Identification of noninvasive diagnostic biomarkers for hepatocellular carcinoma by urinary proteomics.

Abstract Hepatocellular carcinoma (HCC) ranks fourth in cancer mortality worldwide, and third in China. Hepatitis B virus (HBV) infection is a main risk factor for HCC in China, and the early diagnosis of HCC in high-risk population is very important. However, the commonly used diagnostic biomarker alpha-fetoprotein has limitations in clinical practice. In order to identify reliable and noninvasive HCC urinary biomarkers, a high-throughput proteomics streamline was applied in the analysis of urine samples from 74 HCC and 82 high-risk patients with chronic HBV infected liver diseases. Candidate diagnostic markers were screened by feature selection algorithm, and were combined with random forest or simple voting algorithms in the training dataset. Then the multiple feature models were validated in an independent test dataset. The selected features were further verified by Multiple Reaction Monitoring (MRM) in another independent dataset. By integrating 7 features (HPX, GOT1, APOH, GLRX, APCS, NCR3LG1 and PLG) screened in the discovery phase, random forest model achieved AUC of 0.92 and 0.87 in training and test datasets, respectively, while voting model performed better with AUC of 0.94 and 0.90, respectively. In the MRM dataset, the 7 features were targeted quantified, and voting model integrating the 7 features achieved AUC of 0.95. Our work highlights the potential of noninvasive urinary protein biomarkers in HCC diagnosis with high-risk population, which will be beneficial to HCC auxiliary diagnosis and HCC surveillance. Significance A high throughput urinary proteome analysis platform was committed into the discovery of noninvasive HCC biomarkers in high-risk patients with chronic HBV infected liver diseases. The combination of 7 urinary features achieved good performance in distinguishing HCC from high-risk population. The expression of the 7 features was validated by targeted MRM, and the integration of the features also worked well in the MRM dataset. This is the first time that urinary proteomic strategy was applied in discovering HCC biomarkers from high-risk population. This result will be helpful for HCC auxiliary diagnosis and surveillance in a noninvasive way.