Background
Genomic copy number changes are frequently found in different types of cancer and are believed to contribute to their development and progression through inactivation of tumour suppressor genes, activation of oncogenes, or more subtle through gene dosage changes. Comparative genomic hybridization (CGH) [
1] was developed to allow genome-wide screening for such copy number changes. Conventional CGH has a limited resolution and can detect losses of 10 Mb or greater [
2]. High-level amplifications achieve a maximum resolution of 3 Mb [
3]. The resolution of CGH has been improved by replacing the metaphase chromosomes, which have traditionally served as hybridization targets, with mapped and sequenced genomic DNA clones (bacterial artificial chromosomes, P1-derived artificial chromosome or cosmids) arrayed onto glass slides which was named "matrix-CGH" [
4] or "array-CGH" [
5].
Although genomic DNA arrays are considered powerful research tools, their potential to meet specific needs in clinical diagnostics has been debated. Initially, matrix-CGH was restricted to cell lines or investigated inherited diseases, both characterized by a genetically homogeneous cell population [
4,
5]. More recently, it was also used in research studies addressing issues of tumour classification and correlation of gene dosage with gene expression studies [
6].
Genomic arrays allow the identification of genomic copy number alterations that may be suitable for individualized diagnostic, prognostic and therapeutic decision making. This is of particular importance since personalized treatments for cancer patients based on genomic alterations are becoming available. Ideally this type of analysis is performed using DNA from tumour samples freshly collected and snap frozen in liquid nitrogen. However, this type of tissue is not always available in clinical routine. Furthermore, in order to identify prognostic alterations, the analysis of large numbers of well documented tumour samples with ample and accurate clinical follow-up data is required. This information is usually available in controlled multicentric therapeutic studies. Many large-scale multicentric therapeutic studies such as those from the EORTC ("European Organisation for Research and Treatment of Cancer") already mandate the collection and storage of formalin-fixed paraffin-embedded tumour samples in their study protocols [
7], and archival paraffin tissue-blocks can be collected for most patients that participated in past large multicentric trials which would have the additional benefit of an extended follow up. Unfortunately, current protocols for matrix-CGH require microgramm quantities of high quality DNA [
8]. Such material can usually not be obtained from the formalin-fixed, paraffin embedded specimens making these valuable resources unavailable for the search of prognostic genomic alterations. Different protocols to overcome this problem were published [
9,
10]. Recently, DeVries and coworkers [
10] compared hybridizations with DNA extracted from formalin-fixed paraffin-embedded breast cancer tissue samples after manual microdissection and with DNA derived from the same fresh frozen tumour. In this study, reproducible results from archived formalin-fixed paraffin-embbeded tissue samples were obtained using as little as 50 ng input DNA. However, the conditions available in these studies only partly reflect the situation found in routine clinical work or in large multicentric studies. Their tissue blocks were collected by an expert center and were all derived from one institution, thus ensuring comparable and high-quality tissue processing. In most clinical trials, tissue blocks will be collected by a large number of centers, many of them without any expertise in the processing of samples for molecular analyses and using different protocols for fixation and embedding.
So far it is not known, whether DNA of acceptable quality for molecular analyses such as matrix-CGH can be obtained from this type of material in a multicentric setting. In this report, we present a pilot project initiated by the translational research group of the EORTC-GI group addressing this particular question. To test the technical suitability of this type of material, we performed matrix-CGH analysis of macrodissected colorectal tumours collected in the framework of the EORTC-GI PETACC-2 trial. In addition to testing the technical feasibility of this approach, the scientific aim of this pilot project was to identify prognostic genomic signatures differentiating locally restricted (UICC stages II-III) from systemically advanced (UICC stage IV) disease.
Methods
Patients and tumour specimens
29 paraffin-embedded colorectal tumour samples in different stages (10 × UICC II, 11 × UICC III, 8 × UICC IV) were collected in the framework of the PETACC-2 study. 23 of these samples were from patients which participated in the trial, and 6 additional samples were generated from archival material collected by PETACC-2-participants. The samples were provided by the EORTC-GI group. From one tumour (no. 29) we did also obtain snap frozen tissue. The retrospective use of tissue blocks from these patients for translational research was approved by the ethics committee of the "Ärztekammer Niedersachsen" (Berliner Allee 20, 30175 Hannover, Germany, reference number Grae/128/2002).
Tissue samples were collected at four institutions (University of Padua, Italy, Hôpital Ambroise Paré, Boulogne, Paris, France, Department of Pathology, University of Technology, Dresden and Department of Pathology University Hospital of Ulm, Germany). All specimens were fixed in formalin and paraffin-embedded using the standard protocols active in the participating centers. Fixation times were not controlled for and did probably vary. For diagnostic purposes, paraffin sections of about 3 μm were performed and stained H&E. For isolation of tumour-specific DNA, paraffin sections of about 50 μm containing only tumour tissue after macrodissection of adjacent normal tissue were collected.
