This is the first study to conduct a systematic assessment of the interrelationships among circulating levels of IGF-I and IGFBP-3, common genetic variants at these loci, and risk of CAD. By contrast with previous smaller studies that have assessed the relationship between circulating IGF-I and IGFBP-3 levels and risk of cardiovascular disease, our results-from the largest prospective study to date-suggest that circulating levels of IGF-I and IGFBP-3 and genetic variants at the IGF1 and IGFBP3 genes do not materially influence the risk of CAD.
Our findings suggesting a lack of association between circulating IGF-I and IGFBP-3 levels and risk of CAD differ from those in a population -based nested case-control study in Denmark that reported a positive association between circulating IGFBP-3 levels and ischemic heart disease risk [14
] and in the Cardiovascular Health Study, where low circulating IGFBP-3 levels were associated with an increased risk of coronary events in elderly individuals [15
]. The Danish study and other prospective studies have also reported inverse associations between circulating IGF-I levels and risk of ischemic heart disease [14
] and heart failure [20
], although a recent study has found a positive association between IGF-I levels and risk of heart failure, but no association with overall incidence of cardiovascular disease [13
]. However, our results are consistent with three recent reports based on prospective cohort studies, two of which were conducted in women only [17
]. The studies above are small and therefore subject to random variability, with no previous individual study exceeding 550 cases. Furthermore, previous studies have been conducted in relatively disparate populations (e.g. three were conducted in relatively elderly populations [13
]). Such diversity highlights potential limitations in combining the findings of all available studies.
In addition to other known limitations of observational studies, such as incomplete control of confounders and bias, these apparently divergent findings may be due to several factors-heterogeneity in phenotype, age and sex of participants, heterogeneity among IGF-I and IGFBP-3 immunoassays, and pleiotropic effects of IGFs and their binding proteins that may have different actions in different contexts. We did not measure IGFBP-1 in EPIC-Norfolk, and it is possible that circulating levels of this binding protein may modify the relationships of other IGFs and clinical biomarkers on disease, which has been shown in the context of the metabolic syndrome [42
]. Although we measured circulating IGFs in samples that were non-fasting and were measured only once at baseline, our study showed the expected associations between age and sex and circulating IGF-I and IGFBP-3 levels and consistent replication of several SNP associations with circulating levels, thus decreasing the likelihood of substantial random measurement error. Furthermore, intra-individual variation in levels of IGFs is low. Indeed, it has been reported in 249 participants with serial measurements (mean interval three years) that circulating levels of IGFs have high levels of within-person correlation (r > 0.8) over time [15
We identified several SNPs at the IGF1
loci that are associated with circulating IGF-I and IGFBP-3 levels. At the IGF1
locus, five tSNPs were associated with circulating IGF-I levels. The associations for rs1520220, rs6220 and rs3730204 are consistent with previous reports [37
]. In a multivariate analysis of the IGF1
tSNPs, we found that rs1520220 and rs3730204 were sufficient to account for the other three tSNP associations with circulating IGF-I levels. tSNP rs3730204 is located in the 3' untranslated region (UTR) of the IGF1
gene (, thus it is a good functional candidate as it could potentially influence post-transcriptional processing of IGF1
mRNA. However, this SNP is a low-frequency variant (MAF 0.02) and, with only two rare homozygotes in our dataset, it is likely that our observed association is inflated. As rs1520220 is located in intronic sequence it is more likely that this SNP is capturing an untyped functional variant elsewhere in the gene.
We found three tSNPs at the IGFBP3
gene that were associated with circulating IGFBP-3 levels. Again, associations for tSNPs rs2132571 and rs2854744 are consistent with previous reports [37
]. In a multivariate analysis of the three associated IGFBP3
tSNPs we found that rs2132571 explained the other two tSNP associations with circulating IGFBP-3 levels at this locus in EPIC-Norfolk, although the results from our haplotype analysis suggest that rs2132571 and rs3793345 jointly capture an untyped functional variant with a substantial effect on circulating IGFBP-3 levels (). SNP rs2132571 lies 5' of the IGFBP3
locus and is moderately correlated with rs2854744 in EPIC-Norfolk (r2
0.61), thus it may be capturing some residual signal from this SNP that has been shown in cellular expression studies to directly influence promoter activity of the IGFBP3
Some of the disparities in the associations observed between specific SNPs and IGF-I and IGFBP-3 levels among studies are likely due to sampling variation and subtle differences in LD across these loci among populations. This reflects the fact that we are not directly assessing the causal variant(s) underlying these associations, perhaps with the exception of rs2854744 that shows very reproducible associations across multiple studies and ethnic groups and has evidence to support its functionality. Cellular expression studies will be required to assess the functionality of other associated tSNPs at both of these loci.
To our knowledge, our study is the first to assess the risk between common SNPs at the IGF1
loci with risk of CAD in a large prospective study. The strengths of our study are its size, and that we used a systematic tagging approach to capture all common SNPs (MAF ≥ 0.05) at the IGF1
loci. However, it is possible that there are rare variants at these loci that we have not captured in our current analysis and we may not have had sufficient statistical power to detect modest SNP associations with CAD risk. An assessment of the coverage between our SNPs and those in HapMap suggests that, at an r2
threshold of 0.8, we captured 45% and 66% of the variation above MAF 1% in IGF1
respectively spanning +/-10 kb of the loci, although this will be an underestimation because several of our tSNPs are not in HapMap. We may also have missed some additional associations between SNPs and circulating levels due to exclusion of three tSNPs that deviated from HWE in our samples and six SNPs which failed assay design. We did not genotype the microsatellite polymorphism in the promoter region of the IGF1
gene that has been reported to be associated with circulating IGF-I levels and risk of myocardial infarction, although these associations are inconsistent [53
]. A recent paper by Dupuis et al has identified an association between SNP rs35767 and glycaemic traits, with the G allele associated with an adverse glycaemic profile [58
]. Although the association between this SNP and circulating IGF-I levels was not statistically significant in our study, the corresponding allele (coded as C in our data) does show a trend with lower IGF-I levels (), consistent with these findings. Two recent reports by Palles et al and Schumacher et al have found a similar association between SNP rs35767 and IGF-I levels [59
], so our lack of statistical association may have been simply due to insufficient sample size or variation among populations as discussed above. We did not find any evidence for association with SNP rs35767 and CAD risk in around 9,000 cases and 20,000 controls (), but given the small magnitude of the associations with glycaemic traits in Dupuis et al, it is possible that there might be a weak association with CAD that could be detected with a larger sample. Palles et al have also identified three SNPs that lie approximately 70 kb 5' to the IGF1
gene that are associated with circulating IGF-I levels [59
]. These SNPs lie near to a putative transcription factor binding site that regulates IGF1
gene expression [59
], thus there are likely to be additional variants both cis
that might affect the expression of IGFs that will need to be characterised by further resequencingand replication efforts. However, the comprehensive SNP tagging approach that we have used in our analysis is likely to have captured the majority of the common variation at the IGF1
loci that will be relevant to the general population.
In summary, our results suggest that circulating levels of insulin-like growth factor-I (IGF-I), its major binding protein IGF-binding protein 3 (IGFBP-3) and genetic variants at the loci encoding these proteins do not influence the risk of CAD.