Association between VEGF Splice Isoforms and Progression-Free Survival in Metastatic Colorectal Cancer Patients Treated with Bevacizumab

Bevacizumab, a recombinant humanized monoclonal antibody that binds VEGF, has become a standard component of the treatment of patients with metastatic colorectal cancer. In 2004, a landmark study comparing bolus irinotecan, 5-fluorouracil, and leucovorin (IFL) and placebo treatment with IFL and bevacizumab (1) showed a 4.7- and 4.2-month increase in median overall survival (OS) and progression-free survival (PFS), respectively. In 2007, the ECOG E3200 trial of 829 patients with previously treated metastatic colorectal cancer showed that addition of bevacizumab to oxaliplatin, 5-fluorouracil, and leucovorin (FOLFOX4; ref. 2) resulted in a 2.1- and 2.6-month increase in median OS and PFS, respectively. There have as yet been no definitive studies identifying biomarkers that predict outcome to bevacizumab therapy.

VEGF-A is a product of one gene encoding multiple isoforms, identified by their amino acid number and c-terminal sequence. Alternative splicing of exon 8 results in 2 families of proteins (3). Proximal splice site (PSS) usage generates a short open reading frame (ORF) of 6 amino acids common to the pro-angiogenic VEGF isoforms. An alternative distal splice site (DSS) in exon 8 results in a different 6 amino acid ORF, resulting in proteins of the same length as the pro-angiogenic isoforms, but with a different c-terminal sequence. These isoforms that are anti-angiogenic in physiologic systems (4), in experimental VEGF-driven angiogenesis (5), and in pathologic angiogenesis driven by VEGF (6, 7), including in tumors (5, 8–12). These isoforms, the most common of which is VEGF₁₆₅b, form a substantial portion of total VEGF in many tissues including normal colon (12). VEGF₁₆₅b is relatively downregulated in colon cancer but to a variable extent (12). VEGF₁₆₅b binds bevacizumab with the same affinity as VEGF₁₆₅ (12), and the overexpression of VEGF₁₆₅b in human colon carcinomas grown in mice conferred resistance to bevacizumab treatment (12). These findings suggest that VEGF₁₆₅b may interfere with bevacizumab treatment. This led us to propose that bevacizumab may be more effective against tumors in which VEGF₁₆₅b levels are low. We sought to test this hypothesis in samples from patients treated with bevacizumab + FOLFOX in one of the original pivotal phase III clinical trials of bevacizumab + FOLFOX in colorectal carcinoma.

In the clinical trial E3200, eligible patients were randomly assigned to receive FOLFOX4 in combination with bevacizumab (arm A); FOLFOX4 without bevacizumab (arm B); or bevacizumab alone (arm C). Random assignment was stratified on the basis of prior radiation therapy and Eastern Cooperative Group (ECOG) performance status. The primary endpoint of the study was OS. A total of 829 patients were enrolled onto the study between November 2001 and April 2003 from 221 sites in the United States and South Africa. Nine patients who received treatment were determined to be ineligible, leaving 820 randomized patients analyzable. In February 2003, the bevacizumab-alone arm of the study closed to accrual after interim analysis suggested inferior survival when compared with the chemotherapy-containing arms of the study. The final study sample sizes were therefore 286 patients (arm A), 291 patients (arm B), and 243 patients (arm C; see Fig. 1 for breakdown of cases). Human studies were conducted in accordance with the Helsinki Declaration, and expression studies were conducted with local ethics committee approval. Tumor samples were available from 300 patients enrolled in the trial.

