Identification of genetic alterations in pancreatic cancer by the combined use of tissue microdissection and array-based comparative genomic hybridisation

Pancreatic ductal adenocarcinoma (PDAC) is characterised pathologically by a marked desmoplastic stromal reaction that significantly reduces the sensitivity and specificity of cytogenetic analysis. To identify genetic alterations that reflect the characteristics of the tumour in vivo, we screened a total of 23 microdissected PDAC tissue samples using array-based comparative genomic hybridisation (array CGH) with 1 Mb resolution. Highly stringent statistical analysis enabled us to define the regions of nonrandom genomic changes. We detected a total of 41 contiguous regions (>3.0 Mb) of copy number changes, such as a genetic gain at 7p22.2–p15.1 (26.0 Mb) and losses at 17p13.3–p11.2 (13.6 Mb), 18q21.2–q22.1 (12.0 Mb), 18q22.3–q23 (7.1 Mb) and 18q12.3–q21.2 (6.9 Mb). To validate our array CGH results, fluorescence in situ hybridisation was performed using four probes from those regions, showing that these genetic alterations were observed in 37–68% of a separate sample set of 19 PDAC cases. In particular, deletion of the SEC11L3 gene (18q21.32) was detected at a very high frequency (13 out of 19 cases; 68%) and in situ RNA hybridisation for this gene demonstrated a significant correlation between deletion and expression levels. It was further confirmed by reverse transcription–PCR that SEC11L3 mRNA was downregulated in 16 out of 16 PDAC tissues (100%). In conclusion, the combination of tissue microdissection and array CGH provided a valid data set that represents in vivo genetic changes in PDAC. Our results raise the possibility that the SEC11L3 gene may play a role as a tumour suppressor in this disease.

Pancreatic ductal adenocarcinoma (PDAC) is the fourth leading cause of cancer-related death in Western countries. Although the prognosis of patients with many types of cancer has improved recently due to advances in diagnostic and therapeutic modalities, the outlook for patients with PDAC still remains dismal with a median survival of just 6 months from diagnosis and an overall 5year survival rate of less than 5% (Warshaw and Fernandez-del Castillo, 1992;Murr et al, 1994;Jemal et al, 2005). Many previous studies have attempted to elucidate the molecular mechanisms underlying pancreatic tumorigenesis, but it is still not fully understood. Therefore, a better understanding of the genes involved in tumour growth and progression is necessary for the development of novel diagnostic and therapeutic strategies that could improve the outcome of this deadly disease.
Array-based comparative genomic hybridisation (array CGH) is a powerful technique that has been used to detect DNA copy number alterations across the entire genome of malignant tumours (Solinas-Toldo et al, 1997;Pinkel et al, 1998;Pollack et al, 1999;Albertson et al, 2000;Snijders et al, 2001;Fiegler et al, 2003). Compared to conventional CGH, the significantly improved resolution of array CGH permits highly accurate mapping of DNA copy number changes throughout the genome. In cancer, genomic alterations contribute to dysregulation of the expression levels of oncogenes and tumour suppressor genes, the accumulation of which is correlated with tumour progression (Ried et al, 1999). Therefore, array CGH provides a promising starting point for the identification of novel candidate genes affected by such genomic imbalances. Several array CGH investigations of PDAC have already been reported, but all these studies were performed on cell lines or whole tissue samples (Aguirre et al, 2004;Heidenblad et al, 2004;Holzmann et al, 2004;Mahlamaki et al, 2004;Bashyam et al, 2005;Gysin et al, 2005;Nowak et al, 2005). In cell lines, culture-induced genetic adaptation may be induced during the establishment of cell lines in in vitro conditions. On the other hand, PDAC tissues are characterised by a desmoplastic reaction, with neoplastic cells constituting only a small proportion of the tumour mass. Therefore, cytogenetic analysis using bulk tissue samples is invariably hampered by contamination with nonneoplastic cells.
The aim of this study is to identify novel genetic abnormalities that precisely reflect the characteristics of tumour cells in vivo. For this purpose, we applied array CGH to 23 microdissected PDAC tissue samples that consist of purified populations of cancer cells. Then, a subset of identified genetic alterations was evaluated in an independent sample set of 19 PDAC cases using fluorescence in situ hybridisation (FISH) analysis. Finally, in situ RNA hybridisation (ISH) and reverse transcription -PCR (RT -PCR) were performed to assess whether the identified genetic alteration could lead to significant change in transcript level of the gene in question.

