Ewing sarcoma survival splits by neighborhood, not mutation: a 135-patient spatial map yields four prognostic types, with interferon-reactive tumors faring best

Ewing sarcoma looks histologically uniform and carries few driver mutations, yet its outcome differences are now tied to how tumor cells and surrounding cells are arranged in space, not only to genetic change in the tumor cells. Prognosis had been explained mainly by secondary mutations such as STAG2 and TP53 and by chromosomal gains, which cover only a fraction of patients and leave most clinical heterogeneity unexplained; bulk immune deconvolution and single-cell studies of the microenvironment had produced only suggestive results, with no clinically actionable microenvironment marker. Across samples from 135 patients, spatial analysis yielded eight recurrent niches, and four patient types defined by niche composition showed distinct overall survival, a stratification reproduced in two independent bulk transcriptomic cohorts.

Tumor cells do not sit in a single transcriptional guise but distribute across five recurrent states, and these states are bound to particular combinations of neighboring cells that form recognizable tissue regions. Intratumor heterogeneity in Ewing sarcoma had been discussed mainly in terms of high versus low EWSR1::FLI1 fusion activity, and whether fusion activity was the sole determinant remained unclear. Consensus non-negative matrix factorization on tumor cells of each sample, clustered across samples, gave five metaprograms covering fusion activity, cell cycle, stress and hypoxia, protein regulation, and interferon response; four niches were each dominated by one tumor-cell state, while the other four were characterized by microenvironment composition.

The Immune Reactive type is a previously unrecognized class in which tumor cells themselves carry an interferon-activated state and form local contact zones with immune cells. Ewing sarcoma is generally regarded as an immunologically cold tumor, and C1QC+ macrophages are usually linked to poor prognosis, whereas here they show an antigen-presenting, interferon-activated phenotype. Type-I interferon response in tumor cells rose with the number of surrounding C1QC+ and SPP1+ macrophages, T cells were the only lymphocyte population positively associated with both type-I and type-II responses, and in samples where the Immune Reactive and Active Immune niches co-localized, type-I interferon scores and JAK-STAT activity were elevated in the contact regions.

The four types can be partially reconstructed from routine bulk transcriptomes and carry prognostic information independent of established clinical risk factors. Spatial transcriptomics is costly and limited in throughput, so a classification that only works on spatial platforms would be hard to place in routine stratification. On 38 held-out matched samples, a nearest-neighbor classifier reconstructed the spatially defined types with 60.5% accuracy against a permutation null of 24.9%; in Cox models across three cohorts, Immune Reactive versus Proliferative Ewing sarcoma gave a hazard ratio of 0.28 (95% CI 0.12–0.64) and Fusion Active 0.41 (0.17–0.96), comparable in magnitude to and independent of metastasis at diagnosis.

The conclusions rest on spatial transcriptomic data from 135 Ewing sarcoma samples, most of them treatment-naive diagnostic specimens, with relapse or post-treatment samples excluded from survival analysis; the survival differences of the four types were reproduced in a bulk RNA-Seq cohort of 142 patients at initial diagnosis and a gene-expression microarray cohort of 171 patients at initial diagnosis. The report suits readers interested in sarcoma microenvironments, spatial omics and prognostic stratification.

The FGFR2–FGF7 interaction between fibroblasts and tumor cells and its link to reduced fusion activity remain observational and require in vitro and in vivo functional assays; the targeted panel covers only preselected genes, so precise molecular crosstalk between immune cells and immune-reactive tumor cells cannot be inferred, and whether macrophages or T cells actively shape the tumor immunophenotype or are merely recruited remains unresolved; the spatial map rests on two-dimensional sections, with three-dimensional spatial multi-omics still to come; and the utility of the typing in routine clinical stratification still needs prospective validation in larger multicenter cohorts.