Blood-Based Methylation Assay Enables Early, Accurate Detection of Solid Tumors

November 20, 2020
Caroline Seymour
Caroline Seymour

Editor, OncLive®
Caroline Seymour is your initial point of contact for the OncLive® podcast, OncLive On Air™. She joined the company in 2018 as an assistant editor, with expertise in video production and print/digital publication. Email:

The noninvasive blood-based targeted methylation assay ELSA-seq identified early cancers with high specificity and provided accurate predictions of the tissue of origin.

The noninvasive blood-based targeted methylation assay ELSA-seq identified early cancers with high specificity and provided accurate predictions of the tissue of origin, according to findings from the THUNDER-II study presented during the 2020 ESMO Asia Virtual Congress.1

At 99.5% training specificity (95% CI, 96.7%-100%), the cross-validated sensitivity of the assay was 79.9% (95% CI, 74.6%-84.4%). The results proved to be consistent in the validation set, with 98.3% specificity (95% CI, 95.8%-99.4%) and 80.6% sensitivity (95% CI, 76.0%-84.4%) reported across stages and cancer types.

“In general, the performance was consistent in the training and validation set,” Qiang Gao, MD, lead study author and professor in the Department of Liver Surgery and Transplantation, Liver Cancer Institute and Zhongshan Hospital, Fudan University, said during a virtual presentation of the data. “Taken collectively, these findings highlight the potential implementation of ELSA-seq as a noninvasive multicancer detection test.”

Early cancer detection can provide clinical benefits, particularly for those with inadequate screening methods.

ELSA-seq is a sensitive targeted methylation sequencing assay that interrogates epigenetic alterations from circulating cell-free DNA (cfDNA), enabling simultaneous identification and localization of multiple cancer types. The target panel covers 80,672 CpG sites, spanning genomic regions of approximately 1.05 Mb. A total of 8312 co-methylation blocks served as markers to create the classification model.

The THUNDER (THe UNintrusive Detection of Early-stage cancer) program was comprised of 2 phases: THUNDER-I (n = 896) and THUNDER-II (n ~2500). THUNDER-II is a substudy of THUNDER, which is a pilot study designed for the development and validation of ELSA-seq in 6 tumor types: lung, colorectal, liver, ovary, pancreas, and esophagus; these types account for approximately 52% of all cancer cases in China.

The first phase of the research was designed to evaluate the performance of ELSA-seq in high and low concentrations of ctDNA, according to Gao.

Previously reported findings from the THUNDER-I study, which were presented during the 2020 AACR Special Conference on Advances in Liquid Biopsies meeting, demonstratedwith an overall specificity of 95% (n = 215), detection rates of 65% (n = 86), 82% (n = 125), 88% (n = 101), and 94% (n = 84) for stage I to IV solid tumors, respectively.2 The assay also accurately predicted the tissue of origin in 93% (n = 117) of cases in the validation set.

In a substudy, ELSA-seq identified approximately double the number of patients (n = 27) as deep mutation sequencing (n = 15), with no false positives. Moreover, the assay did not require any biopsied tissue to detect ctDNA at an allele frequency as low as 0.02%.

THUNDER-II consisted of 5 steps: marker discovery (n = 247), panel validation (n = 599), model training (n = 469), validation (n = 639), and the blind test, which is ongoing.

In the marker discovery phase of the research, 247 individuals underwent tissue testing (n = 116 patients with cancer; n = 131 non-cancer controls) alongside 5374 individuals from The Cancer Genome Atlas and Gene Expression Omnibus Database (n = 4366 patients with cancer; n = 1008 non-cancer controls).

In the panel validation phase, 599 individuals underwent tissue and blood testing (n = 249 patients with cancer; n = 350 non-cancer controls).

In the final assay validation phase, patients underwent blood testing in 1 of 3 groups: the training set (n = 274 patients with cancer; n = 195 non-cancer controls), the validation set (n = 351 patients with cancer; n = 288 non-cancer controls), and an independent validation set, which is ongoing.

Baseline characteristics indicated that patients with cancer and non-cancer controls had comparable age, gender, and smoking status in the training and validation sets. Moreover, patients with different stages and cancer types were well represented in the cancer group in the training and validation sets, and 80.0% patients were diagnosed at early stages (stage I to III).

Among non-cancer controls in the training set (n = 195), the mean age was 53 (range, 40-72) and the majority were female (n = 128; 70%). Among patients with cancer (n = 274), the mean age was 60 (range, 47-74), the minority were female (n = 110; 40%), and the clinical stage distribution was as follows: stage I (n = 73; 27%), stage II (n = 63; 23%), stage III (n = 97; 35%), and stage IV (n = 41; 15%). The training set included 50 patients with lung cancer, 46 patients with colorectal cancer (CRC), 48 patients with liver cancer, 50 patients with ovarian cancer, 40 patients with pancreatic cancer, and 40 patients with esophagus cancer.

Among non-cancer controls in the validation set (n = 288), the mean age was 54 (range, 40-74) and the majority were female (n = 171; 59%). Among patients with cancer (n = 351), the mean age was 59 (range, 40-75), the minority were female (n = 22; 36%), and the clinical stage distribution was as follows: stage I (n = 83; 23%), stage II (n = 87; 25%), stage III (n = 94; 27%), and stage IV (n = 87; 25%). The validation set included 61 patients with lung cancer, 57 patients with CRC, 57 patients with liver cancer, 53 patients with ovarian cancer, 59 patients with pancreatic cancer, and 64 patients with esophagus cancer.

Additional findings demonstrated that, with regard to the localization of diseased organs, the classifier returned a tissue-of-origin result in 98.6% cases (n = 129). For single organ determinations, the predictive accuracy of the assay was 79% in the training phase and 82% in the validation phase. In the training phase, the predictive accuracy was highest for ovarian cancer (n = 34; 97%), followed by liver cancer (n = 39; 95%), 85% (n = 33; 85%), esophagus cancer (n = 21; 72%), lung cancer (n = 26; 65%), and pancreatic cancer (n = 21; 60%).

In the validation phase, the predictive accuracy was highest for liver cancer (n = 46; 100%), followed by ovarian cancer (n = 36; 97%), CRC (n = 34; 81%), esophagus cancer (n = 39; 80%), pancreatic cancer (n = 41; 72%), and lung cancer (n = 31; 69%).

For the top 2 organ determinations, the predictive accuracy was 89% and 87%, respectively.

“ELSA-seq also enabled accurate prediction of diseased organs, offering guidance for subsequent diagnostic steps,” said Gao.

Similar evaluation of a cfDNA methylation-based model in early cancer detection is underway in the prospective, multicenter, longitudinal PREDICT (Pan-canceR Early Detection projeCT) study.3 The study will enroll approximately 14000 participants. The development and validation of the model will be evaluated in patients with early-stage cancers or benign diseases, along with non-tumor (healthy) individuals in a 2-stage process. The sensitivity and specificity of the model in early cancer detection and the accuracy of the identification of the tissue of origin will be evaluated.


  1. Gao Q, Li B, Cai S, et al. Early detection and localization of multiple cancers using a blood-based methylation assay (ELSA-seq). Presented at: 2020 ESMO Congress; November 20-22, 2020; Virtual. Abstract LBA3.
  2. Li B, Xu J, Wang C, et al. Multiplatform analysis of early-stage cancer signatures in blood. Presented at: AACR Special Conference on Advances in Liquid Biopsies; January 13-January 16, 2020. Abstract 62552825.
  3. Pan-canceR Early Detection projeCT (PREDICT). Updaeted July 22, 2020. Accessed November 20, 2020.