Tissue samples
A total of 23 fresh-frozen PDAC specimens were obtained surgically or at autopsy from Yamaguchi University School of Medicine, Japan, with appropriate ethical approval ( Table 1). All the tissues were confirmed histologically by a pathologist. Tissue microdissection was performed manually to collect more than 90% of tumour cells in all the cases as described previously (Harada et al, 2002a). Briefly, only cancerous regions were microdissected using a sterile 26-gauge needle from several serial tissue sections (20 mm thickness) under visualisation with an inverted microscope (Nikon 66906, Tokyo, Japan). DNA was extracted from at least 5000 tumour cells according to the standard protocol. Reference DNA was obtained from lymphocytes of both healthy male and female volunteers. In addition, another series of 19 formalin-fixed, paraffin-embedded tumour sections (4 mm thickness) were provided from Yamaguchi University (n ¼ 10) and Tohoku University School of Medicine, Japan (n ¼ 9), for FISH and ISH analyses ( Table 1). Owing to the limited accessibility of clinical specimens, the samples used in array CGH were not available for FISH and ISH. For RT -PCR, 16 fresh-frozen PDAC tissues and two normal pancreas tissues were obtained from the Human Biomaterials Resource Centre, Department of Histopathology, Charing Cross Hospital, London, UK, with full ethical approval of the host institution. The clinicopathological information was not available for these anonymous samples. Haematoxylin -eosin-stained slides were examined to ensure a content of 60 -80% tumour cells before use and then, total RNA was extracted directly from homogenised tissues using TRIZOL (Invitrogen Ltd, Paisley, UK).

CGH arrays and image acquisition
The whole-genome CGH arrays were produced at the Wellcome Trust Sanger Institute and consist of 3125 BAC/PAC clones that cover the entire human genome at 1 Mb resolution (Fiegler et al, 2003). All the clone details are available on the Ensembl genome browser (v39, June 2006; http://www.ensembl.org/Homo_sapiens/ index.html). Array CGH was performed as described previously, with minor modifications (Fiegler et al, 2003;Hurst et al, 2004). Briefly, tumour and reference DNAs (300 ng) were labelled with Cy5-dCTP and Cy3-dCTP, respectively. Hybridisation was carried out at 371C for 36 h in a hybridisation chamber (GeneMachines, San Carlos, CA, USA). After washing the slides, fluorescence intensities were measured on an Axon 4000B scanner (Axon Instruments Inc., Burlingame, CA, USA) and the raw images were analysed using the GenePix Pro 4.0 software (Axon). Duplicate hybridisations were performed for each sample to verify the reproducibility of the data except for one case (PC5). The correlation coefficients were calculated on the normalised tumour channel and observed to range from 0.63 to 0.90 (data not shown).

Statistical and data analysis
The CGH arrays were read with the UCSF 'SPOT' software to produce raw text files (Jain et al, 2002). These files were read into R and normalised (using the loess intensity-dependent method), using the 'limma' package (Smyth, 2005; The R Development Core Team, 2006); background correction was omitted in this case, as visual inspection showed it increased the scatter in both the MA and chromosomal-location plots, and the correlation statistics were worse with background subtraction, indicative of low levels of background on the slides being misestimated. The log 2 -transformed normalised data were then pre-processed to average any on-slide replicates using the 'snapCGH' package from Bioconductor and segmented into local regions of constant copy number by circular binary segmentation (Gentleman et al, 2004;Olshen et al, 2004). The sample levels (two replicates for each tumour per clone) were summarised by means to give tumour-level data (one measurement per clone for each tumour). These data were then used in a linear model that estimated the fold change across the tumours, along with a P-value that the average log 2 -fold change between the tumour channel and the normal channel was non-zero (for both the normalised data and the locally smoothed data -data not shown for the latter). Clones that had an uncorrected P-value below 0.001 were considered to be significant candidates.

