|Home | About | Journals | Submit | Contact Us | Français|
Bevacizumab, an anti-vascular endothelial growth factor (VEGF-A) antibody, is used in metastatic colorectal carcinoma (CRC) treatment, but responses are unpredictable. Vascular endothelial growth factor is alternatively spliced to form proangiogenic VEGF165 and antiangiogenic VEGF165b. Using isoform-specific enzyme-linked immunosorbent assay and quantitative polymerase chain reaction, we found that over 90% of the VEGF in normal colonic tissue was VEGFxxxb, but there was a variable upregulation of VEGFxxx and downregulation of VEGFxxxb in paired human CRC samples. Furthermore, cultured colonic adenoma cells expressed predominantly VEGFxxxb, whereas colonic carcinoma cells expressed predominantly VEGFxxx. However, adenoma cells exposed to hypoxia switched their expression from predominantly VEGFxxxb to predominantly VEGFxxx. VEGF165b overexpression in LS174t colon cancer cells inhibited colon carcinoma growth in mouse xenograft models. Western blotting and surface plasmon resonance showed that VEGF165b bound to bevacizumab with similar affinity as VEGF165. However, although bevacizumab effectively inhibited the rapid growth of colon carcinomas expressing VEGF165, it did not affect the slower growth of tumours from colonic carcinoma cells expressing VEGF165b. Both bevacizumab and anti-VEGF165b-specific antibodies were cytotoxic to colonic epithelial cells, but less so to colonic carcinoma cells. These results show that the balance of antiangiogenic to proangiogenic isoforms switches to a variable extent in CRC, regulates tumour growth rates and affects the sensitivity of tumours to bevacizumab by competitive binding. Together with the identification of an autocrine cytoprotective role for VEGF165b in colonic epithelial cells, these results indicate that bevacizumab treatment of human CRC may depend upon this balance of VEGF isoforms.
Solid tumour growth is dependent on the induction of their own blood supply by inducing a proangiogenic state in the tissue environment, regulating this balance between proangiogenic growth factors and antiangiogenic inhibitors (Folkman, 1985, 1995; Boehm et al, 1997). One growth factor that has been shown to be an effective target for antiangiogenic therapy (AAT) is vascular endothelial growth factor-A (VEGF-A). Inhibition of VEGF by humanised monoclonal antibodies has been shown to be effective in increasing the median survival in metastatic colorectal cancer (CRC) when combined with chemotherapy (Hurwitz et al, 2004).
Vascular endothelial growth factor-A is generated by alternative splicing from eight exons within the VEGF-A gene. All isoforms contain exons 1–5 and the terminal exon, exon 8. Exons 6 and 7, which encode heparin-binding domains, can be included or excluded. This gives rise to a family of proteins termed according to their amino-acid number, VEGF165, VEGF121, VEGF189 and so on. Exon 8, however, contains two 3′ splice sites in the nucleotide sequences, which can be used by the cell to generate two families of isoforms with identical length, but differing C-terminal amino-acid sequences (Bates et al, 2002). VEGFxxx, the proangiogenic family of isoforms, is generated by use of the most proximal sequence in exon 8 (resulting from inclusion of exon 8a). The more recently described VEGFxxxb isoforms are generated by the use of a distal splice site, 66bp further along the gene from the proximal splice site. This results in splicing out of exon 8a and the production of mRNA sequences that encode the VEGFxxxb family (Bates et al, 2002). The two resultant families of proteins are of the same length, but with different carboxyl termini. VEGF165b was the first of these exon 8b-encoded isoforms identified and subsequent studies demonstrated the existence of VEGF121b, VEGF183b, VEGF145b (Perrin et al, 2005) and VEGF189b (Miller-Kasprzak and Jagodzinski, 2008).
The functional consequences of this altered C terminus are that VEGF165b homodimers compete with VEGF165 homodimers for binding to their principal receptor, VEGFR-2, at a one-to-one ratio and inhibit endothelial cell proliferation and migration in culture (Woolard et al, 2004; Cebe Suarez et al, 2006). VEGF165b blocks VEGF165-driven angiogenesis in vivo in the rabbit, rat (Woolard et al, 2004), mouse and chick (Cebe Suarez et al, 2006), and human malignant melanomas consisting of cells overexpressing VEGF165b and cells expressing VEGF165 grow slower in nude mice than those consisting of cells expressing VEGF165 alone. Recombinant human VEGF165b is also antiangiogenic in hypoxia-driven angiogenesis in the eye (Konopatskaya et al, 2006).
Although VEGF has been shown to be critical in CRC by inhibition studies, the expression of VEGF in CRC has not been investigated using tools that distinguish between the proangiogenic VEGFxxx isoforms and the antiangiogenic VEGFxxxb isoforms. The vast majority of studies have measured total VEGF levels in plasma, tumours or serum using commercially available antibodies that do not distinguish between pro- and antiangiogenic isoforms, as commercial enzyme-linked immunosorbent assays (ELISAs) detect VEGFxxxb isoforms. Furthermore, it is not known whether the VEGFxxxb isoforms are able to slow or reduce tumour growth if they are highly expressed. To determine whether this antiangiogenic isoform family is expressed in CRC, and how that expression may be regulated, this study compares the balance of expression of the VEGFxxxb family of isoforms in human CRC with paired normal colonic mucosa samples and show how the expression of VEGFxxxb and VEGFxxx is altered during the malignant transformation of colonic adenoma cells in vitro. Furthermore, to determine whether this balance has the potential to regulate tumour growth rates, we have measured VEGF165b functional effects on colon carcinoma growth in animal models where the tumour growth was VEGF-dependent. Furthermore, we assess whether bevacizumab either specifically binds VEGFxxx (proangiogenic) isoforms, or also has cross-reactivity with antiangiogenic VEGF165b and whether it inhibits VEGF165b-expressing tumours.
