Transcriptome analysis of microdissected pancreatic intraepithelial neoplastic lesions


Pancreatic ductal adenocarcinoma (PDAC) carries the most dismal prognosis of all solid tumours. Both the late clinical presentation of patients, due to lack of early symptoms, as well as the rapid and aggressive course of the disease contribute to the extremely high mortality of this malignancy. Recently, a multistep progression model for PDAC integrating morphological, clinical and molecular evidence has been proposed. Putative precursor lesions, termed pancreatic intraepithelial neoplasia (PanIN), are classified into three different grades (PanIN-1 through -3) based on the degree of cellular atypia they display. We have conducted large-scale expression profiling analyses of microdissected cells from normal pancreatic ducts, PanINs of different grades and PDACs using whole-genome oligonucleotide microarrays. Verification of hybridisation results for selected genes was performed using quantitative real-time PCR and immunohistochemical analyses on PanIN tissue microarrays. Comparison of the expression profiles demonstrated that the greatest changes in gene expression occur between PanIN stages 1B and 2, suggesting that PanIN-2 may represent the first truly preneoplastic stage in PDAC progression. Our results identify a large number of potential target genes for the development of novel molecular diagnostic and therapeutic tools for the prevention and early diagnosis of PDAC and provide novel insights into the pathophysiological mechanisms involved in tumour progression in the pancreas.


More than 150 000 deaths per year are estimated to occur due to pancreatic ductal adenocarcinoma (PDAC) worldwide (Parkin et al., 1999). The extremely high mortality rate of PDAC, evidenced by an estimated worldwide 5-year survival rate of 1% (Parkin et al., 1999), is due, among others, to the aggressive behaviour of this tumour, the inadequacy of current therapies and the late clinical presentation. Most patients with PDAC do not develop symptoms, such as jaundice or pain, until the tumour is already in a locally advanced or metastatic stage. Unfortunately, no curative treatment options are available for these patients, and even patients with localized disease where a curative resection is attempted frequently relapse, with mean survival ranging between 8 and 14 months (Conlon et al., 1996; Sperti et al., 1997; Trede et al., 2001). Recent advances in adjuvant therapy have achieved survival rates of 20 months (Neoptolemos et al., 2004), which represents an improvement, though on a low level. On the other hand, curative resection in early-stage PDAC may achieve 5-year survival rates of up 40% (Yeo and Cameron 1998). Thus, early detection of PDAC or detection of preneoplastic lesions allowing curative resection is of paramount importance in order to improve the dismal prognosis of PDAC patients. This is particularly true for patients from high-risk groups, such as smokers or persons with a genetic susceptibility for PDAC. It has been estimated that familial aggregation and genetic susceptibility play a role in 5–10% of patients with PDAC (Lynch et al., 1990, 1996; Lynch, 1994). To date, members of families with genetic syndromes with increased risk for PDAC or with familial PDAC are screened with a combination of imaging and endoscopic techniques such as multislice computed tomography (CT), magnetic resonance imaging (MRI), endosonography (EUS) and enodoscopic retrograde cholangiopancreatography (ERCP) (Brentnall et al., 1999; Hruban et al., 2001c; Canto et al., 2004). However, changes observed in EUS and ERCP are often unspecific and may as well be observed in patients suffering from chronic pancreatitis. Furthermore, the sensitivity of these screening methods for the detection of early neoplastic or precursor lesions remains unclear.

Similar to the adenoma-carcinoma sequence in colon cancer (Fearon and Vogelstein, 1990), a multistep progression model for the development of PDAC has been proposed based on growing morphological, clinical and molecular evidence (Hruban et al., 2000; Luttges et al., 2001; Kloppel and Luttges, 2004). Putative precursor lesions of ductal adenocarcinoma, defined as microscopic papillary or flat noninvasive neoplasms arising in the pancreatic duct, have been termed pancreatic intraepithelial neoplasia (PanIN) (Hruban et al., 2001a, 2004). PanINs are characterised by columnar to cuboidal cells with varying amounts of mucin and increasing degrees of cytological and architectural atypia. PanINs usually involve ducts of less than 5 mm (Hruban et al., 2004). Three PanIN grades are distinguished based on the degree of structural and cellular atypia present in the lesions. PanINs are believed to progress from flat (PanIN-1A) and papillary lesions (PanIN-1B) without dysplasia, to papillary lesions with dyplasia (PanIN-2), to carcinoma in situ (PanIN-3). PanINs, especially the intermediate and higher grade lesions, display a number of genetic abnormalities also observed in invasive cancers, including mutations of the KRAS, CDKN2A/p16INK4A, BRCA2, TP53 and SMAD4 genes (Wilentz et al., 1998, 2000; Luttges et al., 1999, 2001; Goggins et al., 2000; Wilentz et al., 2000; Luttges et al., 2001), which is suggestive of their neoplastic potential (Hruban et al., 2000; Luttges et al., 2001). However, efforts to develop molecular screening approaches based on the detection of some of these molecular alterations have so far been of limited success (see Vimalachandran et al., 2004 for an overview).

