Associations of Genetically Predicted Lipoprotein (a) Levels with Cardiovascular Traits in Individuals of European and African Ancestry

Abstract

Background -

Lipoprotein (a) [Lp(a)] levels are higher in individuals of African ancestry (AA) than in individuals of European ancestry (EA). We examined associations of genetically predicted Lp(a) levels with 1) atherosclerotic cardiovascular disease (ASCVD) subtypes: coronary heart disease (CHD), cerebrovascular disease (CVD), peripheral artery disease (PAD), and abdominal aortic aneurysm (AAA); and 2) non-ASCVD phenotypes, stratified by ancestry.

Methods -

We performed 1) Mendelian randomization (MR) analyses for previously reported cardiovascular associations, and 2) phenome-wide MR (MR-PheWAS) analyses for novel associations. Analyses were stratified by ancestry in electronic MEdical Records and GEnomics, United Kingdom Biobank, and Million Veteran Program cohorts separately and in a combined cohort of 804,507 EA and 103,580 AA participants.

Results -

In MR analyses using the combined cohort, a 1-standard deviation (SD) genetic increase in Lp(a) level was associated with ASCVD subtypes in EA – odds ratio and 95% confidence interval for CHD 1.28(1.16–1.41); CVD 1.14(1.07–1.21); PAD 1.22(1.11–1.34); AAA 1.28(1.17–1.40); in AA the effect estimate was lower than in EA and nonsignificant for CHD 1.11(0.99–1.24) and CVD 1.06(0.99–1.14) but similar for PAD 1.16(1.01–1.33) and AAA 1.34(1.11–1.62). In EA, a 1-SD genetic increase in Lp(a) level was associated with aortic valve disorders 1.34(1.10–1.62), mitral valve disorders 1.18(1.09–1.27), congestive heart failure 1.12(1.05–1.19), and chronic kidney disease 1.07(1.01–1.14). In AA no significant associations were noted for aortic valve disorders 1.08(0.94–1.25), mitral valve disorders 1.02(0.89–1.16), congestive heart failure 1.02(0.95–1.10), or chronic kidney disease 1.05(0.99–1.12). MR-PheWAS identified novel associations in EA with arterial thromboembolic disease, non-aortic aneurysmal disease, atrial fibrillation, cardiac conduction disorders, and hypertension.

Conclusions -

Many cardiovascular associations of genetically increased Lp(a) that were significant in EA were not significant in AA. Lp(a) was associated with ASCVD in four major arterial beds in EA but only with PAD and AAA in AA. Additional, novel cardiovascular associations were detected in EA.

Introduction

Lipoprotein (a) [Lp(a)] is a causal risk factor for atherosclerotic cardiovascular disease (ASCVD)¹, but its physiologic and pathophysiologic roles remain elusive.² Lp(a) promotes atherothrombosis through multiple potential mechanisms including pro-inflammatory (carrier of oxidized phospholipids, stimulating cytokine release from leukocytes, etc.), pro-atherogenic (upregulation of endothelial cell adhesion molecules, smooth muscle cell proliferation, etc.), and pro-thrombotic (decreased plasminogen activation, decreased fibrin degradation, etc.).³ The association of measured circulating Lp(a) levels,^{3, 4} or genetically predicted Lp(a) levels by Mendelian randomization^{5, 6} (MR) with coronary heart disease (CHD), ischemic cerebrovascular disease (CVD), peripheral artery disease (PAD), and abdominal aortic aneurysm (AAA) is established primarily in individuals of European ancestry (EA). Various additional roles for Lp(a) have been speculated,⁷ but a systematic investigation on a phenome-wide basis, stratifying for ancestry, is lacking.

