Identification and validation of endoplasmic reticulum stress-related diagnostic biomarkers for type 1 diabetic cardiomyopathy based on bioinformatics and machine learning

Qiao Tang ¹ Yanwei Ji ¹ Zhongyuan Xia ¹ Yuxi Zhang ¹ Chong Dong ² Qian Sun ^1* Shaoqing Lei ^1*

• ¹ Department of Anesthesiology, Renmin Hospital of Wuhan University, Wuhan, China
• ² Tianjin First Central Hospital, Tianjin, China

Background: Diabetic cardiomyopathy (DC) is a serious complication in patients with type 1 diabetes mellitus and has become a growing public health problem worldwide. There is evidence that endoplasmic reticulum stress (ERS) is involved in the pathogenesis of DC, and related diagnostic markers have not been well-studied. Therefore, this study aimed to screen ERS-related genes (ERGs) with potential diagnostic value in DC.METHODS: Gene expression data on DC were downloaded from the GEO database, and ERGs were obtained from The Gene Ontology knowledgebase. Limma package analyzed differentially expressed genes (DEGs) in the DC and control groups, and then integrated with ERGs to identify ERS-related DEGs (ERDEGs). The ERDEGs diagnostic model was developed based on a combination of LASSO and Random Forest approaches, and the diagnostic performance was evaluated by the area under the receiver operating characteristic curve (ROC-AUC) and validated against external datasets. In addition, the association of the signature genes with immune infiltration was analyzed using the CIBERSORT algorithm and the Spearman correlation test.Results: Gene expression data on DC were downloaded from the GEO database and ERGs were obtained from the Gene Ontology Knowledgebase. Limma package analysis identified 3100 DEGs between DC and control groups and then integrated with ERGs to identify 65 ERDEGs. Four diagnostic markers, Npm1, Jkamp, Get4, and Lpcat3, were obtained based on the combination of LASSO and random forest approach, and their ROC-AUCs were 0.9112, 0.9349, 0.8994, and 0.8639, respectively, which proved their diagnostic potential in DC. Meanwhile, Npm1, Jkamp, Get4, and Lpcat3 were validated by external datasets and a mouse model of type 1 DC. In addition, Npm1 was significantly negatively correlated with plasma cells, activated natural killer cells, or quiescent mast cells, whereas Get4 was significantly positively correlated with quiescent natural killer cells and significantly negatively correlated with activated natural killer cells (P < 0.05).This study provides novel diagnostic biomarkers (Npm1, Jkamp, Get4, and Lpcat3) for DC from the perspective of ERS, which provides new insights into the development of new targets for individualized treatment of type 1 diabetic cardiomyopathy.