Selection of candidate miRNAs from the NCBI database

Using a systematic review of the NCBI database, we selected candidate miRNAs. Figure ​Figure1B1B shows a detailed description of the selection process. A total of 281 relevant full articles related to HCC were retrieved from PubMed up to February 2013. From these reports, we selected 124 candidate miRNAs as shown in Supplementary Table S1. The detailed selection criteria for the candidate miRNAs are described in the Materials and Methods. To identify more sensitive and reliable biomarkers, we focused on miRNAs whose expression levels had been previously validated in HCC tissues. Consequently, 34 miRNAs were excluded. Next, 11 miRNAs previously reported as biomarkers in body fluids were excluded. We limited the target miRNAs to oncogenic miRNAs, as previous studies have revealed that high plasma levels of oncogenic miRNAs might be derived from tumor necrosis, apoptosis and the active release of secretory vesicles, such as exosomes, from cancer cells; thus, these miRNAs in plasma could reflect tumor dynamics in cancer tissues [34–35, 40–41]. Tumor-suppressive miRNAs and functionally unknown miRNAs were excluded in the present study because the origin of these miRNAs is not well known [38]. Consequently, 48 miRNAs were excluded. Furthermore, 27 miRNAs, insufficient for clinical application and reported in only one article, were excluded. After a series of exclusion criteria were applied, we systematically selected four candidate miRNAs: miR-151 [48–49], miR-155 [50–51], miR-191 [52–53], and miR-224 [54–58].

Small-scale analysis of the plasma levels of four candidate miRNAs in HCC patients and healthy volunteers

We next investigated the plasma levels of the four miRNAs in 20 consecutive HCC patients and 20 healthy volunteers using qRT-PCR as a small-scale analysis. Regarding this cohort, the proportion of background hepatic disease was categorized as 40% of HCC patients with non-B, non-C and 60% of HCC patients with viral etiology. The plasma levels of all candidate miRNAs were significantly higher in HCC patients than in healthy volunteers, the plasma miR-224 level was validated to exhibit the most significant difference (Figure ​(Figure2).2). Moreover, the plasma levels of miRNA-151 and miRNA-191 were significantly higher in the plasma of HCC patients with hepatitis B or C than in non-B, non-C patients. Therefore, we excluded these two miRNAs from the candidate miRNAs to select candidates independent of background liver disease with different etiologies (Supplementary Figure S1). Furthermore, we utilized the area under the ROC curve using the Youden index and calculated the value for the AUC for miR-155 and miR-224 [59]. The AUC value for the plasma miR-155 and miR-224 were 0.589 and 0.905, respectively. Thus, we selected the most promising candidate, miR-224, for further analyses.

Confirmation of higher miR-224 expression in HCC tissues and cell lines

Next, we confirmed the expression level of miR-224 in primary HCC tissues and HCC cell lines compared with normal tissues and the fibroblast cell line WI-38. We used qRT-PCR to determine the expression of miR-224 in paired 12 normal hepatic tissues (Figure ​(Figure3A)3A) and in the human HCC cell lines, such as JHH, HLE, Hep-G2 and Alexander cells (Figure ​(Figure3B).3B). The miR-224 expression level was significantly higher in HCC tissues than in normal hepatic tissues (P = 0.0011) (Figure ​(Figure3A).3A). The expression of miR-224 was significantly higher in the HCC cell lines than in normal hepatic tissues and the fibroblast cell line WI-38 (P = 0.0150) (Figure ​(Figure3B).3B). These findings strongly suggested that miR-224 expression was highly expressed in HCC, as shown in previous reports [54–58].

Large-scale analysis of the miR-224 plasma levels in HCC patients

We next validated these observations in large-scale settings. Prior to the analysis of a large number of samples using qRT-PCR, the linearity of qRT-PCR was confirmed using various concentrations of 1 fmol to 0.0001 fmol of each synthetic miRNA, namely, miR-224 (R² = 0.9981) and cel-miR-39 (R² = 0.9977), between the logarithm of the amount of input miRNA and Ct values (Supplementary Figure S3A). Recent reports have demonstrated that some circulating miRNAs might be derived from peripheral blood cells [60]. In the present study, there was no significant correlation between the plasma miR-224 level and the number of any type of peripheral blood cell (Supplementary Figure S3B).

