Integrating Genomic and Fragmentomic Features for Tissue-of-Origin Prediction from Routine Targeted ctDNA Sequencing
This study introduces SHAPE, a novel method that leverages routine targeted ctDNA sequencing to integrate genomic and fragmentomic features for highly accurate, biologically interpretable prediction of cancer tissue-of-origin, potentially extending the clinical utility of existing assays without requiring additional sequencing.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer
When a cancer spreads to other parts of the body, doctors face a critical challenge: they must know where the disease started to choose the right treatment. This starting point is called the tissue of origin. Sometimes, despite scans and biopsies, the origin remains a mystery, leaving patients without a clear path forward. In recent years, a technique called liquid biopsy has emerged, allowing doctors to find traces of cancer in a simple blood sample. These traces are tiny fragments of DNA shed by tumors into the bloodstream. While scientists have long known how to read the genetic code within these fragments to find specific mutations, a new layer of information has been discovered hiding in the physical shape and size of the DNA pieces themselves. Just as a book might be torn in a way that reveals which chapter it came from, these DNA fragments carry subtle clues about the type of cell they originated from.
A team of researchers at Samsung Medical Center in South Korea asked whether they could unlock these hidden clues using the standard blood tests that cancer patients already receive. Currently, hospitals use targeted sequencing panels to look for specific genetic changes in the DNA of circulating tumor cells. These tests focus on a specific set of genes known to be involved in cancer, but they were not originally designed to analyze the physical structure of the DNA fragments. The researchers wondered if they could squeeze extra diagnostic value out of this existing data without needing new, expensive, or time-consuming tests. They developed a method to look closely at how these DNA fragments break apart at specific locations where proteins bind to the DNA, a process that varies depending on the type of tissue the cancer came from.
The team analyzed blood samples from nearly 1,900 patients with metastatic cancer, covering eleven different types of cancer such as lung, colorectal, and pancreatic cancer. They used a routine clinical test that scans 523 genes, which is a standard tool for guiding cancer treatment. From the same data used to find genetic mutations, the researchers extracted five different types of information. They looked at the genetic mutations themselves, the overall number of DNA copies, the depth of coverage across the genes, and two new features related to how the DNA fragments were cut. One of these new features measured the diversity of fragment lengths at specific spots where proteins attach to the DNA, while the other measured how the ends of the fragments were positioned. By combining all these signals, they built a computer model designed to predict the tissue of origin.
The results showed that this combined approach was highly effective. When the model looked at all five types of information together, it correctly identified the cancer's origin in the vast majority of cases, performing significantly better than any single type of information could on its own. The most powerful individual clues came from the physical fragmentation patterns and the overall coverage of the DNA, but the genetic mutations provided a unique boost that the other signals could not replace. The model worked best when the amount of tumor DNA in the blood was high enough to be easily detected, achieving an accuracy of about 80 percent in those cases. Even more importantly, the method was able to distinguish between different cancer types based on the specific biological signatures left behind by the cells, such as which proteins were binding to the DNA in the original tissue.
To ensure their findings were not just a fluke of their specific lab equipment, the researchers tested their method on a completely separate group of patients from a different institution, using a different sequencing platform. The model maintained its high accuracy in this independent group, proving that the approach works across different settings. They also checked if the method could be transferred between two slightly different versions of the clinical test used in their main study. While some features remained stable, others required careful adjustment, highlighting that the physical details of the DNA fragments are sensitive to how the test is run. This suggests that while the method is robust, it needs to be calibrated for each specific laboratory setup.
The study confirms that routine blood tests for cancer contain a wealth of information that goes beyond simple genetic mutations. By analyzing the physical shape and size of DNA fragments, doctors can now potentially determine where a cancer started without needing a new biopsy or a specialized test. This approach turns existing data into a powerful diagnostic tool, offering a way to solve the puzzle of unknown primary cancers using the very same samples patients are already providing. The researchers emphasize that while the results are promising, the method still needs to be tested in large, prospective studies involving patients whose cancer origin is truly unknown before it can become a standard part of clinical care. For now, the work demonstrates that the story of a cancer's origin is written not just in its genetic code, but in the very way its DNA breaks apart.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.