Identification of cis and trans acting genetic variants explaining up to half the variation in circulating VEGF levels

Abstract

Rationale

VEGF impacts angiogenesis, atherosclerosis and cancer. Although the heritability of circulating VEGF levels is high, little is known about its genetic underpinnings.

Objective

Our aim was to identify genetic variants associated with circulating VEGF levels using an unbiased genome-wide approach and explore their functional significance with gene expression and pathway analysis.

Methods and results

We undertook a genome-wide association study (GWAS) of serum VEGF levels in 3,527 participants of the Framingham Heart Study (FHS), with pre-planned replication in 1,727 participants from two independent samples, the STANISLAS Family Study (SFS) and the Prospective Investigation of the Vasculature in Uppsala Seniors study (PIVUS). One hundred and forty SNPs reached genome-wide significance (p<5×10⁻⁸). We found evidence of replication for the most significant associations in both replication datasets. In a conditional GWAS 4 SNPs mapping to 3 chromosomal regions were independently associated with circulating VEGF levels: rs6921438 and rs4416670 (6p21.1, p=6.11×10⁻⁵⁰⁶ and p=1.47×10⁻¹²), rs6993770 (8q23.1, p=2.50×10⁻¹⁶) and rs10738760 (9p24.2, p=1.96×10⁻³⁴). A genetic score including these four SNPs explained 48% of the heritability of serum VEGF levels. Six of the SNPs that reached genome-wide significance in the GWAS were significantly associated with VEGF mRNA levels in PBMCs. Ingenuity pathway analyses showed found plausible biological links between VEGF and 2 novel genes in these loci (ZFPM2 and VLDLR).

Conclusions

Genetic variants explaining up to half the heritability of serum VEGF levels were identified. These new insights provide important clues to the pathways regulating circulating VEGF levels.

Vascular endothelial growth factor (VEGF, also referred to as VEGFA in contrast to other members of the VEGF family) is pivotal in many physiological and pathological processes.¹ It is primarily known for its key role in the stimulation of angiogenesis, with a potent mitogenic effect on vascular endothelial cells from arteries, veins and lymphatics.² VEGF also promotes vasodilatation by inducing the production of nitric oxide and prostacyclin by endothelial cells.³ In addition, VEGF is involved in hematopoietic development and chemotaxis of monocytes, regulation of osteoclast differentiation, stimulation of surfactant production,¹ and has neurotrophic and neuroprotective effects on neuronal and glial cells.⁴ Elevated circulating VEGF levels have been observed in vascular disease (ischemic heart disease,^{5, 6} heart failure,⁷ stroke⁸), and in various other disorders, including diabetes,⁹ cognitive decline and dementia,^{10, 11} reproductive,^12–14 immune-inflammatory disorders,^15,16 and neoplastic diseases.^{17, 18} Administration of VEGF promotes angiogenesis in patients with critical leg ischemia, as well as in animal models of coronary and limb ischemia.¹⁹ VEGF inhibitors such as bevacizumab and sorafenib have been successfully used to inhibit angiogenesis in several tumors,^{20, 21} in macular degeneration²² and in rheumatoid arthritis.²³ However, despite the considerable toxicity associated with VEGF inhibitor drugs,²⁴ there have been no pharmacogenomic studies to identify potential subgroups of responders partly because the genetic determinants of VEGF concentrations remain poorly understood. Indeed, although the heritability of circulating VEGF levels is very high, ranging between 60 and 80%,^25–27 few studies have assessed the relation between circulating VEGF levels and genetic variants, yielding inconsistent results. The aim of the present study was to identify genetic variants associated with circulating VEGF levels using an unbiased genome-wide approach in a large community-based sample.

METHODS

Study populations

The Framingham Heart Study (FHS)

The FHS, initiated in 1948, is a three-generation, community-based, prospective cohort study conducted in Framingham, MA, USA.^33–35 Serum VEGF levels were measured in third generation cohort participants (2002–2005) and genome-wide genotyping was performed on these individuals at Affymetrix (Santa Clara, CA) through an NHLBI funded SNP-Health Association Resource (SHARe) project. We chose not to include participants with cardiovascular disease, as the latter may influence VEGF levels. After excluding participants who had prevalent cardiovascular disease, which may influence their VEGF levels, or failed to meet quality control standards, 3,527 participants were enrolled.

