Integration of cerebrospinal fluid proteomics and clinical scaled scores for subtyping sporadic Parkinson’s disease

Abstract

AbstractBackgroundMolecular pathways and networks associated with sporadic PD (sPD) subtypes are still unclear. We applied a systems biology approach to identifying proteomic signatures associated with sPD, cognitive and motor scaled scores, brain volumetric measurements, and cerebrospinal fluid (CSF) phosphorylated‐tau (p‐tau), total‐tau (t‐tau), alpha‐synuclein (a‐syn), and amyloid‐beta (a‐beta) levels.MethodsWe used CSF proteomic data of sPD from the Parkinson’s Progression Markers Initiative (PPMI) cohort (n = 377 sPD vs. 182 controls) and built an elastic‐net logistic regression model, adjusting for age, sex, and principal components, for sPD risk prediction using glmnet[1]. We performed correlation of baseline proteome with cognitive (MoCA) and motor assessments (UPDRS‐I, II, and III), neuroimaging data (Dopamine transporter scan; DaTscan of mean bilateral caudate, putamen, and striatum thickness (mm)), and CSF biomarkers. We implemented weighted gene co‐expression network analysis (WGCNA) to group co‐expression CSF proteins for sPD and to subtype sPD by CSF biomarkers using WGCNA[2] and ConsensusClusterPlus[3].ResultsWe found a CSF proteomic signature (59 proteins) differentially expressed in sPD compared to controls linked to neurotrophin, endocytosis, and chemokine signaling. There are 82 and 68 proteins differentially expressed in UPDRS‐II and UPDRS‐III scores. Three, 13, and 11 proteins were associated with caudate, putamen, and striatum volumes. We found proteins correlated with p‐tau (n = 1394), t‐tau (n = 1577), a‐syn (n = 1427), and a‐beta (n = 847) levels. We implemented a predictive model with 33 proteins for sPD (AUC = 0.83) that outperformed the models with UPDRS‐I (AUC = 0.71), MoCA (AUC = 0.67), p‐tau (AUC = 0.56), t‐tau (AUC = 0.61), a‐syn (AUC = 0.54), and a‐beta (AUC = 0.57) but underperformed UPDRS‐III (AUC = 0.99) or putamen (AUC = 0.99). WGCNA indicates CSF biomarkers are highly correlated to six protein modules for subtyping sPD: each module was enriched for different pathways, ME1 (SNARE and MAPK signaling), ME9 (autophagy), ME7 (lysosome and MAPK), ME6 (immune response), ME10 (proteostasis and phagocytosis), and ME2 (longevity regulating and mTOR). Those modules defined two subtypes of sPD by p‐tau levels (9.93 pg/mL, AUC = 0.909). MoCA (p = 0.035), t‐tau (p = 3.2×10−60), a‐beta (p = 1.7×10−22), and a‐syn (p = 4.6×10−32) were significantly lower in the subtype with low levels of p‐tau.ConclusionsIntegrating molecular profiling and clinical scale scores improves the prediction of sPD risk and identifies sPD subtypes.