Abstract 3679: Plasma-based multimodal profiling of urothelial carcinoma to characterize tissue-based subtypes

Abstract Background Urothelial carcinoma (UC) is a molecularly and clinically heterogeneous malignancy with notable diversity in patient outcomes and response to treatment such as immune checkpoint blockade. There have been various efforts to identify and enrich for patient populations which benefit most from these immunotherapies. Cell-free DNA (cfDNA) can be utilized as a non-invasive method to detect the extent of residual tumor, and also provide molecular characterization through indirect measures of gene expression. Here, we use a multimodal plasma-based platform on samples from the IMvigor130 trial to explore biomarkers associated with tissue-based UC subtypes, which have been previously shown to have differing responses to checkpoint blockade. Methods Plasma samples from baseline metastatic and locally advanced UC patients with known tissue-defined subtypes (n=120) were evaluated with a targeted sequencing approach that characterizes chromatin accessibility and cfDNA methylation. Fragment length and positioning around each transcription start site were used to indirectly infer gene expression. We also identified UC subtype-associated CpG hypermethylation patterns. We trained binary and multinomial regression models utilizing the individual plasma analytes as well as the combined multimodal features to predict the tissue-based subtypes. Results The luminal marker KRT20 and basal marker CXCL6 were previously observed to be subtype markers in tissue data; here we show similar associations with subtype using cfDNA inferred gene expression. An analysis of differentially methylated promoters revealed many other subtype-specific genes that show potential in plasma cfDNA for UC classification. We observed concordance in predicted gene activity from both inferred gene expression and methylation for the majority of genes (>75%). We compared these signals to subtypes previously identified by tissue RNA expression, which featured luminal, stromal, immune, or basal characteristics. The classes with the strongest classification performance were luminal, with an AUC of 0.66, and basal, with an AUC of 0.72. These groups are characterized by tumor-intrinsic features, as opposed to the stromal and immune-driven subtypes. Notably, the basal subtype is known to be enriched for responders to PD-L1 inhibition. Conclusions Here, we show a proof-of-concept for how a non-invasive multimodal cfDNA assay can be used to interrogate and recapitulate subtypes of UC. Our findings are aligned to the known biology of tissue transcriptomics-defined subtypes and reveal additional insight into potential plasma biomarkers. Finally, the capacity to subtype UC from a blood-based multiomics approach can be leveraged for the longitudinal monitoring of subtype and disease characteristics over the course of treatment. Citation Format: Austin Cauwels, Eric Levy, Alison D. Tang, Victoria Cheung, Rebecca Gupte, Alexander Tseng, Tao Qing, Kimberly Walter, Zoe June Assaf, Sanjeev Mariathasan, Romain Banchereau, Ehsan Tabari, Jimmy C. Lin. Plasma-based multimodal profiling of urothelial carcinoma to characterize tissue-based subtypes [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 3679.