Pathology-defined cell states reveal reproducible transcriptomic signatures across ALS cortical single-nucleus RNA-seq studies
Amyotrophic lateral sclerosis (ALS) is a genetically and biologically heterogeneous neurodegenerative disease in which distinct pathogenic mechanisms operate across patients while overt molecular pathology is confined to only a subset of cells. Such features would act to dilute disease-associated transcriptomic signals and complicate the identification of reproducible molecular signatures across the growing number of ALS single-nucleus RNA sequencing (snRNA-seq) studies. Here, we systematically assessed cross-study reproducibility across four cortical ALS snRNA-seq datasets comprising 140 donors (87 ALS) and tested whether pathology-defined cell states improve detection of conserved molecular signatures. Cell-type annotations were harmonized prior to comparison of cell-type-specific pseudobulk differential expression using gene-level, pathway-level, gene-ranking and alternative polyadenylation analyses. We further examined nuclei exhibiting TDP-43 pathology, identified by expression of the STMN2 cryptic exon. Conventional ALS-versus-control analyses showed limited reproducibility, with minimal overlap of differentially expressed genes or enriched pathways, while fold-change patterns clustered predominantly by study rather than cell type or brain region. Nevertheless, gene-ranking analyses identified reproducible neuronal transcriptional programs, suggesting that biological signal is present but incompletely resolved by current cohort sizes. In contrast, STMN2 cryptic exon-positive nuclei showed substantially greater concordance, revealing robust TDP-43-associated signatures that partially overlapped independent models of TDP-43 dysfunction while also identifying motor cortex-specific changes, including reduced expression of the recently identified ALS risk gene UNC13C. Reproducible ALS-associated alternative polyadenylation changes were not detected, likely reflecting the higher dimensionality and sparsity of polyadenylation site analyses. Together, our findings demonstrate that pathology-defined cell states provide a more reproducible framework for studying ALS transcriptomic alterations than conventional case-control comparisons. We additionally provide an interactive browser to facilitate exploration and comparison of ALS snRNA-seq datasets.