Skip to main content

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

MicroRNA expression alterations are linked to tumorigenesis and non-neoplastic processes in pancreatic ductal adenocarcinoma

Abstract

Pancreatic ductal adenocarcinoma (PDAC) is known for its very poor overall prognosis. Accurate early diagnosis and new therapeutic modalities are therefore urgently needed. We used 377 feature microRNA (miRNA) arrays to investigate miRNA expression in normal pancreas, chronic pancreatitis, and PDAC tissues as well as PDAC-derived cell lines. A pancreatic miRNome was established comparing the data from normal pancreas with a reference set of 33 human tissues. The expression of miR-216 and -217 and lack of expression of miR-133a were identified as characteristic of pancreas tissue. Unsupervised clustering showed that the three pancreatic tissues types can be classified according to their respective miRNA expression profiles. We identified 26 miRNAs most prominently misregulated in PDAC and a relative quantitative reverse transcriptase-polymerase chain reaction index using only miR-217 and -196a was found to discriminate normal pancreas, chronic pancreatitis and cancerous tissues, establishing a potential utility for miRNAs in diagnostic procedures. Lastly, comparing differentially expressed genes from PDAC with predicted miRNA target genes for the top 26 miRNAs, we identified potential novel links between aberrant miRNA expression and known target genes relevant to PDAC biology. Our data provides novel insights into the miRNA-driven pathophysiological mechanisms involved in PDAC development and offers new candidate targets to be exploited both for diagnostic and therapeutic strategies.

Introduction

The hallmarks of the pancreatic ductal adenocarcinoma (PDAC) are late clinical presentation, early and aggressive local invasion, metastatic potential, strong resistance to chemotherapy and radiation therapy, and most importantly, a very poor overall prognosis. The median survival time following diagnosis is only 3–5 months. Surgical resection improves the median survival time to 10–20 months, however, only 20% of PDACs are amenable to surgery (Yeo et al., 2002).

In recent years, a wealth of molecular studies from the level of individual genes/proteins to high throughput gene expression analyses using modern array technology have been applied in the quest for a better understanding of PDAC biology and the identification of new biomarkers and therapeutic targets. Recently, microRNAs (miRNAs), a novel class of 18–23 nt long non-coding RNAs, have gained attention as another family of molecules involved in cancer development (Esquela-Kerscher and Slack, 2006; Sevignani et al., 2006). Upon binding to their target RNAs, miRNAs cause post-transcriptional gene silencing by either cleaving the target mRNA or by inhibiting the translation process (Valencia-Sanchez et al., 2006).

Regulated miRNA expression has been demonstrated in a variety of eukaryotic organisms to be a key player in developmental, cell growth and differentiation processes (Hornstein et al., 2005). It was therefore predicted early on that misregulation of this highly conserved class of genes would likely play a role in cancer development. This hypothesis has been supported by the demonstration of altered miRNA expression in a variety of hematological as well as solid tumor entities (for review see (Esquela-Kerscher and Slack, 2006; Sevignani et al., 2006)). Furthermore, transgenic expression of a single miRNA, mouse miR-155, results in early onset of pre-leukemic proliferation followed by high-grade B-cell malignancy (Costinean et al., 2006). Clearly, the discovery of misregulated miRNAs in a given tumor type not only broadens our biological understanding of the disease, but more importantly opens new avenues for the development of new prognostic and diagnostic strategies. Owing to their high stability even in poorly preserved specimen, miRNAs are expected to be robust clinical analytes, valuable for clinical research and biomarker discovery.

By virtue of their apparent role in tumorigenesis, miRNAs may also have therapeutic potential. A recent publication has shown that anti-miRNA oligonucleotides conjugated with cholesterol, also called antagomirs, are able to inhibit miRNA function in vivo in a mouse model system, indicating that therapeutic adjustment of miRNA overexpression in a given cancer type may become feasible (Krutzfeldt et al., 2005). Similarly, it should be possible to correct miRNAs downregulated in tumors using miRNA expression strategies such as viral delivery systems (Felli et al., 2005).

To date a number of studies analysing miRNA expression profiles across multiple human organs as well as comparing normal vs cancerous tissues have been published. These data provided evidence that miRNA expression profiling can be used to classify human cancers and that there are miRNAs with organ-restricted expression patterns (Liu et al., 2004; Calin et al., 2005; Ciafre et al., 2005; Lu et al., 2005; Shingara et al., 2005; He et al., 2005b; Volinia et al., 2006; Weber et al., 2006). Four studies included also pancreatic tissue, but fall short in sample sizes and/or a systematic comparison of miRNA expression profiles between normal and PDAC tissues precluding the identification of PDAC-related miRNAs (Baskerville and Bartel, 2005; Lu et al., 2005; Sood et al., 2006; Volinia et al., 2006). Only miR-216 was identified in these studies for being pancreas specific (Baskerville and Bartel, 2005; Shingara et al., 2005; Sood et al., 2006). In addition, miR-375 was recently identified as a pancreatic islet-specific miRNA regulating insulin secretion (Poy et al., 2004).

In this study, we establish the first comprehensive miRNA expression profiles in tissues from normal pancreas, chronic pancreatitis and PDACs aiming at identifying miRNA candidates with a potential for future clinical application. In addition, miRNA profiles from primary tissues were complemented by similar analyses in a series of PDAC cell lines, facilitating the future development of model systems to test miRNAs identified in pancreatic tissues. Comprehensive analyses of published genes deregulated in PDACs were applied to select potential target genes for a subgroup of 26 PDAC-related miRNAs identified in our study, helping unravel biological signaling pathways potentially targeted by these miRNAs.

