Vascular Endothelial Growth Factor Pathway Polymorphisms as Prognostic and Pharmacogenetic Factors in Cancer: A Systematic Review and Meta-analysis

Angiogenesis is a critical component in oncogenesis (1). Tumors can promote angiogenesis through the secretion of proangiogenic factors such as VEGF (and its isoforms) and basic fibroblast growth factor (bFGF) or overexpression of associated receptors: VEGFR, VEGFR2/kinase insert domain receptor (KDR), and NRP2 (2). Tumors use this host-mediated process to meet their metabolic needs, to help remove biologic waste products of rapidly dividing cells, and to provide a mechanism for metastasis.

Angiogenic factors interact with a multitude of pathways, including matrix metalloproteinases, inflammatory cytokines, chemokines (e.g., interleukins and interferons), integrins, nitric oxide, and other endothelial-related proteins. Downstream signaling effectors of angiogenesis receptors also lead to the activation of the phosphoinositide 3-kinase/AKT (PI3K/AKT) and mitogen-activated protein kinase (MAPK) pathways.

The VEGF pathway is predominantly driven by host factors. Host genetic variability in this pathway may, therefore, influence angiogenesis-dependent biologic pathways during cancer development and hence influence outcome and sensitivity to various therapies. Associations between genetic polymorphisms in VEGF and the risk of developing various cancers have been reviewed previously (3). Less is known about the role of potential germline polymorphisms as prognostic markers of clinical outcomes or the potential predictive or pharmacogenetic nature of these polymorphisms in relation to drug efficacy and toxicity.

We conducted a systematic review and meta-analysis of the role of VEGF pathway polymorphisms as markers of clinical outcome and toxicity in cancer patients. The 2 aims were as follows: (i) to investigate polymorphisms as pharmacogenetic predictors of drug efficacy and toxicities and (ii) to investigate polymorphisms as general prognostic indicators of survival, independent of therapy.

We conducted a literature search of MEDLINE (host: PubMed) for the period of May 1990 through July 2011 using the keywords and MeSH terms: {“angiogenesis,” “VEGF,” “VEGFR1,” “FLT1,” “KDR,” “VEGFR2,” “FGF2,” “FGF,” “FGFR,” “NRP1,” “Endostatin,” “VEGFA,” “VEGFB,” and “Bevacizumab”} and “cancer” and “polymorphism.” We limited searches to English language articles.

Article eligibility was reviewed independently by 2 reviewers (G. Liu, L. Eng). Eligible articles were original peer-reviewed studies focused on genetic variation in the following angiogenesis pathway genes: VEGFA, VEGFB, and FGF2, and/or genes that code for their direct interactors or receptors: fms-related tyrosine kinase 1 (FLT/VEGFR1), KDR (VEGFR2), neuropilin 1 (NRP1), and FGFR. These genes were selected from The Pharmacogenomics Knowledgebase website (4). We also included endostatin, the 20-kDa C-terminal fragment derived from type XVIII collagen that is a broad-spectrum antiangiogenic factor, which can affect both VEGF and FGF2 pathways and is coded by COL18A1. Eligible studies must have reported cancer outcome (i.e., not risk alone), and may have included either/both prognostic and pharmacogenetic assessments. The outcomes evaluated were of any of the following: overall survival (OS), progression-free survival (PFS), time to progression (TTP), time to recurrence (TTR), disease-free survival (DFS), response rate (RR), and association with treatment toxicity. Finally, all studies must have included genotyping of 25 or more and at least one VEGF pathway polymorphism to warrant inclusion.

Duplicate studies were excluded as were those testing solely the association of polymorphisms and cancer risk/susceptibility, tumor staging/grading, or gene expression. We also excluded case reports, opinion pieces (e.g., letters to the editors), and other review papers. We categorized studies as being pharmacogenetic (effect of polymorphism on response/outcome/toxicity to therapy where treatment included antiangiogenic therapy) or prognostic (effect of polymorphism on outcome irrespective of therapy); data were analyzed for various disease sites: breast, colorectal, gastroesophageal, genitourinary, gynecologic, lung, and other cancers.

