Abstract
Structural variation in retinal blood vessels is associated with global vascular health in humans and may provide a readily accessible indicator of several diseases of vascular origin. We report a meta-analysis of genome-wide association studies (GWAS) for quantitative retinal vascular traits derived using semi-automatic image analysis of digital retinal photographs from the GoDARTS (n=1736) and ORCADES (n=1358) cohorts. We identified a novel genome-wide significant locus at 19q13 (ACTN4/CAPN12) for retinal venular tortuosity, and one at 13q34 (COL4A2) for retinal arteriolar tortuosity; these two loci were subsequently confirmed in three independent cohorts. In the combined analysis, the lead SNP at each locus was rs1808382 in ACTN4/CAPN12 (P=2.39×10−13) and rs7991229 in COL4A2 (P=4.66×10−12). Notably, the ACTN4/CAPN12 locus associated with retinal venular tortuosity traits is also associated with coronary artery disease and heart rate. Our findings demonstrate the contribution of genetics in retinal vascular traits, and provide new insights into vascular diseases.
Retinal vascular traits can be readily measured non-invasively from fundus images and have been linked to a number of clinical conditions associated with vascular health1 including diabetes mellitus2, stroke3, cardiovascular disease4,5, hypertension6, and neurodegenerative disease7,8. Understanding the genetic determinants of retinal vascular traits may contribute to greater understanding of molecular mechanisms involved in determining disease risks and progression. Recent genome-wide association studies (GWAS) reported loci for widely investigated retinal traits including the central retinal vein equivalent (CRVE)9–11 and the retinal arteriolar equivalent (CRAE)10, and optic disc morphology12–15. While retinal vascular tortuosity has to be associated with a range of cardiovascular risk factors16–19, to our knowledge no studies have performed GWAS on this trait.
We carried out a GWAS to examine the underlying genetic factors influencing the retinal tortuosity traits (arteriolar tortuosity (TortA), maximum TortA (TortAmax), venular tortuosity (TortV), and maximum TortV (TortVmax)) and other retinal vascular traits including CRAE, CRVE, Arteriole-to-Venule ratio (AVR), as well as Optic Disc radius (ODradius). Two independent discovery cohorts were included; patients with type 2 diabetes from the Genetics of Diabetes Audit and Research in Tayside Study (GoDARTS, n=1736) and a population-based sample comprising the Orkney Complex Disease Study (ORCADES, n=1358). In both cohorts, traits were measured from retinal fundus images (Figure 1) using VAMPIRE 3.120,21 (Vascular Assessment and Measurement Platform for Images of Retina), which enables efficient, semi-automatic measurement of the retinal vasculature from large numbers of images. The VAMPIRE methodology used in the discovery stage has been reported in detail21–23. The study design and characteristics of the discovery cohorts are shown in Supplementary Figure 1, and Supplementary Table 1.
In the discovery stage, we performed a GWAS using the GoDARTS cohort for each retinal trait separately and tested the additive effect of each variant, adjusted for age, gender and the first three principal components. Similarly, GWAS was performed for the same traits in the ORCADES cohort, using a mixed model to account for kinship. We combined the summary results from these two cohorts for each trait using a fixed effect meta-analysis. Supplementary Table 2 summarize the single nucleotide polymorphisms (SNP) associated with retinal traits from meta-analysis as well as independent cohort results. Manhattan plots, QQ plots and regional plots are shown in Supplementary Figure 2, 3, and 4, respectively.
This analysis revealed one genome-wide significant SNP associated with TortA at 13q34 (lead SNP rs56399312 near COL4A2; Beta=0.182, SE= 0.032, P= 2.70 ×10−8, and another SNP rs9515212 near COL4A2 that was just below the threshold for genome-wide significance; Beta=0.151, SE= 0.028, P= 8.59 ×10−8). Conditional analysis on the lead SNP indicated that these are not independent signals (Supplementary Table 3). Two novel genome-wide significant loci (p<5×10−8) were associated with TortV, including SNPs at 19q13 near ACTN4 (lead SNP rs1808382; Beta=-0.123, SE= 0.022, P= 1.55×10−8), and a SNP at 12q24.33 near TMEM132D (lead SNP rs73157566; Beta= −0.294, SE= 0.054, P= 4.07 ×10−8); these associations have not been reported previously with any retinal vascular parameters.