A unique aspect of Lp(a) biology is that individuals of African ancestry (AA) have up to 4-fold higher mean Lp(a) levels than EA,^8–11 and tend to have smaller isoform size.¹⁰ Despite having higher levels than EA,^{8, 12} it is unclear whether Lp(a) levels are associated with ASCVD in AA.^10–16 Previous studies in EA identified associations with CHD, CVD, PAD, AAA, aortic valve stenosis and regurgitation, mitral valve regurgitation, congestive heart failure, and chronic kidney disease,^{5, 6, 17, 18} but little or no information is available for possible associations in AA for most of these phenotypes.

We investigated differences in the associations of genetically predicted Lp(a) levels with EHR-derived phenotypes between EA and AA, within the electronic MEdical Records and GEnomics (eMERGE) network,¹⁹ United Kingdom (UK) Biobank, and the Million Veteran Program (MVP)²⁰ datasets (combined sample size of 804,507 and 103,580 for EA and AA, respectively). We leveraged EHR-derived phenotypes linked to high-density genotype data²¹ and compared associations of genetically predicted Lp(a) levels with 1) CHD, CVD, PAD, and AAA using 2-sample MR analyses, 2) aortic valve disorders, mitral valve disorders, congestive heart failure, and chronic kidney disease using 2-sample MR analyses, and 3) the entire spectrum of EHR-derived phenotypes without previously known Lp(a) association in both ancestry groups using MR-PheWAS analyses.

Methods

The data that support the findings of this study are available from the corresponding author upon reasonable request. Each study cohort obtained approval from its respective institutional review board and all participants gave informed consent. Demographics of the UK Biobank and MVP replication cohorts are summarized in Table 1. The methods are summarized in Figure 1. All methods are described in the supplemental material.

Table 1.

Demographic characteristics of participants in discovery and replication cohorts

	Discovery Cohort		Replication Cohorts
	eMERGE Network		UK Biobank		MVP
Ancestry	European	African	European	African	European	African
n	65,418	9,850	420,531	6,636	318,558	87,094
Mean age ± SD (years)	64±18*	53±17*	58±7†	51±5†	64±13†	58±12†
Female (%)	54	66	55	61	7	13

eMERGE: electronic MEdical Records and GEnomics, MVP: Million Veteran Program, SD: standard deviation, UK: United Kingdom

^*Age at last follow-up

^†Age at enrollment

Figure 1.

Overview of analyses exploring differences by ancestry in genetically predicted Lp(a) level-phenotype associations. AA: individuals of African ancestry, ASCVD: atherosclerotic cardiovascular disease, EA: individuals of European ancestry, EHR: electronic health record, eMERGE: electronic MEdical Records and GEnomics, ICD: international classification of diseases, IVW: inverse variance weighted, LD: linkage disequilibrium, Lp(a): lipoprotein(a), MAF: minor allele frequency, MR: Mendelian randomization, PC: principal component, SNV: single-nucleotide variant, UK: United Kingdom

Results

ASCVD Subtype MR Analyses

The genetic instruments that passed were selected included 5 SNVs in EA and 5 non-overlapping SNVs in AA that were strongly associated with Lp(a) levels across discovery²² and validation datasets (Table 2) and these explained 14% and 20% of the variation in Lp(a) levels in EA and AA, respectively. Results of the four MR methods for each individual cohort and the combined cohort are presented in Supplemental Excel File I. Only the results from the combined cohort are presented below (n=804,507 and 103,580 for EA and AA, respectively). Across all MR methods, a 1-standard deviation (SD) genetically predicted increase in Lp(a) level was associated with the four ASCVD subtypes in EA, consistent with previous reports⁶ (Figure 2A, blue). In AA, the effect sizes of a 1-SD genetically predicted increase in Lp(a) with CHD and CVD were smaller than in EA and non-significant (Figure 2A, red), and although the Bayesian posterior probability of effect was largely in favor of OR>1, the majority of this mass was concentrated at lower ranges than in EA (Figure 2B). This is in contrast to the PAD and AAA associations in AA which were significant and similar in magnitude to EA for both the frequentist (Figure 2A) and Bayesian (Figure 2B) analyses. The direction of effects and effect size differences in the ancestry-specific estimates were similar in other MR methods (Supplemental Excel File I). Power calculations based on the case and control numbers in the combined cohort analysis are presented in Supplemental Table I.

