Verbal versus Nonverbal Processing Leads to Generalized Hemispheric Laterality Effects that Span Multiple Networks
Curation statements for this article:-
Curated by eLife
eLife Assessment
This important study investigates how verbal and nonverbal working-memory processing is distributed across large-scale functional networks in the human brain using precision fMRI. By leveraging extensive within-subject data and individualized network mapping, the authors provide solid evidence that hemispheric specialization for verbal versus nonverbal information extends across multiple association networks and is reproducible across independent datasets. The use of state-of-the-art precision neuroimaging approaches reveals fine-grained laterality patterns that are likely obscured in conventional group analyses.
This article has been Reviewed by the following groups
Discuss this preprint
Start a discussion What are Sciety discussions?Listed in
- Evaluated articles (eLife)
Abstract
Precision neuroimaging was used to explore specialization for verbal versus nonverbal processing. Consistent with prior findings, individuals exhibited a spatially left-lateralized association language network, with regions in both hemispheres robustly responding to processing of meaning-based sentences. We next examined differential responses to verbal (words) versus nonverbal (faces) materials within the same working memory task. The right hemisphere components of the language network responded more strongly to nonverbal than to verbal materials, splitting the network’s functional profile between the hemispheres. Similar patterns were observed across multiple association networks including putative cognitive-control, action-mode, and attention networks. The hemispheric laterality effect was prospectively replicated in a second independent study. These findings highlight a generalized laterality phenomenon that transcends the specialization of individual networks that has been the recent focus of the human systems neuroscience field, and aligns with a broad mechanism that modulates processing between the hemispheres.
Article activity feed
-
eLife Assessment
This important study investigates how verbal and nonverbal working-memory processing is distributed across large-scale functional networks in the human brain using precision fMRI. By leveraging extensive within-subject data and individualized network mapping, the authors provide solid evidence that hemispheric specialization for verbal versus nonverbal information extends across multiple association networks and is reproducible across independent datasets. The use of state-of-the-art precision neuroimaging approaches reveals fine-grained laterality patterns that are likely obscured in conventional group analyses.
-
Reviewer #1 (Public review):
Summary:
In this manuscript, Sun et al. investigate the hemispheric lateralization of functional brain networks during verbal versus nonverbal working memory tasks. Utilizing state-of-the-art precision neuroimaging in highly sampled individuals, the authors define a set of distributed association networks and examine their task-evoked responses. The authors report a "generalized laterality effect," wherein multiple association networks appear to functionally split across the hemispheres, with left hemisphere components exhibiting a relative preference for verbal stimuli and right hemisphere components preferring nonverbal stimuli.
Strength:
The use of dense-sampling fMRI is a major strength of this study, allowing for a highly accurate, individual-specific mapping of network topologies that group-averaging …
Reviewer #1 (Public review):
Summary:
In this manuscript, Sun et al. investigate the hemispheric lateralization of functional brain networks during verbal versus nonverbal working memory tasks. Utilizing state-of-the-art precision neuroimaging in highly sampled individuals, the authors define a set of distributed association networks and examine their task-evoked responses. The authors report a "generalized laterality effect," wherein multiple association networks appear to functionally split across the hemispheres, with left hemisphere components exhibiting a relative preference for verbal stimuli and right hemisphere components preferring nonverbal stimuli.
Strength:
The use of dense-sampling fMRI is a major strength of this study, allowing for a highly accurate, individual-specific mapping of network topologies that group-averaging typically obscures. Despite the interpretational concerns raised below, this high-quality, within-subject imaging dataset represents a valuable resource for the community. Furthermore, the inclusion of an independent prospective replication dataset provides valuable confidence in the robustness of the core imaging metrics. However, while the data quality is exceptionally high, the conceptual interpretations regarding "network splitting", "preferential recruitment", and the generalizability of the verbal/nonverbal dichotomy require significant refinement. Several methodological and statistical clarifications are needed to fully support the authors' claims.
