METHODS

DNA microarray experiments

The Gene Logic® (Gaithersburg, Maryland, USA) database of Affymetrix® HG-U133 GeneChip® probearray data was screened for probes that corresponded to the VEGF-A and CA9 mRNA sequences (Genbank accession numbers M32977 and NM_001216, respectively). Probes with the highest proportion of samples called “present” were chosen for downstream analysis on 7579 fresh frozen human tissue samples and 225 cell lines. (Gene Logic performed tissue sample preparation and data analysis.) Probeset 210512_s_at was chosen to represent VEGF-A expression and probeset 205199_at to represent CA9 expression.

Tissue microarray experiments

Tissue culture

The human tumour cell lines 786-0 (renal cell adenocarcinoma); A673 (rhabdomyosarcoma); SK-MES-1 (lung squamous cell carcinoma); SW480, KM12, HCT15, HCT116, and COLO205 (all human colorectal adenocarcinomas); SkBr3, MDA-MB-453S, MDA-MB-231, MCF7, and MDA-MB-175-VII (all breast adenocarcinomas); and CALU6 (anaplastic carcinoma) were obtained from the American Tissue Culture Collection (Manassas, Virginia, USA). H322 (lung non-small cell carcinoma) was obtained from the National Cancer Institute (Bethesda, Maryland, USA). Chinese hamster ovary (CHO) cells transfected with human VEGF-A were established as described previously.⁴⁷ Cell lines were cultured in vitro according to the protocols, harvested after the first passage, fixed in 10% neutral buffered formalin, and embedded in paraffin wax.

Selection of human tissues

Formalin fixed, paraffin wax embedded tissue cassettes and corresponding haematoxylin and eosin (H&E) stained sections were retrieved from the histopathology archives at Genentech, South San Francisco, California, USA and the Leeds Teaching Hospitals NHS Trust, UK. (Appropriate ethical approval was obtained for all research conducted using human tissues.) H&E sections were reviewed (KJH) and representative regions of malignant cells were annotated on the slide to aid tissue microarray (TMA) core sampling.

Preparation of synthetic control blocks

cDNA probe templates were generated from human kidney marathon-ready cDNA (BD Clontech, Palo Alto, California, USA). Primers were designed to amplify a 604 bp fragment of VEGF-A mRNA. Sense and antisense primers contained T7 and T3 RNA polymerase initiation sites, respectively (sense, GGATTCTAATACGACTCACTATAGGGCGGGCCTCCGAAACCATGAACT; antisense, CTATGAAATTAACCCTCACTAAAGGGATCCTCCTGCCCGGCTCAC). Each 50 μl polymerase chain reaction (PCR) contained 0.5 ng of cDNA, 33 ng of each primer, 0.6mM dNTPs (0.15mM each dATP, dCTP, dGTP, dTTP), 1× polymerase mix, 1× buffer, and 1.0M GC-Melt (BD Clontech). The thermal cycling conditions were: 94°C for five minutes, followed by 30 cycles of 94°C for one minute and 68°C for 30 seconds. Sense and antisense RNA fragments were transcribed from this template using T7 and T3 RNA Megascript kits, respectively (Ambion, Austin, Texas, USA). Fragments were embedded in 2% agarose at a concentration of 5 μg/ml, and processed as described previously.⁴⁸ Blank 2% agarose controls were processed in an identical manner.

Tissue microarray construction

TMAs were constructed to represent the case series using a Beecher Instruments’ microarrayer (Silver Spring, Maryland, USA), as described previously.⁴⁹ In total, 388 tissues were sampled (three adrenal, 83 breast, two endometrium, three kidney, three liver, 78 lung, 71 ovary, four pancreas, 78 soft tissue, and 63 thyroid), in addition to the cell pellets and synthetic control blocks described above. Tissue sampling was undertaken in triplicate to provide representative data on the parent block⁵⁰; synthetic controls and cell pellets were sampled in duplicate. Sections, 3 μm thick, were cut from the recipient blocks and mounted on to glass slides. H&E staining for verification of the histology (KJH) was performed on the first section cut from each TMA block.

Immunohistochemistry

Tissue sections were dewaxed and heat mediated antigen retrieval was performed in target retrieval solution (Dako Cytomation, Carpinteria, California, USA). Immunohistochemistry (IHC) was carried out as described previously,⁵¹ using a primary antihuman HIF-1α monoclonal antibody at 1 μg/ml, (clone H1α67; Novus Biologicals, Littleton, Colorado, USA). Negative control slides were incubated with a mouse IgG2b immunoglobulin culture supernatant (Dako Cytomation) at 1 μg/ml. Cases were assigned as positive for HIF-1α if one or more of the three cores contained neoplastic cells with positively staining nuclei (KJH).

