RNA foundation model identifies candidate resistance mechanisms in BRAF V600E colorectal cancer from blood samples

Study with Eigen Bio finds that longitudinal blood RNA can reveal biology underlying treatment response and resistance in BRAF V600E colorectal cancer.

NEW YORK, NY, UNITED STATES, October 22, 2026 /EINPresswire.com/ — An RNA foundation model applied to longitudinal blood samples from patients with microsatellite-stable (MSS), BRAF V600E metastatic colorectal cancer identified biological states associated with treatment response and surfaced candidate mechanisms of resistance, without using clinical response labels or predefined response signatures to identify those states.

The study conducted by The University of Texas MD Anderson in collaboration with Eigen Bio and to be presented at the European Society for Medical Oncology (ESMO) Congress 2026, suggests that serial extracellular vesicle RNA (evRNA) may offer a non-invasive window into why patients with the same molecularly defined cancer can respond so differently to treatment.

This question has taken on new relevance following results from the randomized phase II SWOG S2107 trial, which tested encorafenib and cetuximab with or without nivolumab in previously treated patients with MSS, BRAF V600E metastatic colorectal cancer. Adding nivolumab did not improve progression-free or overall survival across the study population, underscoring the biological heterogeneity that remains even among patients whose cancers share the same defining BRAF V600E alteration.

Those results followed earlier work led by UT MD Anderson investigator Van Morris, M.D., whose single-institution phase 1/2 study of encorafenib, cetuximab and nivolumab showed a 50% objective response rate in this difficult-to-treat population. Morris and colleagues had already begun asking why some patients benefited while others did not: their analysis of longitudinal evRNA showed that patients with sustained treatment benefit had distinct transcriptional patterns, suggesting that serial blood samples could capture biology associated with treatment response.

The new study builds on that observation. Researchers applied Eigen-1, Eigen Bio’s RNA foundation model, to paired pretreatment and on-treatment evRNA profiles from the earlier study. Using an unsupervised approach, Eigen-1 identified transcriptional structure in the samples without access to clinical response labels or predefined response signatures; only afterward were the resulting biological states compared with patients’ RECIST outcomes.

The model independently recovered response-associated biology previously identified. Separately, the transcriptional states identified without response labels showed 81% post-hoc concordance with RECIST-defined clinical response, suggesting that clinically meaningful response differences were reflected in the underlying transcriptional structure.
The model then went beyond the previously reported signatures. By mapping the newly identified transcriptional states onto mechanistic networks, researchers identified additional programs associated with treatment resistance, turning patterns in the RNA into hypotheses about why some tumors may escape therapy and where future therapeutic strategies could intervene.

Together, the results point toward a broader use for longitudinal liquid biopsy: not simply monitoring whether a tumor is responding, but learning from response itself to uncover the biology driving sensitivity and resistance, and potentially generate new therapeutic hypotheses grounded in human biological response.

The findings are hypothesis-generating and will require validation in larger, independent cohorts. Future work will test whether the same biological states recur across additional patients and treatment settings.

Sang Lee
Eigen Bio Inc.
slee@eigenbioai.com
Visit us on social media:
LinkedIn

Media gallery