Two-colour FISH
Four human BAC clones (RP11-403N12, RP11-232C20, RP11-8H2 and RP11-350K6) were purchased from BACPAC Resources (Oakland, CA, USA) and these DNAs were labelled with Cy3-dCTP using BioPrime Array CGH Genomic Labeling System (Invitrogen). Centromeric probes for chromosomes 7 and 18 (CEP7 and 18), labelled with SpectrumGreen, were purchased from Vysis (Downers Grove, IL, USA). The specificity of all the probes was confirmed by hybridisation onto Normal Male Metaphase (Vysis). Two-colour FISH was performed as described previously (Lu et al, 1999). DNA copy number was evaluated for each probe by counting spots in at least 100 nuclei. A ratio was calculated between the average copy number of the BAC probes and of corresponding centromeric probes. Based on hybridisation in 10 normal pancreatic tissues (acinar and ductal cells), the threshold of gain and loss was defined as the ratios of 41.16 ( þ 2 standard deviation (s.d.)) and o0.87 (À2 s.d.), respectively (data not shown). Our approach was to use normal samples to estimate overall noise levels: choosing the threshold on the tumour samples corresponding to a 72 s.d. of the normal samples indicates a roughly 5% false-positive rate if the tumour samples were commensurate with diploidy.
In situ RNA hybridisation for SEC11L3 The SEC11L3 probe was amplified by PCR from OriGene clone TC123085 (OriGene Technologies, Inc., Rockville, MD, USA) that encodes full-length cDNA of SEC11L3. The primers used to amplify a 216-bp SEC11L3 product are as follows: forward 5 0 -TTGGATA TCTTCGGGGACCT-3 0 and reverse 5 0 -GTCTTCCCGGAAATTTGT GA-3 0 . The PCR product was cloned into the pCR4-TOPO vector using the TOPO cloning kit (Invitrogen) to create pCR4-SEC11L3-ISH. Positive clones were verified by sequence analysis. The pCR4-SEC11L3-ISH plasmid was linearised with PstI for the sense probe and NotI for the antisense probe. After removing restriction endonucleases, riboprobes were synthesised from 1 mg of template DNA and digoxigenin (DIG) were labelled using a DIG RNA labeling kit (Roche Diagnostics GmbH, Mannheim, Germany). T3 and T7 polymerases were used to synthesise antisense probes and sense probes, respectively. DIG incorporation of riboprobes was verified by DOT blot with anti-digoxigenin-AP Fab fragments (Roche). Antisense and sense riboprobes for SEC11L3 were hybridised to 19 tissue sections using the Ventana Discovery System with Ventana Ribomap and Bluemap kits. Expression of SEC11L3 mRNA in cancer cells was compared to that of nonneoplastic epithelial cells (ductal, acinar, intestinal and hepatic cells) on the identical specimen and judged using a 0 -2 score (0 ¼ no staining, 1 ¼ weak intensity, 2 ¼ intensity comparable to non-neoplastic counterparts).

