White Matter Stratification in Depression Predicts Multidimensional Antidepressant Responses

Reviewer #1 (Public review):

Summary:

This work stratifies depression subgroups based on white matter integrity (Fractional Anisotropy, FA) and evaluates the relationship between white matter (WM) alterations in these subgroups and clinical symptoms. Furthermore, the authors tested these subgroup findings in an independent cohort. This paper provides WM-based depression subtypes that are linked to the clinical symptom profile (anxiety, cognitive, hopelessness, sleep, and psychomotor retardation) and presents the prediction of treatment outcome using these subtypes.

Strengths:

Applying a novel NMF (Non-negative Matrix Factorization) biclustering approach to stratify depression subtypes using white matter integrity. Following the recent functional MRI-based depression subtype stratification, this work provides a structural signature for depression heterogeneity. These subtypes were also tested in an independent cohort, with findings regarding clinical symptom profiles.

Weaknesses:

Although this novel method successfully subgroups depression patients, it is difficult to understand the spatial patterns of WM alteration and which structural connections, such as DMN, SN, ECN, and Limbic, because the findings are distributed across multiple WM bundles in each subgroup. Furthermore, these subtypes fail to predict optimal treatment selection within each group, since all subgroups benefit from different treatments.

Reviewer #2 (Public review):

Summary:

The authors measure the directional consistency of water diffusion in white matter (functional anisotropy: FA) to stratify depression subtypes across young adults. These findings are significant in that they highlight white matter as an underappreciated aspect of neural heterogeneity in major depressive disorder. While the evidence for meaningful, lower-dimensional structure in depression heterogeneity within their Nanjing cohorts is strong, claims that their subtypes are characterized by specific clinical symptom profiles and reflect neuroplasticity reserve are not supported by the same strength of evidence.

Strengths:

Circumscribing analyses to a simple white matter measure, across a sparse skeleton, with explicit sparsity-promoting algorithms yielded heterogeneity subdivisions that are much more interpretable than most depression heterogeneity clustering papers. Replication of their 3-cluster solution in an external dataset bolsters confidence in the existence of these 3 clusters, although generalizability to more diverse populations remains untested. The authors also tested a wide variety of treatment outcomes, which is difficult data to aggregate but ultimately critical for validating the utility of depression subtypes.

Weaknesses:

sCCA and SVR results were less interpretable. In part, this is due to core features of these methods (broad distribution of weights, instability across iterations). However, these inherent components of sCCA and SVR opacity were exacerbated by the opacity surrounding several analytic choices made by the authors and intermediate results associated with them. Without more transparency, it's unclear how these results extend the neuroclinical differentiation established (or not established) by their original NMF analyses.

To be more specific, a central claim of the paper is that their biotypes are "pathophysiologically distinct" and demonstrate "symptom-specific neurobiological substrates". However, only 3/18 pairwise symptom differences generalize across both datasets (Figures 1 and 2), implying that these biotypes have more symptom overlap than distinction. Brain-based distinctions are real and replicable, but because their NMF approach specifically optimizes for separating clusters on the basis of brain features, this is more of a methodological validation than a scientific finding. While several brain-symptom relationships reported later using sCCA and SVR are interesting, it is not currently possible to evaluate the robustness of these relationships and whether or not these relationships are nested within NMF-derived clusters or exist regardless of subtype.

To be clear, the heterogeneity problem in depression is extremely difficult to solve and beyond the scope of this manuscript. Despite the scale of this problem, the authors do report tangible progress in this aim, largely through finding an interpretable set of white matter features distinguishing patient clusters. These findings may lead researchers to meaningfully incorporate white matter features into heterogeneity analyses more in the future. However, many of the claims made are not fully supported, particularly surrounding clinical specificity and neuroplasticity reserve.

Author response:

We sincerely appreciate the constructive comments and valuable suggestions from the editors sand reviewers. We highly value the feedback and will carefully address all concerns in our revised manuscript.

(1) We will supplement more details of the processing steps and key results in the analyses of sCCA and SVR to improve the transparency and reproducibility of our methods.

(2) According to the reviewers’ suggestions, we will adjust and present a more conventional and cautious conclusion regarding clinical specificity and neuroplasticity reserve.

(3) We will supplement the results of structural connections (termed “symptom-related network” in the manuscript) across the three subgroups to strengthen the interpretation of subgroup-specific neurobiological characteristics.

(4) All the suggestions from the reviews will be respected, and we will carefully revise our manuscript to improve its clarity, rigor, and scientific quality.

We believe these revisions will significantly improve the quality of our work.