Although we replicated previously reported loci for CRVE, we did not find any novel genome-wide significant loci for this trait9,11. Furthermore, we did not replicate any of the previously reported SNPs associated with CRAE10. Finally, we replicated a previously reported significant locus for ODradius at 10q21.3 near PBLD (lead SNP rs61854835; Beta= −3.840, SE= 0.575, P= 4.06 ×10−11) and confirmed a number of other loci for this trait 12–15(Supplementary Table 4).
We selected three lead SNPs near ACTN4, TMEM132D, and COL4A2, that all reached significance P ≤ 1.07×10−07 as well as their effect size and direction were similar across the studies, as candidates to carry forward for replication, and confirmed these in three independent cohorts comprised of up to 1398 individuals of European ancestry (Supplementary Table 5); the Lothian Birth Cohort 193624 (LBC1936), Croatia-Korcula, and Croatia-Split. Retinal images from these cohorts had been analyzed by SIVA 3.125,26 (Singapore I Vessels Assessment) software to quantify the tortuosity traits. In overall meta-analysis (discovery and replication stage), only SNPs at 13q34 (TortA) and 19q13 (TortV) were confirmed at genome-wide significance. Although TortA associated SNPs rs7991229 and rs9515212 were not genome-wide significant in the discovery meta-analysis, they did in the overall meta-analysis; Poverall=4.66×10−12 and Poverall=6.52×10−10, respectively. Whereas lead SNP in COL4A2 for TortA (rs56399312) at discovery stage, did not reach genome-wide significance in overall meta-analysis (Poverall=1.95×10−07). For TortV the lead SNPs maintained genome-wide significance rs1808382 (Poverall=2.39×10−13), rs3786835 (Poverall=3.31×10−13) near ACTN4/CAPN12 (Table 1). These SNPs are in tight LD and therefore do not represent independent signals. Regional plots and Forest plots for the genome-wide significant SNPs in the combined analysis are shown in Figure 2 and Supplementary Figure 5, respectively.
COL4A2 encodes collagen type IV alpha 2, one of the six subunits of type IV collagens which are major structural components of basement membranes, forming a thin sheet of fibers under the endothelium controlling passage of vasoactive substances. These are conserved across species and C-terminal non-collagenous domains play a role in angiogenesis27. Several studies suggest that the mutations in COL4A2 and (or) COL4A1 (a paralogue immediately proximal to COL4A2, with which it shares a promoter and is co-expressed) cause a broad spectrum of diseases including retinopathy28, glaucoma28, familial cerebrovascular, small vessel diseases29, and retinal hemorrhages30. In addition, COL4A2 mutations in mice are associated with small vessel disease, intracerebral hemorrhage and retinal changes31. Mendelian variants in COL4A1 underlie syndromes which include tortuous retinal vessels32. Recent GWAS report that common variants around COL4A1 and COL4A2, are associated with coronary artery calcification33, arterial stiffness34, and coronary artery disease35–38 (CAD). Interestingly, gene expression data from GeneAtlas39, a human protein-coding transcriptome study validated the high expression of COL4A2 in retinal micro-vessel endothelial cells (Supplementary Figure 6) whereas COL4A1 is weakly expressed in retina and this supports the specific role of COL4A2 in the retinal vasculature. Genome-wide significant TortA-associated variants near COL4A2 significantly alter the transcription factor binding motifs, overlap enhancer histone marks and have putative effects on transcription as annotated by the ENCODE (Supplementary Table 6). Additionally, expression data from the GTEx database40 confirmed that these significant SNPs, are associated with the expression of COL4A2 in heart left ventricle and artery aorta, shown in Supplementary Table 7, Supplementary Figure 7, and these SNPs are in linkage disequilibrium (LD; r2=0.99, D’=1). Lead SNPs associated with TortA were still significant after conditioning on the previously reported cardiovascular risk variants (rs1161795536, rs477314437, rs951520338) (Supplementary Figure 8, Supplementary Table 8). Conversely the lead SNPs for TortA are not associated with CAD and myocardial infarction (MI) risk in the CARDIoGRAMplus C4D consortium meta-analysis36 (Supplementary Table 9). Finally, the CAD associated variants in COL4A1 from CARDIOGRAMplusC4D are not associated with TortA, whereas CAD associated COL4A2 variants are only weakly associated with TortA (Supplementary Table 10). Together, these data demonstrate that variants in this gene complex may independently influence micro and macrovascular disease.
