Scholarly article on topic 'Changing Zaire to Congo: The fate of no-longer relevant mnemonic information'

Changing Zaire to Congo: The fate of no-longer relevant mnemonic information Academic research paper on "Psychology"

CC BY-NC-ND
0
0
Share paper
Academic journal
NeuroImage
OECD Field of science
Keywords
{}

Abstract of research paper on Psychology, author of scientific article — Johan Eriksson, Mikael Stiernstedt, Maria Öhlund, Lars Nyberg

Abstract In an ever-changing world there is constant pressure on revising long-term memory, such when people or countries change name. What happens to the old, pre-existing information? One possibility is that old associations gradually are weakened and eventually lost. Alternatively, old and no longer relevant information may still be an integral part of memory traces. To test the hypothesis that old mnemonic information still becomes activated when people correctly retrieve new, currently relevant information, brain activity was measured with fMRI while participants performed a cued-retrieval task. Paired associates (symbol–sound and symbol–face pairs) were first learned during two days. Half of the associations were then updated during the next two days, followed by fMRI scanning on day 5 and also 18months later. As expected, retrieval reactivated sensory cortex related to the most recently learned association (visual cortex for symbol–face pairs, auditory cortex for symbol–sound pairs). Critically, retrieval also reactivated sensory cortex related to the no-longer relevant associate. Eighteen months later, only non-updated symbol–face associations were intact. Intriguingly, a subset of the updated associations was now treated as though the original association had taken over, in that memory performance was significantly worse than chance and that activity in sensory cortex for the original but not the updated associate correlated (negatively) with performance. Moreover, the degree of “residual” reactivation during day 5 inversely predicted memory performance 18months later. Thus, updating of long-term memory involves adding new information to already existing networks, in which old information can stay resilient for a long time.

Academic research paper on topic "Changing Zaire to Congo: The fate of no-longer relevant mnemonic information"

Changing Zaire to Congo: The fate of no-longer relevant mnemonic information

Johan Eriksson a,b'*, Mikael Stiernstedta,b, Maria Ohlunda, Lars Nyberga,b,c

a Umea Center for Functional Brain Imaging (UFBI), Umea University, Sweden b Department of Integrative Medical Biology (Physiology), Umea University, Sweden c Department of Radiation Sciences (Radiology), Umea University, Sweden

ARTICLE INFO ABSTRACT

Article history: In an ever-changing world there is constant pressure on revising long-term memory, such when people or coun-

Accepted 21 Jrne 2014 tries change name. What happens to the old, pre-existing information? One possibility is that old associations

Available online 28 June 2014 gradually are weakened and eventually lost. Alternatively, old and no longer relevant information may still be

an integral part of memory traces. To test the hypothesis that old mnemonic information still becomes activated when people correctly retrieve new, currently relevant information, brain activity was measured with fMRI while participants performed a cued-retrieval task. Paired associates (symbol-sound and symbol-face pairs) were first learned during two days. Half of the associations were then updated during the next two days, followed by fMRI scanning on day 5 and also 18 months later. As expected, retrieval reactivated sensory cortex related to the most recently learned association (visual cortex for symbol-face pairs, auditory cortex for symbol-sound pairs). Critically, retrieval also reactivated sensory cortex related to the no-longer relevant associate. Eighteen months later, only non-updated symbol-face associations were intact. Intriguingly, a subset of the updated associations was now treated as though the original association had taken over, in that memory performance was significantly worse than chance and that activity in sensory cortex for the original but not the updated associate correlated (negatively) with performance. Moreover, the degree of "residual" reactivation during day 5 inversely predicted memory performance 18 months later. Thus, updating of long-term memory involves adding new information to already existing networks, in which old information can stay resilient for a long time.

© 2014 The Authors. Published by Elsevier Inc. This is an open access article under the CC BY-NC-ND license

(http://creativecommons.org/licenses/by-nc-nd/3.0/).

|Wj| CrossMark

Introduction

The ability to store information about facts and events is fundamental for every-day functioning. However, many times mnemonic information needs to be revised; your cousin may get re-married and change name, the ruler of Zaire was overthrown and the new government declared the Democratic Republic of the Congo, your favorite artist was "formerly known as Prince", etc. What happens to old, pre-existing information as memories are updated? One possibility, suggested by studies of eyewitness testimony (Loftus, 2005), reconsolidation (Dudai, 2006), and active suppression of memories (Anderson et al., 2004), is that old associations and memory traces gradually are weakened and eventually lost. Alternatively, old and no longer relevant information may still be an integral part of memory traces. Support for the latter view comes from computational (McClelland et al., 1995) and molecular (Tronel et al., 2005) studies. In fact, it might be argued that the pre-existing information

* Corresponding author at: Department of Integrative Medical Biology (Physiology Section), Umea University, S-901 87 Umea, Sweden.

E-mail address: johan.eriksson@physiol.umu.se (J. Eriksson).

serves as a "schema" and facilitates learning and assimilation of the new information (Tse et al., 2007).

