Researchers at Stanford Medicine and the Mayo Clinic have developed a non-invasive blood test capable of mapping the tumour microenvironment, the ecosystem of healthy cells surrounding a tumour that significantly influences how cancers grow and respond to treatment. The findings, published in Nature, represent the first method for studying tumour spatial architecture without tissue biopsy.
Using machine learning tools developed at Stanford, the team analysed gene expression patterns across more than 100 tumour specimens from ten cancer types, including carcinomas and melanomas. Carcinomas account for 80 to 90% of human cancers. The analysis identified nine distinct cellular neighbourhoods, termed spatial ecotypes, each approximately the diameter of a human hair.
These neighbourhoods were found to be consistent across cancer types, with certain ecotypes appearing preferentially at the tumour border and others deeper within the tumour mass. Several correlated with whether a tumour would respond to immunotherapy.
To make this information accessible without a biopsy, the researchers developed a tool called Liquid EcoTyper, which uses artificial intelligence to reconstruct spatial ecotype patterns from methylation signatures in cell-free DNA circulating in the blood. When cells die and release DNA into the bloodstream, chemical markers known as methyl groups indicate which genes were active in the originating cell, allowing the researchers to infer cellular neighbourhoods from a simple blood draw. Because the test can be repeated over time, it offers the potential for real-time monitoring of how the tumour microenvironment evolves during treatment, something biopsy-based approaches cannot practically provide.
The authors suggest this could eventually help clinicians select initial therapies and identify when switching treatment may be appropriate. Further studies are required before the test is approved for routine clinical use.
Source: Zhang W et al. Non-invasive profiling of the tumour microenvironment with spatial ecotypes. Nature (2026). DOI: 10.1038/s41586-026-10452-4
Using machine learning tools developed at Stanford, the team analysed gene expression patterns across more than 100 tumour specimens from ten cancer types, including carcinomas and melanomas. Carcinomas account for 80 to 90% of human cancers. The analysis identified nine distinct cellular neighbourhoods, termed spatial ecotypes, each approximately the diameter of a human hair.
These neighbourhoods were found to be consistent across cancer types, with certain ecotypes appearing preferentially at the tumour border and others deeper within the tumour mass. Several correlated with whether a tumour would respond to immunotherapy.
To make this information accessible without a biopsy, the researchers developed a tool called Liquid EcoTyper, which uses artificial intelligence to reconstruct spatial ecotype patterns from methylation signatures in cell-free DNA circulating in the blood. When cells die and release DNA into the bloodstream, chemical markers known as methyl groups indicate which genes were active in the originating cell, allowing the researchers to infer cellular neighbourhoods from a simple blood draw. Because the test can be repeated over time, it offers the potential for real-time monitoring of how the tumour microenvironment evolves during treatment, something biopsy-based approaches cannot practically provide.
The authors suggest this could eventually help clinicians select initial therapies and identify when switching treatment may be appropriate. Further studies are required before the test is approved for routine clinical use.
Source: Zhang W et al. Non-invasive profiling of the tumour microenvironment with spatial ecotypes. Nature (2026). DOI: 10.1038/s41586-026-10452-4