Weaknesses:
Major:
(1) The manuscript relies heavily on task contrast values (e.g., Face > Word) to conclude that networks functionally "split" their profiles, with specific hemispheres being "preferentially recruited" by either verbal or nonverbal materials. While the data clearly demonstrate relative hemispheric differences, claiming absolute bidirectional specialization and active recruitment appears to overstate the findings in two key ways:
First, statistical evidence for true bidirectional "splitting" is scarce. A significant hemispheric difference confirms a relative shift in processing, but it does not permit claims about absolute preference. When examining the face>word effects against zero in the discovery dataset (Figure 3), the right hemisphere of the LANG, FPN-B, CG-OP, and SAL networks shows no significant preference for faces over words. In the replication dataset (Figure 6), two of the four targeted networks (FPN-A, CG-OP) similarly fail to show a significant right-hemisphere preference. Furthermore, it is unclear whether these tests against zero were corrected for multiple comparisons (e.g., 18 individual tests in Figure 3). Networks reporting significance at the uncorrected p < 0.05 level (such as the left hemisphere of LANG and FPN-B) might not survive standard correction, suggesting that even the left hemisphere's preference for words may be statistically marginal. Second, contrast differences in networks exhibiting negative signals may reflect relative deactivation rather than active recruitment. A mathematically positive contrast value derived from two negative activation states (e.g., Face [-6] > Word [-8]) does not indicate active "recruitment" for face processing. Instead, it merely reflects a relative difference in deactivation. Characterizing this dynamic as "preferential recruitment" misleads the reader regarding the actual physiological state of the network. Furthermore, such asymmetric suppression is frequently driven by generalized differences in task difficulty or cognitive effort, rather than true stimulus-specific processing.
(2) Building on the previous point, it would be highly beneficial to include the behavioral data (such as accuracy and reaction times) for the N-Back conditions, which do not currently appear to be reported in the manuscript. This information is important because if one condition (e.g., the Face N-back) was significantly more challenging or required greater cognitive effort than the other (e.g., the Word N-back), the observed hemispheric dissociations might reflect differences in arousal, effort, or attentional deployment rather than stimulus-specific processing. ion
(3) The manuscript claims a "generalized" hemispheric laterality effect. However, the supplemental figures suggest this effect may be highly sensitive to the specific stimuli used in the main text (unfamiliar Faces vs. rhyming Words), which represent extreme ends of visuospatial and phonological processing. As we talked about earlier, a true functional "split" implies that the hemispheres respond in opposite directions. However, visual inspection of the supplemental graphs reveals that for the vast majority of networks, both hemispheres are actually driven in the exact same direction (Figures S8 and S9).
In the Face > Letter contrast (Figure S8), true bidirectional splits largely disappear. With the exception of FPN-A, the bars for both the left and right hemispheres point in the exact same visual direction for every network (e.g., both hemispheres are visually positive in DN-A and dATN-B, and visually negative in CG-OP and dATN-A). This indicates that the hemispheres actually share the same categorical preference and merely differ in magnitude. Strikingly, the LANG network shows no significant difference between Faces and Letters in either hemisphere, suggesting that the robust leftward shift observed in Figure 3 was possibly driven by the heavy semantic and phonological demands of the rhyming task, rather than a generic preference for "verbal" processing.
Similarly, in the Scene > Word contrast (Figure S9), the hemispheres do not visually diverge in their response direction for most networks. For example, both hemispheres are visually negative (indicating a shared preference for Words) in the LANG, FPN-B, CG-OP, and SAL networks. Conversely, both hemispheres are visually positive (indicating a shared preference for Scenes) in the dATN-B and DN-A network.
Because the hemispheres do not visually diverge in their response direction for most networks across these supplementary contrasts, the claim of a robust, generalized hemispheric "split" is unsupported.
