| [Background]Gastric cancer（GC）is a malignant neoplasm with insidious onset,rapid progress and poor prognosis,being second in mortality among cancers worldwide.Despite great progress in early screening and the diagnosis and therapy of GC in recent years,the 5-year survival rate for GC remains less than 30%.At present,the TNM staging system commonly used for routine prognostication and clinical guidance of GC patients.However,patients in the same clinical stage often have different prognosis,indicating that GC is a highly heterogeneous neoplasm.Germline genetic variant is a vital basis for intratumor heterogeneity,which may affect not only the rate of tumor progression but also the treatment efficiency of patients.Prior studies have identify genetic variations in several genes（TFF1,CLOCK,etc）are significantly associated with survival of GC patients through candidate gene strategy.Due to a limited number of vairants,insufficient sample size and a lack of validation,and thus,no stable prognositic biomarkers have been yet accepted for GC.Genome-wide association study（GWAS）as an efficient molecular epidemiological research method has been widely used for explore the genetic factors involved in the progression of lung and esophageal cancer.Therefore,the GWAS strategy to explore genomic genetic variations related to the prognosis of GC is expected to provide new ideas for elucidating the heterogeneity of GC and provide genetic biomarkers for its prognosis.[Objective]Overarching goal of this study is to seek potential prognostic biomarkers for GC.To achieve this goal,we performed the first GWAS of GC survival by conducting a meta-analysis of two GWASs in Chinese population.Then,we developed a polygenic hazard score（PHS）based on the significant variants and explore whether the PHS could increase the predictive accuracy for GC prognosis.[Methods]In this prospective follow-up study,we recruited patients from three independent GC cohorts,including Jiangsu cohort,Shanghai cohort and TCGA cohort.All subjects restricted to newly diagnosed cases of gastric adenocarcinoma without distant metastasis,and each patient was surgically resected at the collection hospital.Finally,a total of 2454 GC patients recruited from two independent cohorts（1049 from Jiangsu and 1405 from Shanghai）with complete clinical information and survival outcomes were included.There were 343 GC patients with genotype data and survival information and 341 GC patients with gene expression data and survival information in TCGA cohort.All the patients have completed the genome chip detection,and adopted a uniform quality control and imputation procedure.We conducted genome-wide analysis for association with overall survival（OS）using Cox proportional hazards regression in each cohort separately.Genotype data were analyzed using the additive model with adjustment for age at diagnosis,gender,TNM stage,tumor location,history of chemotherapy or radiotherapy,the first ten principal components（PCs）and genotyping platform.Then,a meta-analysis with fixed-effects model was performed and we selected candidate SNPs that met the following criteria for further replication through two strategies in TCGA cohort:（1）P value for GWAS Meta≤1×10-3;（2）The P value<0.05 for both two cohorts and the effect allele of each study was in the same orientation.In the strategy of candidate variant,we identified those variants which were significantly associated with the prognosis of gastric cancer patients in the TCGA cohort（P<0.05）.In the strategy of candidate target gene,we identified functional regions that may regulate gene expression and considered the association between the expression data of target genes and GC survival as an indirect validation.In order to reduce false-positive associations,each target gene was verified in TCGA and Kaplan-Meier Plotter databases（P<0.05）.Finally,SNPs validated at genotype or target gene levels were considered as significant survival-related SNPs in current study.Based on these significant prognosis-related SNPs,we constructed a weighed polygenic hazard score（PHS）and analysed the association between PHS and overall survival in GC patients.Multivariable Cox regression was applied to select independent prognostic predictors for GC,and then we built a nomogram with both PHS and clinical variables,as well as a clinical model for comparison.The prognostic accuracy was evaluated with the time independent receiver operating characteristic（ROC）curve and areas under the curves（AUCs）at 5 years.Net reclassification improvement（NRI）and integrated discrimination improvement（IDI）were used to compare the performance between nomogram and clinical model.[Results]Through the strategy of candidate genetic variation,a total of 21 variants were directly validated in TCGA patients.Notably,the most significant variant association was observed with rs1618332 at 15q15.1.Compared with wild-type allele C,mutant allele T could significantly increase the risk of death in patients with GC(HR=1.30,95%CI:1.20-1.39,P=4.12×10-8).At the candidate target gene level,we identified functional variants in 2q14.3,5p13.3,9p21.3,9q34.11,11p15.4 and 16p12.3regions might affect the prognosis of GC patients by altering the expression of their target genes CYP27C1,ADAMTS12,ELAVL2,PHYHD1,SCUBE2 and ACSM5,respectively.Finally,we identified a total of 26 significant SNPs through the two strategies.Among them,11p15.4 region was verified at both the genotype and target gene levels.We constructed a weight PHS for each individual based on the 26 significant SNPs,and then patients were divided into low hazard,light medium hazard,medium hazard and high hazard groups with the quartile of PHS.A negative correlation between the PHS levels and overall survival of GC patients in both Jiangsu and Shanghai cohort（P<0.001）.Compared with the low hazard group,the increased risk of death of the light medium hazard group（HR=1.51,95%CI:1.21-1.88）,medium hazard group（HR=2.55,95%CI:2.08-3.13）and high hazard group（HR=3.98,95%CI:3.28-4.83）showed a significant dose-response relationship（P<0.001）.Similarly,a significant dose-response relationship was observed in the TCGA cohort（P<0.001）:the light medium hazard group（HR=1.59,95%CI:0.79-3.17）,medium hazard group（HR=2.60,95%CI:1.38-4.90）and high hazard group（HR=6.35,95%CI:3.48-11.60）.In the combined dataset（Jiangsu and Shanghai cohort）,a nomogram was developed based on four independent prognostic factors including PHS,age at diagnosis,TNM stage and history of chemotherapy or radiotherapy.Our nomogram achieved a C-index of 0.771（95%CI:0.757-0.785）,and the calibration curve displayed good agreement between prediction and observation.The AUC at 5 years of PHS in our combined dataset was 0.670,which was higher than that of age at diagnosis（AUC=0.547）and lower than that of TNM stage（AUC=0.734）.Compared with the clinical model（AUC=0.744）,the AUC at 5 years was improved in our nomogram（AUC=0.814）.When PHS was added into the clinical model,the correct proportion and discrimination ability of the model were improved 8.73%（NRI:95%CI:0.044-0.148,P<0.001）and 9.70%（IDI:95%CI:0.078-0.118,P<0.001）,respectively.In TCGA cohort,the AUC at 5 years of PHS was 0.732,which was higher than that of TNM stage（AUC=0.653）.When combined PHS with clinical variables into a model,the AUC at 5 years was 0.790.[Conclusions]The current study identified 26 GC prognostic genetic variants based on a large sample multicenter cohort,highlighting the importance of germline variant in the progression of GC.The PHS based on genetic loci was an independent prognostic predictor in GC patients.The nomogram incorporating the PHS and clinical prognostic factors might facilitate prognostic assessment and personalized decision-making for GC patients.In addition,our results also provided an important theoretical foundation for the subsequent researches on the role of genetic variation in progression of GC. |
PDF Full Text Request