Cell-free DNA (cfDNA) methylation-based multi-cancer early detection (MCED) assay to enable subtyping of lung cancer, breast cancer, and non-Hodgkin lymphoma (NHL).

Abstract

3047 Background: Given that distinct pathological and molecular subtypes of most cancers correspond to different treatment strategies, accurate cancer subtyping is clinically critical. Currently, pathological and molecular subtyping relies primarily on tissue biopsy—a procedure that is not always clinically feasible. Additionally, tumor heterogeneity may compromise the accuracy of subtyping via this approach. Prior studies have shown that cfDNA methylation-based detection is tissue-independent and outperforms mutation-based assays in cancer subtyping. Here, we employed a previously developed targeted methylation MCED assay to further classify the subtypes of lung cancer, breast cancer, and NHL. Methods: Pretreatment blood samples from lung cancer, breast cancer, and NHL patients enrolled in a MCED study were analyzed via the targeted methylation assay. Only samples with detected circulating tumor DNA (ctDNA) were included. The training cohort comprised lung cancer (n=529: 227 adenocarcinomas, 191 squamous cell carcinomas, 111 small cell carcinomas), breast cancer (n=359: 92 triple-negative breast cancers [TNBC], 267 non-TNBC cases), and NHL (n=157: 145 B-cell lymphomas, 12 T-/NK-cell lymphomas). The validation cohort included lung cancer (n=194: 87 adenocarcinomas, 82 squamous cell carcinomas, 25 small cell carcinomas), breast cancer (n=87: 26 TNBC, 61 non-TNBC cases), and NHL (n=60: 48 B-cell lymphomas, 12 T-/NK-cell lymphomas). cfDNA methylation profiles were analyzed to further classify the subtypes of these three malignancies. Results: In the validation cohort, the lung cancer classification model exhibited high overall accuracy 91.8% (178/194), with subtype-specific accuracies of 94.3% (82/87) for adenocarcinoma, 89.0% (73/82) for squamous cell carcinoma, and 92.0% (23/25) for small cell carcinoma. The breast cancer model achieved an overall accuracy of 80.5% (70/87), including 65.4% (17/26) for TNBC and 86.9% (53/61) for non-TNBC. For NHL, the model yielded an overall accuracy of 96.7% (58/60), with 97.9% (47/48) for B-cell lymphoma and 91.7% (11/12) for T-/NK-cell lymphoma. Conclusions: The cfDNA methylation-based MCED assay facilitates accurate subtype prediction for common malignancies without the need for invasive tissue biopsy procedures. This assay achieves high accuracy in lung cancer and NHL, whereas relatively lower accuracy is observed for breast cancer, particularly TNBC, owing to its reliance on molecular subtyping. Future studies will validate this assay in larger cohorts and extend its utility to a broader spectrum of cancer types.