Baseline Serum Metabolites as Predictors of Teplizumab Response in Individuals with Type 1 Diabetes.

Aims/hypothesis: Immunotherapies such as Teplizumab can preserve residual beta cell function in individuals with newly diagnosed type 1 diabetes (T1D), but treatment response is variable. Currently, no biomarker exists to identify individuals most likely to benefit from immunotherapy. We believe that baseline serum metabolomic profiles can distinguish individuals who respond to treatment from nonresponders and predict therapeutic response. Methods: Baseline serum samples from 41 individuals newly diagnosed with T1D enrolled in the AbATE trial (NCT00129259) were analyzed to identify metabolic predictors of response to Teplizumab therapy in the AbATE trial. Responders to Teplizumab, as per study protocol, were defined as individuals who exhibited less than a 40% decline in baseline C-peptide levels at 2 years after start of treatment. We analyzed baseline serum samples using a semi-targeted metabolomics approach via liquid chromatography-high-resolution tandem mass spectrometry. Metabolites that were significantly different between responders and nonresponders were identified (P < 0.05), and the significant metabolites were used to train a supervised Random Forest model to predict treatment response. Model performance was evaluated using a 70/30 training/testing split, 5-fold cross-validation, bootstrap resampling (1,000 iterations), and permutation testing (1,000 permutations). Results: We identified 15 significantly different metabolites at baseline between responders and nonresponders (P < 0.05). These metabolites included amino acids and their derivatives, tricarboxylic acid (TCA) cycle intermediates, and microbially derived metabolites. At baseline, responders exhibited higher levels of TCA cycle metabolites, amino acid derivatives, and microbial metabolites, whereas nonresponders showed elevated levels of glutamate and acylcarnitines. The Random Forest classifier achieved an accuracy of 0.769 and an area under the receiver operating characteristic curve (AUC) of 0.881 in the test dataset. Cross-validation yielded a mean AUC of 0.856 (SD 0.156; 95% CI 0.719-0.992). Bootstrap analysis produced a test AUC 95% CI of 0.619-1.000, and permutation testing confirmed significance (p = 0.012). Conclusions/interpretation: Baseline serum metabolomic signatures can predict responders to Teplizumab with high accuracy. This could potentially be applicable when considering other immunotherapies in preventative efforts in T1D.Trial registration: ClinicalTrials.gov NCT00129259.