Decoding early lung adenocarcinoma progression by single-cell and spatial transcriptomics reveals a CMA-related prognostic signature.

BACKGROUND

Lung adenocarcinoma (LUAD) progression from adenocarcinoma in situ (AIS) to minimally invasive adenocarcinoma (MIA) and invasive adenocarcinoma (IAC) is accompanied by molecular heterogeneity and tumor microenvironment remodeling. Chaperone-mediated autophagy (CMA) regulates tumor cell homeostasis, metabolic adaptation, and stress responses, but its dynamic alterations and prognostic significance during the AIS/MIA-to-IAC progression of LUAD remain unclear.

METHODS

We integrated the single-cell transcriptomic dataset GSE189357 and the spatial transcriptomic dataset GSE189487 with bulk transcriptomic data from TCGA-LUAD, GTEx, and the GEO validation cohorts GSE31210 and GSE50081 to characterize CMA-related features during the AIS/MIA-to-IAC progression of LUAD. CMA activity and myeloid remodeling were analyzed at the single-cell and spatial levels. Candidate genes were identified by combining tumor-normal differential expression analysis in TCGA-LUAD with weighted gene co-expression network analysis. Multiple machine learning algorithms were compared to construct and externally validate a prognostic model. Biological and clinical relevance was further assessed through clinicopathological, pathway, immune, cell-cell communication, drug sensitivity, and in vitro analyses.

RESULTS

CMA-related activity showed marked cell-type specificity and spatial heterogeneity during the AIS/MIA-to-IAC progression of LUAD, with the most prominent changes in the myeloid compartment. Myeloid re-clustering revealed enrichment of cDC2 and APOE+ lipid-associated TAMs in IAC, whereas FABP4+ metabolic TAMs and immature neutrophils decreased. By integrating tumor-normal differential expression analysis with weighted gene co-expression network analysis, 122 candidate genes were identified, and a 15-gene CMA-related prognostic signature was established using a random survival forest model. This signature showed robust prognostic stratification in TCGA-LUAD, GSE31210, and GSE50081. The high-risk group had poorer survival, more advanced stage, and enrichment of malignant pathways including GLYCOLYSIS, G2M CHECKPOINT, MTORC1 SIGNALING, E2F TARGETS, and MYC TARGETS. The low-risk group showed higher stromal and immune scores and stronger immune activity. THBS1 signaling was restricted to high-risk epithelial communication, with fibroblasts as the major signal senders. In vitro experiments showed that MGP overexpression inhibited lung cancer cell proliferation, colony formation, migration, and invasion.

CONCLUSIONS

This study characterized CMA-related heterogeneity during LUAD progression from AIS to IAC and established a robust 15-gene prognostic signature. Fibroblast-derived THBS1 signaling and MGP may contribute to the high-risk phenotype and provide insight into early LUAD evolution and risk stratification.