Study On Feature Integration And Prognostic Model Construction Of Esophageal Cancer Based On Gene Expression And Infiltrating Immune Cells

Background:As one of the most aggressive diseases,esophageal cancer is the sixth leading cause of cancer-related deaths in the world.Although the prognosis of patients diagnosed with esophageal cancer has improved in the past few decades,it is still worse than most other cancers where only 20%of the patients can survive for 5 years.Even for the patients underwent surgery in early stage,there is still a chance of 50%recurrence within 5 years.Clinically,TNM staging is a main predictive index for the survival of patients with esophageal cancer.However,the outcomes of patients at same neoplasm staging who have received similar clinical treatment may be completely different,indicating that this index cannot provide sufficient prognostic information for patients.Therefore,new predictive indicators shall be screened out as supplements of staging system to improve the accuracy of individualized prediction and treatment plan selection.Currently,some studies have looked for potential prognostic indicators of esophageal cancer from the levels of genes and infiltrating immune cells,but there are still some urgent problems to be solved,such as insufficient utilization of negative sample data,few immun e features for prognostic evaluation,and limited prediction accuracy,etc.This study constructed a prognostic model from different perspectives of the biological characteristics of esophageal cancer,and integrated it into the nomogram model,expecting to accurately predict the prognosis of patients with esophageal cancer.Methods:In this study,the gene expression data of esophageal cancer was used to investigate the prognosis of patients.The materials and corresponding clinical information of the samples were downloaded from Gene Expression Omnibu（GEO）and The Cancer Genome Atlas（TCGA）.Then prognosis of esophageal cancer was discussed from two perspectives,including gene expression and immune infiltration,and finally an individualized prediction model of the nomogram was preliminarily constructed for patients with esophageal cancer.The research work was mainly carried out from the following aspects:（1）the construction of risk model based on gene expression data:co-expression network of normal tissues and tumor tissues were constructed by using the algorithm of weighted gene co-expression network analysis（WGCNA）.Followed by dividing them into different modules,and finally the preservation of gene module was evaluated from the perspective of network node density and node connectivity.On this basis,a stepwise forward weighted random survival forest method was proposed to screen critical topology genes of prognosis.（2）Construction of risk model based on infiltrating immune cells:the expression data of genes were used to quantify the immune cells in esophageal cancer samples through CIBERSORT deconvolution algorithm,and x Cell algorithm was adopted to verify the content of immune cells.The prognostic immune features of esophageal cancer were screened by LASSO COX regression model,and the prognostic prediction model based on immune cells was established.（3）Construction of nomogram model combined with multi-indicator:certain indicators related to prognosis of esophageal cancer were integrated into the nomogram model,realizing interactive graphical prediction.ROC curve,concordance index,calibration curve and decision curve were used to evaluate the predictive performance of the model and its clinical application value.Research results（1）32 and 20 gene modules were identified in the co-expression network of normal tissues and tumor tissues by WGCNA algorithm.The results of module preservation analysis showed that 8 gene modules of the normal tissues were not preserved in tumor tissue network,among which the preservation statistics of two modules were less than 2,which were purple(Z_summary=0.200,P_summary=0.102)and midnightblue(Z_summary=0.603,P_summary=0.321).The results of functional analysis showed that genes in the non-preservation modules were significantly enriched in certain biological processes or pathways,such as digestion,the maintenance of gastrointestinal epithelium,digestive tract development,gastric acid secretion,etc.For the genes in the non-preservation modules,the COX model and weighted random survival forest were firstly used to reduce dimensionality,then LASSO method was adopted to refine the features included in the prognostic model,and finally an RNA score model was constructed based on 24 genetic features.The RNA scores achieved an AUC value of 0.805,0.801 and 0.824 for the prediction of the 2^nd,3^rdand 5^th year.The predicted AUC values for the 2^nd,3^rd and 5^th year were 0.701,0.645 and0.811 in the validation cohort.（2）The abundance of 22 types infiltrating immune cells was obtained by CIBERSORT deconvolution algorithm.Top 5 immune cells with the highest content in esophageal cancer tissues were plasma cells,dendritic cells（resting）,CD8⁺T cells,mast cells（activated）and regulatory T cells（Tregs）,with a total average proportion of 64.8%.12 types of immune cells were screened out by LASSO COX regression model to construct an immun oscore model.The immunoscore model achieved an AUC value of 0.733,0.736 and 0.747 for the prediction of the 2^nd,3^rd and 5^th year in the training cohort.The predicted AUC values for the 1^st,2^nd and 3^rd year were 0.649,0.664 and 0.724 in the validation cohort.Enrichment analysis results of gene set showed that the gene set was significantly enriched in the immune-related pathways of low immune score group,including inflammatory response,IL6-JAK-STAT3 signaling,T cell receptor signaling pathway,intestinal immune network for Ig A production,etc.（3）RNA score and immunocore could further divide patients with esophageal cancer into different risk groups in the stratified analysis of different clinical characteristics,indicating that these two indicators can supplement the predictive ability of the TNM staging system.The multivariate regression model showed that RNA score,immunoscore,age and neoplasm staging were independent factors affecting the prognosis of esophageal cancer,and they were integrated into the nomogram model for visualization.The nomogram model achieved an AUC value of 0.851,0.842 and 0.857 in the forecast of the2^nd,3^rd and 5^th year,which maximized the prognostic value.The concordance index showed that the predictive ability of the nomogram model was better than the TNM staging system and the prognostic prediction of a single index.The predicted line in the calibration curve was highly consistent with the ideal line.Decision curve analysis indicated that the nomogram model combined with multiple indicators had higher clinical application values than the TNM staging system.The developed interactive nomogram tool can easily and rapidly evaluate the prognosis of patients.Conclusions:（1）Comparing the network from a systematic point of view can make full use of negative samples,therefore,genetic modules with higher biological relevance can be obtained.（2）Through random survival forest method that integrates network weight information,it will be more convenient to screen out genetic features related to prognosis and have high connectivity in the network.（3）RNA score based on gene expression and immune score based on infiltrating immune cells can be used as reliable indicators for the prognosis of patients with esophageal cancer.（4）The nomogram model combined with multiple indicators can more accurately predict the prognosis of pat ients,compared with traditional TNM staging system.In conclusion,the prognosis prediction model for esophageal cancer constructed in this study is of high accuracy and has important guiding significance for the formulation of personalized treatment plans and clinical decisions.In addition,such analytical thinking of searching for genes that play a special role by comparing the connectivity and modularity of the network,provides a new perspective for the mechanism research of diseases with complex traits and the search of efficacy indicators.