A New Blood Test Aids in the Early Detection of Over 50 Cancers
By Sahana Shankar, Ph.D. Candidate
A collaborative study involving researchers at the Dana Farber Cancer Institute and the Mayo Clinic among others have employed a next-generation sequencing-based test developed by GRAIL Inc. for the analysis of methylation signatures using AI to detect a wide spectrum of cancers with increased specificity, sensitivity and accuracy.
One of the biggest hurdles in combating cancer is its timely diagnosis. It is made harder by the fact that cancer is a multigenic disorder and manifests with different symptoms based on its tissue of origin. Another hallmark of a tumor cell is its markedly different methylation status compared to healthy cells. Hence, information on methylation status is invaluable for prognosis. Techniques like high-throughput genomics have resulted in a database that profiles the methylation status of a wide variety of cells in the human body. A new study looks at these methylation patterns in circulating-free DNA (cfDNA-DNA shed into the bloodstream when cells die) in cancer and non-cancer situations.
DNA Methylation – A Better Predictor of Cancer
All cells in the body contain the same genetic code (DNA). However, not all cells use the code in the same manner. This gives rise to cell diversity. An epithelial cell is distinct from a neuron in its shape, function, turnover rate, etc. One of the major reasons for this is the way the genetic code is utilized in different cells. Differential gene expression is driven by DNA methylation. Each cell type has a distinct methylation pattern which causes the repression of methylated genes, enabling a unique gene expression profile based on cell type and/or cell stage. Tumor cells contain these tissue-specific methylation patterns as well as tumor-specific epigenetic markers, making targeted methylation a better predictive indicator for cancer than mutations.
The Population-Scale Study
The study led by Dr. Geoffrey Oxnard, MD of the Dana-Farber Cancer Institute and Minetta Liu, MD of the Mayo Clinic is a part of the largest clinical genomics program and the first population-scale study of cancer signatures based on circulating free DNA. In their previous report, the investigators established the CGCA- Circulating Cell-Free Genome Atlas by combining whole-genome bisulfite sequencing of cfDNA and machine learning to detect cancer types and their tissue of origin. They used samples from healthy individuals (44%) and over 50 cancer types (56%) to train and validate an algorithm and generated the profiles for whole-genome methylation in both sets. In the second part of the study, published recently in Annals of Oncology, they refine the methylation profiles to look for targeted regions with the most difference between cancer and non-cancer types and use a new set of samples to validate this targeted methylation classifier to detect cancer type and its tissue of origin.
With input from their previous WGBS data and methylation data from the Cancer Genome Atlas, they identified a total of 103,456 distinct methylation regions that can be predictive of cancer and their tissue of origin to develop a custom hybridization capture panel for targeted sequencing. Out of the 30 million CpGs in the genome, approximately 1 million CpGs were considered informative in the context of cancer.
Over 6000 blood samples were collected from healthy individuals and previously untreated cancer patients and divided them into training (n=4780) and independent validation (n=1969) sets. In case of tumors, the tissue biopsy was used too. Targeted probes from the methylation panel pulled out matching regions from the cfDNA from all these groups and the regions were sequenced (139X depth) to check for specific methylation signatures. Since this is a machine learning study, they built custom software with source models and logistic regression to recognize methylation patterns and classify them into cancer/non-cancer types and resolve their tissue of origin.
Accuracy of the Multi-Cancer Early Detection Test
Of the 6689 samples, the study could resolve 99.3% cases correctly and the sensitivity to identify the type of cancer depended on the stage of cancer. Among the 12 most prevalent cancers comprising of breast, colorectal, esophageal, gallbladder, bladder, gastric, ovarian, head and neck, lung, lymphoid leukemia, multiple myeloma, and pancreatic, the sensitivity was 67.3%. Stage IV cancers had a detection rate of >90%, stage III-83%, stage II- 69% and stage I-39%. In cases where the cancer was detected, the prediction of tissue of origin was 90% accurate which is crucial for therapy.
“At GRAIL, we believe that multi-cancer early detection has the potential to significantly reduce cancer mortality,” said Alex Aravanis, MD, Ph.D., Chief Scientific Officer and Head of R&D, and a co-founder of GRAIL. “This is a seminal moment in the field of cancer detection. We’ve built what we believe to be one of the largest clinical study programs ever conducted in genomic medicine, and the data published in Annals of Oncology further support GRAIL’s approach and commitment to clinical and scientific rigor.”
Blood-based cancer detection has been proposed before. But whole-genome sequencing or targeted sequencing to identify mutations have limits in sensitivity and specificity. Methylation is more pervasive in the genome. Hence targeted methylation detection is likely to be more accurate. An added advantage is that epigenetic signatures inherently reflect tissue of origin. The study was applicable to a wide variety of prevalent and less-prevalent cancers and hence can be used for wide population screening. Some of the cancers detected by the study lack standard diagnostic tests and currently the test is available for at-risk individuals in the clinical study at the Dana Farber Institute.
The authors are working on further validating the methods by (a) developing better software to improve detection at earlier stages for better mortality outcomes (b) including more asymptomatic participants, further analyses and follow-up of participants included in the study (c) increasing representation of all cancer types. The study hopes to build a comprehensive cancer detection platform with information about the tissue of origin and malignancy status. This could mean a single test can benefit many, without the need for multiple diagnostics, saving time, physical, psychological and financial costs.
Editor: Rajaneesh K. Gopinath, Ph.D.
©www.geneonline.com All rights reserved. Collaborate with us: firstname.lastname@example.org