A New Tool to Decrease Interobserver Variability in Biomarker Annotation in Solid Tumor Tissue for Spatial Transcriptomic Analysis

Integrating spatial transcriptomic data with immunofluorescence image data is difficult using existing tools because of their difference in spatial resolution. Immunofluorescence informs about protein expression at cellular or subcellular level whereas spatial transcriptomic platforms usually rely on multicellular “spots” for RNA profiling. Our study coupled spatial transcriptomics of irradiated glioblastoma tissues with immunofluorescence for γH2AX, a marker of DNA damage within the nuclei of cells. We then compared gene expression in γH2AX positive and negative regions within the tissue. There was significant interobserver variability in manual annotation of γH2AX positivity in multicellular spots by 3 different researchers (Kappa statistic= 0.345), despite all of them being familiar with γH2AX immunofluorescence and having predefined imaging parameters for annotation. This variability led to different genes being nominated by different researchers, as being associated with DNA repair. To overcome this problem, we developed a new tool using MATLAB. This tool performs “spot”-wise image analysis and uses researcher defined parameters such as immunofluorescent marker intensity threshold and number of positive cells to annotate the “spots” as γH2AX positive or negative. The tissue with most variability in manual annotation was annotated reproducibly by our MATLAB tool leading to reproducible downstream analysis.