ACTN4 encodes alpha-actinin 4, a cross-linking protein belonging to the spectrin superfamily; mutations in this gene cause focal segmental glomerulosclerosis in humans41. ACTN2, a homolog of ACTN4, interacts with ACTN4 and missense mutations in ACTN2 are linked to a range of cardiac diseases42. Annotation by ENCODE43 indicates that the two genome-wide significant variants (rs1808382, rs3786835) associated with TortV near ACTN4 may have direct regulatory effects as they are located within a DNase I hypersensitivity site in multiple cell types and are located in genomic regions enriched for promoter/enhancer histone marks especially in heart tissues (Supplementary Table 6). ACTN4 overlaps with CAPN12 (calcium-activated neural proteases) by 339 bases at their 3’ ends and multi-tissue expression quantitative trait loci (eQTL) analysis confirms that these SNPs in ACTN4 are associated with mRNA expression of both ACTN4 and CAPN12 in aorta, tibial artery, atrial appendage and left ventricle of the heart (Supplementary Table 7, Supplementary Figure 9). Additionally, this analysis indicates that the T allele at rs1808382 is correlated with lower ACTN4 (artery aorta; P=2.1×10−03) and this correlation is even stronger with CAPN12 (artery aorta; P=2.0×10−07). However, while gene expression data using GENEINVESTIGATOR validated the high expression of ACTN4 in arterial tissue, the highest expression of CAPN12 appears to be in the hematopoietic system. Furthermore, lead SNPs in ACTN4 are significantly associated with coronary artery disease in the CARDIoGRAMplus C4D consortium meta-analysis36 (Supplementary Table 9) and are associated with CAD risk factors; HDL cholesterol and triglycerides in the Global Lipid Genetics Consortium analysis44 (Supplementary Table 11). Finally, recent meta-analysis of 35 GWAS studies reported the association of SNP (rs11083475) in the ACTN4 locus with increased resting heart rate45 which may increase cardiovascular disease risk. This signal is the same as that for TortV with strong LD being observed between the lead SNPs for TortV and the index SNP for heart rate. Furthermore we found that these SNPs are associated with heart rate in UK Biobank (Supplementary Table 12, Supplementary Figure 10).
In summary, this first GWAS for TortA and TortV reveals SNPs influencing expression of COL4A2 and ACTN4/CAPN12 respectively. Our results demonstrate that the TortA-associated variants in COL4A2 are not associated with CAD and MI, and point to a selective role of COL4A2 rather than COL4A1 in the retinal vessels. Strikingly, we found TortV-associated ACTN4/CAPN12 SNPs are associated with CAD and heart rate. Detailed investigation of this new finding is essential to elucidate the causal roles of ACTN4 and/or CAPN12 in the observed cardiovascular pathophysiology. These findings highlight the potential genetic impacts of retinal vasculature to provide new insights into wide-range of vascular disease.