Table 2.

SNVs selected as genetic instruments for Lp(a) levels

				Zekavat et al 201822						UK Biobank‡
Chromosome 6, Position*	Variant	Effect allele	Non-effect allele	Beta	SE	P-value	MAF	Sample size	F-statistic	Beta	SE	P-value	MAF	Sample size	F-statistic
European Ancestry
160416196	rs73019695	A	T	0.298	0.049	1.28E-09	0.027	6440	36.84	0.208	0.008	3.57E-150	0.033	318922	681.16
160620141	rs75885118	C	T	−0.321	0.035	4.93E-20	0.054	6440	84.01	−0.165	0.007	5.09E-135	0.051	318922	611.56
160765055	rs572889	G	A	−0.364	0.061	2.42E-09	0.018	6440	35.60	−0.363	0.000	NC†	0.027	318922	NC†
161089307	rs56393506	T	C	0.487	0.021	1.62E-124	0.187	6440	563.28	1.153	0.000	NC†	0.171	318922	NC†
161304242	rs62435349	T	A	−0.362	0.027	2.16E-42	0.104	6440	186.19	−0.212	0.000	NC†	0.099	318922	NC†
Chromosome 6, Position*	Variant	Effect allele	Non-effect allele	Beta	SE	P-value	MAF	Sample size	F-statistic	Beta	SE	P-value	MAF	Sample size	F-statistic
African Ancestry
160901411	rs114067707	A	G	−0.621	0.077	1.08E-15	0.023	2832	64.28	−0.214	0.050	1.90E-05	0.024	5140	18.30
160946747	rs75143493	G	T	0.645	0.099	5.85E-11	0.013	2832	42.87	0.618	0.076	6.06E-16	0.009	5140	65.42
160986915	rs6938647	C	A	0.463	0.059	4.32E-15	0.041	2832	61.55	0.281	0.048	5.39E-09	0.027	5140	34.04
161025797	rs78347018	C	T	−0.760	0.059	2.14E-38	0.041	2832	167.89	−0.272	0.041	2.66E-11	0.031	5140	44.40
161087440	rs41269135	A	G	0.785	0.051	1.28E-53	0.053	2832	237.65	0.587	0.037	1.02E-56	0.039	5140	251.81

NC: not calculated, SE: standard error, MAF: minor allele frequency, UK: United Kingdom

^*Position in human genome assembly hg19 (GRCh37)

^†Unable to calculate as the reported SE in Pan-UK Biobank was provided as 0 at the given precision level in the summary statistics

^‡ https://pan.ukbb.broadinstitute.org

Figure 2.

Ancestry-stratified 2-sample MR estimates of ASCVD associations and other phenotypes previously known to be associated with Lp(a). (A) Forest plots of the combined (eMERGE, UK Biobank, and MVP) cohort MR estimates and (B) Bayesian posterior probability MR estimates for ASCVD subtypes. (C) Forest plots of combined cohort MR estimates and (D) Bayesian posterior probability MR estimates for other phenotypes with previously known associations. For all panels, results are for the inverse variance weighted method. For panels B and D the black circle represents the posterior mean and the interval represents the 95% highest posterior density. ASCVD: atherosclerotic cardiovascular disease, CI: confidence interval, CrI: credible interval, eMERGE: electronic MEdical Records and GEnomics, Lp(a): lipoprotein(a), MR: Mendelian randomization, MVP: Million Veteran Program, PPr: posterior probability, OR: odds ratio, SD: standard deviation, UK: United Kingdom.

*Results for abdominal aortic aneurysm in African ancestry are only for MVP cohort as the other two cohorts did not meet minimum case requirements.

†The Bayesian posterior probability for the indicated OR range.

