Exploration of Genome-Wide Circulating MicroRNA in Hepatocellular Carcinoma: MiR-483-5p as a Potential Biomarker

The incidence of hepatocellular carcinoma in the United States (U.S.) has doubled over the past 20 years (1, 2). Hepatocellular carcinoma prognosis is poor if not diagnosed and treated at an early stage. Current clinical diagnostic approaches for hepatocellular carcinoma are mainly based on imaging techniques (abdominal ultrasound MRI, contrast-enhanced computed tomography scan) and histology (3, 4). Serum α-fetoprotein (AFP) measured in clinically available samples has long been used as an early diagnostic biomarker of hepatocellular carcinoma, but the sensitivity (39%–65%) and specificity (76%–94%) are poor (5, 6). Therefore, identification of novel and reliable biomarkers in easily accessible clinical biospecimens is extremely important to improve the early detection of hepatocellular carcinoma. Circulating microRNAs (miRNA) in cell-free plasma/serum samples have been consistently observed to have high stability and resistance to storage/handling (7–11), suggesting their potential use as diagnostic biomarkers.

miRNAs are a class of small noncoding RNAs that control gene expression by inhibiting translation or inducing cleavage of target mRNAs. miRNAs can regulate diverse biologic processes including DNA repair, apoptosis, cell proliferation, differentiation, and immune function. Aberrant miRNA expression has been associated with a variety of cancers (10, 12), including hepatocellular carcinoma by examining tumor and nontumor tissues (13–15). Several miRNAs with oncogenic characteristics are significantly upregulated in hepatocellular carcinoma tumor tissues compared with nontumor tissues, such as the miRNAs (miR)-17-92 cluster, miR-21, miR-181b, miR-221, miR-222, and miR-500 (16–27). Genome-wide microarray data found overexpression of miR-221 and miR-222 in 50% to 83% of hepatocellular carcinoma tissues compared with matched cirrhotic tissue (28–32). Tumor-site–specific miRNAs can be extracellularly released into the bloodstream via active secretion from tumor tissues in a protein-bound complex (33) or as membrane-bound vesicles (34). Because of high rates of proliferation and cell lysis in tumors, nonspecific passive release of miRNAs is also observed (9). Circulating miRNAs are consistently shown to have high stability due to their protection from RNases (7, 9, 11), and even under severe conditions, such as boiling, very low or high pH levels, extended storage, and multiple freeze–thaw cycles (7). These data indicate that miRNAs are abundant in circulation, and may be useful for early diagnosis of hepatocellular carcinoma.

Investigators, using candidate and genome-wide approaches, have found more than 100 miRNAs that were dysregulated in hepatocellular carcinoma tumor tissue compared with nontumor tissue (35). But only a small proportion of circulating miRNAs (∼1/10) was aberrantly expressed in the plasma of hepatocellular carcinoma cases compared with controls, including overexpression of let-7f, miR-122, miR-192, miR-21, miR-221-224, miR-25, miR-26a, miR-27a, miR-375 miR-500, miR-801, and miR-885 (26, 36–41) and downregulation of miR-16, miR-195, miR-199a, and miR-92a (27, 42). One genome-wide study identified a panel of miRNAs (miR-122, miR-192, miR-21, miR-223, miR-26a, miR-27a, and miR-801) significantly overexpressed in hepatitis B virus (HBV)-related hepatocellular carcinoma compared with controls with unknown viral status (40). Interestingly, different expression patterns were observed for miRNAs in plasma and tumor tissue samples. For example, miR-16 (42, 43) and miR-92a (27) were overexpressed in hepatocellular carcinoma tumor tissues but downregulated in plasma; let-7g (28, 41) and miR-122 (28, 39) were repressed in tissue but upregulated in plasma samples. These results indicate that further studies are needed to validate genome-wide circulating miRNAs as biomarkers to improve early diagnosis of hepatocellular carcinoma. In the current study, we used a two-phase epidemiologic study design to first discover genome-wide circulating miRNAs significantly differentiating hepatocellular carcinoma cases from controls, and then validating candidate miRNAs in the same discovery set and in a validation set including 49 hepatocellular carcinoma cases and matched controls.

