Clinical limit of detection may be a reasonable metric for evaluating cell-free DNA for multi-cancer early detection.
Clinical limit of detection (cLOD) may be a reasonable metric for evaluating cell-free DNA (cfDNA) for multi-cancer early detection, according to recent data from the Circulating Cell-Free Genome Atlas (CCGA) study (NCT02889978).
In particular, a whole-genome methylation assay performed better than whole-genome sequencing or targeted sequencing without the need for additional white blood cell sequencing for clonal hematopoiesis.
Data from the study, which evaluated cell-free DNA multi-omics in prototype cfDNA-based multi-cancer early detection tests, were presented at the 2021 ESMO Congress.
In the study, researchers collected plasma and matched white blood cells before sequencing them from CCGA study participants (n = 2,800). Tumor biopsies were also sequenced when available.
Furthermore, 6 cfDNA-omics were used to detect circulating tumor fraction, including whole-genome methylation data from whole-genome bisulfite sequencing (30x); small somatic variant data from error-corrected targeted sequencing ( 60,000x); somatic copy-number aberration; fragment length; fragment endpoint; and allelic imbalance data from whole-genome sequencing (30x).
Samples were then split into independent training and validation sets, while 10 classifiers were trained to identify solid cancer (carcinomas, sarcomas, lymphomas) using 1 per-omic, 2 corrected for clonal hematopoiesis using germline DNA from paired white blood cell sequencing, 1 pan-omics and 1 clinical data only. These metrics were assessed for cancer detection and cLOD, which was estimated as the probability of detecting cancer as a function of circulating tumor fraction using matched tumor biopsies. Researchers also trained three additional classifiers to predict cancer signal origin.
Of the 2,800 participants, 2,261 had analyzable results, of whom 1,414 were in the training set and 847 were in the validation set. Because training and validation results were similar, results from the validation set are featured in the abstract.
Researchers found that circulating tumor fraction accounted for greater than 72% of the variance in cancer detection scores, while the cLOD was at least 1.5-fold lower for whole-genome methylation than any whole-genome sequencing or targeted sequencing classifier. Cancer signal origin prediction was at least 1.8-fold more accurate using whole-genome methylation than targeted sequencing or somatic copy-number aberration.
“These data informed the design of a significantly improved targeted methylation MCED test for further CCGA substudies to support clinical use,” the researchers wrote in the abstract.