Micro-metastasis And Drug Resistance Predictive Signatures And Reproducible Multi-omics Characteristics For Gastric Cancer

Tumor metastasis(including micro-metastasis)and drug resistance are the main factors resulting in the poor prognosis of gastric cancer patients.In this study,we identified the micro-metastasis and drug resistance predictive signatures and reproducible multi-omics characteristics for gastric cancer.1.Identification of micro-metastasis and postoperative recurrence risk predictive signature and metastasis-related multi-omics characteristics for gastric cancerThe early gastric cancer patients after surgery have a high rate of recurrence.Besides the failure of surgery and low possible tumor regrowth,the main reason of recurrence is the micro-metastasis which is hard to be detected by the clinical imaging and pathological examination.Thus,it is urgent to develop a micro-metastasis and postoperative recurrence risk predictive signature.Additionally,it is hard to identify independently and reproducibly differential expression genes(DEGs)between the metastatic and non-metastatic samples of patients classified by the TNM stage which has high rates of false negative and false positive.Thus,we made the following analyses.(1)Identification of micro-metastasis and postoperative recurrence risk predictive signature: Firstly,using the gene expression profiles of 77 early gastric cancer patients only treated by surgery,we identified a signature consisting of 19 gene pairs involving37 genes such as COL8A1,PTPRE and NALCN,denoted as 19-GPS,whose within-sample relative expression orderings(REOs)could robustly predict the micro-metastasis and postoperative recurrence risk for early gastric cancer through the majority voting rule.The signature was validated in two independent datasets including71 and 46 samples,respectively.(2)Identification of reproducibly metastasis-related multi-omics characteristics: Inabove three independent datasets,with the help of 19-GPS,we excluded the potential false negative(N0M0 samples with high recurrence risk)and false positive(N+/M1 samples with low recurrence risk)metastasis samples and then identified thousands of DEGs between the metastatic and non-metastatic samples in each dataset(limma,FDR<5%).The DEGs extracted from the three independent datasets had high reproducibility.However,we could not detect reproducible DEGs only based on the TNM stage.Combing the 19-GPS and TNM stage,we reclassified the TCGA gastric cancer samples into metastatic and non-metastatic groups.Based on the TCGA multi-omics data,we found the Cp G sites located at the promoter regions of 599 genes such as SERPING1,THBS4 and PODN in the metastatic group were hypomethylated(Wilcoxon rank-sum test,FDR<5%)and the expression of these genes were up-regulated.These genes were significantly enriched in the metastasis-related pathways,such as c AMP and PI3K-Akt signaling pathways.However,no difference of copy number alteration and somatic mutation rate was observed between the two groups(Fisher’s exact test,FDR<5%).2.Identification of 5-FU resistance predictive signature and resistance-related multi-omics characteristics for gastric cancer5-Fluorouracil(5-FU)-based chemotherapy is currently the first-line treatment for gastric cancer.However,the overall response rate is only about 20-40%.Meanwhile,it is hard to identify the drug resistance-related genes due to the weak signal and these genes often fail to be validated in the independent datasets.Thus,we made the following two analyses.(1)Identification of the robust 5-FU resistance predictive signature: Firstly,using the expression profiles of 27 gastric cancer cell lines with 5-FU sensitivity and 35 gastric cancer patients treated by 5-FU-based chemotherapy,we identified a 5-FU resistance predictive signature consisting of two gene pairs(KCNE2 and API5;KCNE2and PRPF3).This signature was validated in an independent dataset including 123 samples,demonstrating that this signature could recognized the patients who could not benefit from 5-FU chemotherapy.(2)Identification of 5-FU resistance-related multi-omics characteristics: Using the5-FU resistance predictive signature,we classified the samples from three independent datasets including TCGA data into 5-FU resistant and sensitive groups.Then,we detected thousands of DEGs between the two groups in each dataset(limma,FDR<20%).The DEGs extracted from the three independent datasets had high reproducibility.Based on the TCGA multi-omics data,we found the Cp G sites located at the promoter regions of 327 genes in the resistant group were hypermethylated and the expression of these genes were down-regulated.Among these genes,19 genes such as PDE1 C,GMPR and CBS were involved in the pyrimidine metabolism and folate metabolism pathways,suggesting the status of DNA hypermethylation was closely associated with 5-FU therapy resistance.Meanwhile,seven regions in the resistant group had significantly high frequencies of copy number gain(Fisher’s exact test,FDR<20%).11 genes such as PMS2,HMGB1 and RFC3 located in these regions were involved in the DNA repair,cell cycle and apoptosis pathways.The mean number of somatic mutation genes in the sensitive group was 366.69 which was significantly higher than the mean number of somatic mutation genes(305.32)in the resistant group(Wilcoxon rank-sum test,p=3.75E-02),suggesting that the patients with instable genome were sensitive to 5-FU chemotherapy.3.New strategy for analyzing the expression profiles of tumor cell line modelsThe tumor cell line experiments,such as gene knockdown,gene transfection and drug treatment,are the commonly used models to validate the function of the metastasis or drug resistance signature genes.To detect the DEGs in small-scale cell line experiments,usually with only two or three technical replicates for each state,the commonly used statistical methods lack statistical power or have a high rate of false positive.Here,we proposed a new REOs-based algorithm for identifying DEGs in such small-scale cell line data.In both simulated and real data,the new algorithm exhibitedhigh precision.In conclusion,we identified a micro-metastasis and postoperative recurrence risk predictive signature and a 5-FU resistance predictive signature for gastric cancer,respectively.With the help of the two signatures,we identified reproducibly metastasis-related and 5-FU resistance-related multi-omics characteristics.We also developed a new algorithm for analyzing the gene expression profiles of small-scale cell line experiments,which provided the basis of validating the function of the metastasis or drug resistance signature genes using the tumor cell line models.