整合代谢组学与机器学习揭示前庭性偏头痛的代谢特征及候选生物标志物

文献信息

中文摘要

背景： 前庭性偏头痛（VM）是一种被低估的疾病，以发作性眩晕为特征，常伴有头痛。尽管其患病率和疾病负担相当大，但其代谢特征仍鲜有描述，且缺乏用于诊断或活动度评估的血清生物标志物。方法：我们在发现队列（33例VM vs 30例健康对照[HCs]）和复制队列（37例VM vs 27例HCs）中进行了非靶向代谢组学，仅在发现队列中应用机器学习进行标志物筛选。结果：在发现队列中，510种代谢物表现出差异丰度（197种上调，313种下调）。京都基因与基因组百科全书（KEGG）富集分析显示，上调代谢物在半胱氨酸/蛋氨酸、组氨酸和苯丙氨酸代谢中呈富集趋势。机器学习识别出三种候选代谢物——苯乙胺、N-乙酰血清素和半胱氨酸-S-硫酸盐——每种均达到曲线下面积>0.9。在发现队列中，半胱氨酸-S-硫酸盐与眩晕视觉模拟量表评分相关，苯乙胺与头痛影响测试-6和广泛性焦虑障碍-7（GAD-7）评分相关。在复制队列中，所有三种标志物均保持差异表达和判别性能；此外，半胱氨酸-S-硫酸盐与头晕残障量表评分相关，苯乙胺与GAD-7评分相关。结论：本研究阐明了VM的代谢景观并识别出有前景的血清生物标志物，但这些发现尚需在更大规模的多中心队列中进一步验证，之后才能进行临床转化。

英文摘要

Vestibular migraine (VM) is an underrecognized disorder characterized by episodic vertigo, often accompanied by headache. Despite its substantial prevalence and disease burden, its metabolic features remain poorly characterized, and serum biomarkers for diagnosis or activity assessment are lacking. We performed untargeted metabolomics in discovery (33 VM vs 30 healthy controls [HCs]) and replication (37 VM vs 27 HCs) cohorts, applying machine learning solely in the discovery cohort for marker selection. In the discovery cohort, 510 metabolites exhibited differential abundance (197 upregulated, 313 downregulated). Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis revealed enrichment trends for upregulated metabolites in cysteine/methionine, histidine, and phenylalanine metabolism. Machine learning identified three candidate metabolites─phenylethylamine, N-acetylserotonin, and cysteine-S-sulfate─each achieving an area under the curve > 0.9. In the discovery cohort, cysteine-S-sulfate correlated with vertigo Visual Analog Scale scores, and phenylethylamine correlated with Headache Impact Test-6 and Generalized Anxiety Disorder-7 (GAD-7) scores. In the replication cohort, all three markers maintained differential expression and discriminative performance; additionally, cysteine-S-sulfate correlated with Dizziness Handicap Inventory scores, and phenylethylamine with GAD-7 scores. This study elucidates the metabolic landscape of VM and identifies promising serum biomarkers, though these findings warrant further validation in larger, multicenter cohorts before clinical translation.