Plasma miR-224 was detected in all samples from the 87 HCC patients and 55 healthy volunteers. Using large validation cohorts, we showed that the plasma level of miR-224 was significantly higher in the HCC patients than in the healthy volunteers (P < 0.0001) (Figure ​(Figure3C).3C). Furthermore, to detect any cut-off points differentiating cancer patients from healthy volunteers, we utilized the area under the ROC curve using the Youden index [59] and calculated the value for the AUC of the validation cohort. The AUC value for the plasma miR-224 analysis was 0.908. The optimal cut-off point was 8.0 in relative expression using the miR-224/cel-miR-39 ratio with a sensitivity of 93.1%, a specificity of 80.0%, and an accuracy of 88.0%.

Because the validation cohorts comprised samples from two independent institutes, we also examined variations using two independent cohorts. The results were nearly identical in each independent cohort (Supplementary Figure S2). In the two independent cohorts, the miR-224 plasma level was significantly higher in the HCC patients than in the healthy volunteers (P < 0.0001). Additionally, the AUC values for the plasma miR-224 analysis were 0.914 and 0.906, respectively. In the first cohort, the cut-off point was 8.0 in relative expression using the miR-224/cel-miR-39 ratio with a sensitivity of 87.7%, a specificity of 86.3%, and an accuracy of 88.5%. In second cohort, the cut-off point was 8.0 in relative expression using the miR-224/cel-miR-39 ratio with a sensitivity of 87.7%, a specificity of 86.3%, and an accuracy of 87.2%. Similar to the small-scale analysis, the consistency of the results using large-scale validation cohorts and two independent cohorts provided strong evidence that the plasma miR-224 level could satisfactorily distinguish HCC patients from healthy volunteers compared with conventional tumor markers.

Furthermore, to demonstrate the clinical utility, we next compared the plasma miR-224 levels between early stage HCC patients and the healthy volunteers in the validation cohort. We defined stage I HCC as early stage HCC, according to the two staging systems such as the LCSGJ and TNM (AJCC/UICC). The results were almost same in both staging systems, namely, the miR-224 plasma level was significantly higher in stage I HCC patients than in the healthy volunteers (P < 0.0001). The AUC values in the LCSGJ and TNM (AJCC/UICC) staging systems for the plasma miR-224 analysis were 0.888 and 0.899, respectively. These results clearly demonstrated that the plasma miR-224 level could satisfactory distinguish early stage HCC patients form healthy volunteers (Figure ​(Figure44).

Open in a separate window

Figure 4

Evaluation of whether the plasma miR-224 level could satisfactory distinguish early stage HCC patients form healthy volunteers

The expression level of plasma miR-224 was significantly higher in stage I HCC patients than in healthy volunteers (P < 0.0001). The AUC values in the LCSGJ and TNM (AJCC/UICC) staging systems for the plasma miR-224 analysis were 0.888 and 0.899, respectively.

Evaluation of whether plasma miR-224 expression reflects tumor dynamics

To validate whether the plasma miR-224 level reflects tumor dynamics during the treatment of HCC patients, we first confirmed that the plasma miR-224 levels were significantly correlated with the tumor tissue levels of miR-224 using paired samples (P = 0.0002) (Figure ​(Figure3D).3D). Second, we evaluated the plasma level of miR-224 in paired samples collected before and almost 1 month after surgery from 8 HCC patients who underwent curative hepatectomy and observed that miR-224 was significantly reduced in these postoperative plasma samples (P = 0.0058) (Figure ​(Figure3E).3E). These results indicated that the plasma miR-224 level precisely reflects the HCC tissue status and could be used to trace tumor dynamics in HCC patients.

