Abstract
Introduction The heterogeneous clinical presentation of SLE is characterised by the unpredictable appearance of flares and remissions of disease activity associated with organ damage and severe symptomatology. Various attempts to classify lupus clinically have not been successful, still burdened by delayed diagnosis and clinical trial failures. Our aim was to develop and validate a robust method to reproducibly stratify patients with lupus according to longitudinal patterns of disease presentation and gene expression data obtained at several points in time.
Methods We calculated correlations among expression values of each gene and SLEDAI across the different time points for each patient. With these, we constructed a bi-dimensional inter-patient matrix. We developed a new approach to select genes strongly correlated with SLEDAI in absolute values across all patients as best genes to stratify patients and filter out the remaining. Finally, we obtain the stratification groups applying consensus clustering that estimates the probability of a patient to belong to a given cluster by random seed permutation.
Results Longitud inally, lupus patients group into three clusters. The three clusters shared the same mean SLEDAI and had no differences in the clinical parameters comprising the score. Functionally however, the clusters had clearly differentiated gene expression profiles and cellular profiles representing three different mechanisms of disease progression. We tested the stability of the clusters by different validation methods and obtained a high reproducibility and robustness. Our stratification method could be used in the future to establish and re-design lupus clinical trials and treatment, and may be used in any disease with measurable but variable patterns of disease progression. This work has received support from the EU/EFPIA/Innovative Medicines Initiative Joint Undertaking PRECISESADS grant n°1 15 565.
