MATERIALS AND METHODS

Tumours

Twenty five GISTs were included in the study. All tumours were immunohistochemically verified as c-KIT-positive, and histologically were in the spectrum of GIST.³ GISTs were divided into three groups by expected biological potential.⁴ At diagnosis, eight tumours were classified as benign based on the mitotic rate not exceeding 5 mitoses per 50 high power field (HPF), and tumour size not exceeding 5 cm (gastric location) or 2 cm (intestinal location). Six tumours were classified as having uncertain malignant potential (UMP) based on tumour size more than 5 cm but not greater than 10 cm (gastric) or more than 2 cm but not greater than 5 cm (intestine). Eleven tumours were classified as malignant based on mitotic activity of more than 5 per 50 HPF or size greater than 10 cm (stomach) or 5 cm (intestine). Details and clinical follow up data are shown in table 1.

Total RNA was extracted from diagnostic samples using Qiagen RNA maxiprep (Qiagen, GmbH, Germany) following the procedure recommended by the manufacturer. Total RNA was treated using Qiagen RNase-Free DNase set (Qiagen) to eliminate any contaminating DNA. The integrity of the RNA was verified on 1% agarose gel electrophoresis.

cDNA array hybridisation

Six malignant and three benign GISTs were included in the gene array analyses (GIST samples Nos 1–9). Expression of 1174 genes was evaluated in each sample (Atlas Human Cancer 1.2K; Clontech Laboratories Inc., Palo Alto, California, USA). Total RNA (3–4 μg) was converted to cDNA and labelled with ³³PdATP using a Clontech cDNA array labelling kit. Hybridisations and washes were performed following the recommended manufacturer’s protocol with minor modifications. Hybridisation images were obtained with a phosphoimager (Bio-Imaging Analyzer, BAS-2500; Fuji, Kanagawa, Japan). Gene expression analyses were performed using Atlas image analysis software (Clontech). The data were viewed using TreeView software (Michael Eisen, Stanford University, California, USA). Each malignant tumour sample was compared with three benign tumour samples (GIST Nos 1–3) to obtain overexpression ratios. Any two compared samples were normalised using nine housekeeping genes that are printed on the array, and a normalisation coefficient that was calculated for each comparison was used to correct signal intensities. Based on our previous studies⁹ and the present report, analysis of any two hybridisation controls obtained from a given sample showed no differences using a ratio of 1.5 for overexpression. However, to increase the stringency of our analyses, we used a value of ⩾2.5 for overexpression. Reproducibility of the results was confirmed using three malignant samples that were hybridised twice, each in a different membrane and in a different experiment.

Quantitative real time RT-PCR

Single stranded cDNA was synthesised using AdvantageTM RT-for-PCR (Clontech). Quantitative real time reverse transcription-polymerase chain reaction (RT-PCR) was performed using iCycler (Biol-Rad, Hercules, California, USA) and threshold cycle number was determined using iCycler software version 3.0. Reactions were performed in triplicate and threshold cycle numbers were averaged. Gene specific primers for ezrin (villin 2 (VIL2)), collagen 8 alpha 1 subunit (COL8A1), G2/mitotic specific cyclin B1 (CCNB1), high mobility group protein (HMG2), TSG101 tumour susceptibility protein, CENP-F kinetochore protein, protein tyrosine kinase 2 (FAK), and protein kinase DYRK2 were designed. The primers used for RT-PCR were obtained from GeneLink (Hawthorne, New York, USA). Gene expression results were normalised to two housekeeping genes (β-actin (ACTB) and hypaxanthine phosphoribosyl-transferase I (HPRTI)). Malignant and UMP GISTs were compared with benign GISTs to obtain an overexpression fold. The overexpression fold was calculated as described previously.¹⁰ Gene overexpression was considered at ⩾5.0-fold based on our previous experience.¹⁰

Statistical methods

T tests were used to examine possible differences in mean gene expression for the nine genes. A natural log transformation of (gene expression+1) was used to stabilise variance. Bivariate correlations were used to examine the relationships between gene expression values for all possible gene pairs. Statistical tests were assessed at a significance level of α = 0.05. Pearson correlations were computed for all possible genes to identify possible correlations between genes.

RESULTS

cDNA array hybridisation

Analyses of nine GIST samples (GIST Nos 1–9) using cDNA arrays showed that the c-KIT oncogene was expressed in all GIST samples, irrespective of histopathology. Comparison of malignant GISTs with benign GISTs detected changes related to tumour progression and revealed 27 overexpressed genes in malignant GISTs compared with benign GISTs (fig 1). These genes included proliferation markers, cell cycle regulators, and several kinases. From the 27 overexpressed genes in the array analyses, we found that eight genes were overexpressed in the majority of malignant tumours (five or more) compared with benign tumours. Thus these genes are likely the most consistent with regard to tumour progression. These genes included VIL2, COL8A1, CCNB1, HMG2, TSG101 tumour susceptibility protein, CENP-F kinetochore protein, protein tyrosine kinase 2 (FAK), and protein kinase DYRK2 (table 2).

Quantitative real time RT-PCR

Quantitative real time RT-PCR using gene specific primers for the eight genes (VIL2, COL8A1, CCNB1, HMG2, TSG101, CENP-F, FAK, and DYRK2) confirmed the gene expression array findings, as shown in table 2. There were few differences in real time RT-PCR expression folds compared with the expression array ratios. These differences were related to the higher sensitivity of real time RT-PCR in quantifying fold changes whereas the array approach is, at best, a ratio based semiquantitative approach.

