The interleukin-6 receptor as a target for prevention of coronary heart disease: a mendelian randomisation analysis

Summary Background A high circulating concentration of interleukin 6 is associated with increased risk of coronary heart disease. Blockade of the interleukin-6 receptor (IL6R) with a monoclonal antibody (tocilizumab) licensed for treatment of rheumatoid arthritis reduces systemic and articular inflammation. However, whether IL6R blockade also reduces risk of coronary heart disease is unknown. Methods Applying the mendelian randomisation principle, we used single nucleotide polymorphisms (SNPs) in the gene IL6R to evaluate the likely efficacy and safety of IL6R inhibition for primary prevention of coronary heart disease. We compared genetic findings with the effects of tocilizumab reported in randomised trials in patients with rheumatoid arthritis. Findings In 40 studies including up to 133 449 individuals, an IL6R SNP (rs7529229) marking a non-synonymous IL6R variant (rs8192284; p.Asp358Ala) was associated with increased circulating log interleukin-6 concentration (increase per allele 9·45%, 95% CI 8·34–10·57) as well as reduced C-reactive protein (decrease per allele 8·35%, 95% CI 7·31–9·38) and fibrinogen concentrations (decrease per allele 0·85%, 95% CI 0·60–1·10). This pattern of effects was consistent with IL6R blockade from infusions of tocilizumab (4–8 mg/kg every 4 weeks) in patients with rheumatoid arthritis studied in randomised trials. In 25 458 coronary heart disease cases and 100 740 controls, the IL6R rs7529229 SNP was associated with a decreased odds of coronary heart disease events (per allele odds ratio 0·95, 95% CI 0·93–0·97, p=1·53×10−5). Interpretation On the basis of genetic evidence in human beings, IL6R signalling seems to have a causal role in development of coronary heart disease. IL6R blockade could provide a novel therapeutic approach to prevention of coronary heart disease that warrants testing in suitably powered randomised trials. Genetic studies in populations could be used more widely to help to validate and prioritise novel drug targets or to repurpose existing agents and targets for new therapeutic uses. Funding UK Medical Research Council; British Heart Foundation; Rosetrees Trust; US National Heart, Lung, and Blood Institute; Du Pont Pharma; Chest, Heart and Stroke Scotland; Wellcome Trust; Coronary Thrombosis Trust; Northwick Park Institute for Medical Research; UCLH/UCL Comprehensive Medical Research Centre; US National Institute on Aging; Academy of Finland; Netherlands Organisation for Health Research and Development; SANCO; Dutch Ministry of Public Health, Welfare and Sports; World Cancer Research Fund; Agentschap NL; European Commission; Swedish Heart-Lung Foundation; Swedish Research Council; Strategic Cardiovascular Programme of the Karolinska Institutet; Stockholm County Council; US National Institute of Neurological Disorders and Stroke; MedStar Health Research Institute; GlaxoSmithKline; Dutch Kidney Foundation; US National Institutes of Health; Netherlands Interuniversity Cardiology Institute of the Netherlands; Diabetes UK; European Union Seventh Framework Programme; National Institute for Healthy Ageing; Cancer Research UK; MacArthur Foundation.

similar), but others recorded only self-reported events. We included all events in the analysis -self-reported and validated -since self-reported data was the lowest common denominator between studies (Supplementary Table 3 Where data allowed, only the first recorded cardiovascular event (of any type) was included and all subsequent events excluded from the analysis. Prevalent events were defined as those occurring prior to study baseline and incident events as any event occurring during follow-up.

SNP selection
We aimed to compile the smallest set of variants to yield the most data on effects of IL6R genotype and therefore act as the most robust instruments for variation in the gene. A number of open-access populations of European ancestry were searched via the Genome Variation Server and the International HapMap Project (Release 23a/Phase 3) 51 for SNP within and 20kbp up-and downstream of IL6R. Using HaploView 52 and existing genotype data from Whitehall II we identified tagging SNPs present on the IBC HumanCVD BeadChip v2 (Cardiochip). SNPs were included on the HumanCVD BeadChip (Cardiochip) platform array -a candidate gene-based SNP array including a high density of markers in nearly 2,000 genes implicated in CVD -in part for their ability to tag genetic variation at this locus. Of those included on Cardiochip, we excluded SNPs whose associated p-value for univariate linear regression of circulating log e IL-6 levels on genotype conducted using PLINK in Whitehall II participants was greater than p=1x10 -5 in an additive model. For each SNP we also assessed the beta-coefficient for the above linear regression model and the minor allele frequency in Whitehall II and the HapMap CEU samples. Using data from the 1000 Genomes Project Pilot 1, we evaluated linkage disequilibrium (LD) between each of the resulting shortlisted SNPs and between those and SNPs identified by previous genome-wide association studies of quantitative traits, giving preference to SNPs capturing data for previously identified SNPs or multiple IL-6-associated SNPs in Whitehall II. Where possible we excluded SNPs in strong LD (r 2 >0.9) to assemble the final subset of SNPs for genotyping across the collaborating studies. The consequence and function of each selected SNP was assessed using HapMap, NCBI SNP database (http://www.ncbi.gov.uk/snp), Ensembl (http://www.ensembl.org), UCSC Genome Browser 53 .

