Article activity feed

  1. Evaluation Summary:

    The goal of this study was to test whether multiple attempts at suppressing the retrieval of an (emotional) memory is associated with degradation of the representation of information in the brain about such memories. A combination of sophisticated computational modelling in fMRI reveals that neural representations of previously suppressed memories are sustainably weakened during memory retrieval attempts. This manuscript is of interest for neuroscientists in the field of motivated forgetting and memory control.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. The reviewers remained anonymous to the authors.)

  2. Reviewer #1 (Public Review):

    Meyer and Benoit provide an elegant and timely approach to understanding the brain mechanisms that underlie suppression-induced forgetting. Using a modified version of the "Think/No-Think" paradigm, the authors explored memory suppression across four study phases: pre-test, suppression phase and post-test. Objects were presented with aversive scene backgrounds and participants studied object-scene associations to a criterion level of 60% accuracy. In the MRI scanner, participants were required to recollect ("think') or suppress ("no-think") the aversive scene associated with the presented object. Results indicate that memory suppression was associated with increased activation of the dorsolateral prefrontal cortex and deactivation of the parahippocampal cortex and hippocampus. Additional analyses indicated that the reduction in memory vividness due to memory suppression was associated with deactivation of the parahippocampal cortex.

    The brain mechanisms enabling us to selectively suppress unwanted memories remain unclear, particularly over longer durations. As such, this study addresses an important topic. There are a number of strengths to this study. By testing memory suppression in the scanner after the initial encoding phase, the authors could identify patterns of brain activation and deactivation when memories were successfully recalled versus successfully suppressed. Importantly, one third of the objects were not shown during the fMRI phase, providing a control baseline to assess how memories weaken in general due to the passage of time. Use of representational similarity analysis further permitted the authors to determine whether the neural activity pattern before and after the suppression phase was statistically different.

    Overall, these findings offer new insights into the processes by which the neural activation for suppressed memories is dampened and how such altered neural representations relate to markers of memory quality (e.g., vividness).

    One aspect that was not discussed is how increased activation of the dorsolateral prefrontal cortex relates to individual differences in cognitive control. It may be that some individuals are more adept at inhibiting unwanted memories and how such individual differences relate to mental imagery, emotion regulation and cognitive control would be important to consider.

  3. Reviewer #2 (Public Review):

    In this paper, the authors examine whether suppression of memory retrieval, as implemented with the Think/No-Think paradigm, results in a degradation of a memory trace. They utilize two multi-voxel techniques, pattern classification and representational similarity analysis (RSA), to examine the nature of the representation of memorial information (rather than just examining the degree of activity in a given brain region related to memory). Furthermore, their investigation determines whether the degradation of the memory representation, as indexed by the information derived from the multi-voxel techniques, is related to reported changes the vividness of the memory. This latter approach is laudable in that it attempts to link their neural measure of a memory's robustness to an index of behavior.

    The methods, approach and statistical analayses that the authors take seem quite sound. They use an object-aversive scene pairing for the cue-target pairing in the Think/No-Think task. They obtain vividness ratings of the target after individuals learn the pairings, and then after they have manipulate the target in their mind, either to Think about it, (i.e., retrieve it) or to Not Think about it ( i.e., suppress it). Finally, a task is given in which individuals see a variety of different aversive scenes as compared to morphed versions of the scenes, which allows them to create a classifier to determine the degree to which scene processing is affected by the Think and No-Think manipulations..

    The first set of findings are confirmatory, indicating that the typical regions that are involved in retrieval suppression becomes activated during the No-Think condition (e..g, lateral prefrontal cortex). They also find that people report memories that were suppressed are less vivid than either those that were retrieved or the control items that were not mentally manipulated. However, in a separate sample of individuals who did a behavior only version of the task, both the control and suppressed items were rated less vivid. Given the demand characteristics (i.e., individuals who were told to Not Think about items, might be inclined to rate them as less vivid to please/go along with the experimenter), these data are not the most convincing.

    The more interesting results come from the multi-voxel approaches. First, it was found the classifier fit for scenes increased with increased vividness rating, suggesting that demand characteristics are not driving the results. With regards to the classifier fits for scenes in general, it is found that the classifier fit is reduced both during active suppression (i.e., during a No-Think trial) and during the subsequent post-test than is observed for baseline trials. Moreover, those participants who showed a greater suppression of scene information during No-Think trials also showed a reduced report of vividness. Much of these results focus on the parahippocampal gyrus, which is known to process scenes. These findings support the idea that retrieval of scene information in general is impacted, and is consistent with prior work that retrieval suppression involves a general (rather than a more specific) inhibition of temporal lobe memory-related regions as to inhibit retrieval.