Sections (5 μm) were cut from all 300 blinded paraffin-embedded tissue blocks taken from participants in E3200 (2). All isoforms of VEGF, termed VEGF_total, were detected using a commercially available anti-VEGF antibody (Dako, mouse IgG1, clone VG1). This antibody detects isoforms generated by PSS selection in exon 8 (VEGF₁₆₅, and other isoforms expressed at lower levels such as VEGF₁₂₁, VEGF₁₈₉, and others, generically termed VEGF_xxx) and those generated by DSS selection (VEGF₁₆₅b, VEGF₁₂₁b, VEGF₁₈₉b, and possibly others, generically termed VEGF_xxxb; refs. 5, 13). VEGF_xxxb expression was examined by immunohistochemistry using a mouse monoclonal IgG1 antibody raised to the terminal nine amino acids of the VEGF₁₆₅b sequence. It was affinity purified against the antigen from conditioned media of hybridoma cells (Abcam 14994). It has previously been shown to have specificity for the VEGF₁₆₅b isoform over the VEGF₁₆₅ isoforms and does not detect VEGF₁₆₅, or VEGF₁₂₁, recombinant protein (5, 13). This antibody has a VEGF₁₆₅b association constant of 3 × 10⁴ (mol/L)⁻¹s⁻¹ and dissociation constant of 0.011s⁻¹ but no affinity with VEGF₁₆₅ (8). It also detects other VEGF_xxxb isoforms such as VEGF₁₂₁b, VEGF₁₄₅b, and VEGF₁₈₉b in human tissues (13). Sections were dewaxed with xylene for 15 minutes before rehydration in graded alcohols and antigen retrieved by microwave treatment for 12 minutes at 800 W in 0.01 mol/L sodium citrate buffer, pH 6.0, with the solution maintained just below the boiling point (96°C ± 2°C). Sections were immersed in 3% hydrogen peroxide for 5 minutes and incubated with 1.5% normal horse serum (NHS) for 30 minutes in a humid chamber at room temperature. After washing, each slide was subject to incubation with 5 μg/mL VEGF_total antibody in NHS or 12 μg/mL anti-VEGF₁₆₅b monoclonal antibody diluted in PBS/T-1% bovine serum albumin (BSA). Sections were incubated overnight at 4°C. The blocking step with NHS was repeated before the addition of 2 μg/mL (for VEGF₁₆₅b) or 2 μg/mL (VEGF_total) biotinylated horse anti-mouse secondary antibody (Vector BA-2000, in NHS, 1:66 in PBS) for 30 minutes. The negative control received a matched concentration of normal mouse immunoglobulin diluted in PBS/T-1% BSA. Avidin-biotinylated enzyme complex was prepared according to the manufacturer's instructions (Elite, Vector Laboratories) and incubated with sections for 45 minutes at room temperature. This was followed by the addition of the visualizing 3,3′-diaminobenzidine (DAB) substrate for all sections and counterstaining in hematoxylin. Sections were dehydrated in graded alcohols (70%, 90%, and 100% ethanol) before final clearing in xylene and mounting with DPX and glass coverslips. Slides were dried overnight before light microscopy examination.

Slides were independently assessed and staining intensity scored (1–6). All sections were examined, conducted by 2 different assessors (D.O. Bates and A.H.R. Varey), blinded to treatment. A third assessor (P. Ramani) was used to assess anomalies. They graded the intensity of DAB staining (1–6) in the normal mucosal tissue and the most poorly differentiated tumor for each section. Stroma was not scored. A score of 1 indicated no staining; 2, weak staining; 3, clear staining; 4, robust staining; 5, strong staining; and 6, intense staining (see Supplementary Figure for examples). To determine the relative intensity of the VEGF isoforms, the staining intensity in the tumor on a scale of 1 to 6 was normalized to that in the peritumoral (histologically normal) colonic mucosa for each stain. The ratio of the 2 scores was then calculated (VEGF₁₆₅b:VEGF_total). Thus, to obtain a score, staining was required in the normal mucosal tissue for both VEGF₁₆₅b and VEGF_total antibody staining. The normal mucosal tissue acted as an internal control for both antibody stains. Scores were sent to the ECOG statistical service (P.J. Catalano) for subsequent analysis. A median split was conducted to separate the samples into 2 groups—high and low VEGF₁₆₅b:VEGF_total. Kaplan–Meier survival curves were calculated for each of the 2 groups. Thus, the actual value of the ratio of VEGF₁₆₅b:VEGFtotal is reduced to a discrete value of greater or less than median. From the 300 samples in the original study, 149 were suitable for use in the final analysis, as they contained both normal mucosa and tumor and successfully stained with both VEGF₁₆₅b and VEGF_total antibodies, as determined by positive staining in the normal mucosal tissue on the slides. However, 52 of these were in arm C and thus were not included in the analysis. Representative images are given in Fig. 2.

Kaplan–Meier methodology was used to present all survival curves and estimate median PFS and OS; standard proportional hazards regression modeling for censored time to event data was used to evaluate the unadjusted and adjusted relationship between VEGF₁₆₅b:VEGF_total isoform ratio and outcome (PFS and OS). As of this analysis, only 11% of the cases are censored for PFS and less than 4% of the cases are censored for OS. All quoted P values are 2-sided, and P values less than 0.05 were considered statistically significant. A priori power analysis showed that with all 175 cases in each arm, an 83% power to detect an interaction effect (ratio of HRs of about 1.8) would be achieved for the PFS endpoint. However, an interaction effect would not be expected to be that large in the OS endpoint because the overall treatment differences were smaller in the clinical study for OS, and the OS power is therefore more limited, only 47% for an interaction effect (ratio of HRs of 1.5). Thus, the study aimed to test the hypothesis that PFS was increased. Thus, the a priori cutoff point for the ratio was the median value of VEGF₁₆₅b (tumor/normal):VEGF_total (tumor/normal). The primary endpoint of this analysis was PFS.