The STANISLAS Family Study (SFS)

The SFS is a 10-year longitudinal survey involving 1,006 volunteer families from Vandoeuvre-lès-Nancy, France, whose members were free of chronic disease (cardiovascular or cancer) between 1993–1995.²⁸ Plasma VEGF levels were measured at the second examination cycle (1998–2000) in a randomly selected subsample; of these 859 persons from 217 families, who also had DNA and met genotyping quality control criteria, were included.

Prospective Investigation of the Vasculature in Uppsala Seniors (PIVUS) study

The PIVUS study is a population-based study that enrolled 1,016 70-year old individuals living in the community of Uppsala, Sweden (2001–2004).²⁹ Of these, 999 persons provided DNA for genetic studies and after exclusions for prevalent cardiovascular disease and inadequate genotyping quality, 868 participants were eligible.

Further details of the study samples are available in the Supplemental Methods, section 1.

Laboratory measurements of VEGF levels

VEGF levels were measured in serum for the FHS and PIVUS and plasma for the SFS (Supplemental Methods, section 2).

Genotyping

FHS

Genome-wide genotyping in the FHS was performed on the Affymetrix GeneChip Human Mapping 500K Array Set® and 50K Human Gene Focused Panel®. Genotyping, quality control and imputation methods are detailed in the Supplemental Methods, section 3.

SFS

The SNPs were genotyped by Genoscreen© (http://genoscreen.fr) using a Sequenom® iPLEX Gold assay – Medium Throughput Genotyping Technology.³⁰

PIVUS study

The SNPs were genotyped as part of a 96-plex assay at the SNP technology platform in Uppsala University (http://www.genotyping.se/) using the Illumina BeadXpress system from Illumina Inc.³¹ Genotyping calls were done with Illumina BeadStudio software.

Statistics

VEGF levels were natural log-transformed to normalize their distribution.

Genome-wide association analysis in the FHS

A linear mixed effects model accounting for familial relatedness was used to evaluate the association of each SNP with VEGF levels.³² An additive genetic model with one degree of freedom was used. In a first step (model A), analyses were adjusted for age, sex, and the ninth principal component (Supplemental Methods, section 4). In a second step designed to explore potential mechanisms, we additionally adjusted our most significant associations for covariates previously found to be associated with serum VEGF levels:²⁵ compared to model A, model B was additionally adjusted for hypertension; model C for smoking; model D for central obesity and model E for the presence of a metabolic syndrome (Supplemental Methods, section 5).

Genetic association study in the SFS and the PIVUS study

In order to confirm our findings in the FHS, we genotyped 25 SNPs in two independent samples. To select a parsimonious number of SNPs for replication we used criteria of strength of association (p-value), whether the SNP was genotyped or imputed, linkage disequilibrium (LD) between SNPs and functionality (Supplemental Methods, section 6). A linear regression model using the same covariates and analytic strategy as in the FHS was implemented.

Joint analysis of the FHS, SFS and PIVUS study

For SNPs that were successfully genotyped in the SFS and the PIVUS study we performed a meta-analysis of the SNP-phenotype associations, using a fixed effects inverse-variance meta-analysis technique for the combination of results from the FHS and the PIVUS study (which had both measured VEGF levels in the serum) and an effective sample size weighted meta-analysis for the combination of results from all three studies, to account for the different scales of VEGF levels in serum and plasma (Supplemental Methods, section 7).

Genetic score

The methods used for computing a genetic score are detailed in the Supplemental Methods, section 8, and Online Table I. The phenotypic variance explained by this genetic score was separately calculated in the FHS, the SFS and the PIVUS study, using regression models that included age and sex as covariates.

VEGF gene expression analysis in peripheral blood mononuclear cells (PBMCs)

Sample preparation and quantification of the PBMC messenger RNA (mRNA) of VEGF spliced forms and statistical analyses of these data are described in the Supplemental Methods (section 10).

Biological pathway analysis

Methods for the biological pathway analysis are provided in the Supplemental Methods, section 9.

RESULTS

Characteristics of the 5,273 study participants are presented in Table 1.

Table 1.