Paired colon samples were from partial colon resection for carcinoma. Samples were obtained by taking biopsies of the fresh specimen from a nonnecrotic central portion of the tumour and from a peripheral part of the macroscopically normal colonic epithelium (n=18 pairs). Samples were collected with Local Ethics committee approval. The mean patient age was 71.5 (range 58–80) years, with 62% male subjects and Duke's staging as follows: 6.7% A, 46.7% B and 46.7% C. Biopsies were immediately frozen in liquid nitrogen and then stored at −80°C until processed. Biopsies were frozen in liquid nitrogen again immediately prior to manual slicing with a sterile blade. The mass of each tissue was recorded, and samples were homogenised and mRNA and protein extracted as described below.
For each of eight pairs of samples, mRNA was extracted from approximately 200mg of tissue (Chomczynski and Sacchi, 1987) and reverse transcribed as previously described (Bates et al, 2002). The cDNA was amplified using primers complementary to VEGF exon 7 and the 3′-UTR downstream of exon 8b, as previously described (Bates et al, 2002). The products were subjected to standard agarose gel electrophoresis and ethidium bromide staining. Protein was extracted from approximately 250mg tissue, from 18 pairs of samples, resolved by 12% sodium dodecyl sulphate-polyacrylamide gel electrophoresis, transferred and immunoblotted as previously described (Woolard et al, 2004). Briefly, membranes containing recombinant human VEGF165 and/or VEGF165b protein (50ng of each) and protein samples extracted from colon (100μg of each) were probed with mouse anti-VEGFxxxb IgG (2μgml−1 A56/1; R&D Systems, cat no. MAB3045) (Woolard et al, 2004) or rabbit anti-VEGF IgG (1μgml−1 A-20; Santa Cruz Biotechnology, Santa Cruz, CA, USA) and detected with horseradish peroxidase-conjugated stabilised goat anti-mouse or anti-rabbit IgG (1/7000; Pierce Biotechnology, Rockford, IL, USA). Visualisation of protein bands was achieved with SuperSignal West Femto Maximum Sensitivity Substrate Kit (Pierce Biotechnology).
Quantitative PCR assays were carried out on cDNA generated as above. An exon 7b forward primer and a 3′-UTR primer (both pan-VEGF quantitative polymerase chain reaction (Q-PCR) from Primer Design, Southampton, UK) or an exon 7a forward primer 5′-TTGCTCAGAGCGGAGAAAGC-3′and a reverse primer specific for exon 8a that did not detect VEGFxxxb isoforms 5′-TCACCGCCTCGGCTTGTCACAT-3′ were used. Reactions were performed using a SmartCyclerII (Cepheid, Sunnyvale, CA, USA) q-PCR machine, with 25μl reaction volumes comprising 12.5μl Quantitect SYBRgreen 2 × master mix (Qiagen, Crawley, UK), 1μl cDNA and 1μl primer mix cycled at 95°C for 15min followed by 50 cycles of 60°C for 30s, 72°C for 60s, 79°C for 15s (reading) and 95°C for 30s. A melt curve was then performed by ramping the temperature from 60 to 95°C at 0.2°C per second, reading throughout. DNA standards used were VEGF165b or VEGF165 cloned into pcDNA3, or oligonucleotides containing the full sequence between the primers (Primer Design).
Figures 1A and B show examples of the reverse transcription-polymerase chain reaction (RT-PCR) curves for VEGF165b and VEGF165 templates respectively using pan-VEGF primers. Figure 1C shows the standard curve generated from cycle threshold for the two templates, showing that there was no difference in the standard curves (n=3). Thus the efficiency of amplification of the two templates is not different. Figure 1D shows the amplification curves for the exon 8a primers using the VEGF165 sequence as a template (VEGF165b template did not result in amplification until 18 cycles later than equivalent VEGF165 concentration). Total VEGF and VEGFxxx copy numbers were calculated for each sample using the calibration curve shown in Figure 1E. The difference between the total VEGF and the VEGFxxx copy number was assumed to be the VEGF165b copy number.