A hospital-based hepatocellular carcinoma case–control study is ongoing in Columbia University Medical Center (CUMC; New York, NY), which is approved by the Institutional Review Board. Written informed consent was obtained from each participant. Patients with hepatocellular carcinoma and controls in the current study were enrolled between October 2008 and August 2011. Cases were newly diagnosed hepatocellular carcinoma patients who were treated in the Hepatobiliary Oncology Clinics, CUMC, and examined pathologically. Histologic evaluation of hematoxylin and eosin stained 4 micron thick sections of frozen tissue stored at −20°C assessed for the presence and percent tumor. Tumor stage was determined according to the American Joint Committee on Cancer criteria (44). Separate blocks of nontumor liver tissues were evaluated with respect to the presence (Batts–Ludwig stage of IV) or absence of cirrhosis (Batts–Ludwig stage<IV). The inclusion criteria for hepatocellular carcinoma cases were pathologically confirmed diagnosis. Cases were excluded from the current study if they had a history of other cancers.

Controls were recruited from volunteers through the Research Recruitment and Minority Outreach (RRMO) core of Herbert Irving Comprehensive Cancer Center (HICCC; New York, NY). Flyers were placed at strategic locations around CUMC where hospital visitors and employees frequent or were handed out at inreach events at the hospital or at outreach events in the community. Interested participants were directed to contact the trained recruitment staff from the RRMO and given further information about participation. Interested participants were excluded from the control group if diagnosed for any kind of cancer or liver disease. Eligible controls were asked to fill out the same demographic and epidemiologic questionnaire as hepatocellular carcinoma cases. In the current study, controls were matched with hepatocellular carcinoma cases on age (±5 years), gender (male/female), and ethnicities (Caucasian/Hispanic/African-American/Asian).

After completing the questionnaire, all participants were asked to provide a 15 mL blood sample, obtained by a trained phlebotomist. Blood samples were processed by the Biomarkers Shared Resource of HICCC according to a standardized protocol (processing blood either the day of collection, normally within 2 hours, or the next morning for samples drawn after 5 pm). Overnight bloods were kept chilled. Time of collection and processing were recorded in the core database. All blood samples in the current study were collected during the daytime, so the time of phlebotomy was not a confounder. Because a few control samples were processed the next day, we compared miRNA profiles for those bloods collected or processed on the same day (16 subjects) and processed the next day (four subjects) to identify potential confounding. No significant difference was observed for the expression of most miRNAs (747 out of total 750 miRNAs). Only three miRNAs (miR-26a, miR-16, and miR-342-3p) were significantly repressed in samples processed the next day (data not shown). These data indicated that no obvious confounding effect exists for different times to blood processing. Samples were stored in −80°C freezers connected to a telephone alarm system. A web-based inventory includes number and volume of aliquots as well as freeze-thaw cycles. In the phase I study, the preoperative plasma samples from 20 histologically confirmed hepatocellular carcinoma cases and 20 matched controls as discovery set were screened for genome-wide miRNA expression. In the phase II study, candidate miRNAs were verified in both the 20 pairs of the discovery set, as well as a validation set of 49 hepatocellular carcinoma cases and 49 matched controls.

A short, self-administered epidemiologic questionnaire was used to collect information on age, gender, race/ethnicity, place of birth of self and parents, height, weight, education, occupation, active and passive smoking, alcohol consumption, and family history of cancer. Information on HBV, hepatitis C virus (HCV) infection, and clinicopathological features including AFP levels, antiviral treatment, cirrhotic status, Milan criteria, Child-Pugh score, tumor stage, tumor size, and survival status for hepatocellular carcinoma cases were obtained from the medical and pathologic records.