Genotyping and genotype data quality control
Genotypes were coded as 0, 1 or 2 homozygous common allele, heterozygous and homozygous minor allele respectively.
We included existing genotype data where available and commissioned new genotyping where necessary.

Participant inclusion criteria
We restricted the analysis to unrelated individuals with available data for genotype at the lead SNP (or one of its proxies), and for at least one biomarker or clinical outcome.

Relatedness
Where study family structure was recorded, in each group of related participants all but the eldest of the group were excluded from the analyses.

SNP associations with continuous risk factors and circulating biomarkers
Analyses were based on an additive genetic model justified on the basis of the association of the three IL6R SNPs including rs7529229 (the most widely typed of these SNPs across all studies) with the circulating concentration of IL-6. Weighted mean difference between genotypes was calculated using the 'metan' command in Stata v 11.2, applying inverse-variance weighted fixed effects meta-analysis to the mean of each biomarker by genotype. Using the common homozygous group as a reference, the difference between the reference and 5 heterozygous and rare homozygous groups was estimated. Where biomarkers were log e transformed, the proportional difference between geometric means was calculated. Linear regression models (using the 'regress' command in Stata and equivalent procedures in other analysis packages) were used to estimate differences (and 95% confidence intervals, CI) in biomarkers per additional minor allele possessed at each SNP. For biomarkers that had been log e transformed, the beta-coefficient from such a regression model was exponentiated to find the proportional difference in geometric means associated with possession of each additional minor allele. Within study estimates of per-allele effects were combined using Mantel-Haenszel fixed effects meta-analysis.

Subgroup analyses
SNP associations with continuous biomarkers and with clinical outcomes were analysed with stratification by the following subgroups: males, females, individuals aged <55 years, 56-65 years and >65 years; individuals with and without CVD at the time of biomarker measurement; those with and without type-2 diabetes mellitus; users and non-users of lipid-lowering drugs; individuals with 10 year Framingham CHD risk of above and below 20%; individuals stratified by tertiles of non-HDL-cholesterol; body mass index <25kg/m 2 , 25-30kg/m 2 and >30kg/m 2 ; genotyping platforms and study designs. Tests of interaction between subgroups were performed using the chi-squared test for heterogeneity within the 'metan' command in Stata.

SNP associations with clinical outcomes
We estimated for each disease endpoint the odds ratio (OR) and 95% CI per minor allele for each SNP (rs7529229, rs4845371, rs12740969) under an additive model, as above, where data were available. Per-allele odds ratios were calculated by logistic regression models (using the 'logit' command in Stata) and resulting odds ratios combined using inverse-variance weighted fixed effects meta-analysis to derive a summary estimate (the 'metan' command in Stata v11.1). These analyses included only studies where at least one case and one control for the relevant outcome were available. Sensitivity analysis using different genetic models (heterozygous vs. common allele homozygous, and minor allele homozygous vs. common allele homozygous) were also conducted (data available on request).
In studies where all participants were diabetic, individuals free of diabetes from EAS were used as controls for ET2DS and from Whitehall-II for EDSC and UDACS when calculating odds ratios by logistic regression with type 2 diabetes as the outcome.

Search strategy
We searched PubMed on 7 September 2010 using the keyword "tocilizumab" and supplemented this by a search of reference lists of existing reviews and a systematic review of tocilizumab treatment published by the Cochrane Collaboration 55 .

Data extraction and analysis
Data were extracted by MVH and information extracted from a random selection of eligible studies was checked by DIS. CRP was selected as a biomarker by which the magnitude of tocilizumab's treatment effect could be indexed since it was the most widely reported at the greatest number of treatment doses and timepoints for which data were available in the genetic studies. For continuous variables other than CRP, the difference from baseline in the active arm was compared to the difference in baseline of placebo for 8mg/kg vs.
placebo RCT only. The most parsimonious time-points were chosen to maximise data yield across studies and also to prevent studies from being represented twice. For CRP, values were extracted for RCTs and cohort studies investigating 4mg/kg, 8mg/kg or 16mg/kg TCZ versus placebo at 8-12 weeks following initial treatment. For RCT, the difference in CRP from baseline in the intervention arm was compared to that of the difference in the placebo group. For cohort studies, only the difference between baseline and follow-up was compared (as there was no comparator arm). We abstracted data for all clinically relevant biomarkers and safety endpoints. Trial results were combined with Mantel-Haenszel (binary traits) and inverse variance (continuous traits) fixed effects meta-analysis performed using Stata 11.2 (StataCorp, Texas, USA).

