MicroRNA-21 is overexpressed in human cholangiocarcinoma and regulates programmed cell death 4 and tissue inhibitor of metalloproteinase 3


  • Potential conflict of interest: Nothing to report.


Cholangiocarcinomas (CCAs) are aggressive cancers, with high mortality and poor survival rates. Only radical surgery offers patients some hope of cure; however, most patients are not surgical candidates because of late diagnosis secondary to relatively poor accuracy of diagnostic means. MicroRNAs (miRs) are involved in every cancer examined, but they have not been evaluated in primary CCA. In this study, miR arrays were performed on five primary CCAs and five normal bile duct specimens (NBDs). Several miRs were dysregulated and miR-21 was overexpressed in CCAs. miR-21 differential expression in these 10 specimens was verified by quantitative reverse transcriptase polymerase chain reaction (qRT-PCR). To validate these findings, qRT-PCR for miR-21 was then performed on 18 additional primary CCAs and 12 normal liver specimens. MiR-21 was 95% sensitive and 100% specific in distinguishing between CCA and normal tissues, with an area under the receiver operating characteristic curve of 0.995. Inhibitors of miR-21 increased protein levels of programmed cell death 4 (PDCD4) and tissue inhibitor of metalloproteinases 3 (TIMP3). Notably, messenger RNA levels of TIMP3 were significantly lower in CCAs than in normals. Conclusions: MiR-21 is overexpressed in human CCAs. Furthermore, miR-21 may be oncogenic, at least in part, by inhibiting PDCD4 and TIMP3. Finally, these data suggest that TIMP3 is a candidate tumor suppressor gene in the biliary tree. (HEPATOLOGY 2009.)

Cholangiocarcinoma, an epithelial cancer of the biliary tree,1 is the second most common primary hepatic malignancy,2 with a United States incidence of 0.82-0.95/100,000.3 Although the rate of extrahepatic CCA has remained stable, intrahepatic CCA has increased by more than 100% in the past 40 years.4 Survival in CCA is dismal, usually measured in months.5 The only potentially curative treatment for CCA is surgical resection.6 Unfortunately, the vast majority of patients are diagnosed at late stages, when surgery is not a viable option.7, 8 This late diagnosis stems from a paucity of disease-specific symptoms in early stages.2 Moreover, in earlier stages of CCA, diagnostic techniques are plagued by low specificity.2 These considerations demand a better understanding of cholangiocarcinogenesis, with an emphasis on biomarkers for earlier diagnosis.

MicroRNAs (miRs) are noncoding RNAs 18 to 25 nucleotides in length.9 The effects of miRs are mediated by binding to target messenger (m)RNAs and either suppressing translation or degrading miR-bound mRNA.10 Groundbreaking research in CCA cell lines has revealed important miR differences among CCA and normal cholangiocytic cell lines.11 MiR-21 was overexpressed in CCA cell lines,11 and this same pioneering study identified the gene PTEN (phosphatase and tensin homolog deleted on chromosome 10) as a potential miR-21 target. Further cutting-edge research found miR-29 to be down-regulated in the CCA cell line, KMCH, and identified Mcl-1 as a target. Finally, in additional studies, down-regulation of miR-370 was discovered in the CCA cell lines MzChA-1 and KMCH-1.12 These seminal in vitro studies laid a foundation for understanding the impact of miRs on cholangiocarcinogenesis. In the current study, we sought to gain insight into miR dysregulation in primary human CCAs. We hypothesized that molecular findings from direct in vivo analyses would complement insights gained from previous fruitful in vitro studies in helping us to understand this deadly disease.


