cfDNA Methylation Is the Best Biomarker for Predicting the Source of Cancer Signals

Abnormal DNA methylation is one of the important epigenetic modifications that drive the occurrence and development of cancer. Almost all human tumors contain abnormal methylation of tumor related genes. Therefore, abnormal DNA methylation is a warning biomarker for the occurrence and progression of cancer. When cancer cells and other human cells apoptosis, their DNA enters the bloodstream, which is cfDNA.

Recently, an article published in Cancer Cell?compared the clinical research results of 10 sequencing analysis techniques targeting cfDNA in early screening of tumors, and showed that cfDNA methylation is the best biomarker for predicting the source of cancer signals.

The title of the article is:?Evaluation of cell-free DNA approaches for multi-cancer early detection。

The research was conducted and published by Prof. Arash Jamshidi's group at GRAIL, a US biotech company.

This is also the largest genome-wide comprehensive comparison of cfDNA methods currently available. The study evaluated several methods for multi-cancer early detection (MCED) testing based on cfDNA, and defined clinical limit of detection (LOD) based on circulating tumor allele fraction (cTAF) to achieve performance comparison.

The DNA of various tissues in the human body is present in the blood and other extracellular fluids, known as cfDNA. This discovery has also enabled the development of many blood based cfDNA tests to detect genomic abnormalities. Chromosome copy number aberrations are used for non-invasive prenatal testing or operable tumor source mutations are used for targeted cancer treatment selection. Early tumor screening is also a key application direction based on blood based cfDNA.

This study evaluated multiple potential blood based multi cancer early screening methods in a cohort of 2800 individuals (1628 cancer patients and 1172 non cancer patients) and randomly assigned them to independent training or validation sets. This includes whole genome sequencing, whole genome methylation sequencing, and targeted sequencing, including methylation, somatic copy number changes, somatic mutations, and a clinical feature classifier.

The evaluation criteria include sensitivity (correctly identifying tumor patients), specificity (defined as 98%, not allowing false positive results to burden the population), and cancer signal origin prediction?(testing the ability to predict the anatomical localization or origin of detected cancer signals).

Research has shown that among the evaluated classifiers, those using whole genome methylation have the highest cancer signal detection sensitivity at a specificity of 98%.?The training and validation sets provide roughly consistent sensitivity results (Figure 2). In contrast, classifiers that only contain clinical features perform poorly (3%).

And as expected, the sensitivity of all cancer classifiers increases with the increase of clinical cancer staging.

In addition, among the evaluated methods, whole genome methylation had the best predictive cancer signal source: WG methylation cancer signal source classifier accurately predicted the cancer signal source of 75% (95/127) of cancer samples, while SCNA and SNV-WBC accurately predicted the cancer signal source of 41% (52/127) and 35% (44/127) of cancer samples, respectively (Figure 4). The accuracy of WG methylation in predicting cancer signal sources is significantly higher than that of SCNA or SNV-WBC.

Researchers based their estimation of the relative number of tumor specific mutations in circulating tumor allele scores?(cTAF; cfDNA), which are closely related to cancer signal detection performance, utilized the concept of limit of detection (LOD)?and created clinical LOD measurements.

The clinical LOD of a single MCED detection is defined as cTAF with a probability of detecting cancer signals of at least 50% and maintaining 98% specificity (clinical LOD should not be confused with analytical LOD, which typically represents the lowest concentration of a specific analyte and can be detected within 95% of the time using known replicates in a dilution series). Clinical LOD reflects relative sensitivity manifestations. Similar to the sensitivity results of cancer signal detection classifiers, WG methylation, SNV-WBC, and pan feature classifiers provided the lowest clinical LOD, but there was no significant difference between them. The SNV classifier needs to remove the CH background to match the LOD performance of the WG methylation classifier (Figure 3). The SCNAWBC classifier has the lowest clinical LOD in the study, followed by the fragment length classifier.

In this study, WG methylation was the most promising overall choice for cross classifier sensitivity, clinical LOD, and tumor signal source prediction!?Part of the reason may be that it is a universal signal of the entire genome (≈ 30 million CpGs). In addition, the methylation pattern contains a strong tumor specific signal along each fragment, making it easy to identify genomic background variations higher than normal, which may result in lower cTAF levels for methylation signal detection compared to cancer genome detection with other features. Based on methylation in ctDNA, it is possible to detect common cancer signals of over 50 types of cancer with 99.5% specificity, and accurately predict the origin of cancer through operable blood sampling.

Early cancer detection is crucial as it can greatly change the lives of cancer patients.

Reference: doi.org/10.1016/j.ccell.2022.10.022