Evaluation of tumor detection based on plasma miR-224 levels independent of chronic hepatic disease and hepatic function

To demonstrate the clinical utility, we validated the plasma level of miR-224 in HCC patients without associated background liver disease and liver function. As shown in the test-scale analysis (Supplementary Figure S1), there were no significant differences in the plasma miR-224 levels among patients with HBV, HCV and other disease in the large-scale analysis (P = 0.3087) (Figure ​(Figure5A).5A). The plasma miR-224 levels were significantly higher in HCC patients compared with non-HCC outpatients among individuals with mild (Figure ​(Figure5B)5B) or severe liver dysfunctions (P = 0.0008) (Figure ​(Figure5C).5C). Furthermore, no significant correlation was observed between the plasma miR-224 levels and other clinical indicators, such as the tumor makers PIVKA-II (DCP) and AFP, the ICG retention rate (ICG 15R), total bilirubin (T-Bil), albumin(Alb), prothrombin time (PT (%)), aspartate transaminase (AST) and alanine transaminase (ALT) (Figure ​(Figure66).

Correlation between the miR-224 plasma levels and the clinicopathological factors in HCC patients

We analyzed whether plasma miR-224 levels were correlated with clinicopathological factors in the validation cohort comprising all 87 consecutive patients with HCC. The results showed that large tumor sizes of more than 20 mm, an advanced pStage (LCSGJ) of more than II and the presence of recurrences were significantly correlated with high levels of plasma miR-224 (P = 0.0005, 0.0382 and 0.0027, respectively), whereas clinical factors, such as gender, age, BMI, diabetes, viral infection and liver cirrhosis, types of recurrence were not correlated with plasma miR-224 levels (Table ​(Table1,1, Supplementary Table S3A). These tendencies were same in two independent cohorts (Supplementary Table S2, S3B).

Comparison of the minimum tumor detectability between plasma miR-224 and conventional serum tumor markers

Next, we compared the minimum tumor detectability between plasma miR-224 and conventional serum tumor markers, such as PIVKA-II and AFP, in 54 consecutive HCC patients who underwent hepatectomy at Kyoto Prefectural University of Medicine (First Cohort) and 33 consecutive HCC patients who underwent hepatectomy at Kyoto Second Red Cross Hospital (Second Cohort). Regarding the cut-off value of plasma miR-224, the most sensitive cut-off value, determined using the ROC curve to detect the smallest tumor, was used (Supplementary Table S4). As shown in Table ​Table2,2, the plasma miR-224 level was more sensitive to the presence of smaller tumors than were conventional tumor markers, such as PIVKA-II and AFP. Specifically, the plasma miR-224 level could detect a small tumor of less than 18 mm with highest AUC value. These results were clearly demonstrated in two independent cohorts.

RNA extraction

Total RNA was extracted from 400 μl of plasma using a mirVana PARIS Kit (Ambion, Austin, TX) and eluted into 100 μl of preheated (95°C) Elution Solution according to the manufacturer's protocol. A volume of 400 μl of plasma was used as the common denominator in each microarray analysis; thus, there was no definite internal control in the plasma miRNA analyses as shown in previous studies [34–41]. Total RNA was also extracted from four 15-μm-thick slices of the formalin-fixed and paraffin-embedded tissue samples (60 μm thickness) using a Recover All Total Nucleic Acid Isolation Kit (Ambion) and subsequently eluted into 60 μl of Elution Solution according to the manufacturer's instructions.

A systematic review of the NCBI database to select candidate miRNAs

We performed a systematic review of the NCBI database to identify novel plasma biomarkers of miRNA in patients with HCC (Figure ​(Figure1B1B and Supplementary Table S1). We searched for all studies related to HCC miRNAs in PubMed up to February 2013. This search was based on the key terms ‘HCC’ and ‘microRNA.’ Any meeting abstracts not accompanied by full articles and other incomplete and non-English articles were excluded. The candidate miRNAs had to meet the following criteria: I) Candidate miRNAs whose overexpression was validated in HCC tissues compared with normal liver or peritumoral non-cancerous tissues; II) Candidate miRNAs not previously reported as biomarkers in body fluids, such as plasma, serum, urine and semen; III) Candidate miRNAs demonstrated to have oncogenic roles (functionally unknown or tumor suppressive miRNAs were excluded); and IV) Candidate miRNAs reported several times with sufficient data in HCC. Two authors independently reviewed all articles identified in this search using these criteria, and a third author resolved any discrepancies.