Immunohistochemistry was performed on formalin-fixed paraffin-embedded normal colonic mucosa, obtained with local ethics committee approval from archival material as previously described (Woolard et al, 2004). Sections were stained with either mouse monoclonal anti-VEGFxxxb IgG (8μgml−1; R&D Systems, cat no. MAB3045) or 5μgml−1 mouse anti-VEGF IgG (Santa Cruz; SC-7269) or a normal mouse IgG (Sigma Aldrich, Gillingham, UK; I8765), as a negative control.
pan-VEGF mouse capture antibody (0.1μg) (Duoset VEGF ELISA DY-293; R&D Systems, Minneapolis, MN, USA) diluted in 1 × phosphate-buffered saline (PBS) (pH 7.4) was adsorbed onto each well of a 96-well sterile plate (Immulon 2HB Thermo Life Sciences, Basingstoke, UK) or, for the VEGFxxxb ELISA, 0.08μg AF293-NA Goat anti-VEGF polyclonal IgG (R&D Systems), overnight at room temperature. The plates were washed three times between each step with 1 × PBS-Tween (0.05%). After blocking with bovine serum albumin in PBS for 30min at 37°C, recombinant human VEGF165 standards or VEGF165b (R&D Systems) diluted in blocking solution (ranging from 62.5pgml−1 to 2ngml−1) or protein sample was added to each well. After incubation for 30min at 37°C with shaking and three washes, 100μl of biotinylated goat anti-human VEGF (50ngml−1 in blocking solution; R&D Systems) or a mouse monoclonal anti-VEGFxxxb-biotinylated IgG (clone 264610/1; R&D Systems) at 0.4μgml−1 was added to each well, and plates left for 30min at 37°C with shaking. Streptavidin-HRP (100μl) (R&D Systems) at 1:200 dilution in PBS was added, plates left at room temperature for 20min and 100μl per well O-phenylenediamine dihydrochloride solution (Substrate reagent pack DY-999; R&D Systems) added, protected from light and incubated for 20min at room temperature. The reaction was stopped with 100μl per well 1moll−1 H2SO4 (10276; BDH Chemicals, Poole, UK), and absorbance read immediately in an Opsys MR 96-well plate reader (Dynex Technologies, Chantilly, VA, USA) at 492nm, with control reading at 460nm.
We used both this antibody (MAB3045) and a separate antibody generated against the terminal nine amino acids of VEGF165b, raised by R&D Systems and biotinylated (clone 264610/1), to quantitate the relative levels of VEGF165b. Both VEGF165b antibodies were as sensitive as the commercially available VEGF ELISA (see Figure 2A). The biotinylated 264610/1 antibody specifically detects VEGF165b, is accurate down to below 62.5pgml−1 VEGF165b and does not detect VEGF165, even at 4ngml−1 in an ELISA (Figure 2B). Both antibodies were used to determine the amount of VEGFxxxb in human tissues from pancreatic islets, placenta, lung, colon and prostate. The two ELISAs did not differ in their results (e.g., colon tissue, 130±40pgmg−1 MAB3045 ELISA 128±20pgmg−1 clone 264610/1 ELISA, N=18). The biotinylated 264610 ELISA was used to quantitate the amount of VEGF in tissue samples, as it was more sensitive, a simpler procedure, could use commercially available capture antibodies and the protocol was most comparable to the commercial VEGF DuoSet ELISA. To determine whether VEGF165 could interfere with this VEGF165b ELISA, serial dilutions of rhVEGF165 were assayed. As seen in Figure 2B, there was no significant change in OD values by the addition of rhVEGF165 as high as 4000pgml−1, indicating that the VEGFxxxb ELISA specifically detects VEGF165b, and is not affected by the conventional VEGF isoform, VEGF165.
To determine whether commercially available ELISAs also detected VEGF165b, we carried out a VEGF ELISA using increasing concentrations of VEGF165b. Figure 2C shows that increasing concentrations of VEGF165b were detected by the R&D Duoset kit – the most widely used ELISA. Interestingly, this ELISA detects VEGF165b at a lower affinity than VEGF165. The ratio of the slopes is 0.42±0.004 or 42±0.4%. To confirm this, we used VEGF165b generated from two different sources – R&D Systems and an in-house production (both proteins were quantitated by Bradford Assay). The R&D Duoset Kit (DY293B) is a second generation ELISA introduced in 2004. We are unaware of published information on sFlt-1 interference in the current DuoSet kit. Figure 2C shows that increasing concentrations of sFlt-1 did not affect the pan-VEGF ELISA at least up to 2000pgml−1 VEGF.
With the previous ELISA, VEGF165b levels were detected at 100% of VEGF165, indicating that the previous ELISA kit had the same affinity for both isoforms. To ensure that this was due to a difference in affinity of the antibodies for the two isoforms, we carried out surface plasmon resonance analysis of binding coefficients.
To compare the binding affinities of VEGF165 and VEGF165b to the pan-VEGF antibody used in the R&D Duoset detection kit, we amine-coupled the latter to a CM5 sensor chip (Biacore AB, Uppsala, Sweden) to an immobilisation level of 630 response units (RU). To compare the binding affinities of VEGF165 and VEGF165b to bevacizumab, the latter was amine-coupled to a CM5 sensor chip (Biacore AB) to an immobilisation level of 580 RU. The coupling was performed using EDC/NHS and 1moll−1 ethanolamine (Biacore) as per the manufacturer's instructions, with the bevacizumab dissolved in 10mmoll−1 sodium acetate (pH 4.5). A blank reference cell was formed by the same activation and deactivation process involved in amine coupling without adding antibody. Samples containing VEGF165 or VEGF165b diluted in HBS-EP sample buffer (Hepes-buffered saline with EDTA and P20 surfactant, Biacore AB) were then run at twofold serial dilutions from 180nmoll−1 down, in random order in duplicate. Injection was performed at 30μlmin−1 for 3min, followed by 6min of buffer only, for monitoring of dissociation. Regeneration between each interaction was performed by injection of 4moll−1 MgCl2 at 20μlmin−1 for 40s, followed by a 2-min period of stabilisation before the next injection. Figure 3A shows the binding curves of VEGF165 to the RnD detection antibody and Figure 3B shows binding of VEGF165b to the same antibody. The RnD detection antibody had a higher association coefficient for VEGF165 than VEGF165b and a lower dissociation coefficient for VEGF165 than VEGF165b (Figure 3C), resulting in an affinity of 602pM for VEGF165, but 3.98nM for VEGF165b, an ~6.6-fold difference in affinity, indicating that the underestimation of the commercial ELISA for VEGF165b was due to a difference in affinity for the antigen.
