Classifying prostate cancer subtypes from liquid biopsies

From the Ha and Nelson labs, Human Biology Division

Early detection and diagnosis of cancer is critical to successful treatment. A common procedure for cancer diagnosis starts with a tissue biopsy from the patient. Needle biopsies are minimally invasive and are usually done as an outpatient procedure.  However, even a needle-based biopsy can be problematic: it is an invasive procedure that carries risk, and the biopsy needle may not be able to reach a subset of tumors. Moreover, biopsies obtained from sites of bone metastasis often suffer from poor sample quality that impedes subsequent diagnostic histology.

There is, however, a promising noninvasive alternative: a “liquid biopsy” from blood, a diagnostic methodology that detects and characterizes DNA released from tumor cells. Although still using needles, the liquid biopsy is a simple blood draw and is less risky, less invasive and generally less traumatic to the patient.  Circulating tumor DNA (ctDNA for short) is cell-free DNA that tumor cells release into the bloodstream upon cell death. Liquid biopsies have already been used to detect genomic changes in both oncogenes and tumor suppressors, including BRCA2.

A team of scientists, including researchers from the Gavin Ha and Pete Nelson labs in the Human Biology Division at Fred Hutchinson Cancer Center, developed a set of computational tools to accurately classify metastatic castration-resistant prostate cancer subtypes from ctDNA. Their results are published in Cancer Discovery.

Like other eukaryotic DNA, most cell-free DNA is wrapped around nucleosomes upon its release from the dying cell into the bloodstream. Work in the past decade has demonstrated differences in transcriptional regulation between circulating, cell-free DNA from healthy individuals and ctDNA from patients with cancer. In this study, the researchers wanted to know if they could use ctDNA to accurately classify tumor phenotypes in patients with prostate cancer.

The team faced an initial logistic hurdle: plasma samples obtained from patients contain low tumor content that makes it challenging to obtain enough sample for analysis. So, they used a tried-and-true combination of indirect methods: a mouse model called a patient-derived xenograft (PDX) followed by bioinformatic analyses. The PDX model takes tumor samples from patients and implants them into humanized mice. This method yields a higher tumor content compared to plasma samples obtained directly from patients. The researchers harvest the mouse plasma, which contains ctDNA from patients, and use bioinformatic tools to distinguish human ctDNA from mouse DNA.

A graphical pipeline of the experiments conducted in this study.
Researchers exploited pre-clinical models to develop computational tools for accurate classification of metastatic castration-resistant prostate cancer subtypes from circulating tumor DNA in patients. Modified from De Sarkar et al., Cancer Discovery 2023

The authors obtained ctDNA from mouse plasma from 24 PDX models of human prostate cancer. These 24 models represented a spectrum of androgen receptor (AR) active and neuroendocrine (NE) prostate cancers. In parallel, they isolated PDX tumor samples from the mice and used CUT&RUN, a method developed by researchers in the Henikoff lab in the Basic Sciences Division at Fred Hutch, to profile histone post-translational modification markers in the patient-derived tumors. They also performed ATAC-seq on the tumor samples to determine chromatin accessibility. Then, they applied this dataset to Griffin, a tool recently developed in the Ha lab to determine tumor subtypes based on nucleosome profiling of cell-free DNA.

To predict whether a patient sample represented AR-active or NE prostate cancer, or a combination of both, the researchers developed two computational models. First, they created a probabilistic model called ctdPheno to classify the two subtypes of prostate cancer in an individual plasma sample. However, a given sample often has a mixture of AR-active and NE prostate cancers. To address this, the researchers created an analytic model to estimate the proportions of each prostate cancer subtypes in a single plasma sample. They validated this model, called Keraon, on plasma samples obtained from three prostate cancer patient cohorts. They found that Keraon achieved 97% accuracy for dominant phenotypes, and 87% accuracy for mixed clinical phenotypes. (In ancient Greek mythology, Keraon, or Mixer, is a demi-god of wine mixing.)

The results from this study suggest that analysis of patient-derived ctDNA provides comparable results with the current diagnostic tool of performing immunohistochemistry or transcriptomic profiling on tissue obtained from needle-based biopsies. “We are now trying to determine if these ctDNA approaches can ultimately serve as predictive biomarkers for specific therapies in prostate cancer, including AR-signaling inhibitors and degraders, as well as recently approved treatments such as PSMA-radioligand therapy (Lutetium-177 PSMA-617),” said Dr. Ha. This is a promising sign for patients and the field of precision oncology. In the future, diagnoses from blood-based “liquid biopsies” may replace invasive needle-based biopsies.

This work was supported by the Pacific Northwest Prostate Cancer SPORE (NIH/NCI), Department of Defense Prostate Cancer Research, NIH Director’s New Innovator Award (DP2), NCI Informatics Technology for Cancer Research (ITCR) algorithm development grant (R21), Brotman Baty Institute for Precision Medicine, and the V Foundation. Researchers also received support from the Prostate Cancer Foundation, Institute for Prostate Cancer Research, Fund for Innovation in Cancer Informatics, Doris Duke Charitable Foundation, Safeway Foundation, Wong Family Awards, Dana-Farber Cancer Institute, H.L. Snyder Medical Research Foundation, Cutler Family Foundation, Claudia Adams Barr Program, American Society of Clinical Oncology, Kure It Cancer Research Foundation, and PhRMA Foundation.

Fred Hutch/University of Washington/Seattle Children's Cancer Consortium members Jay Sarthy, Colin Pritchard, Colm Morrissey, R. Bruce Montgomery, Eva Corey, Steve Henikoff, Pete Nelson, and Gavin Ha contributed to this work.

De Sarkar N, Patton RD, Doebley AL, Hanratty B, Adil M, Kreitzman AJ, Sarthy JF, Ko M, Brahma S, Meers MP, Janssens DH, Ang LS, Coleman IM, Bose A, Dumpit RF, Lucas JM, Nunez TA, Nguyen HM, McClure HM, Pritchard CC, Schweizer MT, Morrissey C, Choudhury AD, Baca SC, Berchuck JE, Freedman ML, Ahmad K, Haffner MC, Montgomery RB, Corey E, Henikoff S, Nelson PS, Ha G. 2023. Nucleosome Patterns in Circulating Tumor DNA Reveal Transcriptional Regulation of Advanced Prostate Cancer Phenotypes. Cancer Discovery. 13(3):632-653.