Background: Systemic autoimmune diseases (SADs) present significant clinical challenges due to their heterogeneity, which complicates patient classification, and delays diagnosis, often leading to late therapeutic interventions and worse outcomes. These challenges could be addressed through the application of new omics technologies and advanced computational tools, which allow for deeper molecular profiling, identification of biomarkers, and the development of predictive models to improve patient stratification, accelerate diagnosis, and guide personalized treatment approaches.

Objectives: To characterize the metabolomic fingerprint of several SADs, aiming to uncover novel molecular insights and enhance patient stratification and diagnosis through the application of machine learning (ML) techniques.

Methods: A total of 716 individuals from the international multicenter study PRECISESADS, in collaboration with the Reina Sofia University Hospital (Cordoba, Spain), were included in the analysis: 272 with rheumatoid arthritis (RA), 183 with systemic lupus erythematosus (SLE), 148 with primary antiphospholipid syndrome (PAPS), 70 with systemic sclerosis (SSc), and 43 healthy donors (HDs). The comprehensive metabolomic profile of serum samples was analyzed using nuclear magnetic resonance (NMR) spectroscopy (Nightingale). In parallel, extensive clinical and analytical data were collected. A combination of unsupervised and supervised ML methods was applied, including k-means clustering for unsupervised patient stratification and Recursive Feature Elimination (RFE) with Logistic Regression using 5-Fold Cross Validation to identify potential diagnostic signatures in a supervised framework. Finally, correlation and association analyses were conducted to investigate relationships between altered metabolomic profiles and clinical parameters.

Results: Several metabolites (metabs) were differentially expressed in each disease compared to HDs, with the highest number of alterations observed in SSc (99 metabs) and APS (68 metabs), followed by SLE (30 metabs) and RA (17 metabs). Notably, some metabolites were simultaneously altered across all diseases, including histidine, albumin, Cholesterol in large LDL (L-LDL-C) and Cholesteryl esters in large LDL (L-LDL-CE) which are key cardiovascular-related metabolites. Each disease also displayed a distinct set of uniquely altered metabolites. ML demonstrated strong diagnostic potential by generating disease-specific signatures based on different metabolites, achieving an AUC above 0.8 in all cases. Unsupervised clustering analysis of the entire cohort identified three distinct patient clusters (C1, C2, C3), with each disease represented across all clusters in varying proportions, highlighting disease heterogeneity and shared metabolic pathways. C1 showed significant differences compared to C3 in both clinical features and metabolomic profiles, while C2 represented an intermediate group. Patients in C1 had increased levels of LDL- and VLDL-related metabolites and decreased levels of HDL-related metabolites compared with C3. In APS patients, C1 presented higher aGAPSS thrombotic risk scores and prevalence of arterial hypertension (HTA), dyslipidemia, diabetes, and arterial thrombosis, compared to C2, which displayed higher proportions of venous thrombosis and pregnancy complications. In SLE patients, C1 had a higher prevalence of dyslipidemia, HTA, atherosclerotic plaques, anti-dsDNA positivity, and lupus nephritis compared to C3. In RA patients, C1 showed higher rates of obesity, atherosclerotic plaques, diabetes, and HTA, while C3 had higher disease activity and positivity for rheumatoid factor and anti-CCP antibodies. Finally, in SSc patients, C1 exhibited higher proportions of dyslipidemia, anti-U1-RNP antibody positivity, and lung fibrosis, whereas C3 showed greater skin involvement and anti-centromere antibody positivity.

Conclusion: Distinct metabolomic profiles were identified across systemic autoimmune diseases, with shared and unique alterations. ML generated accurate diagnostic signatures, and clustering revealed patient subgroups characterized by unique clinical and metabolic profiles across autoimmune diseases, highlighting shared pathogenic mechanisms and disease-specific features. The identification of metabolically distinct patient subsets supports the potential for personalized therapeutic approaches targeting metabolic pathways to better manage complications and improve outcomes across different autoimmune diseases.

REFERENCES: NIL.

Acknowledgements: Supported by CPS: RYC2021-033828-I; PID2022-141500OA-I00; DIN2022-012766 Minister of Science, Innovation and Universities co-financed by the European Union; and CLP: (PI21/00591, PI21/00959, CD21/00187 and RICOR-21/0002/0033), co-financed by European Union. EU/EFPIA-IMI-PRECISESADS (n° 115565)

Disclosure of Interests: None declared.

© The Authors 2025. This abstract is an open access article published in Annals of Rheumatic Diseases under the CC BY-NC-ND license ( http://creativecommons.org/licenses/by-nc-nd/4.0/ ). Neither EULAR nor the publisher make any representation as to the accuracy of the content. The authors are solely responsible for the content in their abstract including accuracy of the facts, statements, results, conclusion, citing resources etc.