Abstract Spatial transcriptomics enables the detection of rare or transitional tumor states not captured by bulk transcriptomics or immunohistochemistry (IHC). However, translating these spatially defined states into clinically relevant biomarkers is challenging, because signals from minor populations are underrepresented or masked in bulk data. Lung adenosquamous carcinoma (ASC), containing intermixed adenocarcinoma and squamous components, provides a model to study lineage transitions and poorly differentiated states unresolved in bulk datasets. Spatial transcriptomic profiling of ASC was integrated with TTF-1 and p40 IHC to detect tumor populations lacking these markers. Gene signatures from IHC-negative populations were projected onto bulk lung adenocarcinoma datasets to assess prognostic relevance. We implemented a color-space uniform manifold approximation and projection visualization (RGB-UMAP) to enhance the spatial mapping of rare transcriptional states that were unresolved by conventional clustering. We identified a TTF-1/p40-negative tumor cell population with a hybrid adenocarcinoma–squamous expression pattern and upregulation of SLC2A1 ( GLUT1 ). RGB-UMAP clarified these states and distinguished them from normal epithelial cells. In bulk lung adenocarcinoma cohorts, high SLC2A1 expression stratified patients with poorer survival. This study demonstrates a bottom-up biomarker discovery strategy translating spatially defined, IHC-negative tumor cell populations in ASC into stratifiers for lung adenocarcinoma.
Hatakeyama et al. (Fri,) studied this question.