Detailed knowledge of molecular changes occurring during PanIN progression is thus urgently required for the development of novel screening strategies or chemopreventive approaches for PDAC. The availability of robust and high-throughput techniques for the analysis of complete transcriptomes of cells and tissues using approaches such as microarry technology or SAGE has made it possible to study the whole spectrum of transcriptional changes involved in the process. At the same time, the introduction of tissue microdissection as well as development of RNA amplification techniques for generating hybridisation probes from as few as several hundred cells offer the chance to perform a global analysis of the transcriptome of microscopic precursor lesions such as PanINs. The aim of the present study was to make use of these novel techniques within the framework of the German Pancreatic Cancer Network (GPCN) funded by the German Cancer Aid foundation (Deutsche Krebshilfe) to perform high-throughput expression analyses of microdissected PanINs using whole-genome oligonucleotide arrays. The results of this study provide a large number of novel potential target genes for the development of molecular diagnostic and therapeutic tools that may be applied to risk populations. Moreover, this study provides insights into interesting pathophysiological mechanisms involved in tumour progression in the pancreas.


Microrray hybridisation and evaluation of expression profiles

For microarray hybridisation, PanIN-1B lesions from 15 patients, PanIN-2 lesions from six patients, PanIN-3 lesions from eight patients and PDAC tissues from eight patients as well as normal ducts and acinar cells from 14 patients were prepared from PDAC tumour specimens and resection margins. Manual microdissection was employed for this purpose, since this method yielded RNA of superior quality and quantity compared to laser capture microdissection (LCM) and laser pressure catapulting (LPC), respectively. A total of 45 microarray hybridisations were performed using pools of at least three individual lesions each. The complete hybridisation data are available as part of the supplementary information at Comparing the expression profiles of PanIN's of different stages or PDAC with that of normal pancreatic ducts, a total of 1251 genes were identified as differentially expressed using the criteria described in the Materials and methods section. Interestingly, the numbers of differentially expressed genes varied greatly across the different stages of preneoplastic and neoplastic lesions. While only 16 genes were upregulated and 31 genes downregulated in PanIN-1B as compared to normal pancreatic ducts (estimated false disovery rate (FDR): 3.2%), these numbers rose sharply to 76 up- and 362 downregulated genes in PanIN-2 (FDR: 1.7%), 418 up- and 160 downregulated genes in PanIN-3 (FDR: 1.4%), and 303 up- and 307 downregulated genes in PDAC (FDR: 1.1%) (Figure 1). There was a prominent overlap, ranging between 30 and 61%, between the genes deregulated in individual PanIN lesions and PDAC.

Figure 1

Number of genes identified as differentially expressed in the various PanIN stages and PDAC as compared to normal pancreatic ducts

Clustering of expression profiles of microdissected cells

Hierarchical cluster analysis was performed using all 1251 genes differentially expressed in at least one of the pairwise comparisons. Expression levels in acini were included in this analysis for control purposes (see below). It was very interesting to note that the expression profiles of the different histological entities were ordered into two main branches of the dendrogram, with PanIN-1B lesions associating closely with normal ducts and acini to form a cluster of benign/hyperplastic tissues, while PanIN-2 and PanIN-3 lesions joined the PDAC samples to form a cluster of dysplastic/neoplastic tissues (Figure 2a).

Figure 2

Hierarchical cluster analysis of differentially expressed genes. Mean expression levels for the different experimental groups are shown in columns, genes are shown in rows. Both the samples (experimental groups) as well as the genes were hierarchically clustered (two-dimensional cluster analysis) using uncentred Pearson correlation as the similarity measure. Red cells indicate high expression, black intermediate expression and green low expression of a gene in the respective group. (a) Complete cluster of all 1251 differentially expressed genes. (b) Subcluster of acinispecific genes. (c) Subcluster of genes downregulated early during carcinogenesis. (d) Subcluster of genes upregulated in advanced PanIN lesions. (e) Subcluster of genes upregulated early during carcinogenesis. Ac=acini, D=ducts, 1B=PanIN-1B, 2=PanIN-2, 3=PanIN-3, Ca=PDAC

Hierachical clustering of genes identified various clusters with distinct expression patterns across the different types of microdissected lesions. The cluster depicted in Figure 2b, which shows that the acini-specific pancreatic enzymes, pancreatic elastases 3A and B, pancreatic lipase (PNLIP) and carboxyl ester lipase (CEL), produced very high hybridisation signals that were strictly confined to the acinar tissue samples, serves to demonstrate the accuracy of the microdissection and RNA amplification procedure.

Figure 2c shows a cluster of genes selectively expressed in ductal cells, thus presumably representing genes whose expression is lost very early in the chain of events leading to tumour formation. These included, among others, the putative cytokine high in normal 1 (SCGB3A1), the amiloride sensitive cation channel 2 (ACCN2), the epithelial transmembrane mucin 13 (MUC13) and the developmental regulator norrie disease (NDP).

Figure 2c shows a cluster of genes upregulated in later stage dysplastic/neoplastic lesions, for example, PanIN-2, PanIN-3 and/or PDAC, thus potentially representing markers of advanced preneoplastic and/or neoplastic lesions. In addition to genes previously implicated in carcinogenesis, for example, serine/threonine kinase 11 (STK11) and fibronectin (FN1), this cluster contained the Ras-GTPase activating protein SH3 domain-binding protein 2 (G3BP2), plastin 3 (PLS3) and the homeobox protein D11 (HOXD11).

