Abstract

Purpose Depression is acknowledged to correlate with the occurrence and progression of multiple cancers. However, no study has yet systematically complied depression-related genes to construct a prognostic signature for lung adenocarcinoma (LUAD). Methods Our study encompasses 1,276 LUAD patients from three cohorts. Consensus clustering was employed to classify patients into different depression subtypes. Then, a variety of machine-learning algorithms were utilized to construct a robust depression-related signature (DRS). Thereafter, a nomogram combining DRS with common clinical characteristics was established for prognosis. The IOBR package was used to quantify the immune landscape, whereas the oncoPredict and Connectivity Map algorithms were employed to evaluate therapeutic response. The Seurat package was applied to process single-cell data, and the Scissor algorithm was used to identify depression-associated cells. Ultimately, depression-like mouse models were constructed to detect alternations in depression-related genes. In vitro experiments were performed to explore the role of PSEN1 in the malignant behaviors of LUAD. Results Unsupervised clustering stratified patients into two subtypes with distinct features. DRS consisting of 14 hub depression-related genes was established using the LASSO + GBM algorithm and served as an independent prognostic indicator. The nomogram constructed with DRS demonstrated robust predictive efficacy, with a C-index of 0.778. LUAD patients in the high-risk group exhibited weaker “immune hot” features and reduced responsiveness to immunotherapy. Additionally, high-risk patients were less sensitive to conventional chemotherapy and targeted therapies. Single-cell analysis revealed that depression-associated high-risk cells displayed more malignant characteristics. Finally, qRT-PCR validated the alternations of depression-related genes in depression-like mouse models, and in vitro experiments confirmed that PSEN1 facilitated