In our newly published study, we describe how this work progressed from analyzing large cancer genome datasets to developing machine learning models, conducting large-scale functional experiments in mice, and ultimately confirming the clinical relevance of these RNAs in nearly 200 breast cancer patients using blood samples.

Cancer-Specific OncRNAs Are Widespread

One of the first major discoveries was that this phenomenon was not limited to breast cancer. By examining small RNA sequencing data from The Cancer Genome Atlas across 32 different cancer types, we identified approximately 260,000 cancer-specific small RNAs. We refer to these molecules as oncRNAs, and they were present across every cancer type analyzed.

Their distribution was not random. Each cancer type displayed its own distinct oncRNA expression pattern. Lung cancers, for example, showed a different set of oncRNAs compared with breast cancers. Using these patterns, machine learning models were able to classify cancer types with 90.9% accuracy. When tested in a separate group of 938 tumors, classification accuracy remained high at 82.1%.

Differences also emerged within individual cancers. Basal breast tumors showed oncRNA patterns distinct from luminal tumors, suggesting additional subtypes that may not yet be fully defined. These findings indicate that oncRNAs reflect fundamental aspects of cancer cell state. Patterns of oncRNA presence and absence function as “digital molecular barcodes” that capture cancer identity at multiple levels, including tumor type, subtype, and cellular state.

Some OncRNAs Actively Drive Tumor Growth

Although oncRNAs serve as powerful biomarkers, we also wanted to understand whether some of them directly influence cancer progression. Specifically, we asked whether cancer cells could use these newly emerged RNA molecules to activate oncogenic pathways.

To test this, we created screening libraries containing about 400 oncRNAs from breast, colon, lung, and prostate tumors. These RNAs were introduced into cancer cells using lentiviral vectors. In half of the cases, we increased oncRNA expression. In the other half, we reduced expression using “Tough Decoy” constructs. The modified cells were then implanted into mice to determine which oncRNAs enhanced tumor growth.

Roughly 5% of the oncRNAs produced clear biological effects in xenograft mouse models. Two breast cancer oncRNAs were examined more closely. One triggered epithelial-mesenchymal transition, an essential step in cancer progression and metastasis. The other activated E2F target genes, promoting cell proliferation. Both significantly accelerated tumor growth and increased metastatic colonization in independent cell line models.

When we examined patient tumor data, we found that tumors expressing these same oncRNAs displayed similar pathway changes. Observing consistent biological patterns in TCGA samples and experimental models strengthened our confidence in the findings.

Cancer Cells Release OncRNAs Into the Bloodstream

Perhaps the most clinically important discovery was that cancer cells actively release many of these oncRNAs into the bloodstream. Tracking these circulating RNAs provides insight into how patients are responding to treatment.

We analyzed cell-free RNA from 25 cancer cell lines across 9 tissue types and found that about 30% of oncRNAs are actively secreted. To confirm their clinical relevance, we studied serum samples from 192 breast cancer patients enrolled in the I-SPY 2 neoadjuvant chemotherapy trial. Blood samples were collected before and after treatment, and we calculated the change in total oncRNA burden (ΔoncRNA below).

That single measurement proved highly informative. Patients with high residual oncRNA levels after chemotherapy had nearly 4-fold worse overall survival. This association remained significant even after accounting for standard clinical indicators such as pathologic complete response and residual cancer burden.

This was our most ambitious goal. Although we knew oncRNAs could be detected in blood, it was uncertain whether they would provide meaningful information in real patient samples. Detecting such a strong signal from just 1 milliliter of serum was unexpected.

A New Approach to Monitoring Minimal Residual Disease

These findings address a significant clinical challenge. Monitoring minimal residual disease in breast cancer using markers such as cell-free DNA is difficult because tumors often release very little DNA into the bloodstream, particularly in early stages. RNA-based monitoring may offer an advantage because cancer cells actively secrete RNA rather than passively shedding DNA.

What Comes Next for OncRNA Research

Important biological and clinical questions remain. How do functional oncRNAs exert their effects? Do they interact with proteins or with other RNAs? Could tracking oncRNA changes in real time guide treatment decisions? Might they help detect recurrence earlier or improve patient stratification? Answering these questions will require more extensive research and larger prospective clinical trials.

At the same time, translation is already underway. The discovery that oncRNAs generate cancer-specific signals in blood is moving toward clinical application. We are collaborating with the biotech company Exai Bio (Hani is a co-founder) to develop oncRNA-based diagnostics. The company has been building artificial intelligence models and assembling diverse datasets to improve cancer detection and classification.

Translational research depends on many contributors. When analyzing tens of thousands of samples computationally, it is easy to forget that each one represents a person who volunteered for research, donated blood, and hoped their participation would help others. Honoring those contributions through careful and rigorous science motivates our entire team.

We believe oncRNAs represent a newly recognized class of cancer-emergent molecules that function both as drivers of disease and as biomarkers. By making this resource openly available, we hope to accelerate progress and open new avenues of research in cancer biology.