Mapping Tumor Microenvironment and Treatment Response of Diffuse Midline Glioma Using Multiplexed Immunofluorescence and AI Models

Background

Despite its clinical promise in non-solid tumor, immunotherapy is yet to show significant clinical efficacy for brain tumors including pediatric diffuse midline glioma (DMG). This indicated the need to fully explore DMG immune tumor microenvironment (TME).

Method

Whole brains (49 DMGs, 20 non-DMG, 10 non-malignant) from 79 pediatric patients were used to establish a tissue microarray (918 cores) representing primary, metastatic, and adjacent healthy sites. CellDIVE MxIF multiplex assay was used to probe for 33 immune and cell type markers. RNA sequencing (n=62 patients) defined additional immune signatures. Findings were validated using patient plasma and DMG PDX models. Our annotated single-cell atlas was used to train a spatial AI model to predict antigens from H&E staining.

Findings

We found enrichment of M1-activated microglia in primary versus adjacent healthy tissue. PD1 positive cells were significantly (p<0.01) higher in tumor compared to adjacent controls. This was validated by mRNA profiling, further indicating two distinct groups with top 35 significant (p<0.05) genes revealing synaptic signature in the metastatic cohort.

We stratified the patient cohort by treatment. Imipridone cohort (n=5) showed decreased progenitor (Nestin+, Vimentin+, and SOX2+) and increased macrophages/microglia infiltration. Increased T and B cells was validated in patient plasma following imipridone therapy. Combination therapy of imipridone and immunotherapy (n=7) resulted in increased myeloid (Iba1, CD68, CD163) and lymphoid (CD3, CD8) cells. Enhanced immune engagement was validated in DMG PDX models. Machine learning resulted in a spatial AI model capable of predicting 22 antigens using H&E slides.

Interpretations

DMG tumors maintain a cold immune microenvironment, which is nevertheless dynamic and responsive to therapy, indicating the need to explore combination therapies. AI-assisted antigen detection is suitable for rapid interpretation of clinical biospecimens.

Funding

This work was supported by Rising Tide, SNF, LilaBean Foundation, Swifty Foundation, Swiss to Cure DIPG and Yuvaan Tiwari Foundation.