Tissue microarrays characterise the clinical significance of a VEGF-A protein expression signature in gastrointestinal stromal tumours

A tissue microarray analysis of 22 proteins in gastrointestinal stromal tumours (GIST), followed by an unsupervised, hierarchical monothetic cluster statistical analysis of the results, allowed us to detect a vascular endothelial growth factor (VEGF) protein overexpression signature discriminator of prognosis in GIST, and discover novel VEGF-A DNA variants that may have functional significance.

The clinical behaviour of gastrointestinal stromal tumours (GIST) is notoriously difficult to predict. The prognostic and therapeutic significance of KIT mutations is somewhat contradictory (Ernst et al, 1998;Lasota et al, 1999;Moskaluk et al, 1999;Taniguchi et al, 1999;Lasota et al, 2000;Hirota et al, 2001;Wardelmann et al, 2002;Koay et al, 2005). Therefore, it appears that new molecular indicators of prognostication are needed. Tissue microarrays (TMA) is a high-throughput method for the analysis of large numbers of formalin-fixed, paraffin-embedded (FFPE) materials with minimum cost and effort (Kononen et al, 1998). Here, we applied the TMA technology to analyse protein expression in GIST. The results were analysed with an unsupervised, hierarchical monothetic cluster statistical method. Those biomarkers with strong clinical significance were tested for mutation status by both PCR-denaturing high performance liquid chromatography (DHPLC) and direct sequencing. By doing so, we identified a VEGF-A protein overexpression signature as a statistically significant predictor of malignancy, discovered VEGF-A ligand DNA variants in GIST, and provided other possible targets in future design of anti-VEGF-directed therapy against GIST.

MATERIALS AND METHODS
We used 50 archival paraffin blocks (Department of Pathology, National University Hospital, Singapore), including 15 cases of GIST with a benign outcome, 17 with a malignant outcome (13 primary neoplasms and four metastases), 10 with no available clinical follow-up, and eight gastrointestinal mesenchymal neoplasms other than GIST, such as leiomyoma (n ¼ 5), leiomyosarcoma, neurofibroma and schwannoma (one of each). The mean clinical follow-up was of 39 months. The overall clinico-pathological characteristics are summarised in Table 3. No chemo or radiotherapy was given to these patients. All the gross and histopathological parameters classically associated with malignant potential were analysed. The findings were similar to those reported in other series (data not shown) and, in themselves, are considered insufficient for single-case prognostication in the clinical setting.
After case review for diagnostic confirmation, the TMA was constructed as reported elsewhere (Zhang et al, 2003a;Salto-Tellez et al, 2004). The 22 antibodies used are 34 BE12, AE 1/3, Bcl-2, CAM 5.2, CD10, CD117, CD34, c-erbB2, CK7, CK20, Desmin, Flk-1, Flt-1, Hep Par1, Ki-67, MNF 116, p53, PCNA, S100, SMA, VEGF-A and Vimentin. Table 4 indicates the antibodies and their technical specifications. In general, these antibodies can be divided into several groups: diagnostic markers, antibodies expressed in a specific differentiation pathway relevant to GIST, proliferative or apoptosis-related markers, angiogenic proteins, and others that may have been associated before with prognostic significance in GIST. The interpretation of the IHC staining results for TMA was confirmed by three independent observers (NME, LCK and MST). Results were interpreted based on previous published experience for each individual antibody.
The concordance between TMA and full sections, tested for five antibodies (Table 5) ranged from 92 -100% in five of six antibodies, excluding S100 (71%), in concordance with previous published results (Zhang et al, 2003a).
The 28 FFPE cases with available clinical follow-up were the subject of genomic DNA extraction (GENTRA DNA Purification Kit -Gentra, Minneapolis, MN, USA), according to the manufacturer's instruction. Mutation analysis was performed by PCR-DHPLC analysis. Briefly, DNA was amplified in 25 ml reactions containing 2 ml DNA template, 1 ml of each forward and reverse primers (10 mM each), 0.5 ml of 10 mM dNTP, 0.2 ml FastStart Taq (Roche, Mannheim, Germany), and 1 Â PCR reaction buffer with MgCl 2 . Primer sequences and cycling conditions are indicated in Tables 6 and 7. The PCR product (8 ml) was denatured at 951C for 5 min followed by gradual re-annealing to room temperature for over a period of 1 h. DHPLC was performed using a fully automated WAVE 3500HT system (Transgenomic, Omaha, NE, USA). The cooled samples were automatically injected into a Table 1 Indication of the immunohistochemistry results based on the groups from the hierarchical cluster analysis (see Figure 1), and highlighting the VEGF protein expression signature Non-coding variants are denoted as 'IVS, exon, nucleotides from exon start: nucleotide change'. concentrations and oven temperatures for optimal heteroduplex separation under partially DNA denaturation was determined using the WAVE Navigator software followed by empirical adjustment. Amplicons from the HeLa cell line were included in each run as a wild-type reference.
Samples showing a dHPLC aberrant elution profile were reamplified and sequenced in both directions. Direct sequencing was performed on the ABI PRISM Model 3100 DNA sequencer (Applied Biosystems, Foster City, CA, USA), using the same primers as were used for amplification. Sequencing reactions were conducted with the ABI PRISM BigDye Terminator v3.1 Cycle Sequencing Kit (Applied Biosystems) according to the manufacturer's instructions.
The monothetic cluster analysis was carried out as reported elsewhere (Zhang et al, 2003b) Significance tests included the student's unpaired t-test (2-tailed) for numerical variables and the Fisher's exact probability test for categorical variables. Significance value for P was taken to be Po0.05.

