Discrimination of Metastasis from Breast and Pancreatic Cancer by MALDI Imaging

Introduction: The therapy of metastatic cancer in organs such as liver and of the tissue of cancer origin (the primary cancer) relies on the accurate classification of the tumor. The identification of the primary site of a metastasis is currently based on clinical information, morphology, immunohistochemistry and may include DNA or RNA analysis. This process is complex and expensive and may not always lead to unambiguous results. The purpose of this study is to test if MALDI imaging allows the differentiation of pancreatic vs. breast cancer, and if the resulting classifiers are also applicable to identify the tumor of origin in liver metastases and to identify potential marker proteins. Methods: All tissues samples were formalin fixed paraffin embedded (FFPE). 96 breast and 92 pancreatic biopsies in a tissue microarray from different patients were included. Additionally, six liver metastases originating from breast or pancreas carcinomas were used for proteomic classification. All samples were subjected to de-paraffinization and antigen retrieval followed by application of trypsin by spray deposition and incubation at 37 ºC for 2 hours. Matrix was deposited using the same sprayer and the tissue was analyzed by MALDI-TOF imaging. Multivariate statistical analyses were applied to detect tryptic peptides as putative classification markers. The peptides were extracted with water and identified using off-line nanoLC-MALDI TOF/TOF fragmentation and Mascot. IDed peptides were matched with the image data to determine the protein localization. Preliminary Data: Cancer related spectra from 29 breast and 32 pancreas tumor samples were used as training set to build a classification model using a support vector machine algorithm (SVM). An SVM classifier based on 17 peptide signals was able to discriminate breast from pancreas carcinomas with high recognition capability (100%) and an overall cross validation of 99.21%. This classifier was applied to two test sets including 67 breast and 60 pancreas primary carcinoma patients. The SVM model classified breast and pancreas carcinoma with an overall accuracy of 83.38%, a sensitivity of 85.95% and a specificity of 76.96%. We applied our model to classify three breast and three pancreatic individual liver metastasis samples. Classification results were concordant to the histopathological diagnosis made by pathologists. The result shows that in fact the proteomics pattern in liver cancer samples varies statistically significant depending of the respective primary tumor so that molecular features detected by MALDI imaging can make sensitive and accurate predictions about the primary tumor site. A number of peptides/proteins were identified on classifier specific peaks. Identified proteins include heat shock protein beta-1(HSPB 27), heterogeneous nuclear ribonucleoprotein A1 (ROA1), heterogeneous nuclear ribonucleoproteins A2/B1 (ROA2), filamin A and SH3 domain-binding glutamic acid-rich-like protein (SH3L1). Novel Aspect: Classification of tumor of origin in metastatic cancer by MALDI imaging