The actual VEGF concentrations (VEGFtotal) in human tissue are the sum of VEGFxxx and VEGFxxxb.
The commercially available pan-VEGF ELISA has a lower affinity for VEGF165b than for VEGF165 by 42%. Therefore the measured VEGF levels (VEGFmeasured) in the commercially available ELISA are the sum of VEGF165 and 42% of VEGF165b.
We used the above correction to estimate tissue concentration of VEGFxxxb and its relative proportion of total VEGF.
Colonic adenoma cells (AAC1) and their in vitro-derived carcinoma cells (10C) were kindly donated by Professor C Paraskeva (Williams et al, 1990). The adenoma cells are a non-tumour-forming clonogenic variant of the PC/AA cell line derived from a polyp from a patient with familial adenoma polyposis (Paraskeva et al, 1988). The cells were cultured to 100% confluence in Dulbecco's modified Eagle's medium supplemented with 20% fetal bovine serum, 1% penicillin/streptomycin and 0.2% Actrapid insulin (Novo Nordisk, Crawley, UK). Protein was extracted from confluent cells in fresh media as above cells were placed in a hypoxic chamber (Billups-Rothenburg, Del Mar, CA, USA) for 5min at 20lmin−1 with 5% CO2/nitrogen gas mixture (BOC) and incubated at 37°C. The gas was changed twice per day.
LS174t human colon carcinoma cell lines were used (ECACC, Salisbury, UK) (Yuan et al, 1996; Lee et al, 2000). Cells were transfected with 1μg of purified plasmid pcDNA3 either as empty vector, with VEGF165, VEGF165b, or both VEGF165 and VEGF165b vectors, using Lipofectamine Plus (Invitrogen, Carlsbad, CA, USA) as per the manufacturer's instructions. Cells were selected using geneticin (500μgml−1). Conditioned media were analysed by ELISAs for VEGF and VEGFxxxb production per cell per 24h, to confirm expression levels of VEGF isoforms. A total of 2 × 106 cells were injected subcutaneously into the lumbar region of nude mice (six per group unless otherwise stated). Mice were then monitored every 2–3 days and tumour length and width measured. When the first tumour reached 16mm in maximum diameter, all mice were killed. Tumour volumes were calculated according to the formula (length × width × (length+width)/2). Injections, measurements and analysis were all carried out with the investigators blinded to group.
Measurement of the cell-cytotoxicity effects of blocking VEGF isoforms was performed using a lactate dehydrogenase (LDH) assay following exposure of the cells to various antibodies and RTKIs as described in the results. Cells were grown to 90–100% confluence in 96-well plates prior to treatment in serum-free medium, against which treatment was compared. A Cytotox 96 Non-Radioactive Cytotoxicity Assay (Promega Corp, Madison, WI, USA, no. G1780) was used according to the manufacturer's instructions.
Means and standard errors are given unless stated otherwise. Tumour volumes, cell cycle parameters and apoptosis were compared by one-way analysis of variance (ANOVA) followed by a Bonferroni post-hoc test. Tumour growth curves were fitted by nonlinear regression using an exponential curve fit in Prism. Doubling times were calculated from 0.69k−1, and are given as mean (95% confidence intervals (CI)), and curve-fitting parameters compared using an F-test. Analysis of ELISA results was performed using Wilcoxon's signed matched ranks at 95% significance level (two-tailed).
To determine whether VEGF165b and VEGF165 mRNA were expressed in normal and cancerous colon, RT-PCR using primers that distinguish between the two families of isoforms was carried out on eight pairs of samples. Reverse transcription-polymerase chain reaction gave two bands, one at ~135bp, consistent with VEGF165b or VEGF189b, and one at ~200bp, consistent with VEGF165 and VEGF189. This size difference was due to the splicing out of exon 8a in the VEGFxxxb family, resulting in the shorter mRNA (although exon 8b is present in the mRNA of the VEGFxxx family, a stop codon in exon 8a prevents its translation). VEGFxxx and VEGFxxxb mRNA expression was detected in both normal and tumour tissue (Figure 4A).
Quantitative PCR on mRNA extracted from seven pairs of colorectal normal and tumour tissue demonstrated that the VEGFxxx mRNA copy number was only 9.1±2.8% of the total VEGF level in normal tissues, indicating that VEGFxxxb species form more than 90% of the mRNA. There was an increase in copy number of all VEGF isoforms from 5±2.2 to 11±3.5 × 103 copies per μg of tissue (Figure 4B). This upregulation was specific for the VEGF165 isoform (0.18±0.32 to 6.0±2.6 × 103 copies per μg, Figure 4C, P<0.01, ANOVA), and not for the VEGF165b isoform (4.8±2.2 to 5.6±2.0 × 103 copies per μg, Figure 4D), such that in cancers, 45±13% of the VEGF mRNA was VEGF165.