-
Reviewer #2 (Public review):
Summary:
In this manuscript, Sun and colleagues use precision fMRI to investigate the spatial patterns of verbal versus nonverbal processing in the cortex. They first show that the left hemisphere tends to be more active for word than face processing, whereas the right hemisphere tends to be more active for face than word processing during a working memory task. They then show that this is not confined to a particular network specialized for verbal vs. nonverbal processing but is a pan-network hemispheric difference across 8/9 association networks. This was unexpected, as prior precision imaging work emphasized network specialization and the authors had hypothesized that verbal vs. nonverbal processing would show bilateral network-level effects in a single, right-lateralized network (FPN-B). They then …
Reviewer #2 (Public review):
Summary:
In this manuscript, Sun and colleagues use precision fMRI to investigate the spatial patterns of verbal versus nonverbal processing in the cortex. They first show that the left hemisphere tends to be more active for word than face processing, whereas the right hemisphere tends to be more active for face than word processing during a working memory task. They then show that this is not confined to a particular network specialized for verbal vs. nonverbal processing but is a pan-network hemispheric difference across 8/9 association networks. This was unexpected, as prior precision imaging work emphasized network specialization and the authors had hypothesized that verbal vs. nonverbal processing would show bilateral network-level effects in a single, right-lateralized network (FPN-B). They then replicated this pan-network effect in an independent dataset.
Strengths:
(1) This paper is neat, clear, and succinct. The authors convincingly show that there exists a hemispheric difference in face vs. word processing during a working memory paradigm across many association networks.
(2) They replicate this result in a fully independent dataset. They do a particularly nice job setting up a prospective replication analysis over 4 distinct association networks.
(3) They do a wonderful job framing the experiment - they provide context on how previous precision imaging findings would have suggested a specialized network for verbal vs. nonverbal processing, they replicate major findings from that work in this data (network laterality and bilateral network activation in multiple task contexts), and then show the surprising pan-network hemispheric laterality in two independent datasets.
(4) This paper adds a unique perspective to precision imaging findings. Most precision imaging has shown that different networks seem to be specialized for different cognitive tasks - individual task activations tend to follow network boundaries and networks tend to be cohesively activated. These types of findings led to the idea that previously reported broad hemispheric effects for verbal vs. nonverbal processing could instead be due to a lateralized network specialized for verbal versus nonverbal processing. However, this paper accounts for individualized network topography and yet finds a hemispheric effect that transcends individual networks. I think this result will have a strong influence on how many readers think about network specialization.
Weaknesses:
(1) The evidence for the main result comes only from a single fMRI task (n-back) with a particular set of stimuli (faces, words) in which processing demands are not matched across verbal and nonverbal conditions (word blocks use rhyming, face blocks use an exact match). While the claim made is fairly broad (hemispheric laterality in verbal vs. nonverbal processing), it isn't clear how well this finding would generalize to other types of stimuli or processing. This was mentioned in the limitations section, but the paper would be strengthened by both additional evidence for a general verbal versus nonverbal processing effect and by a discussion of how these specific stimuli (words and faces) and processing demands (differences between rhyming and exact match) might affect the results. The authors did include supplemental post-hoc analyses of scene and letter conditions that are matched in processing, but did not discuss them in the main text.
(2) While lateralization is seen in 8/9 networks, the effect is much larger in some networks (FP-A, DATN-A) versus others. Similarly, the face > word effect is much stronger in some regions of the cortex than others. For example, all the individuals shown exhibit a patch of RH mid-LPFC cortex with particularly strong face > word preference compared to anywhere else in association cortex and an analogous LH mid-LPFC region that shows a particularly strong word > face preference. While the authors are correct that lateralization is seen across many networks, there does seem to be a topography to it rather than a diffuse hemispheric difference. I wonder to what degree the general hemispheric laterality pattern could be driven by a subset of regions that happen to cross several association networks. Neither the differences between networks nor the finer-scale activation patterns are considered in this paper. The paper would be strengthened by considering them.
(3) The paper is focused on association networks and provides an interesting report on face versus word processing in those networks specifically. In the brain maps, it is possible to see face versus word preference in non-association regions as well. The authors don't discuss these patterns, but the paper would be strengthened by describing the relationship of these patterns with known face and language processing systems in the sensory cortex.