Here we tested the hypothesis that old, no longer relevant mnemonic information still becomes activated when people correctly retrieve newer, currently relevant information. During the first two days of the experiment, participants learned paired associates consisting of either symbol-face (visual-visual) or symbol-sound (visual-auditory) pairs. During days 3-4, half of these associations were updated, such that symbol-face pairs became symbol-sound pairs and vice versa; the other half was unchanged. On day 5, in the fMRI scanner, a memory reactivation paradigm was used. A symbol was presented and the participants were required to indicate whether the symbol was associated with a face or a sound.

It was predicted that retrieval of face information would reactivate face-responsive visual regions, and that retrieval of sound information would reactivate auditory regions (Hofstetter et al., 2012; Nyberg et al., 2000; Polyn et al., 2005; Salami et al., 2010; Wheeler et al., 2000). This was expected to hold for both the associations that remained intact throughout the learning program, and for associations established during the second half of learning. The critical issue was whether the latter class of items would additionally be associated with "residual" activity reflecting previously formed but no longer

http: //dx.doi.org/10.1016/j.neuroimage.2014.06.049

1053-8119/© 2014 The Authors. Published by Elsevier Inc. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/3.0/).

relevant associations. That is, would symbols paired with faces during days 1 -2 and with sounds during days 3-4 be associated with signal change in face-responsive regions at retrieval — despite the fact that the participants correctly reported that these symbols were associated with sounds (and vice versa for symbols initially paired with sounds)?

The experiment also included a learning/updating stage that took place after memory retrieval on a trial-by-trial basis, where a symbol was presented along with the face or sound it had been associated with during training or with a new, unique item. In the latter case, the participants were instructed to form new associations, which made it possible to examine the actual process of updating long-term memory during fMRl scanning (cf., Nyberg et al., 2009). Lastly, to investigate memory-trace durability, participants were re-scanned 18 months after the first session to examine whether reactivation patterns and/or residual activations would endure.

Methods

Participants

Twenty-one right-handed participants were recruited from the Umea University campus area (19-27 years old, 14 women). All gave written informed consent and were paid 600 SKR for participation. Circa 18 months after participation in the first fMRl session all participants were asked to come back for a follow-up fMRl session; 12 participants (23-29 years old, 5 female) accepted and were paid an additional 300 SKR. Due to technical issues, data from behavioral responses during the first fMRl session was lost for three participants, who therefore were excluded from analyses of data from session 1. The study was approved by the local ethics committee.

Procedure

The experiment consisted of 4 days of learning and 1 day of scanning, followed 18 months later with one "booster" session and a second scanning session (Fig. 1A).

Pre-scan learning procedure

During initial learning, Japanese Kanji-signs were presented paired with either a sound (animals, tools, etc.) or a face. During days 1 and 2 (Phase 1) the participants learned 48 symbol-sound and 48 symbol-face pairs. During days 3 and 4 (Phase 2) there was a shift in the training schedule. For half of the items (24 of each kind), the same pairs were presented and further trained. For the remaining half the pairs were changed so new associations had to be formed, such that symbol-sound associations became symbol-face associations (new faces) and vice versa.

Thus, a symbol associated with a face during days 1-2 had to be updated and associated with a novel sound during days 3-4, and vice versa for symbols initially associated with sounds. The participants were explicitly instructed that new associations were to be formed and that the previous association was no longer relevant.

The symbols were Japanese Kanji-signs that were taken from an online site (www.thejapanesepage.com) and were chosen so as to have as little resemblance to each other as possible. The face images were digital color photographs of faces, mainly taken from a set of emotional face expressions (Lundqvist et al., 1998; only emotionally neutral faces were used). The sounds were two-channel wav-files of animals, tools in use, or everyday sounds such as knocking on a door, collected from different sites on the lnternet and chosen to be as distinct from each other as possible. To familiarize participants with the sounds and to ensure that each sound could be meaningfully identified, all sounds and their labels (one or two words describing the sounds) were presented before training on the first and third days. Symbols and faces were presented on a computer screen, and the sounds were delivered through external speakers (Fig. 1A). The specific symbol-sound/face combinations were counterbalanced across participants, such that each symbol was paired with both sounds and faces.

To optimize learning efficiency, each learning session was organized as 3-5 study-test cycles where all items were first studied, then tested, etc. (Karpicke and Roediger, 2008). On the first day of each phase (i.e., days 1 and 3), only failed items were presented on the next study/test round. On days 2 and 4, all items were first studied and tested. ln the following rounds on the same day, only failed items were studied but all items were tested. The participants were explicitly instructed to memorize each pair, and were given unlimited time to study each item (sounds were looped during study). Pilot runs indicated that participants tended to go through the study rounds relatively quickly, resulting in poor memory performance. To encourage participants to attend to the stimulus, they were in the actual experiment asked to make a like/dislike judgment on each study item. During testing, a symbol was presented for 2 s and participants then responded whether the symbol was associated with a sound or face, or if they did not know. lf correctly indicating face/sound, a follow-up 4-alternative forced-choice question was presented (1 correct alternative and 3 incorrect alternatives drawn randomly from the stimulus pool; for sounds, the corresponding labels were presented rather than the actual sounds). The participants were given feedback ("correct"/"incorrect") after each test item. To monitor participants' false response rate, 20 symbols that had not been presented during study (and therefore were not associated with either a sound or a face) were also presented during each test round. A new set of 20 "lures" was added each day, and repeated during each test round on that day.

