Discovery of prognostic and treatment predictive biomarkers in recently diagnosed type 1 diabetes patients treated with anti TNF-alpha

ObjectiveThe T1GER study showed that treatment with the TNF inhibitor golimumab in recently diagnosed type 1 diabetes patients showed better preservation of endogenous insulin production than placebo. However, considerable variation was observed among subjects. Therefore, a range of biomarkers were investigated for their potential to predict treatment response to golimumab. Research Design and MethodsBaseline blood samples from 79 subjects were tested for autoantibodies, microRNA, metabolites, lipids, inflammatory proteins, and clinical chemistry. Univariate analysis was used to identify biomarkers that correlated with C-peptide change. Multivariate analysis was performed to establish a biomarker algorithm predicting the C-peptide response during the study. ResultsMultivariate analysis showed that baseline metabolites and miRNAs best predicted C- peptide responses both for placebo and treatment arms. Lipids, and inflammatory proteins were moderately predictive, whereas autoantibodies and clinical chemistry showed little predictive value. An optimal model combining selected clinical variables and metabolites showed a correlation between predicted and observed C-peptide responses for the overall study up to 52 weeks, with an R2 of 0.85. An LOOCV model was developed as a surrogate validation test, resulting in an R2 of 0.69 overall, and an R2 of 0.76 specifically predicting C-peptide responses at week 38. ConclusionsThe exploratory analysis of the T1GER study resulted in a set of baseline biomarkers with promising performance in predicting future C-peptide responses during the study. If validated in independent cohorts, these prognostic and predictive biomarkers and algorithm carry significant translational impacts that can assist clinicians in making treatment decisions.