Heart disease in women and men
van der Ende, Maaike Yldau
DOI:
10.33612/diss.103508645
IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite from it. Please check the document version below.
Document Version
Publisher's PDF, also known as Version of record
Publication date: 2019
Link to publication in University of Groningen/UMCG research database
Citation for published version (APA):
van der Ende, M. Y. (2019). Heart disease in women and men: insights from Big Data. Rijksuniversiteit Groningen. https://doi.org/10.33612/diss.103508645
Copyright
Other than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons).
Take-down policy
If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.
Downloaded from the University of Groningen/UMCG research database (Pure): http://www.rug.nl/research/portal. For technical reasons the number of authors shown on this cover page is limited to 10 maximum.
CHAPTER 5
Resting heart rate and coronary artery disease: a
genome-wide meta-analysis and Mendelian randomization analysis
• • •
Yordi. J. Van de Vegte#, Ruben N. Eppinga#, M. Yldau van der Ende, Yanick Hagemeijer, IC-RHR investigators*, Harold Snieder, Niek Verweij, Pim van der Harst. #Yordi. J. Van de Vegte and Ruben N. Eppinga contributed equally to the current study. *Investigators of the International Cohorts for Resting Heart Rate. Complete names and author affiliations will be reported in the final manuscript.
ABSTRACT
Resting heart rate (RHR) is linked to cardiovascular diseases in observational studies. However, RHR is influenced by a plethora of mechanisms, which could all confound the association with disease. In this study, we first performed the largest genome-wide association meta-analysis on RHR so far, including 100 studies with a total of 835,465 individuals of European ancestry. In total, 493 genetic variants associated with RHR were identified, of which 276 were novel, and 1,069 candidate causal genes were prioritized. Second, identified genetic loci associated with RHR were used in Mendelian randomization analyses to test whether genetically predicted higher RHR is a causal risk factor for coronary artery disease (CAD) and myocardial infarction (MI). In the independent CARDIoGRAMplusC4D cohort, one standard deviation increase in genetically determined RHR did not increase CAD (odds ratio (OR) 0.994, 95% confidence interval (CI) 0.987-1.001, P=0.091) and MI (OR 0.994, 95% CI 0.986-1.001, P=0.110) risk. Repeating the Mendelian randomization analyses in the UK Biobank, again no associations between RHR and CAD (OR 0.994, 95% CI 0.987-1.000, P=0.064) and MI (OR 0.996, CI 0.987 – 1.006, P=0.431) were observed. The well-known link between higher RHR and CAD and MI may therefore likely be explained by acquiring a higher RHR during live.
INTRODUCTION
Higher resting heart rate (RHR) is associated with both cardiovascular diseases, cancer and total mortality1,2. However, RHR is influenced by a plethora of mechanisms, which
could all confound the association with disease and mortality. Genetic analyses have the potential to overcome this source of bias. In these analyses, genetic variants associated with the exposure (RHR) function as a proxy for this variable. Since genetic variants are randomly assigned when transmitted from parents to offspring, they are therefore mostly unrelated to the presence of confounders.
So far, a total of 64 replicated loci associated with RHR have been discovered and were significantly associated with all-cause mortality, but not with coronary artery disease (CAD) or myocardial infarction (MI). These 64 loci explained only 2.5% of the total estimated variance in RHR, whereas heritability estimates range between the 23% and 70%3–5. More recently, another genome wide association study (GWAS) was performed on
RHR in a larger sample size, but the genetic overlap with chronic obstructive pulmonary disease, and not the biology of RHR itself, was assessed6. Increasing the total estimated
variance by increasing the number of RHR loci could give more accurate insights in the true nature of the association between RHR and CAD and MI. Additionally, associated genetic variants will further broaden our knowledge of the mechanisms underlying RHR and can help to offer entry point into new therapies for personalized medicine. To increase our knowledge on the influence of the human genetic make-up on RHR and whether genetically predicted RHR is causally linked to CAD and MI, we performed a meta-analysis of 100 GWAS in 835,464 participants (Figure 1A). Using the identified genetic variants as instrumental variables, we explored the relationship of RHR with CAD and MI using a Mendelian randomization (MR) strategy (Figure 1B).
UK Biobank data
N = 484,307 with genetic and phenotypic data. 19,400,416 variants after QC filtering on INFO >0.3
and Neff≥25
IC-RHR data
N = 351,158 from 99 different cohorts. 27,082,649 variants after QC filtering on INFO >0.3
and Neff≥25
UK Biobank GWAS analysis
UKB GWAS of HRC imputed SNPs HR ~ SNP + sex + age + age2+ BMI + heart rate
measurement method + 30PCs + GenChip
IC-RHR GWAS analysis
GWAS of imputed SNPs (1000G) Fixed-effect inverse-variance-weighted
meta-analysis; Stringent meta-level QC filtering
Internal replication
1) P < 1 × 10-8 in the discovery (UKB+IC-RHR) meta-analysis 2) support (P < 0.01) in the UKB GWAS alone 3) support (P < 0.01) in the IC-RHR GWAS alone 4) concordant direction of effect between UKB and IC-RHR datasets
D at a & Q C D is co ve ry R ep lic at io n V al id at io n
493 SNPs in 325 loci: 276 novel (outside 1MB of 74 previously discoverd loci)
Locus definition: R2>0.005; 2 MB-region
UK Biobank + IC-RHR (N = 835,465)
332 replicated loci, 257 novel and
replicated 10 newly replicated loci (previouslypublished without replication)
493 SNPs associated with RHR in the
META-analyses
492 SNPs and 1 proxy with beta’s and standard errors from the UK Biobank
484 SNPs and 3 proxies with beta’s and standard errors
from the IC-RHR.
