Prediction of motor and non-motor Parkinson’s disease symptoms using serum lipidomics and machine learning: a 2-year study

Serum lipid prediction of 2 year clinical trajectories in Parkinson's disease
Prediction of motor and non-motor Parkinson’s disease symptoms using serum lipidomics and machine learning: a 2-year study
Like

Lipids in Parkinson's disease

The cause of Parkinson's disease is unknown, however, several genes that increase the risk of Parkinson's disease, such as GBA1, are involved in lipid metabolism or have a lipid-related function. Despite this evidence, how lipids may play a role in the clinical progression of Parkinson's disease remains unclear.

Lipids 101: What are they?

Thousands of lipids exist with diverse functions, ranging from being storage silos for energy to comprising the critical membranes which protect cells and organelles. Notably, some lipids are bioactive signalling molecules in diverse pathways including cell death and other integral cell processes such as mitochondrial function and glucose/insulin signalling. 

Training AI to predict future Parkinson's disease severity using blood lipids

Previously, an untargeted investigation of >1000 lipids in serum from N = 536 participants from the Michael J Fox Foundation LRRK2 Consortium indicated that the serum lipid signature of Parkinson’s disease patients was significantly distinguishable from controls1. A portion of participants with Parkinson's disease from this cohort were studied at baseline (N = 122) and 2 years (N = 67), and their severity scores for nine motor or non-motor clinical scales were recorded.  The importance of a participant's baseline serum lipid levels in predicting their future 2-year disease severity scores was then assessed. The performance of two different prediction algorithms (elastic net and random forest) was also compared. The predictive performance of 995 serum lipids was also compared to 27 serum cytokines previously measured in this cohort2.

Key findings

Baseline serum lipids performed better than cytokines in predicting the 2-year clinical severity of the Geriatric Depression Scale, Schwab and England Activities of Daily Living Scale, University of Pennsylvania Smell Identification Test, Movement Disorder Society Unified Parkinson’s Disease Rating Scale part three (UPDRS III) and the Hoehn and Yahr scale. Demographic information, including age, sex and LRRK2 risk gene mutation status, and clinical data including baseline score of the scale to be predicted after two years were also included in the machine learning models. The top 10 predicting baseline variables and their relative importance to the prediction of future clinical scores are shown in the table below.

The top 10 variables at baseline and their importance scores contributing to the prediction of future Geriatric Depression Scale, The Schwab and England Activities of Daily Living scale, UPSIT and UPDRS III score at 2-year follow-up from the selected lipids plus clinical and demographic data random forest models. N = 122 Parkinson’s disease participants at baseline and N = 67 at 2-year follow-up.

UPDRS III Unified Parkinson’s Disease Rating Scale, UPSIT University of Pennsylvania Smell Identification Test, DG diacylglycerol, LPE lysophosphatidylethanolamine, PAF platelet-activating factor; PC phosphatidylcholine, PE phosphatidylethanolamine, SM sphingomyelin and TG triacylglycerol.

What do these lipids do?

The lipids flagged in this study are implicated in several pathways, including mitochondrial function (diacylglycerol and triacylglycerol), inflammation and oxidative stress (platelet-activating factor), membrane integrity (phosphatidylcholine, phosphatidylethanolamine) and cell death (sphingomyelin). Interestingly, lysophosphatidylethanolamine 16:0 and lysophosphatidylcholine 16:0, which were top predictors of PD scale scores in this study, have recently been shown to inhibit the aggregation of α-synuclein3.

Research wrap up

Despite finding that serum lipids were better predictors of clinical scales than cytokines and in some instances sex, age and mutation status, the prediction model error would need to be addressed before clinical utilisation. It would be interesting to see if this could be improved by controlling for additional genotypes, medication use and increasing the sample size. In conclusion, this untargeted study generated a shortlist of serum lipid candidates that may have a role in the progression of Parkinson's disease.

References:
  1. Galper, J. et al. Lipid pathway dysfunction is prevalent in patients with Parkinson’s disease. Brain https://doi.org/10.1093/brain/awac176 (2022).
  2.  Ahmadi Rastegar, D., Ho, N., Halliday, G.M. et al. Parkinson’s progression prediction using machine learning and serum cytokines. npj Parkinsons Dis. 5, 14 (2019). https://doi.org/10.1038/s41531-019-0086-4
  3. Karaki, T., Haniu, H., Matsuda, Y. & Tsukahara, T. Lysophospholipids–potent candidates for brain food, protects neuronal cells against α-Synuclein aggregation. Biomed. Pharmacother. 156, 113891 (2022).

Please sign in or register for FREE

If you are a registered user on Research Communities by Springer Nature, please sign in

Follow the Topic

Parkinson's disease
Life Sciences > Biological Sciences > Neuroscience > Neurological Disorders > Parkinson's disease
Lipidomics
Physical Sciences > Chemistry > Analytical Chemistry > Mass Spectrometry > Lipidomics
Machine Learning
Mathematics and Computing > Statistics > Statistics and Computing > Machine Learning
Biomarkers
Life Sciences > Health Sciences > Biomedical Research > Biomarkers
Dementia
Life Sciences > Health Sciences > Clinical Medicine > Neurology > Neurological Disorders > Dementia
Neurodegenerative diseases
Life Sciences > Biological Sciences > Neuroscience > Neurological Disorders > Neurodegenerative diseases