续接上篇

Results

Enduring physiological responses after consolidation

First,

we examined physiological responses associated with suppression of aversive memories before and after overnight consolidation. A 2-by-2 repeated-measures analysis of variance (ANOVA) on SCR data, with Suppression (NoThink versus Think) and Time (30 min versus 24 h, referred to as Recent versus Remote), revealed main effects of Suppression (F(1, 17)=5.79, P=0.028, ) and Time (F(1, 17)=6.42, P=0.021, ). Importantly, we observed a significant Suppression-by-Time interaction (F(1, 17)=7.08, P=0.016, ). Post hoc comparisons using two-tailed paired t-tests  revealed that suppression of newly acquired aversive memories led to significantly reduced SCR levels relative to Think condition (t(17)=−3.10, P=0.007, d_av=0.50). This reduction, however, was not found in the suppression of aversive memories following overnight consolidation (t(17)=0.14, P=0.89, power=0.05; Fig. 1e). To mitigate potential confounds due to a difference in general performance between newly acquired and overnight memories, we performed a subsequent SCR analysis by taking random sub-samplings from 30-min conditions to artificially match the number of trials in 24-h conditions. We replicated our results and found that suppression was indeed accompanied by reduced SCR in newly acquired memories (t(17)=−3.33, P=0.004, d_av=0.74), but not for overnight aversive memories (t(17)=0.02, P=0.97, power=0.05; Supplementary Fig. 1). Remarkably, a similar pattern of results was observed in another independent cohort of 25 participants (Supplementary Fig. 2a). To further investigate whether our observed effects were specific to suppression of aversive memories, we conducted an additional control experiment with face and neutral pictures as stimuli in another 30 participants (data from one participant was excluded due to no SCR recording, resulting in 29 participants in total) (Supplementary Fig. 3a–c). Repeated measures ANOVAs revealed neither the main effect of Suppression nor Time-by-Suppression interaction (all F(1, 28)<1.50, P>0.30, <0.05; Supplementary Fig. 3d), and a generally lower SCR level associated with neutral than aversive memories (mean±s.d, 0.15±0.07 versus 0.36±0.07). Altogether, these results indicate that the reduction in physiological responses to emotional memory reactivity is attenuated over a 24-h time period.

Less efficient memory suppression after consolidation

Next, we examined the effectiveness of memory suppression after overnight consolidation compared with newly acquired memories. Note that the type of effectiveness under consideration is referred to the mnemonic aftereffects of suppression. A 2-by-2 ANOVA on subsequent memory performance revealed main effects of Suppression (F(1, 17)=4.60, P=0.047, ) and Time (F(1, 17)=14.75, P<0.001, ; Fig. 1f). Interestingly, we also observed a significant Suppression-by-Time interaction (F(1, 17)=5.86, P=0.027, ) (Fig. 1g). Again, similar results were reproduced in another independent cohort of 25 participants (Supplementary Fig. 2b,c). To examine whether the observed effect was specific to emotional memories, we analysed data from our additional behavioural neutral control experiment. Similar to aversive memories, we found significant main effect of Time (F(1, 28)=62.15, P<0.001, ) and an interaction between Time and Suppression (F(1, 28)=4.89, P=0.035, ; Supplementary Fig. 3e,f). In addition, evidence from our behavioural control experiment with neutral stimuli showed similar effects on the suppression of neutral memories after overnight consolidation (Supplementary Fig. 3e–j). Altogether, these results indicate that overnight consolidation leads to less pronounced suppression-induced forgetting for both emotional and neutral memories.

Hippocampal–neocortical reorganization after consolidation

We then examined functional reorganization of the brain systems involved in retrieval and suppression of aversive memories before and after overnight consolidation. We restricted our analysis to trials that were later remembered and artificially matched the number of trials in the 30 min and 24 h conditions to mitigate potential confounds related to general differences in memory performance and time decay between these conditions. This analysis revealed higher activation in the bilateral hippocampus in the ‘Think’ condition of newly acquired memories relative to the overnight consolidation ones (Fig. 2a,b). The ‘Think’ condition of aversive memories after overnight consolidation, however, was associated with higher engagement in neocortical regions including lateral parietal cortex (LPC) and angular gyrus extending into posterior cingulate cortex (PCC), and middle temporal gyrus (MTG) (Fig. 2c,d; Supplementary Table 1).