MR in CARDIOGRAMC4D cohort. CAD, Ncases = 60,801, Ncontrols = 123,504
MI, Ncases = 43,676, Ncontrols = 128,188
MR in UK Biobank cohort CAD, Ncases = 27,780, Ncontrols = 382,914
MI, Ncases = 12,025, Ncontrols = 398,669
Exposure Outcome
73 SNPs associated with RHR in the
previous GWAS of Eppinga et al. 2016
73 S NP s with beta’s and standard errors from the UK
Biobank
73 S NP s with beta’s and standard errors from the UK
Biobank
MR in CARDIOGRAMC4D cohort. CAD, Ncases = 60,801, Ncontrols = 123,504
MI, Ncases = 43,676, Ncontrols = 128,188
MR in UK Biobank CAD, Ncases = 19,307, Ncontrols = 257,136
MI, Ncases = 8,373, Ncontrols = 268,070
B)
Figure 1. A. Schematic overview of the study design for the discovery and replication of genetic
loci associated with resting heart rate (RHR). BMI = body mass index, GWAS = genome-wide association study, HRC = Haplotype Reference Panel, IC-RHR = International cohorts for resting heart rate, MB = megabytes, N = number, Neff = Effective sample size, PC = principal components, RHR = resting heart rate, SNPs = single nucleotide polymorphisms, QC = quality control, 1000G = 1000 Genomes. B. Schematic presentation of the MR analyses of RHR with CAD and MI. Beta’s and standard errors were taken from the UK Biobank to test the association with CAD and MI in the CARDIoGRAMplusC4D cohort, whereas beta’s and standard errors were taken from the IC-RHR data to test the associations with CAD and MI the UK Biobank. The same was performed using the previously discovered variants. Replication within the UK Biobank was performed on a subset which excluded the participants used for the discovery. CAD = coronary artery disease, IC-RHR = International Cohorts for Resting Heart Rate, MI = myocardial infarction, RHR = resting heart rate, SNPs = single nucleotide polymorphisms.
RESULTS
We performed a meta-analysis of 835,464 individuals (Online Table 1) with a total of 30,458,884 directly genotyped and imputed autosomal genetic variants (Figure 1A). The meta-analyses revealed 493 variants in 352 loci, of which 276 were novel (Figure 2A, Online Table 2), more than a fourfold increase of the number of RHR associated loci known until now. Of the 352 loci, 332 were internally replicated (Figure 2B). This includes replication of 10 out of the 12 previously unreplicated loci. The linkage disequilibrium (LD) score regression (LDSR) intercept (standard error) after the final meta-analysis was 1.051 ± 0.002, suggesting little evidence of genomic inflation (Figure 2C). The estimated single nucleotide polymorphism (SNP) based heritability of RHR as calculated by LDSR in our data was 10%. The International Cohorts for Resting Heart Rate (IC-RHR) beta estimates of the discovered variants explained 5.33% of the variation in RHR within the UK Biobank.
Candidate Genes and Insights into Biology
We explored the potential biology of the 352 identified loci by prioritizing candidate causal genes in these loci (Online Table 2): 407 unique genes were in proximity (the nearest gene and any additional gene within 10kb, Online Table 2) of the lead variant, genes containing 52 unique coding variants in LD (R2>0.80) with RHR lead variants (Online
Table 3), 88 unique genes were selected on the basis of multiple functional expression quantitative trait loci (eQTL) analyses (Online Table 4), and 928 unique genes were taken forward by Data-driven Expression-Prioritized Integration for Complex Traits (DEPICT)
A
analyses (Online Table 5). Of the 1,069 unique candidate causal genes identified, 346 genes were prioritized by multiple methods, which may be used to prioritize candidate causal genes (Figure 2D). Of these genes, PHACTR4, ENO3, B3GNT7, KIAA1755 and SENP2 were prioritized by all four methods.
Pathway analyses and tissue enrichment
Pathway analysis was performed for RHR using DEPICT (Online Table 6). All pathways revolved around common themes including neurological development, neuronal longevity and signaling pathways. Importantly, 1,471 reconstituted gene sets were found to be significantly associated with RHR. The tissue enrichment analyses by DEPICT implicated the cardiovascular system as the most important tissue type, with 8 of the 10 most significantly enriched tissues located within the cardiovascular system (Online Table 7).
Mendelian randomisation analyses
The associations between RHR and CAD and MI were tested within both the UK Biobank and the CARDIoGRAMplusC4D cohort using the 493 SNPs identified in the current study (Figure 1B). The genetic variants used as instrument for the MR analyses are reported in Online Table 8. The estimates of the instruments used for the analyses (P<1 × 10-8) within
the CARDIoGRAMplusC4D study had a combined F-statistic of 63.2, which indicates a low chance of weak instrument bias. F-statistics could only be tested within the UK Biobank using the UK Biobank estimates, since only summary-based statistics of the IC-RHR were available. A series of MR analyses was performed to test the hypothesis that increased RHR is a causal risk factor for CAD and MI. Results of these analyses are displayed in
Figure 2. A. Manhattan plot showing the minus log P-value for the association of identified loci
associated with resting heart rate (RHR). Grey indicates previously identified genetic variants within loci reaching genome-wide significance; red indicates novel genetic variants within loci reaching genome-wide significance. Green indicates previously unreplicated loci that were replicated in the current study. B. Venn diagram of the 352 identified loci. Of the 352 loci, 332 were internally replicated. C. Quantile-quantile (QQ) plot of the meta-analysis of the UK Biobank and IC-RHR data. The black dots represent the observed statistic for the genotyped SNPs against the corresponding expected statistic. The LDSR intercept after the final meta-analysis was 1.051, suggesting little evidence of genomic inflation due to non-polygenic signal. D. Venn diagram of the prioritization of the 1,069 unique candidate causal genes as identified by one or multiple strategies. Venn plot shows overlap of genes tagged by one or multiple strategies. DEPICT = Data-driven Expression Prioritized Integration for Complex Traits, eQTL = expression quantitative trait loci.