Results

miRNA expression profiling in normal and diseased pancreatic samples

miRNA expression data were collected using microarrays carrying 377 individual miRNA probes, including 281 human miRNAs from the miRBase database (http://microrna.sanger.ac.uk), 33 novel human miRNAs and 63 mouse or rat miRNAs from miRBase. Four different sources of RNA were analysed: PDAC cell lines (CL, n=6), normal pancreas (N, n=5), chronic pancreatitis (Ch, n=6) and PDAC (Ca, n=8) tissues. The latter two diseased tissues were macrodissected to trim off to near completion areas of normal pancreatic tissue. For each tissue specimen included in our study, a representative serial section was reviewed by a pathologist before macrodissection (Supplementary Table 1), and the quality of the resulting purified total RNA was assessed (Supplementary Figure 1).

Following array processing and normalization of the raw array data by the variance stabilization (VSN) method, the expressed miRNomes of the 25 different samples were established (Figure 1 and Supplementary Tables 2 and 3). On average, 200 miRNAs were detected above background signal in the tissue samples and 140 in the cell lines, corresponding to 54 and 38% of the miRNA probes present on the microarray, respectively. Unsupervised clustering of samples and miRNA expression levels showed a clear segregation between the four sample types (Figure 1), indicating that miRNA expression profiles were highly reproducible within each sample type.

Figure 1
figure1

miRNA expression profiles classify pancreatic samples. Unsupervised hierarchical clustering of five normal pancreata (Normal), six chronic pancreatitis samples (Chronic), eight PDAC samples (Cancer) and six primary ductal adenocarcinoma cell lines (Cell Line). The normalized expression level (h) of each miRNA is color coded. Calculated average number and percentage of miRNAs detected per sample type are indicated at the bottom.

The pancreatic miRNome

As no comprehensive data set describing miRNA expression in pancreas is currently available, we first sought to characterize the normal pancreatic miRNome. Approximately, 190 miRNAs were found to be reproducibly detected above background signal in all five normal pancreas samples (Supplementary Table 2, Mean (N)>2). These included 158 well-characterized human miRNAs and 21 miRNAs previously identified in mouse or rat. In addition, 2 novel miRNAs (Ambi-miR-7029 and -7058; see sequences in Supplementary Information) were detected at very significant expression levels (Mean (N)>4).

We next performed a global comparison of expression data between the five normal pancreatic samples and a reference set consisting of 33 different human tissues analysed on the same array platform (JJ, AES and EL, unpublished data). The data are summarized in a global graphical representation of mean expression levels and standard deviations within each sample set (Figure 2a). Most miRNAs had similar mean expression levels in both sample sets, and many miRNAs known to be highly expressed in all tissue types, such as hsa-miR-16, -21, -24, -26 and the let-7 family members, were also very abundant in pancreas (Figure 2a, miRNAs with high expression levels and small standard deviations in both sample sets). However, several miRNAs such as hsa-miR-141, -148a, -200a, -200b, -200c, -216, -217 and -375 were clearly enriched in pancreas, whereas others were present at lower levels in our normal pancreatic tissue set (hsa-miR-133a, -143, -145 and -150). Notably, miR-133a was detected at significant expression levels in all of the tissues in our reference set but not in any of the five normal pancreatic tissues. Furthermore, miR-216 and -217 were found to be essentially specific to pancreas as shown by the small mean expression level and standard deviation within the reference set in Figure 2a. The only other tissue where both miRNAs were detected at significant levels was duodenum, but at an expression level 15–25 times lower than in pancreas (data not shown).

Figure 2
figure2

Characterization of the expressed miRNome in normal and PDAC tissues. (a) After global normalization of the raw array data from the 38 samples using the VSN method, the average miRNA expression in the five normal pancreatic tissues (N) was compared against the average miRNA expression in the 33 tissue reference set (Ref). Information on the distribution of expression patterns is embedded in the graph with each miRNA spot size and color coded using the standard deviation of expression values (STDEV) within the reference and normal pancreas sets, respectively. (b) Average miRNA expression in the five normal pancreatic tissues (N) was compared against the average miRNA expression in the eight PDAC samples (Ca) in the context of the 33 tissue reference set. Array data for the 46 samples were globally normalized with VSN. Each miRNA spot is size coded by the difference in average expression between the normal pancreas and 33 tissue reference set (Δh (N vs Ref)), and color coded by the t-test P-value comparing the normal and cancer pancreatic samples (P-value (N vs Ca)).

Identification of miRNAs differentially expressed in PDAC

To uncover potential miRNAs relevant to pancreatic carcinogenesis, we normalized miRNA array data from eight PDAC samples together with the normal pancreas set (Supplementary Table 2). A direct comparison of mean miRNA expression levels showed that miRNA expression is profoundly affected in PDAC (Figure 2b). Interestingly, most of the miRNAs downregulated in PDAC are strongly enriched in pancreas relatively to the 33 human tissues reference set (Figure 2b, large circles corresponding to the variation of mean expression level between normal and reference sets from Figure 2a). Among these miRNAs, the highly expressed, pancreas-enriched miRNAs miR-216 and -217 were downregulated more than 200-fold in PDAC samples, to levels barely detectable on the array.

Further data analysis allowed identification of 84 miRNAs with significantly differential expression between normal pancreas and PDAC (Supplementary Table 4, Flag (N vs Ca)=1). Among these miRNAs, 41 were downregulated and 32 were upregulated by at least twofold in PDAC (Δh(N-Ca)>0.6). Together with miR-216 and -217, a total of 11 miRNAs were strongly downregulated by more than fivefold in PDAC (Δh(N-Ca)>1.6; hsa-miR-29c, -30a-3p, -96, -130b, -141, -148a, -148b, -216, -217, -375 and -494) and 11 others were strongly enriched in PDAC samples (Δh(N-Ca)<-1.6; hsa-miR-31, -143, -145, -146a, -150, -155, -196a, -196b, -210, -222 and -223).