A fourth cluster, depicted in Figure 2e, contained a spectrum of genes showing elevated expression levels from PanIN-1B throughout the PDAC samples, thus presumably representing genes upregulated very early in carcinogenesis. In addition to a large number of new potential markers, this cluster included known markers for pancreatic tumours, such as the S100 calcium-binding protein (S100P) (Crnogorac-Jurcevic et al., 2003), the trefoil factors 1 and 2 (TFF1/2)(Terris et al., 2002) and matrix metalloproteinase 1 (MMP1).

Validation of expression data

Quantitative real-time PCR

Quantitative real-time PCR (qRT–PCR) analysis was conducted for 12 of the genes differentially expressed between normal ducts and the PanIN lesions to confirm differential expression in an independent set of individual microdissected lesions (Figure 3). The results of the qRT–PCR analysis were generally in good agreement with the microarray data. Differential expression between normal ducts and PanIN lesions was confirmed for nine of the 12 genes (PCOLN3, PLAC8, RAI3, TSPAN-1, TFF2, PKIA, CBFB, S100P, Pim1). In three instances, the changes in expression obtained by the microarray experiments were not seen or contradicted by the qRT–PCR results (S100A14, CARD10, Hypothetical protein DKFZP434I0714). The failure rate was thus higher than would be expected from the estimated false discovery rates, which may in part be explained by the fact that in contrast to the microarray experiments, where material from different patients was pooled to obtain sufficient quantities, the qRT–PCR analyses were performed on individual samples, thus altering the potential influence of individual samples on the outcome of the analysis.

Figure 3

Quantitative real-time PCR validation of microarray data. Displayed are mean values (log2 of relative expression=ΔΔCT)±s.e.m. from five individual experiments per gene. Colour panels above the bars symbolise log2-transformed mean expression values obtained in the array hybridisations (see reference colour bar). Expression levels in normal duct cells were arbitrarily set to 0 for both the qRT–PCR and the microarray results

Immunohistochemistry of selected genes using PanIN tissue microarrays (TMAs)

In addition to the qRT–PCR validation experiments, the expression of three upregulated genes (RAB1B, CEACAM5, Cathepsin E) was investigated immunohistochemically using PanIN TMAs (Figure 4). As predicted by the microarray results, staining was weak (CTSE) or absent (RAB1B, CEACAM5) in normal ducts. For all three gene products, the number of positive lesions and/or the staining intensities increased across the PanIN stages, with strongest staining intensities observed in PDAC.

Figure 4

Immunohistochemical analysis of RAB1B (a), CTSE (b) and CEACAM5 (c). Displayed are representative sections of normal duct cells (1), PanIN-1B (2), PanIN-2 (3), PanIN-3 (4) and PDAC (5). The table provides a comparison of immunohistochemistry (IHC) and microarray (MA) results. Listed are the numbers of positive staining lesions per total number of lesions analysed. Microarray results are expressed as fold changes relative to expression in normal duct cells. Asterisks denote differential expression (as defined in the Material and methods section) compared to normal duct cells in the microarray analysis. For all three gene products, the number of positive lesions and/or the staining intensities increased with increasing stages of dysplasia

Functional categories of genes differentially expressed in PanIN lesions

Analysis of the functional roles of the genes differentially expressed in preneoplastic and neoplastic lesions may provide novel insights into the underlying biological mechanisms involved in tumour formation and progression in the pancreas. In order to obtain hints at the cell biological implications of the differences detected between PanIN lesions and normal duct cells, it was of major interest to query the data for accumulations of genes with interesting expression patterns in distinct functional categories. To this end, we used the ‘GoMiner’ tool (Zeeberg et al., 2003) to query the Gene Ontology (GO) database ( for functional categories associated with the 1251 differentially expressed genes. Differentially expressed genes were then organised into functional categories such as development, structure, signal transduction, etc. according to their Gene Ontology annotations and each category analysed separately using one-dimensional hierarchical clustering. Clusters of genes with PanIN stage-dependent expression patterns were identified by visual inspection of the clustering results The ‘structure’ and ‘development/differentiation’ categories, respectively, gave particularly interesting results. Within the ‘structure’ category, clusters of benign/hyperplastic tissue-specific genes and dysplastic/neoplastic tissue-specific genes were readily distinguishable (Figure 5a). While the former included the matrix metalloproteinases 3 and 17 (MMP3, MMP17) as well as laminin gamma 3 (LAMC3), the latter encompassed fibronectin 1 (FN1), keratin 16 (KRT16), plastin 3 (PLS3), matrix metalloproteinase 7 (MMP7) and collagen type III alpha-1 (COL3A1). Similar clusters were observed in the ‘development/differentiation’ category: Here, the benign/hyperplastic tissue-specific cluster contained the homeobox gene HB8 (HLXB8), the NUMB homolog (Drosophila) (NUMB), the ephrin receptor A3 (EphA3), and the protocadherins α4 and β13 (PCDHA4, PCDHB13), while the dysplastic/neoplastic tissue-specific cluster included the genes FN1, homeobox D11 (HOXD11), plastin 3 (PLS3) and core binding factor, beta subunit (CFBF).

