Introduction: Achieving long term glycemic control after islet transplantation depends, in part, on quantity and quality of used islets.  This requires achieving an optimal balance of enzymatic digestion to free islets from surrounding acinar tissue.  Underexposure of digestive enzymes to the pancreas results in trapped or embedded islets, whereas excess digestion can lead to islet fragmentation, injury, and poor islet survival. 

Methods: We have developed a machine learning model to define the most suitable levels of digestive enzyme exposure using results from function assays, gene signature, and isolation parameters to maximize islet yield and in vivo function.  To predict islet yield, the model was trained based on transplant outcome, where the threshold for successful islet isolation was defined as 300,000 IEQ.  Independent parameters were comprised of donor/organ characteristics including BMI, age, HbA1c, cold ischemia time (CI), pancreas weight (PW), and enzyme exposure parameters included amount and duration of Wunsch enzyme used.  To predict in vivo islet function, stabilized blood glucose (BG) averages from NOD/skid mice transplanted with donor islets, within 7 to 30 days post-transplant using the integral area under the curve (AUC) were used as a training set.  Independent variables included isolation parameters used for yield prediction, as well as functional assay results from oxygen consumption rate (OCR)¹ and genetic expression markers shown to be reflective of islet function².

Results: From 246 islet isolations the algorithm correctly identified unsuccessful digests with a hold-out accuracy of 89%, and islet function by predicting when a transplant would fail to reverse diabetes in mice in 84% of cases.  Cluster analysis identified 3 classifications of islet function based on similarity in donor and digest characteristics, OCR, and gene expression.  Group 1 (HbA1c= 5.4% ±0.53, purity = 80% ±6.99%, CI = 395 ±116 minutes) reversed diabetes in 10.6 ±5.5 days with a stabilized BG of 183.6 mg/dl ±125.0; group 2 (HbA1c= 6.6% ±2.17, purity = 77% ±7.47%, CI = 431 ±100 minutes) showed a maximum response at 13.5 ±6.0 days, where the most stabilized BG averaged 404.9 mg/dl ±176.0, and group 3 (HbA1c= 5.3% ±0.32, purity = 81% ±6.48%, CI = 491 ±152 minutes) showed a maximum response at 10.9 ±6.2 days, where the most stabilized BG averaged 300.5 ±163.9 mg/dl.  Group 1 had lower Wunsch exposure, PW, and HbA1c (p=*, **, ***, respectively), and had higher OCR and gene expression for the functional islet gene marker MNX1² (p=***, ***, respectively) than group 2.  Group 3 also had levels indicative of good islet function compared to group 2 (HbA1c, MNX1, OCR p=***, ***, ***, respectively), but may have been susceptible to poor islet function due to high Wunsch amount, CI, and PW (**, **, ***, respectively). 

Conclusion: The model can make inferences on the most suitable levels of digestive enzyme exposure to maximize islet yield and in vivo function.