Hao Lv,1– 3,* Jingkun Wang,2,3,* Yang Wan,4 Yun Zhou1,2 

1Department of Rehabilitation Medicine, The Second Affiliated Hospital of Anhui Medical University, Hefei, 230601, People’s Republic of China; 2Research Center for Translational Medicine, The Second Affiliated Hospital of Anhui Medical University, Hefei, 230601, People’s Republic of China; 3Department of Orthopaedics, The Second Affiliated Hospital of Anhui Medical University, Hefei, Anhui, 230601, People’s Republic of China; 4Department of Hematology/Hematological Lab, The Second Affiliated Hospital of Anhui Medical University, Hefei, 230601, People’s Republic of China

*These authors contributed equally to this work

Correspondence: Yun Zhou, Department of Rehabilitation Medicine, the Second Affiliated Hospital of Anhui Medical University, No. 678 Furong Road, Hefei, People’s Republic of China, Email zhouyunanhui@sina.com

Background: Osteoarthritis (OA), a degenerative and chronic joint disease, is essential for identifying novel biomarkers for the clinical diagnosis of OA.
Methods: We collected 35 OA patients and 32 healthy controls from four clinical cohorts and 8 real-world samples from our institute. The activation status of 7530 signalling pathways was calculated via the gene set enrichment analysis (GSEA) algorithm. Ten machine learning algorithms and 101 algorithm combinations were further applied to recognize the most diagnostic genes. KDELR3 was chosen for further validation via immunohistochemical staining to determine its diagnostic value in real-world samples.
Results: Sixteen pathways, namely, the cellular respiration chain, protein transport, lysosomal and endocytosis pathways, were activated in OA patients. A total of 101 types of algorithm combinations were considered for the diagnostic model, and 58 were successfully output. The two-step model of glmBoost plus RF had the highest average AUC value of 0.95 and was composed of LY86, SORL1, KDELR3, CSK, PTGS1, and PTGS2. Preferable consistency of the diagnostic mole and real conditions was observed in all four cohorts (GSE55235: Kappa=1.000, P< 0.001; GSE55457: Kappa=0.700, P< 0.001; GSE82107: Kappa=0.643, P=0.004; GSE1919: Kappa=1.000, P< 0.001). KDELR3 was expressed at higher levels in OA patients than were the other genes, and with the help of immunohistochemistry (IHC), we confirmed that OA patients presented high levels of KDELR3 in synovial tissues. The infiltration of immunocytes, macrophages, and natural killer T cells was high in OA patients. KDELR3 might be involved in the activation and infiltration of effector memory CD4 T cells (Rpearson = 0.58, P < 0.001) and natural killer T cells (Rpearson = 0.53, P < 0.001).
Conclusion: We constructed and validated a six-gene diagnostic model for OA patients via machine learning, and KDELR3 emerged as a novel biomarker for OA.

Keywords: osteoarthritis, diagnostic model, machine learning, synovial tissue, immunohistochemical staining, KDELR3