In situ hybridisation

For each riboprobe to be synthesised, 12 μl (125 mCi) of [α³³P]-UTP (Amersham Biosciences, Piscataway, New Jersey, USA) was speed vacuumed until dry. Each aliquot was reconstituted in 1× buffer, 4.5mM dithiothreitol, 0.23mM rNTPs (0.08mM each rATP, rCTP, rGTP), 2.3μM rUTP, 1.4 U/μl RNAse inhibitor, 0.05 μg/μl cDNA template, and 0.7 U/μl of either T7 (sense probe) or T3 (antisense probe) RNA polymerase (Promega, Madison, Wisconsin, USA). In vitro transcription took place over one hour at 37°C. Samples were then treated with 0.05 U/μl DNase (Promega) for 15 minutes at 37°C and purified over RNeasy mini columns (Qiagen, Valencia, California, USA). Formalin fixed, paraffin wax embedded tissue sections were dewaxed and treated with proteinase K (20 μg/ml in 2× SSC) at 37°C for 15 minutes. Hybridisation and washes were carried out as described previously.^48,52,53

Tissue sections were dehydrated, air dried, and exposed to a phosphorscreen for 16 hours at room temperature. Immediately after incubation, the phosphorscreen was scanned with a Typhoon 9410 (Amersham Biosciences, Piscataway, New Jersey, USA). Slides were then dipped in NBT2 nuclear track emulsion (Eastman Kodak, Rochester, New York, USA), exposed for four weeks at 4°C, developed, and counterstained with H&E. Background subtraction, gridding, and analysis of the phosphorimages were undertaken with Phoretix Array v.3 (Nonlinear Dynamics, Newcastle upon Tyne, UK). Subsequently, the TMAs were reviewed (KJH) by bright/dark field microscopy for verification of hybridisation. Cores were scored semiquantitatively on a scale of 0 (no expression) to 3 (very strong signal), according to the overall intensity of the hybridisation signal.

Statistical analysis

Statistical analysis was performed using SPSS for Windows (version 11.0; Chicago, Illinois, USA). The Mann-Whitney U test was used to assess the differences between the median values of continuous datasets. Large datasets were subjected to false discovery rate controlling procedures as indicated.⁵⁴ Significance was assumed if the two sided p value was < 0.05 and the false discovery rate controlling procedure did not discount the finding.

RESULTS

DNA microarray data

VEGF-A expression

When compared with normal tissue, median VEGF-A expression was significantly greater in primary malignancies of the breast (1.4 fold), cervix (1.4 fold), colon and rectum (2.2 fold), oesophagus (2.0 fold), head and neck (1.6 fold), kidney (3.2 fold), ovary (3.5 fold), skin (1.6 fold), urinary system (3.1 fold), and white blood cells (7.8 fold; including acute lymphocytic leukaemia and chronic myeloid leukaemia spleen samples) (fig 1; table 1). Malignancies of the kidney showed the greatest range and maximum expression of VEGF-A (fig 1). VEGF-A was further upregulated in metastatic colorectal tumours when compared with primary malignancies with no evidence of metastasis (table 1). This association was not seen in other tumour types. In contrast, significant reductions of VEGF-A, relative to normal tissue, were noted in primary malignancies of the lymphoid system (median 0.24 of normal), prostate (0.65 of normal), stomach (0.77 of normal), testis (0.29 of normal), and thyroid (0.44 of normal) (fig 1; table 1).

VEGF-A v CA9 expression

The degree of CA9 expression showed a linear correlation with that of VEGF-A only in renal tissue (R²  =  0.54; fig 2A). The correlation was strongest in neoplastic tissues (R²  =  0.38), with a weaker association in normal kidney (R²  =  0.26), and no correlation in non-malignant disease (R²  =  0.01). Correlations were not apparent in breast, colorectal, lung, ovary, pancreas, prostate, thyroid, or urinary tissues (fig 2B).

Tissue microarray data

VEGF-A expression

To corroborate the Gene Logic data on tissue homogenates, we undertook an investigation of VEGF-A expression by in situ hybridisation (ISH) on a series of human tumour samples. VEGF-A expression was seen in the stromal and/or neoplastic cell populations of positive (score 1–3) TMA cores covering a range of malignancies (fig 3). The ISH score given by the pathologist correlated closely with the quantitative signal obtained by phosphorimaging, which has been used for the remainder of the analysis (fig 4). The VEGF-A antisense control recorded a positive signal by phosphorimaging, when compared with the VEGF-A sense and blank agarose negative controls (fig 5). Of the cell pellets, the VEGF-A transfected CHO cells showed the highest quantitative signal intensity, with lower signals in 786-0, A673, SKMES, SW480, and KM12 cells (fig 5). The remaining cell lines expressed little or no VEGF-A.

VEGF-A v HIF-1α expression

Of 1025 tumour cores, 316 (31%) scored positive for HIF-1α expression in the nuclei of the stromal and/or neoplastic cell populations. All nine cores of renal cell carcinoma and follicular adenocarcinoma of the thyroid were positive, compared with only 5% of breast ductal adenocarcinomas, 4% of mucinous ovarian adenocarcinomas, 6% of gastrointestinal stromal tumours, and no adenocarcinomas of the endometrium or neuroendocrine tumours of the lung (table 2). HIF-1α positive cores showed significantly higher VEGF-A expression than did HIF-1α negative cores in adenocarcinomas of the lung (1.6 fold), serous adenocarcinomas of the ovary (4.5 fold), liposarcomas (20.5 fold), rhabdomyosarcomas (4.8 fold), and medullary carcinomas of the thyroid (1.7 fold) (table 2).