Hepatitis B surface antigen (HBsAg) and HCV (anti-HCV) status in plasma samples were determined by ELISA Kits (BioChain Institute Inc.) if not available from the medical records. Total RNA, including miRNAs was isolated from 250 μL plasma using miRNeasy Mini Kit (Qiagen) according to the manufacturer's protocol. TaqMan Low Density Arrays (TLDA; Applied Biosystems), including card A and B, were used to measure genome-wide miRNA expression profiles. A three-step process was performed involving a reverse transcription reaction followed by a preamplification reaction and quantitative real-time PCR (qRT-PCR). Cycle threshold (C_t) values were calculated using the SDS2.2.2. The complete set of TLDA data has been deposited in NCBI's Gene Expression Omnibus (45) and are available through series accession number GSE50013 (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?&acc = GSE50013). TaqMan MicroRNA Assays were used for quantification of dysregulated miRNAs identified by TLDA. RT-PCRs were run in duplicate, and the average C_t and SD were calculated. U6 snRNA was used as an endogenous control to normalize the relative expression of target miRNAs and are presented as 2^(-ΔCt) (26). The fold-changes of paired samples were determined by the |$2^{- {\rm \Delta \Delta}C_{\rm t}}$| approach using DataAssist v2.0 (Applied Biosystems; ref. 46). Because of skewing of miRNA levels, all data were log₂ transformed. In a pilot study, we spiked 250 μL plasma with 100 fmol of a chemically synthesized cel-miR-39 (Invitrogen) to normalize the RNA extraction, reverse transcription, and qRT-PCR processes. The reproducibility of measuring cel-miR-39 was excellent with an overall R² = 0.99 (P < 0.01). The average inter- and intra-assay coefficients of variation were 1.47% and 0.84%, respectively, indicating the accuracy of the quantitation.

We applied Biometric Research Branch array tools to conduct data analysis with the original results transformed to log-scale and adjusted for reference scale (47). A two-sample t test was used to compare the difference in geometric mean between cases and controls. A volcano plot was constructed for detectable miRNAs (detected in at least two paired samples). We filtered miRNAs that had missing values for more than 20% of samples, and explored significant miRNAs by the univariate test and Benjamini–Hochberg false discovery rate (FDR) adjustment (P < 0.05). We generated heatmaps and hierarchical plots to check the clustering of samples based on significant miRNAs.

Four miRNAs overexpressed in hepatocellular carcinoma and detectable in every sample were selected for further testing by qRT-PCR in the same 20 pair discovery set and a validation set of 49 paired hepatocellular carcinoma cases and controls. Gene ontology analysis (www.pantherdb.org/) was performed by the PANTHER classification system to compare potential target genes affected by miRNAs with the NCBI reference (human genome build 36; ref. 48). The binomial test was used to identify significantly enriched pathways, biologic processes, molecular functions, cellular components, and protein class terms after Bonferroni correction for multiple comparisons with a cutoff of P ≤ 0.05.

miRNA expression levels were compared by genometric means and using the 75th percentile in controls as cutoff points for overexpression to increase specificity. A two-sample t test assuming equal variance was carried out to compare log₂ miRNA levels by categorical covariates of hepatocellular carcinoma risk factors and clinic-pathological characteristics, including gender (male vs. female), HBsAg (negative vs. positive), anti-HCV (negative vs. positive), cigarette smoking (no vs. yes), alcohol drinking (no vs. yes), AFP (<400 ng/mL vs. ≥400 ng/mL), cirrhosis (no vs. yes), within Milan criteria (yes vs. no), Child-Pugh score (A vs. B), tumor size (<5 cm vs. ≥5 cm), tumor stage (I-II vs. III-IV), and survival (yes vs. no). Logistic regression was used to construct receiver-operating characteristic (ROC) curves by miRNA level adjusted for age and ethnicity (49). The maximum sensitivity and specificity and the area under the curve (AUC) were identified by numerical integration of each ROC curve. A better prediction model was built by fitting miRNAs, HBV, and HCV status into the logistic regression model. The stepwise backward model selection was performed to determine the combinations of miRNA and viral status that significantly discriminate patients with hepatocellular carcinoma from controls. A likelihood ratio test P < 0.05 was considered statistically significant. Statistical analyses were completed using Statistical Analysis System 9.0 (SAS Institute).

