Noninvasive assessment of core metastatic genes in lung adenocarcinoma: development of a predictive model integrating single-cell transcriptomics and radiomics

TL;DR

This study combines single-cell RNA data and CT scans to identify genes linked to lung cancer metastasis and develop a noninvasive model for predicting their expression.

Contribution

A novel noninvasive CT radiomics model is developed to predict metastatic gene expression in lung adenocarcinoma.

Findings

PSMB5, PSMB7, and SLC16A3 are overexpressed in metastatic lung adenocarcinoma and linked to poor prognosis.

A CT radiomics model predicted the combined expression of these genes with an AUC of 0.765.

The genes synergistically promote tumor progression through cell cycle, proteasome activity, and metabolism.

Abstract

Lung adenocarcinoma (LUAD) leads to death primarily due to its high metastatic potential. Risk assessment methodologies currently predicated on histopathological and imaging features possess a limited capacity to predict metastatic potential. Therefore, integrating single-cell transcriptomics and CT radiomics to identify key molecular drivers of metastasis and establishing a noninvasive imaging prediction model for LUAD is important.

Bulk transcriptomic data and single-cell RNA sequencing (scRNA-seq) data were obtained from public database for analysis. Analytical tools (Seurat, inferCNV, Monocle, WGCNA, LASSO regression, GO/KEGG/GSEA, CellChat) were used for cellular profiling, trajectory analysis, gene identification, functional enrichment, and cell–cell communication. Immunohistochemistry (IHC) and RT-qPCR validated candidate genes at protein and mRNA levels. Additionally, a CT radiomics-based predictive model was developed for noninvasive gene expression assessment.

ScRNA-seq analysis revealed a malignant cellular trajectory from primary to metastatic LUAD and identified a metastasis-associated subpopulation. Three consistently overexpressed genes (PSMB5, PSMB7 and SLC16A3) were correlated with poor prognosis. Functional studies indicated their synergistic roles in promoting tumor progression through cell cycle regulation, proteasome activity, and metabolic reprogramming. A CT radiomics model effectively predicted the combined expression of these genes (AUC = 0.765), linking imaging features to molecular phenotypes.

This study reveals that the synergistic expression pattern of PSMB5, PSMB7 and SLC16A3 is closely associated with lung adenocarcinoma metastasis and poor prognosis, confirming their potential value as prognostic biomarkers and therapeutic targets. The CT radiomics model offers a noninvasive tool for molecular phenotyping, aiding in preoperative precision assessment and advancing noninvasive clinical decision-making for LUAD.

Figures