Parkinson’s disease causality and heterogeneity: a proteogenomic view

Abstract

Pathogenesis and clinical heterogeneity in Parkinson's disease (PD) have been evaluated from genetic, pathological, and clinical perspective. Technology allowing for high-throughput proteomic analyses in cerebrospinal fluid (CSF) has opened new opportunities to scrutinize this heterogeneity. This is to date the most comprehensive proteomics profiling in CSF of PD patients and controls, both in terms of subjects (n=1103) and proteins (n=4135). Combining CSF aptamer-based proteomics with genetics across all samples we determined the protein quantitative trait loci (pQTLs) linking genetic variants with protein abundance in CSF. Analyzing pQTLs together with summary statistics from the largest PD genome wide association study (GWAS) led us to identify 68 potential causal proteins by Mendelian randomization. The top PD causal protein, GPNMB, has colocalization support and has been reported to be upregulated in the substantia nigra of PD patients. We also examined three subcohorts of PD patients: LRRK2 variant carriers (LRRK2+), GBA variant carriers (GBA+) and idiopathic PD patients, each with their respective controls. The second goal was to identify proteomic differences between PD patients and controls within and between the three subcohorts. Statistical analyses revealed six proteins differentially expressed when comparing GBA+ PD patients with unaffected GBA+ controls, seven proteins when comparing LRRK2+ PD patients with unaffected LRRK2+ controls and 23 proteins when comparing idiopathic PD patients with healthy controls who did not carry any severe PD mutations. Furthermore a hypothesis-free stratification of idiopathic PD patients based on their proteomic profile revealed two protein modules based on the co-expression network structure. Based on these modules, cluster analysis revealed two patient endotypes for idiopathic PD. Differences in the CSF proteomic signature between subcohorts and between idiopathic endotypes, as well as causal targets identified using both proteomics and genetics may together influence the way we approach the identification of potential therapeutic targets in PD.

Changes in the cerebrospinal fluid proteome precede and stratify the course of Parkinson’s Disease

“…Further, we have shown that non-genetic PD-associated CSF proteomic signatures were correlated with several baseline clinical parameters and CSF biomarkers, and were robustly linked to subsequent motor and cognitive decline in PD, regardless of genetic or non-genetic PD. In conjunction with the findings of an article related to ours, 33 this study provides the most comprehensive proteomic profiling of PD to date. Furthermore, to our knowledge, this is the first study to successfully develop a measure that accurately captures non-genetic PD-associated CSF proteomic signatures and to demonstrate its robust association with the progression of motor and cognitive symptoms of PD, collectively indicating that the CSF proteome encodes information important for both the onset and progression of PD.…”

“…Finally, it should be mentioned that in an article related to ours, 33 it was found that the most differentially expressed proteins in GBA1 -PD, LRRK2 -PD, and non-genetic PD differed from each other. Therefore, at first, we did not expect to observe that CSF proteomic signatures associated with non-genetic PD also existed in patients with GBA1 -PD, LRRK2 -PD, and SNCA -PD.…”

“…For LRRK2 coding variants, p.N1437H, p.R1441C/G/H, p.G2019S, and p.I2020T were considered as LRRK2 mutations. 36 It should be noticed that in an article related to ours, 33 patients with p.E326K, p.T369M, and p.A456P variants in the GBA1 gene were excluded because it focused on severe GBA1 mutations. However, we included these patients in this study because we put focus on examining the utility of the ngPD-ProS in as many patients with PD as possible.…”

“…Most SOMAmers have a one-to-one correspondence with their target protein, but some target the same protein, while others target two (or more) proteins simultaneously. While SOMAmers targeting more than two proteins were also included in the analysis in an article related to ours, 33 we excluded them from the current analysis to enable future application using other platforms. Finally, in this study, we quantified the protein levels of 4071 human proteins targeted by 4716 SOMAmers.…”

“…First, differentially expressed protein analysis was conducted using the linear model implemented in the R package “limma,” adjusted for age, sex, and principal components 1– 4. 33, 55 SOMAmers with q values of ≤ 0.2 were defined as SOMAmers that were likely to be differentially expressed, 56, 57 and proteins targeted by these SOMAmers were defined as proteins that were likely to be differentially expressed. Second, using these SOMAmers, we applied logistic regression with the least absolute shrinkage and selection operator (LASSO) regularization.…”