The demographic and clinicopathologic characteristics of 20 pairs in the discovery set and 49 pairs in the validation set are shown in Table 1. The mean ages, gender, and ethnicity are similar for hepatocellular carcinoma cases and controls in both sets. Most cases are HCV infected (55% and 66%), values significantly higher than among matched controls. Twelve percent and 20% of hepatocellular carcinoma cases are HBsAg positive; this is significantly higher than controls in the discovery set but not in the validation set. For the clinical covariants, there are no statistically significant differences for hepatocellular carcinoma cases in discovery and validation sets for antiviral treatment, AFP levels, Milan criteria, Child-Pugh score, tumor stage, cirrhosis, and survival outcome. Only tumor size significantly differs (P = 0.047) between the two sets, indicating the need for further subgroup analysis.

TLDA data for the 20 pairs in the discovery set showed that 255 miRNAs (34.0% of 750 miRNAs) were detectable in circulation (defined as expression in at least two samples for each group). Analyzing a total of 91 miRNAs with 80% or more detection rates, we found that 66 miRNAs were significantly overexpressed in hepatocellular carcinoma cases compared with controls after adjusting for FDR (Supplementary Table S1). Hierarchical cluster analysis found that miRNA expression patterns in hepatocellular carcinoma cases were significantly different from controls, that is, genome-wide miRNA overexpression was more common in hepatocellular carcinoma cases compared with controls (Supplementary Fig. S1). A volcano plot shows that seven miRNAs (miR-150, miR-30b, miR-30c, miR-376a, miR-483-5p, miR-520b, and miR-720) have more than 2-fold case–control differences with a P < 0.01 (Fig. 1). The log₂ fold-changes of these seven miRNAs between 20 paired hepatocellular carcinoma cases and controls are shown in Supplementary Fig. S2. Overexpression (log₂ fold-change > 0) is consistently observed for most hepatocellular carcinoma cases (75%–90%). No significantly downregulated miRNAs were observed.

There were 36 and 30 miRNAs significantly associated with HBsAg or HCV-positive hepatocellular carcinoma, respectively, compared with viral negative controls at the nominal 0.05 level of the univariate test (Supplementary Table S1). No miRNA remained statistically significant after FDR adjustment, except for miR-1243 among HbsAg-positive hepatocellular carcinoma cases (P = 0.014). To examine viral-related miRNAs, we compared expression profiles by viral status (HBsAg or HCV positive) in patients with hepatocellular carcinoma. Seven miRNAs were specifically associated with viral infection, but unrelated with hepatocellular carcinoma (P < 0.05, data not shown).

We further verified four miRNAs (miR-150, miR-30c, miR-483-5p, and miR-520b) by qRT-PCR in the same 20 pairs of discovery set, and in another validation set including 49 cases and 49 matched controls. The four miRNAs were selected because they were detectable by TLDA with varied expression levels (high, moderate, or low) in all subjects. Expression of miR-520b was undetectable for all plasma samples by qRT-PCR, and was omitted in the data analysis. Through PANTHER ontology analysis, we examined the biologic characteristics of 1,497 conserved genes potentially targeted by the three detectable miRNAs (miR-150, miR-30c, and miR-483-5p). Enriched genes were significantly associated with more than 20 biologic pathways, including pathways of Hedgehog, interleukin, TGF-β, EGF receptor/Wnt signaling, p38 mitogen—activated protein kinase, Ras, p53, phosphoinositide 3-kinase, etc. (Supplementary Table S2). Some of these genes were previously identified as mutated or with copy number changes in hepatocellular carcinoma (50, 51). Our data confirm the broad range of targets for the three miRNAs potentially involved in the pathophysiology of hepatocellular carcinoma.