IL-6, CRP, fibrinogen and blood lipid concentrations
Four studies were conducted exclusively in men (BRHS, CaPS, HIFMECH, and NPHS-II) and two exclusively in women (BWHHS and WHI), all remaining studies contributed participants from both genders (with the proportion of men in the range 18% and 75%).
Study-specific geometric mean concentrations (approximate SD) were in the range 1.01 (0.66) to 3.49

SNP selection, quality control & genotype
The Genome Variation Server identified 44 SNPs and the HapMap database 67 SNPs in the region of IL6R, of which 26 and 24 respectively passed the MAF threshold of 30%. Of the HapMap variants, 13 were designated tagging SNPs. Of the 42 SNPs in IL6R included on the CardioChip, 12 met the p-value threshold of p<1x10 -5 for association with circulating log e IL-6 at phase 3 of the Whitehall II study. Strong linkage disequilibrium (r 2 >0.9) was noted between one of the CardioChip SNPs and SNPs identified by GWAS studies for C-reactive protein 56 and soluble IL-6 receptor 57 . Evaluation of effect size (indicated by linear regression beta-coefficient and R 2 values) and linkage disequilibrium between SNPs around the IL6R gene (Supplementary figure 3) suggested an ultimate subset of 3 SNPs. Of these, 2 (rs7529229 and rs4845371) were located within IL6R itself and one (rs12740969) in TDRD10, 34kbp from IL6R. The latter SNP was selected because it captured information on variation at the 3' end of IL6R. Carriage of the minor allele at rs7529229 resulted in higher circulating levels of IL-6, whilst the minor alleles at rs4845371 and rs12740969 were associated with lower IL-6 levels. 8 The ass ociations o f t he o ther IL6R S NPs ( rs4845371, rs12 740969) with IL-6, CRP and fibrinogen w ere consistent with but opposite in direction to the findings for the rs7529229 variant (Supplementary figure 8).

Reporting of IL6R variants associated with safety endpoints.
The NHGRI cat alogue of genom e-wide association studies lists 8 G WA studies of colorectal cancer (searched April 2011), one study reported SNPs in IL6R though there was no association of those with cancer 58 (Figure   4).

Association of IL6R variants with CHD events
Estimates for the association of the IL6R variant with risk of CHD were consistent between the present de novo analysis and previously published case-control study data for the rs4537545 variant in LD (r 2 =1.0 in the HapMap CEU sample) with rs7529229 1 (Figure 3). Collaborating study

Footnote:
All cohort studies were prospective The MRC Fenland Study is a single cohort though was genotyped using two platforms with different individuals in each group and therefore is reported as two distinct samples here. C-C : case-control; C-Ct: case-cohort; HSE -Health Survey for England *Mean age in controls n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a SNPs are ranked on p-value and β-coefficient; β-coefficients are reported on the log scale.
*There was no variation in genotype at these very rare SNPs in the Whitehall II study sample and consequently a regression model could not be fitted.    Proteolytic cleavage of the mIL-6R is increased with carriage of the Ala allele resulting in greater soluble concentration of the soluble receptor and reduced numbers of functioning mIL-6R. Both classical (through the membrane-bound IL-6R) and trans-signalling (through the soluble IL-6R) pathways are active with either allele, however the preferential production of sIL-6R with the Ala allele results in increased trans-signalling and reduced classical signalling. The phenotypic effects associated with the Ala allele (and the minor allele at rs7529229) are consequences of this alteration in receptor production.

Supplementary
b. Tocilizumab blocks both the mIL-6R and the sIL-6R, inhibiting IL-6 signalling via both classical and trans-signalling pathways.
c. Comparison between the directions of effects of the IL6R variant and tocilizumab reflect the similarities and differences in the biological processes of the two scenarios.   Estimates are based on pairwise comparison of individuals heterozygous or homozygous for the variant allele with reference to the wildtype homozygous group. The total number of participants (studies) is also shown.  Figure 9 -Stratified associations of IL6R rs7529229 with fatal and non-fatal coronary heart disease. Estimates are stratified by participant age, sex, the presence of CVD or diabetes at baseline, BMI, estimated risk of CVD based on the Framingham equation and by tertile of non-HDL-C and log CRP. The p-value from a chi-squared test of heterogeneity is also presented.