Reverse transcription -PCR for SEC11L3
cDNAs were synthesised from 1 mg of total RNA using an oligo dT primer and the Multiscribe reverse transcription kit (Applied Biosystems, Warrington, Cheshire, UK) as instructed by the manufacturer. Reverse transcription was followed by 30 PCR cycles (1 min of denaturation at 941C, 1 min of annealing at 551C and 1 min of extension at 721C). Primers for SEC11L3 are the same as those designed in ISH. Primers for 18S ribosomal RNA, which was used as an endogenous control for normalisation, are as follows: forward 5 0 -CGCCGCTAGAGGTGAAATTC-3 0 and reverse 5 0 -CATTCTTGGCAAATGCTTTCG-3 0 . Amplified products were separated on 1% agarose gels and visualised with ethidium bromide.

Comparison of array CGH profiles in microdissected tissues and cell lines
A total of 23 microdissected PDAC tissues were analysed by array CGH. Applying highly stringent statistical conditions (Po0.001), we could identify the regions of nonrandom genomic changes in PDAC. Figure 1 shows overall copy number changes for each chromosome and the entire data set of all clones is available in Supplementary Table 1 (the raw data set for all the experiments is shown in Supplementary Table 2). A total of 1015 clones met the statistical criterion; 17% of clones (508 clones including 698 candidate genes) showed genetic gain and 17% (507 clones including 1254 genes) showed loss. The array CGH profiles were compared to the previous reports in which PDAC-derived cell lines were analysed (Aguirre et al, 2004;Bashyam et al, 2005;Gysin et al, 2005;Nowak et al, 2005). Although they displayed similar spectrum patterns of genetic alterations overall, we found that there were apparently different breakpoints in our profiles. Our data showed several segmented gains on chromosome 2, which have rarely been observed in cell lines. In contrast, losses of 4q and 13q and gains of 11q and 20q in cell lines were not as frequent as in our microdissected samples. Next, we focused on individual clones harbouring the genes that are known to be dysregulated in cell lines. Increased copy numbers were detected in the regions including MYC (8q24.21) and NCOA3/AIB1 (20q13.12), while genetic losses were observed in the regions containing SMAD4 (18q21.1), TP53 (17p13.1), MAP2K4 (17p11.2) and RUNX3 (1p36.11). However, using the rigorous statistical conditions employed, we identified neither genetic gains of KRAS (12p12.1), MYB (6q23.3), EGFR (7p11.2) and ERBB2 (17q12) nor losses of MLH1 (3p22.2), BRCA2 (13q13.1) and CDH1 (16q22.1). (All the genes cited here are depicted in Figure 1.)

Contiguous regions of nonrandom copy number changes
In addition to numerous localised alterations, we detected a total of 41 contiguous regions (43.0 Mb) of nonrandom genomic changes ( Table 2). For instance, increased copy number was detected in the 26.0 Mb region of 7p22.2 -p15.1 that contains 48 known or hypothetical protein-coding genes. We also defined the regions of genetic gains on 1q, 3q, 5p, 5q, 8q and 12p, which may represent loci for candidate oncogenes in PDAC. The largest region of copy number loss was from 17p13.3 to 17p12 (13.6 Mb), which covers a total of 53 candidate genes including TP53 (17p13.1) as well as MAP2K4 (17p11.2). We delineated three contiguous regions of genomic loss on 18q, which is known to be a site of frequent deletions in PDAC, 18q21.2 -q22.1 (12.0 Mb), 18q22.3 -q23 (7.1 Mb) and 18q12.3 -q21.2 (6.9 Mb). The region of 18q21.2 -q22.1 harbours 16 candidate genes in addition to SMAD4 (18q21.1) that is one of the most recurrently inactivated tumour suppressor genes in PDAC. The region of 18q12.3 -q21.2 contains a total of 23 putative target genes, whereas seven genes are included in the 7.1 Mb region of 18q22.3 -q23. Interestingly, the clone encompassing DCC (18q21.2) has shown an apparent discontinuity between two broad regions of genetic loss (18q12.3 -q21.2 and 18q21.2 -q22.1) in our 23 microdissected PDAC sample set ( Figure 1).