Figure 3 and Online Table 9, and are discussed below. All excluded variants in the PRESSO analyses are shown in Online Table 10. We investigated heterogeneity in MR-inverse variance weighted (IVW) analyses using the I2 index and Cochran’s Q and in
MR-Egger analyses using Rucker’s Q in order to gain insights in potential pleiotropy (Table 1). Heterogeneity and thus potential pleiotropy was found using I2 index and Cochran’s Q
within all the MR-IVW analyses, indicating that the assumption of absence of pleiotropy was not fulfilled, which is necessary for the MR-IVW fixed effects model. Therefore, the MR-IVW random effects model was considered to be a better approach. Rucker’s Q was significant in all analyses, again indicating potential pleiotropy. However, Rucker’s Q was not significantly lower than Cochran’s Q (Table 1), providing evidence for absence of directional pleiotropy. This indicates that the MR-IVW random effects approach has the potential to give an unbiased estimation of effect. The variation between individual genetic variant estimates for each exposure (I2
GX) of all MR analyses was >95%, which
indicated a low chance of weak instrument bias within the MR-Egger analyses. Using the IVW-MR random effects approach with the new variants, showed that one standard deviation (SD) increase in genetically determined RHR did not significantly increase CAD (Odds ratio (OR) 0.994, 95% confidence interval (CI) 0.987-1.001, P=0.091) and MI (OR 0.994, 95% CI 0.986-1.001, P=0.110) risk within the CARDIoGRAMplusC4D cohort (Figure 3). Performing the analyses in the UK Biobank, again no associations between RHR and CAD (OR 0.994, 95% CI 0.987-1.000, P=0.064) and MI (OR 0.996, CI 0.987 – 1.006, P=0.431) were observed. For comparison, we tested the 76 SNPs from the previous RHR GWAS within the UK Biobank and the CARDIoGRAMplusC4D cohort as well. Again, the Rucker framework indicated the MR-IVW random effects model to be the preferred method for the main analyses. The association between a one SD increase in genetically determined RHR with CAD (OR 0.995, 95% CI 0.984-1.005, P=0.321) and MI (OR 0.999, 95% CI 0.987-1.011, P=0.911) risk within the CARDIoGRAMplusC4D cohort was similar to the main analyses, but with broader CIs (Figure 3). Performing the analyses in the UK Biobank, again no links between RHR and CAD (OR 0.996, 95% CI 0.985-1.006, P=0.414) and MI (OR 0.992, 95% CI 0.977-1.006, P=0.260) were observed.
Method
MR−IVW (fixed effects) MR−IVW (random effects) MR Egger MR−PRESSO MR−PRESSO (outlier−corrected) Weighted median Weighted mode MR−IVW (fixed effects) MR−IVW (random effects) MR Egger MR−PRESSO MR−PRESSO (outlier−corrected) Weighted median Weighted mode MR−IVW (fixed effects) MR−IVW (random effects) MR Egger MR−PRESSO MR−PRESSO (outlier−corrected) Weighted median Weighted mode MR−IVW (fixed effects) MR−IVW (random effects) MR Egger MR−PRESSO MR−PRESSO (outlier−corrected) Weighted median Weighted mode 472 472 472 472 465 472 472 472 472 472 472 467 472 472 75 75 75 75 74 75 75 75 75 75 75 74 75 75 0.994 (0.989 − 0.998) 0.994 (0.987 − 1.000) 0.993 (0.980 − 1.007) 0.994 (0.987 − 1.000) 0.996 (0.990 − 1.002) 0.995 (0.987 − 1.004) 1.001 (0.988 − 1.014) 0.996 (0.989 − 1.003) 0.996 (0.987 − 1.006) 1.005 (0.987 − 1.023) 0.996 (0.987 − 1.006) 0.996 (0.987 − 1.004) 0.995 (0.983 − 1.008) 1.006 (0.986 − 1.026) 0.996 (0.989 − 1.003) 0.996 (0.985 − 1.006) 0.997 (0.969 − 1.025) 0.996 (0.985 − 1.006) 0.998 (0.989 − 1.007) 0.999 (0.988 − 1.010) 1.002 (0.985 − 1.020) 0.992 (0.981 − 1.003) 0.992 (0.977 − 1.006) 1.013 (0.975 − 1.053) 0.992 (0.977 − 1.006) 0.994 (0.981 − 1.007) 1.007 (0.990 − 1.025) 1.014 (0.990 − 1.039) 0.006 0.064 0.328 0.064 0.186 0.290 0.845 0.301 0.431 0.573 0.431 0.340 0.464 0.554 0.240 0.414 0.812 0.417 0.628 0.849 0.795 0.132 0.260 0.501 0.264 0.387 0.424 0.248 0.9 1.0 1.1 456 456 456 456 448 456 456 454 454 454 454 445 454 454 72 72 72 72 69 72 72 69 69 69 69 65 69 69 0.994 (0.989 − 0.999) 0.994 (0.987 − 1.001) 0.990 (0.975 − 1.005) 0.994 (0.987 − 1.001) 0.995 (0.988 − 1.001) 0.996 (0.987 − 1.005) 0.995 (0.984 − 1.006) 0.994 (0.989 − 0.999) 0.994 (0.986 − 1.001) 0.990 (0.975 − 1.006) 0.994 (0.986 − 1.001) 0.995 (0.988 − 1.002) 1.000 (0.990 − 1.010) 1.000 (0.987 − 1.014) 0.995 (0.988 − 1.001) 0.995 (0.984 − 1.005) 0.999 (0.969 − 1.029) 0.995 (0.984 − 1.005) 0.996 (0.988 − 1.005) 0.999 (0.988 − 1.010) 1.002 (0.987 − 1.018) 0.999 (0.992 − 1.007) 0.999 (0.987 − 1.011) 1.008 (0.975 − 1.041) 0.999 (0.987 − 1.011) 1.003 (0.993 − 1.012) 1.002 (0.989 − 1.014) 0.999 (0.982 − 1.017) 0.012 0.091 0.182 0.092 0.103 0.394 0.376 0.023 0.110 0.220 0.111 0.142 0.979 0.979 0.119 0.321 0.946 0.325 0.420 0.857 0.764 0.863 0.911 0.657 0.912 0.579 0.797 0.951 0.9 1.0 1.1
SNPs from current meta-analyses, CAD
SNPs from Eppinga et al (2016), CAD
SNPs from Eppinga et al (2016), MI
OR (95% CI) P value
Nsnps Nsnps OR (95% CI) P value
Outcomes in the UK Biobank Outcomes in CARDIoGRAMplusC4D
SNPs from current meta-analyses, MI
Figure 3. Results of the Mendelian randomization analyses of resting heart rate (RHR) on coronary
artery disease (CAD) and myocardial infarction (MI). Results are given for the genetic variants associated with RHR in the current meta-analyses, as well as for the genetic variants previously associated with RHR (Eppinga et al. (2016)). On the left side, analyses on CAD and MI in the UK Biobank are displayed. On the right side, analyses on CAD and MI in CARDIoGRAMplusC4D are displayed. CAD = coronary artery disease, CI = confidence interval, IVW = inverse variance weighted, MI = myocardial infarction, MR = Mendelian randomization, MR-PRESSO = MR pleiotropy residual sum and outlier, OR = odds ratio, SNPs = single nucleotide polymorphisms.
