Open reviews
July 17, 2021 — Open review of preprint by Liu and colleagues: PubPeer
Liu, L., Li, H., Ren, Z., Zhou, Q., Zhang, Y., Lu, C., Qiu, J., Chen, H., & Ding, G. (2021). The “two-brain” approach reveals the active role of task-deactivated default mode network in speech comprehension. bioRxiv. DOI
Liu and colleagues provide evidence demonstrating that the putatively “task-negative” posterior DMN (pDMN) plays an active role in narrative speech comprehension. The authors rally a handful of related fMRI results in support of this claim. First, two “sanity checks” replicate the task-negative profile pDMN: (a) the ICA-based pDMN defined here is shown to be anticorrelated with executive control (ECN) and language (LN) networks; (b) pDMN also displays a task-negative contrast in both story-listening vs. a brief fixation baseline and in an independent (visual) task; next, two speaker-listener ISC analyses suggest a positive role for pDMN in speech comprehension: (c) speaker-listener lagged ISC in pDMN is correlated with behavioral comprehension scores; and (d) the degree of anticorrelation between pDMN and ECN was correlated with speaker-listener ISC in pDMN. The study includes a handful of control analyses (e.g. control story stimulus, control ICA network, etc), as well as a dynamic causal modeling (DCM) analysis positioning anterior DMN (aDMN) between the ECN and LN networks and pDMN. This study replicates and expands on the relationship between speaker-listener ISC and comprehension (focusing in particular on pDMN) with a considerably larger sample size (N ≈ 60) than early studies on this topic (e.g. Stephens et al., 2010; N ≈ 11—yikes). My main complaint in reading this manuscript is that it’s fairly difficult for the reader to synthesize all these results as currently written. I have a few questions/suggestions (that hopefully don’t inflate the already wide array of analyses) as well as some wording suggestions or clarity.
Major comments:
While reading, I found it difficult to synthesize each of the results presented in this study into a coherent story about the DMN’s active role in narrative comprehension. The results sometimes read as a catalogue of findings. Any effort the authors could make to streamline the interpretation and guide the reader through the results would be much appreciated! For example, when introducing a new Results section, it might be worth spending an additional sentence providing (more) conceptual motivation for why this analysis is necessary (i.e. what outstanding question/concern does it address?) and linking it to the preceding analyses. The interpretation felt a little convoluted despite attempts to pull everything together in the Discussion.
In one of the initial analyses, the authors contrast a brief (20-second) fixation baseline with roughly 10 minutes of story listening to show that pDMN is deactivated during story-listening relative to baseline whereas LN is activated during story-listening relative to baseline. This doesn’t seem like a particularly meaningful analysis, given the short duration of the baseline (and much longer and more variable activity during story-listening). Authors use a separate task to validate pDMN deactivation, and also demonstrate that pDMN is anticorrelated with LN and ECN. These latter results provide a more convincing “sanity check” for the classical “task-negative” character of DMN, and the “story-listening vs. 20-second baseline” analysis could probably be deemphasized.
Related to the previous comment, you describe contrasting baseline with story-listening as “[(task-baseline) / baseline]” where “task and baseline denote the mean of time points within the task phase and baseline phase.” I’m curious why you divide by mean baseline activity in this example rather than simply subtracting baseline from task (maybe there’s a precedent for this that I’m not aware of).
I’m curious whether the anticorrelations observed between e.g. pDMN and LN in Figure 2B would hold up in an intersubject functional connectivity analysis (among listeners; ISFC; Simony et al., 2016; Nastase et al., 2019). Within-subject functional connectivity is notoriously susceptible to head motion and other confounds. On the other hand, ISFC analysis effectively filters out subject-specific idiosyncratic fluctuations and yields stimulus-driven connectivities. I don’t think this analysis is strictly necessary or that it would meaningfully change the story, but it might even strengthen the relative network relationships observed with within-subject connectivity.
The observation that greater anticorrelation between pDMN and ECN was correlated with increased speaker-listener coupling in pDMN was difficult for me to interpret. Does the degree of pDMN and ECN anticorrelation also correlate with comprehension?