URLs
SNPTEST V 2.5.2, https://mathgen.stats.ox.ac.uk/genetics_software/snptest/snptest.html; SHAPEIT v2, https://mathgen.stats.ox.ac.uk/genetics_software/shapeit/shapeit.html; IMPUTE v2 https://mathgen.stats.ox.ac.uk/impute/impute_v2.html; 1000 Genomes Project, http://www.1000genomes.org/; Vampire, http://vampire.computing.dundee.ac.uk/index.html; GWAMA http://www.well.ox.ac.uk/gwama/; R statistical program package, http://www.r-project.org/; BEDTools, http://bedtools.readthedocs.org/en/latest/; LocusZoom, http://csg.sph.umich.edu/locuszoom/; UCSC Genome Browser, https://genome.ucsc.edu/; HaploReg, http://www.broadinstitute.org/mammals/haploreg/haploreg.php; PLINK, https://www.cog-genomics.org/plink2/; Genotype-Tissue Expression (GTEx) project, http://www.gtexportal.org/home/; ENCODE, http://www.genome.gov/encode/ and http://genome.ucsc.edu/ENCODE/; RegulomeDB, http://www.regulomedb.org/; EIGENSTRAT, http://genetics.med.harvard.edu/reich/Reich_Lab/Software.html; Type2Diabetes Knowledge portal, http://www.type2diabetesgenetics.org/; Genevisible, https://genevisible.com/search; GoDARTS, http://diabetesgenetics.dundee.ac.uk/; ORCADES, http://www.orcades.ed.ac.uk/orcades/index.html; LBC1936, http://www.lothianbirthcohort.ed.ac.uk/; UK Biobank, http://www.ukbiobank.ac.uk/.
AUTHOR CONTRIBUTIONS
The study was designed by C.NA.P, A.SF.D, and E.T for GoDARTS cohort, J.F.W for ORCADES cohort, I.J.D for LBC1936 cohort, C.H, O.P for Croatia-Split, and Croatia- Korčula cohort. VAMPIRE software was designed and developed by E.T, T.M, D.R, L.B, S.K.V, E.B and B.D. Retinal images were collected and analysis was performed by E.T, T.M, J.F.W, L.B, M.K, D.R, and V.V. Genotype data processing and statistical analysis was conducted by A.V, K.E.S, P.K.J, H.C, L.B, M.K, S.H, V.V and K.Z. Bioinformatics analysis was performed by A.V. The manuscript was drafted by A.V, C.NA.P A.SF.D, and revised by E.T, J.F.W, T.M, I.J.D, S.H, L.B, O.P, E.R.P and K.Z. All the authors reviewed the manuscript.
Competing interest
None
ONLINE METHODS
Study participants
Discovery Cohorts
Participants in the discovery phase of this study were obtained from the two independent cohorts, the Genetics of Diabetes Audit and Research in Tayside (GoDARTS) and the Orkney Complex Disease Study (ORCADES). GoDARTS comprises individuals of European-heritage from Tayside, Scotland who provided a sample of blood for genetic analysis and consent to link their genetic information to the anonymized electronic health records. Approval for recruitment to GoDARTS was obtained from the Tayside Committee on Medical Research Ethics. 18,190 individuals were recruited with approximately half having type 2 diabetes at the time of recruitment with the other half being diabetes free. 7,290 individuals currently have genome-wide data for analysis. ORCADES is a family-based study of 2078 individuals aged 16-100 years recruited between 2005 and 2011 in the isolated Scottish archipelago of Orkney46. Genetic diversity in this population is decreased compared to Mainland Scotland, consistent with the high levels of endogamy historically. Fasting blood samples were collected and over 300 health-related phenotypes and environmental exposures were measured in each individual. All participants provided written informed consent and the study was approved by Research Ethics Committees in Orkney and Aberdeen.