Quantification of miRNA by qRT-PCR

The amounts of miRNAs were quantified by qRT-PCR using a human TaqMan MicroRNA Assay Kit (Applied Biosystems, Foster City, CA). A reverse transcription reaction was performed using a TaqMan MicroRNA Reverse Transcription Kit (Applied Biosystems) in 5 μl of solution containing 1.67 μl of extracted RNA, 0.05 μl of 100 mM dNTPs, 0.33 μl of Multiscribe Reverse Transcriptase (50 U μl^-1), 0.5 μl of 10× Reverse Transcription Buffer, 0.06 μl of RNase inhibitor (20 U μl^-1), 1 μl of gene-specific primer (hsa- miR-151, Assay ID: 000596; hsa-miR-155, Assay ID: 000479; hsa-miR-191, Assay ID: 000490; hsa-miR-224, Assay ID: 000599; cel-miR-39, Assay ID: 000200; and RNU6B, Assay ID: 001093), and 1.39 μl of nuclease-free water. To synthesize cDNA, the reaction mixtures were incubated at 16°C for 30 min, 42°C for 30 min, and 85°C for 5 min, after which the reactions were held at 4°C. Next, 0.67 μl of cDNA was amplified using 5 μl of TaqMan 2× Universal PCR Master Mix with no AmpErase UNG (Applied Biosystems), 0.5 μl of gene-specific primers/probe, and 3.83 μl of nuclease-free water in a final volume of 10 μl. Quantitative PCR was run on a StepOnePlus PCR system (Applied Biosystems), and the reaction mixtures were incubated at 95°C for 10 min, followed by 40 cycles of 95°C for 15 sec and 60°C for 1 min. The cycle threshold (Ct) values were calculated using StepOne Software v2.0 (Applied Biosystems).

As previously reported [22], we used an approach to data normalization based on spiking the sample with a synthetic RNA oligonucleotide, cel-miR-39, which does not exist in the human genome. C. elegans cel-miR-39 was purchased as a custom-made RNA oligonucleotide (Qiagen, Valencia, CA). The oligo used for spiking, as a mixture of 25 fmol of the oligonucleotide in a total volume of 5 μl of water, was introduced after the addition of 2X Denaturing Solution (Ambion) to the plasma or serum sample to avoid degradation by endogenous plasma RNases. As a control for each RNA sample, cel-miR-39 was used for TaqMan qRT-PCR assays (Applied Biosystems) as previously described. The data were normalized across samples using the 2^-ΔΔCt method relative to cel-miR-39. The miRNA expression from tissue samples and cultured cells was normalized using the 2^-ΔΔCt method relative to U6 small nuclear RNA (RNU6B). The ΔCt was calculated after subtracting the Ct values of cel-miR-39 or RNU6B from those of the miRNAs of interest. The ΔΔCt was subsequently calculated after subtracting the mean of ΔCt of the plasma of healthy volunteers or normal pancreatic tissue samples from the ΔCt of HCC tissues. Changes in gene expression were calculated using the equation 2^-ΔΔCt [91–92].

Culture of HCC cell lines

The HCC cell lines JHH-6, HLE and Alexander were purchased from the JCRB Cell Bank (Osaka, Japan), and HepG2 was purchased from the RIKEN cell bank (Tsukuba, JAPAN) and cultured in Roswell Park Memorial Institute (RPMI)-1640 medium (Sigma, St Louis, MO), Dulbecco's Modified Eagle Medium (DMEM, Nacalai Tesque, JAPAN) and William's Medium E (LANZA, Switzerland) supplemented with 10% FBS (Trace Scientific, Melbourne, Australia). All cell lines were cultured in 5% carbon dioxide at 37°C in a humidified chamber.

The authors thank Yoshio Sumida, MD, Ph.D. and Michihisa Moriguchi, MD, Ph.D. (Department of Molecular Gastroenterology and Hepatology, Kyoto Prefectural University of Medicine, Kyoto, Japan), and Momoko Todo, MD, Ph.D. and Yutaka Kondo, MD, Ph.D. (Department of Surgery, Kyoto Second Red Cross Hospital, Kyoto, Japan) for supporting this research.