The argument that “neural coupling between a speaker-listener pair is likely less susceptible to the influences of task difficulty, since speech production is typically more cognitively demanding than speech comprehension” was not very convincing to me. Even if speech is generally more cognitively demanding than listening, as you admit (line 526), there may be parts of the spoken narrative that are concurrently more or less demanding for both the speaker and listener—which would undermine this argument. On the other hand, I don’t find the original argument that DMN ISC may reflect fluctuations in generic “task difficulty” during language processing very compelling either. Real-world language probably does naturally vary in cognitive demand; and this varying cognitive demand may properly be a part of language comprehension. Using experimental manipulations to try to rule out this putative “confound” will by necessity introduce an artificiality to the linguistic paradigm that limits generalizability to real-world language.
Some of the terminology is confusing here. For example, in the Introduction paragraph starting at line 108, the text seems to contrast intersubject correlation (ISC) with “coupling”—but coupling here is defined as “the correlation of brain activities over time,” which is the typical definition of ISC. ISC computed between listeners, or between speaker and listener, is still ISC. I understand the authors want to use “coupling” to refer specifically to the listener-speaker correlation, but this should be made explicit at each usage.
Another bit of terminology I found hard to follow: the “successful” and “failed” terminology for communication conditions with the Chinese and Mongolian speakers should be introduced early on when the stimuli/design are described. This also made terms like the “change (successful > failed) of pDMN-ECN functional connectivity” (line 467) hard to parse at first.
Minor comments:
Line 57: “despite they have mastered” > “despite having mastered”
Line 124: “The group Independent…” > “Group Independent…”
Line 161: “To make the listeners attend” > “To encourage listeners to attend”
Line 207: “successful and failed communication conditions”—make sure to introduce these condition labels early on with reference to the Chinese and Mongolian speakers
Line 213: I’m not very familiar with the pipeline for group ICA and dual regression, so maybe this is my ignorance—but the description here wasn’t entirely clear. Are the individually-defined PCs aligned prior to concatenating data across subjects? Or is this not necessary? Is the “group dimension reduction” mentioned at line 215 referring to ICA, or a second group-level PCA step? At line 219, maybe specify that 20 components might be appropriate for naturalistic data (as opposed to e.g. resting-state data).
Figure 1: the orange text doesn’t seem particularly well-aligned here
Line 310: “lied” > “lie”
Figure 2: panel B may benefit from a correlation matrix to depict network pairs (although no error bars in that case)
Line 385: “lagged by that in the speaker’s” > “lagged relative to the speaker’s”
Lines 422 and 425: why was the 6-second lag chosen here? (presumably the peak lag)
548: “and not” > “and was not”
References: A couple references jumped out to me as missing, but take with a grain of salt—my reading list is biased and many of these are papers published by my current lab (don’t feel pressured to cite). For example, when discussing brain-to-brain coupling as a system or transmitting linguistic and social information between brains, Hasson et al., 2012, came to mind. When describing the DMN as the interface between high-level semantic content and internal content, a recent review by Yeshurun et al. (2020) came to mind. When describing the DMN’s role in narrative comprehension, Baldassano et al., 2017, Chen et al., 2017, and Zadbood et al., 2017, came to mind (although these studies rely on intersubject pattern similarity rather than time-series ISC).
I encourage the authors to share the code associated with this paper (e.g. via GitHub). This seems like a very rich and useful dataset, so I hope ultimately the authors will share the data as well!
References:
Baldassano, C., Chen, J., Zadbood, A., Pillow, J. W., Hasson, U., & Norman, K. A. (2017). Discovering event structure in continuous narrative perception and memory. Neuron, 95(3), 709–721. DOI
Chen, J., Leong, Y. C., Honey, C. J., Yong, C. H., Norman, K. A., & Hasson, U. (2017). Shared memories reveal shared structure in neural activity across individuals. Nature Neuroscience, 20(1), 115–125. DOI
Hasson, U., Ghazanfar, A. A., Galantucci, B., Garrod, S., & Keysers, C. (2012). Brain-to-brain coupling: a mechanism for creating and sharing a social world. Trends in Cognitive Sciences, 16(2), 114–121. DOI
Nastase, S. A., Gazzola, V., Hasson, U., & Keysers, C. (2019). Measuring shared responses across subjects using intersubject correlation. Social Cognitive and Affective Neuroscience, 14(6), 667–685. DOI
Yeshurun, Y., Nguyen, M., & Hasson, U. (2021). The default mode network: where the idiosyncratic self meets the shared social world. Nature Reviews Neuroscience, 22(3), 181–192. DOI
Zadbood, A., Chen, J., Leong, Y. C., Norman, K. A., & Hasson, U. (2017). How we transmit memories to other brains: constructing shared neural representations via communication. Cerebral Cortex, 27(10), 4988–5000. DOI