Table 1: Het er ogeneit y (I 2, C ochr an ’s Q , Rucker ’s Q and Q -Q ’), pleiotr op y (MR-E gger in ter cept) and w eak instrumen t sta tistics in the MR-E gger analy ses (I 2 GX ) in the e xamined associa tions bet w een RHR, and C AD and MI. A sso cia tion I 2 Co chr an ’s Q Ruck er ’s Q Q-Q ’ MR E gger in ter cept I 2 GX inde x 95% CI min 95% CI max Q df P v alue Q df P v alue Q df P v alue In ter cept se P v alue A1 CAD 0.547 0.497 0.592 1000 470 6.2×10 -45 1000 469 4.2×10 -45 9.3×10 -4 1 0.980 2.3×10 -5 0.001 0.984 0.965 A1 MI 0.419 0.351 0.480 810 470 1.9×10 -20 810 469 2.3×10 -20 2.20 1 0.130 -0.002 0.002 0.255 0.965 A2 CAD 0.545 0.494 0.591 1000 460 6.8×10 -43 1000 459 5.7×10 -43 0.78 1 0.380 0.001 0.002 0.551 0.984 A2 MI 0.506 0.449 0.556 920 450 1.3×10 -33 920 449 1.1×10 -33 0.55 1 0.460 0.001 0.002 0.602 0.984 B1 CAD 0.594 0.475 0.687 180 71 9.2×10 -11 170 70 6.2×10 -11 0.24 1 0.620 -0.002 0.006 0.758 0.983 B1 MI 0.585 0.458 0.682 160 68 7.2×10 -10 160 67 5.5×10 -10 0.67 1 0.410 -0.003 0.007 0.603 0.983 B2 CAD 0.518 0.371 0.630 150 74 1.7×10 -07 150 73 1.2×10 -07 0.01 1 0.920 -3.8×10 -4 0.005 0.945 0.982 B2 MI 0.440 0.262 0.574 130 74 3.9×10 -05 130 73 5.1×10 -05 2.50 1 0.110 -0.009 0.008 0.240 0.982 On the lef t side , the t est ed associa tion is r epor ted . T he lett ers stand f or the e xposur e, in which “A ” denot es the genetic v ar ian ts f or RHR disc ov er ed in the cur ren t study and “B ” denot genetic v ar ian ts pr eviously associa
ted with RHR (Eppinga et al
. (2016)).
The number stands f
or the out come , in which “1” stands f or the analy
ses in the UK Biobank and
“2” stands f or the analy ses on the C ARDIoGR A M plusC4D c ohor t. C AD = c or onar y ar ter y disease , CI = c onfidenc e in ter val , D f = D eg rees of fr eedom, MI = m yocar dial infar ction. MR= M endelian r andomiza tion, se = standar d er ror
DISCUSSION
In this GWAS meta-analysis, performed in 835,465 individuals, 493 genetic loci associated with RHR are identified, of which 276 are novel. These loci account for 5.33% of the variation in RHR within the UK Biobank. We expect the explained variation to be higher, as only estimates of the IC-RHR and not of the complete meta-analyses were used for calculating the variation of RHR in the UK Biobank, since another replication cohort was not available. The sample size represents an expansion of 315% over the previous analysis and increased the explained variance by at least twofold. A total of 1,069 candidate causal genes were prioritized, providing a comprehensive catalogue of data for future studies on RHR offering potential new insights into its biology. In our analyses, four strategies were employed to prioritize candidate causal genes and PHACTR4, ENO3, B3GNT7, KIAA1755 and SENP2 were highlighted by all of the strategies. The gene PHACTR4 has been shown to interact with actin in mouse and is involved in processes ranging from angiogenesis to cell cycle regulation7. PHACTR4 has been
associated with pulse pressure and systolic blood pressure in a previous GWAS8. ENO3
has a function in striated muscle development and regeneration and maps to the region encoding the gene cluster for the sarcomeric myosin heavy chains9. ENO3 has
been associated in a GWAS with red cell distribution width10. B3GNT7 and KIAA1755 have
been linked to RHR in previous GWAS11, and the latter has also been associated with
pulse pressure12 and heart rate variability13. SENP2 has been associated with systolic
and diastolic blood pressure14, type 2 diabetes15, the conduction system of the heart16,17
and estimated glomerular filtration rate18. Our analyses do not provide functional
experimental validation and future studies are necessary to elucidate the role and function of the identified genetic variants and genes in relation to RHR.