The following data were extracted from individual publications: title, authors, disease site, country, sample size, and/or actual number of patients genotyped, patient inclusion/exclusion criteria, treatment/intervention details, histologic subtypes, grade, stage, tissue used for genotyping (e.g., whole blood vs. archival normal tissue), and polymorphisms that were genotyped. Associations between polymorphisms and outcome data were recorded as OR, HR, or RR, along with their respective confidence intervals (CI). We noted whether studies were analyzed using only univariate (or crude) analysis or in multivariate (or adjusted) analysis. Haplotype analyses, if conducted, were also recorded. A common problem was the use of different nomenclature and names for the same polymorphism. Thus, all identified polymorphisms sharing the same RefSNP (rs) numbers were renamed to share common names throughout this manuscript [Supplementary Table S1 provides the common and Human Genome Variation Society nomenclature; ref. (5)].

Meta-analyses were conducted to evaluate the overall prognostic associations of the most commonly evaluated VEGF polymorphisms when there were an adequate number of individual studies reporting similar outcomes to justify such an analysis. Only studies reporting HR for OS were included. Analyses were undertaken separately for each polymorphism and were conducted in 2 phases. First, study data reporting outcome by wild-type or variant polymorphisms irrespective of zygosity were assessed together (i.e., pooled). Second, data were assessed separately for homozygous and heterozygous genotypes. Differences between homozygous and heterozygous subgroups were assessed using methods described by Deeks and colleagues (6). Data were analyzed using RevMan 5.1 analysis software (The Cochrane Collaboration). Pooled estimates of HR were computed using generic inverse variance and a random-effects model (7, 8). All statistical tests were 2-sided, and statistical significance was defined as P < 0.01, which corrected for multiple statistical testing using the Bonferroni method.

A total of 614 studies were initially identified (Fig. 1). Of these, 186 were duplicate reports/studies, 130 were related to cancer development risk, and 238 further studies were excluded because they did not include at least 25 patients, did not include polymorphisms of interest, or did not assess any pharmacogenetic or prognostic outcomes of interest. Sixty studies met the inclusion criteria and were included in the final analysis.

Most of the included studies had sample sizes of under 150 patients, and were secondary analyses of phase I/II clinical trials. Other characteristics of included studies are shown in Table 1. The majority of trials were prognostic (n = 48), based primarily on data from case series [n = 33 (69%)]. Most pharmacogenetics studies (n = 12) were nested case–control analyses of clinical trials [n = 9 (75%)].

Nine studies investigated the predictive effect of polymorphisms on response to bevacizumab in numerous cancer sites (Table 2, VEGF pathway polymorphic predictive associations of survival and response; additional data in Supplementary Table S2; refs. 9–17). Among those investigating bevacizumab response, Schneider and colleagues (N = 341) reported that the C allele of VEGF −2578C>A and G allele of VEGF −1154G>A (alternatively known as −116G>A) were independently associated with poorer OS in patients treated with bevacizumab (9). Similarly, a smaller study investigating 9 polymorphisms, by Formica and colleagues (N = 40), reported the same finding for VEGF −1154G>A with respect to PFS (corrected for multiple comparisons) and also found the C allele for VEGF +405G>C (alternatively known as +634G>C) associated with improved overall radiologic RR (12). Another small study by Argiris and colleagues (N = 28), found the C/− genotypes of VEGF +405G>C significantly associated with OS and TTP in multivariate analysis, but only 1 patient was present in the G/G genotype group (14).

In all other studies, associations have yet to be validated: (i) one report by Etienne-Grimaldi and colleagues (N = 138), identified the C allele of VEGF +936C>T associated with a shorter TTP; (ii) Loupakis and colleagues (N = 111), found the C allele of VEGF −460C>T (alternatively known as −1498T>C) associated with improved PFS; and (iii) Schultheis and colleagues (N = 70) reported that the C/T genotype of C>T polymorphism in the 3′ UTR of NRP1 was associated with improved PFS, after adjustment for 30 comparisons (10, 11, 15). In 3 other studies by Jain and colleagues (n = 178), Zhang and colleagues (n = 65; using both univariate and multivariate analyses), and Smerdel and colleagues (n = 38), VEGF, KDR, and NRP1 polymorphisms were found to have no association with survival (13, 16, 17).

