Integrative computational analysis of urinary PAH biomarkers and Parkinson's disease-related molecular signatures

Polycyclic aromatic hydrocarbons (PAHs) are widespread environmental contaminants with potential neurological effects, yet the molecular links between PAH-related internal exposure and Parkinson's disease (PD)-associated biology remain poorly characterized. We developed an integrative computational framework combining population biomonitoring, transcriptomics, machine-learning prioritization, single-nucleus transcriptomics, molecular docking, and molecular dynamics (MD) simulations. Five urinary PAH metabolites representing complementary biomarkers within the same PAH exposure domain were evaluated in 11,512 adults from NHANES 2005-2018 using complex survey-weighted models. Among 107 participants meeting a prescription-based, single-nucleus transcriptomics, molecular docking, and molecular dynamics (MD) simulations. Five urinary PAH metabolites representing complementary biomarkers within the same PAH exposure domain were evaluated in 11,512 adults from NHANES 2005-2018 using complex survey-weighted models. Among 107 participants meeting a prescription-based antiparkinson medication proxy, 1-hydroxynaphthalene was the only metabolite associated with the outcome after false-discovery-rate correction (OR per doubling, 1.16; 95% CI, 1.05-1.29; P = 0.005; q = 0.027), with similar estimates across sensitivity analyses. At the molecular level, the five-metabolite biomonitoring panel was retained to identify shared or convergent target features across PAH-related internal-exposure markers. Panel decomposition and leave-one-metabolite-out analyses showed that the prioritized candidates were not dependent on any single metabolite. Integration of 441 predicted PAH-associated targets with 624 PD-associated differentially expressed genes yielded 15 candidate genes. Integration of 441 predicted PAH-associated targets with 624 PD-associated differentially expressed genes yielded 15 candidates genes, from which full-data LASSO and SVM-RFE prioritized AKR1C1, CXCR2, GPRC5A, PGR, and PSPH as candidate features. Repeated nested cross-validation of the locked five-gene logistic model yielded an out-of-fold AUC of 0.731 (95% CI, 0.602-0.853), whereas resampling-based feature re-selection was unstable; independent substantia nigra datasets provided partial expression-level replication. Functional analyses implicated immune-inflammatory, neuroactive, metabolic, and cellular-stress pathways, while single-nucleus transcriptomics localized these genes across vascular, immune, neuronal, astrocytic, and oligodendroglial populations. Docking of 1-hydroxypyrene predicted interactions with all five prioritized proteins, and 100-ns MD simulations of AKR1C1, CXCR2, PGR, and PSPH characterized persistent ligand association and target-specific conformational behavior. Collectively, the study distinguishes a metabolite-specific population signal from a broader panel-level molecular landscape and identifies computationally prioritized PAH-related molecular features relevant to PD-associated biology.