Fabrizio Carlo, Termine Andrea, Caltagirone Carlo
Data Science Unit, Santa Lucia Foundation IRCCS, Rome, Italy.
Department of Clinical and Behavioral Neurology, Santa Lucia Foundation IRCCS, Rome, Italy.
J Cent Nerv Syst Dis. 2025 Jan 27;17:11795735241286821. doi: 10.1177/11795735241286821. eCollection 2025.
Parkinson's Disease (PD) varies widely among individuals, and Artificial Intelligence (AI) has recently helped to identify three disease progression subtypes. While their clinical features are already known, their gene expression profiles remain unexplored.
The objectives of this study were (1) to describe the transcriptomics characteristics of three PD progression subtypes identified by AI, and (2) to evaluate if gene expression data can be used to predict disease subtype at baseline.
This is a retrospective longitudinal cohort study utilizing the Parkinson's Progression Markers Initiative (PPMI) database.
Whole blood RNA-Sequencing data underwent differential gene expression analysis, followed by multiple pathway analyses. A Machine Learning (ML) classifier, namely XGBoost, was trained using data from multiple modalities, including gene expression values.
Our study identified differentially expressed genes (DEGs) that were uniquely associated with Parkinson's disease (PD) progression subtypes. Importantly, these DEGs had not been previously linked to PD. Gene-pathway analysis revealed both distinct and shared characteristics between the subtypes. Notably, two subtypes displayed opposite expression patterns for pathways involved in immune response alterations. In contrast, the third subtype exhibited a more unique profile characterized by increased expression of genes related to detoxification processes. All three subtypes showed a significant modulation of pathways related to the regulation of gene expression, metabolism, and cell signaling. ML revealed that the progression subtype with the worst prognosis can be predicted at baseline with 0.877 AUROC, yet the contribution of gene expression was marginal for the prediction of the subtypes.
This study provides novel information regarding the transcriptomics profiles of PD progression subtypes, which may foster precision medicine with relevant indications for a finer-grained diagnosis and prognosis.
帕金森病（PD）在个体间差异很大，人工智能（AI）最近有助于识别三种疾病进展亚型。虽然它们的临床特征已经为人所知，但其基因表达谱仍未得到探索。
本研究的目的是（1）描述由人工智能识别的三种PD进展亚型的转录组学特征，以及（2）评估基因表达数据是否可用于在基线时预测疾病亚型。
这是一项利用帕金森病进展标志物倡议（PPMI）数据库的回顾性纵向队列研究。
对全血RNA测序数据进行差异基因表达分析，随后进行多种通路分析。使用包括基因表达值在内的多种模式的数据训练了一种机器学习（ML）分类器，即XGBoost。
我们的研究确定了与帕金森病（PD）进展亚型独特相关的差异表达基因（DEG）。重要的是，这些DEG以前未与PD相关联。基因通路分析揭示了各亚型之间既不同又共享的特征。值得注意的是，两种亚型在免疫反应改变所涉及的通路上表现出相反的表达模式。相比之下，第三种亚型表现出更独特的特征，其特点是与解毒过程相关的基因表达增加。所有三种亚型在与基因表达调控、代谢和细胞信号传导相关的通路上均表现出显著调节。机器学习显示，预后最差的进展亚型在基线时预测的曲线下面积（AUROC）为0.877，但基因表达对亚型预测的贡献很小。
本研究提供了关于PD进展亚型转录组学谱的新信息，这可能有助于推进精准医学，为更精细的诊断和预后提供相关指征。
