Methylation Is Best for cfDNA Multicancer Early Detection Tests

Cell-free DNA (cfDNA) present in circulating blood can be used for the early detection of cancer across multiple tumor types. One multicancer early detection (MCED) test is commercially available in the United States, and many are currently under development. Each of these tests uses different information from cfDNA, but no rigorous or systematic analysis comparing these tests has been performed to date. To determine the most promising cfDNA MCED test with a low false-positive rate and sufficient sensitivity for cancer signal detection, Jamshidi and colleagues conducted the first substudy of the Circulating Cell-free Genome Atlas (CCGA) in which next-generation sequencing, coupled with machine learning, was used to detect cancer signal from patient data. Ten different machine learning classifiers using different cfDNA features were trained on the same samples and independently validated. When evaluated at high (98%) specificity, the top classifiers for cancer signal detection were trained on data from whole-genome methylation (WG methylation), data from single-nucleotide variant with paired white blood cell background removal (SNV-WBC), and the pan-feature classifier. Evaluation of classifier performance across methods based on circulating tumor allele fraction (cTAF), which is strongly associated with cancer signal detection, indicated that the clinical limit of detection (LOD) was lowest in WG methylation, SNV-WBC, and pan-feature classifiers. Of the three top-performing classifiers, WG methylation had the most accurate prediction of cancer signal origin (CSO). Moreover, cTAF variation was observed between tumor type and clinical stage, which supports both cTAF as a predictor of classifier performance and the use of clinical LOD for assessing cancer signal detection. Lastly, WG methylation was selected for further development, as it was among the most sensitive, had leading CSO accuracy, did not require WBC sequencing, and could be further improved by targeted sequencing. Using a targeted methylation approach, clinical LOD improved by almost an order of magnitude over the original top classifiers. In summary, this CCGA substudy provided a comparison of cfDNA-based detection methods, revealing methylation as the most promising approach, and also defined the cTAF-based clinical LOD metric as broadly applicable to compare cfDNA-based tests.

Jamshidi A, Liu MC, Klein EA, Venn O, Hubbell E, Beausang JF, et al. Evaluation of cell-free DNA approaches for multi-cancer early detection. Cancer Cell 2022;40:1537–49.e12.

Note:Research Watch is written by Cancer Discovery editorial staff. Readers are encouraged to consult the original articles for full details. For more Research Watch, visit Cancer Discovery online at http://cancerdiscovery.aacrjournals.org/CDNews.