Abstract P3-05-06: Automated analysis of Her2 FISH using combined Immunofluorescence and FISH signals

Abstract

Abstract Introduction: Qualifying patients for Her2 targeted therapy is currently done by detecting Her2 protein overexpression or gene amplification using immunohistochemistry and/or FISH. We have recently developed a method for detecting both signals on the same tissue section allowing direct correlation of protein expression and gene copy number on a cell by cell basis. Accurate assessment of Her2 gene copy number is critical and can pose a challenge due to tumor heterogeneity. This paper reports the accuracy of a proprietary FISH dot counting algorithm on a cell-by-cell basis, potentially allowing analysis of thousands instead of dozens of tumor cells. Method: Automatic FISH signal counts were compared to manual counts of 888 cells selected from 19 invasive ductal breast carcinoma samples exhibiting varying degrees of Her2 expression collected between June 2011 and March 2012. Tissue sections (4 µm) were mounted on positively charged slides, baked and processed through deparaffinization, rehydration and antigen retrieval, then stained for immunofluorescence (IF) using Cy5 labeled Her2 and Cy3 labeled cytokeratin antibodies, counterstained with DAPI, and imaged using InCell 2000 analyzer with GE-proprietary acquisition and processing software. Images were collected at 10x magnification and digitally stitched to span the entire tissue section. A pathologist then selected separate tumor and adjacent normal epithelium regions for subsequent imaging at 40x magnification. Slides were subsequently processed for FISH by pepsin digestion and then subjected to FISH by using PathVysion kit (Abbott Molecular, Des Plaines, IL). After hybridization and subsequent high stringency washes, samples were DAPI stained and mounted for microscopy. Samples were imaged at 40x at the same regions recorded for 40x IF acquisition, using filtersets appropriate for FISH fluorophores and DAPI. A proprietary automated processing algorithm was used to analyze combined IF and FISH signals and derive case specific Her2 score from the tumor and/or adjacent normal epithelium. Cell-level dot counting accuracy was assessed using two metrics comparing automated counts to manual counts: cell classification agreement, where a normal cell was defined as having 3 or less Her2 and Cep17 dots; and dot-counting match, where a difference of more than 20% in absolute counts was considered an error. Result: Our automatic results gave an overall cell-by-cell classification agreement of 88% (range 71% to 98% by case). Combining classification agreement and counting match, our algorithm gave an overall accuracy of 81% (range 63% to 97% by case). Restricting to tumor tissues (as judged by pathologist review of IF) classification agreement and accuracy were 84% and 72%, respectively. Conclusion: The observed variability in algorithm performance between the different cases was due to the fact that error root causes were case dependent. For instance, the main cause of over-counting errors was image noise and artifacts. On the other hand, the main cause of under-counting was low image contrast, especially in highly amplified cases. These results are an early indication of the promise of automatic dot counting applied to breast cancer slides multiplexed for Her2 IF and FISH. Citation Information: Cancer Res 2012;72(24 Suppl):Abstract nr P3-05-06.