To determine whether protein expression of VEGFxxxb isoforms was present in normal and tumour tissues, western blotting and ELISA were carried out using antibodies specific to VEGF165b and antibodies that do not distinguish between isoforms. The previously characterised anti-VEGF165b antibody (R&D Systems, cat no. MAB3045) demonstrates highly specific binding to recombinant human VEGF165b but not to recombinant human VEGF165 (Pritchard-Jones et al, 2007). Figure 5A shows that both VEGF121b and VEGF165b were expressed in both normal colonic mucosa and in carcinomas as detected by western blotting. To determine which cells in colonic epithelium expressed the VEGFxxxb isoforms, we used immunohistochemistry to stain the normal colon samples. Figure 5B shows that VEGFxxxb was highly expressed in colonic epithelial cells of the brush border (especially cells at the apex of the villi), cells within the lamina propria with plasma cell morphology and goblet cells. Staining of sections with the VEGF antibody that recognises all VEGF isoforms (including VEGFxxxb isoforms, Figure 5C) showed that VEGF was expressed in all the places where VEGF165b staining was apparent. Nonspecific isotype control IgG did not stain (Figure 5D).
To determine whether the previously reported increase in VEGF expression (Ellis et al, 2000; Reinmuth et al, 2003) in CRC was due to changes in VEGFxxxb or VEGFxxx, the levels of VEGF and VEGFxxxb were determined by ELISA and isoform-specific ELISA respectively. Total VEGF was significantly upregulated in CRC (288±48pg VEGF per mg total protein) compared to normal colon (122±23pgmg−1, n=18; P<0.01, Wilcoxon Figure 5E). VEGFxxxb concentrations, however, were no different in tumour extracts (160±50pgmg−1) than in controls (130±23pgmg−1; P=0.96, Figure 5F). Since the total VEGF is the sum of the pro- and antiangiogenic isoforms, the level of VEGFxxx expression was calculated in these samples. The mean VEGFxxx concentration in normal mucosa was not significantly different from zero. In CRC tissue, however, the mean VEGFxxx level was 128±53pg per mg total protein, indicating that the difference in total VEGF expression was due to an upregulation in VEGFxxx production alone. Thus, there was a shift from predominantly VEGFxxxb protein in controls to VEGFxxx in CRC (mean±s.e.m. ratio VEGFxxxb/total VEGF: 112±12% in controls, 59±12% in tumours; P<0.001 paired t-test). Figure 5G shows that the ratio of VEGFxxx to VEGFxxxb in the tumours was highly variable with some patients having no excess of VEGFxxx, and others up to 57-fold excess of VEGFxxx. There was no relationship between Dukes' staging and VEGFxxxb concentration or proportion (P=0.75; one-way ANOVA, data not shown).
As VEGF expression is predominantly proangiogenic in CRC and predominantly antiangiogenic in normal colonic epithelium, cell lines at different stages of the malignant transformation may reflect this. Cells along the adenoma-carcinoma sequence, grown in 100% confluent monolayers, were assayed for VEGF expression as shown in Figure 6. Although the total VEGF expression increased along the adenoma-carcinoma sequence, the degree to which it did so was very variable (Figure 6A) and only the LS174t cells demonstrated a significant such increase. In addition, the VEGFxxxb expression was reduced in the carcinoma cells compared to the adenoma cells, but this only reached significance in the HT29 cell line (Figure 6B). Overall, the effect of a shift from anti- to proangiogenic VEGF expression predominance in CRC cell lines was due to a combination of reduced VEGFxxxb expression and increased VEGFxxx expression, resulting in a shift from 82±10% VEGFxxxb in adenoma cells to between 5.7±0.3 and 53±5.8% in the carcinoma cell lines (P<0.0001, one-way ANOVA, Figure 6C). There was thus a predominance of VEGFxxxb in this adenoma cell line, and a variable switch towards predominance of VEGFxxx in the colonic carcinoma cell lines.
To test whether hypoxic upregulation of VEGF expression induced a switch in the relative expression of pro- and antiangiogenic VEGF required the use of cell lines that produced predominantly VEGFxxxb in normoxia, such as AAC1 and 10C. The exposure of the AAC1 cells to hypoxia resulted in a switch from predominantly antiangiogenic VEGFxxxb (82±10% VEGFxxxb) to predominantly proangiogenic VEGFxxx (65±1% VEGFxxx; P<0.01, Figure 6D). This change in the AAC1 cells was due to an upregulation in the proangiogenic isoforms of VEGF, since total VEGF increased (P<0.01, Table 1) but VEGFxxxb isoforms did not significantly alter. Neither the total VEGF nor the VEGFxxxb expression in 10C cells was significantly altered by hypoxia.
Prior to injection of the tumour cells into the mice, expression of VEGF isoforms was measured both by western blotting and by ELISA. Vascular endothelial growth factor concentrations of the LS174t human colonic carcinoma cell line transfected to overexpress VEGF165, VEGF165b, VEGF165 and VEGF165b (VEGF165/165b), or with the empty expression vector (pcDNA3) are given in Table 1.
