3667 - Revealing the Microenvironment of Radiation-Induced Lung Fibrosis through Bulk RNA Decomposition
Presenter(s)
J. Liu1, M. T. Islam2, and L. Xing2; 1Stanford University, Palo Alto, CA, 2Department of Radiation Oncology, Stanford University, Stanford, CA
Purpose/Objective(s): Radiation-induced lung injury (RILI) is a significant complication of lung cancer radiotherapy, with 16%-28% of patients developing pulmonary fibrosis. The mechanisms driving fibrosis remain unclear, making risk prediction challenging. While single-cell RNA sequencing (scRNA-seq) can reveal cellular heterogeneity and biomarkers, its high cost limits large-scale use. In contrast, bulk RNA sequencing is more affordable but lacks cellular resolution. We hypothesize that a novel deep learning-based bulk RNA deconvolution method, incorporating gene-gene interactions, can effectively characterize the post-radiation lung microenvironment and identify biomarkers predictive of fibrosis risk. This approach enables biomarker discovery, ultimately aiding personalized treatment strategies.
Materials/Methods: We obtained public datasets comprising both bulk RNA sequencing and scRNA-seq of mouse lungs following radiation treatments of 65 Gy or 75 Gy. The data was processed using our previously developed 'genoMap' framework. Specifically, the count matrices were transformed into image representations based on gene-gene interactions, with highly interacting genes repositioned proximally in the image domain. This approach generated gene interaction patterns that were subsequently analyzed using a convolutional deep learning model. The lung tissue scRNA-seq data served as a reference, encompassing endothelial cells, fibroblasts, myofibroblasts, and immune cells. We then deconvoluted the bulk RNA sequencing samples into these cell types to obtain their respective cell fractions. Finally, the cell fractions and genetic profiles for each cell type were compared among the control, 65 Gy, and 75 Gy groups.
Results: Our preliminary findings demonstrate the application of genoMaps in visualizing bulk RNA sequencing data from mouse lungs collected post-radiation. We notice the control group of weeks two and six have the same image patterns (the minor difference is likely due to the batch effect or individual mouse heterogeneity), while the radiation group have significant different image patterns. These observed changes in gene expression patterns are directly attributable to radiation exposure and the progression of lung fibrosis. We quantified the genes significantly contributing to these altered patterns and identified fibrosis-related genes, including ADCY8, CHIL3, CSF3R, and TSPAN17.
Conclusion: The preliminary results have shown a promising path to quantitatively visualize and identify the key biomarkers leading to the radiation-induced pulmonary fibrosis, assisting in early prediction of patients at risk, which has the potential for more precise and individualized treatment strategies in oncology.