New computational tools widen horizons for liquid biopsies

Methods allow scientists to use gene regulation patterns to detect cancer subtype in cell-free DNA
An illustration of how the arrangement of nucleosomes changes the pattern of cell-free DNA in the blood and sheds light on which genes are turned on and off.
How DNA packaging proteins are arranged on DNA changes the pattern of cell-free DNA fragments in the blood, and can tell scientists more about which genes are turned on and off in cancer cells. Sites where transcription factors bind (TFBS) are free from nucleosomes and form gaps in cfDNA. Doebley et al., Nature Communications 2022

Tumor DNA floating in our blood offers doctors a way to gain information about cancer even when standard needle-based biopsies aren’t an option. Over the past decade, scientists have refined their ability to get important tumor information from mutations in these DNA snippets. More recently, researchers have been exploring ways to go beyond the DNA sequences to learn about which genes are turned on and off in tumors. The genes identified using liquid biopsies can reveal more about a tumor’s subtype, and how it may progress or respond to treatment.

A suite of new computational tools, developed at Fred Hutchinson Cancer Center and recently written up in Cancer Discovery and Nature Communications, could help scientists use liquid biopsies to get more information about tumor gene regulation.

“These methods provide new ways for us to better capture signals from cell-free DNA,” said Fred Hutch computational biologist Gavin Ha, PhD, who led development of the suite of new tools. The first novel tool, dubbed Griffin and described in a paper published online in Nature Communications, improves the level of gene expression-information that can be gleaned from fragments of tumor DNA in patients’ bloodstreams.

“In the clinic, we already use the identification of mutations to direct cancer therapy,” said Fred Hutch prostate cancer researcher and Ha collaborator Peter Nelson, MD.

But tumors are a mixture of different types of cancer cells.

“And what we recognize now is that tumors can change what genes they express in ways that aren’t necessarily dictated by mutations,” Nelson said. “These changes might really influence tumor biology and also therapy.”

“The cornerstone of Griffin is its ability to subtype tumors,” Ha said.

First, Ha and his team showed that Griffin could be used to develop predictive models that can determine whether a metastatic breast tumor is estrogen receptor positive or not — critical information that’s needed to properly treat breast cancer patients.

Next, Ha and Nelson groups used Griffin to build new machine learning methods, called ctdPheno and Keraon, that could identify two subtypes of prostate tumors, each of which require different treatment strategies. These findings were published online in the journal Cancer Discovery in November.

Liquid biopsies have the potential to be more than a less-invasive alternative to needle-based biopsies, said Nelson, who treats patients with prostate cancer. Blood samples can be taken more often and when tumors can’t be reached by needles, which means that they could be used for early diagnosis, long-term monitoring of disease and therapy response, and as tools to help select a patient’s best treatment.

Several liquid biopsy-based tests already on the market assess tumor mutations, but tumors contain even more kinds of important molecular information that could guide treatment. Ha is among the growing number of scientists who are seeking to mine these fragments for information beyond their DNA code. He, Nelson and their collaborators hope that Griffin and similar approaches will allow doctors to use low-cost and safe blood-based diagnostics to gain the information they need to optimize individual patients’ cancer care.

Drs. Gavin Ha (left) and Pete Nelson (right) teamed up to improve the capabilities of liquid biopsies to distinguish cancer subtypes with differing gene expression patterns.
Drs. Gavin Ha (top) and Pete Nelson (bottom) teamed up to improve the capabilities of liquid biopsies to distinguish cancer subtypes with differing gene expression patterns.

Fred Hutch file photos

Griffin: Getting more information from cell-free DNA

Early liquid biopsy strategies focused on the sequence of the fragments of cell-free DNA released by tumors. The DNA of cancer cells can be rearranged or mutated in characteristic ways, allowing scientists to detect the presence of cancer or important mutations that could reveal treatment vulnerabilities. But the information in cell-free DNA goes far beyond its DNA sequence. Scientists are delving into the epigenetic information — information about how DNA is packaged and which genes are likely turned on or off — that lie hidden in these snippets, Ha said.

DNA packaging influences gene expression. Our DNA isn’t a tangled mess because our cells organize it by wrapping it around wheel-shaped proteins called nucleosomes. In general, genes that are expressed are more loosely bundled and the nucleosomes shift to allow transcription factors, the molecules that turn genes on, access to DNA. Cells help keep genes turned off by tightly bundling them around more nucleosomes, which also prevents transcription factors from binding DNA and turning genes on.

