Multimodal convergence is needed before inferring neuroaxonal injury in migraine

Abstract

Gallardo and colleagues reported higher serum neurofilament light (sNf-L), greater susceptibility-weighted imaging signal in the left rostral anterior cingulate cortex, and lower fractional anisotropy in the right cuneus in patients with high-frequency migraine. Their prospective phenotyping and multimodal approach are valuable. However, these modality-specific findings do not necessarily indicate a shared neuroaxonal-injury process. Serum Nf-L is anatomically non-specific, susceptibility-weighted signal is an indirect marker of iron-related changes, and fractional anisotropy reflects several aspects of tissue architecture. In addition, the imaging findings were focal, cross-modal associations were not reported, and the classification model had a small held-out test set with a wide confidence interval. We therefore suggest describing the results as independent candidate markers associated with high-frequency migraine. Evidence of multimodal convergence would require replicated cross-modal relationships or related longitudinal changes. This distinction preserves the value of the study while avoiding a mechanistic interpretation that extends beyond the available data.

Clinical trial number Not applicable.

Main text

To the Editor: Gallardo and colleagues prospectively compared serum and MRI measures in 68 adults with high-frequency migraine and 66 healthy controls [1]. The integration of electronic diaries, serum assays, susceptibility-weighted imaging (SWI), diffusion MRI, and six-month treatment follow-up is ambitious and clinically relevant. Higher serum neurofilament light (sNf-L), greater left rostral anterior cingulate SWI signal, and lower right cuneus fractional anisotropy (FA) survived modality-wise correction. We suggest that these findings are best regarded as complementary candidate markers, rather than convergent evidence of one neuroaxonal-injury process.

The distinction matters because each measure answers a different question. Serum Nf-L is sensitive to neuroaxonal injury across many neurological conditions, but it is not anatomically specific and is influenced by age, renal function, body composition, and other systemic factors [2]. SWI signal is affected by susceptibility sources, T2* effects, acquisition, and vendor reconstruction; unlike quantitative susceptibility mapping, it does not provide a direct quantitative estimate of iron [3]. FA is a summary of diffusion-tensor anisotropy and can change with fibre crossing, orientation dispersion, packing, extracellular water, and registration, rather than axonal loss alone [4]. A significant group association in each modality therefore does not establish that all three reflect the same biological lesion.

Spatial and cross-modal coherence would strengthen that claim. The serum marker has no location, whereas the imaging effects occurred in different regions and represented one surviving SWI region of interest and one surviving FA region of interest. The article did not report whether individual sNf-L, rostral anterior cingulate SWI signal, and cuneus FA values correlated after adjustment for shared covariates. Without such relationships, a model in which independently selected variables contribute to case–control discrimination demonstrates complementary statistical information, not a common mechanism. The absence of group differences in serum tau, glial fibrillary acidic protein, and ubiquitin carboxy-terminal hydrolase L1 also argues for caution before invoking a broader neurodegenerative process.

Prediction requires similar restraint. In the MRI subcohort of 102 participants, a 25% held-out test set contains only about 25 individuals. The reported accuracy of 0.739 had a 95% confidence interval of 0.516–0.898, showing substantial uncertainty. Repeated cross-validation is useful for exploratory internal assessment, but selection of candidate variables based on group associations should be repeated inside every resampling loop to avoid information leakage. External validation is needed before the variables are described as discriminative biomarkers, especially because depressive symptoms were the strongest predictor and the imaging measures may vary with scanner and processing choices.

A practical hierarchy could clarify future multimodal studies. First, report each modality as an independently associated candidate marker with its measurement limitations. Second, test pre-specified cross-modal associations or a latent-variable model after controlling for age, renal function, migraine state, and mood. Third, demonstrate temporal convergence: related within-person changes across modalities with chronification, remission, or treatment. Fourth, show incremental clinical value beyond readily available clinical variables in an external cohort. This hierarchy distinguishes co-occurrence, statistical complementarity, biological convergence, and clinical utility.

The authors appropriately use the modifier “possible” in several places and acknowledge the non-specificity of SWI and FA [1]. Extending that caution to the central label would make the message even clearer. “Independent serum and MRI markers associated with high-frequency migraine” accurately describes the current data. “Multimodal neuroaxonal injury” should be reserved for a replicated pattern in which the modalities demonstrably converge on the same process. Such precision is important in migraine, where evidence of altered brain biology should not be inadvertently read as proof of progressive neurodegeneration.

Abbreviations

FA

Fractional anisotropy

MRI

Magnetic resonance imaging

sNf-L

serum neurofilament light

SWI

Susceptibility-weighted imaging

Author contributions

YZ: Conceptualization, Writing – original draft, Writing – review & editing. JJ: Conceptualization, Supervision, Writing – review & editing. Both authors read and approved the final manuscript.

Funding

This work received no funding.

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

Not applicable. This Matters Arising article discusses previously published literature and reports no new research involving human participants, human data, human tissue, or animals.

Consent for publication

Not applicable.

Authors’ information

Not applicable.

Competing interests

The authors declare no competing interests.