SEC11L3 mRNA downregulation detected by ISH and RT -PCR
Subsequently, the SEC11L3 mRNA level was evaluated by ISH ( Figure 3). Firstly, we tested several different types of normal epithelial cells from the pancreas, intestine and liver. Normal pancreatic tissues showed strong mRNA expression of SEC11L3 in both ductal and acinar cells (score 2), whereas there was weak expression in islet cells (score 1). Strong expression was also observed in normal intestinal and hepatic cells (score 2). Of 19 PDAC cases, SEC11L3 mRNA was downregulated (score 0 -1) in 11 cases (58%), whereas it was unchanged (score 2) in eight cases  (42%) ( Table 4). All 11 cases with downregulated SEC11L3 demonstrated genetic losses by FISH. Despite decreased copy numbers, SEC11L3 was expressed in two samples (PC46 and PC51). A significant correlation between deletion and expression levels was found (P ¼ 0.001, Fisher's exact test), indicating that the mRNA level of this gene was highly dependent on its DNA copy number. RT -PCR was performed further to confirm downregulated mRNA of SEC11L3 in PDAC. Figure 4 shows SEC11L3 expression in normal pancreas and PDAC tissue samples. Compared to two normal pancreas tissues, SEC11L3 was found to be downregulated in 16 out of 16 PDAC samples (100%). In particular, the SEC11L3 transcript was virtually almost absent in four PDAC tissues (25%; lane 5, 6, 7 and 15).