Comparison with chronic pancreatitis samples

As many of the miRNAs expression changes identified above may be in fact related to non-neoplastic processes, we next added to our analysis miRNA expression data from a control set consisting of six chronic pancreatitis samples. Hierarchical clustering analysis (Figure 1) showed that miRNA expression profiles in chronic pancreatitis are distinct from the normal and cancer profiles, and overall more similar to normal pancreas than PDAC. Pair-wise comparison of Pearson correlation values, calculated using mean normalized data for the 377 individual miRNAs, confirmed that global miRNA expression levels in pancreatitis tissues are intermediate between the normal and PDAC tissues (Table 1). Consistent with this observation, 45 out of the 52 miRNAs differentially expressed between chronic pancreatitis and cancer samples (Supplementary Table 4, Flag (Ch vs Ca)=1) were also deregulated in cancer respectively to normal tissues (Figure 3a, top). The expression levels in chronic pancreatitis of all of these miRNAs were intermediate between levels in the normal and PDAC tissues.

Table 1 Paired Pearson correlation values calculated using mean normalized miRNA expression levels
Figure 3
figure3

miRNAs differentially expressed in normal (N), chronic pancreatitis (Ch) and PDAC samples (Ca). (a) Venn diagrams illustrating the relationships between sets of differentially expressed miRNAs. Circles include the total number of differentially expressed miRNAs in the direct pairwise comparison indicated. Intersection areas correspond to the number of differentially expressed miRNAs shared between each comparison. (b) PCA on the 94 differentially expressed miRNAs from Supplementary Table 4. (c) Top 20 miRNAs differentially expressed with Δh>1.6 and P-value<0.0001 between at least two sample types. The graph shows mean normalized expression values and standard deviations for the 20 indicated miRNAs in the three sample types.

A total of 94 miRNAs were found differentially expressed at significant level between any two of the three tissue types (Supplementary Table 4). Among the 68 miRNAs differentially expressed in chronic pancreatitis vs normal tissues (Supplementary Table 4, Flag (Ch vs N)=1), 61 were also misregulated in cancer vs normal (Figure 3a, bottom). These miRNAs whose expression is affected in both diseased tissues are more likely to reflect the desmoplastic reaction of the tumor rather than changes specific to PDAC. Nonetheless, principal component analysis (PCA) using the 94 miRNAs identified above showed a perfect segregation of the PDAC samples away from the normal and chronic groups showing that miRNA expression classifies pancreatic tissues (Figure 3b).

miRNA biomarker candidates for pancreatic diseases

In an effort to further identify the best markers for PDAC, we selected the miRNAs most differentially expressed between at least two sample types using stringent parameters:Δh>1.6 (fivefold) and P-value<0.0001 (Figure 3c). As previously, a clear discrimination between the tissue types could be achieved using this subset of 20 miRNAs. For example, expression of miR-29c, -96, -143, -145, -148b and -150 was affected in both chronic pancreatitis and cancer samples whereas miR-196a, -196b, -203, -210, -222, -216, -217 and -375 were misregulated only in PDAC samples. These data indicate that expression levels of only a few miRNAs can be used to classify normal, chronic pancreatitis and cancerous tissues, and discriminate between neoplastic and non-neoplastic processes in PDAC.

miRNA expression in pancreatic carcinoma cell lines

Pancreatic carcinoma cell lines are currently the best available cell model systems for in vitro analyses of miRNA function. Therefore, the same array profiling strategy was deployed to characterize miRNA expression in six PDAC cell lines. To compare miRNA expression profiles in cell lines and primary tissues, we normalized miRNA array data from the cell lines together with the 19 tissue set (Supplementary Table 3). A total of 140 miRNAs were detected in cell lines, and each of them was also expressed in pancreatic tissues (Supplementary Table 3). Hierarchical clustering analysis on the global miRNA population as well as clustering and principal component analyses on the differentially expressed miRNAs showed a clear segregation of the cell line samples away from the primary tissues (Figure 1 and Supplementary Figure 2a and b). This divergence stemmed mainly from the lack of detectable expression of 60 miRNAs in cell line samples relative to tissue samples, including seven candidates from our list of top 20 differentially expressed miRNAs in PDAC (miR-143, -145, -150, -216, -217, -223 and -375; Supplementary Figure 3).

Overall, the mean miRNA expression levels in cell lines correlated better with PDAC than with normal or chronic pancreatitis tissue (Pearson correlation values of 0.84, 0.78 and 0.79, respectively; see Table 1). Furthermore, 12 out of the top 20 miRNAs differentially expressed in PDAC had expression levels in cell lines similar to PDAC (Supplementary Figure 3). Thus, miRNA expression profiles in cell lines more closely resemble those from primary tumors than those from normal pancreatic tissues.

Identification of miRNA biomarker candidates overexpressed in PDAC and cell lines

Consistent with the differences described above, 87 miRNAs were identified as differentially expressed between PDAC and cell lines, and 108 between normal pancreas and cell lines (Supplementary Table 5, Flag (Ca vs CL)=1 or Flag (N vs CL)=1). Strikingly, one miRNA, miR-205, was highly overexpressed by more than 600-fold in cell lines vs normal tissue, and was also upregulated in five out of eight PDAC tissues relative to chronic or normal tissues. This observation prompted us to look for other miRNAs highly expressed in cell lines and upregulated in cancer that could have escaped our initial stringent screening because of a high P-value or low Δh. By this approach, we further identified five miRNAs: miR-18a, -31, -93, -221 and -224 (Figure 5). These six miRNAs are overexpressed in both primary neoplastic ductal cells and in PDAC cell lines and therefore represent potential additional biomarkers for PDAC.