Characteristics of Study Participants

Characteristics	FHS	SFS†	PIVUS Study
Number of participants	3,527	859	868
Mean circulating VEGF level (ng/L), median (IQR) *	280 (294.7)	27.4 (28.2)	187.5 (210.6)
Mean age (SD) at VEGF measurement, mean (SD)	40.0 (8.7)	29.83 (14.5)	70.2 (0.2)
Women (%)	1890 (53.2)	428 (49.8)	454 (52.3)
Cardiovascular risk factor at VEGF measurement
Systolic blood pressure, mean (SD)	116.7 (14.0)	120.3 (12.8)	149.6 (22.7)
Hypertension (%)	561 (15.9)	23 (2.7)	606 (69.8)
Diabetes mellitus (%)	89 (2.5)	0	68 (7.8)
Current smoker (%)	544 (15.3)	188 (21.9)	93 (10.7)
Central obesity (%)	1315 (37.2)	37 (44.3)	266 (30.6)
Metabolic syndrome (%)	693 (19.6)	19 (2.2)	198 (22.8)

IQR: Inter-Quartile Range; SD: Standard Deviation;

^*Serum levels for the FHS and the PIVUS study, and plasma levels for the SFS;

^†by design, SFS participants were free of chronic disorders (cardiovascular or cancer) and had no personal history of cardiovascular disease at the time of inclusion (VEGF levels and covariates for the present analysis were measured during the second examination cycle in 1998 – 2000); all individuals with CVD (cardiovascular disease), defined in the FHS as presence of stroke, congestive heart failure, coronary heart disease or intermittent claudication, were excluded before analyses in FHS and PIVUS

GWAS of VEGF levels in the FHS

The quantile-quantile plot showed an excess of extreme p-values but no evidence of systematic inflation of the genomic control inflation factor (λ=1.02) (Online Figure I). The genome-wide plot of p-values for the individual SNPs against their genomic position is shown in Online Figure II. A total of 140 SNPs cleared the threshold for genome-wide significance at 5×10⁻⁸ (Supplemental Table II). These were located in three chromosomal regions: 6p21.1, 8q23.1, 9p24.2 (Table 2). The most significant association was found with rs6921438 on chromosome 6p21.1 (p=6.11×10⁻⁵⁰⁶), at 171 kb downstream of the VEGF gene, and close to the mitochondrial ribosomal protein L14 gene (MRPL14) and the MCG45491 gene (C6orf223), encoding an uncharacterized protein. Sixty-seven other SNPs on chromosome 6p21.1 were also associated with VEGF levels at p<5×10⁻⁸ (Figure 1a). When running a conditional GWAS adjusting for rs6921438, one other SNP in 6p21.1 (rs4416670) still yielded a genome-wide significant association, suggesting that two variants in this region independently modulate VEGF levels. In the 8q23.1 region the SNP yielding the most significant association with VEGF levels (rs6993770, p=2.50×10⁻¹⁶) is located in the zinc finger protein, multitype 2 (ZFPM2) gene and 980.4 kb away from the low-density lipoprotein receptor-related protein 12 gene (LRP12). Forty-three SNPs in LD with rs6993770 were also associated with VEGF levels at p<5×10⁻⁸ (Figure 1b). A conditional GWAS adjusting for rs6993770, rs6921438 and rs4416670 did not yield any other genome-wide significant association in chromosome 8q23.1. The most significant association on 9p24.2 was observed with rs10738760 (p=1.96×10⁻³⁴), located close to the very low density lipoprotein receptor (VLDLR) and potassium voltage-gated channel subfamily V, member 2 (KCNV2) genes. Twenty-nine SNPs in LD with rs10738760 were also associated with VEGF levels at p<5×10⁻⁸ (Figure 1c). None reached genome-wide significance in a conditional GWAS adjusted for rs6921438, rs4416670, rs6993770 and rs10738760. We computed a genetic score including the four SNPs yielding genome-wide significant associations with VEGF levels in the conditional GWAS (Online Table I). This score explained 47.6% of serum VEGF variability (p=2.19×10⁻⁶⁴⁴).

Table 2.

