Multiscale Transcriptomic Architecture of the ALS Motor Cortex

Abstract

Amyotrophic lateral sclerosis (ALS) exhibits marked clinical and molecular heterogeneity, but how this heterogeneity is organized within disease-relevant central nervous system regions remains incompletely understood. Because pooled cross-region analyses may confound regional variation with subtype structure, we focused on the motor cortex, a key site of ALS pathology, to resolve its internal transcriptomic organization. Using approximately 600 postmortem motor cortex transcriptomes from the Target ALS resource, we applied gene set variation analysis to generate sample-level pathway activity profiles and performed consensus clustering in pathway space. We identified seven reproducible ALS motor cortex subtypes characterized by distinct inflammatory, damage-response, metabolic, glial, and neuronal/synaptic programs. Donor-held-out projection recovered the seven subtype assignments with a mean balanced accuracy of 0.939, supporting the reproducibility of the pathway-defined subtype structure in unseen donors. Mapping these bulk-defined programs to an independent motor-cortex single-nucleus RNA-seq dataset revealed distinct cellular contexts, while cell-type marker adjustment showed that cellular differences contributed to, but did not fully account for, subtype-associated pathway variation. The seven subtypes also recapitulated major molecular programs reported in previous ALS subtype frameworks while resolving them into finer motor-cortex-specific states. Rather than forming isolated molecular classes, these subtypes occupied a structured pathway-space landscape with both discrete subtype-centered peaks and continuous biological gradients. Principal component analysis and representative pathway overlays defined a major axis from an upper inflammatory/damage region to a lower neuronal/metabolic/remodeling region, with C7 forming a neuronal/synaptic connector within the lower landscape. Co-expression network analysis further supported this multiscale organization, revealing family-level network backbones, subtype-specific modular refinements, and bridge-like rewiring. Metadata-informed analyses showed that this molecular architecture was not primarily organized by disease duration. Instead, the landscape was more clearly associated with clinical and genetic-background variables, including ALS/FTD family history and mutation-related signals. In particular, family-history-positive and C9orf72 repeat expansion-positive samples were more represented in the lower neuronal/metabolic/remodeling region, whereas family-history-negative samples and C9orf72-negative cases were more represented in the upper inflammatory/damage region. These findings support a model in which ALS motor cortex heterogeneity is organized as a multiscale transcriptomic architecture more closely associated with genetic-background variables than with disease duration. This framework provides a basis for subtype-aware biological comparison, future patient-stratification, biomarker, or therapeutic target-prioritization studies.