Figure 5
figure5

Comparison between array and qRT-PCR data. Real time RT-PCR were performed using 25 ng of total RNA input from the 19 tissue samples previously profiled plus two normal (N), two PDAC (Ca) and one chronic pancreatitis (Ch) samples. miRNA expression data obtained with primer sets specific for the indicated miRNAs were normalized to 5S rRNA expression level for each sample (miRNA Ct – 5S Ct). The graphs show the individual normalized miRNA expression levels determined by array (19 samples) or qRT-PCR (24 samples), and associated P-values.

Target prediction for the top 26 miRNA candidates

The identity, associated array data and chromosomal location of the top 26 miRNA candidates identified in our study are summarized in Table 2. To gain further insights into the biological pathways potentially regulated by miRNAs during pancreatic carcinogenesis, we next performed a comprehensive comparison between the published genes known to be misregulated in PDAC and the predicted target genes for our top 26 miRNA candidates according to PicTar (http://pictar.bio.nyu.edu). Relying on the current view that miRNAs are negative regulators of gene expression, and therefore on an inverse correlation between miRNA expression and predicted target gene expression, we identified 246 unique genes targeted by at least one miRNA and misregulated in PDAC (Supplementary Table 6). Importantly, about 37% of these genes were predicted to be targeted by multiple miRNAs, with seven genes each targeted by up to five different miRNAs.

Table 2 Top 26 miRNAs identified in this study

Microarray data validation by qRT-PCR

To verify the accuracy of our array data we performed quantitative RT-PCR (qRT-PCR) experiments for five miRNAs showing very distinct expression patterns within the three tissue types (miR-143, -155, -196a, -217 and -223) and one miRNA with no significant variation (miR-16). We analysed the 19 total RNA samples previously profiled as well as an independent set of tissues consisting of two normal pancreas, two PDAC and one chronic pancreatitis samples that had a greater extent of RNA degradation (N6, N7, Ca9, Ca10 and Ch7; Supplementary Figure 1 and Supplementary Table 1). The relative variations of miRNA expression levels were similar for the normalized array and qRT-PCR data (Figure 5). Moreover, all of the samples had the expected miRNA expression patterns characteristic of their corresponding disease state (normal, cancer or chronic pancreatitis), thus validating our array data and further illustrating the stability of mature miRNA molecules.

Pancreatic tissue classification by qRT-PCR

Analysis of global miRNA expression profiles (Figure 1), differentially expressed miRNAs (Figure 3b and Supplementary Figures 2a and b), and the top 26 miRNA candidate markers (Figures 3c, 4 and 5) indicates that miRNAs expression can classify normal and diseased pancreatic tissues. To further demonstrate the utility of miRNAs as diagnostic analytes, we searched for the minimal set of miRNAs that could discriminate between neoplastic and non-neoplastic tissues using qRT-PCR analysis of mature miRNAs. We found that the difference between raw Ct values of two miRNAs (miR-196 and -217) provided a simple index to identify diseased tissues independently of the total RNA sample input (Figure 6a). The same analysis performed on the 24 independent tissue samples showed a perfect segregation between normal, cancer and chronic pancreatitis samples with a P-value of 1.77E-13 (Figure 6b).

Figure 4
figure4

miRNAs overexpressed in PDAC and cell line samples. The graphs show the individual miRNA normalized expression levels and associated P-values in five normal (N), six chronic pancreatitis (Ch), eight PDAC (Ca) and six cell lines (CL) samples.

Figure 6
figure6

miR-196a and -217 expression classifies normal and diseased pancreatic tissues. (a) Real-time RT-PCR were performed with primer sets specific miR-196a and -217 using the indicated total RNA input from one normal (N5), one PDAC (Ca3) or one chronic pancreatitis (Ch1) sample. Raw Ct values were directly used to calculate the ratio of miR-196a to miR-217 expression, that is, miR-196a Ct – miR217 Ct in the logarithmic space. (b) Same experiment as in (a) with the indicated 24 individual tissue samples and 25 ng of total RNA input.

Discussion

With the recent discovery of miRNA, a new class of gene regulators has been identified, adding another layer of complexity to our understanding of gene expression control. Physiologic expression of miRNAs affects a variety of cellular processes including cell growth, differentiation and apoptosis. Importantly, miRNA mutation and/or misexpression might turn miRNAs into ‘oncomirs’ supporting tumor growth. Therefore, the knowledge of miRNomes expressed under normal conditions and conditions of neoplastic growth for any given tissue is not only an important prerequisite to a comprehensive understanding of cancer cell biology, but it also offers the chance to identify new targets that can be explored both for diagnostic and therapeutic purposes.

In the present work, we generated the first detailed description of the expressed miRNome of normal and diseased pancreatic tissues including miRNA expression data on 377 known and novel miRNAs. Comparing the normal pancreatic miRNome with a miRNome of a reference tissue set containing 33 different human tissues we identified several miRNAs with expression patterns distinctive for pancreas, suggesting a possible role of these miRNAs in pancreas differentiation.

Among these pancreas-specific miRNAs were human miR-216 and -217 that were only expressed in pancreas and, to a lesser extend, in duodenal tissue, thus confirming previously published studies (Baskerville and Bartel, 2005; Shingara et al., 2005; Sood et al., 2006). In addition, we found a very low to absent expression of miR-216 and -217 in pancreatic carcinoma tissues and cell lines. As normal pancreas consists of about 90% acinar cells, this observation suggests that these two miRNAs are primarily expressed in acinar cells. Therefore, it can be speculated that these miRNAs could play an important role in maintaining the acinar differentiation status.

Similarly, miR-375 expression was high in normal pancreas but was significantly lower in both diseased tissues and absent in cell lines. miR-375 has been previously described to be expressed in mouse pancreatic islet cells suppressing glucose-induced insulin secretion (Poy et al., 2004). Thus, it is likely, that the lower content of miR-375 in chronic pancreatitis and PDAC tissues reflects the reduced number of islet cells present in these tissues.