Both healthy and tumor cells release DNA into our bloodstreams (most cell-free DNA actually comes from immune cells). When this happens, the short stretches wrapped around nucleosomes are protected against degradation.  This results in a confetti of DNA fragments in the blood. The pattern of confetti can tell researchers about where nucleosomes are located: like a negative image, there’s more confetti from silent, nucleosome-rich genes and less confetti from expressed genes with fewer nucleosomes.

Ha and his team work on computational methods to build a picture of active and quiet regions of DNA. Griffin enhances this ability by helping to correct for a source of noise, or irrelevant information, found in sequencing strategies for cell-free DNA.

Anna-Lisa Doebley, PhD, who spearheaded the project as a graduate student in Ha’s lab, developed a way to overcome a source of sequencing noise called GC bias. DNA fragments with extra-high or extra-low amounts of guanine and cytosine bases (G and C) can be over- or under-represented in next-generation sequencing data. This can make it difficult to accurately determine how much of a given fragment is in a liquid biopsy. Doebley designed Griffin to correct the GC bias in cell-free DNA.

It’s ability to correct for GC bias makes Griffin much better at revealing “accessible” stretches of DNA where transcription factors sit, Ha said. This paints a more detailed picture of the nucleosome (and by extension, gene expression) pattern of a patient’s tumor.

Doebley and Ha put Griffin through its paces in two different applications. To test Griffin’s potential as a diagnostic, they used a public data set of cell-free tumor DNA sequences from people with different kinds of early-stage cancer. Griffin was able to detect tumor-specific nucleosome patterns and identify early-stage tumors from blood samples.  

Then, to test the tool’s potential to help guide therapy, they tested Griffin against blood samples from patients with metastatic breast cancer. The presence or absence of different hormone receptors on breast tumors can make them more or less likely to respond to specific treatments, making this information critical when choosing appropriate therapies. Doebley and the team showed that as long as at least 5% of the cell-free DNA in a blood sample was tumor DNA, Griffin could be used to predict the estrogen receptor status of a patient’s tumor, at up to 92% accuracy.

To see if Griffin could detect more complex genetic patterns, Ha teamed up with Nelson to dig deeper into prostate cancer.

Griffin distinguishes prostate tumor subtypes

In the last decade, androgen receptor-targeted therapies have dramatically extended the lives of men with advanced prostate cancer. But these new therapies aren’t cures, and they’re driving new developments in prostate cancer.

In response to AR-targeting drugs, “prostate tumor cells lose the features that comprise the original cell type and gain new characteristics,” Nelson said.

In 2017, his group reported on this phenomenon, called trans-differentiation. A growing percentage of recurrent prostate tumors are taking on characteristics of neuroendocrine cells, which allows them to thrive without relying on the androgen receptor. These NE-type tumors are vulnerable to different drugs. And because gene expression changes drive protein changes, they could have important implications for immune-based cancer therapies and imaging, Nelson said.

Right now, clinicians do not have a non-invasive approach to identify tumors that have undergone trans-differentiation, Nelson said. These tumors must still be detected using a standard needle-based biopsy. But trans-differentiation is driven by changes in gene expression — exactly what Griffin is designed to ferret out.

Nelson Lab staff scientist Navonil De Sarkar, PhD, who also contributed to the Nature Communications paper, teamed up with Robert Patton, a post-doctoral trainee jointly mentored by Ha and Nelson, and applied Griffin to subtyping prostate tumors from blood samples. They used cell-free DNA from patient-derived xenograft models, in which human tumor tissue is grown in mice, to help develop novel predictive models that can zero in on patterns in tumor gene usage.

A schematic of how the computational analysis of PDX cell-free DNA enabled the development of tools to distinguish different prostate cancer subtypes.
Growing human tumor tissue in mice makes it easier to reveal relevant gene expression patterns in tumor cells. De Sarkar was able to use human tumor tissue to analyze the packaging and gene-regulation modifications in different subtypes of human prostate cancer (bottom) while zeroing in on the "confetti" of human cell-free DNA in mouse blood (top). Applying Griffin to this data allowed the researchers to develop two predictive models that could determine whether a patient's cell-free DNA came from one of two prostate tumor subtypes. Modified from De Sarkar et al., Cancer Discovery 2022

Using this strategy, the team developed two models, ctdPheno and Keraon, for accurately predicting androgen receptor active and neuroendocrine prostate cancer subtypes. These methods can also predict mixtures of both subtypes that can emerge during trans-differentiation.

