关键词: 早发性卵巢功能不全, 细胞衰老, 内质网应激, 生物信息学, 机器学习

Abstract: BACKGROUND: Ovarian granulosa cell senescence and endoplasmic reticulum stress are closely related to the development and progression of premature ovarian insufficiency; however, the underlying regulatory mechanisms remain unelucidated. 
OBJECTIVE: To identify potential biomarkers associated with cellular senescence and endoplasmic reticulum stress in granulosa cells in premature ovarian insufficiency using bioinformatic analysis and machine learning algorithms, with subsequent validation in animal experiments. 
METHODS: The premature ovarian insufficiency dataset GSE201276 was downloaded from the GEO database. Differentially expressed genes were screened, and weighted gene co-expression network analysis was performed to identify module genes. Gene sets related to cellular senescence and endoplasmic reticulum stress were obtained from the GeneCards database and intersected with the differentially expressed genes and module genes. Consensus clustering analysis was then conducted to identify subtype-specific differentially expressed genes, followed by Gene Ontology and Kyoto Encyclopedia of Genes and Genomes enrichment analyses and immune infiltration analysis. Two machine learning algorithms were applied to screen for key genes associated with granulosa cell senescence and endoplasmic reticulum stress in premature ovarian insufficiency. A diagnostic model was constructed and validated. Finally, a C57BL/6J mouse model of premature ovarian insufficiency was established. The model was evaluated by estrous cycle monitoring, hematoxylin-eosin staining, and serum ELISA. The expression of key genes was further verified using quantitative real-time PCR and western blot assay.
RESULTS AND CONCLUSION: (1) Consensus clustering analysis identified 911 subtype-specific differentially expressed genes associated with cellular senescence and endoplasmic reticulum stress. Gene Ontology enrichment analysis indicated that these differentially expressed genes were primarily involved in biological processes such as negative regulation of the cell cycle, meiosis, and female gonad development. Kyoto Encyclopedia of Genes and Genomes pathway analysis revealed significant associations with oocyte meiosis, progesterone-mediated oocyte maturation, and the transforming growth factor-β signaling pathway. Immune infiltration analysis demonstrated significantly elevated levels of M1 macrophages and resting dendritic cells in the premature ovarian insufficiency group (P < 0.05). (2) Four key genes were screened using machine learning algorithms. Diagnostic model and calibration curve results indicated that aurora kinase A and actin-binding protein exhibited strong predictive performance. (3) Animal experiments showed that compared with the blank group, the model mouses exhibited disrupted estrous cycles, a decreased number in primary, secondary, and antral follicles, and an increased number of atretic follicles (P < 0.01). Serum follicle-stimulating hormone levels were significantly elevated, while anti-Müllerian hormone levels were markedly reduced, with significant differences (P < 0.01). Compared with the blank group, both mRNA and protein expression levels of aurora kinase A and actin-binding protein were significantly downregulated in the ovarian tissues of the model group, with significant differences (P < 0.05). (4) These findings suggest that aurora kinase A and actin-binding protein may contribute to the pathogenesis of premature ovarian insufficiency by regulating granulosa cell senescence and endoplasmic reticulum stress. Their specific regulatory roles and molecular mechanisms merit further experimental investigation. 

Key words: premature ovarian insufficiency, cellular senescence, endoplasmic reticulum stress, bioinformatics, machine learning