Next, we compared the expressed miRNomes of normal and diseased pancreatic tissues aiming at the identification of miRNAs potentially involved in pancreatic carcinogenesis. Our microarray profiling experiments combined with statistical analyses allowed us to identify the miRNAs differentially expressed between normal and PDAC or cell line samples (Supplementary Table 5, Flag (N vs Ca)=1 or Flag (N vs CL)=1). From this list, we further selected 26 miRNAs (Table 2 and Figure 3c), including six miRNAs overexpressed in neoplastic ductal cells both in vivo and ex vivo (Figure 4), and representing novel potential candidates for biomarker and therapeutic target selection for PDAC. Of note, four of these 26 miRNAs (miR-155, -146, -221 and -223) have previously been identified as being upregulated in different epithelial tumors including breast, endocrine pancreatic, prostate, stomach and colon carcinomas (Iorio et al., 2005; Volinia et al., 2006), suggesting that these miRNAs may play a more general role in carcinogenesis.

PDACs are typically a mixture of mostly neoplastic and stromal cells owing to the inherent strong desmoplastic reaction accompanying cancer development. Therefore, without microdissection it is difficult to specifically analyse the neoplastic epithelial cells in primary tissue specimens. Microdissection in combination with mRNA amplification methods have successfully been used to establish the transcriptome of various pancreatic cell types using serial analysis of gene expression and microarrays (Heidenblut et al., 2004; Buchholz et al., 2005). Unfortunately, a similar approach is currently not practicable for miRNA owing to the lack of robust and accurate global miRNA amplification methods. This restriction prompted us to include tissues from patients with chronic pancreatitis and pancreatic cancer cell lines as a mean to control for miRNA expression likely originating from the ductal or stromal cell compartment.

From our top 26 miRNA candidate list, overexpression of miR-143, -145 and -150 was observed in both chronic pancreatitis and cancer samples but their expression was not detectable in cancer cell lines. This result suggests that altered expression of these miRNAs is more likely a consequence of the desmoplastic/inflammatory reaction of the tumor than epithelial cell dedifferentiation. An alternative explanation would be that loss of expression of these miRNAs is the consequence of culturing epithelial ductal tumor cells. In contrast, miR-93, -196a, -196b, -203, -205, -210, -221, -222 and -224 may be directly related to neoplastic processes in the epithelial cells as they were found upregulated only in PDAC samples. Among these deregulated miRNAs the mode of expression of miR-196a and -196b is especially interesting. These miRNAs are completely absent in normal and chronic pancreatitis tissues but significantly upregulated in PDAC and they are also expressed in cell lines. This pattern is highly suggestive that miR-196 overexpression is specific to PDAC and thus may represent an excellent biomarker and target for therapeutic intervention.

Currently, a major drawback for functional studies of miRNAs is the difficulty in determining the specific target genes regulated by a given miRNA at the transcriptional or translational level. Available prediction algorithms frequently predict over 200 target genes for any single miRNA, and it is likely that this high number of genes contain a significant fraction of false-positive genes. Therefore, it was not unexpected that the top 26 miRNAs identified in our study are predicted by PicTar to potentially be able to target 4207 individual genes. In order to reduce this excessively high number of target genes and enrich for target genes with a potential relevance in pancreatic tumor biology, we performed a comparison of currently published genes known to be differentially regulated in PDAC with the list of predicted target genes for the top 26 miRNAs (Supplementary Table 6). This approach reduced the list of target genes to 246 known genes. More than a third of these 246 candidate target genes were also predicted targets for multiple miRNAs from our top 26 candidate list, with seven genes targeted by up to five different miRNAs. Furthermore, among the 402 individual target predictions included in Supplementary Table 6, 273 predictions reached an intermediate to high PicTar score (>3), and of these, 31 target gene predictions had a high PicTar score (>10). Clearly, although the PicTar scores above 3 should markedly improve the signal-to-noise ratio of the target prediction, it remains difficult to estimate the true false-positive rate of current target prediction algorithms. Therefore, it will be an important next step to experimentally validate a representative number of candidate target genes identified by our target enrichment strategy. Nevertheless, our target gene list combined with published data can be used to shape some initial hypotheses on how alteration of miRNA expression may be directly involved in PDAC development.

Among the pancreatic cancer related genes, we identified beta-catenin as a potential target gene for miR-217. The importance of altered beta-catenin signalling has been demonstrated not only for PDAC but also for pancreas development (Al-Aynati et al., 2004; Dessimoz et al., 2005; Heiser et al., 2006; Zeng et al., 2006). Forced expression of beta-catenin in mice led to an increase in pancreatic organ size whereas loss of beta-catenin signaling impaired the growth of the exocrine pancreas. These data support a model where controlled expression of miR-217, which is likely acinar specific according to our data, may be an additional mechanism for acinar cells to balance their beta-catenin expression level.

Only recently it was shown that nuclear factor kappaB (NF-κB) is able to induce miR-146 expression (Taganov et al., 2006). In our study, we identified this miRNA to be overexpressed sixfold in PDAC compared to normal pancreatic tissue. The importance of miR-146 is further underscored by its upregulation evidenced in a number of epithelial tumors (Volinia et al., 2006). Interestingly, activated NF-κB signalling has been demonstrated in PDAC and is likely the result of an activated phosphatidylinositol 3-OH kinase (PI3K)/AKT signalling pathway (Altomare et al., 2003; Schlieman et al., 2003). It will be important to identify and test miR-146 downstream target genes, such as NUMB, SP8 or KLF7 (Supplementary Table 6), to gain insights into the signalling pathways altered by the aberrant expression of this miRNA.