Seven studies investigated bevacizumab toxicity in various cancer sites (9–12, 14, 15, 17). Three additional studies investigated the predictive effect of VEGF polymorphisms on toxicity of non-VEGF targeting drugs (Table 2, VEGF pathway polymorphic predictive associations of toxicity; additional data in Supplementary Table S3; refs. 18–20). The experimental paclitaxel and bevacizumab arm of a breast cancer study (n = 180) identified the C allele in VEGF −460C>T and the G allele in VEGF +405G>C as being associated with a greater frequency of grade 3 or 4 hypertension (9). In contrast, Etienne-Grimaldi and colleagues, found the G allele of VEGF +405G>C associated with decreased toxicity (which included any signs of hypertension, arterial or venous thromboemboli, or hemorrhages; ref. 10). Four smaller studies (N = 28–111) involving various combinations of bevacizumab with chemotherapeutic drugs found no associations of any evaluated polymorphisms with any of the toxicities (11, 12, 14, 15). In Jain and colleagues' investigation of KDR polymorphisms, the A allele in KDR 1718T>A (alternatively known as 1719A>T, Q472H, 1416A>T, or Exon 11T>A) was found associated with increased frequency of hypertension and hand–foot skin reactions in patients treated with sorafenib and bevacizumab (17).

Three other studies investigated the role of VEGF polymorphisms in the toxicities in non-VEGF targeting drugs (18–20). In a 49 patient study focused on drug toxicity of 5-fluorouracil and cisplatin chemotherapy, a multitude of nominally significant VEGF polymorphic associations with various toxicities such as cheilitis, leukopenia, and stomatitis was reported (18). Two studies, 1 investigating danusertib (a multikinase aurora inhibitor) and another on telatinib found no associations between any KDR or FLT polymorphisms and toxicity (19, 20).

Studies that evaluated VEGF pathway polymorphisms as prognostic markers of cancer outcome are described in Supplementary Tables S4–S7 (21–68).

VEGF +936C>T was associated with OS, DFS, or TTR in the 3 largest colorectal cancer studies involving; however, 2 showed improved outcome by the T allele while 1 showed worse outcome by the T allele (27, 29, 32). Among these 3 studies, the 2 showing improved outcome by the T allele involved patients from the United States and Greece and both used blood as their tissue source while the one showing worse outcome was from Korea and used fresh tumor tissue as their tissue source. There was more consistency with VEGF +405G>C, where 2 of these studies showed improved outcome in patients carrying the C allele (a third smaller study reported no significant association; refs. 27, 32). Of the 2 studies showing improved outcome, both included cases from all stages; Dassoulas and colleagues evaluated a Greek population while Kim and colleagues focused on a Korean population that specifically underwent surgical resection.

There was no consistency in the relationship between various VEGF polymorphisms and clinical outcome in gastroesophageal cancers. For example, 1 gastric cancer study (N = 137) found a strong relationship between carrying at least one A allele in COL18A1 (Endostatin) +4349G>A (alternatively known as D104N) and worse outcome [adjusted HR (aHR) for TTR, 2.28; 95% CI, 1.26–4.11; P = 0.004) while a larger 239 patient study in distal esophageal and gastroesophageal junction adenocarcinomas found no relationship (35, 39).

Of 5 breast cancer studies, VEGF polymorphisms, −7C>T and +405G>C, and the KDR polymorphism, +1192A>G, were associated with either OS or DFS in single, unvalidated studies (21, 22, 24). In genitourinary and gynecologic cancers, associations were either of borderline statistical significance or involved crude, unadjusted analyses, with 2 exceptions: VEGF +405C/C genotype was associated with poorer OS in a cervical cancer study (n = 199) from Korea; and minor alleles of VEGFC polymorphisms (rs17697305 and rs1485766) were associated with altered OS outcomes in a single ovarian cancer study that evaluated more than 1,400 polymorphisms across multiple pathways in 325 patients (52, 54). Separate lung cancer studies reported worse survival with VEGF +405G/G, VEGF −1154A/-, and VEGF −460T/T; none being validated (55, 56, 58).

