0
0Researchers have developed a multimodal machine-learning framework consisting of Spatial EcoTyper and Liquid EcoTyper to identify and profile spatial ecotypes (SEs), which are organized multicellular units within the tumour microenvironment (TME). By analyzing over 10 million cells across various human cancers, the study identified nine conserved SEs with distinct geospatial features and significant clinical outcome associations, including predictions for immunotherapy response. A major breakthrough of this work is the ability to recover these spatial signatures from plasma cell-free DNA (cfDNA) using deep learning, effectively creating a non-invasive liquid biopsy for TME assessment. Testing on patients with melanoma showed that cfDNA-derived SE levels strongly correlate with patient survival and their likelihood of benefiting from immune checkpoint inhibitors. This technology offers a promising alternative to invasive biopsies by providing a more accessible way to personalize cancer treatment and monitor disease progression. Ultimately, the study reveals that these multicellular ecosystems are fundamental units of tissue organization that can be accurately tracked through both solid and liquid analytes.
References:
Zhang W, Brown E L, Usmani A, et al. Non-invasive profiling of the tumour microenvironment with spatial ecotypes[J]. Nature, 2026: 1-12.