Figure 5

Clusters of structure (a) and development/differentiation (b) related genes. Genes were assembled in functional categories according to their Gene Ontology annotations and each category analysed separately by one-dimensional hierarchical clustering of genes. Clusters with clear distinctions between benign/hyperplastic tissues (ducts and PanIN-1B) and dysplastic/neoplastic tissues (PanIN-2, PanIN-3 and PDAC) are obvious in both categories

Potential early diagnostic target genes

The most promising candidates to serve as markers for the early detection of PDAC would be genes that are significantly upregulated in preneoplastic and neoplastic lesions from early stages on. Since our results indicate that PanIN-2 is the first truly preneoplastic stage in PDAC progression, we have compiled a list of potential early diagnostic target genes by selecting genes that displayed low expression in acini and normal ducts, which were upregulated (as defined by the criteria described in the Materials and methods section) in PanIN-2 lesions as compared to normal ducts and which retained their high expression levels throughout the progression to PDAC. Application of these criteria to the analysis of the differentially expressed genes resulted in the identification of 22 known genes and eight ESTs (Figure 6). In addition to genes previously not implicated in carcinogenesis, these included a number of candidates that have previously been suggested as markers for pancreatic tumours, for example, S100 calcium-binding protein (S100P) (Crnogorac-Jurcevic et al., 2003) and Trefoil factor 1 (TFF1) (Terris et al., 2002), or which have been implicated in cancerogenesis, for example, interferon alpha-inducible protein 27 (IFI27) (Suomela et al., 2004), cell division cycle 37 homolog (CDC37) (Stepanova et al., 2000) and anterior gradient 2 homolog (AGR2) (Kristiansen et al., 2004).

Figure 6

Potential early diagnostic target genes. Selection of genes upregulated from PanIN-2 (presumably the first truly preneoplastic stage in pancreatic carcinogenesis) throughout PDAC resulted in the identification of 30 candidate target genes. Displayed are mean normalised expression values for the different tissue types. Red indicates high expression, black intermediate expression and green low expression in the respective tissue


The progressive accumulation of morphological changes and the presence of typical mutations strongly argue for a role of PanINs as precursor lesions of PDAC and a progression from normal pancreatic ducts to PDAC via the different PanIN stages (Hruban et al., 2000; Luttges et al., 2001; Kloppel and Luttges, 2004). Our results provide for the first time global expression profiles of highly pure preparations of normal, preneoplastic and neoplastic pancreatic duct cells as well as a comprehensive analysis of the gene expression changes associated with the different PanIN stages. One striking result of this analysis was the observation that in comparison to more advanced PanIN lesions or PDAC, PanIN-1B was associated with very few expression changes (47 differentially expressed genes as compared to 438 genes in PanIN-2, 578 in PanIN-3 and 610 in PDAC). Consequently, PanIN-1B lesions clustered very closely with normal duct cells in the hierarchical cluster analysis. Earlier reports have shown that PanIN-1A and B lesions are frequently observed in non-neoplastic pancreata (Luttges et al., 1999; Hruban et al., 2001b), strongly suggesting that the risk of progression to invasive carcinoma is very low for PanIN-1. Together, these results indicate that PanIN-2 rather than PanIN-1B represents the earliest truly preneoplastic lesion in the pancreas.

It was very interesting to note that while the number of differentially expressed genes steadily increased with more advanced stages of dysplasia, the gene expression changes were not simply additive but showed extensive fluctuations across the different tissue types. In addition to genes whose expression was constantly or progressively changed, a significant proportion of the genes was deregulated in a stage-specific manner, with transcript levels returning to control levels in later PanIN stages and/or PDAC. This was somewhat unexpected, since earlier studies investigating the mutation/deregulation of limited numbers of individual genes in PanIN lesions had suggested an ordered sequence of cumulative mutations associated with the stepwise progression to PDAC (Heinmoller et al., 2000; Hruban et al., 2000; Luttges et al., 2001; Maitra et al., 2003). It can be speculated that the transient expression changes detected in our analysis reflect the activation of counteractive mechanisms in the cells in response to early ‘gatekeeper’ mutations (Vimalachandran et al., 2004), which impair the normal function of the cells. During the progression towards PDAC, many of the control mechanisms that are operative in normal pancreatic duct cells become inactivated, thus resulting in the reversion of some of the gene expression changes seen in earlier PanIN stages.

While transient changes can provide hints at counter-regulatory mechanisms, persistent expression changes are more likely to reveal basic genetic principles underlying malignant transformation in the pancreas. Persistent or progressive changes in transcript levels were in particular detected for structure- and development/differentiation-related genes, respectively. Profound changes in the composition and turnover of extracellular matrix (ECM) components, produced both by tumour cells and surrounding stromal cells, is a hallmark feature of PDAC (Bramhall et al., 1997; Ellenrieder et al., 2000; Buchholz et al., 2003). Our results demonstrate that significant changes in the composition of ECM components, for example, upregulation of MMP7, fibronectin and type 3 collagens as well as downregulation of MMP17, are already detectable in the PanIN-2 stage. While the ECM has traditionally been regarded as an inert scaffold providing structural support for the functional cells within an organ or tissue, it has recently become increasingly evident that ECM components can actively regulate growth, death, adhesion, migration, invasion, gene expression and differentiation in adjacent cells (Liotta and Kohn, 2001; Pupa et al., 2002). This is additionally emphasised by the fact that many of the differentially expressed structural genes also appear in the gene ontology-derived list of development/differentiation-related genes. Active modification of the microenvironment is therefore likely to be an important early event during cancerogenesis in the pancreas.