Replication Cohorts
Lothian Birth Cohort 1936 (LBC1936) comprised of 1091 participants who were born in 1936, most of whom took part in the Scottish Mental Survey of 1947. At a mean age of 69.5 years (SD 0.8), between 2004 and 2007, they were recruited to a study to determine influences on cognitive aging24. The CROATIA- Korčula study sampled from the Adriatic island of Korčula, between the ages of 18 and 88. The fieldwork was performed in 2007 in the eastern part of the island, targeting healthy volunteers who underwent complete eye examination and provided their blood sample for genetic analysis from the town of Korčula and the villages of Lumbarda, Žrnovo and Račišće. The Croatia-Split study included inhabitants of the Croatian coastal city of Split, aged 18 to 93. The sampling scheme was similar to Croatia- Korčula, and it took place during 2008 and 2009.
Retinal Vascular Parameters Measurement
Retinal image analysis
Standard digital retinal photographs used for routine diabetic retinopathy screening were obtained from the clinical record in 2,104 participants in GoDARTS. Images of the right eye of usable quality, defined using criteria reported in21,22, were selected and categorized into two datasets based on the image pixel resolution: GoDARTS dataset 1 (n=788) and GoDARTS dataset 2 (n=1288). Finally, 661 images from the GoDARTS dataset 1, and 1083 images from GoDARTS dataset 2 were included after quality control (QC). 28 individual’s images from the GoDARTS were excluded due to inadequate resolution. Standard fundal retinal photographs centred between the macula and optic disc were obtained using digital fundus camera from 1,743 participants in ORCADES. After image processing and QC, 1595 individual’s retinal images were used for this study.
VAMPIRE 3.0, was used to measure retinal vascular traits in fundus images (see Figure 1) from both GoDARTS and ORCADES. The measurement process is organized as a sequence of automatic and manual stages. Manual stages allowed correction of errors made by the automatic software (e.g. vessel labeling as artery or vein) and to minimize their impact on statistical analysis. The standard protocols were followed to measure the retinal vessel parameters. Briefly, after automatic detection of the optic disc and its radius (ODradius), the 6 thickest arterioles and 6 thickest venules appearing in a zone extending from the optic disc boundary to 2 optic disc diameters out were sampled to calculate the median (TortA) and maximum (TortAmax) arteriolar tortuosity and the median (TortV) and maximum (TortVmax) venular tortuosity. CRAE, CRVE and the Arteriole-to-Venule ratio (AVR; CRAE/CRVE) qualify vessel calibers and were measured in a zone 2 to 3 optic disc radii from the center of the optic disc. Among the eight parameters, TortA, TortAmax, TortV and TortVmax mean values were natural log transformed for association analysis. Supplementary Table 1 shows descriptive statistics for retinal parameters and population characteristics for both discovery cohorts.
Replication Cohorts
Standard retinal fundus images using digital fundus camera from 1091 individuals from LBC1936 were collected at the recruitment stage and three years later, retinal traits were measured at a subsequent wave of testing using SIVA v3.1, at a mean age of 72.5 years (SD 0.7). A total of 897 and 976 individual’s retinal fundus images centred between the macula and optic disc from Croatia- Korčula and Croatia-Split cohorts were collected using digital fundus camera and retinal traits were quantified using SIVA v3.125,26. SIVA is a semi-automated software which can be used to measure the retinal vascular parameters including retinal vascular tortuosity and vascular caliber from retinal images. After automatic detection of the optic disc, it placed a grid with reference to the center of the optic disc. Then the tortuous vessels were identified and tortuosity traits including TortA, and TortV were measured using the standard grading protocol by the software; this process was monitored by trained graders and adjusted manually if necessary. Supplementary Table 2 shows descriptive statistics of variables.