Genome-wide significant single nucleotide polymorphism (SNP)-phenotype associations in genome-wide association analysis of circulating VEGF levels (p<5×10⁻⁸)

SNP	Chr	Position*	Function	CAF	Coded Allele	strand	beta† (FHS)	SE (FHS)	p (FHS)||	p (PIVUS)||	p (SFS)||	Dir	Meta-p (FHS+PIVUS)‡	Meta-p (all)§	Gene1	Distance (kb)	Gene2	Distance (kb)
rs6921438	6	44033585	intergenic	0.51	G	+	0.72	0.01	6.11×10−506	NA	2.93×10−39	-?-	6.11×10−506	4.44×10−524	MGC45491	42.7	MRPL14	155.8
rs4513773	6	44033504	intergenic	0.47	G	+	−0.70	0.01	2.08×10−482	6.23×10−139	NA	++?	4.45×10−619	1.10×10−584	MGC45491	42.8	MRPL14	155.8
rs9472159	6	44027673	intergenic	0.50	C	+	0.76	0.02	4.30×10−452	3.27×10−109	4.11×10−35		1.61×10−557	8.16×10−553	MGC45491	48.6	MRPL14	161.7
rs9369434	6	44026385	intergenic	0.53	C	+	0.84	0.02	2.15×10−442	1.43×10−63	5.31×10−28		1.43×10−490	1.21×10−496	MGC45491	49.9	MRPL14	163.0
rs1776717	6	44059314	intergenic	0.21	A	+	−0.23	0.02	8.10×10−20	3.74×10−4	9.75×10−6		3.28×10−22	1.07×10−26	MGC45491	17.0	MRPL14	130.0
rs1776721	6	43998961	intronic	0.31	T	+	−0.18	0.02	1.52×10−19	3.43×10−8	0.02		5.38×10−26	4.23×10−26	MGC45491	77.3	VEGF	136.8
rs1886979	6	44012879	3′UTR	0.41	G	+	0.17	0.02	3.71×10−19	3.23×10−6	0.01		6.55×10−24	1.70×10−24	MGC45491	63.4	VEGF	150.7
rs9472155	6	44005705	intronic	0.22	T	+	−0.20	0.02	4.45×10−19	3.93×10−9	0.01		2.50×10−26	1.51×10−26	MGC45491	70.6	VEGF	143.5
rs844294	6	44008685	intronic	0.52	C	+	−0.15	0.02	1.19×10−14	2.25×10−5	0.09	+++	1.41×10−18	2.46×10−18	MGC45491	67.6	VEGF	146.5
rs4416670	6	44058431	intergenic	0.55	T	+	0.13	0.02	1.47×10−12	0.10	2.87×10−4	+++	1.44×10−12	2.08×10−15	MGC45491	17.9	MRPL14	130.9
rs910611	6	44058829	intergenic	0.08	C	+	−0.26	0.04	2.61×10−10	6.36×10−6	0.11	+++	9.92×10−15	1.94×10−14	MGC45491	17.5	MRPL14	130.5
rs6993770	8	106650704	intronic	0.32	T	+	−0.17	0.02	2.50×10−16	3.99×10−8	0.02	+++	2.60×10−22	4.71×10−23	ZFPM2	0	LRP12	980.4
rs16873402	8	106658423	intronic	0.33	T	+	−0.15	0.02	1.97×10−14	9.49×10−9	0.16		1.10×10−20	5.32×10−20	ZFPM2	0	LRP12	988.1
rs16873365	8	106627411	intronic	0.22	T	+	−0.16	0.02	5.65×10−12	2.09×10−6	0.37		4.10×10−16	2.27×10−15	ZFPM2	0	LRP12	957.1
rs7013321	8	106662734	intronic	0.49	A	+	−0.14	0.02	6.75×10−12	NA	0.01	-?-	6.75×10−12	4.49×10−13	ZFPM2	0	LRP12	992.4
rs6993696	8	106650460	intronic	0.46	A	+	−0.13	0.02	8.54×10−12	1.49×10−4	0.05		6.18×10−15	2.12×10−15	ZFPM2	0	LRP12	980.1
rs16873291	8	106597206	intronic	0.31	T	+	−0.13	0.02	5.30×10−11	7.65×10−7	0.07		1.12×10−15	4.75×10−16	ZFPM2	0	LRP12	926.9
rs1349319	8	106625810	intronic	0.39	A	+	0.11	0.02	3.59×10−8	1.32×10−3	0.05	+++	1.99×10−10	3.53×10−11	ZFPM2	0	LRP12	955.5
rs10738760	9	2681186	intergenic	0.49	A	+	0.28	0.02	1.96×10−34	1.12×10−8	0.03	+++	4.46×10−41	9.93×10−40	KCNV2	26.3	VLDLR	36.7
rs6475920	9	2663933	intergenic	0.36	A	+	−0.24	0.02	3.76×10−32	2.40×10−8	0.02		6.11×10−39	7.93×10−38	VLDLR	19.4	KCNV2	43.6
rs4741756	9	2658187	intergenic	0.28	C	+	−0.25	0.02	2.95×10−31	8.64×10−5	0.09	+++	4.45×10−34	3.41×10−32	VLDLR	13.7	KCNV2	49.3
rs2375980	9	2682622	intergenic	0.42	G	+	−0.25	0.02	1.30×10−27	2.25×10−8	0.02	+++	2.55×10−34	1.01×10−33	KCNV2	24.9	VLDLR	38.1
rs10122587	9	2681951	intergenic	0.28	T	+	−0.22	0.02	3.02×10−24	NA	0.02	-?-	3.02×x10−24	4.67×10−24	KCNV2	25.6	VLDLR	37.5
rs10967492	9	2671175	intergenic	0.21	A	+	−0.22	0.02	1.02×10−21	NA	0.10	-?-	1.02×10−21	1.25×10−20	VLDLR	26.7	KCNV2	36.3
rs10967470	9	2665698	intergenic	0.24	G	+	−0.22	0.02	1.17×10−21	NA	0.04	+?+	1.17×10−21	2.79×10−21	VLDLR	21.2	KCNV2	41.8