The correlations in expression levels (log₂ transformed) between TLDA and qRT-PCR assays for the three miRNAs were significant (Supplementary Fig. S3). The r were 0.57, 0.60, and 0.79 for miR-30c, miR-150, and miR-483-5p, respectively (P ≤ 0.0001), indicating excellent correlations if the miRNAs were detected by both assays. However, when separately comparing the qRT-PCR data for the three miRNAs between hepatocellular carcinoma cases and controls in the discovery and validation sets, only miR-483-5p is significantly overexpressed in hepatocellular carcinoma cases, which is consistent with the result from the TLDA data (Table 2). The fold-change in miR-483-5p was 3.8 and 9.4, suggesting the potentially important role of miR-483-5p upregulation in hepatocellular carcinoma. No consistent expression patterns were observed for miR-30c and miR-150 between the discovery and validation sets.

To increase the specificity of dysregulated miRNA in detecting hepatocellular carcinoma, we used the 75th percentile among controls as a cutoff point for overexpression. Only miR-483-5p is significantly associated with increased hepatocellular carcinoma risk compared with controls in the validation set [OR = 9.7; 95% confidence interval (CI), 2.6–35.3; P = 0.0006; Table 3] after adjustment for age, gender, ethnicity, HBV, and HCV status, which is consistent with the finding in discovery set (Supplementary Table S3). No significant case–control differences were observed for miR-150 and miR-30c in the discovery or validation sets (Table 3 and Supplementary Table S3).

To examine the potential influence of hepatocellular carcinoma risk factors and clinic-pathological factors on expression of the three miRNAs, we separately conducted subgroup analyses for those covariates in the discovery and validation sets using the qRT-PCR data. Among hepatocellular carcinoma cases, we found no significant differences in miRNA expression by HBV, HCV, antiviral treatment, cirrhosis, tumor stage, or survival subgroups (Supplementary Tables S4 and S5). Although differences were seen with respect to certain covariates in one set, no consistent influences were found in both sets. For example, miR-483-5p differs by Child-Pugh score in the discovery but not in validation set; miR-30c and miR-483-5p differ by AFP and tumor size in the validation but not in discovery set. The three miRNAs also display no significant difference in expression by HBV or HCV status among controls (data not shown). These data suggest that the influence of the covariates on miRNAs may be minor, although the small sample size in subgroup analysis limits the interpretation of the results.

To assess the predictive accuracy of miRNAs in detecting hepatocellular carcinoma cases, multivariable logistic regression was used to construct ROC curves in combination with HBV and HCV infection status. Using a stepwise selection model to gradually eliminate nonsignificant covariates, the best predictive model including miR-483-5p and HCV has a predictive accuracy of 0.908 (P < 0.0001), a sensitivity of 75.5%, and a specificity of 89.8% (Fig. 2). HCV had an AUC of 0.796 (P < 0.0001), and either viral infection (HBV, HCV) had an AUC of 0.815 (P < 0.0001). MiR-483-5p expression can significantly differentiate hepatocellular carcinoma cases from controls with an AUC of 0.827 (P < 0.0001), which is better than the accuracy of using viral status alone. The sensitivity and specificity for miR-483-5p alone are, respectively, 55.1% (27/49) and 85.7% (42/49; Supplementary Table S6).

We used low-density array to first examine genome-wide expression for 750 miRNAs, and then validated three candidate miRNAs in the same discovery set and a validation set by qRT-PCR. A total of 66 miRNAs were significantly differentially expressed between hepatocellular carcinoma cases and controls after adjusting for FDR (Supplementary Fig. S1 and Supplementary Table S1). More importantly, 59 miRNAs were initially identified as circulating biomarkers for hepatocellular carcinoma, and seven (miR-19b-1, miR-24, miR-29c, miR-376a, miR-378, miR-520c-3p, and miR-92a) were consistent with previous findings. A volcano plot indicated that seven miRNAs (miR-150, miR-30b, miR-30c, miR-376a, miR-483-5p, miR-520b, and miR-720) had more than 2-fold case–control differences with an adjusted P < 0.01 (Fig. 1). To our knowledge, no previous study has characterized these miRNAs in hepatocellular carcinoma. In a validation study, miR-483-5p overexpression was significantly associated with increased hepatocellular carcinoma risk (OR = 6.8; 95%CI, 2.1–22.2; P = 0.002; Table 3). Multivariable logistic regression indicated that miR-483-5p expression and HCV status could significantly differentiate hepatocellular carcinoma cases from controls with an AUC of 0.908 (Fig. 2). These data suggest that circulating miR-483-5p may be a useful biomarker for hepatocellular carcinoma detection.