Phase 1

Day 1 Day 2

Original

96 comb.

Phase 2 Day 3 Day 4 1st fMRI Day 5

Original - Original Original

48 comb. 24 comb.

\ Updated while in scanner

24 comb.

Updated - Updated - Updated

2nd fMRI

18 months

Original

Updated

Re-encode or update, 2 s

48 comb.

Fig. 1. A, Overview of experimental phases. Days 1 and 2 consisted on establishing symbol-sound and symbol-face associations. During days 3 and 4, half of the associations were updated, such that symbol-face associations became symbol-sound associations and vice versa. fMRl scanning was performed on day 5 and also 18 months later. B, Scanner procedure during day 5. Each trial consisted of two parts: one cued retrieval part and one re-encoding/updating part. The scanner procedure at 18 months only included the first part (cued retrieval).

Scanner procedures

During the first fMRI session, each trial was divided into two parts: one cued retrieval part to investigate memory associations and one learning/updating part to investigate the processes of updating memories (Fig. 1B). During the retrieval part, a symbol was displayed for 2 s and the participants indicated whether this symbol was associated with (i) a face, or (ii) a sound, or (iii) if they did not remember, by pressing one of three buttons with their right hand. Participants were explicitly instructed that only the most recently learned association was relevant. The learning/updating part, which immediately followed the retrieval part on a trial-by-trial basis, was separated from the retrieval part by display of a fixation circle centered on the screen for 2.7-8.7 s. During updating/learning, the symbol was again presented (2 s) but now together with either the previously learned associate (continued learning; no symbol was presented together with the old associate of an updated pair) or with a sound/face not previously associated with a symbol (updating). This made it possible to examine the actual process of updating long-term memory during fMRI scanning. The participants were instructed to indicate whether the pair was old/new and to remember new associations. Each trial (cued retrieval + learning/ updating) was separated by a 3-9 s fixation cross. To avoid novelty effects during updating, all new sounds/faces had been presented in each study session during the learning procedure (preceding days), where the participants were asked to make a like/dislike judgment to ensure that they attended to the stimuli.

During the second fMRI session, 18 months after the first, a similar procedure as for the first session was used but with the following changes. Each item was presented three times (randomly intermixed with other items) to increase statistical power. No updating procedure was used and no item that was updated during the first scanner session was presented. Thus, for each trial a symbol was displayed and the participants indicated whether this symbol was associated with a face or a sound, or if they did not remember, by pressing one of three buttons with their right hand. The second fMRI session was preceded by a "booster" session, where participants were presented with all symbols (N = 72) except those that had been updated during the first scanner session, and indicated with a button press whether each symbol was associated with a face or a sound, or if they did not remember. The purpose of this booster session was to remind the participants of the materials. It also offered unlimited time for recall to enable as many recalled items as possible, as the response time during scanning was limited to 2 s. Thus, the booster session served to minimize trials where participants would have remembered the associated item if only given enough time during scanning.

Data acquisition and analyses

Data collection (both fMRI sessions) was made on a 3 T Phillips Achieva scanner (Philips Medical Systems, Netherlands). Functional T2*-weighted images were obtained with a single-shot GE-EPI sequence for BOLD imaging (TR = 1500 ms; flip angle = 70 deg; echo time = 30 ms; 31 slices acquired: 3.44 x 3.44 mm in-plane x 4.65 mm thick; an 8-channel SENSE head coil and a SENSE-factor of 2.6 was used). To eliminate signals arising from progressive saturation, ten dummy scans were performed prior to image acquisition.

Data were analyzed using multiple regression implemented in SPM8 (Wellcome Department of Cognitive Neurology, London, UK). Image preprocessing consisted of realignment and unwarping, slice-timing correction, normalization to MNI space, and smoothing using an 8.0 mm FWHM Gaussian kernel. Temporal autocorrelations were estimated using a first-order autoregressive model. Data were scaled over sessions and high-pass filtered (128 s cut-off). The retrieval part of each trial was modeled with 4 regressors: symbols associated with sounds (S) throughout all 4 days of learning (SSSS, where the first S represents "sound association day 1 ", the second S "sound association day 2" etc.), symbols first associated with sounds but then changed

their associate to a face (F) during days 3-4 (SSFF), and the symbol-face counterparts (FFFF and FFSS). The learning/updating part of each trial was modeled with 6 regressors: 4 corresponding to continued learning (FFFF-F, SSFF-F, SSSS-S, and FFSS-S) and 2 for memory updating (SSSS-F and FFFF-S). The second fMRI session (18 months) only had the retrieval trial part and was modeled accordingly. Unsuccessful trials (wrong answers or not remembered) were modeled with separate regressors (both trial parts). All regressors were convolved with a canonical HRF as implemented in SPM8. Head-motion parameters were included as effects of no interest. Model estimations (restricted maximum likelihood) from each individual were taken into second level random-effects analyses (one-sample t-tests) to account for inter-individual variability.