DISCUSSION

The measurement of VEGF-A expression in tumours has become an important methodological issue in determining prognosis,^28–40 and may affect the sensitivity of individual tumours to targeted antibody and small molecule treatments. At present, there is a large body of literature describing the changes in VEGF-A expression during tumorigenesis, using many different techniques, such as ISH,^48,55 IHC,^{30,37,40,56,57} enzyme immunoassay,⁵⁸ reverse transcription-PCR,^36,40 and western blotting.⁴⁰ However, none of the techniques used to measure VEGF-A expression has been widely applied,⁵⁹ and there is no accepted gold standard. This limits the external validity of studies and prohibits reliable interlaboratory comparisons. In addressing this concern, we used both DNA microarray and TMA technology to investigate the expression of VEGF-A in a large series of diverse human tumours. Furthermore, the data yielded information on the proportion of tumours in which the upregulation of VEGF-A expression is associated with markers of hypoxia.

A key aim of our study was to appraise the prevalence and degree of VEGF-A upregulation in a large series of diverse tumour types. DNA microarray analysis demonstrated a significant increase in median VEGF-A expression in many prevalent types of malignancy (fig 1; table 1). The data fall within the range of VEGF-A expression reported for these tumour types in the literature.^28–40,48 In addition, there is a significant increase in VEGF-A expression in colorectal adenomas compared with normal colorectal tissue, which has been reported previously by our group⁴⁸ and by independent investigators.⁶⁰ However, not all tumours upregulate VEGF-A expression at the mRNA level. For example, DNA microarray analysis demonstrated a significant downregulation of VEGF-A expression in benign prostatic hypertrophy and primary malignancies of the lymphoid system, prostate, stomach, testis, and thyroid. Furthermore, the Gene Logic series of metastatic breast cancers did not show a significant increase in VEGF-A expression when compared with primary malignancies or normal breast. These observations are not supported by the literature,^61–68 which may be a reflection of publication bias towards positive findings, demographic differences, or experimental idiosyncrasies. Limitations of the Gene Logic data include low sample numbers for certain tumour types. Moreover, many samples of normal tissue were obtained from tumour resection margins, which may be under the influence of paracrine and hormonal factors secreted by the tumour. For example, in testicular cancer this adjacent tissue frequently shows other histopathological changes. Nevertheless, analyses on subsets of normal tissue of the colon, ovary, breast, lung, and cervix showed no significant differences in VEGF-A expression between tissue from patients with or without cancer (data not shown). This paper also focuses on the expression of VEGF-A mRNA, in contrast to the published literature, which has relied more heavily on immunohistochemistry.^61–66 Although mRNA transcript abundance is not always a true reflection of changes in protein expression, the interpretation of immunohistochemistry is complicated by the specificity and sensitivity of the antibody used. In addition, it is possible that tumours in which VEGF-A is not upregulated trigger the angiogenic switch by either downregulating antiangiogenic factors or upregulating other proangiogenic factors, circumventing their requirement for VEGF-A. For example, angiopoietin-2, a proangiogenic ligand, is upregulated in thyroid cancers and thrombospondin-1, an antiangiogenic extracellular matrix component, is downregulated.⁶¹

“DNA microarray analysis demonstrated a significant downregulation of vascular endothelial growth factor A expression in benign prostatic hypertrophy and primary malignancies of the lymphoid system, prostate, stomach, testis, and thyroid”

In the canonical model outlined above, the expression of VEGF-A is a result of increased transcriptional activity in response to HIF-1α activation under hypoxic conditions.⁶ However, the data presented here indicate that a large proportion of malignancies exhibit VEGF-A upregulation in the absence of these markers of hypoxia. Although the canonical model holds true in some tumours, it seems short sighted to view it as the only means of VEGF-A upregulation in human cancer. In adenocarcinomas of the kidney, the relatively strong association of HIF-1α and CA9 with VEGF-A may result from reduced HIF-1α turnover as a result of mutations or promoter methylation of the VHL gene, reported in 56% of sporadic clear cell carcinomas.⁶⁶ A similar association between HIF-1α and VEGF-A expression was seen for lung tumours, serous adenocarcinomas of the ovary, liposarcomas, rhabdomyosarcomas, and medullary carcinomas of the thyroid. However, the last three tumour types were not well represented in our study and our findings need to be confirmed in a larger series. Furthermore, we suggest that many tumours may exploit hypoxia independent mechanisms to increase their expression of VEGF-A, which is also regulated by certain growth factors and/or hormones known to be manipulated during tumorigenesis.^25–27

In summary, the data presented here represent the largest single series of human tissues and pathologies profiled for VEGF-A expression in a uniform and quantitative manner. The findings may prove useful in establishing a reference for VEGF-A expression in malignant disease.