Proliferation of LS174t cells after transfection with VEGF165, VEGF165b, both or control vector was measured by flow cytometry. Histogram analysis of propidium iodide-labelled cells showed that there was no significant difference between the cell cycle stage of the groups (control 79±2.3%, VEGF165 80±3%, VEGF165b 83±5% and VEGF165/165b 77±0.7% in G0/G1; P>0.1). Apoptosis and necrosis were also measured by dual PI/Annexin V staining. There were no significant differences between the four groups (data not shown).
To determine whether VEGF165b expression by the tumour cells inhibited tumour growth in vivo, LS174t human colon carcinoma cells stably transfected to express VEGF165, VEGF165b, both isoforms (VEGF165+VEGF165b), or a control vector were injected into nude mice. Figure 7A shows that control-transfected cells formed large vascular solid tumours within 15 days. Tumours formed from cells overexpressing VEGF165b were smaller and less vascularised (Figure 7B) and had a softer texture. Comparison of tumour volumes made by caliper measurements in live animals showed that VEGF165b-expressing cells formed significantly smaller tumours than control cells (P<0.01, Figure 7C). Overexpression of VEGF165 in LS174t cells resulted in large, vascular tumours (Figure 7D). However, injection of cells expressing both VEGF165 and VEGF165b resulted in smaller, paler tumours than VEGF165 alone (P<0.001, Figures 7E and F). VEGF165b tumours (0.33±0.22cm3) were significantly smaller than the VEGF165-expressing tumours (1.61±0.57cm3; P<0.001) at 15 days. Exponential curve fitting to the tumour growth curves was used to calculate the doubling time of the tumour groups. The mean (95% CI) doubling times for pcDNA3, VEGF165 overexpression and VEGF165/165b overexpression were 2.1 (1.9–2.4), 1.9 (1.9–2.1) and 2.4 (2.2–2.6) days respectively. In comparison, the doubling time for VEGF165b overexpression was 3.0 (2.5–3.6) days, which was statistically significantly different (P<0.001) from the other groups. After excision, tumour sections were stained with haematoxylin and eosin, and areas of necrosis quantified (Figure 7G). VEGF165b-expressing tumours had significantly greater areas of necrosis compared with other groups (Figure 7H; P<0.05 ANOVA).
To determine whether bevacizumab bound VEGF165b, a western blot of recombinant human VEGF165 and VEGF165b was carried out by immunodetection with bevacizumab and secondary antibodies to human IgG (Figure 8A). Bevacizumab appeared to detect efficiently recombinant human VEGF165b. To determine the efficiency of binding, and its affinity for VEGF165b, we carried out Biacore analysis of bevacizumab covalently bound to a surface plasmon resonance sensor. Figure 8B shows the association and disassociation kinetics of VEGF165, and Figure 8C those for VEGF165b across this chip (Figure 8D). The ratio of these two gives the overall affinity (KD). The KD was similar for VEGF165 (2.5nM) and VEGF165b (6.8nM).
To determine whether the effect of bevacizumab on tumour growth was dependent on VEGF165b expression, nude mice were injected with either VEGF165b-transfected LS174t colon cancer cells expressing 95% VEGF165b (n=14), or control LS174t cells expressing 94% VEGF165 (n=12). Twenty-four hours after tumour cell injection, bi-weekly treatment with 50μg bevacizumab or saline was started, and tumour sizes measured every 3–4 days. Figure 9A shows that bevacizumab significantly inhibited the growth of VEGF165-expressing colon cancer cells (P<0.05) within 15 days. Figure 9B shows that even after 35 days of treatment, bevacizumab had no effect on tumour growth in cells expressing predominantly VEGF165b. To determine whether the effect of bevacizumab on previously established tumours was modified by VEGF165b, cells were injected as before, and tumours allowed to grow to 4mm in diameter before treatment with bevacizumab as above. Figure 9C shows that bevacizumab inhibited the growth of established tumours compared with saline treatment (P<0.05), whereas Figure 9D shows that it did not affect the growth of VEGF165b-expressing tumours. Figure 9E shows that VEGF165b-expressing tumours grew faster than those not expressing VEGF165b when treated with bevacizumab (P<0.05).
VEGF165b is strongly expressed in normal colonic tissue and by AAC1 colonic adenoma cells and has been shown to be cytoprotective in renal epithelial cells. Therefore, the sequestration of VEGF165b by addition of either a specific anti-VEGFxxxb antibody or a more general anti-VEGF antibody could be cytotoxic. To explore this, AAC1 adenoma cells (normally express ~85% of their VEGF as VEGF165b) were treated with either an anti-VEGF165b-specific antibody (R&D Systems, cat no. MAB3045) or bevacizumab at increasing doses for 48h. Cytotoxicity was measured by assaying supernatant for lactate dehydrogenase. Figure 10A shows that VEGF165b inhibition was toxic to AAC1 cells in a dose-dependent manner, increasing the cytotoxicity by 14.4±1.1-fold (P<0.001). Bevacizumab also significantly increased cytotoxicity 4.9±0.5-fold (P<0.001). A nonspecific IgG (1mgml−1) resulted in a modest 2.2±0.25-fold increase in cytotoxicity over baseline in AAC1 cells (P<0.01). Much smaller increases in cytotoxicity were seen when LS174t colonic carcinoma cells were treated with the anti-VEGF165b antibody (4.0±0.35-fold) or bevacizumab (3.2±0.6-fold) (Figure 10B, no increase seen with nonspecific IgG). To confirm that the cytotoxicity effects of anti-VEGF antibodies were due to inhibition of VEGFxxxb, the effects of supplementing the media with rhVEGF165b were measured (Figure 10C). Addition of 40ngml−1 rhVEGF165b protein abolished cytotoxicity induced by the VEGF165b antibody (P<0.01, one-way ANOVA).
