Non-invasive profiling of the tumour microenvironment with spatial ecotypes | Nature

Download PDF

Subjects

- Biomarkers

- Cancer microenvironment

- Genomics

- Immunotherapy

- Machine learning

Abstract

Multicellular programs in the tumour microenvironment (TME) drive cancer pathogenesis and response to therapy but remain challenging to identify and profile clinically1,2,3. Here, we present a machine-learning framework for multi-analyte profiling of spatially dependent cell states and multicellular ecosystems, termed spatial ecotypes (SEs). By integrating over 10 million single-cell and spot-level spatial transcriptomes from diverse human carcinomas and melanomas, we identified nine SEs with broad conservation, each of which has unique biology, geospatial features and clinical outcome associations, including several linked to immunotherapy response. Notably, SEs were distinguishable by DNA methylation profiling and were recoverable from plasma cell-free DNA (cfDNA) using deep learning. In cfDNA from nearly 100 patients with melanoma, SE levels exhibited striking associations with immunotherapy response. Our data reveal fundamental units of TME organization and demonstrate a multimodal platform for profiling solid and liquid TMEs, with implications for improved risk stratification and therapy personalization.

Main

Multicellular ecosystems are fundamental units of tissue organization and key elements of phenotypic variation. In cancer, such ecosystems—arising from immune, stromal and/or malignant cells—form dynamic signalling hubs that powerfully influence disease progression, immune evasion and response to therapy1,2,3. Although single-cell and bulk genomics studies have revealed crucial insights into multicellular ecosystems in cancer1,2,3,4,5,6,7, also known as tumour ecotypes, relatively little is known about their phenotypic diversity in relation to key geographical features in the TME. An unbiased pan-cancer survey of spatially defined cell states, their patterns of co-association into SEs8 and their clinical relationships would illuminate TME biogeography and aid the discovery of improved diagnostics and therapeutic targets9.

Two main factors currently hinder the identification and clinical application of spatially resolved ecotypes in cancer. First, SEs are challenging to profile using existing methods, which are either limited in breadth to a modest number of predefined markers (for example, multiplexed protein imaging)1,5,10, ignore spatial information (for example, EcoTyper)4,6,7 or are unable to perform integrative analyses across diverse samples, cancer types and genomic platforms11,12,13. A second key challenge is the requirement for invasive tumour biopsies to analyse SEs in clinical settings. In particular, solid-tumour biospecimens are subject to considerable sampling bias and are generally restricted to a single diagnostic biopsy14,15. Although cfDNA has emerged as a promising non-invasive analyte with the potential to address these problems16,17, no liquid biopsy assay has yet been described for non-invasive TME assessment.

Here, we introduce a new multimodal framework for solid and liquid profiling of SEs in the TME (Fig. 1a). Our approach combines data fusion, statistical learning and deep learning to overcome critical barriers in both the detection and the recovery of SEs across genomic platforms and bodily compartments. Starting with over 10 million single-cell and spot-level spatial transcriptomes from human carcinomas and melanomas, we identified nine SEs with highly conserved cellular, spatial and clinical features across cancer types, including several that are predictive of response to immune checkpoint inhibitors (ICIs). We then tested whether SEs are recoverable by liquid biopsy. From whole-genome cfDNA methylation profiles of nearly 100 patients with melanoma, we observed striking concordance between plasma-derived SE levels, tumour biopsy-confirmed SE levels and known ICI outcomes. Our study demonstrates a unified approach for granular spatial profiling of the TME, with implications for improved forecasting and spatiotemporal monitoring of therapy response.

Fig. 1: Multimodal profiling of SEs in human cancer.The alternative text for this image may have been generated using AI.Full size image

a, Schematic description of the study. Top: discovery and clinical characterization of spatially colocalized cell states in human tumours (SEs). Bottom: recovery of SEs in plasma cell-free DNA and the use of non-invasive SE profiling for immunotherapy response assessment. b, Compendium of human tumour ST samples (left) and single-cell-scale expression profiles (right) curated and analysed in this work (Supplementary Tables 1 and 2). Inner and outer rings denote platform and cancer type proportions, respectively. Tumour-type abbreviations are defined in Supplementary Table 1. c, Main cell types and key geographic regions in representative breast cancer specimens profiled by MERSCOPE (left; n = 365,811 cells) and 10x Genomics Visium (right; n = 16,860 cells). The