Background: Systemic lupus erythematosus (SLE) exhibits significant heterogeneity in clinical progression and treatment response, posing challenges in both diagnosis and therapeutic interventions.

Objectives: To explore mechanisms underlying relevant clinical endotypes in SLE patients though the integrated analysis of the serum proteomic and metabolomic profiles, employing advanced machine learning approaches.

Methods: Proteomic and metabolomic analyses assessed 184 proteins (Olink) and 250 metabolites (NMR, Nightingale) in 100 SLE patients and 27 healthy donors (HD). A comprehensive clinical profile complemented the analysis. Molecular profiles were established using clustering, differential protein expression, pathway enrichment (STRING and Metascape), univariate logistic regression, and advanced machine learning models integrated with clinical data. A validation SLE cohort (n=41) from the University College London Hospital (UCLH) was included.

Results: Unsupervised hierarchical clustering of proteomic data identified two patient groups. Forty-seven proteins were elevated in patients within Cluster-1 vs Cluster-2, while 31 proteins were upregulated in Cluster-1 relative to HD. Cluster-1 was characterised by a heightened disease severity and prevalence of cardiovascular (CV) risk factors, including prolonged disease duration, high disease activity (SLEDAI >5), prevalence of lupus nephritis (LN), hypertension, abnormal lipid profile and obesity. Differentially regulated proteins were enriched in pathways associated with nephritis, CV-risk, and immune response.

A univariate logistic regression (LR) analysis of metabolomic data between clusters identified a distinctive metabolite signature, which included elevated levels of Acetoacetate, Citrate, Creatinine, and various triglycerides. Moreover, a neural network machine learning model applied to metabolomic and clinical data classified patients into Cluster-1 and -2 with high accuracy (Area Under the Curve=0.77).

Hierarchical clustering applied to proteins enriched in both CV and nephritis pathways confirmed the distinct SLE subgroups. Furthermore, correlation analysis between regulated proteins and metabolites identified a positive relationship between creatinine and proteins within the CV [Natriuretic peptide precursor C, Mevalonate kinase, Placenta growth factor, and CD40] and nephritis [CD40 and IL17C] pathways.

Finally, comparison of patients with and without LN identified 40 upregulated metabolites, including several CV markers: low-density lipoprotein subsets, fatty acids and apolipoprotein-B. This signature was validated in an external SLE cohort of patients stratified for the presence/absence of LN (UCLH), where 35 out of 40 (87.5%) LN-associated metabolites were also detected, underscoring the robustness of our findings.

Conclusion: The present study revealed a distinct molecular signature associated with novel SLE subgroups defined by high disease activity, elevated CV risk and incidence of LN. This integrative approach, harnessing proteomic and metabolomic methodologies, could enhance molecular characterization of novel SLE patient endotypes, refine the understanding of patient clinical profiles and potentially unveil novel disease biomarkers, paving the way for improved therapeutic interventions.

Supported by the EU/EFPIA Innovative Medicines Initiative Joint Undertaking 3TR, Projects no. PI21/0591 & CD21/00187 funded by Instituto de Salud Carlos III (ISCIII) and co-funded by the European Union. Project no. RD21/0002/0033 funded by ISCIII and funded by the European Union-NextGeneration EU, via Plan de Recuperación, Transformacion y Resiliencia (PRTR) and MINECO (RYC2021-033828-I, and PID2022-141500OA-I00).

REFERENCES: NIL.

Acknowledgements: NIL.

Disclosure of Interests: None declared.