We conducted parallel analyses for all trials irrespective of final recall status to examine general changes in brain systems involved in suppression of aversive memories following overnight consolidation compared with newly acquired condition. We identified a set of widely distributed brain regions previously reported^4,11, including the inferior frontal gyrus, DLPFC and posterior parietal cortex (Supplementary Figs 4,5, Supplementary Table 2). Further analysis of anatomically defined regions of interest (ROIs) revealed significantly higher engagement in the right (t(17)=2.32, P=0.033, d_av=0.37) but not the left (t(17)=1.13, P=0.27) DLPFC in suppression of aversive memories after overnight consolidation, compared with newly acquired condition (Fig. 2e,f).

To better understand neural mechanisms underlying the effects of overnight consolidation on emotional memory suppression, we contrasted suppression (that is, NoThink trials) with retrieval (that is, Think trials) of aversive memories in overnight consolidation versus the newly acquired condition to explore a Suppression-by-Time interaction effect. This analysis revealed significant clusters in the bilateral hippocampus and amygdala (Fig. 3a,b), and other brain regions (Supplementary Table 3).

Follow-up ROI analyses revealed that suppression relative to retrieval of newly acquired aversive memories led to significant reductions in activation of the bilateral hippocampus and amygdala (Fig. 3c,d; all t(17)>2.5, P<0.01, d_av>0.43).

These effects, however, dissipated after overnight consolidation after which no differences were found between Think and NoThink conditions in the bilateral hippocampus or amygdala (both t(17)<1.3, P>0.2). Further prediction analysis based on machine learning algorithms (see Methods section) revealed that hippocampal activity during the NoThink trials was negatively predictive of the suppression-induced forgetting score for newly acquired memories (r_{(predicted, observed)}=−0.55, P=0.028) but not for the overnight condition (r_{(predicted, observed)}=−0.21, P=0.65).

To further investigate neural systems underlying suppression-induced voluntary (intentional) forgetting (that is, NoThink trials later forgotten in post-scan test) and incidental forgetting (i.e., Think trials later forgotten in post-scan test) for overnight and newly acquired memories (Supplementary Fig. 6), we performed additional analysis only for NoThink and Think trials that were later forgotten in the 30 min and 24 h conditions (see Methods section for details). We artificially matched the number of trials between conditions and excluded four participants for their lack of forgotten trials. Again, we observed similar Suppression-by-Time interaction effects in the right DLPFC, the hippocampus and amygdala. Follow-up paired t-tests revealed higher activation in the right DLPFC for intentional relative to incidental forgetting for overnight memories (t(13)=3.36, P=0.01, d_av=0.43), but not for newly acquired memories (t(13)=−0.38, P=0.71; Supplementary Fig. 7a).

The differences in DLPFC engagement involved in intentional relative to incidental forgetting were higher for overnight memories than newly acquired ones (t(13)=3.02, P=0.016, d_av=0.41; Supplementary Fig. 7b). Interestingly, we observed the opposite pattern in the hippocampus and amygdala, with significant decreased activation during intentional forgetting, relative to incidental forgetting of newly acquired memories (both t(13)<–2.5, P<0.04, d_av>0.30), but not for overnight memories (both t(13)<1, P>0.35; Supplementary Fig. 7c,d).

Taken together, these results indicate higher DLPFC engagement and less concomitant hippocampal and amygdala disengagement in suppression of overnight-consolidated aversive memories, even when considering only intentional forgetting.