To define regions related to sensory processing of sounds/faces, we contrasted all symbol-sound presentations with all symbol-face presentations during the learning/updating part of each trial (p < .001, k > 10 voxels), where sounds and faces were actually presented. The resulting regions were then used as regions of interest for analyzing brain activity related to the retrieval trial part (p < .05 uncorrected) during both fMRI sessions 1 and 2. The statistical thresholds were motivated by the a priori predictions of effect locations. Reactivation of sensory regions during retrieval was revealed by contrasting SSSS > FFFF (original sound associations) and FFSS > FFFF (updated sound associations that were symbol/face associations during days 1 and 2) for auditory cortex; FFFF > SSSS (original) and SSFF > SSSS (updated) for face-processing regions. Residual activity related to no longer relevant associations was revealed by FFSS > SSSS and SSFF > FFFF, respectively.

Brain activity related to updating of memory representations (only fMRI session 1) was investigated by contrasting the two conditions where associations changed during scanning with the two conditions in which the associate never changed: (SSSS-F + FFFF-S) > (SSSS-S + FFFF-F). Thus, the updating process (continued learning/updating part of each trial) captured by this contrast reflected activity independent of modality of the associate.

Results

After two days of training on the 48 symbol-face and 48 symbolsound pairs, participants had reached about 90% correct retention (Fig. 2; Phase 1). At the final test on day 2, symbol-sound pairs were remembered better than symbol-face pairs (t[17] = 2.38, p = .029). During Phase 2 (days 3 and 4), where half of the symbol associations were updated, it was found that the learning of new associations was faster than during initial learning (initial test performance after the first studyroundday3vs. day1:F[1,17] = 105.47, p < .001; learning rate between tests 1 and 2 on day 3 vs. day 1: F[1,17] = 11.41, p = .004).Also, there was a significant interaction between updating/continued learning

Phase 1 Phase 2 fMRI

Day 1 Day 2 Day 3 Day 4 Day 5 18 months

60 40 20

C3 i , W t sa

-•- Face Original -o- Sound-To-Face Update Sound Original Face-To-Sound Update

Fig. 2. Learning curves for intact (solid lines) and updated (dashed lines) associations across repeated study-test runs during each day of learning, and memory performance during scanning.

and material type, such that the transition from face to sound was more difficult than that from sound to face (performance at first test day3: F[1,17] = 9.36, p = .007). During the final test on day 4, performance was 96 and 94% correct retention for original and updated associations, respectively (SE = 1.2 and 1.7). There was no significant effect of either material type (face vs. sound) or update history, or the interaction thereof (material-by-update-history repeated-measures ANOVA, all p > .19). This was true also for performance in the scanner during day 5 (all p > .46), where performance was still high (original = 91% [2.9], updated = 90% [2.6]).

Correctly remembering that a symbol was associated with a face or a sound was associated with increased BOLD signal in visual and auditory regions, respectively (Fig. 3). As predicted, this was true for both the original associations and those updated during training. Most critically, residual activation was also found in face- and sound-responsive regions. Coordinates for peak voxels within the functionally defined visual cortex ROI (MNI xyz: 42 - 78 - 24; - 38 - 72 -18) matched well (within 10 mm for any coordinate) with the location of the "occipital face area", as described in previous research (Chen et al., 2010: left - 37 - 69 - 7, right 37 - 71 - 7; Rotshtein et al., 2005: left -45 - 78 - 15, right 42 - 69 - 18).

The results thus demonstrate that cued retrieval of associations updated during training reactivates both visual and auditory cortices. To investigate how the brain "decides" which representation is correct, we compared retrieval of updated associations with retrieval of original associations, across both symbol-face and symbol-sound associations (whole-brain conjunction analysis between "updated face" and "updated sound", p < .001, k > 10). This revealed a single cluster in the left inferior frontal sulcus (xyz: -42 18 36, k = 15, t = 3.95).

Separate from memory reactivation (the retrieval part), the process of updating memory (updating vs. continued learning during the second part of each trial) engaged fronto-parietal cortical areas (Fig. 4, Table 1), including a left lateral frontal region that partially overlapped at the dorsal end with the left frontal region from the "updated vs. original retrieval" analysis above.

Eighteen months later

Eighteen months after the first scanning session, memory performance during scanning was greatly reduced compared with the initial

session, and only original symbol-face pairs were identified better than chance (M = .64%,SE = .05, t[11] = 2.66, p = .022). Interestingly, the performance on updated symbol-sound pairs was significantly worse than chance (M = .40[.03], t[11] = - 3.26, p = .008; for original symbol-sound: M = .56 [.05], t[11] = 1.26, p = .23, for updated symbol-face: M = .53 [.04] t = 0.75, p = .47). That is, for symbols that had initially been paired with faces but during days 3-4 were changed to symbol-sound pairs, participants tended to respond that they had been paired with faces, as if the original symbol-face association dominated over the more recently formed symbol-sound association.

