One of the most common sites of extra-thoracic distant metastasis of nonsmall-cell lung cancer is the brain. Our study was performed to discover genes associated with postoperative brain metastasis in operable lung adenocarcinoma (LUAD). RNA seq was performed in specimens of primary LUAD from seven patients with brain metastases and 45 patients without recurrence. Immunohistochemical (IHC) assays of the differentially expressed genes were conducted in 272 surgical-resected LUAD specimens. LASSO Cox regression was used to filter genes related to brain metastasis and construct brain metastasis score (BMS). GSE31210 and GSE50081 were used as validation datasets of the model. Gene Set Enrichment Analysis was performed in patients stratified by risk of brain metastasis in the TCGA database. Through the initial screening, eight genes (CDK1, KPNA2, KIF11, ASPM, CEP55, HJURP, TYMS and TTK) were selected for IHC analyses. The BMS based on protein expression levels of five genes (TYMS, CDK1, HJURP, CEP55 and KIF11) was highly predictive of brain metastasis in our cohort (12-month AUC: 0.791, 36-month AUC: 0.766, 60-month AUC: 0.812). The validation of BMS on overall survival of GSE31210 and GSE50081 also showed excellent predictive value (GSE31210, 12-month AUC: 0.682, 36-month AUC: 0.713, 60-month AUC: 0.762; GSE50081, 12-month AUC: 0.706, 36-month AUC: 0.700, 60-month AUC: 0.724). Further analyses showed high BMS was associated with pathways of cell cycle and DNA repair. A five-gene predictive model exhibits potential clinical utility for the prediction of postoperative brain metastasis and the individual management of patients with LUAD after radical resection.
Keywords: RNA-seq; brain metastasis; immunohistochemical assays; lung adenocarcinoma; predictive models.
© 2020 UICC.