CCA, cholangiocarcinoma; Ct value, cycle passing threshold; FDR, false discovery rate; HCC, hepatocellular cancer; miR-21, MicroRNA-21; miR21i, miR-21 inhibitor; NBD, normal bile duct specimen; NSI, nonspecific miR inhibitor; PDCD4, programmed cell death 4; PTEN, phosphatase and tensin homolog deleted on chromosome 10; qRT-PCR, quantitative reverse transcriptase polymerase chain reaction; ROC, receiver operating characteristic; SAM, significance analysis of microarrays; SD, standard deviation; TIMP3, tissue inhibitor of metalloproteinases 3.

Patients and Methods

Human Tissues.

The first five CCA specimens were obtained at surgeries performed at Johns Hopkins University (Baltimore, MD). We also collected five primary normal bile duct (NBD) specimens from surgical resections performed for pancreatic cancer. These patients underwent a Whipple's procedure. Bile ducts contained in the excision specimens were confirmed histologically to be free of tumor. After the initial stage of the study, an additional 18 patients were included (Table 1). These patients underwent resections for CCA. From 12 of these patients, paired normal and cancer tissues were obtained. From the remaining six patients, only cancer tissues were obtained. Therefore, a total of 30 specimens was added to the study: 18 cancers and 12 normal tissues. The normal tissues were confirmed histologically to not contain any cancer. Informed consent for surgical procedures and for using the specimens for research was obtained from all patients under approved Johns Hopkins University Institutional Review Board (IRB) protocols.

Table 1. Clinicopathologic Data for CCAs
  1. Location: intrahepatic (I) or extrahepatic (E). Two cancers originated in the perihilar region and were labeled extrahepatic-Klatskin tumor (E-K) to differentiate them from the distal extrahepatic cancers (E). Age: age at surgery. Gender: female (F) and male (M). Size: measured in centimeters (cm) at the time of resection. Staging: TNM staging. Differentiation: well (W), moderately (M), moderately-poorly (MP), and poorly (P) differentiated. NA: not available.

T1I58F5.5 cmT1NxMxNA
T2I73M4.5 cmT3N1MxM
T3I77F7 cmT3N0MxNA
T4E84F2.2 cmT3N1MXP
T5I66F3.2 cmT1N0M0M
T6E64M1.5 cmT3N1M0M
T7E-K46F3.5 cmT3NxMxM
T8E84F3.5 cmT3N1MxP
T9I57M8 cmT1N0M0M
T11E66M2.5 cmT3N1MxP
T12E85F4 cmT3N1MxM
T13E70M2 cmT3N1MxP
T14I61M6 cmT3N1MxM-P
T17I71F19 cmT3N0MxP
T18I54F5.4 cmT4N0M1M-P
T19I36F0.3 cmT1N0MXM-P
T20I49F3 cmT2NXMXM
T21I50F7 cmT1N0MoM
T22E47M3 cmT3N1MxM-P
T23E-K54M4 cmT2N0MxW

Cell Lines.

CAK-1 and HuCCT1 were generated at Johns Hopkins University from extrahepatic CCAs.13 TFK1 was established from a common bile duct CCA resected surgically.14

DNA and RNA Extraction.

Total RNA was isolated using TRIzol reagent (Invitrogen, La Jolla, CA).

MicroRNA Arrays.

In all, 100 ng of total RNA for each specimen was used for miR arrays. We employed an Agilent Human miR array chip (Agilent, Santa Clara, CA) containing 15,000 probes corresponding to 470 unique human miRs. Data were extracted using Feature Extraction Software 9.3 and GeneSpring software (Agilent). A raw array value below five was considered background level. Array data were normalized to the control small RNA species printed on the slide. Data from miRs with at least 5 of 10 values above background were used for further analyses. Data were analyzed using significance analysis of microarrays (SAM, Stanford University, Stanford, CA). A false discovery rate (FDR) less than 2% was considered acceptable.

Quantitative RT-PCR (qRT-PCR) for miR Expression.