DNA extraction and MSI-testing
DNA extraction of the paraffin embedded tumours was done using a standard protocol with proteinase K digestion and phenol/chloroform extraction [
11], which yielded between 3 and 15 μg DNA from one section. A detailed protocol is provided as part of the supplementary material (additional file
1). In tumour no. 29 DNA was also extracted from the snap frozen tissue using a standard protocol [
12]. As a control we used DNA from peripheral leucocytes of healthy volunteers as previously described [
13]. Tumour microsatellite instability status was determined using the BAT26 microsatellite marker following a standard protocol [
14]. Tumours were defined as having high-frequency microsatellite instability (MSI-H) if change of length mutations were detected in BAT26 when compared with DNA from colon cancer cell lines with known MSI-status. Because of the quasi monomorphic nature of the BAT26 polyA tract (size variation is uncommon between germline alleles), this marker can be used to screen initially for MSI without matching normal DNA [
15].
Array production
Two different types of microarrays were used. DNA from all tumours was hybridized to a chip previously described by members of our group [
13] containing 644 DNA targets. For the fine mapping of a selected gain on chromosome 13q in one tumour we used an additional chip containing 6400 DNA targets. These comprise 3200 clones with a genome-wide resolution of 1 MB (Sanger-Institute, Cambridge, Great Britain) and 2800 additional clones of regions showing frequent genomic alterations in cancer or that are known to harbor oncogenes and tumour suppressor genes. Lists of the clones included in the arrays together with their genomic localization are provided as supplementary material (additional files
2 and
3). Clone preparation and spotting was done as previously described by our group [
6,
13].
Array hybridization
Tumour and control DNA (each 250 ng) were labelled with Cy3 and Cy5-conjugated dCTP by random priming. Each DNA sample was labeled separately with both fluorophores and used for independent array-hybridizations ("Colour-switch"). Labelled tumour and control DNA was hybridized to the chip. Detailed protocols are available as supplementary material on the journals webpage (additional file
4).
Data Acquisition and Evaluation
Images of fluorescence signals were acquired by a dual laser scanner (GenePix 4000 A, Axon Instruments, Foster City, CA). Assessment of fluorescence signal intensities was done using GENEPIX PRO 4.0 imaging software. To identify imbalanced genomic sequences we used a specialized algorithm previously developed and validated by our group [
9]. Array experiments were declined for further analysis if the R value of the Gauss fit did not reach 0.96 (see [
9]). In contrast to this protocol, we made no grading of imbalances: all imbalances were described as gains or losses, simple gains were not differentiated from amplifications.
Fluorescence in-situ hybridization (FISH)
Fluorescence in-situ hybridization (FISH) was performed as previously described for metaphases [
16] and interphases [
17]. Two-color FISH to 2 μm-sections of the formalin-fixed samples was performed with the Rhodamin-labelled BAC-clones RP11-89P22 (27.7 MB) and RP11-8C15 (21 MB) with corresponding FITC-labelled BAC-clone RP11-9F13 (40.7) as a control. To verify their location on 13q all these probes were hybridized with leucocytes of healthy volunteers before interphase FISH. The gain was shown in relation to the control by visual analysis of the hybridization signals. Criteria for gene amplification were: tight clusters of signals in multiple cells or at least three times more test probe signals than control signals per cell in > 10% of the tumour cells (a minimum of 100 cells were counted).
Discussion
The presented data demonstrates that it is possible to extract DNA of sufficient quality to perform matrix-CGH from the majority of archival formalin-fixed, paraffin-embedded tumour samples collected at different institutions. This reflects the situation encountered in most multicentric therapeutic clinical trials where ample clinical and follow-up data is available, but tissue samples are either not collected systematically or at least not using a standardized protocol for tissue processing, fixation and embedding. In this study, 66% (19/29) of the tissue blocks, irrespective of the institutional origin or the age of the blocks, delivered DNA suitable for array hybridizations employing a simplified method for tumour cell enrichment involving a minimal number of tissue handling steps. The procedure of paraffin embedding and the fixation time have been shown to be able to influence the DNA quality. However, these factors could not be evaluated retrospectively. Further improvements in data quality can be realized with snap frozen samples or with paraffin-embedded tissue, once procedures for tissue processing have been standardized to ensure optimal conditions for molecular analyses and are routinely used in multicentric therapeutic trials to collect tissue. Such an approach is increasingly used by the EORTC-GI group for the PETACC-trials of adjuvant treatment for colorectal cancer patients, and among others the present study was done to verify and optimize tissue sampling protocols for genomic analyses (for example PETACC-4). Our study indicates that it should be technically feasible for the time being to use the large collections of formalin-fixed paraffin embedded tumour tissues for genomic analyses collected by the local pathologists during the routine workup of resected tumour samples of patients included in closed multicentric therapeutic trials. This possibility opens the road for systematic analyses of these archival materials with an excellent long-term follow-up to identify prognostic genomic signatures e.g. by the use of matrix-CGH immediately without the need to wait for the follow up of ongoing prospective studies.
