Functional Gradients of the Neuraxis: Reorganization from Cortex to Spinal Cord

Technical Summary: Functional Gradients of the Neuraxis

Problem Statement
While functional connectivity gradients have become a powerful tool for summarizing high-dimensional brain organization into low-dimensional manifolds, this approach has remained largely cortex-centric. Although recent studies have extended gradient analyses to the cerebellum, brainstem, and hippocampus, the spinal cord—a critical hub for sensorimotor integration and the primary conduit for ascending and descending pathways—has been absent from this unified, neuraxis-wide framework. Furthermore, while resting-state corticospinal fMRI has demonstrated somatotopic correspondence between the cortex and spinal cord, the joint topography of these structures within a shared low-dimensional space has not been described. The central question addressed is how neuraxis-wide principles of functional organization are expressed beyond the cortex and how bidirectional coupling between the sensorimotor cortex (SMC) and the cervical spinal cord shapes their intrinsic gradient manifolds.

Methodology
The study utilized simultaneous resting-state fMRI of the whole brain and cervical spinal cord (C3–C8) in 20 healthy adults. Data were acquired on a 3T Siemens Prisma system using a single field-of-view multiband EPI sequence.

• Parcellation:
  • Cortex: 62 bilateral sensorimotor cortex (SMC) regions were defined using the Schaefer atlas.
  • Spinal Cord: Two parcellation schemes based on the PAM50 atlas were tested: a fine-grained 33-region scheme and a coarser 8-region scheme (collapsing regions into dorsal horn, ventral horn, intermediate zone, and white matter tracts).
• Connectivity Matrices: Functional connectivity (FC) matrices were constructed for:
  1. Intrinsic SMC (62 × 62).
  2. Intrinsic Spinal Cord (48 × 48 for the 8-region scheme).
  3. Corticospinal (62 × 110 and 48 × 110 matrices combining intra-structure and cross-structure connections).
• Gradient Analysis: Diffusion map embedding (via the BrainSpace toolbox) was applied to the FC matrices to derive low-dimensional gradients.
  • SMC matrices were sparsified (top 10% connections retained).
  • Corticospinal matrices were left unsparsified due to inherently low-amplitude cross-level connectivity.
  • Procrustes alignment was used to align subject-specific gradients to a common reference.
• Analysis:
  • Clustering: Spectral clustering (K=4) was applied to SMC gradients to assess recovery of somatotopic classes (Lower Limb, Upper Limb, Trunk, Face).
  • Reorganization: Cluster dispersion (mean radius in gradient space) was compared between "SMC-only" and "Corticospinal" conditions to quantify manifold reorganization.
  • Geometry: Canonical Correlation Analysis (CCA) was used to test the alignment between the functional gradient manifold and a one-dimensional anatomical distance metric along the sensorimotor strip.
  • Spinal Stability: Fisher's J separation scores and gray matter (GM) compactness scores were used to evaluate the stability of spinal gradients across subjects and parcellation scales.

Key Results

1. Intrinsic SMC Organization: The first two intrinsic SMC gradients recapitulated canonical somatotopic organization. G1 varied along the mediolateral axis (segregating lower limb from upper limb/trunk/face), while G2 provided a complementary reweighting. Spectral clustering successfully identified four distinct functional clusters corresponding to body parts without prior labeling.
2. Corticospinal Reorganization of Cortex: When spinal connections were included in the SMC matrix, the SMC gradient manifold underwent significant reorganization. Functional clusters (Face, Lower Limb, Upper Limb) expanded and became more differentiated in gradient space (increased cluster radii), indicating that spinal input introduces within-cluster variability. However, the overall manifold remained aligned with cortical geometry, as confirmed by CCA, though the variance explained by the anatomical distance metric decreased slightly.
3. Spinal Gradient Stability and Parcellation: Intrinsic spinal gradients were highly sensitive to parcellation scale. The fine-grained 33-region scheme showed clear tissue-type separation at the single-subject level but exhibited high inter-subject variability at the group level. The coarser 8-region scheme yielded significantly more stable and reproducible gradients across subjects, with clearer separation of gray/white matter and ascending/descending pathways.
4. Asymmetric Coupling: Introducing SMC connections into the spinal matrix resulted in only modest changes to the spinal gradient manifold. Centroid positions shifted minimally, and cluster dispersion remained statistically unchanged. This contrasts sharply with the pronounced reorganization observed in the cortex when spinal connections were added.
5. Anatomical Consistency: Both cortical and spinal manifolds maintained orderly, anatomy-consistent axes. The spinal gradients exhibited orderly separation of gray and white matter and of ascending/descending pathways when the geometry respected major anatomical compartments.

Significance and Claims
The paper claims to incorporate the spinal cord into the gradient-based toolkit for mesoscale brain mapping, extending this approach beyond the cortex. The primary contribution is the demonstration that corticospinal organization can be described not as isolated regions but as coupled manifolds spanning the neuraxis.

Key claims include:

• Asymmetric Embedding: The study reveals an asymmetric embedding of the cortex and cord at rest. Spinal input broadens and differentiates cortical gradient axes, whereas cortical input preserves spinal gradient geometry. This suggests that while the spinal cord's intrinsic organization is robust and governed by local segmental architecture, cortical gradients are more sensitive to cross-anatomy connectivity.
• Unified Framework: The work provides an integrated framework where sensorimotor cortex and cervical spinal cord are placed within a common corticospinal manifold, validating the use of gradient mapping for structures beyond the brain.
• Methodological Insight: The findings highlight that parcellation resolution is a critical design decision for spinal gradient analysis, with coarser, anatomy-informed schemes providing greater group-level stability than fine-grained tract-wise parcellations.

The authors remain modest regarding the interpretation of these findings, noting that the observed asymmetry may be influenced by state-dependence (resting state vs. task), signal-to-noise limitations in spinal fMRI, and specific preprocessing choices. They posit that while the results suggest a dominant role for local spinal architecture in shaping spinal gradients, stronger effects on spinal geometry might be revealed under task-based paradigms or with higher-SNR acquisitions.