The fMRl data were consistent with the behavioral pattern: only original face-associations elicited significant BOLD signal change in face-responsive regions (Fig. 3B). Moreover, the magnitude of signal change in face-responsive regions was negatively correlated across participants with performance during recall of updated symbol-sound associations (beta values for "residual" activity, r[12] = — .82, p < .001; Fig. 3C). To investigate whether brain activity already during the first session was predictive of this "reversed" memory performance, we regressed memory performance 18 months later on BOLD signal during initial retrieval of updated sound associations. There was a significant positive relation between performance and BOLD signal in auditory cortex (r[10] = .73, p = .009), and also a significant negative relation in face-responsive regions (r[10] = — .72, p = .009; Fig. 3C), such that lower auditory cortex activity and higher residual face activity during retrieval of updated symbol-sound pairs during the first scanner session predicted worse memory performance 18 months later.

Discussion

In line with previous research, we show that content-specific parts of the cortex are reactivated during memory retrieval (Hofstetter et al., 2012; Nyberg et al., 2000; Polyn et al., 2005; Salami et al., 2010). Specifically, the BOLD signal in the visual cortex increased when participants recalled face information, whereas BOLD signal in the auditory cortex increased during recall of sound information. Critically, we found evidence suggesting that representations of no longer relevant mnemonic information were also reactivated during retrieval of altered/updated memory associations. Thus, even though the residual BOLD signal response was generally weaker compared with reactivation related to currently relevant information, the results give little reason to

Auditory cortex

Visual cortex: 42 -78 -24

Day 5 18 months

Auditory cortex: -66 -44 6 OUR

Face-to-sound update fMRI Day 5

0.0 Beta values

fMRI 18 months

-1.0 0.0 Beta values

Fig. 3. A, Reactivation (pink) for original (O) and updated (U) items and residual (R) activation in face-responsive (upper) and sound-responsive (lower) regions (blue) during cued retrieval (1st fMRI session). B, BOLD signal change during day 5 (three leftmost bars, values are averages across voxels within the cluster) and 18 months later (three rightmost bars). Error bars signify 1 standard error. C, Correlations between memory performance for updated symbol-sound associations 18 months later and BOLD signal in visual cortex (dark green line and dots) and auditory cortex (purple line and dots) during fMRI scanning day 5 (upper plot) and 18 months later (lower plot). Dashed lines signify chance performance.

believe that old information is lost. Indeed, as is apparent from the interference from previously learned associations across time (18 months), old and no longer relevant information can be long lasting. These findings support a view of memory updating where old information is part of the newly formed representational structure rather than being overwritten or treated as a separate memory trace. This view is further supported by the behavioral results from day 3, where learning of new associations was faster compared with initial learning, in line with findings that new learning can be facilitated by already established knowledge structures (Tse et al., 2007). Retaining previously learned mnemonic information as part of a representational structure may not only help during learning of new information, but also make sense from an ecological perspective in that old information may be useful in future situations. However, such benefits may be partly counterbalanced by possible interference effects between related associates, as was evident in the results from the 18-month follow-up.

The present findings are in line with studies showing that fear extinction (i.e., unlearning of a previously learned stimulus-threat association) is usually incomplete and unstable over time, such that unlearned memory associations reoccur (see Bouton, 2002, for review). However, recent research has indicated that previously learned associations can be essentially deleted through new learning, provided that the new learning occurs within a specific time window following memory reactivation, such that the new learning thereby interferes with the so-called reconsolidation process (Agren et al., 2012; Monfils et al., 2009; Schiller et al., 2010). Reconsolidation is taken to be a renewed state of instability of a previously consolidated memory trace through trace reactivation (Dudai, 2004), and provides an opportunity to update the memory trace with information during the reactivation period

Table 1

Brain regions showing increased BOLD signal for updating > re-encoding (second part of each trial).

Brain region x y z t-Value Cluster size BA

Inferior frontal gyrus 52 28 30 10.34 371 44

-52 24 26 6.55 731 44

48 -34 16 5.53 36 47

Inferior parietal cortex 42 -68 46 8.89 492 39/7

-46 -72 34 7.94 587 39

Precuneus 10 -60 32 8.52 602 7

Superior frontal gyrus -14 62 18 6.87 120 10

18 60 2 5.09 16 10

Posterior cingulate cortex -4 -30 -32 6.81 95 23

Temporal pole 46 16 -38 5.90 35 20

30 20 -32 4.27 10 38

Cerebellum -12 -88 -36 4.80 34 -

Calcarine/cuneus -16 -66 22 4.53 14 17

Supplementary motor area -8 22 54 4.48 10 8

Note. xyz = MNI coordinates, BA = Brodmann area.

