Dipg-66. Multiplexed Immunofluorescence Mapping and AI Model Prediction of Tumor Microenvironment in Diffuse Midline Gliomas

Abstract BACKGROUND Diffuse midline glioma (DMG) is a fatal childhood brain cancer with a survival rate of less than one year from diagnosis. Pharmacological approaches and immunotherapy have failed to make a clinical impact. Incomplete understanding of the tumor microenvironment (TME) and tumor associated antigens have contributed to the observed poor prognosis. Therefore, mapping TME is urgently needed. METHODS Whole brains were collected at autopsy from 80 pediatric subjects, including patients diagnosed with DMG (n=50), GBM (n=9), non-malignant controls (n=10), ependymoma (n=5), ATRT (n=2). Up to four brain anatomical sites were selected: primary tumor, metastatic and adjacent healthy control sites, and processed for staining with hematoxylin and eosin (H&E). After reviewing the tumor and healthy regions, three punch cores were obtained resulting in a total of 918 core punches and two TMA blocks. Multiplexed immunofluorescence (MxIF) technology was used to probe 43 biomarkers on a single TMA slide, focusing on immune cell types and activation thereof. Clinical data including genomics alterations were obtained from each patient for downstream analyses. RESULTS Highest expressed immune markers across all patients were CD8 and CD3 (T-cells), CD68, Iba1 and CD163 (microglia). In DMGs, T-cells were mostly detected and increased in ONC201 and immunotherapy treated patients compared to only immunotherapy treated patients. Analysis of Iba1 confirmed a higher difference in activation in primary tumor compared to metastatic and adjacent healthy tissue, whereas in contrast, CD68 was significantly increased in metastatic sites. Patient plasma and RNA transcriptome analysis confirmed biomarkers and immune related pathways. Further, immune competent murine models were used to validate immune cell infiltration. AI deep learning technology modeling is ongoing now to, in future, be able to predict cell type composition and neighborhoods mapping. CONCLUSION We posit that this comprehensive data and characterization of the DMG TME will help to identify biology-informed targeted treatments.