67 patients with HCC and 47 matched control subjects without HCC were enrolled. Their clinical characteristics and biochemical parameters are listed in Table 1. Age; sex; hypertension; smoking history; drinking history; Hepatitis history (Hepatitis B and Hepatitis C) were not different between HCC and control groups. In addition, the level of AFP in 31.13% of HCC patients was lower than 20 ng/ml, but it was detected in subjects without cancers.
Analysis of gene or micro RNA expressions of HCC
Microarray data of GDS4882 including 10 patients with HCC and 10 normal subjects were obtained from the Gene Expression Omnibus database. Genes of HCC samples were analyzed using the online OmicsBean bioinformatics resource (http://www.omicsbean.cn), which is a multi-omics data analysis tool that can be applied to dynamic results. The results showed that 64 genes were significantly up-regulated and 5 genes were down-regulated with over 1.5 fold change (p < 0.01) (Figure.1A and table. 1S). Among them, THBS-1 attracts more attention as it has been found to participate in liver dysfunction and HCC development.
We further assayed THBS-1 related microRNAs by the TargetScan database and found that there was the only miR-338 that could bind to converse and no-converse sites located with the 3’UTR of THBS-1 mRNA (nt 38–44, 559–665, respectively) and with a high target score (Figure.1B). In addition, miR-194 was changed in HCC and promoted angiogenesis by inhibiting THBS-1, indicating that miR-194 might be used as a biomarker or therapeutic target for HCC .
The plasma levels of THBS-1, miR-194, and miR-338-3p
To explore the biomarkers for HCC, we tested plasma levels of THBS-1 and miR-194, miR-338-3p, which was reported or predicted to target THBS-1, respectively. The results showed that plasma THBS-1 was significantly lower in HCC patients, compared with control subjects (765.96 ± 612.28 vs 265.49 ± 163.99 pg/ml, p < 0.01) (Figure.2A), The plasma miR-194 was decreased, while the plasma miR-338-3p was increased in HCC patients compared with control subjects (0.36 ± 0.22 vs 1.00 ± 0.70 fold, 3.46 ± 4.24 vs 1.00 ± 0.89 fold, p < 0.05, p < 0.05, respectively) (Figure.2B and Figure.2C). Bootstrap analysis with 1000 iterations showed that there was significance of THBS-1 between two groups, whereas there were no significances of miR-194, miR-338-3p between groups (p = 0.38, p = 0.17 respectively).
The correlations of THBS-1 and miR-194, miR-338-3p in patients with HCC
MiR-194 and miR-338-3p were reported or predicted to target THBS-1, respectively. Thus, we also investigated the correlations of THBS-1 with miR-194 or miR-338-3p. The results showed that plasma THBS-1 positively correlated with plasma miR-194 (R = 0.23, p < 0.05) (bootstrapped 95% CI:0.04, 0.28) (Figure.3A), but negatively correlated with miR-338-3p (R= -0.21, p < 0.05) (bootstrapped 95% CI: 0.00, 0.06) (Figure.3B).
Comparison of the predictive powers of THBS-1, miR-194, and miR-338-3p for HCC
As plasma THBS-1, miR-194 or miR-338a were significantly changed in HCC patients, we evaluated the predictive powers of THBS-1, miR-194, miR-338-3p, and hepatitis history by binary logistical regression analysis (Table 2). The results showed that plasma THBS-1 (odds ratio: 1.00; 95% CI: 0.99, 1.00; p < 0.01) (bootstrapped 95% CI: 0.00, 0.06; p < 0.01), miR-338-3p (odds ratio: 1.77; 95% CI: 1.05, 3.89; p < 0.05) (bootstrapped 95% CI: 0.00, 0.06; p < 0.05), were significantly correlated with HCC status. However, there were no significance of miR-194 and Hepatitis between two groups. These results suggested that plasma THBS-1, miR-338-3p showed a highly significant diagnostic value in discriminating between HCC patients and control subjects. Notably, AFP as positive control was also evaluated and displayed a highly significant diagnostic value (odds ratio: 1.32; 95% CI: 1.04, 1.67; p < 0.05) (bootstrapped 95% CI: 0.11, 1.27; p < 0.05).
We next performed ROC analysis containing ROC curve and overall model quality to evaluate the predictive powers of plasma THBS-1, miR-338-3p, and AFP for HCC. ROC analysis (Figure.4A) showed that the area under the ROC curve (AUC) for 1/THBS-1 was 0.73 (95% CI: 0.66, 0.83; p < 0.01) and the optimal cut-off value was 328.28 pg/ml with sensitivity and specificity of 59.6 and 62.7%, respectively. The AUC for miR-338-3p was 0.72 (95% CI: 0.62, 0.81; p < 0.01 and the optimal cut-off value was 1.06 fold with sensitivity and specificity of 71.6 and 68.2%, respectively. The AUC for AFP was 0.92 (95% CI: 0.86, 0.97; p < 0.01) and AFP displayed higher predictive powers than that of 1/THBS-1 or miR-338-3p (p < 0.01, p < 0.01, respectively). Overall model quality displays the value of the lower bound of the confidence interval of the estimated AUC, and predictive model is considered good when the value is over 50%. As shown in figure.4B, the overall model quality in all data is over 50%.
We further tested whether the predictive powers of combination of 1/THBS-1 or miR-338 with AFP by ROC analysis. The resulted (figure.4C and figure.4D) showed that the AUC of 1/THBS-1 plus miR-338 was 0.82 (95% CI: 0.74–0.89; p < 0.01) and higher than that of 1/THBS-1 or miR-338 alone (p < 0.05, p < 0.01, respectively). And, the overall model quality of combination of miR-338 with 1/THBS-1 was higher than them alone (0.74 versus 0.62 or 0.63, respectively). The AUC of 1/THBS-1 plus AFP was 0.93 (95% CI: 0.74–0.89; p < 0.01) and the overall model quality of combination of 1/THBS-1 with AFP was 0.88, there was no statistical significance between combination of 1/THBS-1 with AFP than AFP alone. AUC of miR-338 plus AFP was 0.95 (95% CI: 0.91–0.99; p < 0.01), and higher than AFP alone, but there was no statistical significance. The overall model quality of combination of miR-338 or 1/THBS-1 with AFP were higher than AFP alone (0.91 versus 0.86).