Parallel Wires: A Conserved Principle of Contralateral-Ipsilateral Segregation in the Visual Corpus Callosum

Reviewer #1 (Public review):

Summary:

This study combined high-field fMRI with computational modelling (including a Bayesian population receptive field [pRF] model and functional gradient analysis) in humans to demonstrate that the architecture of the corpus callosum (CC) and its interhemispheric connections is organized into parallel ipsilateral and contralateral streams, rather than functioning as a mixed integration of inputs from both hemispheres. The human findings were validated through preclinical experiments in mice using viral axonal tracing, which revealed a non-overlapping laminar arrangement of axons carrying left and right visual field information.

These results suggest that the CC operates as a set of parallel, segregated pathways, with each stream independently conveying information from one side of the visual field. This organization preserves the spatial origin of visual signals within the white matter. Although the overall concept of interhemispheric parallel pathways is not entirely unexpected, this refined understanding of callosal organization provides important scientific and clinical insights in relation to pathway-specific perturbations and in neurological disorders.

Strengths:

The manuscript is well written, the methodology is sound, and the analyses are carefully conducted. I particularly appreciate the effort to integrate functional and structural approaches and to validate the human neuroimaging findings with more sensitive preclinical techniques, such as viral tracing.

Weaknesses:

Several points require clarification to allow a more complete interpretation of the results. In addition, some further analyses are necessary to fully substantiate the claims made in the manuscript. These are detailed below

Comment 1:

BOLD signals in white matter remain a matter of debate, although this is not the central focus of the present study. Nevertheless, it is important to establish whether the underlying data have sufficient tSNR to support robust pRF estimation in white matter. In Figure 1, the EV appears relatively robust; however, it seems that only the best-fitting examples are shown. In contrast, the group-average EV reported in Figure 2, and the individual maps in the Supplementary Information indicate very low EV values, typically below 5%. In conventional fMRI analyses, thresholds of approximately 15-20% EV are often applied to exclude poor fits that may bias pRF parameter estimates. It appears that no such threshold was applied here. Interestingly, in Figure S6, the average EV for dual pRF models appears to be approximately 17%. Do dual and triple pRF models systematically produce higher EV compared to single pRF models? Additionally, Figure 2 suggests the presence of baseline activation that is captured by the model. Could this be related to a delayed or altered hemodynamic response function (HRF) in white matter? Clarification would be helpful. To better assess the robustness of the reported findings, the authors should provide quantitative measures of tSNR within the white matter tracts where the pRF model was fitted. Furthermore, a plot showing the average BOLD signal during visual stimulation versus baseline in those tracts would greatly strengthen confidence in the signal quality.

Although the reported linear relationship between pRF size and eccentricity, as well as the test-retest reliability analyses, suggest the presence of consistent receptive field estimates, these analyses are based on distributions and may lack the sensitivity required to differentiate single, dual, and triple pRF models. Moreover, the pRF estimates within the FMA appear noisy, particularly at the individual level (Figure S4), making it difficult to clearly dissociate information originating from the left and right hemifields.

Comment 2.1:

The Bayesian modelling approach is interesting and robust. However, as I understand it, the authors must specify a priori the number of pRFs to be estimated. This introduces a strong assumption about the expected underlying receptive field structure. An alternative Bayesian approach, such as micro-probing (Carvalho et al., 2020), does not require prior assumptions regarding the number or shape of pRFs. Instead, it estimates receptive field profiles in a more data-driven manner and provides a direct visualization of the pRF structure. Implementing such an approach, or at least comparing it with the current modelling strategy, could yield more reliable and potentially less biased estimates of multiple pRFs, particularly in white matter where signal quality is limited.

Comment 2.2:

Some clarifications regarding the pRF model are needed: in the Methods section, the authors mention the use of a Difference-of-Gaussians (DoG) model. However, it appears from the Results that the analyses were performed using a single-Gaussian model. Additionally, in Section 5.6, the authors state that six different pRF models were tested. Which specific models were included in this comparison? A clear description of each model, along with justification for the final model selection criteria, would help better understand the study

Comment 3:

Throughout the manuscript, the authors repeatedly refer to laminar-specific findings. However, the reported functional resolution of 1.6 mm isotropic is insufficient to reliably resolve cortical layers. Given this limitation, the laminar interpretations appear overstated. For example, in the Discussion section titled "Integrating White Matter with Laminar-Resolved Function", the authors state: "The combination of anatomically segregated white matter pathways with functionally specific cortical laminae presents a powerful synergy for human brain circuit research." Given the spatial resolution of the functional data, how are laminar-specific functional claims justified?

Similarly, the authors suggest that: "It becomes possible to assess not just if the CC is damaged, but precisely which directional pathways are compromised-either the pathways projecting from the lesioned hemisphere, or those projecting to the other, or both." It is unclear to me how the current methodology uniquely enables this level of directional specificity, and whether this was not already feasible using existing structural and diffusion-based approaches. The authors should clarify what is genuinely novel in this study.