We also identified cyclin-dependent kinase inhibitor 1B (p27, Kip1) and 1C (p57, Kip 2) as targets of miR-221 and -222 (Supplementary Table 6). These findings open the possibility that upregulation of specific miRNAs may be involved in the well-documented loss of expression of cyclin-dependent kinase inhibitors in PDAC which was also linked to a poor prognosis of this disease (Lu et al., 1999; Hu et al., 2000). Interestingly, overexpression of miR-221 and -222, together with miR-146, was also reported in papillary thyroid cancer and was further correlated with loss of Kit expression (He et al., 2005a).

Lastly, miR-196a and -196b, with a 14-fold overexpression in PDAC compared to normal tissues, were shown to regulate HOXB8 expression involved in limb development (Yekta et al., 2004; Hornstein et al., 2005). A role for HOXB8 in pancreatic carcinogenesis has not been demonstrated so far, however, our data analysis suggests that HOXA1 could be a promising target gene to be evaluated (Supplementary Table 6). Additional homeobox genes predicted to be targeted by miR-196 and previously shown to be implicated in PDAC development are HOXA5 and B6 (Prasad et al., 2005; Segara et al., 2005). Therefore, it will be important to test the effect of miR-196 knockdown on any of these target genes. The miRNA expression data of pancreatic cancer cell lines collected in our study should expedite the selection process of appropriate cell culture models to test these candidates as well as for other functional studies using gain- or loss-of-function strategies.

In an effort to devise a simple assay to discriminate between normal and PDAC tissues, we found that the difference between the raw Ct values of miR-196 and -217 provided a simple index to identify diseased tissues independently of the total RNA sample input (Figure 6a and b). This result suggests that a direct measure of few miRNA biomarkers by qRT-PCR may prove useful for pancreatic tissue classification and potentially other applications such as monitoring of cancer and chronic pancreatitis progression or PDAC aggressiveness and prognosis. Furthermore, the small size and high stability of miRNAs will enable retrospective molecular analysis of archived clinical samples and formalin-fixed, paraffin-embedded (FFPE) tissues. Pilot studies using endoscopic ultrasound guided fine needle aspiration and FFPE specimen will help evaluate the diagnostic value of the miRNA biomarker candidates identified in this study.

Materials and methods

Sample collection and RNA isolation

Surgical pancreatic resection specimens were immediately placed on ice, and subsequently snap-frozen and stored at −80°C. The tissue collection was performed according to a protocol approved by the ethics committee of the University Hospital Schleswig-Holstein, (permission no. 110/99). To prevent RNA degradation by endogenous ribonucleases present in the normal tissues with a high content of acinar cells, normal pancreatic specimen were saturated with RNAlater (Ambion, Austin, TX, USA) according to the manufacturer's instructions before long-term storage. Tissue from normal pancreas (n=7), chronic pancreatitis (n=7) and PDAC (n=10) were included in our study. RNA isolation from cryostat sections and tumor cell lines was performed using the mirVana miRNA Isolation Kit (Ambion) according to the manufacturer's instructions. Integrity of the isolated RNA was analysed using a standard 1% formaldehyde agarose gel (Supplementary Figure 1) and confirmed using the Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA). Purified total RNA was quantified using Nanodrop ND1000 (NanoDrop Technologies, Wilmington, DE, USA). Details about the cell lines and the human reference set containing 33 distinct tissues are given in the Supplementary Information.

miRNA expression analyses

miRNA expression profiling was performed essentially as described previously (Shingara et al., 2005) except that the miRNA fractions recovered from 10–15 μg total RNA were labeled with Cy5 fluorescent dye (GE Healthcare Life Sciences, Piscataway, NJ, USA) and hybridized to mirVana miRNA Bioarrays (Ambion) according to the manufacturer's instructions.

qRT-PCR reactions were performed using SuperTaq Polymerase (Ambion) and the mirVana qRT-PCR miRNA Detection Kit and Primer Sets (Ambion) following the manufacturer's instructions. qRT-PCR were performed with 5–50 ng of total RNA input on an ABI7500 thermocycler (Applied Biosystems, Foster City, CA, USA).

Array data processing

The raw array data were normalized with the VSN method (Huber et al., 2002). In short, VSN is a global normalization process that stabilizes the variance evenly across the entire range of expression. It involves calibration of signal followed by a transformation to a generalized natural logarithmic space in lieu of the traditional logarithm base 2 transformation. Absolute and differences in VSN transformed expression are denoted by h and Δh, respectively, and were used for all subsequent data analyses. Differences in normalized expression values between samples (Δh) can be transformed to a generalized fold change via exponentiation base e. These values will exhibit a compression for small differences in expression. For an overview of miRNA array processing and analysis, see Davison et al. (2006).

For each array, the minimum observable threshold was determined by examining the foreground minus background median intensities for ‘EMPTY’ spots. The minimum threshold was defined as the 5% symmetric trimmed mean plus 2 standard deviations across all ‘EMPTY’ spots on individual array.

Data analyses

A one-way analysis of variance (ANOVA) model was used to test the hypothesis that there was no difference in expression between groups for each miRNA on the array. Pair-wise comparisons for differentially expressed genes identified by ANOVA were then performed to measure relative differences. For each pair of sample types (normal, cancer, chronic and cell line), a two-sample t-test assuming equal variance was carried out for every miRNA and multiplicity correction (Benjamini and Hochberg, 1995) was included to control the false discovery rate at 0.05%. Equivalent t-tests were performed on the Ct values for individual qRT-PCR reaction without correction for multiple testing.

Two-dimensional unsupervised hierarchical clustering using average linkage and the correlation distance metric was performed on all the miRNA normalized expression values (Figure 1) or on those miRNAs determined to be differentially expressed by one-way ANOVA (Supplementary Figure 2a).