To test these eight genes in a clinical setting, we obtained diagnostic samples of 16 additional GISTs (GIST 10–25) that were classified, at diagnosis, as benign, malignant, and UMP GISTs (table 1). On performing quantitative real time RT-PCR, using the same eight genes, we noted remarkable gene overexpression in all five malignant GISTs (fig 2). In contrast, gene expression of the six UMP GISTs suggested two different groups that corresponded to the follow up data that were later obtained (table 1).

Statistical analyses of gene expression levels in all GIST tumours (table 3) indicated that six genes were significantly overexpressed in malignant tumours: VIL2 (p<0.001), CCNB1 (p<0.001), HMG2 (p<0.001), TSG101 (p<0.02), CENP-F (p<0.001), and FAK (p<0.03). Furthermore, computing Pearson correlations revealed significant correlations between overexpression of several gene pairs in malignant GISTs. We found the strongest correlations (ρ>0.70) among the significant correlations (p<0.01) between CCNB1-CENP-F (ρ = 0.87) and CCNB1-FAK (ρ = 0.73). Further assessments of associations were not possible due to the relatively limited sample.

DISCUSSION

The gene expression array analyses revealed that the c-KIT oncogene was overexpressed in all GISTs, irrespective of histopathology, which is in agreement with our immunohistochemistry staining results. We developed our analyses based on comparing malignant tumours with benign tumours to detect those changes that may be related to tumour progression as it was not feasible to analyse the true normal cell of origin—“stem cells of Cajal”. In our analyses we used cases with verified long follow up data in order to minimise artefacts. We found 27 genes that were variably overexpressed in two or more malignant GISTs compared with benign GISTs. These genes included several tyrosine kinases, MAP kinases, growth factors, and cell cycle regulators (fig 1). Therefore, the known functions of these genes support their overexpression in tumours with malignant behaviour.

Eight genes were frequently overexpressed in malignant GISTs. These included VIL2, COL8A1, CCNB1, HMG2, TSG101, CENP-F, FAK, and DYRK2. Quantitative real time RT-PCR with gene specific primers and statistical analyses of gene expression levels in all GISTs (table 3) indicated that six of these genes were significantly overexpressed in malignant tumours: CCNB1 (p<0.001), CENP-F (p<0.001), FAK (p<0.03), HMG2 (p<0.001), TSG101 (p<0.02), and Ezrin (p<0.001). Thus it is most likely that these genes are related to the malignant behaviour of GISTs.

Interestingly, testing gene expression levels in 16 additional diagnostic GIST samples was in agreement with the clinical outcome after several years (fig 2). Three UMP GISTs (GIST Nos 12, 14, 15) had gene overexpression (⩾5 fold) of at least four genes, thus suggesting a true malignant potential. The follow up data of these three patients revealed that all had indeed developed recurrences or metastases (table 1). GIST 15 was diagnosed as UMP and had a mitotic index of 1 at diagnosis. This patient developed aggressive tumour recurrence at 12 months with a dramatic increase in the mitotic index up to >100. This sample showed the highest overexpression fold noted in all cases. GIST 13 had low gene expression levels and continued free of disease for 43 months. GIST 25, who was diagnosed with benign GIST and died a month later from other causes, had low gene expression levels consistent with benign behaviour. Although GISTs 10 and 11 had expression patterns suggesting malignant potential, they have continued without disease to date (table 1). It remains possible that these patients may develop recurrences on longer follow up.

Several of the aforementioned overexpressed genes are involved in signalling pathways relevant to cancer. Ezrin is an important signal transduction protein that undergoes phosphorylation and translocation on stimulation by growth factors. Ezrin is a downstream effector of trafficking PKC-integrin complexes involved in the control of cell motility. Ezrin phosphorylation and translocation are thought to be correlated with cell motility, invasion, and carcinoma metastasis.¹¹¹² During mitosis, CENP-F kinetochore protein is associated with kinetochores from prometaphase until early anaphase¹³ whereas CCNB1 has a cell cycle regulation function and its cyclin kinase activity appears only at the G2 to M transition.¹⁴ CCNB1 is responsible, at least in part, for radioresistance after fractionated irradiation of tumours.¹⁵ Accordingly, overexpression of CENP-F and CCNB1 may be related to the increased mitotic activity during malignant transformation. The enzymatic properties of DYRK related kinases have the ability to catalyse tyrosine directed autophosphorylation as well as phosphorylation of serine/threonine residues in exogenous substrates that are involved in the regulation of cellular growth and/or development.¹⁶¹⁷ Recently, TSG101 has been shown to be essential for the growth, proliferation, and survival of mammary epithelial cells.¹⁸ Overexpression of TSG101 has been shown in human papillary thyroid carcinomas and progression of cervical neoplasia.¹⁹²⁰ Nevertheless, FAK is a member of a growing family of non-receptor protein tyrosine kinases and is centrally implicated in the regulation of cell motility and adhesion.²¹ FAK is involved in tumour progression and its overexpression in subpopulations of tumour cells leads to populations of cells with a high propensity towards invasion and metastasis.²² This observation has been reported in oral, ovarian, and prostate cancer.^23–25 Therefore, we have shown, in malignant GISTs, overexpression of genes that are involved in oncogenesis related multiple signal transduction pathways.

In summary, our results suggest that expression profiling of GISTs may be used as a complementary diagnostic tool for the prognostication of GISTs and for the differentiation between benign and malignant tumours. We further highlighted expression of several genes that may be related to the malignant behaviour of GISTs. Therefore, testing the expression profile of a number of genes may segregate GIST into groups of different tumour behaviour.