Genotyping, quality control and imputation
GoDARTS samples were genotyped using the Affymetrix 6.0 (n=927) and Illumina Human Omni Express (n=809) platforms. The poor quality variants, samples were excluded based on the quality control (QC) criteria included the following: SNPs call rate < 95%, Hardy–Weinberg equilibrium (HWE) P value < 10−6, sample call rate < 95%, sample relatedness (IBD >0.8), and mismatch between reported and genotypic gender information. QC’d genotype data was imputed using IMPUTE247,48 on the basis of 1000 Genome Projects reference panel for all population. Finally, ancestry information of the individuals were derived using EIGENSTRAT49 and first three principal components (PCs) were used for the association analyses to adjust the population stratification. ORCADES samples were genotyped with either the Illumina HumanHap300 bead chip (n=890) or the Illumina Omni1 (n=304) or Illumina Omni Express bead chips (n=1073). Alleles were called in Bead Studio/Genome Studio (Hap300/Omni) using Illumina cluster files. Subjects were excluded if they fulfilled any of the following criteria: genotypic call rate <98%, mismatch between reported and genotypic sex, unexpectedly low genomic sharing with first or second degree relatives, excess autosomal heterozygosity, or outliers identified by IBS clustering analysis. We excluded SNPs on the basis of minor allele frequency (<0.01/monomorphism), HWE (P<10^-6), call rate (<97%). Given the very high overlap in SNPs between the two Omni chips, the intersection of QC’d SNPs was used to impute and phase individuals’ genotyped on the Omni arrays together, whilst the Hap300 individuals were phased and imputed, separately. Samples were phased using Shapeit v250. Imputation was carried out using IMPUTE2 and the 1,000 genomes reference panel. All ancestries phase1 integrated v3 reference panel, with a secondary reference panel of local exome sequences, sequenced using the Agilent Sure Select All Exon Kit v2.0 and Illumina 100 bp paired end reads (average 30x depth), derived from 90 ORCADES subjects chosen to optimally represent the haplotypes present. Imputations for the Hap300 and Omni subjects were then combined to form a combined panel of 37.5m SNPs for 2222 subjects51. Imputed genotypes for 658, 1078, 1358 individuals from the GoDARTS dataset 1, GoDARTS dataset 2 and ORCADES cohorts, respectively, were used for the three independent GWAS analysis. Additional information on the characteristics of both study cohorts and study design can be found in Supplementary Figure 1 and Table 1.
Replication Cohorts
LBC1936 samples were genotyped at the Wellcome Trust Clinical Research Facility, Edinburgh, using the Illumina Human 610Quad BeadChip. Individuals were excluded based on unresolved gender discrepancy, relatedness, call rate (≤0.95), and evidence of non-Caucasian descent. SNPs were included if they met the following conditions: call rate ≥ 0.98, minor allele frequency ≥ 0.01, and Hardy-Weinberg equilibrium test with P ≥ 0.001. Imputation to the 1000 Genomes (March 2012 release) reference set was performed using minimac software. A total of 1398 participants from the two independent Croatian replication cohorts were available for the analysis and subjects were genotyped on different genotyping platforms including Illumina CNV370v1 and CNV370-Quadv3 for Croatia-Korčula (n=378), and Illumina CNV370-Quadv3 and IlluminaOmniExpressExome-8v1_A for Croatia-Split (n=376). Samples and markers were excluded based on the following QC metrics; SNPs call rate < 98%, Hardy–Weinberg equilibrium (HWE) P value < 10−6, sample call rate < 97%, MAF < 1%, outliers identified by IBS clustering analysis and unresolved gender discrepancy. Imputation was carried out using IMPUTE2 software and 1000G Phase I v3 (March 14, 2012) reference panel.
