Skip to main content
Premium Trial:

Request an Annual Quote

New Liquid Biopsy Data at ASCO Highlights Commercial Push Into Early Cancer Detection

NEW YORK – As the clinical applications for liquid biopsy testing begin to widen in the cancer space, biotech companies have started to further invest in DNA fragmentation- and methylation-based platforms to detect a variety of early-stage cancers.

At the American Society of Clinical Oncology virtual annual meeting last week, some firms presented data on different techniques to analyze a patient's blood sample for epigenetic modifications, in particular DNA methylation, in their circulating tumor DNA, or ctDNA.

Methylation analysis is commonly accomplished by chemically treating cell-free DNA, or cfDNA, with sodium bisulfite to convert unmethylated cytosine to uracil and identifying changes with next-generation sequencing.

Guardant Health, which previously noted that it is shifting to developing early-stage cancer detection assays, presented data on its LUNAR-2 blood-based colorectal neoplasia diagnostic assay that is in development. The assay assesses somatic mutations and methylation and fragmentation patterns in ctDNA to improve sensitivity for early-stage CRC detection.

The Redwood City, California-based firm gathered plasma samples from 434 patients with CRC prior to resection and 271 aged-matched controls and analyzed them with the LUNAR-2 assay.

Training the assay on a separate 614-sample set, Guardant found an overall CRC specificity of 94 percent, with an 88 percent sensitivity in stage I-II patients and a 93 percent sensitivity in stage III patients. While there were no differences in sensitivity between asymptomatic and symptomatic CRC patients, the firm found higher cfDNA tumor fractions in the symptomatic cohort.

Expecting the test to have clinically "meaningful performance in an average risk screening population," Guardant has since launched a prospective registration study to evaluate the test in an average risk CRC screening cohort.

Johns Hopkins University spinout Delfi Diagnostics, which is developing an assay based on DNA fragmentation patterns, also provided new data on its platform at the conference. In its study, the firm evaluated the cfDNA fragmentation screening method in plasma samples from 281 patients with and without cancer who had been diagnosed with additional comorbidities using low-coverage whole genome sequencing.

Applying an approach called DELFI, which stands for DNA evaluation of fragments for early interception, the company and its academic partners found 74 patients with one of 16 different stage I-IV solid cancers (including CRC and lung cancer).

In an email, Delfi said that within the group of patients that did not have cancer, 24 had cardiovascular disease, 23 had diabetes, and the remainder had other comorbidities. After measuring cfDNA fragmentation in patient plasma and using a cross-validation machine learning model, the team found that the assay identified cancer patients from non-cancer patients with an AUC of 0.92, including for common tumors like CRC and lung cancer.

The group also saw that higher DELFI scores were linked with decreased overall survival, independent of cancer stage.

In another study, Delfi researchers modeled cfDNA fragment length distribution and looked at how the parameters could help further distinguish individuals with and without cancer. The team examined the number of cfDNA fragments by size and used statistical methods to approximate the frequency distribution of fragment lengths. To validate the method's performance, the team used a cross validation model with the DELFI approach, parameters from the statistical methods, and a combination of both approaches.

Applying this to a cohort of 215 cancer patients and 208 cancer-free individuals, Delfi's researchers observed cross-validated AUCs of 0.94, 0.95, and 0.97 among the three approaches.

Delfi therefore believes that its platform can both distinguish abnormal cfDNA fragmentation from fragmentation patterns of patients with comorbidities, as well as help detect cancer when combined with the mixture models.

Spanish cancer startup Universal Dx, meanwhile, has been developing its own methylation analysis approach, which uses methylation-sensitive restriction qPCR, or MSRE-qPCR, to detect early-stage cancers. After revealing data on MSRE-qPCR's potential application for lung cancer detection earlier this year, Universal Dx presented new data on the platform's use in CRC screening at ASCO, based on plasma samples from patients scheduled for a colonoscopy or prior to colon surgery for primary CRC.

In the study, Universal Dx first selected and filtered differentially methylated regions (DMRs) from 169 CRC and matched control tissue samples, 21 buffy coat samples, and healthy cfDNA on whole genome bisulfite sequencing. Establishing a CRC-specific signal scoring threshold for individual reads, the team used a panel of methylation scores from 203 DMRs to build a targeted bisulfite hybrid capture sequencing assay and validated it in a test cohort of patients.

Calculated scores were used to train a machine learning model on 68 ctDNA samples from 18 early-stage and 16 late-stage CRC patients and 34 controls. The team then applied the model to an independent set of subjects, including 36 stage I-IV cancer patients and 159 age- and sex-matched controls.