Table 1 shows the breakdown of cases by treatment arm. A total of 820 cases were included in the original E3200 clinical dataset, hence 671 clinical cases were not in the VEGF isoform ratio analysis and 149 cases were included in the VEGF₁₆₅b:VEGF_total isoform ratio analysis.

There were no appreciable differences in treatment arm, sex, age, race, performance status, and the number of metastatic sites at study entry for the study overall compared with those samples included in the VEGF₁₆₅b:VEGF_total isoform ratio analyses (see Supplementary Table S1).

Figure 2 shows representative samples of tissue stained for VEGF₁₆₅b or VEGF_total from normal tissue and 2 stages of differentiation (poor and well) of tumor. Figure 3A shows the distribution of ratios for all samples. The median score was 1.0, indicating that the VEGF₁₆₅b staining relative to the VEGF_total staining varied around equivalence. Figure 3B shows the distribution of scores. Table 2 shows the results of proportional hazards regression modeling of PFS and OS by VEGF₁₆₅b:VEGF_total isoform ratio (predictive index or PI) in tumors, for both unadjusted and adjusted analyses where the adjustment in the models included the following demographic and clinical on-study variables: sex, race, age as a continuous predictor, performance status, and the number of metastatic sites as a continuous predictor. Formal tests for the interaction between low/high VEGF₁₆₅b:VEGF_total isoform ratio and treatment arm were conducted using the proportional hazards regression model. The interaction test of the ratio by arm was significant (P = 0.024) for PFS but did not attain statistical significance (P = 0.18) for OS. These effects were also present after adjustment for demographic and clinical variables (listed above, P_interaction = 0.072).

Table 2 shows that in unadjusted analyses, significant treatment differences were observed between arm A (bevacizumab + FOLFOX4) and arm B (FOLFOX alone) for PFS (P = 0.018) among patients with a low VEGF₁₆₅b:VEGF_total isoform ratio (Fig. 4A). In contrast, among patients with a high VEGF₁₆₅b:VEGF_total isoform ratio, there was no significant treatment difference between the 2 arms (P = 0.40 Fig. 4B). Although the HRs were in the same direction, no significant associations were observed for the OS endpoint (Fig. 4C and D).

The response rate [partial response (PR) or complete response (CR)] in the low VEGF₁₆₅b:VEGF_total isoform ratio group treated with bevacizumab + FOLFOX was 20.8% but 7.7% in those treated with FOLFOX4 alone. This compares with a response rate of 5.0% in the bevacizumab + FOLFOX–treated group with high VEGF₁₆₅b:VEGF_total isoform ratio and 7.4% in the FOLFOX4 alone–treated high VEGF₁₆₅b:VEGF_total isoform ratio group. The study was not powered to detect a difference in response rate, and the difference was not significant by Fisher exact test.

As the clinical use, expense and adverse effect profile of bevacizumab (and other anti-angiogenic agents such as sunitinib, sorafenib, pazopanib) has grown, the need to identify predictive markers to guide the clinical use of these agents has gained importance and urgency. Initial studies of the phase III clinical trials of bevacizumab showed no predictive value for total VEGF expression or microvessel density (14), suggesting that it was not the VEGF levels that determined outcome. Correlations between outcomes with bevacizumab and single-nucleotide polymorphisms (SNP) of VEGF in breast cancer (15); ICAM, interleukin (IL)8, and VEGF in non–small cell lung cancer (16); VEGFR1 in pancreatic and renal carcinoma (17); and hypertension in non–small cell lung cancer (18); and between VEGF SNPs and hypertension in patients with metastatic breast cancer (15) have been shown. In colonic carcinoma, low-serum angiopoietin-2 (Ang-2), an inhibitory ligand of endothelial Tie-2 receptor, has been proposed as a possible predictive marker (19), and in gastric cancer, a nonsignificant trend was seen in patients with high baseline serum VEGF (20). In breast cancer, a retrospective analysis of the mRNA or protein expression of 11 vasoactive or cell-cycle–related factors found that the low expression of delta-like ligand (Dll) 4, VEGF-C, and neuropilin-1 may identify bevacizumab-sensitive patients, although statistical significance was lost following correction for multiple hypothesis testing (21).