Statistical analyses
We performed association analyses with each data sets from GoDARTS separately for each of the eight retinal traits using SNPTEST V2.547, linear regression assuming an additive genetic model, adjusting for 3 ancestry PCs, age at eye examination and gender. Subsequently, markers with low imputation quality scores (< 0.4) and minor allele frequency cutoffs (< 0.03) were filtered from each GWAS summary output data separately. Then we performed the meta-analysis using a fixed-effects model in GWAMA52 with the QC filtered data sets. Association analysis in ORCADES was performed for each of the eight retinal traits, using linear mixed modelling to account for relatedness and assuming an additive genetic model, adjusting for 3 ancestry PCs, age at eye examination and gender, using MMscore in ProbABEL53. As in GoDARTS, markers with low imputation quality scores (< 0.4) and minor allele frequency cutoffs (< 0.03) were filtered and meta-analysis was performed with the GoDARTS and ORCADES results using GWAMA. The strand alignment and build check between studies were performed prior to meta-analysis. Also, the genomic inflation factor (λ) was estimated by GWAMA (λ=0.99). All statistical analyses and QCs were performed using SNPTEST v2.547, ProbABEL53, GWAMA52, PLINK v1.0954, EIGENSTRAT49, custom shell scripts, and R scripts. Manhattan plots, Quantile-Quantile plots and forest plots were generated using in-build R scripts, and metafor - R package55. Regional plots were generated using the Locus Zoom tool56 and other data processing was performed using R scripts. Conditional analyses were performed in SNPTEST v2.5 using the genome-wide associated loci in the COL4A2 region, conditioned on lead SNPs (rs56399312). Also, this new locus was conditioned on previously reported genome-wide significant SNPs (rs4773144, rs11617955, rs9515203) associated with coronary artery disease (CAD).
Replication-analyses
Top three SNPs (P ≤ 1.07×10−07) near ACTN4, TMEM132D, and COL4A2 from the discovery stage for the tortuosity traits were taken forward for examination in three replication cohorts of European ancestry. In LBC1936 cohort, association analysis was performed for arterial and venular tortuosity traits using linear regression model adjusting for age at eye examination, sex, and 3 ancestry PCs, using mach2qtl. Similarly, in the Croatia – Split, - Korčula cohorts, association analysis were performed for each traits separately using the mixed model in R - hglm package to account for kinship derived using gkin function of the GenABEL package57.
Then we combined the summary association statistics for lead SNPs associated with TortA and TortV from the two discovery and three replication cohorts and effect estimates from each study cohorts were presented in the forest plots using metafor - R package. Due to the difference in the units of the beta and standard errors between the discovery and replication studies arising from different approaches to measurement we standardized the effect estimates (using Cohen’s d) from each individual’s study results.
In-silico look-ups of the novel variants for clinical outcomes
We performed in-silico look-ups of variants of interest for cardiovascular related outcomes including coronary artery disease, myocardial infarction, hypertension, HDL, and triglycerides from the CARDIoGRAMplus C4D consortium36 and Global Lipid Genetics Consortium analysis44. The CARDIoGRAMplusC4D 1000 Genomes-based meta-analysis data comprised of 60,801 CAD cases and 123,504 controls from European, South Asian, and East Asian descent. In the Global Lipid Genetics Consortium, genetic data from 188,577 individuals of European, East Asian, South Asian, and African ancestry were used to examine the genetic loci associated with blood lipids levels. We retrieved summary association results for the index SNPs from these studies to investigate the association of the lead SNPs for TortA, and TortV with cardiovascular outcomes.
A recent study reported the association of ACTN4 locus with heart rate. In order to examine whether the lead SNPs associated with TortV in ACTN4 were also associated with heart rate, we checked the LD (r2>0.8) between our SNPs and the index SNP (rs11083475) for heart rate. Furthermore, we selected these SNPs and investigated the association of these SNPs with pulse rate in UK Biobank data. This data comprised of 112,008 participants who had a measure of pulse rate at the main interview and had genotype data. We extracted the imputed genotypes for the these SNPs from the interim release data set of the UK Biobank58 and performed multiple linear regressions including covariates of age, gender, and the first ten principal components obtained using EIGENSTRAT.