DISCUSSION
It is well known that a strong desmoplastic reaction is a typical feature of PDAC tissues. A dense stromal component, which occupies larger parts of the tumour mass, significantly reduces the sensitivity and specificity of cytogenetic analysis. Tissue microdissection is laborious, but the practical method available to enrich the tumour cell population in clinical specimens. In the present study, we first identified genomic abnormalities that represent the characteristics of tumour cells in vivo by combining CGH arrays with tissue microdissection. This approach led to more precise definition of chromosomal breakpoints in a panel of 23 PDAC tissue samples. To identify nonrandom genomic changes in PDAC, we applied a P-value rather than a fixed cutoff value because we found that concomitant lack of power in dichotomising the data at an early stage in the analysis provided poor resolution with which to distinguish between clones. Taking a statistical threshold approach allows us to take account of different clones' differing variance across samples that a fixed fold-change approach does not.
We compared the CGH profiles to the previously published cell line data (Karhu et al, 2006). Despite overall similar spectrum patterns, there were clear differences between both profiles. It is important to take into account that the resolution of CGH arrays used and the type of statistical analysis employed vary widely between the reports. However, our results indicated that some recurrent genetic alterations, such as losses of 4q and 13q and gains of 11q and 20q, seem to be relatively unique to cell lines, implying that these genetic changes may have been artificially acquired through the establishment of cell lines or in the course of culturing. In addition, our data did not demonstrate significant copy number changes of some known genes, such as KRAS, ERBB2, MLH1 and CDH1. This is probably due to the fact that intragenic mutation or promoter methylation is more likely to occur in these genes (Lemoine et al, 1992;Scarpa et al, 1994;Rozenblum et al, 1997;Ueki et al, 2000). Alternatively, this discrepancy could be explained by the intratumoral heterogeneity that is characteristically observed in PDAC cells in vitro as well as in vivo, or may reflect the differences of the geographic origin of the tumours used in this study (a total of 42 Japanese samples were  analysed by array CGH, FISH and ISH) (Scarpa et al, 1994;Harada et al, 2002b). Similarly, deletion of the DCC gene was not recurrent in our sample set. However, a larger scale of study using much higher-resolution genomewide microarrays (tiling-path CGH arrays or single nucleotide polymorphism arrays) is required to conclude whether this genetic alteration is critically involved in the pathogenesis of PDAC.
As our array CGH profiles revealed that the regions of 7p22.2 -p15.1, 18q12.3 -q21.2 and 18q21.2 -q22.1 are nonrandomly altered, four clones from those regions were validated by FISH experiments. The results showed that genetic alterations for these clones were observed in 37 -68% of tumours in a separate sample set, supporting the validity of our CGH results. Moreover, copy number changes of those four clones were detected in 18 out of 19 PDAC cases (95%), implying the potential clinical applicability. Among the candidate genes verified by FISH, BCMP11 is newly detected in PDAC, although Fletcher et al (2003) demonstrated that mRNA and protein of this gene were overexpressed in oestrogen receptor-positive breast cancer. This gene lies adjacent to the AGR2 gene (7p21.1) and both genes are classified as members of the same family (AGR family) due to highly similar (approximately 70%) protein sequences. Interestingly, several gene expression analyses have shown that AGR2 is upregulated in the majority of PDACs as well as pancreatic intraepithelial neoplasia (PanIN) lesions (Crnogorac-Jurcevic et al, 2003;Iacobuzio-Donahue et al, 2003a;Iacobuzio-Donahue et al, 2003b;Missiaglia et al, 2004;Buchholz et al, 2005). Taken together, BCMP11 is also likely to be involved in the development of PDAC. On the other hand, SCAP2 was first described in a recently published report,  showing that mRNA of this gene is frequently overexpressed in PanIN lesions (Buchholz et al, 2005). The protein encoded by this gene belongs to the src family of kinases. Takahashi et al (2003) demonstrated that SCAP2 functions as a downstream target of c-Src under various stress conditions (UV light, tumour necrosis factor-a and osmotic stress). Therefore, SCAP2 also seems to work as a cell-signalling molecule in cancer cells. However, the biological function and putative role of these two genes have not been investigated in cancer.
Remarkably, SEC11L3 was deleted in approximately 70% of tumours and its expression level was significantly correlated with its DNA copy number. Despite the high frequency of this genetic abnormality, this gene has not been described in any type of cancer. Previous cytogenetic analyses have revealed a frequent deletion of 18q in PDAC, but neither the chromosomal breakpoints nor the candidate genes included could be clearly identified due to technical limitations of the technology employed (Griffin et al, 1995;Hahn et al, 1995;Fukushige et al, 1997Fukushige et al, , 1998Mahlamaki et al, 1997;Hoglund et al, 1998;Schleger et al, 2000;Harada et al, 2002a;Iacobuzio-Donahue et al, 2004). SMAD4 (18q21.1) has been reported to be deleted or inactivated in about 50% of PDACs and, therefore, it is considered to be one of the most likely candidate tumour suppressor genes at this locus (Hahn et al, 1996;Rozenblum et al, 1997). However, we propose that SEC11L3 (18q21.32) could be an equally promising candidate gene on 18q because the significance of genetic loss of this gene is comparable to that of the SMAD4 gene. In addition, both ISH and RT -PCR independently confirmed that SEC11L3 mRNA is downregulated in PDAC tissues at a high frequency (53% and 100%, respectively), suggesting that dysregulation of this gene is likely to be associated with the development of PDAC. Little is known about the biological role of this gene, although it belongs to the peptidase S26B family and functions as part of the signal peptidase complex.
In summary, we could successfully identify genetic alterations that reflect the intrinsic characteristics of PDAC cells in vivo by combining array CGH with tissue microdissection. These results provided a valid data set to search for novel candidate genes involved in pancreatic carcinogenesis. The specificity of our array CGH results was confirmed by interphase FISH in an independent sample set. Among the identified candidates, we are particularly interested in the SEC11L3 gene that is located on 18q21.32. FISH and ISH analyses for this gene demonstrated a significant correlation between genetic deletion and the corresponding mRNA downregulation, raising the possibility that the SEC11L3 gene may play a putative role as a tumour suppressor. For these reasons, we propose that SEC11L3 should be considered as a potential marker gene for the molecular diagnosis of PDAC and a possible candidate target for therapeutic intervention.