Limited data are available for miR-483-5p expression in human cancer. Because pre-miR-483-5p maps to intron 2 of insulin-like growth factor II (IGF-II; ref. 20), a gene that is highly expressed in adrenocortical carcinoma (52, 53) and pheochromocytomas (54), it is reasonable to assume that miR-483-5p may be coexpressed with its host gene (55). In support of this hypothesis, several previous studies found positive correlations between miR-483 expression and IGF-II mRNA levels in Wilms tumor, colorectal cancer, malignant pheochromocytoma, and hepatocellular carcinoma tissues (52–55). One study also found elevated expression of miR-483-5p in serum from hepatocellular carcinoma cases (36), which is consistent with our current observation. The pathogenic role and molecular mechanism of action of miR-483-5p in tumorigenesis remain unknown. miR-483-3p was found to function as an antiapoptotic oncogene in cancer cell lines (HEPG2, liver carcinoma and HCT116, colorectal carcinoma) (53). An in vitro study of adrenocortical carcinoma revealed a growth-promoting role for miR-483-5p (56). These data suggest a potential carcinogenic role for miR-483-5p in tumorigenesis.

Despite the promising finding that elevated plasma miR-483-5p can differentiate hepatocellular carcinoma cases from controls, it is still unknown whether this is due to active secretion from the tumor tissues. This is one of the major limitations in the current study. In an ongoing study, we found that miR-483-3p and miR-483-5p were highly expressed in hepatocellular carcinoma tumor tissues (2.8–13.9-fold) compared with adjacent nontumor tissues, consistent with our current finding. From the methodologic point of view, another limitation is the preamplification step involved in the discovery phase using the TLDA arrays. The qRT-PCR assays used in the validation approach do not require preamplification. This may be one reason for the inconsistent results for miR-30c and miR-520b in the discovery and validation studies. More sensitive assays or protocols to enrich for circulating miRNAs may be required. The cross-sectional study design with no postsurgical plasma samples collected at different time points prevents us from obtaining information on the causal association between aberrant miRNAs and hepatocellular carcinoma. Therefore, it is imperative to establish longitudinal biospecimen repositories to clarify the critical role of circulating miRNAs in the long-term process of hepatocarcinogenesis.

In summary, our study suggests that circulating miR-483-5p may be a potential biomarker for less invasive detection of hepatocellular carcinoma. These data support the idea that blood is a promising resource for novel miRNA biomarker discovery in addition to its ability to monitor genetic variation and DNA methylation alterations. Further evaluation of the biologic functions of candidate miRNAs in tumorigenesis, such as angiogenesis, chronic inflammation, or cellular proliferation, will provide more evidence to understand the role of miRNAs as hepatocellular carcinoma biomarkers.

No potential conflicts of interest were disclosed.

Conception and design: J. Shen, R.M. Santella

Development of methodology: J. Shen

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): I. Gurvich, A.B. Siegel, H. Remotti, R.M. Santella

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J. Shen, A. Wang, A.B. Siegel

Writing, review, and/or revision of the manuscript: J. Shen, A.B. Siegel, H. Remotti, R.M. Santella

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J. Shen, Q. Wang, I. Gurvich

The authors thank Dr. Victor R. Grann and Kazeem Abdul in the Research Recruitment and Minority Outreach Core, Herbert Irving Comprehensive Cancer Center (HICCC) for recruitment of control subjects for the current study and all subjects who participated by donating blood for this study.

This work was financially supported by NIH grants R03 CA156629 (to J. Shen), R01 ES005116 (to R.M. Santella), P30 ES009089 (to R.M. Santella), P30 CA013696 (to R.M. Santella), and a pilot of NIEHS Center for Environmental Health in Northern Manhattan (to J. Shen).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.