Harnessing single-cell and multi-omics insights: STING pathway-based predictive signature for immunotherapy response in lung adenocarcinoma

Abstract

Background: Lung adenocarcinoma is the most prevalent type of small-cell carcinoma, with a poor prognosis. For advanced-stage patients, the efficacy of immunotherapy is suboptimal. The STING signaling pathway plays a pivotal role in the immunotherapy of lung adenocarcinoma; therefore, further investigation into the relationship between the STING pathway and lung adenocarcinoma is warranted.

Methods: We conducted a comprehensive analysis integrating single-cell RNA sequencing (scRNA-seq) data with bulk transcriptomic profiles from public databases (GEO, TCGA). STING pathway-related genes were identified through Genecard database. Advanced bioinformatics analyses using R packages (Seurat, CellChat) revealed transcriptomic heterogeneity, intercellular communication networks, and immune landscape characteristics. We developed a STING pathway-related signature (STINGsig) using 101 machine learning frameworks. The functional significance of ERRFI1, a key component of STINGsig, was validated through mouse models and multicolor flow cytometry, particularly examining its role in enhancing antitumor immunity and potential synergy with α-PD1 therapy.

Results: Our single-cell analysis identified and characterized 15 distinct cell populations, including epithelial cells, macrophages, fibroblasts, T cells, B cells, and endothelial cells, each with unique marker gene profiles. STING pathway activity scoring revealed elevated activation in neutrophils, epithelial cells, B cells, and T cells, contrasting with lower activity in inflammatory macrophages. Cell-cell communication analysis demonstrated enhanced interaction networks in high-STING-score cells, particularly evident in fibroblasts and endothelial cells. The developed STINGsig showed robust prognostic value and revealed distinct immune microenvironment characteristics between risk groups. Notably, ERRFI1 knockdown experiments confirmed its significant role in modulating antitumor immunity and enhancing α-PD1 therapy response.

Conclusion: The STING-related pathway exhibited distinct expression levels across 15 cell populations, with high-score cells showing enhanced tumor-promoting pathways, active immune interactions, and enrichment in fibroblasts and IFI27+ inflammatory macrophages. In contrast, low-score cells were associated with epithelial phenotypes and reduced immune activity. We developed a robust STING pathway-related signature (STINGsig), which identified key prognostic genes and was linked to the immune microenvironment. Through in vivo experiments, we confirmed that knockdown of ERRFI1, a critical gene within the STINGsig, significantly enhances antitumor immunity and synergizes with α-PD1 therapy in a lung cancer model, underscoring its therapeutic potential in modulating immune responses.

Keywords: lung adenocarcinoma; machine learning; non-small cell lung cancer; prognosis signature; single-cell RNA sequencing.

Figure 1

Cell clustering and score of STING Pathway. (A) Cell clustering by Seurat. (B) Integration of 11 samples from lung adenocarcinoma. (C) Annotation of cell clusters. (D) The expression of 14 marker genes in cell clusters. (E) The distribution of high-score cells and low-score cells. (F) Scores of STING Pathway distribution in clusters.

Figure 2

Cell-cell communication analysis between Hight-score and Low-score cells. (A) Number of interactions in Low-score cells. (B) Number of interactions in Hight-score cells. (C) Number of inferred interactions and interaction strength in Hight-score and Low-score cells. (D) Interaction strength of 15 cell clusters in Hight-score and Low-score cells. (E) Differences of Cell-Cell Communication.

Figure 3

Similarities and differences between Hight-score and Low-score cells. (A) The difference of GSVA scores. The correlation of cell clusters in all cells. (B) The correlation of cell clusters in Hight-score cells. (C) The correlation of cell clusters in Low-score cells. (D) Sample distribution across datasets.

Figure 4

Development of a prognostic signature. (A) A consensus deubiquitination-related signature, by 101 machine-learning algorithms. (B–G) Survival curve of high-risk samples and low-risk samples across 6 datasets.

Figure 5

Prediction value and immune microenvironment. (A) ROC curve for 1-year, 3-year, and 5-year survival predictions. (B) PCA of high-risk samples and low-risk samples across 8 datasets.

Figure 6

Construction and validation of STINGsig using machine learning. (A) Immune infiltration analysis of high-risk samples and low-risk samples. (B) Evaluation of the correlation between cancer immune cycle, immunotherapy pathways, and PIS using GSVA. (C) Prediction of IPS for TCGA-LUAD patients using The Cancer Immunome Atlas, consistently indicating higher IPS and greater sensitivity to immunotherapy in the low-risk group. LUAD, lung adenocarcinoma; IPS, immunophenoscore; TCGA, The Cancer Genome Atlas. *** indicates statistical significance with p < 0.001. NS indicates “not significant” (no statistical significance).

Figure 7

ERRFI1 loss enhances anti-tumor immunity and synergizes with α-PD1 therapy in lung cancer model. (A) Schematic of the experimental design. Mice were intravenously injected with M109-Luc cells (3×10^5) on day 0, followed by treatments (Veh., shERRFI1, α-PD1, or Comb.) from day 7 to day 15. Analysis was performed on day 18. (B) Kaplan-Meier survival curves show enhanced survival in mice treated with shERRFI1, α-PD1, or the combination, with the greatest survival benefit observed in the combination group. (C, D) Tumor burden assessed by bioluminescence imaging (photoflux) on day 18. The combination of shERRFI1 and α-PD1 significantly reduced tumor burden compared to individual treatments or vehicle control (****P < 0.0001). (E–H) Flow cytometry analysis of immune cell infiltration in the tumor microenvironment. (E) Percentage of CD45+ leukocytes. (F) Percentage of CD8+ T cells among CD45+ cells. (G) Percentage of TNF-α+ cells among CD8+ T cells. (H) Percentage of GZMB+ cells among CD8+ T cells. * p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001.