To evaluate whether the VEGF165b required for AAC1 cell survival was acting through VEGFR1 or VEGFR2, cells were treated with receptor tyrosine kinase inhibitors to selectively inactivate each of these receptors. Selective inhibition (Glass et al, 2006) of either VEGFR1 (10nM SU5416) or VEGFR2 (200nM ZM323881) was not toxic to the AAC1 cells (Figure 10D), but combined inhibition of both receptors induced a significant increase in cytotoxicity (1.8±0.29-fold; P<0.05, one-way ANOVA), an effect that could be rescued by the addition of 100ngml−1 of either rhVEGF165 or rhVEGF165b protein.
Vascular endothelial growth factor has been identified in thousands of studies as being altered in tumours, and able to affect tumour growth. VEGF165 was originally identified from tumours and tumour cells, showed angiogenic and propermeability activity and was generated from the sequence now described as exon 8a. In 2002, we described VEGF165b, encoded by an alternative sequence in exon 8 (exon 8b), resulting in a protein of identical length but different amino-acid sequence to that encoded by exon 8a. This different C terminus is also found in other VEGF isoforms, resulting in a family of VEGFxxxb splice variants (Perrin et al, 2005) that is expressed in many normal tissues and mirrors the conventional, angiogenic VEGFxxx isoforms.
Homodimeric VEGF165b is a competitive antagonist of VEGFR2, binding to it with affinity equal to that of VEGF165 (Woolard et al, 2004). Moreover, VEGF165b is not angiogenic, and can act as an antiangiogenic agent in VEGF-mediated angiogenesis, such as in the eye and in the mesentery (Woolard et al, 2004). We show here that these antiangiogenic isoforms of VEGF are expressed as mRNA and protein in both normal colonic epithelial cells and colonic carcinomas. Not only is their expression dominant in normal colonic epithelium, but also remains relatively constant in the carcinomas. Many previous studies have identified VEGF upregulation in colon carcinomas either without distinguishing between the two families of isoforms or by methods that only detect the proangiogenic isoforms. Previously described total VEGF levels seen in normal and CRC samples are similar to those described here (40 and 220pgmg−1, respectively (Konno et al, 1998)). However, this is a significant underestimate of total VEGF, as the commercially available ELISAs for VEGF have an affinity for VEGF165b of only 42% compared with VEGF165. Thus, a more accurate estimation of VEGF concentrations in tissues requires a correction for this affinity. Here we show that the measured amount of VEGFxxx, but not VEGFxxxb, is increased in CRC, supporting the idea that there is a proangiogenic switch involving upregulation of both VEGF overall and, crucially, of only the proangiogenic splice variants.
However, our results show that the VEGF165b levels in some tissues (approximately half of the normal samples) exceed those of the total VEGF levels, measured by the pan-VEGF ELISA, even after adjustment for the poor affinity of the pan-ELISA for VEGF165b. There are a number of possible reasons for this difference, although none has been clearly proven, mainly relating to the lack of accuracy of the commercial pan-VEGF ELISA when the VEGFxxxb isoforms are considered. These include (a) endogenous heterodimerisation, (b) other isoforms and (c) interference by other as yet unknown proteins and fats. Endogenous heterodimerisation may yield intermediate forms such as VEGF165b:VEGF165 or VEGF121b:VEGF165b. The affinity of the pan-ELISAs for heterodimers has not yet been shown, but may be intermediate between the two homodimers. If half the VEGF165 is dimerised with VEGF165b, then the true total VEGF value may be still higher. Other isoforms of VEGFxxxb exist (Perrin et al, 2005), but the affinity of the ELISAs for these (particularly the commercially available pan-VEGF ELISA) has not been measured. If the affinity of the commercial pan-VEGF ELISA for VEGF121b or VEGF121 was less than that for VEGF165b or VEGF165, then this would also result in an underestimate of the total VEGF levels. Finally, the soluble splice variant of VEGFR1 (sFlt-1) has been shown to inhibit detection of VEGF in previous commercially available pan-VEGF ELISAs (Maynard et al, 2003). We showed that sFlt-1 does not interfere with the current R&D pan-VEGF ELISA, ruling this out as an explanation for the discrepancy, but other proteins, such as soluble VEGFR2, may affect binding. We show here that VEGFxxxb is localised to the same regions of the colonic mucosa as that previously thought to be for proangiogenic VEGF, the lamina propria (Griga et al, 2002), goblet cells and glandular cells of the mucosa (Griga et al, 1999) and our findings for the pan-VEGF stain are consistent with this. The detection of the same cells by the antibody raised specifically against VEGF165b, as by an antibody to all VEGF isoforms, concurs with our ELISA findings that in normal colonic mucosa it is the VEGFxxxb isoforms that predominate. This has significant implications for interpretation of all previous studies investigating VEGF expression in the colon, not only for tumour studies (Hurwitz et al, 2004), but also for collagenous (Griga et al, 2004) and ischaemic colitides (Okuda et al, 2005).
