Liquid biopsy refers to the analysis of cancer-derived materials found in blood or other body fluids, offering the possibility of diagnosing, monitoring, and characterizing cancer without invasive tissue sampling. For endometrial cancer - a disease where no routine blood-based screening test currently exists - liquid biopsy technologies could transform early detection, particularly for women who may not yet have symptoms.
Two promising liquid biopsy approaches are studied in this paper. Tumor-educated platelets (TEPs) are blood platelets that have been exposed to tumor signals and incorporate tumor-derived RNA into their contents, creating a detectable molecular signature. Circulating tumor DNA (ctDNA) consists of small fragments of DNA shed from tumor cells into the bloodstream, carrying the specific genetic mutations and methylation patterns characteristic of the tumor of origin.
Both approaches rely on the fact that tumors actively communicate with - and alter - the broader biological environment of the body, including the blood. By reading these molecular signals, researchers can develop tests that distinguish cancer patients from healthy individuals or those with benign conditions, potentially enabling detection at earlier stages when treatment is most effective.
For the tumor-educated platelet (TEP) analysis, blood samples were collected from 295 subjects: 53 endometrial cancer patients, 38 women with benign gynecologic conditions (such as fibroids or polyps), and 204 healthy women. Platelets were isolated from blood, their RNA was extracted and sequenced using RNA sequencing (RNA-seq) technology, and the resulting gene expression profiles were analyzed with a machine learning classifier called imPlatelet.
For the ctDNA analysis, the researchers used a large existing dataset of 519 primary endometrial tumor DNA methylation profiles to train a random forest classifier - a machine learning algorithm that combines predictions from many individual decision trees to make robust classifications. This classifier was then tested on 16 plasma samples from endometrial cancer patients to assess real-world performance in detecting tumor-derived ctDNA in blood.
Both classifiers were evaluated using receiver operating characteristic (ROC) curve analysis and area under the curve (AUC) metrics - standard methods for quantifying how well a diagnostic test distinguishes between positive and negative cases, where an AUC of 1.0 represents perfect discrimination and 0.5 represents chance-level performance.
The imPlatelet classifier applied to tumor-educated platelet RNA achieved an AUC of 97.5% when distinguishing endometrial cancer patients from healthy women - a near-perfect diagnostic performance. This means that in 97.5% of cases, the system correctly ranked cancer patients as more likely to have cancer than healthy women based on their platelet RNA profiles alone.
When the comparison was made against the more clinically challenging scenario - distinguishing endometrial cancer from women with benign gynecologic conditions (who might also present with abnormal bleeding and undergo similar clinical workups) - the AUC was 84.1%. This is still a strong performance given that benign gynecologic conditions can create some degree of biological overlap with cancer in terms of systemic inflammatory and hormonal signals.
The distinction between cancer and benign conditions is arguably more clinically relevant than the cancer-versus-healthy comparison, because most screening scenarios involve evaluating symptomatic women who may have various non-malignant explanations for their symptoms. An AUC of 84.1% in this setting indicates the TEP approach has genuine diagnostic utility, though it would need to be combined with other clinical information for definitive diagnostic decisions.
The random forest ctDNA classifier trained on primary tumor methylation profiles achieved an impressive AUC of 96% when classifying the 519 primary tumor samples. This demonstrates that ctDNA methylation patterns contain highly discriminatory information about endometrial cancer identity - the molecular signals are strong and consistent across tumor samples.
However, when the same classifier was applied to actual plasma samples from cancer patients (the real-world clinical scenario), performance dropped to an AUC of 69.8%, correctly classifying 11 out of 16 samples. This performance gap between tissue-based training and plasma-based testing is a well-recognized challenge in ctDNA liquid biopsy research.
The reason for this gap is that ctDNA in plasma is present in very small quantities - typically only a small fraction of total circulating DNA in the blood comes from tumor cells. The concentration of tumor-derived signal is therefore much lower in plasma than in primary tumor tissue, making classification harder. Early-stage cancers and low-grade tumors tend to shed even less ctDNA, compounding the challenge.
The comparison between TEP and ctDNA approaches reveals important trade-offs. TEPs achieved higher performance in actual blood samples, likely because platelets actively accumulate cancer-derived RNA and amplify the signal, while ctDNA is a passive shedding process diluted by the large background of normal cell-free DNA. TEPs may therefore be more practical for detection of early-stage or low-grade tumors where ctDNA shedding is minimal.
On the other hand, ctDNA carries the actual tumor genetic code - specific mutations, chromosomal abnormalities, and methylation patterns that are unique to the tumor. This makes ctDNA potentially valuable not just for detection but for tumor profiling - understanding what type of endometrial cancer a patient has, what genetic changes drive it, and how it might respond to specific targeted therapies. TEPs provide a more generalized cancer signal but less tumor-specific molecular information.
A combined approach using both TEP RNA profiling and ctDNA analysis might offer complementary advantages - TEPs for sensitive detection and ctDNA for molecular characterization. Such multi-modal liquid biopsy strategies are increasingly explored in cancer research, though they add complexity and cost that must be justified by demonstrated clinical benefit.
This study provides strong evidence that liquid biopsy can detect endometrial cancer from blood samples with clinically meaningful accuracy. The TEP approach in particular - with its AUC of 97.5% against healthy controls and 84.1% against benign conditions - demonstrates that blood-based endometrial cancer detection is feasible and warrants further development.
The potential impact is substantial given that endometrial cancer currently lacks any routine screening test. A reliable blood test could be used to evaluate symptomatic women before invasive procedures, to screen high-risk populations such as those with obesity or Lynch syndrome, and potentially in population-level screening programs in the future - though the latter application would require much higher specificity to avoid excessive false positives and unnecessary interventions.
Key next steps include validation in larger, prospectively collected cohorts across diverse populations, assessment of performance specifically in early-stage disease where the clinical need for sensitive detection is greatest, and head-to-head comparison with existing diagnostic approaches such as endometrial biopsy and transvaginal ultrasound. Regulatory approval for clinical use will require robust evidence from such studies before these tests can be routinely recommended.