Screening Of Autoantibodies To Tumor-associated Antigens Based On Protein Microarray And Construction Of Diagnostic Models For Lung Cancer

Lung cancer(LC)is one of the most common malignant tumors in the world.Due to the lack of effective early diagnosis methods,its five-year survival rate is low,only 19.8%,it’s high time to look for a more effective early detection method with low-loss and low-cost.In recent years,studies have shown that abnormal expression of tumor associated antigens(TAAs)in the serum of various tumor patients can trigger autoimmune reactions and produce corresponding TAAb.TAAbs are more stable and longer lasting in serum than TAAs and have the advantage of becoming tumor serum markers.PurposeThrough protein microarray screening and indirect ELISA experiments,we obtain candidate TAAbs with diagnostic value for LC.Building LC autoantibody diagnostic model by a variety of data mining techniques,and compareing the diagnostic value of each model to obtain a model that can effectively diagnose early LC.Methods1)Screening of TAAbs by protein microarrayA customized protein microarray based on 138 cancer-driven genes was used to detect autoantibodies in the sera from 100 LC patients and 50 NC.Screening was performed through a variety of data analysis methods to obtain candidate TAAbs with potential diagnostic value.2)Detection of TAAbs by indirect ELISAIndirect ELISA was used to detect the levels of TAAbs in serum samples from 155 newly diagnosed LC patients and 155 NC,and the difference of TAAbs levels between LC patients and NC were compared.The diagnostic value of candidate TAAbs was evaluated by ROC curve analysis.AUC>0.5 and P<0.05 was used to screen TAAbs as an inclusion criteria.3)Validation of TAAbs by indirect ELISA in a large sample size populationIndirect ELISA was used to detect the concentrations of autoantibodies in serum samples from 300 newly diagnosed patients with LC,144 patients with BLD and 300 NC.The levels of TAAbs between LC patients and NC,LC patients and BLD patients were compared respectively.The diagnostic value of TAAbs was further verified.4)Establishment and evaluation of diagnostic model of LCSubjects(300 LC patients and 300 NC)in the validation group were randomly divided into a training set(n=414)and a validation set(n=186).The model was built by training set and verified by validation set.We used TAAbs to construct LC diagnosis model based on data mining methods,including Logistic regression analysis,Fisher discriminant analysis,DT C5.0,ANN-MLP,ANN-RBF and SVM.ROC curve was used to evaluate the diagnostic value of each model,and the AUC,sensitivity,specificity and the number of TAAbs was used to select the optimal model.5)Application of the optimal model in early stage of LC and BLDThe optimal combination model was first applied to early stages and late stages of LC to evaluate the diagnostic value,and then the model was applied to the differential diagnosis of BLD.6)Statistical methodsSPSS 21.0,SPSS Modeler 18.0,GraphPad Prism 5.0 and MedCalc 11 were used for statistical analysis of experimental data.Nonparametric tests were used to compare the difference of the expression level of TAAbs in serum between LC and NC(BLD).The AUC,sensitivity,specificity,and 95%CI of each autoantibody were obtained by ROC curve.The studies were two-sided tests,which were statistically significant when P<0.05.Results1)Twelve TAAbs were screened by protein microarray,including TP53,P62,NPM1,Survivin,GNA11,SRSF2,HIST1H3B,FGFR2,PBRM1,JAK2,TSC1,and PIK3CA.2)Eight TAAbs(TP53,NPM1,GNA11,SRSF2,HIST1H3B,FGFR2,TSC1 and PIK3CA)were screened by indirect ELISA,the screening conditions was set at AUC>0.5 and P<0.05.The diagnostic AUC(95%CI)of eight TAAbs for LC ranged from 0.591(0.528-0.654)-0.802(0.753-0.850).3)Eight TAAbs all met AUC>0.5 and P<0.05 through validated in a large sample size by indirect ELISA.TP53 showed the highest diagnostic AUC(95%CI)of 0.751(0.710-0.793),and FGFR2 showed the lowest diagnostic AUC(95%CI)of 0.556(0.509-0.602).4)The above eight kinds of TAAbs were combined to construct diagnostic model of LC by data mining.After comparison,the DT C5.0 model has a highest diagnostic value.A total of seven TAAbs(TP53,NPM1,FGFR2,PIK3CA,GNA11,HIST1H3B,and TSC1)were included.In the training set,the AUC(95%CI)was 0.897(0.863-0.924),and the sensitivity and specificity were 94.4%and 84.9%,respectively,and in the validation set,the AUC(95%CI)was 0.838(0.777-0.888),and the sensitivity and specificity were 89.4%and 78.2%,respectively.5)The DT C5.0 model achieved an AUC(95%CI)of 0.886(0.845-0.926)and 0.864(0.826-0.902)in the diagnosis of early and advanced LC and NC,and AUC(95%CI)in the differential diagnosis of LC and BLD was 0.576(0.510-0.642).Conclusions1)Eight TAAbs(TP53,NPM1,GNA11,SRSF2,HIST1H3B,FGFR2,TSC1 and PIK3CA)might be identified as potential markers for the diagnosis of LC by protein microarray technology screening and indirect ELISA validation.2)The DT C5.0 model based on seven TAAbs(TP53,NPM1,FGFR2,PIK3CA,GNA11,HIST1H3B,and TSC1)has higher diagnostic value for LC,higher diagnostic value for early LC,and also has a certain discrimination effect for BLD.