Article activity feed

  1. Evaluation Summary:

    This paper reports a methodologically rigorous investigation into the neural mechanisms supporting encoding and retrieval of specific and general information in the context of memory schemas for events, or "scripts." Its findings will be of general interest to neuroscientists and cognitive psychologists who work both with typical young adults (as studied int he present works) and in particular populations (e.g., development and/or aging; patients with brain damage). The work is particularly comprehensive in how it links both specific and general narrative representation at both encoding and retrieval with later memory behavior, which is a notable strength.

    (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. Reviewer #1 agreed to share their name with the authors.)

  2. Reviewer #1 (Public Review):

    A primary strength of the manuscript is that the authors use very innovative and sophisticated analyses to characterize neural representations of schemas in rich, naturalistic stimuli. Schemas for how events unfold (scripts) are a very fundamental kind of schema, but most schema studies to date simply cannot address these kinds of representations given the nature of the stimuli they use. Likewise, the current study seeks to directly capture schema information encoded in patterns of activity as opposed to detecting coarser univariate responses that differ according to schema information.

    The dissociation between anterior and posterior hippocampus (where schema representations in posterior hippocampus are negatively related to recall) is striking and very nicely relates to current theories about functional differences along the long axis of the hippocampus.

    One of the main points of emphasis in the manuscript is that different brain regions are implicated during encoding vs. recall in terms of schema representations that support behavior. While this is qualitatively true, this claim would be stronger if more direct statistical comparisons were applied.

    One limitation is that the behavioral measures were based on story-specific recall and not on schematic elements. Thus, as the authors acknowledge, it is hard to know whether mPFC (or other neural measures) might have been correlated with behavior if the behavioral metric was different. More generally, there is a slight mismatch in that the neural measures separate schema representations from event-specific representations, but the behavioral measures lack this distinction.

  3. Reviewer #2 (Public Review):

    Masís-Obando and colleagues describe a study investigating the neural basis of specific (story) and general (schema) representations of naturalistic narratives (movie/audio clips). Narratives were of one of two types (airport, restaurant) about which participants would likely have rich past experience and knowledge, which allowed the researchers to ask what features were shared among different narratives that depicted the same "script." The researchers characterized the degree to which neural patterns reflected unique, story-specific codes (there is correspondence across people at the particular narrative level) versus general, schema codes (airport patterns are more similar to one another than they are to restaurant patterns). They were moreover interested in understanding how these representations were leveraged at both encoding and retrieval separately to guide free recall of each particular narrative's events. The main hypotheses were surrounding the involvement of medial prefrontal cortex (mPFC) and hippocampus (HPC) in this process. mPFC overall represented both schema at story at encoding, but neither at retrieval; a follow-up analysis revealed different effects in anterior versus posterior mPFC clusters, with anterior showing a greater relationship (than posterior) between schema representation at encoding and behavior that was mediated by specific story reinstatement in posterior medial cortex (PMC). Consistent with ideas about differences in representation across hippocampal long axis, they also found anterior HPC showed schema effects, whereas story effects (at encoding only) were more prominent in posterior. Beyond their a priori regions of interest, the researchers also report widespread cortical involvement for many of these analyses. The main take-home appeared to be that these networks differed between encoding and retrieval.

    Overall, the findings are compelling and align with prior work, while also providing new insights in the context of a more naturalistic memory task. For example, lack of mPFC involvement (schema or story representation) during retrieval was unexpected, and may inform future work on this topic (e.g., through encouraging more fine-grained consideration of mPFC sub-divisions). I moreover appreciated the author's transparency about their hypotheses and clear acknowledgement of the relationship between this data set and an existing paper. The work appears to be carefully done, and the paper is generally clear and well-written. I do however have a few questions and suggestions for the authors, as follows:

    1. From a theoretical perspective, I am struggling with the behavioral outcome measures being exclusively at the "specific" story level, and whether/how that should impact our interpretation of the findings. In other words, the behavioral outcome of interest has to do with participants' ability to recall story-specific details, and a score was given to each subject for each story to summarize the quality of their memory for that particular narrative. By necessity, of course, this means knowledge at the "schematic" level is not tested or operationalized in any way. (In fact, it would I believe be impossible to do this on a narrative-by-narrative basis.) The authors address this in their setup, discussing how a schema can be used to guide the retrieval of details, and also touch upon this in the Discussion (lines 404-410). However, I am struggling with the contrast between the memory ~ encoding and memory ~ reinstatement findings being whopping and widespread for the story neural representation (Figure 3A, C), and much smaller (and nonsignificant in many ROIs) for the schema neural representation (Figure 3B, D). Is this showing us that (detailed, specific) story representation supports recall of (detailed, specific) memories, and (general, abstracted) schema representation does not? Does that mean schema representation does not relate to memory, or just that it doesn't relate to *specific* memory (i.e., but could have in theory been related to schematic memory, had that been tested)? I suppose from some vantage points, it could be viewed as merely a replication of many other findings that representing specific memories at either encoding or retrieval is helpful for recall of those details. And similarly, one could argue that schema representations haven't been given a fair shake because the behavior was tested at a different level of specificity. In other words, in their analysis for Figure 3 B and D the authors separately considered the relationship between schema representation and behavior, without simultaneously considering the level of specific story representation, which is a bit hard to reconcile with the framework that schemas would guide retrieval via reinstatement of specific details (i.e., theoretically, should we expect that they can support detail recall on their own? or should it be that schema representation supports specific memory, but only when detail recall is also high?). With the exception of the mediation analysis in Figure 5 (which I think does speak to this point in a nice way), the earlier, primary analyses do not take this complexity into account. To be clear, I am not sure answering these questions requires new analyses, and am not asking the authors to change their approach. I am more hoping the authors could provide us more of their thoughts on these points in the paper and perhaps soften their conclusions if appropriate.

    2. It was not clear to me how the audio vs. movie difference was worked into the analysis, or why for the schema scores, different-modality patterns were not also considered. It would seem as though comparing patterns derived from the presentation of movie vs. audio as part of the schema measure would allow the researchers to get around potential confounds like visual presentation of the same type of stimuli across narratives of the same type to drive the "schema" representation (e.g., restaurant movies presumably show a lot of the same types of objects as one another, but those same objects would not be presented visually in the audio clips). Similarly, perhaps audio clips contained similar words for a given schema. It seems as though airport 1 movie being more similar to airport 1-4 audio than it is to restaurant 1-4 audio (all different modality comparisons) would be a powerful way to demonstrate schema representation (I believe the authors have done this in past work; Baldassano et al. 2018 J Neuro). In any case, I think this detail and reasoning should be added to the main paper, and potentially worked into the visualizations.

    3. It was unclear to me from the methods how the models relating neural scores with behavioral performance were set up. It sounds as though perhaps the researchers ran a simple linear regression, such that all participants' data was combined into a single model but subjects were not treated as random effects. If this were the case, then variability in memory performance across subjects is going to contribute to the estimate of the within-subject relationship between neural scores and memory performance on a story-by-story basis. It seems from the paper as though the authors are more interested in the within-subject variability. Can the authors clarify this point (e.g., by expanding the methods section beginning on line 585)?