The detection of circulating tumor DNA allows to non-invasively retrieve tumor molecular profiles and follow disease evolution. It promises optimal and individualized management of patients with cancer. However, despite remarkable progress, several technological obstacles still limit liquid biopsy widespread application. Indeed, detecting small fractions of tumor DNA released when the tumor burden is reduced remains a challenge and detectable recurrent mutations do not cover all patients.
We aimed to assess the universal potential of DNA methylation as circulating tumor biomarker using new highly sensitive strategies to detect common cancer-specific signatures in blood. We targeted hypomethylation of LINE-1 elements, a shared feature of multiple cancers, using a multiplex PCR-based targeted bisulfite method coupled to deep sequencing, together with computational tools to accurately align sequencing data in a genome reference-free manner. We implemented machine learning-based classifiers, integrating methylation patterns at single CpG sites and at the single molecule level, to discriminate cancer from healthy plasma samples.
We detected 30-40,000 LINE-1 elements scattered throughout the genome, covering 82-125,000 CpG sites. Methylation of these LINE-1 elements, showed an extremely performant ability to discriminate between healthy and tumor plasmas from 6 different types of cancers with an area under the curve (AUC) of 0.95 (NHealthy = 123; NCancers = 393). This includes metastatic colorectal, breast, lung and uvea cancers but also non metastatic ovarian, gastric and breast cancers. These results have been validated on an independent cohort (NHealthy = 30; NCancers = 160) including metastatic colorectal, breast, gastric and lung cancers and non-metastatic ovarian cancers (AUC = 0.98).
Our method allows to dramatically increase the sensitivity of ctDNA detection in a cost-effective manner, providing an optimal trade-off between the number of targeted regions and sequencing depth. These results have important biomedical implications and should lead to the development of more efficient non-invasive diagnostic tests adapted to all types of cancers, based on the universality of these factors.
Legal entity responsible for the study
ANR-10-EQPX-03 Equipex; Laboratoire d’Excellence (LABEX) DEEP (11-LBX0044); CNRS Prematuration Program 2021; SiRIC Grant «INCa-DGOS-Inserm_12554».
All authors have declared no conflicts of interest.