We performed miR qRT-PCR to confirm the expression of candidate miRs. TaqMan MiR assays, human (Applied Biosystems, Foster City, CA) were used. Cycle passing threshold (Ct) was recorded and normalized to RNU6B expression. Relative expression was calculated as 2Ct_miR-21-Ct_RNU6B. PCR reactions were carried out in duplicate. All qRT-PCR values were calculated as ratios to the U6-normalized qRT-PCR miR-21 value in the first normal specimen. For calculating the sensitivity and specificity of miR-21 to diagnose CCAs versus noncancerous tissues, a cutoff of 4.5 was used.

Quantitative qRT-PCR for mRNA Expression.

iQ SYBR Green Supermix (Bio-Rad, Hercules, CA) was used. Primer sequences are available in Supporting Table 1. PCR products were confirmed by melting curve analysis. Beta-actin was used to normalize mRNA expression levels. Relative expression was calculated as 2Ct_target gene-Ct_beta-actin. PCR reactions were carried out in duplicate.

Transfection of miR-21 Inhibitor.

Synthesized RNA duplexes of miR-21 inhibitor were purchased from Dharmacon (Lafayette, CO). Thirty to 50% confluent cells were transfected with 60 nM of miR-21 inhibitor, or inhibitor-negative control using Lipofectamine RNAi MAX (Invitrogen). RNA and proteins were harvested 72 hours after transfection. To document that miR-21 inhibition did not affect the levels of unrelated microRNA species, we measured the level of an unrelated microRNA, miR-590-5p. Although the HuCCT1 cells treated with miR-21 inhibitor showed a significant decrease in the level of miR-21 compared to untreated cells, the level of miR-590-5p was the same in cells treated with miR-21 inhibitor as in the mock condition (Supporting Fig. 1). To document that miR-21 transfection did not affect the levels of unrelated proteins, we verified that the level of p21 was not affected. Supporting Fig. 2 shows that transfection with miR-21 inhibitor does not affect levels of p21 in TFK1 CCA cells.

Western Blot.

Cells were lysed in Laemmli sample buffer (Bio-Rad) supplemented with a protease inhibitor (Complete, EDTA-free; Roche, Nutley, NJ). Protein concentration was measured using a BCA Protein Assay kit (Pierce, Rockford, IL). Cell lysates (50 μg) were electrophoresed on 10% to 20% polyacrylamide gels (Bio-Rad) and transferred to Immobilon-PSQ membranes (Millipore, Bedford, MA). The membranes were blocked with TBS containing 5% skim milk and 0.1% Tween-20, then incubated with the primary antibody. Anti-PDCD4, anti-TIMP3, and anti-p21 antibodies produced in rabbit (Sigma-Aldrich, St. Louis, MO, 63178 for PDCD4 and TIMP3, and Zymed, San Francisco, CA) were used according to the manufacturer's instructions. The membranes were incubated after washing with the secondary antibody, horseradish peroxidase-conjugated goat antirabbit IgG (Calbiochem, Gibbstown, NJ) and analyzed using enhanced chemiluminescence-plus reagent (GE Healthcare, Buckinghamshire, UK).


MicroRNA Arrays Identify Differentially Expressed miRs in Primary CCA Specimens.

Microarrays were performed on five NBD and five CCAs. In all, 221 miRs exhibited at least 50% of values above background. With the FDR set at 1.44, SAM identified 20 overexpressed and 112 underexpressed miRs in CCAs versus NBDs. This dysregulation imbalance agreed with previous in vitro CCA studies.11 The top 10 miRs in each category are listed in Table 2A and B. MiR-200b, overexpressed in CCA cell lines Mz-ChA-1 and TFK,11 was not overexpressed in our primary CCAs. Similarly, although overexpressed in cell lines,11 miR-141 was not overexpressed in our primary CCAs. Conversely, miRs-93, -25, -21, and -27a, were found overexpressed both in CCA cell lines11 and in our primary CCAs. Of note, the miR 25-93-106b cluster was recently found overexpressed and involved in gastric cancer.15 Interestingly, miR-106b, of the same family with miRs-93 and -25, was also overexpressed in our primary CCAs. Among miRs underexpressed in primary CCAs, miR-560 has not been previously described in human cancers. In contrast, miR-370 was found underexpressed in both our primary CCAs and in cell lines.12 Analogously, miR-198, underexpressed in our primary CCAs, is underexpressed in hepatocellular carcinomas (HCCs).16