Among the control cohort, 87 patients had a negative colonoscopy finding, 19 had hyperplastic polyps, 37 had small non-advanced adenomas, and 16 were diagnosed with other benign gastrointestinal diseases. Meanwhile, the model correctly classified 92 percent of CRC patients: Applying a specificity of 97 percent, the model's sensitivity per cancer stage ranged from 83 percent for stage I, 92 percent for stage II, and 92 percent for stage III to 100 percent for stage IV.

In addition, using a specificity of 97 percent, the model had a sensitivity of 91 percent and 93 percent for proximal and distal cancers, respectively.

The firm believes that integrating methylation sequencing data analyzed using the read-wise scoring approach with machine learning could be useful for detecting early-stage CRCs. It envisions using the method as the basis for a "highly accurate and minimally invasive" test for CRC screening.

Startup Avida Biomed also presented initial validation data, on an updated version of its Point-N-Seq targeted methylation sequencing, or TMS, dual analysis assay, which now includes both genomic and epigenomic analysis without sample splitting.

Performing validation and multi-center, multi-operator reproducibility studies, the team used spike-in titrations of cancer cell line genomic DNA with known mutations and methylation profiles and achieved a detection level down to .003 percent of tumor DNA.

The Fremont, California-based firm first integrated a colorectal adenocarcinoma TMS panel, covering 560 methylation markers, with a mutation panel with more than 350 hotspot mutations in 22 genes in its dual assay. Using 1 ml plasma from a cohort of CRC patients, the firm and its scientific collaborators detected cancer-specific methylation signals in all samples, as well as oncogenic mutations.

In a cohort of stage I/II CRC patients, the group compared tumor-informed, personalized-mutation panels (100 SNVs) and tumor-independent CRC methylation panels for each patient. The team saw that the integrated assay achieved similar detection as the personalized tumor-informed approach, with an AUC of 0.91. 

In the pilot study, the team collected plasma samples from 75 stage I, 46 stage II, 24 stage III, and 23 stage IV patients. At a specificity of 91 percent, the assay had sensitivity of 70 percent for stage I, 94 percent for stage II, 96 percent for stage III, and 96 percent for stage IV CRC.

In addition, the firm chose to integrate cfDNA size into the analysis to improve the assay's sensitivity.

Avida envisions the TMS dual assay being used to help improve early cancer detection, minimal residual disease detection, and patient monitoring, both for research and clinical applications.

Meanwhile, Chinese molecular diagnostics company AnchorDx, which is developing a ctDNA platform called "Aurora," also provided new details from a validation study involving a cohort of cancer patients. The test uses a DNA methylation panel that tracks between 100 and 200 methylation biomarkers in early-stage lung, breast, CRC, gastric, and esophageal cancer.

Collecting 109 gastric cancer, 177 breast cancer, and 329 healthy plasma samples from five clinical sites in China, AnchorDx's scientists placed the samples in training and validation cohorts. The firm then sequenced the training set on Illumina's MiSeq platform. After building a classifier to separate malignant and normal cases, the firm then evaluated the classifier in the validation cohort and calculated a probability score for each sample.

In the validation set, the Aurora-based breast cancer and gastric cancer classifier had an AUC of 0.93 and 0.95, respectively.

AnchorDx expects to complete validating a larger independent cohort containing CRC and lung cancer — in addition to the modeling for esophageal cancer — later this year. The firm has also launched a large, prospective clinical study to further validate the methylation-based Aurora multi-cancer classifier in asymptomatic patient populations.

San Diego-based Bluestar Genomics is also eyeing the DNA methylation-based cancer testing space, as it provided new data on its liquid biopsy assay for breast, colorectal, lung, ovarian, and pancreatic cancer. The firm's assay tracks 5-hydroxymethylcytosine, or 5hmC, DNA modifications in a patient's blood sample.

Bluestar scientists isolated DNA from 176 fresh frozen tissues from stage I to IV cancer patients and collected cfDNA from the plasma of 783 non-cancer controls and 567 cancer patients. The firm then enriched for 5hmC using chemical labeling, sequenced the samples, and aligned the results to a reference genome to build feature sets of 5hmC patterns.

By analyzing 5hmC across tumor and normal tissues, Bluestar's team identified specific and discrete tumor and normal tissue gene-based features. The firm then applied a machine learning algorithm to the cfDNA dataset, spotting a signature that allowed it to classify non-cancer versus cancer patients. On a cancer-specific level with a specificity of 99 percent, the assay had a sensitivity of 43 percent for CRC, 52 percent for lung cancer, 75 percent for ovarian cancer, 57 percent for pancreatic cancer, and 30 percent for breast cancer.

Based on the study's initial results, Bluestar believes that the combined epigenomic and genomic profiles can distinguish between cancer and normal tissues and between different cancer types in asymptomatic high-risk individuals.