RESULTS
VEGF protein expression signature and its prognostic significance in GIST Table 1 shows the whole protein expression results. Figure 1 shows the cluster diagram obtained upon monothetic hierarchical cluster analysis, including IHC of representative cases. From the cluster analysis, two main groups emerged, based on reactivity for the VEGF-A ligand antibody. Group 1 includes all the VEGF-A ligand expressing cases; out of the 20 GISTs with known clinical outcome, 15 were malignant (75%). In group 2 (VEGF-A negative), only 2/11 of the cases (18%) had a malignant outcome. The difference was statistically significant (P ¼ 0.003). Within group 2, the two malignant cases are further subclassified into a cluster arm, which is positive for flt-1, a receptor for VEGF. Hence, all 17/17 malignant cases were positive for either VEGF-A ligand or the VEGF-A receptor, flt-1, as compared to 8/15 of the benign cases (P ¼ 0.002). In all, 13/17 malignant cases were positive for both these markers as compared to 4/15 benign cases (P ¼ 0.006). Indeed, concomitant expression of VEGF ligand and VEGF receptor

New VEGF-A variants are discovered as a result of mutation analysis
Those GIST samples with known clinical follow-up underwent genomic analysis. In view of the evidence of KIT mutations in GIST and their possible prognostic value (as well as their relation to imatinib therapeutic response) (Lasota et al, 1999;Heinrich et al, 2003), exons 9, 11, 13, 17 of KIT (which are those related to prognosis in the literature) were analysed in the same methodological manner. The results are summarised in Table 2. Variants identified included non-coding IVS1 -7:C-T changes in five (18%) samples (Figure 2), IVS4 -28:C-T changes in seven (25%) samples and coding codon 48A:G-T (Q-H) and codon 91A:G-A (G-D) changes in one sample each (Table 2). A total of 12 (43%) cases had variants in KIT, all in exon 11 (Figure 3). VEGF IVS4 -28:C-T variants were more frequent in samples with low (5/7, 71%) than high (2/7, 29%) VEGF-A expression. The VEGF codon 48 and 91 mutants were present in samples with high Group 1 Figure 1 In red are the study cases with malignant behaviour, in blue are those cases with benign behaviour; cases without available follow-up and non-GISTs are in black. The TMA immunohistochemistry results are included. The asterisk indicates cases reflected in the photomicrographs. Other abbreviations are similar to those described in Table 1. VEGF1 is equivalent to VEGF-A in this figure.
VEGF-A expression, KIT mutations and of a malignant phenotype.
KIT mutations were more frequent in samples with high (10/22, 46%) than low (2/6, 21%) KIT expression. Nevertheless, none of the associations between sequence variants and the expression of their respective proteins, or with the presence of each other, were significant, presumably due to the limited number of samples in this series. The only parameter significantly associated with malignancy in these selected 28 cases was, as expected, VEGF-A protein expression (P ¼ 0.020). Of interest, there was no association with exon 11 KIT mutations and survival in our series.

DISCUSSION
The uncertain prognosis of GIST, both before (Nilsson et al, 2005) and after (Kosmadakis et al, 2005) imatinib treatment, indicate the need for the search of other molecular prognostication biomarkers.
GIST are highly vascularised neoplasms and VEGF-A is a major antiangiogenic therapeutic target (Ferrara and Kerbel, 2005). Recently, anti-VEGF-A therapy has been successful in the treatment of GIST (Marx 2005), with drugs such as Sutent and Sorafenib. Our results indicate that a combined VEGF-A ligandreceptor protein expression signature is a determinant of clinical behaviour in GIST. This is obvious in our study because (a) there is a relation between protein overexpression of the VEGF-A ligand and Flt-1 proteins and benign/malignant behaviour; (b) novel variants in the VEGF-A ligand gene are characterised, some of which appear related to a malignant behaviour (such as VEGF-A exon 3); and (c) in general, these VEGF-ligand variants localise to areas of the VEGF protein with functional significance. The role of flt-1 in this context is unclear; it could be related to the induction of metalloproteinases (Hiratsuka et al, 2002), or to chemotactic signals (Wey et al, 2005).
The role of the detected VEGF-A ligand variants in protein overexpression and GIST tumorigenesis can only be a matter of speculation, based on the scant information available. The IVS4 -28:C-T variant is also identified in phenotypically normal gastrointestinal tissue, thus may not be relevant. The two other variants, however, may have functional implications. The IVS1 -7:C-T variant lies within a GC box that binds the transcriptional repressor protein methyl CpG binding protein-2 (Lapchak et al, 2004), and was found to be associated with higher levels of VEGF mRNA in colorectal cancer (Yamamori et al, 2004), increasing the risk of liver metastasis and worsening its prognosis. In addition, two missense mutations (unreported to date) were discovered in exon 3, coding codon 48A:G-T (Q-H) and codon 91A:G-A (G-D), both in malignant GIST and both showing VEGF-A ligand protein overexpression. In any case, the evidence points to the novel hypothesis that VEGF-A ligand mutations may play a role if the biology and prognosis of GIST.
There has been a previous suggestion that VEGF-A ligand protein expression may be related to prognosis (Takahashi et al, 2003). However, the strength of our unsupervised hierarchical cluster analysis, comparing the expression of an antibody in the context of another 21 biomarkers, delineates 'biological groups' and establishes more complete 'prognostic signatures', which in our study, shown the importance of including protein expression of both VEGF-A ligand and flt-1 receptor in the characterisation of malignant behaviour.