Testing AI predictions that hyperglycemia-induced non-enzymatic glycation of immune system proteins impairs diabetic immunity

Abstract

People with type 1, type 2 and gestational diabetes, and newborns of diabetic mothers, have increased risks of contracting microbial infections such as SARS-CoV-2, amplifying comorbidity and mortality risks. Hyperglycemia and its associated hyperinflammation mediate these risks but by unknown mechanisms. We hypothesize that hyperglycemia results in glycation (non-enzymatic addition of sugars to free amino groups such as lysines) of antimicrobial and anti-inflammatory proteins and peptides that mediate innate and adaptive immunity. While glycation of hemoglobin A1c is used as a benchmark for glycemic control, studies of the effects of the glycation of immune system molecules are almost non-existent. Three AI protein-glycation prediction models demonstrate that immune system proteins and peptides such as serotransferrin, lactotransferrin, lysozyme, complement proteins, immunoglobulins, Toll-like receptors, etc., probably glycate at multiple sites affecting function. However, the AI-predicted glycation sites differed significantly among the three models employed. Therefore, three tests of these predictions were carried out. Predicted glycation sites were first evaluated in terms of clinically or experimentally identified glycation sites of homologous proteins and known effects of lysine substitutions at predicted glycation sites. Second, mass spectrometry was used to identify glycation sites on lactoferrin and human lysozyme. Third, the effects on the antimicrobial activity of lactotransferrin and lysozyme glycation were tested using a Candida albicans assay. Results of tests demonstrate that protein glycation occurs and interferes with immune function but that AI glycation prediction models are very inaccurate. Suggestions for improving AI accuracy are provided.