CAF: Coded Allele Frequency; Chr: chromosome; Dir: Direction of association in FHS, PIVUS, SFS; Gene1: closest referenced gene; Gene2: second closest referenced gene; Meta-p: meta-analysis p-value; SNP: Single Nucleotide Polymorphism; CSE: Standard Error;

^*genome build 36.3;

^†effect estimate for the minor allele;

^‡inverse variance meta-analysis;

^§effective sample size weighted meta-analysis;

^||model A: adjusted for age and gender, as well as for the ninth principal component in FHS

Figure 1.

Regional plot for associations in region centered on rs6921438 (1a), rs6993770 (1b) and rs10738760 (1c). All SNPs (triangles) are plotted with their meta-analysis p-values against their genomic position. The color of the triangles represents the linkage disequilibrium between each of the SNPs in the region with rs6921438: purple: r2 ≤0.05, light blue: 0.05<r2 ≤0.10, green: 0.10<r2≤0.30, yellow: 0.30<r2≤0.60, orange: 0.60<r2≤0.80, red: r2>0.80. Light blue line represents estimated recombination rates. Genes are shown as dark green arrows. LD and recombination rates were drawn from Hapmap (release #22).

Replication studies

We sought to replicate our most significant results in two independent cohorts. Of the 25 SNPs selected for replication, 24 were successfully genotyped in the SFS and 20 in the PIVUS study (Table 2). Of these, 17 and 20 respectively reached nominal significance in association with VEGF levels, with the same direction of effect (Table 2). When meta-analyzing the results of the FHS and the PIVUS study, which both used serum VEGF levels, for the 19 SNPs genotyped in both studies, all 19 SNPs were associated with VEGF levels at p<0.05 (Table 2). The joint meta-analysis of results from all three studies, using an effective sample size weighted meta-analysis, is displayed in Table 2. There was statistically significant heterogeneity between studies for a few but not all SNPs in each locus, due to differences in effect size, but not in direction of effects (Supplemental Table III). The genetic score explained 16.6% (p=1.75×10⁻³⁶) of observed plasma VEGF variability in the SFS and 48.4% (p=3.31×10⁻¹⁸⁰) of observed serum VEGF variability in the PIVUS study. The observed associations remained unchanged in each of the three cohorts after adjusting for hypertension, current smoking, central obesity and metabolic syndrome (Online Table IV).

VEGF gene expression analysis

In order to better characterize the functional role of the SNPs identified in the GWAS we quantified mRNA expression of the two splice variants corresponding to the diffusible isoforms of VEGF, VEGF₁₂₁ and VEGF₁₆₅, in PBMCs of 220 SFS participants. The association of VEGF mRNA levels with the 24 SNPs successfully genotyped in the SFS was assessed.

At the nominal significance level, 1 SNP on chromosome 6p21.1, 4 SNPs on chromosome 8q23.1 and 1 SNP on chromosome 9p24.2 were associated with VEGF₁₂₁ mRNA levels (Table 3).