(Lee, 2009). The present experiment, with cycles of study and test periods, gave plenty of opportunity for memory updating within the reconsolidation time window, which has been shown to be open for several hours (Nader et al., 2000; Schiller et al., 2010). Still, the current results demonstrate that outdated memories need not be replaced by new and more relevant information, but instead that updated and outdated mnemonic information can coexist. Thus, the mechanisms for learning a new associate to a memory token in the present paradigm seems to be more like consolidation than re-consolidation (Tronel et al., 2005).

Possibly, a critical determinant of reconsolidation is memory strength, as it has been suggested that only incompletely learned associations are subject to reconsolidation processes, whereas new learning in relation to well-learned associations may instead lead to parallel memory traces (Lee, 2009). In the current experiment, the paired associates had been learned to near-asymptote levels after the first two days (~90% correct performance), thereby making initial memory strength a plausible explanation of the durability of old associations in the face of new learning and possible reconsolidation processing. Moreover, such strong initial establishment of the original associations gives credence to a possible generalization of the current results to real-world situations outside the laboratory, e.g., when long-established knowledge such as names of countries changes.

Given strong initial memory traces, it is critical to consider how efficient the updating component of the experiment was. It would, after all, not be unexpected to find evidence for residual memories if the original associations were strong but the memory updating procedure was weak. However, the behavioral performance at the final test on day 4 as well as during scanning (day 5) showed that recall of updated associations was not significantly different from original associations (94 vs. 96% on day 4, 90 vs. 91% during scanning). Thus, the two days of establishing new memory associations (days 3-4) seem to have led to as strong learning as for the initial associations. The current results thereby substantiate and extend previous indications of residual activation, where AB-AC procedures have been used to study memory interference (with only one updating opportunity; Kuhl et al., 2011, 2012; Waldhauser et al., 2012).

Memory interference between old and new associations was evident also in the present results, in that stronger residual reactivation during retrieval of updated memories correlated with poorer performance 18 months later, in line with previous research on extinction (reoccurrence of unlearned behavior, Bouton, 2002). Similarly, Kuhl et al. (2012) recently demonstrated that category-specific brain activity during memory updating predicted later memory performance, such that stronger reactivation of old memory traces was correlated with less accurate new memories. Comparing retrieval of updated associations with associations that remained intact throughout the learning process revealed BOLD signal change in left prefrontal cortex, which may reflect

interference resolution between competing mnemonic representations (Badre and Wagner, 2007). Competition between memory associations is known to reduce memory performance (Danker et al., 2008), and inhibition of alternative memory associations has been suggested as a mechanism of forgetting (of the non-recalled associate; Anderson and Spellman, 1995; see Kuhl et al., 2011; Oztekin and Badre, 2011; Waldhauser et al., 2012, for further neurophysiological support). Inhibition between old and new memories is consistent with the current cross-over correlation pattern (Fig. 3C), where memory performance at 18 months on updated symbol-sound associations correlated positively with auditory, but negatively with visual cortex BOLD signal change at day 5, suggesting that stronger currently relevant mnemonic information inhibited old information more.

Surprisingly, in the present experiment it was the new and currently relevant associations that were forgotten. This result was material-specific in that behavioral performance as well as BOLD signal changes indicated that only symbol-face associations remained after 18 months (original symbol-face associations, as well as interfering face associations to updated symbol-sound items). It is unclear why face associations were remembered better over time compared to sounds. During the acquisition phase (days 1-4), there was no evidence for superior performance for symbol-face associations. Indeed, at the final test on day 2, symbol-sound associations were remembered slightly better than faces. However, there was a significant material-by-update-history interaction at the first test day 3, showing that it was harder to learn symbol-sound associations if the previous association had been with a face than vice versa. Possibly, the more durable symbol-face associations reflect the privileged status that visual information seems to have (Standing, 1973). Regardless of the reason for the difference between symbol-sound and symbol-face associations, the effect of type of association on the long-term effects of memory updating calls for cautiousness regarding the generalizability of the current results to other types of materials.

During memory updating in the scanner, BOLD signal increases were evident in frontal and parietal cortices, bilaterally, compared to re-encoding. Increased frontal cortex activity is a consistent finding in relation to updating, both of working (Dahlin et al., 2008; Nee et al., 2013) and long-term (Dolan and Fletcher, 1997; Nyberg et al., 2009) memories, which indicates shared processes (Nyberg et al., 2003). Computational models of memory updating suggest involvement of both frontal and subcortical structures, as well as involvement of dopaminergic neurotransmission (O'Reilly, 2006). Correspondingly, a relation between striatal dopamine functioning and frontal cortex activity during long-term memory updating has been demonstrated (Nyberg et al., 2009). A causal relation between lateral frontal cortex and (working) memory updating, as well as correlations between behavioral performance, activity in brainstem dopamine nuclei, and the lateral frontal cortex has also been shown (D'Ardenne et al., 2012). The inferior frontal gyrus (IFG) has been suggested to be involved in interference resolution between competing mnemonic information during both long-term memory encoding (Fletcher et al., 2000) and retrieval (Badre and Wagner, 2007), in line with the partial overlap with the PFC cluster found to differentiate between retrieval of updated vs. intact memory associations discussed above. In support for IFG involvement during encoding, Kuhl et al. (2012) demonstrated that higher activity in the left IFG during memory updating predicted less competition between old and new memory information. Contrary to the interference resolution suggestion, recent research has indicated that the lateral prefrontal cortex (PFC) may not be called upon to resolve interference, but rather to instantiate it (see Robertson, 2012, for review). For example, transcranial magnetic stimulation of the dorsolateral PFC after learning a word list and a motor skill task reduced interference otherwise evident between the two memory tasks (Cohen and Robertson, 2011). According to this view, memory interference may not be a by-product of competing memories, but instead add functional significance, speculatively related to integration of different memory traces.

