We knew we had something interesting with T3p, a single small RNA found in breast cancer but absent from normal tissue. After being described in 2018 , this molecule took our team on a six-year journey to systematically map orphan non-coding RNAs (oncRNAs) across all major cancer types, understand which ones actually drive disease, and demonstrate their utility in monitoring patients through simple blood tests.
In a paper published today , we show how we went from mining cancer genome data to building machine learning classifiers, developing large-scale functional screens in mice, and finally validating these RNAs as blood-based biomarkers in nearly 200 breast cancer patients.
Orphan RNAs exist everywhere in cancer
Our first major finding was that what we saw in breast cancer wasn't unique. We systematically analyzed small RNA sequencing data from The Cancer Genome Atlas across 32 cancer types and found approximately 260,000 of these cancer-specific small RNAs—which we call oncRNAs—distributed across every cancer we examined.
The appearance of these RNAs isn't random and each cancer type produces a distinct pattern of oncRNA expression. For example, lung cancer cells express a different set of oncRNAs than breast cancers. These patterns can be leveraged by machine learning models to classify cancer type identities with 90.9% accuracy. We validated these patterns in an independent cohort of 938 tumors, achieving 82.1% accuracy.
Variations also exist within specific cancers: Basal breast tumors produce different oncRNA patterns than luminal tumors, indicating there may be subdivisions we don't even know about yet. This suggests that these molecules are telling us something fundamental about cancer cell state. In other words, patterns of oncRNA presence and absence serve as "digital molecular barcodes" that capture cancer cell identity, from cancer type to subtype and even cellular states.
Some oncRNAs actually drive cancer
While this new set of cancer biomarkers turns out to be very useful, we were also interested in knowing whether cancer cells can adopt some of these cancer emergent RNA species to engineer new oncogenic pathways.
To find out, we created large-scale screening libraries encoding roughly 400 oncRNAs across breast, colon, lung, and prostate tumors and introduced them into cancer cells using lentiviral vectors. For half, we overexpressed the oncRNAs. For the other half, we used "Tough Decoy" constructs to knock them down. We then injected these cells into mice and looked for which oncRNAs gave cells a competitive advantage during tumor growth.
About 5% showed clear phenotypes in xenograft mouse models. We validated two breast cancer oncRNAs in detail. One promoted epithelial-mesenchymal transition, a critical process for cancer progression and metastasis. The other activated E2F target genes, driving proliferation. Both significantly increased tumor growth and metastatic colonization in independent cell line models.
When we looked back at patient tumor data, we found that tumors expressing these same oncRNAs showed similar pathway modulations. Seeing the same biological patterns in TCGA samples and our own models gave us confidence in our results.
They're secreted into blood
Most clinically exciting is that cancer cells release these RNAs into the bloodstream, and tracking them can tell you how patients are doing.
We profiled cell-free RNA from 25 cancer cell lines across 9 tissue types and found that about 30% of oncRNAs are actively secreted. To validate, we analyzed serum from 192 breast cancer patients in the I-SPY 2 neoadjuvant chemotherapy trial , collecting blood before and after treatment and calculated the change in total oncRNA burden (ΔoncRNA below).
.That simple metric was remarkably predictive. Patients with high residual oncRNA after chemotherapy had nearly 4-fold worse overall survival. This remained true even when we controlled for standard clinical measures like pathologic complete response and residual cancer burden.
This was our moonshot. We knew these RNAs showed up in blood, but whether they'd be informative in actual patient samples was unknown. The fact that we needed only 1 milliliter of serum and still got a strong signal was surprising.
This addresses a real clinical problem: monitoring minimal residual disease in breast cancer with markers such as cell-free DNA is challenging because tumors don't shed much DNA, especially in early stages. RNA-based monitoring might offer a way forward because cancer cells actively secrete RNA rather than passively shedding DNA.
What's next?
There are still big open biological and clinical questions. How exactly do these functional oncRNAs work? Are they interacting with proteins? With other RNAs? Could oncRNA dynamics be used in real-time to guide treatment decisions? Could it catch cancer recurrence earlier? Could they help clinicians better stratify patients to guide treatment?These will require deeper study and larger, prospective trials to answer.
But the core findings that oncRNAs provide cancer-specific signals in blood are already moving toward translation. We're closely involved with a biotech company Exai Bio (Hani is a co-founder) to develop oncRNA-based diagnostics, and the company has been building AI models and diverse datasets for cancer detection and classification.
Any good translational work really does take a village, but when you're processing tens of thousands of samples computationally, it's easy to lose sight of the fact that each one represents a patient's story. These samples come from people who enrolled in trials, gave their blood, and participated in research hoping it would help others. Being able to honor that through rigorous science is what motivates our whole team.
We think oncRNAs represent a new class of cancer-emergent molecules that function as both drivers and biomarkers. We hope that this resource, which we've made open source, opens new directions for the field.