-
Reviewer #3 (Public review):
Summary:
This work takes a precision neuroscience approach to examining how different task demands elicit lateralized vs. bilateral activity in large-scale brain networks. Using ~35-150+ minutes of resting state data per person, the authors identify individualized networks and map their degree of laterality. They find that all association networks are bilateral, with some networks. such as the language network (left) and fronto-parietal B network (right), showing some marked degree of laterality. A sentence reading task evokes activity bilaterally in the language networks, while working memory load during an n-back task evokes bilateral activity largely in fronto-parietal, dorsal attention, and salience networks. Interestingly, though, the authors find that when digging into the N-back task and contrasting …
Reviewer #3 (Public review):
Summary:
This work takes a precision neuroscience approach to examining how different task demands elicit lateralized vs. bilateral activity in large-scale brain networks. Using ~35-150+ minutes of resting state data per person, the authors identify individualized networks and map their degree of laterality. They find that all association networks are bilateral, with some networks. such as the language network (left) and fronto-parietal B network (right), showing some marked degree of laterality. A sentence reading task evokes activity bilaterally in the language networks, while working memory load during an n-back task evokes bilateral activity largely in fronto-parietal, dorsal attention, and salience networks. Interestingly, though, the authors find that when digging into the N-back task and contrasting blocks that have rhyming words vs. faces, this elicits a more lateralized pattern of activity; left activation for rhyming and right activation for faces. This lateralization is seen across multiple association networks, not just the language or fronto-parietal networks. These findings are then replicated in another precision data set.
Strengths:
This work has several notable strengths. This study boasts a lot of data for each individual, allowing them to examine individualized functional networks and task activations. Given the marked individual differences in laterality (see Figures S4 and S5), a group-averaged network atlas and group-averaged activation maps would likely muddle some of the interesting effects.
The authors also use an elegant approach for individualized network estimation, multisession hierarchical Bayesian modeling, or MSHBM. MSHBM allows vertex-based functional connectivity to steer individualized networks while also incorporating meaningful priors to identify comparable networks across individuals. This is a nice approach combining elements of a common atlas across individuals with data-driven approaches.
The authors have a rich working memory N-back task with four different stimuli types with which they can examine processing different types of stimuli.
A major strength of this work is the replication. The authors take their surprising findings and replicate the entire experiment in another sample. The replication sample notably has even more data per person than the discovery sample, allowing for robust and precise estimation of individual networks and task activity.
Weaknesses:
I'd like to frame this section as 'unsolved challenges' rather than 'weaknesses'.
One of the strengths of this paper is that the working memory task that incorporates so many different stimuli and conditions also poses a caveat for interpreting the task stimuli effects. The 'word' condition requires multiple cognitive demands - working memory and rhyming/phonological processing. This makes it a little hard to draw complete parallels between the word and face conditions. It does, however, provide a useful backdrop to explain why there are such strong left laterality effects for the word condition. I would expect that explicit phonological processing that is required for a rhyming task would elicit more left lateralized activation than the sentence processing task, as fluent readers are likely not using explicit phonological processing for sentence processing. This could explain why, for the sentence reading task, they find more bilateral activation, but a task that specifically targets phonological processing (i.e., the 2-back word condition) would evoke more left lateralized activation above and beyond the activation supporting working memory, which is held constant across both the 'words' and 'faces' conditions. Ultimately, I believe this supports the broad idea of this work - that large-scale association networks can show lateralization, even beyond language network boundaries. However, the fact that there is a dual-task load that evokes phonological processing for the word condition is important for contextualizing these findings.
When looking at plots of individual differences, it is clear that some people are more bilateral than others in certain networks (and sometimes overall). Because of the amount of data per person in this precision study, it naturally limits the overall sample size (N=29). This makes it difficult to interrogate individual differences in laterality and what might predict this effect across people.
-
