Bruce Bissonnette

M.D.

University of Chicago

Associate Professor

PLCO (Learn more about this study)

2024-0105

Jul 10, 2025

Non-invasive blood test for multi-cancer detection and tissue of origin determination preceding overt diagnosis

There were ~18.1 million new cancer cases in 2020 world-wide, and 28.4 million are projected for 2040. In the US, >700,000 new cases are estimated in 2023 for prostate, lung, colorectal (CRC), and ovarian cancer. Treatment options for advanced cancers remain limited with few cancers (e.g., CRC) have screening options. Most common cancers are detected late with symptoms, making treatment less effective. Non-invasive blood tests that could detect early-stage cancers could be paradigm-shifting. Current multiple cancer detection approaches, e.g., Galleri (Grail Inc.) have disadvantages: lack of evidence for “predicting” cancer occurrences when cancer is not yet “overt”; requirements for prohibitive amounts of DNA. The PLCO biorepository offers a unique opportunity to investigate the possibility of using our established 5hmC-Seal technique to 1) explore cancer-specific early epigenetic biomarkers; and 2) develop a non-invasive multi-cancer detection tool with high sensitivity and specificity. We have demonstrated 1) feasibility of 5hmC-Seal assay to profile PLCO plasma samples; 2) 5hmC-based model that could detect CRC months or years before tumor diagnosis using PLCO samples; 3) altered 5hmC signatures associated with “overt” cases, such as CRC, lung cancer, and liver cancer that showed biomarker potential for early cancer detection; 4); genome-wide 5hmC shows tissue-specificity; and 5) success of 5hmC-Seal protocol with heparin-stored PLCO samples. Our primary goal is to investigate early epigenetic signatures for prostate, lung, colorectal, and ovarian cancers tracked in the PLCO Trial to develop a multi-cancer detection tool for “pre-clinical” cases. We propose: Specific Aim 1 Train and validate models for individual cancer-types using PLCO samples. For each cancer type, we will randomly split PLCO samples into training and validation sets and profile genome-wide 5hmC using the 5hmC-Seal technique and NGS. We will train and validate a machine-learning (ML) model (e.g., the elastic net regularization) for predicting tumor occurrence for each cancer type. We will evaluate cancer-specific early signatures and explore implicated epigenetic pathways for individual cancer types. Hypothesis: Altered 5hmC in “pre-clinical” cases can be exploited as non-invasive biomarkers for detecting cases months or even years prior to diagnosis. Aim 2. Develop a multi-cancer discriminator for detecting “pre-clinical” cases in PLCO samples. We will combine the genome-wide 5hmC data for all 4 cancer types in the PLCO Trial to develop a multi-cancer classification rule and determine tissue of origin (TOO) using ML approaches. The model will be trained and validated using randomly grouped PLCO cases and controls. Hypothesis: Specific 5hmc signatures associated with “pre-clinical” cancer cases can be exploited to develop a multi-cancer detection algorithm and determine TOO. Aim 3. Evaluate the PLCO-trained algorithm for detection of early cancers in prospectively collected samples. We will prospectively collect ~400 prostate cancer, lung cancer, CRC, and ovarian cancer (stage I-II or equivalent) as well as healthy controls from UChicago Medicine. The 5hmC-Seal profiling in cfDNA will be performed to evaluate the performance of the PLCO-trained algorithm for early detection. Hypothesis: The PLCO-trained model for “pre-clinical” cancer cases has value for early detection of “overt” cancer cases as well.

Therefore, with the primary goals of investigating early epigenetic signatures for common cancers, specifically Prostate Cancer, Lung Cancer, CRC, and Ovarian Cancer, which are covered by the PLCO Trial, and developing a multi-cancer detection tool for “pre-clinical” cases, we propose the following Specific Aims:

◉ Aim 1. Train and validate individual models for common cancers in PLCO samples. For each cancer type, we will randomly split the PLCO samples into a training set and a validation set and profile genome-wide 5hmC using the 5hmC-Seal technique and NGS. We will train and validate a machine-learning model (e.g., the elastic net regularization) for predicting cancer occurrence for each cancer type. We will evaluate cancer-specific early signatures and explore implicated epigenetic pathways for individual cancer types. Hypothesis: Altered 5hmC in “pre-clinical” cases can be exploited as non-invasive biomarkers for detecting cases months or even years earlier than clinical diagnosis.

◉ Aim 2. Develop a multi-cancer detection algorithm for detecting “pre-clinical” cases in PLCO samples.
We will combine the genome-wide 5hmC data for all of the four cancer types: prostate cancer, lung cancer, CRC, and ovarian cancer from the PLCO Trial to develop a multi-cancer classification rule and determine tissue of origin (TOO), using machine-learning approaches. The PLCO samples will trained and validated in randomly grouped cases and controls. Hypothesis: Specific 5hmc signatures associated with “pre-clinical” cancer cases can be exploited to develop a multi-cancer detection algorithm and to determine TOO.

◉ Aim 3. Evaluate the PLCO-trained algorithm for early detection of multiple cancers in prospectively collected samples. We will prospectively collect prostate cancer (n=100), lung cancer (n=100), CRC (n=50), and ovarian cancer (n=100) (stage I-II or equivalent) as well as healthy controls from UChicago Medicine. The 5hmC-Seal profiling in cfDNA will be performed to evaluate the performance of the PLCO-trained algorithm for early detection of these common cancers. Hypothesis: The PLCO-trained model for “pre-clinical” cancer cases has value for early detection of “overt” cancer cases as well.
Impact: This U01 leverages the PLCO biospecimens and our clinical resources as well as our previous studies of biomarker discovery2-5, 8, 9 using our highly sensitive 5hmC-Seal technique2, 10, 11 to fill a critical gap in knowledge required to improve our understanding of epigenetic signals associated with cancer early development. This research offers a timely strategy to develop novel epigenetic tools for cancer early detection.