Table 3.

Significant associations between SNPs and VEGF transcripts

phenotype	SNP	Chr	position	CA	CAF	beta †	SE	p	h2q (%)
mRNA_121	rs16873365	8	106627411	T	0.22	22.71	7.22	0.002	4.73
mRNA_121	rs16873402	8	106658423	T	0.33	12.15	5.10	0.017	2.84
mRNA_121	rs6993770	8	106650704	T	0.32	12.06	5.23	0.021	2.82
mRNA_121	rs16873291	8	106597206	T	0.31	11.95	5.37	0.026	2.47
mRNA_121	rs2375980	9	2682622	G	0.42	10.15	4.75	0.032	2.03
mRNA_121	rs910611	6	44058829	C	0.08	−19.47	9.49	0.040	2.13

^*log-transformed;

^†effect estimate for the minor allele; CAF: Coded Allele Frequency; Chr: chromosome; h²_q: variance explained; SE: standard error

Biological pathway analysis

Using the Ingenuity Pathway Analysis software (IPA, Ingenuity Systems, www.ingenuity.com) we explored functional relationships between VEGF and the genes closest to the SNPs on chromosome 8q23.1 and 9p24.2 that were significantly associated with circulating VEGF levels. In each case we selected the genes closest to the identified SNPs, as in Table 2, to identify plausible biological pathways. We selected five focus genes: VEGF, ZFPM2, LRP12, VLDLR, KCNV2. The IPA network analysis identified relationships among three of these five focus genes (VEGF, ZFPM2 and VLDLR) as part of a larger network of 35 genes. The probability of finding 3 or more focus genes in a set of 35 genes randomly selected from the Global Molecular Network was p=10⁻⁸, suggesting that the presence of three of our five focus genes in this network was unlikely to occur by chance. In Figure 2 we present a subset of this network, including only interactions between VEGF and the two other focus genes in the network, with 2 or fewer intermediate nodes.

Figure 2.

Putative Gene Network based on Ingenuity Path Analysis. Edges are displayed with labels describing the nature of the relationship between the nodes. The lines between genes represent known interactions and the nodes are displayed using various shapes which represent the functional class of the gene product (legend).

DISCUSSION

Principal findings

In this first GWAS of circulating VEGF levels undertaken in 3,527 community individuals of European descent, we identified novel genetic associations: 140 SNPs reached genome-wide significance. Of these, 4 SNPs were independently associated with VEGF levels (rs6921438 and rs4416670 on chromosome 6p21.1, rs6993770 on chromosome 8q23.1 and rs10738760 on chromosome 9p24.2). We found evidence of replication for selected SNPs in 1,727 individuals of European descent from two independent community-based samples. The SNPs are located close to the VEGF and MRPL14 genes (chromosome 6p21.1), within the ZFPM2 gene (chromosome 8q23.1), and between the VLDLR and KCNV2 genes (chromosome 9p24.2). In a subset of participants we found that 6 of 25 selected SNPs yielding genome-wide significant associations with circulating VEGF levels were also associated with VEGF mRNA levels (VEGF₁₂₁ splice variant) in PBMCs.

In the context of the current literature

Genetic association study

Candidate gene studies exploring associations between VEGF polymorphisms and circulating VEGF levels have yielded controversial results (Supplemental Methods, section 11; Online Table V). Eight studies have found significant associations with candidate polymorphisms (rs699947, rs1570360, rs833061, rs2010963, rs3025039 and -2549 18bp I/D) in the promoter, 5′ and 3′ untranslated region of the VEGF gene.^33–40 However, several other studies did not identify any association with these and other VEGF SNPs (Online Table V). Using a hypothesis-free genome-wide approach, the present study revealed novel associations with 140 SNPs. Of these, 68 SNPs are located on chromosome 6 approximately 150 kb downstream from the 3′ end of the VEGF gene, far from previously tested candidate SNPs. Although we do replicate previously described associations of 2 VEGF promoter polymorphisms (rs699947 and rs833061) with serum VEGF levels at p<5×10⁻⁷ (Supplemental Methods, section 11; Online Table V; Online Figure III), none of the SNPs that reached genome-wide significance in our analysis, on chromosome 6p21.1, 8q23.1 and 9p24.2, had been examined previously in relation with circulating VEGF levels.

