Screening Of Tumor Associated Antigens And Evaluating And Modeling Of Their Autoantibodies As Diagnostic Markers For Hepatocellular Carcinoma

ObjectiveThe high mortality of hepatocellular carcinoma(HCC)in China is one of the most important public health issues and needed to be addressed urgently.The keys to solve this problem should be focused on early diagnosis,and the improvement in treatment methods.As to early diagnosis of HCC,substantial evidence manifests the occurrence of autoantibodies to tumor associated antigens(TAAs)in early stage of tumorigenesis.In this study,we aimed to identify TAAs in HCC by using serological proteome analysis(SERPA)and protein microarray,and further to construct and develop a diagnostic model for the early immunodiagnosis of HCC.Methods1.Screening TAAs for HCC(1)SERPA was used to screen TAAs in HCC.Firstly,positive sera were screened out by Western blotting on HepG2 cell lysate from 100 HCC sera and 50 health control sera.Secondly,using two-dimensional electrophoresis(2DE)technology,the proteins from HepG2 cell lysate in three equal parts were separated,then were transferred onto PVDF membrane for Western blotting or stained with Coomassie blue R-250.Comparing and matching the visualized protein spots on each membrane reacted with positive HCC sera and health control sera to the equivalent protein spots on the Coomassie blue-stained 2DE gel,the immunoreactive spots of difference between the membranes were excised from Coomassie blue-stained SDS-PAGE gel and the interesting ones were identified by Matrix-Assisted Laser Desorption/Ionization Time of Flight Mass Spectrometry(MALDI-TOF MS).(2)Protein microarrays were customized to screen TAAs in HCC.Protein microarrays,containing 143 proteins encoded by cancer-driver genes and 11 proteins used in our previous studies,were customized to screen TAAs in serum samples from 100 patients with HCC and 50 health controls.Mann Whitney-U test,ROC curve and positive rate were used to analyse the difference between HCC group and health control group;Logistic regression was used for further screening of multivariates.Finally,the results of AUC and Logistic regression were combined to determine the TAAs to be further verified.2.Indirect enzyme-linked immunosorbent assay(ELISA)was performed to test titers of anti-TAAs autoantibody(TAAb)in two independent phases.(1)Autoantibodies against screened TAAs were tested in sera from 286 patients with HCC and 286 health controls in the first phase by ELISA.In the second phase,autoantibodies in sera from 160 patients with HCC,160 health controls,and 127 patients with liver cirrhosis(LC)were tested.(2)Mann Whitney-U test was performed to compare the difference of TAAbs level between HCC patients and health controls both in first and second phase.ROC curves were drawn to evaluate the diagnostic value of each TAAb.TAAbs with higher level in HCC patients than that in health controls were selected to construct diagnostic model in next step.(3)The trend of TAAbs among groups of health control,liver cirrhosis,and TNM stages of HCC,were observed based on samples of the second phase.3.Multiple models were constructed and evaluated(1)Based on subjects in the first phase,Logistic regression,LASSO regression,parallel diagnostic test,classical decision tree model,random forest model,and support vector machine were used to construct diagnostic models for HCC.Patients with HCC and health controls in the second phase were used to verify each model.By comparing the constancy of each model,an optimal model was selected.(2)The positive rates between subgroups were compared,including subgroups of HCC patients of early stage or late stage,HCC patients with or without metastasis,HCC patients with or without history of hepatitis B,and HCC patients with or without history of family history.(3)Patients with liver cirrhosis were used to evaluate performance of the selected model.(4)Difference between the selected model and AFP was analysed.Results1.Six TAAs including GAPDH,ENO1,HSPD1,PGK1,TPM3,HSP90 were selected by SERPA.By using protein microarray,11 TAAs were selected including GNA11,IDH1,PTEN,NPM1,Survivin,MSH2,SRSF2,PTCH1,PAX5,GNAS and TP53,with AUCs of 0.749,0.712,0.693,0.691,0.685,0.685,0.656,0.658,0.616,0.618,and 0.631 respectively.Totally,17 candidate TAAs were screened out by both SERPA and protein microarray,one of them(TPM3)is not available on market,so 16 candidates got into the subsequent verification phase.2.It was found that titers of 11 TAAbs(Survivin,TP53,NPM1,IDH1,MSH2,GNAS,SRSF2,GNA11,PTCH1,ENO1 and HSP90)were higher in patients with HCC than that in health controls,the performances of GNAS,Survivin and TP53 were better than others,with AUCs of 0.738,0.737 and 0.705,respectively.The results from the second phase confirmed the findings from the first phage,the levels of all 11 TAAbs were still higher in patients with HCC than that in health controls,the AUCs of GNAS,Survivin and TP53 were still more than 0.700.With the development of HCC,it was found that the levels of 11 TAAbs were higher in patients with liver cirrhosis and patients with HCC in clinical stage I than other groups.3.The consistence rates of Logistic regression,LASSO regression,parallel diagnostic test,classical decision tree model,random forest model,and support vector machine in the two phases were 76.4%,75.5%,71.3%,74.3%,78.3%,92.8%and 76.6%,76.6%,66.6%,62.8%,73.1%,63.1%respectively.It was found that Logistic regression and random forest model were more stable than other models.However,the diagnostic performance of random forest model was not significantly better than that in Logistic regression and the variables included in this model were more than that in Logistic regression.The Logistic regression was finally selected as the optimal model,with the calculation formula of P(HCC)=1/(1+Exp(2.309-6.391 ×Survivin-4.409 × TP53+6.696 × NPM1-12.056 × GNAS+12.380 × SRSF2-3.471 × PTCH1+4.274 × ENO1-5.021 × HSP90)).There was no significant difference for the model’s diagnostic efficiency among clinical subgroups of HCC.Based on samples included in the second phase,if the case group was set as HCC patients group,and the control group was set as LC group,LC and health group,and healthy group respectively,AUC of Logistic regression model were 0.638,0.753,0.844,respectively.The diagnostic value showed no difference between AFP negative HCC patients and AFP positive HCC patients.The combination of this model with AFP to diagnose HCC could effectively improve the diagnostic value of both phases,the consistent rates could be elevated to 88.3%and 89.3%in the first phase and the second phase.Conclusion1.Eleven TAAs(TP53,NPM1,IDH1,MSH2,GNAS,SRSF2,GNA11,PTCH1,ENO1 and HSP90)were screened out by SERPA and protein microarray and validated in two independent phases,their autoantibodies can be considered as potential diagnostic biomarker in early screening of HCC and the diagnosis of AFP negative HCC.2.The level of each autoantibody was higher in patients with LC or TNM-I HCC than others,it was speculated that these autoantibodies might appear in the transformation from LC to HCC,and could be used as indicators for early HCC.3.By comparing multiple diagnostic models,the Logistic regression model was selected as an optimal one in this study.This model could be used in the diagnosis of HCC,especially for AFP-negative HCC and the early screening in subjects without clinical symptoms.