One of the main goals of this study was the identification of potential new targets for early detection and/or intervention in PDAC. As mentioned above, our results strongly suggest that PanIN-2 represents the first truly preneoplastic stage in the process of cancerogenesis. Ideal targets should therefore be significantly and continuously upregulated from PanIN-2 through PDAC to ensure applicability of potential new procedures independently of the exact type of lesion(s). We have identified a total of 30 genes satisfying these criteria, including genes that have previously been suggested as markers of pancreatic tumours, for example, S100 calcium-binding protein (S100P) (Crnogorac-Jurcevic et al., 2003) and trefoil factor 1 (TFF1) (Terris et al., 2002), or genes that have been suggested as targets for tumour therapy, for example, the interleukin receptor alpha 1 (Kawakami et al., 2001). Of special interest for diagnostic purposes were the many extracellular and cell surface genes contained within this set. A particularly promising candidate may be the secreted proteinase inhibitor cystatin C (CST3), which was strongly upregulated from PanIN-1B on. Cystatin C has previously been shown to be linked to prognosis in breast, lung, colorectal, brain and head and neck cancer (Kos et al., 2000) and serum levels of cystatin C have been used as markers for diagnosis and prognosis in melanoma, colorectal and head and neck cancer (Kos et al., 2000; Strojan et al., 2004). In addition, cystatin C as well as S100P have recently been identified as markers of PDAC in a global proteome analysis of pancreatic juice samples (Gronborg et al., 2004). Other candidates identified in our analysis included the extracellular protease inhibitor 15 (PI15), which has been proposed as a marker for neuroblastoma and glioblastoma (Yamakawa et al., 1998), galectin 4 (LGALS4), which is secreted in a soluble form by many epithelial cancer cells and has been proposed as a marker for breast and liver tumours (Huflejt and Leffler, 2004), and the procollagen (type III) N-endopeptidase (PCOLN3), a matrix metalloproteinase not previously implicated in carcinogenesis.

In summary, the combined use of tissue microdissection, RNA amplification and global microarray analysis has proven to be a powerful tool for profiling the molecular changes associated with the progression of normal pancreatic duct cells to invasive ductal adenocarcinoma. Our results both provide valuable new insights into the biology of PanINs and identify a host of candidate target genes for the development of novel strategies for early detection and treatment of PDAC.

Materials and methods

Microdissection of pancreatic tissues

Tissue was obtained from 51 patients with PDAC in the head of the pancreas. Informed consent was obtained from all patients undergoing surgery and the trial was approved by the ethics committees at the Universities of Kiel and Ulm.

Immediately after surgical resection, pancreas specimens were placed on ice and tissue from the carcinoma, the peritumoural parenchyma and from the resection margin was removed, snap frozen and stored at −80°C. For the identification of acini, normal ducts and the various PanIN lesions, 5 μm thin frozen sections were prepared from tissue blocks from peritumoral pancreatic parenchyma, in particular tissue from resection margins. They were briefly placed in RNAse-free ethanol (Merck, Darmstadt, Germany), stained with H&E and subsequently reviewed. Duct lesions were classified as PanINs according to the criteria of the WHO classification (Hruban et al., 2001a). Tissue blocks that were found to harbour the required tissue components were serially sectioned, the slides were stained using H&E and immediately stored at −20°C. PanIN lesions from the stained serial sections were manually microdissected under microscopic control (BH2, Olympus, Wetzlar, Germany) using a sterile injection needle (size 0.65 × 25 mm, Fa. Braun, Melsungen, Germany). Intermediate ducts were preferably chosen in order to avoid contamination by acinar tissue. Microdissected cells were sampled in a 100 μl reaction tubes containing 50 μl extraction buffer (MWG; Münich, Germany) and placed on ice.

Total RNA was isolated from 1000 to 10 000 microdissected cells using the Pico Pure RNA Isolation Kit (Arcturus, Mountain View, USA), according to the manufacturers instructions.


The Human Genome Oligo-Set-Version 2.0 (Operon, Germany) representing 21 329 genes in the form of optimized 70-mer oligonucleotides was spotted onto GAPSII Slides (Corning, USA) using a OmniGrid Microarrayer (GeneMachines, San Carlos, USA), equipped with Stealth SMP3 Micro Spotting Pins (Telechem, Sunnyvale, CA, USA) at the Chip-Facility of the University of Ulm. Printing concentration of the oligos was 40 μ M in 3 × SSC, 1.5 M Betain. Information about each oligo and its representative gene is available online at

Oligonucleotides were immobilized on the slides by 15 min incubation at 80°C, followed by irradiation with UV light at 254 nm with an energy output of 120 mJ/cm2 in a Stratalinker Model 2400 UV illuminator (Stratagene).

Linear amplification and hybridisation probe generation

In order to obtain sufficient amounts of RNA for hybridisation, 5–50 ng of purified total RNA was linearly amplified under the presence of UTP and 5-(3-aminoallyl)-UTP (each 3.75 mM) using the MessageAMP™ aRNA Kit (Ambion, Woodward, Austin, USA), according to the manufacturer's instructions. The quality and quantity of the total and amplified RNA samples was determined with a 2100 Agilent Bioanalyser (Agilent Technologies, Palo Alto, CA, USA).

