Abstract The levels of circulating tumour DNA (ctDNA) in cancer patients are associated with response and survival outcomes. Analysis of ctDNA generally requires knowledge of tumour mutations or the application of expensive next generation targeted sequencing. Here we utilize a cutting-edge approach, combining cell-free DNA molecules (cfDNA) fragmentomics and machine learning, to infer the presence of ctDNA to predict response and survival to immune checkpoint inhibitors in stage III unresectable and IV melanoma patients. Genome libraries were constructed from cfDNA derived from 48 (discovery cohort) and 96 (validation cohort) stage III unresectable and IV melanoma patients prior to the commencement of immune-checkpoint inhibitors. The discovery cohort included any-line therapy, patients had a median age of 66, 69% were male, 33% were BRAF V600 mutant, and 88% had an ECOG of 0-1. The validation cohort included first-line therapy, patients had a median age of 69, 59% were male, 22% were BRAF V600 mutant, and 89% had an ECOG of 0-1. Genome libraries were subjected to shallow whole genome sequencing (8x) and the following metrics were calculated: tumour content prediction, aneuploidy levels, various proportions of fragments ranging from 20 bp to 320 bp, the proportion of mitochondrial DNA, and the amplitude of 10 bp size oscillations between 75-150 bp, and the Shannon diversity of trinucleotide motifs at each end of the read. Non-penalized logistic modelling was used to predict clinical benefit, and the resulting logistic model was transformed into a “white-box” model. The white-box model was applied to the cohorts and used to test progression-free and overall survival. The removal of reads with inserts <90 and >150 bp greatly improved chromosome aneuploidy detection from 19% to 31%. Logistic modelling of the fragmentomic features of cfDNA yielded an AUROC of 0.83 for prediction of clinical benefit in the discovery cohort. A cutoff was set for the fragmentomic model score and used to stratify groups for survival analysis. Patients with a low score showed significantly worse progression- free (mPFS 931 vs 286 days, HR 3.4, 95% CI 1.55-7.46, p<0.002) and overall (mOS NR vs 363 days, HR 8.1, 95% CI 2.7-24.3, p<0.001) survival. These results were replicated in the validation cohort with an AUROC of 0.78 for prediction of clinical benefit and significant associations with progression-free (mPFS 1986 vs 926 days, HR 2.33 95% CI 1.27-3.92, p=0.005) and overall (mOS NR vs 1238 days, HR 4.51 95% CI 1.7-11.96, p=0.002) survival. Fragmentomic features of plasma cfDNA enable the prediction of clinical benefit, progression-free, and overall survival in stage IV melanoma patients treated with immune-checkpoint inhibitors. Citation Format: Aaron B Beasley, Leslie Calapre, Ashleigh Stewart, Anna L Reid, Russell Diefenbach, Wei Yen Chan, Rebecca Auzins, Luisa Pinnel, Sanjeev Adhikari, Muhammad A Khattak, Peter Lau, Tarek M Meniawy, Jenny H Lee, Lydia Warburton, Michael Millward, Alexander M Menzies, Richard A Scolyer, Georgina V Long, Helen Rizos, Elin S Gray. Fragmentomic features of cell-free DNA predict late-stage melanoma treatment benefit and survival [abstract]. In: Proceedings of the AACR Special Conference: Liquid Biopsy: From Discovery to Clinical Implementation; 2024 Nov 13-16; San Diego, CA. Philadelphia (PA): AACR; Clin Cancer Res 2024;30(21_Suppl):Abstract nr B044.
Read full abstract