Kai Wang,1,2 Rujie Zhu,1 Min Xu,3 Kexin Zhu,1 Ju Li,2 Chang Li,4 Deqian Meng,2 Hongwei Chen,1,3,5 Lingyun Sun1,5 

1Department of Rheumatology and Immunology, Nanjing Drum Tower Hospital Clinical College of Nanjing Medical University, Nanjing, 210008, People’s Republic of China; 2Department of Rheumatology and Immunology, The Affiliated Huai’an No.1 People’s Hospital of Nanjing Medical University, Huai’an, 223001, People’s Republic of China; 3Department of Rheumatology and Immunology, Nanjing Drum Tower Hospital, Clinical College of Nanjing University of Chinese Medicine, Nanjing, 210008, People’s Republic of China; 4Department of Medical Laboratory, The Affiliated Huai’an No.1 People’s Hospital of Nanjing Medical University, Huai’an, 223001, People’s Republic of China; 5Department of Rheumatology and Immunology, Nanjing Drum Tower Hospital, The Affiliated Hospital of Nanjing University Medical School, Nanjing, 210008, People’s Republic of China

Correspondence: Lingyun Sun, Department of Rheumatology and Immunology, Nanjing Drum Tower Hospital Clinical College of Nanjing Medical University, Zhongshan Road 321, Nanjing, 210008, People’s Republic of China, Email lingyunsun@nju.edu.cn Hongwei Chen, Department of Rheumatology and Immunology, Nanjing Drum Tower Hospital, The Affiliated Hospital of Nanjing University Medical School, Nanjing, 210008, People’s Republic of China, Email chenhw@nju.edu.cn

Objective: This study aimed to identify potential diagnostic biomarkers for systemic lupus erythematosus (SLE) using metabolomics approaches and machine learning algorithms, and to evaluate therapeutic targets for SLE treatment.
Methods: Serum samples from 44 SLE patients with lupus nephritis, 40 rheumatoid arthritis patients, 39 primary Sjögren’s syndrome patients, and matched healthy controls were analyzed using ultra-performance liquid chromatography-high resolution mass spectrometry (UPLC-HRMS). Eight machine learning algorithms were employed to establish diagnostic models. Partial least squares discriminant analysis (PLS-DA) and orthogonal PLS-DA (OPLS-DA) were used to identify differential metabolites. The therapeutic potential of identified metabolites was validated in MRL-Faslpr mice through histological examination, flow cytometry, and biochemical analysis.
Results: A total of 129 metabolites were detected, with machine learning models achieving area under the curve (AUC) values > 0.8. The principal component regression model performed best with AUC values of 0.99 and 0.96 for training and test datasets, respectively. Two key metabolites, tryptophan and beta-alanine, showed significantly decreased levels in SLE patients compared to healthy controls (both p< 0.05), while exhibiting opposite patterns in other autoimmune diseases. In the mouse model, tryptophan supplementation improved renal histology, reduced proteinuria, increased naïve T cells and central memory T cells, and decreased effector T cell frequencies in both peripheral blood and spleen.
Conclusion: This study demonstrates the successful application of machine learning algorithms to metabolomics data for SLE classification and identifies tryptophan and beta-alanine as potential SLE-specific biomarkers. Tryptophan supplementation shows therapeutic promise in lupus mouse models through immunomodulatory effects on T cell subsets and renal protection.

Keywords: SLE, metabolomic, tryptophan, beta-alanine