A machine-learning approach that reads tumor microenvironment patterns from blood-derived DNA has shown promise for predicting immunotherapy response in melanoma and could expand the role of liquid biopsy in cancer diagnostics.
Researchers reporting in Naturedescribed a framework that identifies “spatial ecotypes” — recurring groups of immune, stromal, and tumor-associated cells that occupy specific regions within tumors. By analyzing more than 10 million single-cell and spatial transcriptomic profiles from 132 tumor samples across 10 cancer types, the team defined nine conserved ecotypes linked to tumor biology, survival outcomes, and response to immune checkpoint inhibitors.
The study’s main diagnostic advance was the ability to detect these tumor microenvironment signatures from plasma cell-free DNA (cfDNA). Using a deep-learning system called Liquid EcoTyper, investigators inferred tumor ecotypes from cfDNA methylation patterns in patients with melanoma. The liquid biopsy results closely matched tissue-based spatial transcriptomic findings and showed strong associations with immunotherapy outcomes.
Senior research Aadel Chaudhuri, Professor of Radiation Oncology at Mayo Clinic, said, "This is the first time we've been able to noninvasively profile the tumor microenvironment at this level."
Several ecotypes correlated with treatment response. High levels of two inflammatory ecotypes, termed SE7 and SE8, were associated with durable benefit and longer survival after immune checkpoint inhibitor therapy, whereas another ecotype, SE4, linked to wound-healing and hypoxia-related stromal activity, was associated with resistance and poorer survival.
Importantly, the cfDNA-based approach outperformed established biomarkers such as tumor mutational burden and PD-L1 expression in several analyses. The authors suggest this reflects the broader biological information captured by spatial ecotypes, which integrate immune, stromal, and vascular features rather than measuring tumor cells alone.
The framework also addresses limitations of conventional tissue biopsy, including sampling bias and the difficulty of repeated tumor collection during treatment. Because cfDNA can be obtained serially from blood samples, the method may support longitudinal monitoring of tumor microenvironment changes during therapy.
"This work opens up an entirely new way of thinking about disease," said Chaudhuri. "We've essentially uncovered a world that was invisible to us before – and now we can access it with a simple blood test."
Although the findings require validation in larger prospective cohorts, the work points toward a future in which liquid biopsy assays provide functional insight into tumor architecture and immune activity, potentially improving patient stratification and treatment monitoring in oncology diagnostics.
Newsletters
Receive the latest pathologist news, personalities, education, and career development – weekly to your inbox.
