Cell-Type-Resolved Transcriptomic Signatures of Parkinson's Disease: Discovery in a Primary Cohort and Independent Cross-Cohort Validation

Parkinson's disease (PD) involves coordinated dysfunction across multiple central nervous system cell types, but the extent to which cell-type-specific molecular signatures replicate across independent single-nucleus RNA sequencing (snRNA-seq) cohorts remains incompletely characterized. Here we present a discovery-and-validation study using two independent, publicly available human postmortem snRNA-seq datasets. In a primary discovery cohort (GSE157783; Smajic et al., 2022; postmortem midbrain, 11 donors: 5 PD, 6 control), we performed cell-type-resolved differential expression (DE) analysis between PD and control nuclei across microglia, astrocytes, oligodendrocytes, and dopaminergic neurons, identifying 1,935, 1,844, 5,254, and 236 significantly differentially expressed genes, respectively (adjusted p < 0.05, |log2 fold-change| > 0.25). This analysis nominated a candidate panel of five cell-type-specific markers: NEAT1 (astrocytes, log2FC = 1.24), MOBP (oligodendrocytes, log2FC = -1.25), and PBX1 (log2FC = 1.44), EN2 (log2FC = 4.13), and LINGO1 (log2FC = -1.95) in dopaminergic neurons (all padj < 1e-7). Gene Ontology and gene set enrichment analysis implicated proteostasis and lysosomal biology in microglia, lipid/cholesterol metabolism in astrocytes, myelin and phospholipid metabolism in oligodendrocytes, and synaptic/axonal regulation in dopaminergic neurons; donor-level pseudobulk testing (DESeq2) within the primary cohort was underpowered at this donor number and did not independently corroborate these nucleus-level signals. We then tested this panel in an independent validation cohort (GSE184950; Wang et al., 2024; postmortem substantia nigra, 34 samples, ~84,000 nuclei), using a sex-corrected model (logistic regression with donor sex as a latent variable). Replication was partial: NEAT1 (log2FC = 1.49, padj = 7.85e-21) and MOBP (log2FC = -1.13, padj = 1.13e-120) replicated robustly in the same direction and remained highly significant. PBX1 and EN2 did not replicate in dopaminergic neurons (opposite direction, padj = 1.0 for both). LINGO1 could not be tested in its designated cell type in the validation cohort, falling below the minimum expression threshold. TH, examined separately, was significantly downregulated in the validation cohort (log2FC = -1.20, padj = 4.08e-14) but significantly upregulated in the primary cohort (log2FC = 3.19, padj = 8.1e-10), a genuine, quantitatively confirmed cross-cohort discordance that we discuss as most likely a small-sample effect in the 11-donor primary cohort. Together, these results support robust, replicated glial (astrocyte, oligodendrocyte) dysfunction signatures in PD, while underscoring that dopaminergic-neuron-specific candidate markers identified in a small discovery cohort do not generalize to an independent cohort, a finding we consider as scientifically important as full replication would have been. Analysis code and differential expression outputs: https://github.com/Noahfereydouni/pd-snrnaseq-discovery-validation Preprint. This manuscript has not been peer reviewed.