When applied to clinical samples, Keraon was able to distinguish AR-active from NE-type prostate tumors in human blood samples with 97% accuracy when one or the other was the predominate subtype, and with 87% accuracy when blood samples came from patients whose tumors displayed characteristics of both subtypes.

“To me, the beauty of the assay — and where it's going to go next, in part — is trying to understand and measure tumor heterogeneity, which is really difficult to do noninvasively,” Nelson said.

Tumor heterogeneity, or a mixture of subtypes within a patient, is an important frontier for cancer doctors. It can drive drug resistance, since only a subset of cancer cells will be vulnerable to a given drug. But needle biopsies only take tissue from one region of a single tumor, and patients with metastatic disease may have tumors spread through the body. This makes it less likely that they’ll be able to pick up on the presence of multiple tumor subtypes within an individual patient, Nelson said.

A non-invasive way to capture the range of cancer subtypes within an individual patient would be an important boon to tailoring their treatment.

Next steps: Refining the approach

Next, Ha and his team want to make Griffin more sensitive and widen its applicability. The 5% tumor DNA minimum is also a limitation the team is working to overcome, so that Griffin can be used to extract tumor molecular information from a wider range of patient samples, Ha said. These efforts will include improving its sensitivity and adapting the method for data obtained using deeper sequencing than with the breast cancer samples used in Doebley’s initial studies.

Griffin may also give researchers information that is both clinically useful and reveals deeper insights into tumor biology. De Sarkar and Patton showed that in addition to distinguishing two tumor subtypes, Griffin could profile nucleosome position patterns in key genes and show which genes are more transcriptionally active, or “on.” The team was able to correlate this information with tumors’ levels of proliferation.  

“We're working towards a signature in cell-free DNA that may be able to give us extra information on tumor aggressiveness,” Ha said.

As part of a clinical trial testing a strategy to improve small cell lung cancer response to immunotherapy, Ha is collaborating with Fred Hutch lung cancer researcher Dr. David MacPherson to use liquid biopsies to assess how the experimental treatment may alter the molecular characteristics of lung tumors.

Before Griffin, ctdPheno and Keraon can be used in the clinic to guide treatment for prostate cancer patients, the team will need to confirm their ability to distinguish tumor subtypes and show that using the tools to tailor treatment improves patients’ outcomes, Nelson said.

He also anticipates that molecular information about tumor heterogeneity could merge with advances in cancer imaging to make it faster, easier and cheaper to monitor and treat patients.

Ha also envisions using Griffin to add new gene-expression capabilities to current clinical cancer mutation panels, like the University of Washington’s OncoPlex, that assess a range of potential cancer mutations.

“That would then really broaden the scope and maximize the impact of this tool with a path towards accelerating its translation into the clinic,” he said.

These projects were funded by the National Institutes of Health, the V Foundation, the Prostate Cancer Foundation, the Fund for Innovation in Cancer Informatics, the Department of Defense and the Pacific Northwest Prostate Cancer SPORE.

Sabrina Richards, a staff writer at Fred Hutchinson Cancer Center, has written about scientific research and the environment for The Scientist and OnEarth Magazine. She has a PhD in immunology from the University of Washington, an MA in journalism and an advanced certificate from the Science, Health and Environmental Reporting Program at New York University. Reach her at

Are you interested in reprinting or republishing this story? Be our guest! We want to help connect people with the information they need. We just ask that you link back to the original article, preserve the author’s byline and refrain from making edits that alter the original context. Questions? Email us at

Related News

All news
Unlocking more information from liquid biopsies Dr. Gavin Ha receives $1.5M NIH Director’s New Innovator Award for creative young scientists with high-risk, high-reward projects October 3, 2022
Targeted therapies push metastatic prostate cancer in new directions Combination therapy could help stave off new “escape variant” of disease that arises in response to powerful treatments November 14, 2017
Identifying the root of prostate cancer treatment resistance using liquid biopsy Dr. Gavin Ha receives 2019 V Scholar Grant to study whether noninvasive procedure can help monitor and predict how patients will respond to therapy January 30, 2020

Help Us Eliminate Cancer

Every dollar counts. Please support lifesaving research today.