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

We analyzed baseline serum samples from 41 individuals newly diagnosed with type 1 diabetes (T1D) enrolled in the Autoimmunity-blocking Antibody for Tolerance (AbATE) trial (NCT00129259) to identify metabolic predictors of teplizumab response. Responders to teplizumab were defined as individuals who exhibited <45% decline in baseline C-peptide levels at 2 years after start of treatment. We used a semitargeted metabolomics approach via liquid chromatography-high-resolution tandem mass spectrometry. We identified 15 significant (P < 0.05) metabolites, including amino acids and their derivatives, tricarboxylic acid (TCA) cycle intermediates, and microbially derived metabolites. Responders exhibited higher levels of TCA cycle metabolites, amino acid derivatives, and microbial metabolites, whereas nonresponders showed elevated glutamate and acylcarnitines. These metabolites were used to train a supervised random forest (RF) model to predict treatment response. Model performance was evaluated using a 70:30 training-to-testing split, fivefold cross-validation, bootstrap resampling (1,000 iterations), and permutation testing (1,000 permutations). The RF classifier achieved an accuracy of 0.769 and an area under the receiver operating characteristic curve of 0.881 in the test data set. These findings suggest baseline serum metabolomic signatures have the potential to predict responders to teplizumab with accuracy. This could potentially be applicable to other immunotherapies in T1D preventative efforts. Further validation of our findings is needed.

We believe that baseline serum metabolomic signatures can predict response to immunotherapies to identify which individuals will benefit most from treatment. We aimed to identify whether baseline serum metabolomic profiles can distinguish responders from nonresponders to teplizumab. Fifteen metabolites were identified as significantly different. Our findings suggest that baseline serum metabolomic signatures could be used to predict which newly diagnosed type 1 diabetes patients will respond to teplizumab, enabling more personalized treatment decisions.