Memory updating was also related to BOLD signal change in the parietal cortex. During the updating phase, new faces/sounds were presented and paired with pre-learned symbols. This may have induced some form of novelty response and attentional processes, which can be expected to involve frontoparietal regions (Corbetta and Shulman, 2002; Petersen and Posner, 2012). Novelty effects as such should be limited given that all faces and sounds had been presented during the preceding days. Rather, attention may be seen as part of the updating process and the left ventral parietal cortex in particular has been suggested critical for attention to memory (Cabeza et al., 2012).

In conclusion, we show that no-longer relevant mnemonic information is activated during retrieval of related information — even if this information is not actively sought after. As such, the revision of long-term memory can be seen as adding information to already existing networks where the old, pre-existing memory traces still have a high probability of being activated when some units of the network that represent more recent or relevant information are activated, and may even come to dominate over more recent learning.

Acknowledgments

This work was supported by the Swedish Science Council and the Goran Gustafsson award. LN was supported by awards from Torsten and Ragnar Soderberg's Foundation and Knut and Alice Wallenberg's (KAW) Foundation.

Conflict of interest

References

Agren, T., Engman, J., Frick, A., Bjorkstrand, J., Larsson, E.-M., Furmark, T., Fredrikson, M., 2012. Disruption of reconsolidation erases a fear memory trace in the human amygdala. Science 337,1550-1552.

Anderson, M.C., Spellman, B.A., 1995. On the status of inhibitory mechanisms in cognition: memory retrieval as a model case. Psychol. Rev. 102,68-100.

Anderson, M.C., Ochsner, K.N., Kuhl, B., Cooper, J., Robertson, E., Gabrieli, S.W., Glover, G.H., Gabrieli, J.D.E., 2004. Neural systems underlying the suppression of unwanted memories. Science 303,232-235 (80-.).

Badre, D., Wagner, A.D., 2007. Left ventrolateral prefrontal cortex and the cognitive control of memory. Neuropsychologia 45,2883-2901.

Bouton, M.E., 2002. Context, ambiguity, and unlearning: sources of relapse after behavioral extinction. Biol. Psychiatry 52, 976-986.

Cabeza, R., Ciaramelli, E., Moscovitch, M., 2012. Cognitive contributions of the ventral parietal cortex: an integrative theoretical account. Trends Cogn. Sci. 16, 338-352.

Chen, J., Zhou, T., Yang, H., Fang, F., 2010. Cortical dynamics underlying face completion in human visual system. J. Neurosci. 30, 16692-16698.

Cohen, D.A., Robertson, E.M., 2011. Preventing interference between different memory tasks. Nat. Neurosci. 14, 953-955.

Corbetta, M., Shulman, G.L., 2002. Control of goal-directed and stimulus-driven attention in the brain. Nat. Rev. Neurosci. 3, 201-215.

D'Ardenne, K., Eshel, N., Luka, J., Lenartowicz, A., Nystrom, L.E., Cohen, J.D., 2012. Role of prefrontal cortex and the midbrain dopamine system in working memory updating. Proc. Natl. Acad. Sci. U. S. A. 109,19900-19909.

Dahlin, E., Neely, A.S., Larsson, A., Backman, L., Nyberg, L., 2008. Transfer of learning after updating training mediated by the striatum. Science 320,1510-1512 (80-.).

Danker, J.F., Gunn, P., Anderson, J.R., 2008. A rational account of memory predicts left prefrontal activation during controlled retrieval. Cereb. Cortex 18,2674-2685.

Dolan, R.J., Fletcher, P.C., 1997. Dissociating prefrontal and hippocampal function in episodic memory encoding. Nature 388, 582-585.

Dudai, Y., 2004. The neurobiology of consolidations, or, how stable is the engram? Annu. Rev. Psychol. 55, 51-86.

Dudai, Y., 2006. Reconsolidation: the advantage of being refocused. Curr. Opin. Neurobiol. 16,174-178.

Fletcher, P.C., Shallice, T., Dolan, RJ., 2000. "Sculpting the response space" — an account of left prefrontal activation at encoding. Neuroimage 12,404-417.

Hofstetter, C., Achaibou, A., Vuilleumier, P., 2012. Reactivation of visual cortex during memory retrieval: content specificity and emotional modulation. Neuroimage 60, 1734-1745.

Karpicke, J., Roediger, H., 2008. The critical importance of retrieval for learning. Science 319, 966-968 (80-.).

