续接02

Taken together, our findings highlight the interaction between memory consolidation and suppression, and invite consideration of their possible clinical significance. Aberrant memory consolidation has been implicated in cognitive models of PTSD, and sleep is believed to play a vital role in this process6,8. These models posit that the emotional intensity of a traumatic memory is reduced during sleep through an active decoupling process whereby the emotional arousal is reduced and the non-emotional memory content is consolidated. If this process is prevented, for instance by disturbed sleep, the emotional ‘charge’ of a traumatic memory remains, resulting in an inability to forget the trauma which is the hallmark of PTSD8. Related cognitive models of PTSD have led to promising therapeutic approaches10,34,45. For example, sleep deprivation immediately after traumatic experiences may prevent traumatic memories from being consolidated into stabilized representations and thus provide the opportunity to block the formation of traumatic memories46. In addition, it is also possible that our observed effects might simply be dependent on time rather than overnight sleep, or both. Our current design cannot separate general time- from sleep-dependent effects on memory consolidation. This possibility should be considered, since previous studies have showed that arousal-mediated consolidation effects are dependent on time, not sleep47 (but see Wilhelm et al.48). Future studies including a wake control group are needed to disentangle these effects.

In conclusion, our study demonstrates that suppression of consolidated aversive memories, compared with newly acquired memories, is characterized by less suppression-induced forgetting in conjunction with enduring emotional reactivity, which is associated with a higher prefrontal inhibitory control over hippocampal memory and amygdala emotional systems and a shift away from hippocampal-dependent representational patterns to distributed neocortical representational patterns. Our findings point towards a neurobiological model through which overnight consolidation assimilates aversive memories into more distributed neocortical representations, and makes these memories more resistant to suppression through the prefrontal–hippocampal inhibitory pathway. Our study underlines the importance of memory consolidation in understanding the resistance to suppression of emotional memories, which is a cardinal feature of affective disorders.

Methods

Participants

Twenty-one young healthy, male, right-handed college students (mean age±s.d., 21.6±0.81 years ranged from 20 to 24) with normal or corrected-to-normal vision participated in this study. Participants followed regular daily work-rest schedule, and reported no sleep-related problems during the last two months⁴⁹. All participants reported good quality sleep during the night after training on day 1, with 7–9 h sleep time (mean±s.d.=7.35±0.37). Informed written consent was obtained from all participants before the experiment, and the study protocol was approved by the Institutional Review Board for Human Subjects at Beijing Normal University. Data from three participants were excluded from fMRI analyses due to excessive head movement during scanning. Note that another independent cohort of 25 participants (age range from 20 to 24 years old) was recruited for a parallel study using the same paradigm to examine the reproducibility of our behavioural and SCR findings. An additional behavioural control experiment with neutral stimuli was conducted in another independent 30 participants (age range from 20 to 24) to examine whether the observed effects of overnight consolidation on aversive memories were generalizable to neutral memories.

General procedure

The whole experiment consisted of three phases: memory acquisition, memory suppression, and memory testing (Fig. 1). The memory acquisition phase comprised two sessions on day 1 and day 2, which occurred about 24 h (that is, training on day 1) and 30 min (that is, training on day 2) before fMRI scanning. On day 1, participants were extensively trained to learn and remember a set of 26 face-aversive picture associations. On day 2, they returned and were trained to remember another set of 26 face-picture associations. About 30 minutes after the training on day 2, participants underwent the scanning during which they were instructed to perform the memory suppression task using ‘Think/NoThink’ paradigm⁴ in conjunction with concurrent recording of SCRs. Following that, participants performed a post-scan memory test for face-picture associations to assess their subsequent memory performance and effectiveness of memory suppression.

Stimuli

Fifty-two face-aversive picture pairs were used in the present study. Fifty-two faces (26 males and 26 females) were carefully selected from 100 colour photographs of Chinese individuals unknown to the participants⁵⁰. To standardize the stimuli and minimize potential confounding factors, faces were selected under the following criteria suggested by previous studies^51,52: direct gaze contact, no strong emotional facial expression (which was rated as having a neutral expression in a pilot study on separate 9-point scales: 1=‘extremely sad’ or ‘not arousing at all’, 9=‘extremely happy’ or ‘extremely arousing’, with mean valence=5.16±0.53 and mean arousal=5.03±0.43), no headdress, no glasses, no beard, etc. There was no significant difference in terms of arousal, valence, attractiveness, and trustworthiness between male and female faces (all P>0.05). Subsequent analysis also revealed no difference in memory accuracy between male and female faces (t(17)=−0.88, P=0.39). Fifty-two aversive pictures from the International Affective Picture Series⁵³ were carefully selected as having minimal relatedness in content with each other as possible, with a highly negative level of emotional valence (mean valence=2.37±0.69) and a high level of negative arousal (mean arousal=7.89±0.55) as measured on 9-point scales. Faces and aversive pictures were randomly paired across participants to create 52 face-picture associations. Associations were randomly assigned to day 1 and day 2. For behavioural control experiment (testing for neutral memories), 52 neutral pictures were chosen from International Affective Picture Series and online resources. They were carefully matched on complexity, luminance and contrasts with the aversive pictures, and have modest level of emotional valence (M=5.32±0.47) and low level of arousal (M=2.52±0.36) based on ratings from an independent sample.

