Abstract 4707: Quantifying cellular composition using laser microdissection regions of interest via automated digital image analysis of co-registered tissue thin sections

Low purity tumors often portend poor disease outcome. Understanding the compositional heterogeneity of complex tumors, such as high-grade serous ovarian cancers (HGSOC) is an area of intense study. Laser microdissection affords decoupled molecular profiling analyses of tumor and non-tumor cells not otherwise possible using routine methods. Automated documentation of LMD collection activities are needed to support the development of high stringency quality control and quality assurance (QA/QC) procedures. This study was conducted to demonstrate that histology combined with digital image analysis can be utilized to quantify tumor and stroma cell content within LMD regions of interest (ROI). A single frozen high grade serous ovarian cancer tissue specimen was acquired under an IRB-approved protocol and thin sectioned onto charged glass or polyethylene naphthalate (PEN) membrane slides followed by staining with hematoxylin and eosin H&E. Tumor and stromal cell populations were harvested from membrane slides in a locoregional manner by LMD (LMD7, Leica) at ~150 µm intervals. Digital whole slide images (WSI) (Aperio ScanScope XT slide scanner, Leica) of HGSOC tissue were provided for analysis consisting of matched reference H&E sections and serial sections following enrichment of tumor (n=15) or stromal cell populations (n=3). Image analysis was performed by OracleBio using the Indica Labs HALO™ platform. Classification algorithms were developed and applied to each LMD image to identify the ‘Dissection Area’ and ‘All Remaining Tissue’ as two separate ROIs. A separate cytonuclear algorithm was developed using the H&E stained sections to detect cell nuclei. Each LMD image was co-registered with their respective H&E image to overlay classified ROI before analysis for nuclei detection. Dissected ROIs revealed median tumor cell areas were 27.5% and median stroma cell areas were ~50% of the total tissue area observable across sections. These were consistent with histopathological estimates of tumor (~36%, R2=0.91) and stromal (~52%, R2=0.96) cellularity across sections. Data generated from co-registered post-LMD dissected Area ROI within H&E sections serial to the LMD tissues highlighted that the median number of nuclei per sample within tumor (n=15) and stroma (n=3) ROI was 7,527 ± 665 and 3,470 ± 476 per mm2 respectively. The average size of nuclei detected within tumor and stroma ROI was 24.2 ± 1 and 21.4 ± 0.5µm2 respectively; a difference of ~12% between these ce