Non-invasive Profiling of Tumour Microenvironment via Spatial Ecotypes
A groundbreaking study published in Nature introduces a multimodal machine-learning framework for non-invasively profiling the tumour microenvironment (TME). Researchers identified nine conserved 'spatial ecotypes' (SEs)—distinct multicellular ecosystems within tumours—by analyzing over 10 million single-cell and spatial transcriptomes from human carcinomas and melanomas. These SEs exhibit unique biological features and correlate with clinical outcomes, particularly responses to immunotherapy. Crucially, the study demonstrates that these spatial profiles can be detected non-invasively through liquid biopsies. By applying deep learning to plasma cell-free DNA (cfDNA) methylation data from nearly 100 melanoma patients, the team successfully recovered SE levels that matched tissue biopsy results and predicted immune checkpoint inhibitor responses. This innovation addresses significant limitations of traditional invasive biopsies, such as sampling bias, offering a unified platform for granular TME assessment. The findings pave the way for improved risk stratification, spatiotemporal monitoring of therapy, and personalized cancer treatment strategies without the need for repeated surgical interventions.
Wire timeline
Non-invasive Profiling of Tumour Microenvironment via Spatial Ecotypes
A groundbreaking study published in Nature introduces a multimodal machine-learning framework for non-invasively profiling the tumour microenvironment (TME). Researchers identified nine conserved 'spatial ecotypes' (SEs)—distinct multicellular ecosystems within tumours—by analyzing over 10 million single-cell and spatial transcriptomes from human carcinomas and melanomas. These SEs exhibit unique biological features and correlate with clinical outcomes, particularly responses to immunotherapy. Crucially, the study demonstrates that these spatial profiles can be detected non-invasively through liquid biopsies. By applying deep learning to plasma cell-free DNA (cfDNA) methylation data from nearly 100 melanoma patients, the team successfully recovered SE levels that matched tissue biopsy results and predicted immune checkpoint inhibitor responses. This innovation addresses significant limitations of traditional invasive biopsies, such as sampling bias, offering a unified platform for granular TME assessment. The findings pave the way for improved risk stratification, spatiotemporal monitoring of therapy, and personalized cancer treatment strategies without the need for repeated surgical interventions.
Nature