Review Papers Claim by Claim

Concise take

This study of 120 HCP subjects reports a strong relationship between cortical low frequency power and white matter BOLD projections (r = 0.811, p < 0.0001) and shows network specific spatial patterns and a positive correlation with cortical myelin (r = 0.7199) — evidence that resting cortical activity modulates white matter BOLD in a network and myelin dependent mannerPrimary paper: main results

Cortical modulation of resting state BOLD signals in white matter — detailed review and critique

Summary of main claims

• The authors report that across 75 cortical ROIs (derived from an atlas) the subject-averaged cortical fALFF predicted the mean white matter projection with correlation r = 0.8110, p < 0.0001Cortex WM correlation
• Each of 13 cortical functional networks produced a distinct white matter spatial projection map; basic sensory/motor networks had larger white matter projections than higher order networksNetwork differences
• White matter projection maps resemble diffusion tractography structural connectivity (Dice > 0.81 for tested seeds), and cortical myelin content correlates with projection magnitude (r = 0.7199, p < 0.0001)Structural similarity and myelin correlation

Methods and controls (what they did well)

Key strengths: the authors used HCP resting fMRI with fast TR and long runs (TR = 720 ms, Nd = 1200) and a conservative white matter mask plus erosion to reduce gray matter contamination; they computed fALFF using Welch periodograms, used time delay correction (1–10 TR) moving from superficial to deep white matter, applied physiological regressors including cardiac, respiratory and mean CSF, and included two negative controls: skull bone marrow and time-shuffled white matter signalsMethods and controls

Interpretation and mechanistic plausibility

The authors interpret the results as evidence that spontaneous cortical activity (low frequency BOLD fluctuations) modulates measurable low frequency fluctuations in white matter BOLD, in patterns consistent with structural connectivity and cortical myeloarchitecture — a plausible conclusion supported by (a) topographic match to tractography, (b) myelin correlation, and (c) negative controls that reduce chances of trivial global or nonneural confoundsMain interpretation

Context in the literature

• Independent lines of work report measurable functional signals in white matter (electrophysiology, transcriptomics, diffusion-fMRI analyses), supporting that WM BOLD can carry neural informationWM functional gradients studyADC-fMRI preprint
• Computational and intracranial recordings show very-slow cortical fluctuations can be amplified and modulated by network state; such dynamics are a plausible neural origin for low frequency BOLD patterns observed hereCriticality model and iEEG

Principal strengths

1. Large, high quality HCP dataset with fast sampling and long runs that improve low frequency estimation and temporal resolutionHCP acquisition details
2. Multiple, sensible negative controls (time shuffle; skull bone marrow) that argue against trivial systemic or scanner-only correlationsControls used
3. Anatomical triangulation using diffusion tractography and myelin maps increases biological plausibility and reduces likelihood of purely vascular artifactsAnatomical triangulation

Key limitations, blindspots and alternative explanations

Despite strengths, several important limitations reduce the certainty of a definitive neural interpretation:

• Residual gray matter signal contamination remains possible. Although the authors eroded superficial WM and used a tight WM mask and smoothing of only 2 mm, limited spatial resolution at 2 mm isotropic and imperfect registration can leak cortical signal into adjacent WM voxels; the authors acknowledge this and demonstrate the correlation persists in a 57 voxel deep WM mask but note lower SNRDeep WM mask result
• Vascular and draining vein effects cannot be fully excluded by the present analyses. The authors argue separate venous and arterial systems for GM and WM reduce draining confounds and they regressed CSF rather than global signal, but residual global or physiological artifacts (slow systemic oscillations, spatially structured physiology) may partially contributeGlobal signal discussion
• Cross sectional use of a single dataset and a specific parcellation means findings may depend on atlas choice, thresholds, and preproc steps; reproducibility requires code and open processed maps (authors make processed data available on request but have not released full code/data at the time of acceptance)Data availability statement
• Temporal causality is not established: correlations and inner products at zero lag (with time delay correction) are indicative of association but do not prove that cortical neural events generate WM BOLD; electrophysiological or interventional work is required for stronger causal claimsTime delay correction and zero lag

Practical reproducibility assessment

Reproducibility is moderate: the authors report detailed preprocessing and parameters (TR, voxel size, bandpass ranges, smoothing, mask threshold, tractography parameters), and used publicly available HCP raw data; however, processed maps and analysis code are only available on request rather than as a public repository, limiting immediate reproducibility. Reproducibility score estimate: 6/10Reproducibility notes

What would disprove the main claim?

Strong, falsifying outcomes would include (1) replication on independent datasets with identical processing producing null cortex–WM projection correlations, (2) demonstration that nonneural tissues or synthetic noise produce similar network-specific projection maps, or (3) invasive electrophysiology showing WM BOLD fluctuations are not temporally coupled to neural spiking or LFP changes in corresponding cortical networks — any of which would reduce confidence in a neural origin for WM BOLD fluctuationsAuthors on what would falsify

Suggested next experiments (concrete)

1. Simultaneous intracranial electrophysiology and fast fMRI targeting cortical regions and deep WM regions to test time-locked neural-WM BOLD coupling and latency structure; compare with superficial WM and deep WM to rule out partial volume effectsiEEG model support
2. Repeat analyses on alternative contrast modalities such as ADC-fMRI or multi-echo BOLD with careful vascular denoising to test whether WM functional patterns persist when vascular sensitivity is reduced;cite comparative ADC-fMRI evidence that ADC can be more WM sensitiveADC-fMRI comparison
3. Open release of analysis pipelines, test-retest assessment, and cross-atlas replication (vary parcellation and WM mask thresholds) to quantify robustness to preprocessing choicesReplication and openness suggestion

Useful links and next actions