PCA for differentially expressed miRNAs identified by one-way ANOVA was performed using covariance for the dispersion matrix on the mean-centered data set (Figure 3b and Supplementary Figure 2b). This analysis illustrates the level of spread between the samples and experimental groups.

Details on the identification of miRNA predicted target genes deregulated in PDAC (Supplementary Table 6) are given in the Supplementary Information.

References

  1. Al-Aynati MM, Radulovich N, Riddell RH, Tsao MS . (2004). Epithelial-cadherin and beta-catenin expression changes in pancreatic intraepithelial neoplasia. Clin Cancer Res 10: 1235–1240.

    CAS  Article  Google Scholar 

  2. Altomare DA, Tanno S, De Rienzo A, Klein-Szanto AJ, Skele KL, Hoffman JP et al. (2003). Frequent activation of AKT2 kinase in human pancreatic carcinomas. J Cell Biochem 88: 470–476.

    CAS  Google Scholar 

  3. Baskerville S, Bartel DP . (2005). Microarray profiling of microRNAs reveals frequent coexpression with neighboring miRNAs and host genes. RNA 11: 241–247.

    CAS  Article  Google Scholar 

  4. Benjamini Y, Hochberg Y . (1995). Controlling the false discovery rate: a practical and powerful approach to multiple testing. J Roy Stat Soc 57: 289–300.

    Google Scholar 

  5. Buchholz M, Braun M, Heidenblut A, Kestler HA, Kloppel G, Schmiegel W et al. (2005). Transcriptome analysis of microdissected pancreatic intraepithelial neoplastic lesions. Oncogene 24: 6626–6636.

    CAS  Article  Google Scholar 

  6. Calin GA, Ferracin M, Cimmino A, Di Leva G, Shimizu M, Wojcik SE et al. (2005). A microRNA signature associated with prognosis and progression in chronic lymphocytic leukemia. N Engl J Med 353: 1793–1801.

    CAS  Article  Google Scholar 

  7. Ciafre SA, Galardi S, Mangiola A, Ferracin M, Liu CG, Sabatino G et al. (2005). Extensive modulation of a set of microRNAs in primary glioblastoma. Biochem Biophys Res Commun 334: 1351–1358.

    CAS  Article  Google Scholar 

  8. Costinean S, Zanesi N, Pekarsky Y, Tili E, Volinia S, Heerema N et al. (2006). Pre-B cell proliferation and lymphoblastic leukemia/high-grade lymp. Proc Natl Acad Sci USA 103: 7024–7029.

    CAS  Article  Google Scholar 

  9. Davison TS, Johnson CD, Andruss BF . (2006). Analyzing micro-RNA expression using microarrays. Methods Enzymol 411: 14–34.

    CAS  Article  Google Scholar 

  10. Dessimoz J, Bonnard C, Huelsken J, Grapin-Botton A . (2005). Pancreas-specific deletion of beta-catenin reveals Wnt-dependent and Wnt-independent functions during development. Curr Biol 15: 1677–1683.

    CAS  Article  Google Scholar 

  11. Esquela-Kerscher A, Slack FJ . (2006). Oncomirs – microRNAs with a role in cancer. Nat Rev Cancer 6: 259–269.

    CAS  Article  Google Scholar 

  12. Felli N, Fontana L, Pelosi E, Botta R, Bonci D, Facchiano F et al. (2005). MicroRNAs 221 and 222 inhibit normal erythropoiesis and erythroleukemic cell growth via kit receptor down-modulation. Proc Natl Acad Sci USA 102: 18081–18086.

    CAS  Article  Google Scholar 

  13. He H, Jazdzewski K, Li W, Liyanarachchi S, Nagy R, Volinia S et al. (2005a). The role of microRNA genes in papillary thyroid carcinoma. Proc Natl Acad Sci USA 102: 19075–19080.

    CAS  Article  Google Scholar 

  14. He L, Thomson JM, Hemann MT, Hernando-Monge E, Mu D, Goodson S et al. (2005b). A microRNA polycistron as a potential human oncogene. Nature 435: 828–833.

    CAS  Article  Google Scholar 

  15. Heidenblut AM, Luttges J, Buchholz M, Heinitz C, Emmersen J, Nielsen KL et al. (2004). aRNA-longSAGE: a new approach to generate SAGE libraries from microdissected cells. Nucleic Acids Res 32: e131.

    Article  Google Scholar 

  16. Heiser PW, Lau J, Taketo MM, Herrera PL, Hebrok M . (2006). Stabilization of beta-catenin impacts pancreas growth. Development 133: 2023–2032.

    CAS  Article  Google Scholar 

  17. Hornstein E, Mansfield JH, Yekta S, Hu JK-H, Harfe BD, McManus MT et al. (2005). The microRNA miR-196 acts upstream of Hoxb8 and Shh in limb development. Nature 438: 671–674.

    CAS  Article  Google Scholar 

  18. Hu YX, Watanabe H, Li P, Wang Y, Ohtsubo K, Yamaguchi Y et al. (2000). An immunohistochemical analysis of p27 expression in human pancreatic carcinomas. Pancreas 21: 226–230.

    CAS  Article  Google Scholar 

  19. Huber W, von Heydebreck A, Sultmann H, Poustka A, Vingron M . (2002). Variance stabilization applied to microarray data calibration and to the quantification of differential expression. Bioinformatics 18 (Suppl 1): S96–S104.

    Article  Google Scholar 

  20. Iorio MV, Ferracin M, Liu CG, Veronese A, Spizzo R, Sabbioni S et al. (2005). MicroRNA gene expression deregulation in human breast cancer. Cancer Res 65: 7065–7070.