To ensure adequate representation of different patients and to allow for replicate hybridisations, amplified RNA samples were combined to form pools of at least three different patients (see supplementary material for numbers of patients per individual pool). At least two pools per type of lesion were analysed, and all pools were hybridised at least in triplicate (see supplementary material).

For each microarray hybridisation, 700 ng of amplified aminoallyl RNA was resuspended in 75 mM sodium carbonate buffer (pH 9.0) and coupled with 7.3 pM/5.6 μg/μl Cy5 monoreactive dye (Amersham Biosciences, Uppsala, Sweden) in DMSO for 1 h at room temperature in the dark. The reaction was quenched by adding 1.3 M hydroxylamine and incubation for 15 min at 25°C in the dark. Uncoupled dyes were removed by RNeasy-Kit (Qiagen, Hilden, Germany) and the final volume adjusted to 30 μl.

Reference cDNA was reverse transcribed from 10 μg of universal human reference RNA (Stratagene, La Jolla, CA, USA) using Superscript III Reverse Transcriptase (Invitrogen, Carlsbad, CA, USA), oligo-d(T)-primers and dTTP/aminoallyl-dUTP at a ratio of 1 : 2, according to the manufacturer's instructions. The aminoallyl cDNA was coupled with 5.6 μg/6l Cy3 monoreactive dye (Amersham Biosciences, Uppsala, Sweden) and purified as described above.

Microarray hybridisations

Following preincubation for 1 h at 42°C in prehybridisation buffer (3 × SSC, 0.25% SDS, 1% BSA), slides were denatured at 75°C for 1 min. In total, 30 μl each of the labelled experimental and reference samples were hybridised for 14–18 h at 37°C in hybridisation buffer (3 × SSC, 0.25% SDS, 0.3 μg/μl poly dA, 0.5 μg/μl yeast tRNA, 1 μg/μl salmon sperm DNA) using a GeneTac hybridisation chamber (Genomic Solutions, Cambridgeshire, UK). Slides were washed three times in 2 × SSC, 5% formamide, 0.1% Tween 20 (pH 7.0) at 37°C, once in 1 × PBS (pH 7.4), 0.05% Tween 20 (pH 7.0) at 25°C, once in 1 × PBS (pH 7.4), 0.1% Tween 20 (pH 7.0) at 25°C for 3 min and once in 0.5 × PBS (pH 7.4), 0.05% Tween 20 (pH 7.0) at 25°C for 5 min in the dark. Slides were dried in a centrifuge by spinning for 5 min at 1250 g.

Image and data analysis

Hybridisation signals were visualised using a dual laser scanner (Axon 4000B) and analysed with GenePix Pro 4.0 imaging software (Axon Instruments, Union City, CA, USA). On visual inspection, spots of insufficient quality were excluded from further analysis. For the purposes of this study, we analysed the signal intensities from the Cy5 channel only, since low signal intensities in the reference channel for individual spots can lead to loss of data points when using the common reference ratio method. Print-tip LOESS-normalised ratios of experimental and common reference samples obtained with the LIMMA software package (Smyth, 2004) for comparison with external data sets are available as part of the supplementary data (

To correct for differences between the microarray slides and for gradients within a slide, a block normalisation was performed. Following local background correction, signal intensities were normalised to the average of medians of all spots within individual 4-by-4 subarray blocks on each slide. Block normalised expression values of all individual hybridisations are available as part of the supplementary data accompanying this paper.

For the identification of differentially expressed genes, the block normalised data were first filtered to include only genes that exceeded a mean normalised expression value of 1 in at least one of the experimental groups. Genes were defined as differentially expressed between normal ducts and PanIN lesions or PDAC if (1) the difference between the mean normalised expression values was at least two-fold; and (2) a two-sided T-test yielded a P-value of <0.01. False discovery rates were estimated by analysing all possible permutations of group assignments (Tusher et al., 2001). Lists of genes differentially expressed between the different sample sets are available as part of the supplementary data.

To further analyse the biological function of differentially expressed genes, the GoMiner package (Zeeberg et al., 2003) was used to organize lists of genes for biological interpretation in the context of the Gene Ontology (GO) data base. Hierarchical cluster analysis of expression profiles was performed using the ‘Genesis’ software tool (Sturn et al., 2002) with uncentred Pearson correlation as the similarity measure.

Data validation by real-time RT–PCR

Real-time PCR was performed using the comparative CT method on the ABI Prism Sequence Detection System (PE Applied Biosystems, Forster City, CA, USA), according to the manufacturer's instructions. In brief, cDNA was reverse transcribed from 0.2 μg of amplified antisense RNA using Superscript III Reverse Transcriptase (Invitrogen, Carlsbad, CA, USA) and random decamers, according to the manufacturer's instructions. Primers for 12 genes were defined using the PrimerExpress program (PE Applied Biosystems) and are listed in Table 1. Samples were amplified using the SYBR Green PCR Master Mix system (PE Applied Biosystems). The human cyclophilin gene (RefSeq ID NM_021130) was used as internal standard. Variation was assessed by calculating standard errors of the mean of all possible differences of individual ΔCT-values between two sets of samples according to the following formula:

with Xj representing the different stages of disease progression.