In the current work, we validated previous findings that an increase in genetically determined RHR does not increase the risk for CAD an MI11. Studies observing the
phenotype of RHR documented a link between RHR and cardiovascular disease, including CAD and MI1,2. Since this data is observational, unmeasured mechanisms
could have confounded the association. Using genetic variants associated with RHR in MR-analyses we were able to estimate the relationship between RHR and disease in the absence of confounding. Higher RHR is thought to be a marker of an imbalance between the parasympathetic and sympathetic system. RHR is influenced by a plethora of mechanisms and can increase independently of the genetic profile of RHR due to several modifiable and non-modifiable factors, such as obesity, stress and socio-economic status19,20. In the current study, we were unable to show a causal pathway
studies and the current and previous MR analyses, we conclude that RHR probably reflects other cardiovascular risk factors, but is not on the causal pathway leading to CAD or MI.
In conclusion, this GWAS meta-analysis and subsequent bioinformatic analyses and MR-analyses, provides several novel insights into the biology of RHR and the link between RHR and CAD and MI. We reported 276 novel loci associated with RHR and provided a large list of potential candidate genes. We did not find a link between higher genetically predicted RHR and risk CAD and MI risk. The well-known link between higher RHR and cardiovascular disease may therefore likely be explained by ‘environmentally’ acquiring a higher RHR during life.
ONLINE METHODS
Populations
This analysis included 100 studies with data on RHR in up to 835,465 individuals of European ancestry. Further details are provided in Online Table 1.
Ascertainment of resting heart rate
Resting heart rate (RHR) was obtained from electrocardiogram (ECG) in 52 studies (52%), from pulse rate in 16 studies (16%) (of which two were self-measured by the participants) and was derived from a blood pressure monitor in 13 studies (13%). In three (3%) studies, heart rate was derived from electronic medical records. In 16 studies (16%) it was unknown how heart rate was derived (Online Table 1).
Statistical Analysis
All studies performed genetic variant association analyses with RHR using linear regression assuming an additive genetic model (Online Table 1). No transformation of heart rate was performed and extreme (>4SD) phenotypic outliers were excluded. The model was adjusted for: age, age2, body mass index (BMI), sex and study specific
covariates (e.g. principal components, genotyping array and RHR measuring method in case multiple RHR methods were used within a study).
UK Biobank Data
The Wellcome Trust Centre for Human Genetics performed genotyping, quality control before imputation and imputed to Haplotype Reference v1.1 panel (HRC). Analysis
has been restricted to variants that are in the HRC v1.1. Quality control of samples and variants, and imputation was performed by the Wellcome Trust Centre for Human Genetics, as described in more detail elsewhere21.
GWAS in UK Biobank was performed using BOLT-LMM v2.3beta2, employing a mixed linear model that corrects for population structure and cryptic relatedness22. We
performed a GWAS in 484,307 UK Biobank participants. All genetic variants were excluded if they had poor imputation quality score (Info < 0.3) and effective sample size (Neff) < 25 for the SNP computed as (number of samples) x Info x 2 x minor allele frequency (MAF) x (1 - MAF), after which a total of 19,400,416 remained.
The genomic inflation lambda was 1.765 for RHR in the UK Biobank. LDSR intercept showed no genomic inflation due to non-polygenic signals (1.132 ± 0.017)23. The
attenuation ratio statistic indicated polygenicity, not population stratification, to be the main driver of the observed inflation of test statistics for RHR. The QQ-plot for the RHR GWAS within the UK Biobank can be found in Online Figure 1.
International cohorts for resting heart rate (IC-RHR) Data
Samples were genotyped on different genome-wide genotyping arrays and were imputed for 1000 Genomes Phase 1 and 3 (Online Table 1). On cohort level, we performed quality control by: 1) re-formatting and SNP-name harmonization; 2) checking the used reference panel by plotting effect allele frequency plots using 1000G as a reference; 3) check for inflation by plotting QQ plots; 4) check the betas by plotting histograms of the beta of frequency and info; 5) comparison of the expected P-value based on beta and standard error versus reported P-values.
Fixed effect meta-analyses using the inverse variance method in METAL24 were
performed on all cohorts. Genomic control was applied at study-level. All genetic variants were filtered on poor imputation as stated above, after which 27,082,649 variants remained. The genomic inflation lambda was 1.320 for RHR in the 99 cohorts including a total of 351,158 individuals. LDSR intercepts showed no genomic inflation due to non-polygenic signals (1.020 ± 0.010)23. The QQ-plot for the RHR GWAS within
the UK Biobank can be found in Online Figure 2.
Meta-analysis of UK Biobank and IC-RHR data
Again, a fixed effect meta-analyses using the inverse variance method in METAL24 was
performed to pool the data from the UK Biobank and IC-RHR, for up to N=835,465 participants and ~30 M SNPs. We corrected for genomic inflation based on the LDSR
intercepts in the UK Biobank (1.132 ± 0.017) and the IC-RHR (1.020 ± 0.010). The LDSR intercept (standard error) after the final meta-analysis was 1.051 ± 0.002, suggesting little evidence of genomic inflation (Figure 2B).
The PLINK clumping procedure was used to prune genetic variants at a stringent LD of R2<0.005 within a five megabase window into a set of independently associated
variants. Genetic loci were determined by assessing the most significant variants in a one megabase region at either side of the independent variants.
