Causal role of the angular gyrus in insight-driven memory reconfiguration

Reviewer #1 (Public Review):

Summary:
In the present manuscript, the authors present the results of a well-designed, thoughtful, and well-motivated study, targeting the role of angular gyrus in insight-based memory gains. The study is well conducted, timely, and presents clear-cut behavioral results. However, the analysis of the EEG-data lacks clarity and leaves many open questions - especially with regard to the representational similarity analyses. (Nevertheless, analogous concerns with regard to the focus on the three-way interaction and the comparison of linked vs. non-linked events pertain similarly to the connectivity analyses.)

Strengths:
- Well-conducted study with a proper sham-controlled TMS design.
- Clever insight-based memory task.
- Interesting behavioral findings.

Weaknesses:
- "We then calculated Pearson's correlations to compare the power patterns across theta frequency between the time points of linked events (A with B), as well as between the time points of non-linked events (A with X) for the pre- and the post-phase separately, separately for stories linked via imagination and via observation." (p.34)

The RSA basically asks on the lowest level, whether neural activation patterns (as measured by EEG) are more similar between linked events compared to non-linked events. At least this is the first question that should be asked. However, on page 11 the authors state: "We examined insight-induced effects on neural representations for linked events [...]". Hence, the critical analysis reported in the manuscript fully ignores the non-linked events and their neural activation patterns. However, the non-linked events are a critical control. If the reported effects do not differ between linked and non-linked events, there is no way to claim that the effects are due to experimental manipulation - neither imagination nor observation. Hence, instead of immediately reporting on group differences (sham vs. control) in a two-way interaction (pre vs. post X imagination vs. observation), the authors should check (and report) first, whether the critical experimental manipulation had any effect on the similarity of neural activation patterns in the first place.

Overall, the focus on the targeted three-way interaction is poorly motivated. Also, a functional interpretation is largely missing.

- "Interestingly, we observed a different pattern of insight-related representational pattern changes for non-linked events."

It is not sufficient to demonstrate that a given effect is present in one condition (linked events) but not the other (non-linked events). To claim that there are actually different patterns, the authors would need to compare the critical conditions directly (Nieuwenhuis et al., 2011).

- "This analysis yielded a negative cluster (p = 0.032, ci-range = 0.00, SD = 0.00) in the parieto-temporal region (electrodes: T7, Tp7, P7; Fig. 3B)." (p. 11)

The authors report results with specificity for certain topographical locations. However, this is in stark contrast to the fact that the authors derived time X time RSA maps.

"These theta power values were then combined to create representational feature vectors, which consisted of the power values for four frequencies (4-7 Hz) × 41 time points (0-2 seconds) × 64 electrodes. We then calculated Pearson's correlations to compare the power patterns across theta frequency between the time points of linked events (A with B), as well as between the time points of non-linked events (A with X) for the pre- and the post-phase separately, separately for stories linked via imagination and via observation. To ensure unbiased results, we took precautions not to correlate the same combination of stories twice, which prevented potential inflation of the data. To facilitate statistical comparisons, we applied a Fisher z-transform to the Pearson's rho values at each time point. This yielded a global measure of similarity on each electrode site. We, thus, obtained time × time similarity maps for the linked events (A and B) and the non-linked events (A and X) in the pre- and post-phases, separately for the insight gained through imagination and observation." (p. 34+35)

If RSA values were calculated at each time point and electrode, the Pearson correlations would have been computed effectively between four samples only, which is by far not enough to derive reliable estimates (Schönbrodt & Perugini, 2013). The problem is aggravated by the fact that due to the time and frequency smoothing inherent in the time-frequency decomposition of the EEG data, nearby power values across neighboring theta frequencies are highly similar to start with. (e.g., Schönauer et al., 2017; Sommer et al., 2022)

Alternative approaches would be to run the correlations across time for each electrode (resulting in the elimination of the time dimension) or to run the correlations at each time point across electrodes (resulting in the elimination of topographic specificity).

At least, the authors should show raw RSA maps for linked and non-linked events in the pre- and post-phases separately for the insight gained through imagination and observation in each group, to allow for assessing the suitability of the input data (in the supplements?) before progressing to reporting the results of three-way interactions.