Table 2A. MiRs Overexpressed in Human CCA vs. Normal Specimens
Gene IDScore(d)Fold Change
Table 2B. MiRs Underexpressed in Human CCA vs. Normal Specimens
Gene IDScore(d)Fold Change
  1. The 10 most overexpressed miRs and the 10 most underexpressed miRs in CCA vs. normal specimens are listed. Score(d), SAM score.


MiR-21 Is Overexpressed in Human CCA Specimens.

In our array data, miR-21 was the most-fold overexpressed microRNA and had the second-highest SAM score (Table 2, Fig. 1). Its average expression was 7.7-fold greater in CCAs than in NBDs. This ratio resembles the in vitro difference reported for miR-21 in Mz-ChA-1 (2.74-fold) and in TFK (4.35-fold)11 and agrees with similar findings in HCC tissues.17 Moreover, a similar fold difference (4.7 to 10-fold) was reported in HCC cell lines as well as in HCC versus normal liver tissues (2 to 65-fold).18

Figure 1.

Array data and qRT-PCR data for miR-21 in 10 primary tissue specimens. X-axis, specimens: N1-5, normal bile duct specimens; T1-5, primary CCAs. Y-axis, miR-21 values from array and qRT-PCR data. Array and qRT-PCR data are ratios to specimen N1. Solid bars, array data; open bars, qRT-PCR values.

qRT-PCR Data for miR-21 Validate Array Findings.

Our qRT-PCR data closely matched our array data (Fig. 1). The Pearson correlation coefficient between array and qRT-PCR was 0.91. The average expression of miR-21 measured by qRT-PCR was 5.9-fold higher in CCAs versus NBDs, confirming our array data.

Differential qRT-PCR Expression for miR-21 Is Validated Prospectively (Fig. 2A).

To prospectively validate miR-21 overexpression in human CCA, we verified its expression in the remaining 30 specimens included in the study: 18 CCAs and 12 normal tissues. Normal tissues displayed uniformly low expression of miR-21, with a standard deviation (SD) of 1.34. In contrast, CCAs displayed more variable expression (SD 11.19). No correlation was found between the level of miR-21 expression and cancer location (extrahepatic vs. intrahepatic), TNM stage, or histologic grade. All but one cancer displayed miR-21 levels greater than the highest value in normal specimens. The fold difference between the cancer and normal groups was 5.4, consistent with our comparison of five cancers versus five normal tissues.

Figure 2.

(A) qRT-PCR data for miR-21 in 20 primary CCAs and 14 normal tissues. Blue triangles, normal tissues; red diamonds, cancers. Y-axis, miR-21 qRT-PCR values as ratios to the N1 normal specimen; P-value obtained by Student's unpaired t test. (B) ROC curve built using miR-21 qRT-PCR data from all 34 specimens. X-axis, 1-specificity; Y-axis, sensitivity. The area under the curve equaled 0.995 (asymptotic 95% confidence interval: 0.981-1.008).

MiR-21 Accurately Discriminates Between CCA and Normal Bile Duct.

miR-21 correctly diagnosed 22 of 23 cancers and 17 of 17 normal specimens, yielding a sensitivity of 95% and specificity of 100%. The area under the ROC curve (AUROC) was 0.995 (Fig. 2B).

MiR-21 Targets Programmed Cell Death 4 (PDCD4) in CCA.