One-stage replication
A one-staged replication analysis was performed. All of the following criteria had to be satisfied for a signal to be reported as a novel replicated signal for RHR:
1. the sentinel SNP has P <1 × 10-8 in the discovery (UKB+ IC-RHR) meta-analysis;
2. the sentinel SNP shows support (P <0.01) in the UKB GWAS alone; 3. the sentinel SNP shows support (P<0.01) in the IC-RHR GWAS alone;
4. the sentinel SNP has concordant direction of effect between UKB and IC-RHR datasets;
The sentinel SNP must not be located within any of the 76 previously reported SNPs described above. We selected all P-value thresholds to be an order of magnitude more stringent than a genome-wide significance P-value to ensure robust results and to minimize false positive findings.
Functional annotation of genes and pathway analyses
For all independent genetic variants, that were genome-wide significantly associated in the final meta-analysis, candidate causal genes were prioritized as follows: 1) by proximity, the nearest gene or any gene within 10 kb; 2) genes containing coding variants in LD with RHR genetic variants at R2>0.8; 3) eQTL genes in LD (R2 >0.8) with RHR
(described below); and 4) DEPICT gene mapping using variants that achieved P<1 × 10-8
(described below).
eQTL analyses
Functional effects of genetic variants associated with RHR were investigated using multiple expression quantitative trait loci (eQTL) mapping. This was done using summary data based Mendelian randomization analysis25 (SMR) in data repositories from GTEx
V726, GTEx brain26, Brain-eMeta eQTL27 and blood eQTL from Westra28 and CAGE29. eQTL
Bonferroni correction for the amount of eQTL’s tested, in this case P<0.05/188,737=2.65 × 10-7. In addition, genes were only taken forward in case they passed the HEIDI test of
P>0.05 and if the lead variants of the eQTL genes were in LD (R2 > 0.8) with the queried
variants.
Pathway analyses by DEPICT
DEPICT was used to find genes associated with identified variants, enriched gene sets and tissues in which these genes are highly expressed. DEPICT.v1.beta version rel137 (obtained from https://data.broadinstitute.org/mpg/depict/) was used to perform integrated gene function analyses as stated above. DEPICT was run using all genetic variants that achieved P<1 × 10-830.
Mendelian randomization analyses
Genetic instruments
All independent variants at P <1 × 10-8 in the final meta-analysis were taken forward. To
ensure the absence of overlap between the cohorts used to gain insights in the exposure and the outcome, beta’s and standard errors were taken from the IC-RHR data to test the associations within the UK Biobank, whereas beta’s and standard errors were taken from the UK Biobank to test the association with CAD and MI in the CARDIoGRAMplusC4D cohort (Figure 1B). Eight variants could not be found within the IC-RHR data and proxies were searched for these variants (Online Table 2 & 11) . One genetic variant could not be found within the UK Biobank cohort and hence we searched for a proxy (Online Table 2 & 12). The same approach was used to test the association between the 76 previously established RHR variants with CAD and MI.
Analyses in the UK Biobank
The prevalence and incidence of CAD and MI in the UK Biobank were captured through data collected at the Assessment Centre in-patient Health Episode Statistics (HES) in combination with data on cause of death from the National Health Service (NHS) Information Centre. HES data was available up to 31-03-2017 for English participants, 29-02-2016 for Walsh participants and 31-10-2016 for Scottish. Information on cause of death was available for participants from England and Wales until 31-01-2018, and from the NHS Central Register Scotland for participants from Scotland until 30-11-2016. Cox regression analyses were performed to obtain the beta’s and standard errors of all the variants from the meta-analyses. Similar to the GWAS analyses, age, age2, BMI, sex
and study specific covariates, principal components and genotyping array were used as covariates. After mandatory genetic exclusions due to high heterozygosity, missingness
or differences between reported and inferred gender (n= 1,341), exclusions based on familial relatedness (n = 74,471) and missing phenotype data (n = 1,778), data of 410,694 individuals was available for Cox regression analyses. Among these individuals, 27,780 and 12,025 had CAD and MI, respectively.
Similar analyses were performed to obtain the beta’s and standard errors of the previously discovered variants. However, we excluded the 134,251 participants used for the discovery of these variants, leaving 19,307 CAD- and 8,373 MI cases.
Analyses in CARDIoGRAMplusC4D
For MR-analyses of RHR on CAD and MI, the CARDIoGRAMplusC4D consortium’s 1000 genomes-based meta-analysis were used. In total, this study includes 60,801 cases of CAD and 123,504 controls31. CAD events were defined as a documented diagnosis
of CAD, such as acute coronary syndrome (including MI), chronic stable angina, or >50% stenosis of at least one coronary vessel, as well as those who had undergone percutaneous coronary revascularization or coronary artery bypass grafting. In addition, the CARDIoGRAMplusC4D consortium included a subset of 43,676 individuals with a reported history of MI and 128,199 controls.
Weak instrument bias in MR analyses
Potential weak instrument bias was explored by calculating the F statistic using the equation F=((n-k-1)/k) × (R2/(1−R2)) 32. Here, n is the sample size, k is the number of SNPs
and R2 is the percent variance explained in RHR by all SNPs combined32. We tested the
F-statistic using the effect size of the variants in the UK Biobank, since only phenotype data of the UK Biobank was at our disposal. The most widely used cut-off for the F statistic to prevent weak instrument bias and thus violation of the `NO Measurement Error’ (NOME) assumption is a F statistic of >10 32. To evaluate weak instrument bias
in the MR-Egger regression analysis, the variation between individual genetic variant estimates for RHR was calculated (I2
GX) 33. An I2GX of >95% was considered low risk of
measurement error.
Pleiotropy analyses in MR analyses
Pleiotropy in MR analyses occurs when genetic variants exert their effect on a phenotype through multiple pathways. Those genetic variants may therefore influence the outcome independent of the phenotype studied. Therefore, careful consideration of confounding and bias of MR estimates must be taken into account and investigation of pleiotropy is therefore essential. Heterogeneity, an indicator of potential pleiotropy since low heterogeneity indicates that estimates between genetic variants should vary
by chance only, was assessed by I2-index34 and Cochran’s Q35 statistics in the
fixed-effect variance weighted analyses. An I2 index>25% and Cochran’s Q P-value of <0.05
were considered as an indication of heterogeneity and, as a consequence, of potential pleiotropy.
