Development of a class prediction model that discriminates pancreatic adenocarcinoma from neuroendocrine tumors

Background: Pancreatic neuroendocrine tumors (pNETs) increased considerably in the last two decades, with about 60-70% of patients having a metastatic disease at diagnosis. Although the vast majority of pancreatic cancer cases are pancreatic adenocarcinoma (pADC), pNETs make up 3% to 5% of all pancreas tumors. Both entities can be distinguished by immunohistochemistry. We would like to show that discrimination of neuroendocrine tumors and pancreatic carcinoma can be easily done by MALDI imaging mass spectrometry (IMS). Methods: We investigated pNET (n= 174) and pADC (N=78) needle core biopsies from individual patients with primary tumor. Sections (5 μm thick) were on-tissue trypsin digested and sprayed with an organic matrix solution using an automatic vibrational sprayer devise. Data were acquired using a MALDI-TOF/TOF at a spatial resolution of 150 µm. The dataset was subsequently preprocessed and loaded into SCiLS Lab software (SCiLS, Bremen) for classification analysis. Following the MALDI measurement, sections were stained with hematoxylin & eosin and examined by a pathologist to confirm the presence of intact neoplasm, and to correlate MALDI data with the histological features in the same section. Results: Linear Discriminant Analysis was used to generated a classification model that could stratify pADC (n= 35) and pNET (n= 92) patients with a cross validation accuracy of 99.46%. The model was validated with an indipendent testing set including thirty-four pNET and forty-three pADC patients. All pNET patients were classified correctly and 37/43 (86%) of the pADC patients were predicted correctly. Conclusions: Our findings demonstrated that MALDI IMS may provide a reliable and practical method in discriminating pNETs from pancreatic ADC.