Wu Li, Zhang Renxin, Wang Yichao, Dai Shaoxing, Yang Naixue
State Key Laboratory of Primate Biomedical Research, Institute of Primate Translational Medicine, Kunming University of Science and Technology, Kunming, Yunnan, China.
Yunnan Key Laboratory of Primate Biomedical Research, Kunming, Yunnan, China.
Front Aging Neurosci. 2025 May 30;17:1571783. doi: 10.3389/fnagi.2025.1571783. eCollection 2025.
Data-driven omics approaches have rapidly advanced our understanding of the molecular heterogeneity of Alzheimer's disease (AD). However, limited by the unavailability of brain tissue, there is an urgent need for a non-invasive tool to detect alterations in the AD brain. Cell-free RNA (cfRNA), which crosses the blood-brain barrier, could reflect AD brain pathology and serve as a diagnostic biomarker.
Here, we integrated plasma-derived cfRNA-seq data from 337 samples (172 AD patients and 165 age-matched controls) with brain-derived single cell RNA-seq (scRNA-seq) data from 88 samples (46 AD patients and 42 controls) to explore the potential of cfRNA profiling for AD diagnosis. A systematic comparative analysis of cfRNA and brain scRNA-seq datasets was conducted to identify dysregulated genes linked to AD pathology. Machine learning models-including support vector machine, random forest, and logistic regression-were trained using cfRNA expression patterns of the identified gene set to predict AD diagnosis and classify disease progression stages. Model performance was rigorously evaluated using area under the receiver operating characteristic curve (AUC), with robustness assessed through cross-validation and independent validation cohorts.
Notably, we identified 34 dysregulated genes with consistent expression changes in both cfRNA and scRNA-seq. Machine learning models based on the cfRNA expression patterns of these 34 genes can accurately predict AD patients (the highest AUC = 89%) and effectively distinguish patients at early stage of AD. Furthermore, classifiers developed based on the expression of 34 genes in brain transcriptome data demonstrated robust predictive performance for assessing the risk of AD in the population (the highest AUC = 94%).
This multi-omics approach overcomes limitations of invasive brain biomarkers and noisy blood-based signatures. The 34-gene panel provides non-invasive molecular insights into AD pathogenesis and early screening. While cfRNA stability challenges clinical translation, our framework highlights the potential for precision diagnostics and personalized therapeutic monitoring in AD.
数据驱动的组学方法迅速提升了我们对阿尔茨海默病（AD）分子异质性的理解。然而，由于脑组织难以获取，迫切需要一种非侵入性工具来检测AD大脑中的变化。可穿过血脑屏障的游离RNA（cfRNA）能够反映AD大脑病理学特征，并可作为一种诊断生物标志物。
在此，我们将来自337个样本（172例AD患者和165例年龄匹配的对照）的血浆来源cfRNA测序数据与来自88个样本（46例AD患者和42例对照）的脑来源单细胞RNA测序（scRNA-seq）数据进行整合，以探索cfRNA谱分析用于AD诊断的潜力。对cfRNA和脑scRNA-seq数据集进行了系统的比较分析，以鉴定与AD病理学相关的失调基因。使用所鉴定基因集的cfRNA表达模式训练机器学习模型，包括支持向量机、随机森林和逻辑回归，以预测AD诊断并对疾病进展阶段进行分类。使用受试者操作特征曲线下面积（AUC）对模型性能进行严格评估，并通过交叉验证和独立验证队列评估其稳健性。
值得注意的是，我们鉴定出34个在cfRNA和scRNA-seq中具有一致表达变化的失调基因。基于这34个基因的cfRNA表达模式的机器学习模型能够准确预测AD患者（最高AUC = 89%），并有效区分AD早期患者。此外，基于脑转录组数据中34个基因的表达开发的分类器在评估人群中AD风险方面表现出强大的预测性能（最高AUC = 94%）。
这种多组学方法克服了侵入性脑生物标志物和基于血液的嘈杂特征的局限性。34基因面板为AD发病机制和早期筛查提供了非侵入性分子见解。虽然cfRNA稳定性对临床转化构成挑战，但我们的框架突出了AD精准诊断和个性化治疗监测的潜力。
