Screening Of Tumor Associated Antigens(TAAs) And Their Autoantibodies Based On Protein Microarray And Evaluation Of The Autoantibodies As Diagnostic Biomarkers For Hepatocellular Carcinoma

BackgroundHepatocellular carcinoma（HCC）is one of the common malignant tumors with a highly mortality rate in China.Timely treatment followed early accurate diagnosis of HCC is an effective measure to reduce its mortality.Diagnostic methods such as needle biopsy and surgical exploration are traumatic and difficult to obtain and they are not easy to popularize.Exploring serum detection technology with high sensitivity and specificity is helpful for early screening and diagnosis of HCC.A number of studies have shown that the tumor associated autoantibody（TAAb）produced by the stimulation of tumor associated antigen（TAA）in the sera of tumor patients have became potential biomarkers for early diagnosis of tumors.ObjectiveThis study aims to use Hu Prot^TM human protein microarray V3.1 technology to screen potential TAAbs in sera from HCC patients,and verified the expression level of TAAbs in sera from the patients with HCC,patients at high risk of HCC and normal controls by indirect enzyme-linked immunosorbent assay（ELISA）,and then to evaluate the diagnostic value of individual TAAbs for HCC.At same time,with data mining technology,we try to establish and verify multiple autoantibody combined diagnosis models,followed by selecting the optimal model and evaluating its diagnostic value for HCC,so as to provide some worthwhile theories for the serological diagnosis and early screening technology of HCC.Methods1.Using protein microarray to screen TAAbs for HCCThe Hu Prot^TM human proteomic microarrays were used to screen autoantibodies in 10 mixed HCC serum samples and 10 mixed healthy controls serum samples.Non-parametric test,KEGG analysis of String database,positive rate were used to screen out potential diagnostic TAAbs for HCC.2.Using ELISA for TAAbs detection1)Using small sample size for detection of TAAbs:In this study,a two-stage verification strategy was designed in the ELISA experiment.The PASS software was used to calculate the sample size according to the sensitivity and specificity based on the results of the protein microarray.HCC patients and healthy controls were matched1:1 by gender and age（±3 years）.Sera from 75 HCC patients and 75 healthy controls were included in small-sample size verification stage.Sera from 111 HCC patients,111 healthy controls,157 liver cirrhosis patients,and 92 hepatitis B patients were recruited in the large-samples size validation stage.2)Using large sample size for verification of TAAbs:The meaningful TAAbs screened out in the small sample size stage were further verified in the large sample stage.Mann-Whitney U test was used to compare the difference of TAAbs expression levels between HCC patients and healthy controls.The receiver operating characteristic（ROC）curve was applied to evaluate the diagnostic value of TAAbs.Kruskal-Wallis test and Dunn’s Multiple Comparison test were used to compare the differences of TAAbs expression levels in HCC group,healthy controls group,liver cirrhosis group,and hepatitis B group.3.Construction and verification of diagnostic models for HCC1)Model construction:111 HCC patients and 111 healthy controls in ELISA were used as training set.Logistic regression analysis,support vector machine（SVM）,and C5.0 decision tree were used to construct HCC diagnostic models to realize the combined diagnosis of TAAbs for HCC.ROC curve was used to evaluate the diagnostic value of each model.2)Model validation:75 HCC patients and 75 healthy controls were used as the validation set,which was used to evaluate the diagnostic values of different models in the training set.Sensitivity,specificity and agreement rate were employed to verify the diagnostic value of these three models.The optimal and stable model was selected according to the results of model construction,validation and evaluation.3)The diagnostic values of TAAbs and the optimal model were evaluated in HCC clinical subgroups:Epidemiological diagnostic test evaluation system was applied to evaluate the diagnostic ability of the TAAbs and optimal model for HCC across 3 clinical subgroups divided by clinical staging,AFP level and tumor sizes.The difference of AUC between different subgroups was detected by Delong test.Results1.The results of protein microarray:Ten TAAbs were screened out by protein microarray,including BRMS1L,TAF7L,RUNX1T1,NFKB1,MAGEA12,ERK1,APPL1,DKK4,CCDC6 and NCOA3.2.The results of detection and verification by ELISA:From the results of ELISA in the small sample group,the expression levels of 9 TAAbs（BRMS1L,TAF7L,RUNX1T1,NFKB1,MAGEA12,ERK1,APPL1,DKK4,CCDC6）in HCC group were higher than that in healthy controls group.The