Memory acquisition

During the acquisition phase, participants performed a training session on day 1 and another on day 2 outside the scanner, which occurred 24 h and 30 min before the Think/NoThink task respectively. We carefully controlled the time interval between 30 min and 24 h on day 1 and day 2 to minimize variability across participants⁵⁴. We thus restricted the first training phase to start at 16:00 hours on day 1, the scanning during ‘Think/NoThink’ phase to 16:00–17:00 hours on day 2. The day 2 training started a half hour before scanning. In each training session, participants were trained to memorize 26 face-aversive picture pairs using multiple study–recall cycles. In each study–recall cycle, each association was presented for 4 s, and participants were encouraged to remember the association in detail. After presentation of all associations, participants were then shown a face and asked to recall details of the corresponding associated picture. This study–recall cycle was repeated 3–5 times until each participant could recall correctly all 26 face-picture associations. For each association, we required the participants to give detailed descriptions about the associated images when faces were presented as cues. The participants were asked to give enough details of that image to enable that it to be uniquely identified. This procedure was used to insure that participants formed vivid episodic memories for all associations rather than vague impressions based on familiarity^4,55,56. Participants who required more than five training cycles to meet the criterion of 100% accuracy were excluded from the experiment. Note that we restricted the training phase within 3–5 cycles to avoid extensive training which may result in memories too strong to suppression (3,5).

TNT phase

In the TNT Phase, participants underwent fMRI with concurrent recording of SCRs while performing the TNT task for face-picture associations acquired on day 1 and day 2. Each trial started with presentation of a face for 4 s, and was followed an inter-trial interval with a fixation cross for 2–6 s (average duration=4 s). The ‘Think’ and ‘NoThink’ trials were pseudo-randomized across participants and interleaved by a fixation period. The instructional cues (i.e., green and red rectangles) indicated ‘Think’ and ‘NoThink’ trials respectively, which appeared simultaneously with the face presentation. The presentation of only faces in this phase ensured that participants manipulated associated memories of the target picture^4,57. When seeing an instructional cue ‘Think’ (green rectangle), participants were required to recall and think of the previously learned picture, and when seeing an instructional cue ‘NoThink’ (red rectangle), they were instructed to not let the associated picture enter consciousness. After the presentation of a face, a fixation was presented during an inter-trial interval which served as a low level baseline for the experimental trials⁴. The total duration of the task was 19.2 min with 144 trials in total with 36 trials in each condition. Participants were shown 36 out of 52 faces in total, with half of them acquired on day 1 around 24 h before the task (that is, overnight condition) and the other half acquired on day 2 about 30 min before the task (that is, newly acquired condition). Cues for baseline pairs were not presented in this phase. Half of these faces were randomly assigned to either Think (that is, ‘T) or NoThink (that is, ‘NT’) condition, resulting in four experimental conditions in a 2-by-2 full factorial design (that is, memory suppression: NoThink versus Think, acquisition time: 30 min versus 24 h) with 9 faces per condition. Each face repeated four times, resulted in 144 trials in total. The remaining 16 faces (half of them acquired 24 h before the task) were not included in the TNT task, served as a behavioural baseline. Before the fMRI scanning, participants were trained twice using 10 trials that were not used in the actual experiment. Participants were explicitly instructed to directly suppress unwanted aversive memories by attempting to exclude a memory from awareness, rather than occupying awareness with another competing thought (i.e., thought substitution)¹³.

Post-scan memory test

In the testing phase (Fig. 1d), memory performance for face-picture associations was assessed by a cued-recall task, in which all 52 faces (learned on day 1 and day 2) were included. Each trial started with a face as a cue, and participants were encouraged to recall details of the associated picture. On each trial, participants were given a maximum of 30 s to verbally describe the associated pictures in as much detail as possible. The description was only scored as correct (that is, remembered) if it included enough details for the specific scene to be uniquely identified⁵⁷. Incorrect or vague descriptions were treated as forgotten. Three raters who were blinded to the experiment reviewed participants’ answers independently. The final test scores were cross-validated by three raters. A final judgment on each item was only made when a consensus was reached among the three raters. If there was any disagreement, the three raters discussed them and made a collective decision. In the TNT phase, the baseline items were not present in TNT task and thus with less presentation times. This has been widely used to provide a baseline measure of memory performance that is not directly affected by ‘Think’ or ‘NoThink’ manipulation^3,4,11.

Behavioural data analysis

Memory accuracy was submitted to a 2-by-3 repeated-measures ANOVA with memory suppression (Memory: Baseline versus versus Think versus NoThink) and acquisition time (Time: 30 min versus 24 h) as within-subject factors. Similar to Levy and Anderson¹⁸, we also computed the suppression score for overnight (24 h) and newly acquired (30-min) memories by subtracting corresponding memory accuracy of ‘NoThink’ items from their respective ‘Baseline’ items to provide a measure of suppression efficiency. Individual participant’s suppression score was then Z-normalized for further brain-behavioural prediction analyses.