By in silico searches, we identified PDCD4 as a potential miR-21 target. PDCD4 is down-regulated in HCC tissues19 and is involved in TGF-beta1–induced apoptosis in an HCC cell line19; however, it has not been previously implicated in cholangiocarcinogenesis. By extension, we hypothesized that PDCD4 may be a tumor-suppressor gene in CCA, and that its expression could be regulated by miR-21. Transfecting the CCA cell lines CAK1, TFK1, and HuCCT1 with a miR-21 inhibitor (miR-21i) resulted in a dramatic increase in PDCD4 protein levels (Fig. 3A). This finding strongly suggests that PDCD4 is regulated by miR-21 in CCA.

Figure 3.

Western blot using (A) anti-PDCD4 antibody and (B) anti-TIMP3 antibody. NT, untreated cells; NSI, cells treated with control nonspecific miR inhibitor; 21i, cells treated with specific miR-21 inhibitor. Lower panels, beta-actin.

MiR-21 Regulates PDCD4 at the Level of Protein Translation in CCA Specimens.

To elucidate the mechanism by which miR-21 regulates PDCD4, we measured mRNA levels of PDCD4 in normal liver and CCA specimens. Figure 4A shows that PDCD4 mRNA was approximately equal in CCA and normal tissues (mean level 4.32 in normal vs. 4.4 in CCAs). These findings argue that miR-21 inhibits PDCD4 protein production, rather than degrading its mRNA. Figure 4B shows that there was no effect on PDCD4 mRNA levels following miR-21i transfection of HuCCT1 and TFK1 CCA cells. This finding further suggests that, in CCA, miR-21 regulates PDCD4 at the level of protein translation. After we performed our own experiments, a similar miR-21/PDCD4 interaction was recently observed in the human embryonic kidney cell line HEK-293T.20

Figure 4.

(A) mRNA qRT-PCR for PDCD4. X-axis, normal (blue triangles) and cancerous (red diamonds) primary specimens. Y-axis, PDCD4 mRNA qRT-PCR value relative to specimen N1. P-value, Student's unpaired t test. (B) PDCD4 mRNA qRT-PCR data after treatment with specific versus nonspecific miR inhibitors. Solid bars, PDCD4 mRNA levels in cells treated with control nonspecific miR inhibitor. Open bars, PDCD4 mRNA levels in cells treated with specific miR-21 inhibitor. Experiments were performed in triplicate in HuCCT1 as well as in TFK1 CCA cell lines.

Tissue Inhibitor of Metalloproteinases 3 (TIMP3) mRNA Is Statistically Significantly Underexpressed in CCAs.

By employing in silico searches, TIMP3 was also identified as a miR-21 target in CCA. The sole study suggesting TIMP3's involvement in cholangiocarcinogenesis reported that 8.9% of 79 intrahepatic CCA tissues displayed TIMP3 promoter hypermethylation.21 In contrast, 42% of HCCs analyzed showed TIMP3 hypermethylation,22 suggesting that this mechanism may be more important in HCCs. Moreover, up-regulation of TIMP3 expression in the HCC cell line HCC-7721 inhibits invasion in vitro and metastasis in nude mice.23 To assess the expression of TIMP3 in our CCA specimens, we performed qRT-PCR of TIMP3 mRNA. Figure 5 shows that TIMP3 mRNA was significantly higher in normal than in CCA tissues. This finding strongly suggests, for the first time, that TIMP3 is an important tumor suppressor gene in cholangiocarcinogenesis.

Figure 5.

TIMP3 mRNA qRT-PCR data. X-axis, normal (blue triangles) and cancerous (red diamonds) primary specimens. Y-axis, TIMP3 mRNA qRT-PCR levels. P-value, Student's unpaired t test.

MiR-21 Targets TIMP3 in CCA.