Transcriptomic analysis

While several studies have examined the association of candidate genetic variants with VEGF gene expression in pathological tissues,^{41, 42} little is known about the genetic variants influencing VEGF expression in normal cells. Our data suggest that six of the SNPs associated with circulating VEGF levels in our GWAS also modulate the expression of the VEGF₁₂₁ splice variant in PBMCs of community-based persons. The diffusible VEGF isoforms, VEGF₁₆₅ and VEGF₁₂₁, are released by a variety of tumor and normal cells, including PBMCs.⁴³ VEGF₁₂₁ lacks a heparin-binding domain and has a higher migration but lower mitogenic potency than VEGF₁₆₅.⁴⁴

Potential mechanisms mediating observed genetic associations

Our data suggest that almost half the inherited component of circulating VEGF levels is explained by genetic variants located downstream from the VEGF gene on chromosome 6p21.1. The conditional GWAS suggests that this region could harbor at least two distinct loci that are independently associated with circulating VEGF levels. Although located relatively far from the VEGF gene, results from our transcriptomic analysis indicate that this region could indeed contain functional variants modulating VEGF gene expression. Genome-wide significant associations with circulating VEGF levels were also identified for SNPs located on chromosome 8q23.1 and 9p24.2. Although these trans effects explain a much smaller proportion of the heritability of VEGF levels, they provide important clues about the pathways involved in the regulation of VEGF expression. The SNPs on chromosome 8q23.1 are located in introns 4 and 5 of the ZFPM2 gene. This gene encodes a widely expressed member of the Friend of GATA family of transcription factors that modulate the activity of the GATA family proteins, which are important regulators of embryogenesis and also seem to play a significant role in endothelial cell biology.^45–47 The second closest gene to the SNPs identified on chromosome 8q23.1 is LRP12, encoding a low-density lipoprotein receptor-related protein that interacts with proteins related to signal transduction pathways and is differentially expressed in many cancer cells. The SNPs on chromosome 9p24.2 are located between the VLDLR and KCNV2 genes. VLDLR encodes a lipoprotein receptor involved in the metabolism of apolipoprotein-E-containing triacylglycerol-rich lipoproteins. Like VEGF, VLDLR appears to modify the risk of developing age-related macular degeneration,⁴⁸ and recent data suggest that VLDLR could play a central role in a network of interacting angiogenic genes activated in response to hypoxia.⁴⁹ KCNV2 encodes a member of the potassium voltage-gated channel subfamily V involved in regulation of neurotransmitter release, neuronal excitability and heart rate. While the present data does not permit us to formally determine which of these genes underlie the observed SNP associations with circulating VEGF levels, our in silico biological pathway analysis suggests that ZFPM2 and VLDLR are the most likely candidates. Further research is needed to explore the mechanisms underlying the associations of cis and trans acting genetic variants with circulating VEGF levels, such as modulation of gene expression, differential splicing or mRNA degradation.

Strengths and limitations

The findings from this first GWAS of circulating VEGF levels emphasize the importance of screening for genetic variation modulating biomarker levels not only within and in close proximity to the gene encoding the protein under investigation, but also in more distant potentially regulatory regions, including on other chromosomes. The strength of the observed associations and the fact that we were able to replicate our findings in two independent cohorts suggest that these associations are real. This is further supported by the association of several of these genetic variants with VEGF gene expression in PBMCs. Our study also had several limitations. Whereas focusing on white populations of European descent has the advantage of minimizing potential population stratification issues, our findings cannot be generalized to other ethnic groups. We may not have identified the true causal variants but merely SNPs in LD with the latter. Plasma levels of VEGF were measured in one of the replication cohorts (SFS), while serum VEGF levels had been measured in the discovery cohort. Serum VEGF concentrations are higher than plasma concentrations due to the release of VEGF from platelets during the clotting process.⁵⁰ Although the vast majority of associations found in the FHS did replicate in the SFS, suggesting an important overlap between genetic susceptibility factors of serum and plasma VEGF levels, the lower proportion of VEGF variability explained by the genetic score in the SFS compared to the FHS and PIVUS study may be related to differences in plasma and serum VEGF concentrations. Finally, although our transcriptomic analysis does provide some support for a functional role of SNPs associated with mRNA levels of VEGF₁₂₁, these results are exploratory and were not corrected for multiple testing. We may have been underpowered for this analysis due to limited sample size and also because VEGF expression was measured on PBMCs only, which are not the sole contributors to circulating VEGF levels. Further studies looking at the association of genetic markers with VEGF expression in other cell types, including endothelial cells, would be of great value.