Next, MR-Egger test was performed, which in contrast to the IVW method, does not assume absence of pleiotropic effects for all included genetic variants36. A variable
intercept of the y-axis is allowed in the MR-Egger regression, allowing genetic variants to be invalid. Hence, the intercept of the y-axis provides additional information, as large deviations from the non-zero intercept represent large average horizontal pleiotropic effect36. A difference of the intercept with an otherwise zero intercept was tested
and a P-value threshold of >0.05, was considered to provide evidence for absence of horizontal pleiotropic bias. The MR-Egger follows the InSIDE assumption, which states that the association of genetic variants with the exposure are independent of the direct effects of the genetic variants on the outcome36. Under the InSIDE assumption, the
slope coefficient from the MR-Egger regression is a consistent estimate of the causal effect. Again, heterogeneity was assessed, this time using Rucker’s Q35. We adopted
the Rucker framework to move between different MR methods. In case a significant difference (P<0.05) between heterogeneity in the IVW analyses (Cochran’s Q) and MR-egger analyses (Rucker’s Q) was observed, i.e. a significant Q-Q’35, the MR-Egger test was
used study the genetic association between the exposure and outcome. In addition, MR pleiotropy residual sum and outlier (MR-PRESSO) was used to test for pleiotropic effects. MR-PRESSO can identify outlier variants based on their observed distance from the regression line, as compared with their expected distance based on the assumption of no horizontal pleiotropy37,38. Pleiotropic effects can be detected, and re-analyses after
outlier removal can be performed, correcting for these pleiotropic effects.
Sensitivity analyses in MR analyses
Additionally several sensitivity analyses were performed, including weighted median analysis38 (which allows up to 50% of information from variants to violate MR
assumptions) and weighted mode-based estimator MR analyses39 (which allows even
the majority of all variants to be invalid in case the largest number of those which produce similar MR results are valid).
REFERENCES
1. Jouven X, Empana J-P, Schwartz PJ, Desnos M, Courbon D, Ducimetière P. Heart-Rate Profile during Exercise as a Predictor of Sudden Death. N Engl J Med. 2005;352(19):1951-1958. doi:10.1056/NEJMoa043012
2. Zhang D, Wang W, Li F. Association between resting heart rate and coronary artery disease, stroke, sudden death and noncardiovascular diseases: a meta-analysis. Can Med Assoc J. 2016;188(15):E384-E392. doi:10.1503/cmaj.160050
3. Jensen MT, Wod M, Galatius S, Hjelmborg JB, Jensen GB, Christensen K. Heritability of resting heart rate and association with mortality in middle-aged and elderly twins. Heart. 2018;104(1):30-36. doi:10.1136/heartjnl-2016-310986
4. Russell MW, Law I, Sholinsky P, Fabsitz RR. Heritability of ECG measurements in adult male twins. J Electrocardiol. 1998;30 Suppl:64-68. http://www.ncbi.nlm.nih.gov/pubmed/9535482. Accessed April 9, 2019.
5. De Geus EJC, Kupper N, Boomsma DI, Snieder H. Bivariate Genetic Modeling of Cardiovascular Stress Reactivity: Does Stress Uncover Genetic Variance? Psychosom Med. 2007;69(4):356-364. doi:10.1097/PSY.0b013e318049cc2d
6. Zhu Z, Wang X, Li X, et al. Genetic overlap of chronic obstructive pulmonary disease and cardiovascular disease-related traits: a large-scale genome-wide cross-trait analysis. Respir
Res. 2019;20(1):64. doi:10.1186/s12931-019-1036-8
7. Huet G, Rajakyla EK, Viita T, et al. Actin-regulated feedback loop based on Phactr4, PP1 and cofilin maintains the actin monomer pool. J Cell Sci. 2013;126(2):497-507. doi:10.1242/ jcs.113241
8. Evangelou E, Warren HR, Mosen-Ansorena D, et al. Genetic analysis of over 1 million people identifies 535 new loci associated with blood pressure traits. Nat Genet. 2018;50(10):1412-1425. doi:10.1038/s41588-018-0205-x
9. Peshavaria M, Day IN. Molecular structure of the human muscle-specific enolase gene (ENO3). Biochem J. 1991;275 ( Pt 2)(2):427-433. doi:10.1042/bj2750427
10. Kichaev G, Bhatia G, Loh P-R, et al. Leveraging Polygenic Functional Enrichment to Improve GWAS Power. Am J Hum Genet. 2019;104(1):65-75. doi:10.1016/j.ajhg.2018.11.008
11. Eppinga RN, Hagemeijer Y, Burgess S, et al. Identification of genomic loci associated with resting heart rate and shared genetic predictors with all-cause mortality. Nat Genet. 2016;48(12):1557-1563. doi:10.1038/ng.3708
12. Giri A, Hellwege JN, Keaton JM, et al. Trans-ethnic association study of blood pressure determinants in over 750,000 individuals. Nat Genet. 2019;51(1):51-62. doi:10.1038/s41588-018-0303-9
13. Nolte IM, Munoz ML, Tragante V, et al. Genetic loci associated with heart rate variability and their effects on cardiac disease risk. Nat Commun. 2017;8(1):15805. doi:10.1038/ ncomms15805
14. Hoffmann TJ, Ehret GB, Nandakumar P, et al. Genome-wide association analyses using electronic health records identify new loci influencing blood pressure variation. Nat Genet. 2017;49(1):54-64. doi:10.1038/ng.3715