Based on the qRT-PCR assays for TIMP3 mRNA and miR-21, we hypothesized that miR-21 is a major regulator of TIMP3. To test this hypothesis, we transfected TFK1 and HuCCT1 with NSI or miR-21i and performed qRT-PCR for TIMP3 mRNA. For both TFK1 and HuCCT1, TIMP3 did not amplify to 50 cycles in either untreated or NSI-treated cells. Upon treating with miR-21i, the mRNA for TIMP3 was amplified at cycle 42 in TFK1 cells and 35 in HuCCT1 cells. We then transfected CAK1, TFK1, and HuCCT1 with miR21i or NSI. Figure 3B shows that miR-21 inhibition resulted in a remarkable increase in TIMP3 protein levels. These results show, for the first time, that miR-21 is an important regulator of TIMP3 protein level in CCA. A recent study performed in glioma cell lines demonstrated that miR-21 does not directly bind to the TIMP3 3′UTR. This finding suggests that the interaction between miR-21 and TIMP3 in glioma cell lines is likely indirect.24


Previous seminal microRNA research in CCA has been performed in cell lines.11, 12, 25 Interestingly, work from Dr. Tushar Patel's laboratory revealed that microRNA profiles are not identical in CCA cell lines.11 This finding suggests that different miRs exert unique carcinogenic contributions, depending on the particular environment in which they occur. Therefore, in the first stage of our study, we used human primary CCA specimens rather than cell lines in an attempt to appreciate the impact that miRs may exert on cholangiocarcinogenesis in vivo. Our results confirmed many, although not all, of the profiling results found in cell lines. Among miRs overexpressed in primary human CCA tissues, miR-21 had also been reported in previous cell line research,11 perhaps because of its paramount involvement in cholangiocarcinogenesis. Although miR-21 has been reported as overexpressed in a variety of tumors,18, 24, 26–29 it has not been reported in primary human CCA tissues. Moreover, our study suggests that miR-21 is uniformly overexpressed in human CCA, in view of our finding that it was 95% sensitive and 100% specific in diagnosing human CCA. This finding may address the existing difficulty of diagnosing CCA by current means, such as imaging.

PDCD4 was initially identified as an up-regulated apoptosis-related protein,30 suggesting its role as a tumor suppressor gene.31 Although PDCD4 was recently reported as a potential tumor suppressor in hepatocarcinogenesis,19 our work is the first reporting its involvement in cholangiocarcinogenesis. Moreover, this is the first report of PDCD4 being inhibited by miR-21 in CCA. The interaction between miR-21 and PDCD4 was previously reported in breast cancer20, 32, 33 and colon cancer,34 suggesting that this interplay may be a general carcinogenic pathway, rather than a tissue-specific mechanism. Furthermore, because the mRNA levels of PDCD4 were similar in human CCA and normal specimens, we conclude that miR-21-induced PDCD4 inhibition is posttranscriptional (Fig. 4A). Confirming our findings, a similar level of mRNA for PDCD4 was found in colon cancer and normal specimens in a previous study.34 As shown before in Colo206f, HeLa, and HEK-293T cells, miR-21 directly binds to its binding site in the 3′UTR of PDCD4.20, 32–34

Matrix metalloproteinases play a crucial role in cancer invasion and metastasis.35 Among the four members of the family, TIMP3 is uniquely proapoptotic.35 For example, the adenoviral transfer of TIMP3 into HeLa, HT1080 fibrosarcoma cells, and melanoma cells reduces their invasiveness and stimulates apoptosis.36, 37 More recently, herpes simplex-mediated transfer of TIMP3 into neuroblastoma and malignant peripheral nerve tumor xenografts reduced tumor growth and reduced the density of the tumor vascular network.38 Nonetheless, there are no previous reports regarding the role of TIMP3 in CCA. A single article reported low-level promoter methylation of TIMP3 in intrahepatic CCA.21 Our study is the first to report (1) lower levels of mRNA for TIMP3 in human CCA specimens compared to normal human specimens, and (2) an inhibitory effect of miR-21 on TIMP3 in CCA cell lines. Taken together, these findings provide support for our hypothesis that TIMP3 is a tumor suppressor gene in CCA and that its activity is closely regulated by miR-21.