Distinct hippocampal–prefrontal pathways after consolidation

To investigate how the hippocampal memory and prefrontal inhibitory control systems functionally coordinate to carry out suppression of aversive memories after overnight consolidation, we conducted a psychophysiological interaction (PPI) analysis (see Methods section) to identify functional coupling of the hippocampus with every other voxel of the brain, with a particular focus on the prefrontal inhibitory systems (Fig. 4a). This analysis revealed significant interaction effects in the DLPFC, the inferior frontal gyrus, and other regions (Supplementary Table 4). Further analysis revealed significantly lower hippocampal functional coupling with these regions in the suppression of aversive memories after overnight consolidation compared with the newly acquired condition (all t(17)>2.87, P<0.012, d_av>0.45).

The opposite pattern of results was observed in suppression of newly acquired aversive memories. Notably, increased hippocampal functional coupling with the bilateral DLPFC was predictive of more effective suppression of newly acquired memories (r_{(predicted, observed)}=0.47, P=0.023; Fig. 4b,d).

In contrast, decreased hippocampal functional coupling with the left DLPFC was predictive of more effective suppression of overnight consolidated memories (r_{(predicted, observed)}=−0.70, P=0.001; Fig. 4c,e).

In addition, we also performed hippocampal-seeded functional connectivity analysis only for forgotten trials by taking memory status into account. This analysis revealed a similar pattern of Suppression-by-Time interaction in the DLPFC (Supplementary Fig. 8).

Altogether, these results indicate distinct hippocampal-prefrontal functional connectivity involved in suppression of overnight and newly acquired memories

Distinct representational patterns after consolidation

To further our understanding of how aversive memories become resistant to voluntary suppression after overnight consolidation, we investigated multivoxel activity patterns associated with individual aversive memories in newly acquired and overnight consolidation conditions. We implemented a multivariate pattern analysis that provides a measure of neural pattern dissimilarity, by examining inter-item correlational dissimilarity of multivoxel activity patterns within each condition. This analysis revealed a significant main effect of Time in the anatomically defined hippocampus (F(1, 17)=11.2, P=0.004, ; Fig. 5a), with lower multivoxel pattern dissimilarity (that is, greater pattern similarity) for aversive memories after overnight consolidation relative to the newly acquired condition for both recall (t(17)=2.54, P=0.021, d_av=0.43) and suppression (t(17)=2.42, P=0.03, d_av=0.42). Further analysis (see more details in the Supplementary Methods) revealed that the hippocampal representational dissimilarity was higher for intentional forgetting (NoThink trials later forgotten, NTf) versus incidental forgetting (Think trials later forgotten, Tf) of newly acquired aversive memories, but not for that of overnight aversive memories (Supplementary Fig. 9).

Using the searchlight algorithm²³, we also performed a whole-brain exploratory analysis to identify changes in neural pattern dissimilarity associated with individual aversive memories after overnight consolidation. This analysis revealed significant clusters in the left and right hippocampus (peak at (−33, −27, −12), and (33, −36, 0) in MNI coordinates; Supplementary Table 5) that showed more generalized multivoxel activation patterns for aversive memories after overnight consolidation (Fig. 5b,c).

Importantly, the hippocampal pattern dissimilarity was negatively predictive of the right DLPFC engagement in memory suppression for both newly acquired (r_{(predicted, observed)}=−0.74, P=0.006) and overnight (r_{(predicted, observed)}=−0.67, P=0.012) memories (Fig. 5d,e). There was no difference in the prediction slopes across newly acquired or overnight memories (P=0.89).

Critically, further analyses revealed that higher multivoxel pattern dissimilarity in the hippocampus was predictive of higher suppression-induced forgetting for newly acquired memories (r_{(predicted, observed)}=0.66, P=0.013; Fig. 6a,b), whereas higher pattern dissimilarity in the neocortex (that is, LPC) was predictive of more effective suppression of overnight consolidated memories (r_{(predicted, observed)}=0.69, P=0.002; Fig. 6c,d).

Altogether, converging results from both ROI and whole brain analyses indicate neural activity patterns associated with individual aversive memories become more generalized (ie, less separable) in the hippocampus after overnight consolidation. Specially, distinct fine-tuned neural activity patterns in the hippocampus and the neocortex are predictive of suppression-induced forgetting of newly acquired and consolidated aversive memories, respectively.