The results here show that VEGF overexpression can alter the rate of human xenografted tumour growth in vivo and moreover that VEGF165b can antagonise the effects of VEGF165, thus confirming the role of the C terminus of VEGF in determining its function and the importance of the ratio of VEGFxxxb to VEGFxxx in the progression of tumour growth. The ability of AAT to inhibit xenografted tumour growth has been demonstrated previously (Kendall and Thomas, 1993; Kim et al, 1993; Kanai et al, 1998; Wildiers et al, 2003). The effectiveness of AAT has translated into the clinic with bevacizumab in the treatment of renal (Yang et al, 2003), breast (Ramaswamy et al, 2006) and CRCs (Hurwitz et al, 2004). Furthermore, the ability of VEGF165b to inhibit VEGF165-mediated angiogenesis in a rat mesenteric and rabbit corneal assay (Woolard et al, 2004) and mouse retinal angiogenesis assay (Konopatskaya et al, 2006) suggests that it may serve as a novel AAT agent. Consistent with this, we show here that the overexpression of VEGF165b in human CRC can inhibit tumour growth in vivo, further supporting the potential role of relative VEGFxxxb downregulation as part of the angiogenic switch in tumour progression. Furthermore, neither VEGF165 nor VEGF165b overexpression alters the in vitro proliferation or apoptosis rates of cells, suggesting that the mechanism of action of VEGF in altering tumour growth rate is not through an autocrine pathway, but likely to be via its known antiangiogenic effects. Furthermore, the antagonistic effects of VEGF165b overexpression on tumour growth when co-overexpressed with the potent proangiogenic VEGF165 and the increased tumour necrosis observed when VEGFxxxb was overexpressed further suggests that VEGF165b inhibits tumour growth through antiangiogenesis.
Bevacizumab binds to all the conventional isoforms of VEGF (Kim et al, 1992) via an epitope on the common region of the protein family, adjacent to the receptor-binding site (Muller et al, 1997). In CRC, the response to bevacizumab is limited to a small subset of patients, approximately 11–12%, but to date this subset has not been shown to be predictable (Jubb et al, 2006). The results here show that the affinity of bevacizumab was similar for VEGF165 and VEGF165b and concur well with the results by Presta et al (1997). Thus the variability in response to bevacizumab could be explained by VEGFxxxb expression. The relative expression levels of the VEGFxxx isoforms in human colon carcinoma vary from only 27% of the VEGFxxxb isoforms to 60-fold excess, whereas in normal colon, we did not detect excess of VEGFxxx over VEGFxxxb. This can explain why in some patients bevacizumab could be a highly effective antiangiogenic agent – those in which the vast majority of the VEGF in their carcinomas is VEGFxxx (the proangiogenic family), whereas in most, those with significant VEGFxxxb expression, bevacizumab may be less effective. To date, no biomarkers have been identified that can predict response to bevacizumab (Ince et al, 2005; Jubb et al, 2006). The differential response in the LS174t tumours to bevacizumab therapy in mice depending upon VEGF165b-to-VEGF165 ratio shown here is suggestive of its role as a potential biomarker in CRC and possibly other cancers that are responsive to bevacizumab. It remains to be seen whether patients with high levels of VEGF165b respond poorly to bevacizumab, but this needs to be tested.
Vascular endothelial growth factor has been shown to be a survival factor for several carcinoma types, including colon. Transfection of RKO colon cancer cells with short interfering RNA against the coding region of VEGF reduced cell proliferation by 67% (Mulkeen et al, 2006). However, given that normal colonic epithelial cells produce principally VEGFxxxb and the toxicity of blocking VEGFxxxb in AAC1 adenoma cells, the autocrine role, may be predominantly due to VEGFxxxb in colonic epithelial cells, suggesting for the first time a physiologically active role of VEGFxxxb. Furthermore, the general anti-VEGF antibody bevacizumab was also cytotoxic to AAC1 cells, suggesting that bevacizumab could potentially have negative effects on normal colonic epithelium, which expresses high levels of VEGFxxxb. A life-threatening complication of bevacizumab therapy is gastrointestinal perforation, the cause of which is unknown (Hurwitz et al, 2004), but could be due to its effects on survival of normal colonic epithelial cells, as suggested by the results shown here.
In summary, we show here that in normal colonic epithelium, the antiangiogenic isoforms form the majority of VEGF, and VEGF upregulation in CRC is unique to the proangiogenic isoforms. These results indicate that the role of VEGF in the normal function of the colonic mucosa may depend either on the function of VEGFxxxb, which is still unknown, or the effect of the balance between the isoforms. We also show that this switch appears to be an endogenous one to transformation, although environmental cues such as hypoxia can also induce the switch. Furthermore, VEGF165b inhibits both colorectal tumour growth (in a VEGF-dependent manner) and the effect of bevacizumab on that tumour growth. We further show that bevacizumab binds VEGF165b and that VEGF165b is an autocrine survival factor for colonic epithelial cells. These results suggest that anti-VEGF therapy for CRC may be better targeted to patients with significant excess of proangiogenic isoforms over antiangiogenic isoforms, and that therapies that specifically target the proangiogenic isoforms may be more effective.
This work was supported by the Royal College of Surgeons of England (AHRV), Cancer Research UK (AHRV, ER, ABH), The British Heart Foundation BB2000003 (DOB), North Bristol Research Foundation, Specific Cancer Research Fund and the Richard Bright VEGF Research Trust.