    Probably the most novel part of the study involves using representational similarity analysis to examine the degree to which a specific scene (rather than scenes more generally) is suppressed and the relationship of such suppression to the post-test vividness rating of the item. Here the results are marginal which makes an interpretation difficult. Had the effect been found it would have suggested an item-specific effect that has not been observed previously.

    Overall, this is a solid study with interesting results that expand our knowledge of the neural mechanisms that support the retrieval and suppression of memories.

  4. Reviewer #3 (Public Review):

    This fMRI study uses classification analysis and representational similarity analysis (RSA) to test the hypothesis that memory suppression sustainably deteriorates the neural representations of previously suppressed memories during their subsequent retrieval. In this study, 33 participants repeatedly suppressed memories of aversive scenes during a Think/No-Think (TNT) paradigm. Before and after the TNT phase, participants also had to recall the scenes and to rate the vividness of their recollection (pre- and post- tests). Classification analysis using the decoding toolbox (Hebart et al., 2015) was computed to distinguish between intact aversive scenes and their morphed versions. The weight pattern obtained was then used to quantify the degree of reactivation during pre-, post- test and suppression phase of the TNT. RSA was performed to obtain the similarity of each scene with itself (same-item) and with the average of the scene with the other scenes (different-item) of the same category (i.e., Think, No-think or Baseline). The difference score between same-item and different-item similarity enabled to assess whether the retrieval of a given scene is associated with similar neural activity pattern before and after suppression (i.e., reinstatement of its unique representation). Results evidenced that the process of memory suppression rendered the memories less vivid and diminished the reactivation of the scenes both globally across the brain and locally in the parahippocampal cortices. The decline in vividness was associated with weaker reinstatement of memory representations in the parahippocampal cortex. These results support the hypothesis that suppression sustainably reduces the neural reactivation of memory traces.

    The conclusions of this paper are well supported by the data.

    One of the strengths of this study is that it does not only examine whether there is less reactivation of memory traces during suppression attempts (as previously shown), but whether this effect maintains during the subsequent reactivation of the previously suppressed memories. This question is tested using innovative multivariate pattern analyses that are not traditionally used in the Think/No-Think literature and that may provide new promising approach for the analysis of such data. Moreover, RSA investigates the neural reinstatement of unique representations, rather than more global category of items (i.e., Think, No-Think and baseline categories).

    Although multivariate methods are promising tools in neuroimaging, the specific advantages of the methods used in this paper, in comparisons to more classical methods (e.g., difference in activation at pre- vs post- tests), are not sufficiently highlighted.

    The approach used in the manuscript do not allow to have a full picture of the neural network involved in the reduction of scene reactivation at the post-test. Authors focused a priori on the parahippocampal cortex for its involvement in memories for complex scenes and in mnemonic vividness. However, other regions such as the hippocampus or the amygdala also play an important role. Concerning the hippocampus, the authors speculate that the hippocampal representations may be largely protected from interference (discussion, page 11), suggesting a different disruption process of these representations in comparison to the parahippocampal cortex. Amygdala is crucial for the type of material used (i.e., aversive scenes from the IAPS) in this study. From a clinical perspective, decreasing the reactivation of the amygdala is important as it should subsequently decrease the negative emotional content of the memory trace. Despite the interest of the hippocampus and amygdala, these regions were not included in the analyses. Focusing on a single region or more globally across the whole grey matter (without distinguishing the involved brain regions) prevents to make a more integrated discussion on the neural mechanisms that may help to sustainably reduce memory reactivation.

    At the end of the experiment, participants had to recall the 48 object-scene pairs to investigate whether suppression during the TNT phase induced forgetting. However, results did not evidence significant suppression-induced forgetting at final recall (Appendix 3). Authors suggest that the additional retrieval tests before and after the TNT may have reduced the forgetting. Additionally, the classifier training task (aversive scenes used), presented before the final recall, may also have introduced interference, and dampened the differences between the conditions (Think, No-think, and baseline). Consequently, the authors cannot rely on this objective measure of forgetting but can only rely on the subjective measure of vividness. This limitation is not sufficiently discussed in the manuscript.

    Although the representational similarity analysis was promising to evidence the reinstatement of unique representations, it did not evidence a significant reinstatement for suppressed memories, and it was not weaker than for baseline memories (Figure 4b). This suggests a lack of power for such neural measure.