A public-data spatial transcriptomic framework for invasive lung adenocarcinoma tumor ecosystems and prognosis
Background Public transcriptomic cohorts rarely contain audited spread through air spaces (STAS) labels or recurrence endpoints. We therefore evaluated a transparent invasive lung adenocarcinoma (LUAD) tumor-ecosystem score as an exploratory cross-cohort framework rather than a clinical STAS predictor. Methods We integrated GSE131907 single-cell RNA sequencing, E-MTAB-13530 spatial transcriptomics, an exploratory HRA001238/ST_LUAD processed subtype resource, and TCGA-LUAD bulk RNA sequencing. A fixed, equal-weight compact tumor-ecosystem score combined tumor epithelial, myofibroblast, invasive epithelial, CAF/ECM, and proliferation programs while subtracting alveolar differentiation. The formula was not optimized against survival, and no external preregistration or timestamped commit was available. Analyses benchmarked the score against ESTIMATE stromal and immune scores, an eight-collagen average, individual score components, and rank-based single-sample gene-set enrichment scoring (ssGSEA). Spatial sensitivity analyses accounted for repeated sections within patients, and TCGA models compared purity adjustment within an identical 155-patient subset. Results The GSE131907 reference contained 82,884 cells. The 40 E-MTAB-13530 sections came from eight patients and two donors. The mean-expression compact score differed by tissue class (Kruskal-Wallis H(2)=30.30, P = 2.63e-07); all eight patients had a higher patient-mean score in tumor than matched adjacent lung (paired Wilcoxon P = 0.0078), and the tumor coefficient remained positive with patient-clustered standard errors (6.27, 95% CI 4.22–8.31, P = 1.99e-09). Rank-based ssGSEA reproduced the spatial finding. The HRA001238/ST_LUAD proxy analysis was non-significant and provided no support for the primary framework. In TCGA-LUAD, the compact score was associated with overall survival after adjustment for age, stage, and sex (n = 492; HR per SD = 1.34, 95% CI 1.15–1.56; P = 1.58e-04). ESTIMATE StromalScore and the collagen average were not significant in corresponding models, and adding the compact score improved model fit. Within the identical 155-patient purity-complete subset, the score HR was 1.14 (95% CI 0.90–1.46; P = 0.282) without purity and 1.27 (95% CI 0.96–1.68; P = 0.098) with purity. Conclusions The compact score is not reducible to a generic ESTIMATE stromal score, a simple collagen average, or one component, and its cross-method reproducibility supports a coordinated multi-program signal. Nevertheless, tumor-purity independence is not established, thematic circularity with stromal biology remains, and no analyzed cohort validates STAS prediction. The contribution is therefore an interpretable, benchmarked public-data framework with explicit compositional limits, not a clinical biomarker.
Authors
- 曲明江
- Chunyi Jia (ORCID: https://orcid.org/0000-0002-6598-6902)
- Qinguang Zou
- Xuejuan Gong
- Yinghao Chang
Institutions
- Jilin Province Tumor Hospital (CN)
Publication Details
- Journal
- PLoS ONE
- Published
- 2026-11-19
- DOI
- https://doi.org/10.1371/journal.pone.0357864
- Primary Topic
- Single-cell and spatial transcriptomics
- Type
- article
- Field-Weighted Citation Impact
- 0.00