Molecular Classification And Prognosis Prediction Of Colorectal Cancer Based On Gene Expression Profiling

Backgrounds and ObjectivesColorectal cancer is the third most common cancer and the fourth most common cancer cause of death globally.In China,the mortality of colorectal cancer is also increasing,currently ranking between the 4th and the 5th.Despite of clinical and pathological parameters are available for the classification and prognostic stratification of cancer,they may be inadequate due to the great biologic and genetic heterogeneity of colorectal cancer.Molecular biology provides a global and detailed view on the molecular change involved in tumor progression,leading to better understanding of carcinogenesis process,to discovering new prognostic markers and novel therapeutic targets.In this study,the application of gene expression profiling on carcinogenesis studies purposes to identify the specific alteration on gene expression according to tumor development and to diagnose and classify tumors on the basis of molecular features.Furthermore,we tried to use prognostic index(PI)to predict prognosis for patients.And individual hazard model was built to predicted patient's hazard probability.Material and MethodsWe collected 127 paired cancer-normal matched colorectal tissue samples.As result of mRNA sequencing of 6 paired tissue samples among them,97 top candidate different expression genes(DEGs)based on largest fold change between cancer tissues and normal tissues were found.Then we expand samples to validate this result.Consequently,75 DEGs were incorporated into our next research.Cumulative survival rate was calculated using life-table methods.Survival curve was drawn using the univariate survival analysis based on Kaplan-Meier methods.The statistically significant prognostic factors identified by univariate analysis were then analyzed by multivariate Cox proportional hazard model,and backward stepwise method was used to bring variable into model.According to the regression coefficients determined by Cox proportional hazard model,the prognostic index(PI)was calculated.Two-step cluster analysis was used to classify CRC.Comparison of the resultant clusters was made by ?2 test or Fisher's exact test.Receive-operation characteristic(ROC)analysis was used to evaluate the accuracy of PI on predicting prognosis.ROC analysis was used to evaluate the sensitivity and specificity of LASSO regression model and Logistic regression model.ResultsAmong 75 DEGs,18 DEGs were identified as the prognostic factors by univariate survival analysis,including CPNE8,LOC646627,CDKN2A,ATP6V1A,CA1,SCARA5,BEST4,SCNN1B,KLF9,DNMT3B,ANLN,DNMT1,DNMT3A.Only those genes,age and gender were entered into multivariate analysis.Multivariate Cox proportional hazard analysis showed that DNMT3B,LOC646627,SCARA5,CDKN2A,ATP6V1A were independent prognostic factors for survival.18-gene signature including MLH1,PLOD3,TGM2,TP6V1A,SQLE,MET,S100P,MT1M,BEST4,CA7,LOC646627,ANPEP,P2RX1,FOXF2,GAB3,ABI3BP,SCARA5,ADAMDEC1 was identified by LASSO regression model.This 18-gene signature was able to separate normal and tumor tissues with sensitivity 98.43%,specificity96.85%,and 97.6%accuracy.Unsupervised cluster analysis was used to classify 127 CRC tissues.We identified molecular classification based on 75-gene(total DEGs),13-gene(prognostic factors),5-gene(independent prognostic factors).By univariate survival analysis,only classification based on 5-gene had statistical significant on prognosis.PI based on 13-gene(prognostic factors),5-gene(independent prognostic factors)was calculated according to the regression coefficient generated by univariate and multivariate Cox proportional hazard model.By ROC curves analysis,there was no significant difference on accuracy in predicting 1-year,3-year and 5-year survival status between them.But PI based on 5-gene had significant higher accuracy than TNM stage in predicting prognosis.Then we establish PI stage by the cut-off value produced by ROC curve.At last,we built joint predictor by combining 5-gene stage and TNM stage.Because of the addition of 5-gene stage,the area under ROC(AUC)was increased by 20.98%?29.51%?26.77%individually in predicting 1-year,3-year and 5-year survival status.We choose category-free net reclassification improvement(cfNRI)to measure the usefulness of PI stage.The addition of PI stage adds the ability of 1-year,3-year,5-year prognostic prediction by get cfNRI = 0.381,0.507 and 0.465.Conclusion18-gene signature selected by LASSO regression model could separate normal and tumor tissues with high accuracy.Unsupervised classification on gene expression profiling is superior to understanding of carcinogenesis process.Prognostic index has robust performance in prediction 1-year,3-year and 5-year survival status.The PI grade based on prognostic index is more accurate than TNM stage in predicting prognosis.The joint predictor combining PI stage and TNM stage could increase the accuracy in predicting prognosis.