Because of the large number of studies (n = 42) evaluating 5 individual VEGF polymorphisms (VEGF +936C>T, VEGF −460T>C, VEGF −1154G>A, VEGF −2578C>A and VEGF +405G>C; individual details in Supplementary Tables S4–S7), meta-analyses were conducted separately on these 5 polymorphisms. When the homozygous and heterozygous polymorphisms were assessed together (i.e., pooled; Table 3, Combined heterozygosity and homozygosity results), the C allele of VEGF +405G>C was found significantly associated with improved OS (HR, 0.74; 95% CI, 0.60–0.91; P = 0.004). The corresponding forest plot is shown in Table 4, Forest plots for pooled and subgroup analysis of VEGF polymorphisms as prognostic biomarkers for VEGF +405G>C. Among studies in the zygosity-segregated analysis (Table 3, Zygosity-segregated meta-analysis results), the same variant of VEGF +405G>C also showed a statistically significant improvement in OS (HR, 0.79; 95% CI, 0.67–0.94; P = 0.007) that was not different by zygosity (Table 4, Forest plots for pooled and subgroup analysis of VEGF polymorphisms as prognostic biomarkers for VEGF +405G>C). In addition, a nonsignificant trend was identified where VEGF +936C>T variants (regardless of zygosity) were associated with worse OS (HR, 1.38; 95% CI, 0.96–1.97; P = 0.08; Supplementary Fig. S1B). For other polymorphisms, there was no evidence of association with survival in either pooled or zygosity-segregated analyses (Supplementary Figs. S2–S4). Assessment of funnel plots in both pooled and zygosity-segregated analysis revealed no any evidence of publication bias (data not shown).

Given that 2 of these 5 VEGF polymorphisms (VEGF −460C>T and VEGF −2578C>A) were in strong linkage disequilibrium with each other [(D′ > 0.90) on the basis of HapMap data, and with several other polymorphisms (rs833068 G>A, rs2146323 C>A, and VEGF −7C>T)], a meta-analysis was conducted on this haplotype block, pooling results from all studies involving any of these polymorphisms and assigning association direction on the basis of the most common haplotype (69). No significant association with survival outcome was found.

Antiangiogenic drugs are part of the clinical armamentarium for numerous solid tumors. Results of clinical trials of antiangiogenic agents have shown variable results (70). The need for predictive and prognostic markers for antiangiogenic therapy is, therefore, acute. There were too few studies focusing on the same polymorphism and outcome to assess consistency of results across studies. Of the 2 largest studies, Jain and colleagues pooled data from multiple cancer sites across several phase I and II trials of sorafenib and bevacizumab and found no significant results (17). In contrast, Schneider and colleagues evaluated 341 breast cancer patients treated with paclitaxel with or without bevacizumab and found 2 variants (VEGF −2578C>A and −1154G>A) associated with improved OS, whereas 2 different variants (VEGF +405G>C and −460C>T) were significantly associated with hypertension; the VEGF −1154G>A and −460C>T are, in general, highly linked in the Caucasian population (9). Both the A allele of VEGF −1154G>A, and the C allele of VEGF −460C>T, were associated with improved outcome in 2 separate pharmacogenetics studies (11, 12). However, these relationships await validation. In the future, comprehensive analyses of polymorphisms in either carefully designed observational studies or in larger phase III trials of bevacizumab and other VEGF-targeting agents in ovarian, colorectal, breast, and other cancers could be substantially more useful in delineating the role of these polymorphisms as pharmacogenetic markers.

