Integrated analysis of single-cell and bulk RNA-sequencing identifies a metastasis-related gene signature for predicting prognosis in lung adenocarcinoma.

Background: Metastasis has been documented as an independent and significant prognostic feature of lung adenocarcinoma (LUAD) patients. However, the underlying genetic and molecular mechanisms responsible for LUAD metastasis and their prognostic significance are not exactly defined.

Methods: The single-cell transcriptomic profiles of primary and metastatic LUAD samples were integrated as a whole dataset. Enrichment analysis and pseudotime trajectory analysis were performed to illustrate the cellular origins and changes during the metastatic process. The LUAD metastasis-related genes (LMRGs) molecular cluster and signature was constructed through unsupervised consensus clustering and ten machine-learning algorithms in The Cancer Genome Atlas (TCGA) LUAD cohort using ten machine-learning algorithms. Validation of the signature was conducted using four independent cohorts from the Gene Expression Omnibus (GEO) database. Kaplan-Meier, ROC, univariate and multivariate Cox-regression analyses were performed to test the stability of the signature. The gene CCT6A was subjected to knockdown, followed by validation through western blot analysis, flow cytometry, wound healing and transwell-migration assays to determine its potential significance.

Results: First, the signaling pathway networks remodeling and metabolic reprogramming were demonstrated to be involved in the metastasis of malignant LUAD cells, which facilitate their extravasation and adaptation to other organs. Furthermore, distinct subtypes of malignant LUAD cells exhibit tissue-specific patterns. Then, two distinct molecular patterns of LMRGs were established, which showed diverse prognoses. A LUAD metastasis-related gene signature (LMRGS) was constructed via a multiple machine-learning-based integrative procedure, which possesses distinctly superior accuracy than most common clinical features and 69 published prognostic signatures. The patients stratified by the signature into high-risk group had a significantly poorer prognosis compared to those in the low-risk group, and this was well validated across different clinical subgroups. In addition, the risk score calculated by LMRGS remained an independent prognostic parameter in both univariate and multivariate Cox regression. Notably, knockdown of CCT6A gene promoted cell apoptosis and decelerated the cell migration obviously.

Conclusion: LMRGS could serve as a novel and promising tool to improve clinical outcomes for individual LUAD patients.