Kuhl, BA, Rissman, J., Chun, M.M., Wagner, A.D., 2011. Fidelity of neural reactivation reveals competition between memories. Proc. Natl. Acad. Sci. U. S. A. 108,5903-5908.

Kuhl, B.A., Bainbridge, W.A., Chun, M.M., 2012. Neural reactivation reveals mechanisms for updating memory. J. Neurosci. 32, 3453-3461.

Lee, J.L.C., 2009. Reconsolidation: maintaining memory relevance. Trends Neurosci. 32, 413-420.

Loftus, E.F., 2005. Planting misinformation in the human mind: a 30-year investigation of the malleability of memory. Learn. Mem. 12, 361-366.

Lundqvist, D., Flykt, A., Ohman, A., 1998. The Karolinska Directed Emotional Faces—KDEF. CD-ROM from Department of Clinical Neuroscience, Psychology Section, Karolinska Institutet, Stockholm, Sweden.

McClelland, J.L., McNaughton, B.L., O'Reilly, R.C., 1995. Why there are complementary learning systems in the hippocampus and neocortex: insights from the successes and failures of connectionist models of learning and memory. Psychol. Rev. 102, 419-457.

Monfils, M.-H., Cowansage, K.K., Klann, E., LeDoux, J.E., 2009. Extinction-reconsolidation boundaries: key to persistent attenuation of fear memories. Science 324, 951-955 (80-.).

Nader, K., Schafe, G.E., LeDoux, J.E., 2000. Fear memories require protein synthesis in the amygdala for reconsolidation after retrieval. Nature 406, 722-726.

Nee, D.E., Brown, J.W., Askren, M.K., Berman, M.G., Demiralp, E., Krawitz, A., Jonides, J., 2013. A meta-analysis of executive components of working memory. Cereb. Cortex 23, 264-282.

Nyberg, L., Habib, R., Mcintosh, R., Tulving, E., 2000. Reactivation of encoding-related brain activity during memory retrieval. Proc. Natl. Acad. Sci. U. S. A. 97,11120-11124.

Nyberg, L., Marklund, P., Persson, J., Cabeza, R., Forkstam, C., Petersson, K.M., Ingvar, M., 2003. Common prefrontal activations during working memory, episodic memory, and semantic memory. Neuropsychologia 41, 371-377.

Nyberg, L., Andersson, M., Forsgren, L., Jakobsson-Mo, S., Larsson, A., Marklund, P., Nilsson, L.-G., Riklund, K., Bâckman, L., 2009. Striatal dopamine D2 binding is related to frontal BOLD response during updating of long-term memory representations. Neuroimage 46,1194-1199.

O'Reilly, R.C., 2006. Biologically based computational models of high-level cognition. Science 314,91-94 (80-.).

Oztekin, I., Badre, D., 2011. Distributed patterns of brain activity that lead to forgetting.

Front. Hum. Neurosci. 5,86. Petersen, S.E., Posner, M.I., 2012. The attention system of the human brain: 20 years after.

Annu. Rev. Neurosci. 35, 73-89. Polyn, S.M., Natu, V.S., Cohen, J.D., Norman, KA, 2005. Category-specific cortical activity

precedes retrieval during memory search. Science 310,1963-1966 ( 80-.). Robertson, E.M., 2012. New insights in human memory interference and consolidation.

Curr. Biol. 22, R66-R71. Rotshtein, P., Henson, R.N. a, Treves, A., Driver, J., Dolan, R.J., 2005. Morphing Marilyn into Maggie dissociates physical and identity face representations in the brain. Nat. Neurosci. 8,107-113.

Salami, A., Eriksson, J., Kompus, K., Habib, R., Kauppi, K., Nyberg, L., 2010. Characterizing the neural correlates of modality-specific and modality-independent accessibility and availability signals in memory using partial-least squares. Neuroimage 52, 686-698.

Schiller, D., Monfils, M.-H., Raio, C.M., Johnson, D.C., Ledoux, J.E., Phelps, E.A., 2010. Preventing the return of fear in humans using reconsolidation update mechanisms. Nature 463,49-53. Standing, L., 1973. Learning 10,000 pictures. Q.J. Exp. Psychol. 25,207-222. Tronel, S., Milekic, M.H., Alberini, C.M., 2005. Linking new information to a reactivated memory requires consolidation and not reconsolidation mechanisms. PLoS Biol. 3, e293.

Tse, D., Langston, R.F., Kakeyama, M., Bethus, I., Spooner, P.A., Wood, E.R., Witter, M.P., Morris, R.G.M., 2007. Schemas and memory consolidation. Science 316,76-82 (80-.). Waldhauser, G.T., Johansson, M., Hanslmayr, S., 2012. Alpha/beta oscillations indicate

inhibition of interfering visual memories. J. Neurosci. 32,1953-1961. Wheeler, M.E., Petersen, S.E., Buckner, R.L., 2000. Memory's echo: vivid remembering reactivates sensory-specific cortex. Proc. Natl. Acad. Sci. 97,11125-11129.