References:
Nieuwenhuis, S., Forstmann, B. U., & Wagenmakers, E.-J. (2011). Erroneous analyses of interactions in neuroscience: A problem of significance. Nature Neuroscience, 14(9), 1105-1107. https://doi.org/10.1038/nn.2886
Schönauer, M., Alizadeh, S., Jamalabadi, H., Abraham, A., Pawlizki, A., & Gais, S. (2017). Decoding material-specific memory reprocessing during sleep in humans. Nature Communications, 8(1), 15404. https://doi.org/10.1038/ncomms15404
Schönbrodt, F. D., & Perugini, M. (2013). At what sample size do correlations stabilize? Journal of Research in Personality, 47(5), 609-612. https://doi.org/10.1016/j.jrp.2013.05.009
Sommer, V. R., Mount, L., Weigelt, S., Werkle-Bergner, M., & Sander, M. C. (2022). Spectral pattern similarity analysis: Tutorial and application in developmental cognitive neuroscience. Developmental Cognitive Neuroscience, 54, 101071. https://doi.org/10.1016/j.dcn.2022.101071

Reviewer #2 (Public Review):

The formation of long-term memory representations requires the continuous updating of ongoing representations. Various studies have shown that the left angular gyrus (AG) may support this cognitive operation. However, this study demonstrates that this brain region plays a causal role in the formation of long-term memory representations, affecting both the neural and behavioural measures of information binding.

A significant strength of this work is that it is the first one to test the hypothesis that the left angular gyrus has a causal role in the reconfiguration and binding of long-term memory representations by comparing when insights are primarily derived from direct observation versus imagination. Consequently, the results from this manuscript have the potential to be informative for all areas of cognitive research, including basic perception, language cognition, and memory.

Furthermore, this study presents a comprehensive set of measurements on the same individuals, encompassing various task-related behavioural measures, EEG data, and questionnaire responses.

There are, however, some weaknesses. One of them pertains to the link between the observed results and the conclusions. While the observed memory reconfiguration/changes are attributed to the angular gyrus in this study, it remains unclear whether these effects are solely a result of the AG's role in reconfiguration processes or to what extent the hippocampus might also mediate these memory effects (e.g., Tambini et al., 2018; Hermiller et al., 2019).

Another weakness in this manuscript is the use of different groups of participants for the key TMS intervention, along with underspecified or incomplete hypotheses/predictions. Furthermore, in some instances, the types of analyses used do not appear to be suitable for addressing the questions posed by the current study, and there is limited explanation provided for the choice of analyses and questionnaires.

Reviewer #3 (Public Review):

Summary:
Grob and colleagues investigated the causal role of the angular gyrus in insight-driven memory reconfiguration. Participants watched unrelated movie scenes while EEG was recorded prior to receiving either active or sham continuous theta burst stimulation (cTBS) over the left angular gyrus. Following stimulation, participants either observed or imagined links or non-links between scenes watched before stimulation. Next, participants rated their comprehension of the links. Following this part, participants completed questionnaires for 30 minutes, followed by a free recall test of details from the videos. Subjects then watched the videos again while EEG was recorded and engaged in a recognition test to determine whether they retained information about the linking events. Participants showed strong evidence of insight-driven linking between videos. The results indicate that overall memory of video details was stronger for the Sham group compared to the cTBS group, but only for the linked videos. An RSA analysis using pre- and post-video observation indicated that similarity increased for imagined and linked videos for the sham group, but not for the cTBS group, in sensors in parieto-temporal regions. Similarity for imagined, non-linked videos increased for the cTBS group, but not for the sham group, in frontal sensors. Coherence between fronto-parietal sensors decreased during the viewing of videos linked by imagination for the cTBS group, but not the sham group. Coherence between the same sensors increased while watching videos that were linked by observation in the cTBS group, but decreased for the sham group. The authors conclude that the angular gyrus is causally related to memory-insight reconfiguration.

Strengths:
The paper is nicely written, and the rigor of the experimental design is strong. The paper is pre-registered, and the authors used a double-blind sham-controlled design to eliminate the possibility of bias and non-specific effects of rTMS on their results. The behavioral results are striking and provide strong evidence that their intervention significantly decreased memory for details of linked events. The authors also took care to leave time between stimulation and recall to reduce the influence of carry-over rTMS effects on memory. There are also strong behaviorally-relevant neural changes.

Weaknesses:
My major criticism relates to the main claim of the paper regarding causality between the angular gyrus and the authors' behavior of interest. Specifically, I am not convinced by the evidence that the effects of stimulation noted in the paper are attributable specifically to the angular gyrus, and not other regions/networks.