Multi-omics profiling identifies neuroinflammation-related genes and exosomal miRNA as robust diagnostic signatures for Parkinson’s disease

Neuroinflammation is a pivotal driver that amplifies the pathogenic cascade within the Parkinsonian brain. Nevertheless, the pathogenic drivers connecting neuroinflammation to PD pathogenesis remain unclear. To elucidate their diagnostic and therapeutic implications, this research sought to identify key neuroinflammation-related genes (NIRGs) and exosomal miRNAs in PD.

To comprehensively identify neuroinflammation-related genes (NIRGs) in Parkinson’s disease (PD), we conducted an integrated multi-omics analysis. Publicly available transcriptomic data encompassing microarray (GSE75249, GSE22491), high-throughput RNA-seq (GSE269775), and scRNA-seq (GSE223138) profiles were obtained from the GEO repository. We performed differential analysis to screen for significant transcriptional variations, encompassing both mRNA (DEGs) and miRNA (DE-miRNAs).

Functional enrichment analyses were conducted, encompassing pathway analysis via the Kyoto Encyclopedia of Genes and Genomes (KEGG), ontological annotation through Gene Ontology (GO), and pre-ranked gene set enrichment analysis (GSEA). Potential protein-level interactions were explored by constructing a protein-protein interaction (PPI) network with the STRING database. Based on the overlap between DEGs and NIRGs, a machine learning framework incorporating ten machine learning algorithms and their 101 combinations was constructed.

Subsequently, a quantitative nomogram was constructed for diagnosis in clinical practice. Additionally, the CellChat and Monocle packages were employed to investigate intercellular signaling and cellular differentiation trajectories, respectively. GeneMANIA, Friends analysis, regulatory network, immune infiltration, drug sensitivity, and molecular docking were also investigated.

Bulk RNA-seq data were examined, revealing 426 DEGs. Following intersection analysis and the application of a machine learning framework, we generated a diagnostic model utilizing the expression patterns of five signatures (PTGDS, RTN3, MAG, PROK2, and CNTNAP2). The five-gene signature achieved AUC values of 0.797–0.901 in the training cohort and 0.800–1.000 in the validation cohort, with corresponding sensitivity and specificity ranges of 0.500–0.786 and 0.769–1.000, respectively.

The robustness of the model was substantiated through cross-validation with internal and external datasets. The scRNA-seq data analysis revealed seven distinct cell clusters, with monocytes being identified as the predominant cell population. Pseudotime trajectory analysis further elucidated the developmental dynamics of the major monocyte lineage.