We also reviewed VEGF pathway polymorphisms serving as general prognostic markers of outcome in patients not treated with VEGF-targeted agents. In this analysis, a large number of prognostic studies across many different cancers including ovarian, cervical, lung, breast, colorectal, gastric, esophageal/gastroesophageal junction, renal, lung, and liver cancers focused on 5 specific VEGF polymorphic variations. A meta-analysis was conducted on these polymorphisms, which identified the variant of VEGF +405G>C showing a highly significant prognostic role in improving the survival of cancer patients. As expected, given the varied cancer sites there was heterogeneity across individual studies (I² = 72%). However, the use of a pooled analysis and random effects modeling has been particularly useful for identifying significant relationships in the face of heterogeneous data.

VEGF +405G>C is located in the 5′-untranslated region of the VEGF gene and is in modest linkage disequilibrium with VEGF +936C>T (D′ = 0.65–0.74); this may help explain the nonsignificant trend found with VEGF +936C>T (34, 71). Genotype–phenotype analyses suggest that the C allele VEGF +405G>C is associated with decreased promoter activity and decreased expression of VEGF, as the C allele reduces binding of the transcription factor, MZF1, which then subsequently reduces gene expression (72–74). Hence, the C allele VEGF +405G>C may confer a survival advantage by reducing a tumor's ability to stimulate angiogenesis.

The quality of a systematic review and meta-analysis is partly related to the quality of its underlying studies. The included studies have limitations. First, studies were extremely varied in sample size: some studies had as few as 28 patients and the largest involved more than 1,100 patients. Our meta-analyses included weighting on the basis of sample size. Second, most studies used hypothesis-driven candidate approaches, whereby only specific genes and polymorphisms from a selected list were analyzed. This is a common study design and can be most useful for meta-analyses, but suffers from an inability to identify novel biomarkers. Third, statistical significance of the VEGF +405G>C polymorphism for cancer prognosis required the pooling of heterogeneous studies across disease sites that had different underlying populations (i.e., race/ethnicity), study designs, sample sources (i.e., blood vs. tumor-based tissue), and statistical analyses (including whether there was adjustment for multiple comparisons, and using multivariate vs. univariate analyses). However, this heterogeneity has advantages in that the significance of this finding was robust to all these aforementioned issues. Finally, several of the measured phenotypes, such as therapeutic response and drug toxicity, were defined differently across studies, rendering it difficult for cross-comparison, interpretation, and evaluation in pooling or meta-analytic analysis. For example, some studies explored each toxicity separately while other studies analyzed multiple toxicities as a composite measure. Hence, no meta-analyses were conducted for the pharamacogenetic arm of the study.

In conclusion, there have been a large number of published studies exploring the association of germline polymorphisms of the VEGF pathway with cancer outcomes and treatment toxicities. These studies have substantial heterogeneity both in terms of methodology and in terms of effect size. When studies using similar methodology and involving similar outcome were assessed in meta-analyses, the variants of VEGF +405G>C were highly significantly associated with a 26% relative improvement in OS across a variety of tumor types, even after accounting for the number of comparisons conducted. Future prospective studies should extend this research to include pathway analysis, haplotype analysis, and discovery-focused genome-wide association studies, followed by appropriate validation or meta-analytic analyses.

No potential conflicts of interests were disclosed.

Conception and design: L. Eng, A.K. Azad, S. Habbous, A.H. Maitland-van der Zee, E. Amir, G. Liu

Development of methodology: L. Eng, A.K. Azad, E. Amir, G. Liu

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L. Eng, A.K. Azad, H.J. Mackay, G. Liu

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L. Eng, A.K. Azad, S. Habbous, W. Xu, A.H. Maitland-van der Zee, S. Savas, H.J. Mackay, E. Amir, G. Liu

Writing, review, and/or revision of the manuscript: L. Eng, A.K. Azad, S. Habbous, V. Pang, A.H. Maitland-van der Zee, S. Savas, H.J. Mackay, E. Amir, G. Liu

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): L. Eng, A.K. Azad, S. Savas, G. Liu

Study supervision: A.K. Azad, E. Amir, G. Liu

This work was sponsored in part by the Alan B Brown Chair in Molecular Genomics, CCO Chair in Experimental Therapeutics and Population studies, and Posluns Family Foundation.