Multi-Omics Identification of Key Immune Molecules in Gestational Diabetes Mellitus: FKBP5 and HLA-DQA1 as Candidate Biomarkers

Ju Zhang,1 Yuehan Qin,2 Nuo Chen,3 Jiao Wang4 1Department of Obstetrics, Guizhou Provincial People’s Hospital, Guiyang, 550000, People’s Republic of China; 2Department of Obstetrics,Guiyang Maternal and Child Health Care Hospital, Guiyang, 550000, People’s Republic of China; 3Department of Obstetrics, Guiyang Nanming District People’s Hospital, Guiyang, 550000, People’s Republic of China; 4Department of Obstetrics,The Affiliated Hospital of Guizhou Medical University, Guizhou Medical University, Guiyang, 550000, People’s Republic of ChinaCorrespondence: Jiao Wang, Department of Obstetrics, The Affiliated Hospital of Guizhou Medical University, No. 86, Beijing Road, Yunyan District, Guiyang, Guizhou Province, 550000, People’s Republic of China, Email wangjiao1585455223@163.comBackground: Gestational diabetes mellitus (GDM) is a major metabolic complication of pregnancy in which immune dysregulation has been implicated. The Systemic Immune-Inflammation Index (SII) has been significantly associated with GDM risk, highlighting the importance of the placental immune microenvironment in GDM pathogenesis. Yet comprehensive cross-modality integration of immune molecular data remains limited. This study aimed to systematically identify and validate key placental immune molecules in GDM.Patients and Methods: Bulk transcriptome data (training set: 10 GDM/10 controls; validation set: 32 GDM/31 controls) and single-cell data (4 donors, 2 GDM/2 controls) were obtained from GEO. Differential expression, GO/KEGG enrichment, and ssGSEA-based immune infiltration analyses were performed. Three machine-learning algorithms (LASSO, SVM-RFE, Random Forest) were applied for feature selection, and consensus genes were validated in the independent validation cohort and in clinical samples by qRT-PCR (30 GDM/30 controls). A nomogram model was built and assessed by AUC, calibration curves, the Hosmer-Lemeshow test, and the Brier score. Candidate drugs were identified via CMap, with molecular docking and 100-ns molecular dynamics simulations.Results: Bulk analysis identified 378 differentially expressed genes (190 up, 188 down) enriched in immune response, cytokine production, and insulin signalling. Three-algorithm consensus nominated CLEC12A, FKBP5 and HLA-DQA1; however, CLEC12A failed to replicate in the independent validation cohort and was therefore not advanced to qRT-PCR. FKBP5 and HLA-DQA1 retained significant downregulation in GDM placentas across the training set, validation set, and clinical samples (qRT-PCR P< 0.001). Both genes correlated with immune infiltration patterns including activated B cells. The AUCs in both training and validation sets were modest, indicating preliminary discriminative ability. The nomogram showed acceptable calibration (Hosmer-Lemeshow P=0.342; Brier score=0.118).Conclusion: FKBP5 and HLA-DQA1 are candidate biomarkers with preliminary diagnostic evidence warranting validation in larger multi-centre cohorts.Keywords: gestational diabetes mellitus, immune molecules, single-cell transcriptome, machine learning, immune microenvironment, precision medicine