15. Mahajan A, Wessel J, Willems SM, et al. Refining the accuracy of validated target identification through coding variant fine-mapping in type 2 diabetes. Nat Genet. 2018;50(4):559-571. doi:10.1038/s41588-018-0084-1
16. Prins BP, Mead TJ, Brody JA, et al. Exome-chip meta-analysis identifies novel loci associated with cardiac conduction, including ADAMTS6. Genome Biol. 2018;19(1):87. doi:10.1186/ s13059-018-1457-6
17. van Setten J, Brody JA, Jamshidi Y, et al. PR interval genome-wide association meta-analysis identifies 50 loci associated with atrial and atrioventricular electrical activity. Nat Commun. 2018;9(1):2904. doi:10.1038/s41467-018-04766-9
18. Wuttke M, Li Y, Li M, et al. A catalog of genetic loci associated with kidney function from analyses of a million individuals. Nat Genet. 2019;51(6):957-972. doi:10.1038/s41588-019-0407-x
19. Shigetoh Y, Adachi H, Yamagishi S -i., et al. Higher Heart Rate May Predispose to Obesity and Diabetes Mellitus: 20-Year Prospective Study in a General Population. Am J Hypertens. 2009;22(2):151-155. doi:10.1038/ajh.2008.331
20. Zhang A, Hughes JT, Brown A, et al. Resting heart rate, physiological stress and disadvantage in Aboriginal and Torres Strait Islander Australians: analysis from a cross-sectional study. BMC
Cardiovasc Disord. 2016;16(1):36. doi:10.1186/s12872-016-0211-9
21. Bycroft C, Freeman C, Petkova D, et al. The UK Biobank resource with deep phenotyping and genomic data. Nature. 2018;562(7726):203-209. doi:10.1038/s41586-018-0579-z
22. Loh P-R, Tucker G, Bulik-Sullivan BK, et al. Efficient Bayesian mixed-model analysis increases association power in large cohorts. Nat Genet. 2015;47(3):284-290. doi:10.1038/ng.3190 23. Loh P-R, Kichaev G, Gazal S, Schoech AP, Price AL. Mixed-model association for biobank-scale
datasets. Nat Genet. 2018;50(7):906-908. doi:10.1038/s41588-018-0144-6
24. Willer CJ, Li Y, Abecasis GR. METAL: fast and efficient meta-analysis of genomewide association scans. Bioinformatics. 2010;26(17):2190-2191. doi:10.1093/bioinformatics/btq340
25. Wu Y, Zeng J, Zhang F, et al. Integrative analysis of omics summary data reveals putative mechanisms underlying complex traits. Nat Commun. 2018;9(1):918. doi:10.1038/s41467-018-03371-0
26. Aguet F, Brown AA, Castel SE, et al. Genetic effects on gene expression across human tissues.
Nature. 2017;550(7675):204-213. doi:10.1038/nature24277
27. Qi T, Wu Y, Zeng J, et al. Identifying gene targets for brain-related traits using transcriptomic and methylomic data from blood. Nat Commun. 2018;9(1):2282. doi:10.1038/s41467-018-04558-1
28. Westra H-J, Peters MJ, Esko T, et al. Systematic identification of trans eQTLs as putative drivers of known disease associations. Nat Genet. 2013;45(10):1238-1243. doi:10.1038/ng.2756 29. Lloyd-Jones LR, Holloway A, McRae A, et al. The Genetic Architecture of Gene Expression in
30. Pers TH, Karjalainen JM, Chan Y, et al. Biological interpretation of genome-wide association studies using predicted gene functions. Nat Commun. 2015;6(1):5890. doi:10.1038/ ncomms6890
31. Nikpay M, Goel A, Won H-H, et al. A comprehensive 1000 Genomes–based genome-wide association meta-analysis of coronary artery disease. Nat Genet. 2015;47(10):1121-1130. doi:10.1038/ng.3396
32. Burgess S, Thompson SG. Avoiding bias from weak instruments in Mendelian randomization studies. Int J Epidemiol. 2011;40(3):755-764. doi:10.1093/ije/dyr036
33. Bowden J, Del Greco M. F, Minelli C, Davey Smith G, Sheehan NA, Thompson JR. Assessing the suitability of summary data for two-sample Mendelian randomization analyses using MR-Egger regression: the role of the I2 statistic. Int J Epidemiol. 2016;45(6):dyw220. doi:10.1093/ ije/dyw220
34. Greco M F Del, Minelli C, Sheehan NA, Thompson JR. Detecting pleiotropy in Mendelian randomisation studies with summary data and a continuous outcome. Stat Med. 2015;34(21):2926-2940. doi:10.1002/sim.6522
35. Bowden J, Del Greco M F, Minelli C, Davey Smith G, Sheehan N, Thompson J. A framework for the investigation of pleiotropy in two-sample summary data Mendelian randomization. Stat
Med. 2017;36(11):1783-1802. doi:10.1002/sim.7221
36. Bowden J, Davey Smith G, Burgess S. Mendelian randomization with invalid instruments: effect estimation and bias detection through Egger regression. Int J Epidemiol. 2015;44(2):512-525. doi:10.1093/ije/dyv080
37. Verbanck M, Chen C-Y, Neale B, Do R. Detection of widespread horizontal pleiotropy in causal relationships inferred from Mendelian randomization between complex traits and diseases.
Nat Genet. 2018;50(5):693-698. doi:10.1038/s41588-018-0099-7
38. Bowden J, Davey Smith G, Haycock PC, Burgess S. Consistent Estimation in Mendelian Randomization with Some Invalid Instruments Using a Weighted Median Estimator. Genet
Epidemiol. 2016;40(4):304-314. doi:10.1002/gepi.21965
39. Hartwig FP, Davey Smith G, Bowden J. Robust inference in summary data Mendelian randomization via the zero modal pleiotropy assumption. Int J Epidemiol. 2017;46(6):1985-1998. doi:10.1093/ije/dyx102
