2019 AACR | Breakthrough in Application of Methylation Ultra-deep Sequencing in Liquid Biopsy for Early Lung Cancer Screening

Var Date:2019-08-26


At the ongoing 2019 annual meeting of American Association for Cancer Research (AACR), Burning Rock was said to demonstrate ELSA-seqTM, a hypersensitive ctDNA methylation analysis method developed by Burning Rock for data of research on early lung cancer screening. This study was co-conducted by Burning Rock, UC Berkeley, Peking Union Medical College Hospital and Shanghai Chest Hospital. Research findings show that capture of tumor-related apparent genetic change by single-base resolution is promising in application to early noninvasive lung cancer testing!


Based on ELSA-seqTM, the method can successfully produce sufficient single-base resolution sequencing libraries by only 1ng free plasma DNA (cfDNA). Subject to mining results in the public database (cancer cell count=4,772, normal cell count=411 and leukocyte count=656) and Burning Rock’s methylation genome sequencing database, Burning Rock screened 80,637 typical lung cancer and pan-cancer differential methylation loci (DML) for designing overlapped capture probes to detect cancer-related methylation variation in the blood.


The efficient and accurate ssDNA database building and targeted capture laid a solid foundation for ELSA-seqTM ultra-deep methylation sequencing -- under median sequencing depth of 6,000X, the effective sequencing amount kept mounting, enabling effective correction of sampling errors concerning low-frequency signals, PCR amplification and sequencing errors.


 

Burning Rock’s R&D team designed a detailed analytical performance for ELSA-seqTM. In the simulated data, ELSA-seqTM testing susceptibility hit 0.00001 (1/100,000); in the cancer cell mixing test, the susceptibility remained stabilizing at 0.0005 (5/10,000) despite repetition of many operators. Compared with the traditional antibody capture or digestive enzyme methylation sequencing, the ELSA-seqTM method showed a more satisfactory repeatability.



For the sake of clinical validation of the method, researchers applied ELSA-seqTM to two separate cohorts: 398 early lung cancer patients (78% of them were stage I patients) and 501 healthy persons:

Cohort from center 1: the cohort was divided into a training group (n=314) and holdout verification group (n=309) at random;

Independent test group from center 2: sample collection, processing and data analysis were separately done. Pathological results were not given until the end of analysis.



In this study, for stage I/II/III non-small cell lung cancer patients and healthy persons, the susceptibility hit 64%, 80% and 87% respectively and specificity hit 96% in this method, significantly higher than the blood-based early lung cancer screening platform reported earlier.


It is also the biggest ctDNA methylation ultra-deep sequencing cohort for early lung cancer patients in China and around the world.


Figure A. Testing Results of Training Sample Set (Left) and ROC Curve (Right), n=314

AUC of the training set equals to 0.93 (95% CI: 0.88, 0.95), indicating effective model training.


Figure B. Testing Results of Verification Sample Set (Left) and Predicted Probability of All Samples by Stage and Group in Pathology (Right), n=309

The holdout verification set showed approximate classification effects, and its predicted probability kept rising as the cancer’s stage progressed, which conformed to biological features and displayed the model’s accurate simulation of features and effective control of overfitting.


Figure C. Testing Results of Independent Test Group (Left) and Gene Mutation Testing Results of Some Patients (>10,000X) in Parallel Comparison (Right):

A total of 28 early lung cancer patients with sufficient plasma were tested in methylation and somatic mutation. ELSA-seqTM successfully determined 27 patients (27/28), 16 of whom were specific. It, once again, verified the high susceptibility of methylation ultra-deep sequencing.


According to ELSA-seqTM’s dual-center clinical verification results (n=899): signals of early lung cancer in plasma can be captured by ultra-deep methylation targeted sequencing. Upon a large sample training, the machine learning model showed sound accuracy and generalization, indicating its huge potential for application in the clinical environment.


There is a long way to go on the path of verification for transferring scientific research findings into clinical products. Burning Rock’s R&D and medical teams will continue to work on development, optimization and clinical multiple-cohort verification of DNA-based methylation products for early lung cancer screening. It is expected that better testing techniques can embark on a new era of early cancer diagnostics and screening.