diagnostic AUC range of 9TAAbs was 0.637～0.708,the sensitivity range was 22.7%～56.0%,the specificity range was 80.0%～94.7%.In the large sample group,the expression levels of 9 TAAbs were higher in HCC group than in healthy controls group（P<0.05）.Their AUC range was 0.591～0.712,the sensitivity range was 19.8%～55.0%,and the specificity range was 80.2%～95.5%.3.The comparison of TAAbs in HCC,liver cirrhosis,hepatitis B,and healthy controls:Using Kruskal-Wallis test and Dunn’s Multiple Comparison test to compare the expression levels of 9 TAAbs among HCC,liver cirrhosis,hepatitis B,and health controls groups.The levels of autoantibodies of BRMS1L,TAF7L,and ERK1 in HCC group were higher than those in liver cirrhosis group,simultaneously,their expression levels were higher in liver cirrhosis group than those in hepatitis B group.The expression levels of 9 TAAbs in hepatitis B group and healthy controls group were lower than those in liver cirrhosis group or HCC group.The levels of autoantibodies of NFKB1,APPL1 in liver cirrhosis group were higher than those in HCC group.The expression levels of RUNX1T1,MAGEA12,DKK4,and CCDC6 between liver cirrhosis group and HCC group were not statistically different.The expression levels of RUNX1T1,NFKB1,MAGEA12 and CCDC6 had no statistical difference between hepatitis B group and healthy controls group.The levels of BRMS1L,TAF7L,ERK1and DKK4 in healthy controls group were higher than those in chronic hepatitis B group.The expression level of APPL1 in chronic hepatitis B group was higher than that in healthy controls group.4.The results of establishment and verification of diagnostic models in HCC:Logistic regression analysis,SVM,and C5.0 decision tree were established according to data mining technology.The AUC values in the training set for those three models were 0.753,0.745,0.851 respectively,and the agreement rates were 72.5%,66.2%,74.8%respectively.The AUC values in the validation set for those three models were0.707,0.717,0.678,respectively,and the agreement rates were 63.4%,67.3%,66.7%respectively.The Logistic regression analysis model was selected as the optimal model for this study.A total of 3 TAAbs（RUNX1T1,MAGEA12,DKK4）were included in Logistic regression analysis model.The sensitivity and specificity of this model were 68.5%and 76.6%respectively in the training set and 52.0%,74.7%respectively in the validation set.5.The diagnostic values of TAAbs and optimal model in various clinical subgroups of HCC:The diagnostic AUC range of 9 TAAbs for HCC at early stage was 0.629～0.722,with sensitivity range of 22.6%～60.4%,and specificity range of80.1%～93.6%.The diagnostic AUC range for 9 TAAbs at advanced stage of HCC was 0.592～0.671,the sensitivity range was 27.5%～46.3%,and the specificity range was 80.1%～92.5%.The AUC range for 9 TAAbs of AFP-negative HCC was0.610～0.721,the sensitivity range was 26.9%～62.7%,the specificity range was80.1%～89.3%.Similarly,for AFP-positive HCC,the AUC range was 0.611～0.693,the sensitivity range was 26.9%～46.2%,the specificity range was 80.1%～87.6%.For patients with tumor size≤5 cm,the AUC range was 0.621～0.733,the sensitivity range was 18.3%～58.5%,and the specificity range was 80.1%～93.6%.For patients with tumor size>5 cm,the diagnostic AUC range was 0.607～0.700,the sensitivity range was 31.7%～51.0%,and the specificity range was 80.1%～92.5%.When the logistic regression model was selected as the optimal model,the diagnostic AUC for early stage HCC was 0.749,with a sensitivity of 66.0%and a specificity of 75.8%.As for advanced stage HCC,the diagnostic AUC,sensitivity and specificity were 0.706,56.3%and 75.8%respectively.Referring to AFP-negative HCC,the diagnostic AUC was 0.739,the sensitivity was 68.7%,and the specificity was 75.8%.As for AFP-positive HCC,the diagnostic AUC was 0.726,the sensitivity was 58.0%,and the specificity was 75.8%.For patients with tumor size≤5 cm,the diagnostic AUC was 0.736,the sensitivity was 65.0%,and the specificity was 75.8%.For patients with tumor size>5 cm,the diagnostic AUC was 0.726,the sensitivity was 60.6%,and the specificity was 75.8%.There were no statistically significant difference in AUC of both TAAbs and optimal model across 3 clinical subgroups divided by clinical staging,AFP level and tumor size.Conclusions1.According to protein microarray technology and ELISA test,9 TAAbs（BRMS1L,TAF7L,RUNX1T1,NFKB1,MAGEA12,ERK1,APPL1,DKK4,CCDC6）were identified to have potential application value in the immunodiagnosis of HCC.Among them,autoantibodies against RUNX1T1 and MAGEA12 had high diagnostic values and stability,which can be used as prodential tumor markers for the diagnosis of HCC.2.The logistic regression model constructed by 3 TAAbs（RUNX1T1,MAGEA12,DKK4）had a certain diagnostic value for early stage HCC and AFP negative HCC.