On the basis of evidence that the VEGF-A isoform splice variant VEGF₁₆₅b shows anti-angiogenic properties in animal systems (12), we postulated that patients with colorectal cancer with tumors expressing low VEGF₁₆₅b levels would be more likely to benefit from bevacizumab + FOLFOX–based therapy. Our findings suggest that a low VEGF₁₆₅b:total VEGF ratio may be a marker for clinical effect from bevacizumab + FOLFOX–based therapy.

There are a number of interpretations of the mechanisms underlying this finding. We previously showed that bevacizumab binds VEGF₁₆₅b with similar affinity to VEGF₁₆₅ (12)_. It is possible that in patients with a higher VEGF₁₆₅b level, the effective dose of bevacizumab in the tumor is lowered, as quantitatively more of the antibody is bound to VEGF₁₆₅b and not VEGF₁₆₅, thereby diminishing the anti-angiogenic effect of VEGF₁₆₅b. However, the concentration of therapeutic antibody is thought to be in excess of the VEGF levels. Conversely, patients with lower VEGF₁₆₅b levels may have a more angiogenic tumor, which would in turn allow for more rapid progression. Thus, a low VEGF₁₆₅b level could render the tumor more susceptible to bevacizumab + FOLFOX than the high VEGF₁₆₅b group, as angiogenesis is more likely to facilitate the progression. Finally, it is possible that binding VEGF₁₆₅b negates a more dominant anti-angiogenic effect than can be achieved by inhibiting the pro-angiogenic effect of VEGF₁₆₅ (12, 22, 23).

There are a number of limitations of our study. First, despite starting with a large number of samples (300), only 97 were able to be used for all 4 staining markers. Thus, the ability to detect meaningful differences in all clinically relevant endpoints, such as OS, was reduced. In addition, a larger cohort would have allowed for evaluation of the predictive function of even lower quartiles of the VEGF₁₆₅b:VEGF_total ratio. Second, the use of an immunohistochemical scoring system is subjective, and future validation using an automated quantitative system should be considered. mRNA studies would be more quantitative, although quantitative PCR for VEGF₁₆₅b needs to be undertaken with great care to avoid artifactual mispriming and amplification of VEGF₁₆₅. It is necessary to use appropriate controls to ensure specificity, the lack of which has caused misinterpretation of negative data in some recent studies (24). Third, our findings are in previously treated patients with advanced colorectal cancer and may not necessarily extrapolate to other tumors. And, finally, given the complexity of angiogenesis, it is unlikely that one marker will serve a uniformly predictive function across disease types and their respective stages.

The relationship between low VEGF₁₆₅b:VEGF_total ratio and PFS in patients treated with bevacizumab + FOLFOX in E3200 supports its further evaluation as a potential biomarker and provides new insight into the tumor angiogenic network.

This work is original and has not previously been published. Its contents are solely the responsibility of the authors and do not necessarily represent the official views of the National Cancer Institute. B.J. Giantonio has received honoraria from Roche and Genentech, ABB research funding and consultancy from Genentech. P.J. O'Dwyer has received research support from Genentech. N.J. Meropol received consulting fees from Precision Therapeutics. D.O. Bates and S.J. Harper are inventors on the patent describing VEGF₁₆₅b and use of antibodies, including as biomarkers. No potential conflicts of interest were disclosed by the other authors.

Conception and design: D.O. Bates, A.H. Varey, P.J. O'Dwyer

Development of methodology: D.O. Bates, P.J. O'Dwyer

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): D.O. Bates, K.E. Symonds, A.H. Varey, P. Ramani, P.J. O'Dwyer, B.J. Giantonio, N.J. Meropol, A.B. Benson

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): D.O. Bates, P.J. Catalano, P.J. O'Dwyer, B.J. Giantonio, N.J. Meropol

Writing, review, and/or revision of the manuscript: D.O. Bates, P.J. Catalano, A.H. Varey, P. Ramani, P.J. O'Dwyer, B.J. Giantonio, N.J. Meropol, A.B. Benson, S.J. Harper

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): D.O. Bates, K.E. Symonds, P.J. O'Dwyer, S.J. Harper

Study supervision: D.O. Bates, P. Ramani, P.J. O'Dwyer, B.J. Giantonio, N.J. Meropol, S.J. Harper

This study was coordinated by the Eastern Cooperative Oncology Group (Robert L. Comis, M.D.) and supported in part by Cancer Research UK CF11392/A8451, Association for International Cancer Research (07-0605), Public Health Service grants CA23318, CA66636, CA21115, CA15488, CA27525, CA17145 CA139003-01, and from the National Cancer Institute, NIH, and the Department of Health and Human Services, and Richard Bright VEGF Research Trust and North Bristol Cancer Research Fund.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.