Metabolic Profiling of CSF from People Suffering from Sporadic and LRRK2 Parkinson's Disease: A Pilot Study

Abstract

CSF from unique groups of Parkinson's disease (PD) patients was biochemically profiled to identify previously unreported metabolic pathways linked to PD pathogenesis, and novel biochemical biomarkers of the disease were characterized. Utilizing both ¹H NMR and DI-LC-MS/MS we quantitatively profiled CSF from patients with sporadic PD (n = 20) and those who are genetically predisposed (LRRK2) to the disease (n = 20), and compared those results with age and gender-matched controls (n = 20). Further, we systematically evaluated the utility of several machine learning techniques for the diagnosis of PD. ¹H NMR and mass spectrometry-based metabolomics, in combination with bioinformatic analyses, provided useful information highlighting previously unreported biochemical pathways and CSF-based biomarkers associated with both sporadic PD (sPD) and LRRK2 PD. Results of this metabolomics study further support our group's previous findings identifying bile acid metabolism as one of the major aberrant biochemical pathways in PD patients. This study demonstrates that a combination of two complimentary techniques can provide a much more holistic view of the CSF metabolome, and by association, the brain metabolome. Future studies for the prediction of those at risk of developing PD should investigate the clinical utility of these CSF-based biomarkers in more accessible biomatrices. Further, it is essential that we determine whether the biochemical pathways highlighted here are recapitulated in the brains of PD patients with the aim of identifying potential therapeutic targets.

Keywords: 1H NMR; machine learning; metabolic pathways. Parkinson’s disease; metabolomics; targeted mass spectrometry.

Figure 1

(a) The O-PLS-DA score plot of LRKK control vs. LRRK2 sufferers; (b) top 13 metabolites on loadings driving the separation; (c) O-PLS-DA score plot of sporadic Parkinson’s disease (sPD) controls vs. sPD sufferers; (d) top 13 metabolites on loadings driving the separation; (e) the O-PLS-DA score plot of sPD vs. LRKK2 sufferers; (f) top 13 metabolites on loadings driving the separation.

Figure 2

AUC values of various machine learning algorithms evaluated for the prediction of LRRK2 PD as compared LRRK2 controls, sPD as compared to sPD controls, and LRRK2 PD as compared to sPD on both test and training sets, respectively.

Figure 3

Violin plots comparing the median concentrations of the different metabolite species as identified in the CSF of sPD (n = 20), sPD controls (n = 20), LRRK2 PD (n = 20), and LRRK2 controls (n = 20). The data were analyzed using a Student’s t-test where ns: p > 0.05, *: p ≤ 0.05, **: p ≤ 0.01, ***: p ≤ 0.001. A point in one of the violin plots is representative of a patient sample. Each p-value shown on a plot is an ANOVA p-value for all the patient groups.