O18 Data-driven clustering of cerebrospinal fluid proteome reflects clinical phenotypes of systemic lupus erythematosus

Objective Neuropsychiatric (NP) symptoms are frequent in patients with systemic lupus erythematosus (SLE) and signs of neuronal damage can be present in patients without evident NP involvement. The cerebrospinal fluid (CSF) protein patterns may reveal insights to the pathogenesis of NPSLE. We applied a data-driven approach to investigate the clinical differences in patients with SLE, clustered by their CSF proteomic profile. In addition, we explored the association between groups of proteins and clinical and serological data. Methods CSF samples from a cross sectional cohort of 29 female outpatients recruited irrespectively of disease activity and organ involvement, were analyzed using label-free quantification liquid chromatography tandem mass spectrometry. Hierarchical clustering of proteomic data was used to identify sample clusters and clusters were analyzed for variance of clinical traits using Kruskal-Wallis and Wilcoxon tests. Proteins were clustered in modules using Weighted Gene Co-expression Correlation Network Analysis (WGCNA). Protein modules were analyzed for correlation to clinical traits using Pearson correlation coefficient. Results Patient cluster 1 showed highest frequency of nephritis, depression and cognitive impairment. Cluster 2 was characterized by alopecia, SSA-antibodies, and low frequency of cognitive impairment. Cluster 3 had a clinical profile of autonomic neuropathy, lupus headache and increased neurofilament light concentrations in CSF. The protein modules (M1-M6) were characterized by nervous tissue proteins (M1), lipid lifecycle proteins (M2), macrophage derived proteins (M3), plasma proteins (M4), immunoglobulins (M5), intracellular metabolic proteins (M6). Modules 1 and 2 were associated with nephritis, depression, longer disease duration and cognitive impairment, and this pattern was most pronounced in patient cluster 1. The opposite clinical profile was associated with M4 and M5, which showed inverse correlation to cognitive impairment and brain atrophy and was most distinct in patient cluster 2. Conclusion Data-driven clustering of patients using their CSF proteome forms subgroups reflecting clinical phenotypes of SLE. Two clinical phenotypes appear, where age of disease onset, level of disease severity, renal involvement and degree of neuronal damage differentiates the phenotypes. Variances in CSF proteomic patterns may represent differences in the SLE disease process. Acknowledgements Medical Faculty of Lund University, Skåne University Hospital Research Funding, Anna-Greta Crafoord Foundation, Greta and Johan Kock Foundation, King Gustav V 80th Birthday Foundation, Magnus Bergwall foundation, Alfred Österlund Foundation, Swedish Research Council, and Swedish Rheumatism Association