Evidence that contextual representations increase intrusive perceived emotion. Proceedings of the National Academy of Sciences of
memories. Journal of Behavior Therapy and Experimental Psychiatry, the United States of America, 111, E3110 –E3119. http://dx.doi.org/10
43, 573–580. http://dx.doi.org/10.1016/j.jbtep.2011.07.009 .1073/pnas.1323342111
Rimmele, U., Davachi, L., Petrov, R., Dougal, S., & Phelps, E. A. (2011). Yonelinas, A. P., & Ritchey, M. (2015). The slow forgetting of emotional
Emotion enhances the subjective feeling of remembering, despite lower episodic memories: An emotional binding account. Trends in Cognitive
accuracy for contextual details. Emotion, 11, 553–562. http://dx.doi.org/ Sciences, 19, 259 –267. http://dx.doi.org/10.1016/j.tics.2015.02.009
10.1037/a0024246 Zeithamova, D., Dominick, A. L., & Preston, A. R. (2012). Hippocampal
Roozendaal, B., & McGaugh, J. L. (2011). Memory modulation. Behav- and ventral medial prefrontal activation during retrieval-mediated learn-
ioral Neuroscience, 125, 797– 824. http://dx.doi.org/10.1037/a0026187 ing supports novel inference. Neuron, 75, 168 –179. http://dx.doi.org/10
Rubin, D. C., Boals, A., & Berntsen, D. (2008). Memory in posttraumatic .1016/j.neuron.2012.05.010
stress disorder: Properties of voluntary and involuntary, traumatic and
nontraumatic autobiographical memories in people with and without Received February 13, 2017
posttraumatic stress disorder symptoms. Journal of Experimental Psy- Revision received June 28, 2017
chology: General, 137, 591– 614. http://dx.doi.org/10.1037/a0013165 Accepted June 28, 2017 䡲