Clinical Implications

VEGF plays a key role in various diseases including atherosclerosis, inflammatory and neurodegenerative disorders, and cancer.¹ Anti-VEGF and proangiogenic VEGF based treatments have recently been developed for several therapeutic indications.^19–23 The identification of polymorphisms linked to VEGF levels could help in identifying patients who are more likely to respond favorably to such treatments. These therapies can have major side-effects,²⁴ and optimizing the risk-benefit ratio of their administration could lead to substantial improvements in patient care. The discovery of trans-acting genetic variants influencing VEGF levels could also spur the discovery of new molecular targets for pro- or anti-angiogenic therapies.

Conclusions

In a large population-based sample of European ancestry we identified novel genetic variants associated with circulating VEGF levels, on chromosome 6p21.1, 8q23.1 and 9p24.2, which explain almost half of the observed phenotypic variation.

Novelty and Significance.

What is known?

• Circulating level of Vascular Endothelial Growth Factor (VEGF) is a heritable trait (60–80% heritability). The underlying genetic variants have not been identified.

• VEGF has important physiologic and pathophysiological roles in promoting and supporting angiogenesis, as a vasodilator and neurotrophic factor. It also promotes atherosclerosis.

• VEGF antagonists are used to treat macular degeneration and certain cancers.

• VEGF agonists may help treat limb ischemia.

What new information does this article contribute?

• Four independent novel loci are identified, using an unbiased genome-wide association study (GWAS) approach combined with mRNA expression and Ingenuity pathway analysis.

• The novel loci explain nearly half the observed variation in serum VEGF levels, in population-based samples of European descent, who were free of cardiovascular disease.

• Two novel loci are downstream of the VEGFA gene on chromosome 6 and may be transcription regulators.

• Two other novel ‘trans’ loci are located on chromosomes 8q23.1 and 9p24.2. Potential candidate genes at these loci are zinc finger protein multitype 2 (ZFPM2) and very low density lipoprotein receptor (VLDLR).

VEGF has a pathophysiologic role in atherosclerosis, inflammatory and neurodegenerative disorders, and cancer. Circulating VEGF levels are strongly heritable trait but the underlying genetic variants are unknown. Understanding the genetic determinants of VEGF levels is of potential pharmacogenomic importance as VEGF antagonists (with inter-individual differences in toxicity and efficacy) are used to treat macular degeneration, colon cancer and other conditions. We undertook GWAS of serum VEGF levels in 3527 community-based, Framingham Heart Study participants and replicated our findings in two other healthy samples, the Stanislas Family Study and the Prospective Investigation of Vasculature in Uppsala Seniors. We identified 4 novel loci across 3 chromosomes: 6p21.1, 8q23.1 and 9p24.2 that together explain 48% of the observed variability in serum VEGF. In exploratory analyses 6 of 24 SNPs studied were associated with PBMC mRNA expression of VEGF. Further Ingenuity pathway analyses revealed that genes adjacent to the two trans loci, ZFPM2 on chromosome 8 and VLDLR on chromosome 9 are linked to VEGF along plausible biological pathways These findings highlight the potential importance of distant regulatory regions in determining biomarker levels. These results might help target VEGF based treatments and spur the discovery of new molecular targets for pro- or anti-angiogenic therapies.

Non-standard Abbreviations and Acronyms

FHS

Framingham Heart Study

GWAS

Genome-Wide Association Study

IPA

Ingenuity Pathway Analysis

KCNV2

Potassium voltage-gated Channel subfamily V, member 2

LD

Linkage Disequilibrium

LRP12

Low-density lipoprotein Receptor-related Protein 12

mRNA

messenger RNA

MRPL14

Mitochondrial Ribosomal Protein L14

PBMS

Peripheral Blood Mononuclear Cells

PIVUS

Prospective Investigation of the Vasculature in Uppsala Seniors study

SFS

STANISLAS Family Study

SHARe

SNP-Health Association Resource

SNP

Single Nucleotide Polymorphism

VEGF

Vascular Endothelial Growth Factor

VLDLR

Very Low Density Lipoprotein Receptor

ZFPM2

Zinc Finger Protein, Multitype 2