Analysis of the array hybridizations revealed consistent regions of copy number changes. Many of the findings were in agreement with those observed previously in conventional CGH [
18]and matrix-CGH [
19‐
21]. investigations, thus confirming the validity of our hybridizations with DNA extracted from archival formalin-fixed, paraffin-embedded tumour tissue collected at multiple centers. This includes the most frequent gains observed in our study over all different tumour stages such as 20q, 8q, 7p, 13q, and 17q, as well as the most frequent losses on 1p, 18q, 8p and 17p.
Since the list of candidate genes involved in these alterations is very large and has been discussed elsewhere [
19‐
21], we will not elaborate in a detailed discussion of individual candidates. The scientific goal of this pilot matrix-CGH study was to identify genomic signatures associated with systemically advanced colorectal tumours. Since only 15% of UICC III patients will benefit from the routinely administered systemic adjuvant chemotherapy [
22], genomic signatures could serve to select patients with a high risk of recurrence who would have the major benefit from an adjuvant treatment.
Genomic profiles from primary tumours of patients with clinical evidence of metastases (UICC stage IV) differed from those of locally restricted tumours (UICC II/III) only by the type of chromosomal imbalances. The average total number of gains and losses per tumour were not different between different tumour stages, which is in agreement with a recent study of Dukes C colorectal tumours using conventional CGH. In this study, Rooney and coworkers found that the number of aberrations was highly variable. 21% of the tumours showed no aberration, whereas 41% displayed 1–8 and 38% 11–20 aberrations [
23]. Interestingly, in their study patients with more than 2 aberrations appeared to have a better survival than patients with fewer regions of losses and gains.
In contrast, we found a number of imbalances that were more frequent in the UICC IV tumours. The most frequent imbalances were loss on 18q (100% UICC IV vs. 58.3% UICC II/III) and gain on 13q (85.7% UICC IV vs. 58.3% UICC II/III). Loss on 18q included the well known 18q21.1 area which is known to harbour the tumour suppressor genes
SMAD2 [
24] and
SMAD4 [
25], which function in the TGFβ-pathway e.g. to mediate the growth-inhibitory effects of this cytokine. Allelic loss at 18q21.1 and mutations of
SMAD4 are common alterations in colorectal cancer [
26]. Loss of 18q has as well been found in the recently published array-CGH analyses of colorectal cancer [
19‐
21], though an association to more advanced stages has not yet been reported in these studies. LOH-analyses have shown that 18q allelic loss is a strong predictive factor [
27]. Moreover, Dukes C tumours with SMAD4 expression show a significantly longer disease-free survival [
28] and significantly more benefit from 5-fluorouracil-based chemotherapy [
29].
Since at the time when we started this study little detail was available concerning the 13q gain, we decided to further characterize this alteration by high-resolution mapping using high-density genomic arrays and interphase FISH-analyses. Interphase FISH showed the typical pattern of a high-level chromosomal amplification. After finishing the experimental part of the project, a number of manuscripts were published describing genomic profiles of colorectal tumours obtained with matrix-CGH [
19], all confirming the high incidence of 13q gains. However, the area described in these studies was very large and was covered by more than 20 BAC clones in most studies [
20,
21]. Our high resolution mapping allowed us to define the distal border of this gain to the area between 30.01 and 30.67 Mb. However, the aberration involved all proximal BAC clones available on the high density array, thus still leaving a chromosomal area of at least 11.51–11.17 Mb (between 18.50 MB and 30.01–30.67) as minimally amplified region. This area includes numerous candidate oncogenes such as
FLT3, a tyrosine kinase receptor in which activating mutations have been found in acute myeloid leukemia [
30],
FLT1, a vascular endothelial growth factor receptor found to be expressed in gastric and breast carcinoma cells [
31], and
FGF9. FGF9 is a potent mitogen that stimulates normal and cancer cell proliferation [
32] and appears to be involved in the pathogenesis of a number of tumours such as prostate cancer [
33], melanomas [
34], brain tumours [
35], and breast cancer [
36].
Competing interests
The author(s) declare that they have no competing interests.
Authors' contributions
HF was involved in the design of the study, carried out the array hybridizations, data Acquisition and Evaluation, the FISH-experiments, the MSI-testing and drafted the manuscript.
BR was responsible for the array production.
JS performed the macrodissection of the tumour specimen.
MB was involved in the FISH experiments and in drafting the manuscript.
DEA, CJ, BN and CB carried out tumour sampling and provided information about the study samples.
FR provided tumor samples, performed the DNA-extraction of the fresh frozen tissue and was involved in the design of the study.
EVC participated in the design of the study.
CHK was involved in tumor sampling and participated in the design of the study.
HAK was involved in data acquisition and evaluation.
CS was involved in the array production.
MN participated in data acquisition and evaluation.
MPL participated in the design of the study and drafting the manuscript.
PL participated in the design of the study, in array production and evaluation.
TMG conceived the study and participated in its design and coordination and in drafting the manuscript.
All authors read and approved the final manuscript.