Comment 4:

In the Discussion, the authors state: "These findings fundamentally reframe our understanding of interhemispheric communication, moving beyond static connectivity to reveal a dynamic, directionally specific highway where spatial location encodes the origin of information. This framework provides a novel blueprint for decoding directional information flow in the living human brain." Based on the analyses presented, it is unclear how the findings of this study demonstrate dynamic connectivity or true directional specificity. The reported results appear to characterize spatial organization and segregation of callosal pathways, but they do not measure the directionality of information flow, temporal dynamics, or causal directionality between hemispheres. To substantiate claims regarding dynamic or directional communication, additional analyses, such as connective field model (Haak et al.2013), effective connectivity modelling, time-resolved approaches, or perturbation-based methods (neuromodulation) would be required. As currently presented, the findings seem to support structural and functional segregation rather than dynamic or directionally resolved interhemispheric information transfer. The authors should either provide stronger evidence for these claims or moderate them.

Comment 5:

I agree with the authors that pooling of information across hemispheres represents a plausible explanation for the presence of dual pRFs. As discussed in the manuscript, such an effect would be expected to predominantly affect pRFs located near the vertical meridian. However, Figures S6C and S6D do not appear to demonstrate that bilateral pRFs are preferentially located along the vertical meridian.

Reviewer #2 (Public review):

Summary:

The manuscript proposes a "parallel wires" architecture for the visual corpus callosum, suggesting that contralateral and ipsilateral visual streams remain spatially segregated into distinct anatomical channels. The authors use a cross-species approach, combining Bayesian population receptive field (pRF) modeling in humans with dual-color viral tracing in mice. The analysis of the publicly available human fMRI dataset indicates a 92% probability of single-hemifield representation, arguing for functional segregation. The mouse mesoscale tracing data support the idea of anatomical parallel wires by displaying dorso-ventral segregation of callosal axons post-midline crossing.

Strengths:

The primary strength of this study is its cross-species integration. Observing that functional segregation in humans is mirrored by specific anatomical pathways in the mouse provides a convincing, multimodal argument for the "parallel wires" hypothesis. The data is generally well-presented, and the Bayesian modeling of the human data is a robust methodological choice.

Weaknesses:

There are weaknesses in the description, presentation, and methodological details of the mouse tracing data. First, the authors must provide detailed information regarding spectral unmixing, intensity normalization, and threshold-sensitivity analyses. These factors are critical as they directly influence the Dice and Jaccard overlap estimates that underpin the study's primary conclusions. Second, it is unclear which cortical layers have been virally labelled as there is no quantification of the spatial extent of the injection site, and there is ambiguity regarding the dorso-ventral stereotaxic coordinates.

Reviewer #3 (Public review):

Summary:

This manuscript describes a study into the functional organization of the forceps major (FMA). The authors present a Bayesian population receptive field (pRF) analysis of group-averaged HCP fMRI retinotopic mapping data, focusing on voxels within the FMA. This is unconventional because pRF modelling is usually limited to gray matter voxels, where synaptic activity underlying neural computation is the highest. Nevertheless, some previous work suggests that meaningful fMRI signals can also be gleaned from white matter voxels, where the signals are thought to reflect metabolic activity from action potentials that travel along axons. However, these signals are generally much noisier, and possible confounding effects due to partial voluming, draining veins, and different hemodynamics must be carefully ruled out. Based on the Bayesian pRF analysis, the authors claim evidence of segregated contralateral and ipsilateral representations of the visual field in the FMA. Anatomical tract tracing based on HCP diffusion MRI data from seeds identified using the pRF analysis further suggests that these representations are underpinned by separate fiber bundles, which also appear to be consistent with the results of viral tracing in mice. The results of this study could mean an important step forward in understanding transcallosal signaling.

Strengths:

The study treads uncharted territory, leveraging multiple data modalities across species and advanced analytical approaches.

Weaknesses:

The study does not address potential confounds related to BOLD imaging in white matter structures. If the fMRI results can be explained based on neighboring grey matter responses, the evidence that remains is limited to an apparent anatomical segregation of white matter bundles that appear to be present in both mice and humans.

Further details are also missing regarding the Bayesian pRF approach, including the priors used for the pRF model. These are important as they will dominate the estimates when the data are very noisy, and the authors have adopted unconventional, more complex pRF models compared with earlier work employing Bayesian pRF analyses.

It appears that the authors have not applied any statistical thresholding to ensure that only good-quality model fits are entered into subsequent analyses (i.e., the reported probabilities pertain to model comparisons, not goodness of fit). From Figure 2, it appears that the majority of the FMA voxels, barring those adjacent to visual gray matter, do not exhibit more than a few percentage points of explained variance (EV). In fact, a common threshold is >15% EV, but it looks like none of the FMA voxels exceed this threshold.