Table 1 qRT–PCR primer sequences


Protein expression of the genes that was found to be differentially expressed was tested using a TMA that included the various grades of PanINs, normal pancreatic tissue and PDAC. Briefly, cores of 0.2 μm diameter that included the requested PanIN lesion were taken from the donor block and transferred to the recipient block using a microarrayer (Beecher Instruments, Silver Springs, MD, USA) (Bubendorf et al., 2001). For histological investigation, 5 μm sections were cut and transferred to uncovered superfrost slides (Menzel, Gläser, Braunschweig, Germany). The various antibodies were tested using the required different control tissues. In most cases, antigen demasking was necessary and performed by the pressure cooker method. Immunostaining was carried out by the ABC method as described previously (Luttges et al., 2004). Briefly, the slides were incubated for 45 min with the primary antibodies followed by an incubation with a biotinylated secondary antibody (5 μg/ml, Vector Laboratories, Burlingame, CA, USA) and avidin-biotin-peroxidase (ABC ELITE, Vector Laboratories). Diaminobenzidine served as chromogen for antigen detection. Subsequently, the slides were briefly counterstained with haematoxylin. For CEACAM5, the APAAP method was applied as previously described (Luttges et al., 2000). For the negative control the primary antibody was omitted.

Accession codes




  1. Bramhall SR, Neoptolemos JP, Stamp GW and Lemoine NR . (1997). J. Pathol., 182, 347–355.

  2. Brentnall TA, Bronner MP, Byrd DR, Haggitt RC and Kimmey MB . (1999). Ann. Intern. Med., 131, 247–255.

  3. Bubendorf L, Nocito A, Moch H and Sauter G . (2001). J. Pathol., 195, 72–79.

  4. Buchholz M, Biebl A, Neesse A, Wagner M, Iwamura T, Leder G, Adler G and Gress TM . (2003). Cancer Res., 63, 4945–4951.

  5. Canto MI, Goggins M, Yeo CJ, Griffin C, Axilbund JE, Brune K, Ali SZ, Jagannath S, Petersen GM, Fishman EK, Piantadosi S, Giardiello FM and Hruban RH . (2004). Clin. Gastroenterol. Hepatol., 2, 606–621.

  6. Conlon KC, Klimstra DS and Brennan MF . (1996). Ann. Surg., 223, 273–279.

  7. Crnogorac-Jurcevic T, Missiaglia E, Blaveri E, Gangeswaran R, Jones M, Terris B, Costello E, Neoptolemos JP and Lemoine NR . (2003). J. Pathol., 201, 63–74.

  8. Ellenrieder V, Alber B, Lacher U, Hendler SF, Menke A, Boeck W, Wagner M, Wilda M, Friess H, Buchler M, Adler G and Gress TM . (2000). Int. J. Cancer, 85, 14–20.

  9. Fearon ER and Vogelstein B . (1990). Cell, 61, 759–767.

  10. Goggins M, Hruban RH and Kern SE . (2000). Am. J. Pathol., 156, 1767–1771.

  11. Gronborg M, Bunkenborg J, Kristiansen TZ, Jensen ON, Yeo CJ, Hruban RH, Maitra A, Goggins MG and Pandey A . (2004). J. Proteome Res., 3, 1042–1055.

  12. Heinmoller E, Dietmaier W, Zirngibl H, Heinmoller P, Scaringe W, Jauch KW, Hofstadter F and Ruschoff J . (2000). Am. J. Pathol., 157, 83–92.

  13. Hruban RH, Adsay NV, Albores-Saavedra J, Compton C, Garrett ES, Goodman SN, Kern SE, Klimstra DS, Kloppel G, Longnecker DS, Luttges J and Offerhaus GJ . (2001a). Am. J. Surg. Pathol., 25, 579–586.

  14. Hruban RH, Adsay NV, Albores-Saavedra J, Compton C, Garrett ES, Goodman SN, Kern SE, Klimstra DS, Kloppel G, Longnecker DS, Luttges J and Offerhaus GJ . (2001b). Am. J. Surg. Pathol., 25, 579–586.

  15. Hruban RH, Canto MI and Yeo CJ . (2001c). Dig. Dis., 19, 76–84.

  16. Hruban RH, Takaori K, Klimstra DS, Adsay NV, Albores-Saavedra J, Biankin AV, Biankin SA, Compton C, Fukushima N, Furukawa T, Goggins M, Kato Y, Kloppel G, Longnecker DS, Luttges J, Maitra A, Offerhaus GJ, Shimizu M and Yonezawa S . (2004). Am. J. Surg. Pathol., 28, 977–987.

  17. Hruban RH, Wilentz RE and Kern SE . (2000). Am. J. Pathol., 156, 1821–1825.

  18. Huflejt ME and Leffler H . (2004). Glycoconj. J., 20, 247–255.

  19. Kawakami K, Kawakami M, Joshi BH and Puri RK . (2001). Cancer Res., 61, 6194–6200.

  20. Kloppel G and Luttges J . (2004). Curr. Gastroenterol. Rep., 6, 111–118.

  21. Kos J, Werle B, Lah T and Brunner N . (2000). Int. J. Biol. Markers, 15, 84–89.

  22. Kristiansen G, Pilarsky C, Wissmann C, Kaiser S, Bruemmendorf T, Roepcke S, Dahl E, Hinzmann B, Specht T, Pervan J, Stephan C, Loening S, Dietel M and Rosenthal A . (2004). J. Pathol., 205, 359–376.