    CAS  Article  Google Scholar 

  21. Krutzfeldt J, Rajewsky N, Braich R, Rajeev KG, Tuschl T, Manoharan M et al. (2005). Silencing of microRNAs in vivo with/‘antagomirs/’. Nature 438: 685–689.

    Article  Google Scholar 

  22. Liu CG, Calin GA, Meloon B, Gamliel N, Sevignani C, Ferracin M et al. (2004). An oligonucleotide microchip for genome-wide microRNA profiling in human and mouse tissues. Proc Natl Acad Sci USA 101: 9740–9744.

    CAS  Article  Google Scholar 

  23. Lu CD, Morita S, Ishibashi T, Hara H, Isozaki H, Tanigawa N . (1999). Loss of p27Kip1 expression independently predicts poor prognosis for patients with resectable pancreatic adenocarcinoma. Cancer 85: 1250–1260.

    CAS  Article  Google Scholar 

  24. Lu J, Getz G, Miska EA, Alvarez-Saavedra E, Lamb J, Peck D et al. (2005). MicroRNA expression profiles classify human cancers. Nature 435: 834–838.

    CAS  Article  Google Scholar 

  25. Poy MN, Eliasson L, Krutzfeldt J, Kuwajima S, Ma X, Macdonald PE et al. (2004). A pancreatic islet-specific microRNA regulates insulin secretion. Nature 432: 226–230.

    CAS  Article  Google Scholar 

  26. Prasad NB, Biankin AV, Fukushima N, Maitra A, Dhara S, Elkahloun AG et al. (2005). Gene expression profiles in pancreatic intraepithelial neoplasia reflect the effects of hedgehog signaling on pancreatic ductal epithelial cells. Cancer Res 65: 1619–1626.

    CAS  Article  Google Scholar 

  27. Schlieman MG, Fahy BN, Ramsamooj R, Beckett L, Bold RJ . (2003). Incidence, mechanism and prognostic value of activated AKT in pancreas cancer. Br J Cancer 89: 2110–2115.

    CAS  Article  Google Scholar 

  28. Segara D, Biankin AV, Kench JG, Langusch CC, Dawson AC, Skalicky DA et al. (2005). Expression of HOXB2, a retinoic acid signaling target in pancreatic cancer and pancreatic intraepithelial neoplasia. Clin Cancer Res 11: 3587–3596.

    CAS  Article  Google Scholar 

  29. Sevignani C, Calin GA, Siracusa LD, Croce CM . (2006). Mammalian microRNAs: a small world for fine-tuning gene expression. Mamm Genome 17: 189–202.

    CAS  Article  Google Scholar 

  30. Shingara J, Keiger K, Shelton J, Laosinchai-Wolf W, Powers P, Conrad R et al. (2005). An optimized isolation and labeling platform for accurate microRNA expression profiling. RNA 11: 1461–1470.

    CAS  Article  Google Scholar 

  31. Sood P, Krek A, Zavolan M, Macino G, Rajewsky N . (2006). Cell-type-specific signatures of microRNAs on target mRNA expression. Proc Natl Acad Sci USA 103: 2746–2751.

    CAS  Article  Google Scholar 

  32. Taganov KD, Boldin MP, Chang K-J, Baltimore D . (2006). NF-{kappa}B-dependent induction of microRNA miR-146, an inhibitor targeted to signaling proteins of innate immune responses. Proc Natl Acad Sci USA 103: 12481–12486.

    CAS  Article  Google Scholar 

  33. Valencia-Sanchez MA, Liu J, Hannon GJ, Parker R . (2006). Control of translation and mRNA degradation by miRNAs and siRNAs. Genes Dev 20: 515–524.

    CAS  Article  Google Scholar 

  34. Volinia S, Calin GA, Liu C-G, Ambs S, Cimmino A, Petrocca F et al. (2006). A microRNA expression signature of human solid tumors defines cancer gene targets. Proc Natl Acad Sci USA 103: 2257–2261.

    CAS  Article  Google Scholar 

  35. Weber F, Teresi RE, Broelsch CE, Frilling A, Eng C . (2006). A limited set of human microRNA is deregulated in follicular thyroid carcinoma. J Clin Endocrinol Metab 91: 3584–3591.

    CAS  Article  Google Scholar 

  36. Yekta S, Shih I-H, Bartel DP . (2004). MicroRNA-directed cleavage of HOXB8 mRNA. Science 304: 594–596.

    CAS  Article  Google Scholar 

  37. Yeo T, Hruban R, Leach S, Wilentz R, Sohn T, Kern S et al. (2002). Pancreatic cancer. Curr Prob Cancer 26: 176–275.

    Article  Google Scholar 

  38. Zeng G, Germinaro M, Micsenyi A, Monga NK, Bell A, Sood A et al. (2006). Aberrant Wnt/beta-catenin signaling in pancreatic adenocarcinoma. Neoplasia 8: 279–289.

    CAS  Article  Google Scholar 

Download references

Acknowledgements

We thank Bernard Andruss for critical review of the manuscript and Matthias Becker for his excellent technical assistance. This work was supported in part by the Deutsche Krebshilfe (BS and SAH, 70-2988).

Author information

Affiliations

Authors

Corresponding author

Correspondence to S A Hahn.

Additional information

Supplementary Information accompanies the paper on the Oncogene website (http://www.nature.com/onc).

Supplementary information

Rights and permissions

Reprints and Permissions

About this article

Cite this article

Szafranska, A., Davison, T., John, J. et al. MicroRNA expression alterations are linked to tumorigenesis and non-neoplastic processes in pancreatic ductal adenocarcinoma. Oncogene 26, 4442–4452 (2007). https://doi.org/10.1038/sj.onc.1210228

Download citation

Keywords

  • pancreatic ductal adenocarcinoma
  • chronic pancreatitis
  • microRNA
  • expression profiling

Further reading

Search

Quick links