In-silico functional annotation
The sentinel genome-wide significant variants were mapped to the gene, 20 kb upstream/downstream using BEDTools59, and UCSC Genome Browser60. Top SNPs were queried in the HaploReg v4.1 database61 to catalogue the all SNPs near noncoding variants with r2 > 0.8, and RegulomeDB62, and GWAS catalog databases63 used to explore the known and predicted regulatory elements and relevant genetic association studies. Functional effects of the top genes were predicted using the Encyclopedia of DNA Elements43 (ENCODE) project and Roadmap Epigenomics projects which aggregate the information about the transcription factor, motifs, histone modification, and chromatin states. Additionally, functional elements were investigated using HaploReg, UCSC Genome Browser, and RegulomeDB. We used the expression Quantitative Trait Loci (eQTL) browser database in Genotype-Tissue Expression40 (GTEx) to examine the cis-eQTLs for the top retinal traits associated SNPs mapped to the gene within the genomic region. Gene Visible web database from GENEINVESTIGATOR which integrates manually curated gene expression data from microarray and RNAseq experiments, was used to find the expression level of the genes, associated with tortuosity traits, in the human tissues.
ACKNOWLEDGEMENTS
We are grateful to all the participants in the GoDARTS study, the general practitioners, the Scottish School of Primary Care for their help in recruiting the participants, and to the whole team, which includes interviewers, computer and laboratory technicians, clerical workers, research scientists, volunteers, managers, receptionists, and nurses. The study complies with the Declaration of Helsinki. We acknowledge the support of the Health Informatics Centre, University of Dundee for managing and supplying the anonymized data and NHS Tayside, the original data owner. The Wellcome Trust United Kingdom Type 2 Diabetes Case Control Collection (GoDARTS) was funded by The Wellcome Trust (072960/Z/03/Z, 084726/Z/08/Z, 084727/Z/08/Z, 085475/Z/08/Z, 085475/B/08/Z) and as part of the EU IMI-SUMMIT program.
ORCADES was supported by the Chief Scientist Office of the Scottish Government (CZB/4/276, CZB/4/710), the Royal Society, the MRC Human Genetics Unit, Arthritis Research UK and the European Union framework program 6 EUROSPAN project (contract no. LSHG-CT-2006-018947). DNA extractions were performed at the Wellcome Trust Clinical Research Facility in Edinburgh. We would like to acknowledge the invaluable contributions of the research nurses in Orkney, the administrative team in Edinburgh and the people of Orkney.
We thank the Lothian Birth Cohort 1936 (LBC1936) participants and team members who contributed to these studies. Phenotype collection was supported by Age UK (The Disconnected Mind project). Genotyping was funded by the BBSRC (BB/F019394/1). The work was undertaken by The University of Edinburgh Centre for Cognitive Ageing and Cognitive Epidemiology, part of the cross council Lifelong Health and Wellbeing Initiative (MR/K026992/1). Funding from the BBSRC and Medical Research Council (MRC) is gratefully acknowledged.
The Croatia- Korčula and Croatia-Split study were funded by grants from the Medical Research Council (UK), European Commission Framework 6 project EUROSPAN (Contract No. LSHG-CT- 2006-018947), European Commission Framework 7 project BBMRI-LPC (FP7 313010), the Republic of Croatia Ministry of Science, Education and Sports research grant (216-1080315-0302) and the Croatian Science Foundation (grant 8875). We would like to acknowledge the staff of several institutions in Croatia that supported the field work, including but not limited to the University of Split and Zagreb Medical Schools and Croatian Institute for Public Health. The SNP genotyping for the Korčula cohort was performed in Helmholtz Zentrum München, Neuherberg, Germany.
VAMPIRE team: Parts of the VAMPIRE software and its use for measuring the image set described here was funded by the Leverhulme Trust project RPG-419 “Discovery of retinal biomarkers for genetics with large cross-linked data sets”. VAMPIRE 3.1 has been developed under funding from EPSRC (EPSRC EP/M005976/1), the EU (REVAMMAD ITN).
For the analysis of the association of the identified genetic variants with heart rate, this research has been conducted using the UK Biobank Resource under Application Number 20405.