  23. Liotta LA and Kohn EC . (2001). Nature, 411, 375–379.

  24. Luttges J, Diederichs A, Menke MA, Vogel I, Kremer B and Kloppel G . (2000). Cancer, 88, 2495–2504.

  25. Luttges J, Galehdari H, Brocker V, Schwarte-Waldhoff I, Henne-Bruns D, Kloppel G, Schmiegel W and Hahn SA . (2001). Am. J. Pathol., 158, 1677–1683.

  26. Luttges J, Reinecke-Luthge A, Mollmann B, Menke MA, Clemens A, Klimpfinger M, Sipos B and Kloppel G . (1999). Virchows. Arch., 435, 461–468.

  27. Luttges J, Stigge C, Pacena M and Kloppel G . (2004). Cancer, 100, 173–182.

  28. Lynch HT . (1994). Arch. Surg., 129, 266–268.

  29. Lynch HT, Fitzsimmons ML, Smyrk TC, Lanspa SJ, Watson P, McClellan J and Lynch JF . (1990). Am. J. Gastroenterol., 85, 54–60.

  30. Lynch HT, Smyrk T, Kern SE, Hruban RH, Lightdale CJ, Lemon SJ, Lynch JF, Fusaro LR, Fusaro RM and Ghadirian P . (1996). Semin. Oncol., 23, 251–275.

  31. Maitra A, Adsay NV, Argani P, Iacobuzio-Donahue C, De Marzo A, Cameron JL, Yeo CJ and Hruban RH . (2003). Mod. Pathol., 16, 902–912.

  32. Neoptolemos JP, Stocken DD, Friess H, Bassi C, Dunn JA, Hickey H, Beger H, Fernandez-Cruz L, Dervenis C, Lacaine F, Falconi M, Pederzoli P, Pap A, Spooner D, Kerr DJ and Buchler MW . (2004). N. Engl. J. Med., 350, 1200–1210.

  33. Parkin DM, Pisani P and Ferlay J . (1999). CA Cancer J. Clin., 49, 33–64 1.

  34. Pupa SM, Menard S, Forti S and Tagliabue E . (2002). J. Cell Physiol., 192, 259–267.

  35. Smyth GK . (2004). Statistical Applications in Genetics and Molecular Biology, 3, Article 3.

  36. Sperti C, Pasquali C, Piccoli A and Pedrazzoli S . (1997). World J. Surg., 21, 195–200.

  37. Stepanova L, Yang G, DeMayo F, Wheeler TM, Finegold M, Thompson TC and Harper JW . (2000). Oncogene, 19, 2186–2193.

  38. Strojan P, Svetic B, Smid L and Kos J . (2004). Clin. Chim. Acta, 344, 155–161.

  39. Sturn A, Quackenbush J and Trajanoski Z . (2002). Bioinformatics, 18, 207–208.

  40. Suomela S, Cao L, Bowcock A and Saarialho-Kere U . (2004). J. Invest. Dermatol., 122, 717–721.

  41. Terris B, Blaveri E, Crnogorac-Jurcevic T, Jones M, Missiaglia E, Ruszniewski P, Sauvanet A and Lemoine NR . (2002). Am. J. Pathol., 160, 1745–1754.

  42. Trede M, Richter A and Wendl K . (2001). Surg. Clin. N Am., 81, 595–610.

  43. Tusher VG, Tibshirani R and Chu G . (2001). Proc. Natl. Acad. Sci. USA, 98, 5116–5121.

  44. Vimalachandran D, Ghaneh P, Costello E and Neoptolemos JP . (2004). Cancer Control, 11, 6–14.

  45. Wilentz RE, Geradts J, Maynard R, Offerhaus GJ, Kang M, Goggins M, Yeo CJ, Kern SE and Hruban RH . (1998). Cancer Res., 58, 4740–4744.

  46. Wilentz RE, Iacobuzio-Donahue CA, Argani P, McCarthy DM, Parsons JL, Yeo CJ, Kern SE and Hruban RH . (2000). Cancer Res., 60, 2002–2006.

  47. Yamakawa T, Miyata S, Ogawa N, Koshikawa N, Yasumitsu H, Kanamori T and Miyazaki K . (1998). Biochim. Biophys. Acta, 1395, 202–208.

  48. Yeo CJ and Cameron JL . (1998). Langenbecks Arch. Surg., 383, 129–133.

  49. Zeeberg BR, Feng W, Wang G, Wang MD, Fojo AT, Sunshine M, Narasimhan S, Kane DW, Reinhold WC, Lababidi S, Bussey KJ, Riss J, Barrett JC and Weinstein JN . (2003). Genome Biol., 4, R28.

Download references


We thank Susanne Braun, Karin Lanz, Claudia Ruhland, Pat Schreiter and Britta Redeker for excellent technical assistance. The work of the GPCN is supported by a multicentre grant form the Deutsche Krebshilfe, Bonn, Germany.

Author information

Correspondence to Thomas M Gress.

Rights and permissions

Reprints and Permissions

About this article

Cite this article

Buchholz, M., Braun, M., Heidenblut, A. et al. Transcriptome analysis of microdissected pancreatic intraepithelial neoplastic lesions. Oncogene 24, 6626–6636 (2005).

Download citation


  • PanIN progression
  • gene expression
  • early diagnosis
  • target genes

Further reading