PL01.3.A Radiomic features and DNA methylation attributes in primary CNS lymphoma. (5th September 2022)

• Record Type: Journal Article
• Authors: Nenning, K; Gesperger, J; Nemc, A; Roetzer-Pejrimovsky, T; Choi, S; Preusser, M; Nam, D; Bock, C; Langs, G; Woehrer, A
• Is Part Of: Neuro-oncology. Volume 24(2022)Supplement 2
• Page Start: ii1
• Page End: ii1
• Publication Date: 2022-09-05
• Subjects: Brain Neoplasms -- Periodicals; Brain -- Tumors -- Periodicals; Brain -- Cancer -- Periodicals; Nervous system -- Cancer -- Periodicals; 616.99481
• DOI: 10.1093/neuonc/noac174.000
• Languages: English
• ISSNs: 1522-8517

Abstract

Abstract: Background: Clinical and laboratory markers have been exploited to model risk in patients with primary CNS lymphoma (PCNSL), but the derived risk models do not fully explain the observed variation in outcome. Here we present an extended framework of phenotype-epigenotype correlations that reveal novel prognostic constellations and enable prioritizing epigenetic therapy. Material and Methods: In this retrospective discovery and validation study, we leverage radiomic feature-driven analysis of medical images and supervised bioinformatic integration of DNA methylation profiles. We integrate both data modalities synergistically using machine learning-based prediction and cross-domain alignment. Ultimately, we validate the most relevant biological associations in tumor tissues and cell lines. Results: We leverage a cohort of 191 patients across 9 sites in Austria and an external validation site in South Korea, and use T1-weighted contrast-enhanced magnetic resonance imaging to derive a radiomic risk score that consists of 20 mostly textural features. We determine the risk score as strong and independent predictive factor (multivariate HR=6.56), and confirm its prognostic value in an external validation cohort. Radiomic features align with DNA methylation sites in distinct, biologically meaningful ways, and radiomic risk is predictable from selected DNA methylation sites (AUC=0.78). Ultimately, gene-regulatory differences between radiomically-defined risk groups converge on bcl6 binding activity, which is posed as testable treatment strategy in a subset of patients. Conclusion: The radiomic risk score is a robust and complementary predictor of survival and is reflected at the level of DNA methylation in PCNSL. Assessing risk and selecting epigenetic treatment based on imaging phenotypes represents a huge step forward, and the ability to define radiomic risk groups provides a concept on which to advance prognostic modeling and precision therapy for this aggressive brain cancer.