To assess whether the association estimates differed by ancestry, we quantified heterogeneity when combining EA and AA results for each ASCVD phenotype using I² and Cochran Q heterogeneity test (Table 3). I² across the MR methods ranged between 0.62–0.90 in CHD and 0.45–0.92 in CVD, indicating moderate-to-high heterogeneity. In PAD and AAA, there was no measurable heterogeneity (except when using the MR-Egger method for PAD, I²=0.38). MR sensitivity analyses for ASCVD associations are presented in Supplemental Figures I–IV and Supplemental Excel File I.

Table 3.

Heterogeneity of MR estimates with combined results from both ancestry groups

Phenotype (phecode)	MR Method	I 2	Heterogeneity P-value
CHD (411)	IVW	0.69	0.071
	MR-Egger	0.83	0.015
	Weighted median	0.62	0.107
	Weighted mode	0.90	2.03E-03
CVD (433)	IVW	0.54	0.139
	MR-Egger	0.92	3.56E-04
	Weighted median	0.45	0.178
	Weighted mode	0.87	5.66E-03
PAD (443)	IVW	0	0.549
	MR-Egger	0.38	0.205
	Weighted median	0	0.588
	Weighted mode	0	0.693
AAA (442.11)	IVW	0	0.648
	MR-Egger	0	0.629
	Weighted median	0	0.406
	Weighted mode	0	0.766
Aortic valve disorders (395.2)	IVW	0.64	0.095
	MR-Egger	0.91	8.00E-04
	Weighted median	0	0.336
	Weighted mode	0.87	5.82E-03
Mitral valve disorders (395.1)	IVW	0.70	0.069
	MR-Egger	0.11	0.290
	Weighted median	0.64	0.096
	Weighted mode	0.59	0.120
Congestive heart failure (428)	IVW	0.70	0.069
	MR-Egger	0.58	0.125
	Weighted median	0.74	0.048
	Weighted mode	0.72	0.060
Chronic kidney disease (585.3)	IVW	0	0.728
	MR-Egger	0	0.921
	Weighted median	0	0.968
	Weighted mode	0	0.935

AAA: abdominal aortic aneurysm, CHD: coronary heart disease, CVD: cerebrovascular disease, IVW: inverse variance weighted, MR: Mendelian randomization, PAD: peripheral artery disease

MR Analyses of Other Previously Described Lp(a)-Associated Phenotypes

Additional associations have been previously demonstrated with genetically predicted Lp(a) levels in EA, but little is known about whether these associations are also present in AA. Therefore, we performed 2-sample MR analyses with these phenotypes (aortic valve disorders, mitral valve disorders, congestive heart failure, and chronic kidney disease) in both ancestry groups at the cohort level as well as with the combined cohort (Supplemental Excel File I). In the AA combined cohort analysis (Figure 2C, red), per 1-SD genetically predicted increase in Lp(a), associations with all phenotypes were weaker than in EA except chronic kidney disease, which although not significantly associated, had a similar magnitude of association as in EA (Figure 2C, blue); this was similar across other MR methods (Supplemental Excel File I). Power calculations based on the case and control numbers in the combined cohort analysis are given in Supplemental Table I.

As before, we quantified the heterogeneity when combining EA and AA results for these phenotypes (Table 3) and found I² across all MR methods to range between 0.64–0.91 (except weighted median, which showed no heterogeneity) in aortic valve disorders, 0.11–0.70 in mitral valve disorders, and 0.58–0.74 for congestive heart failure, indicating moderate-to-high heterogeneity in these three phenotypes. In chronic kidney disease there was no measurable heterogeneity with any method. Most of the mass of the posterior probability distribution of ORs for the associations with aortic valve disorders, mitral valve disorders, and congestive heart failure was located in the higher ranges in EA than in AA. For chronic kidney disease, EA and AA had similar posterior probabilities of having a modest OR (Figure 2D). MR sensitivity analyses for these associations are presented in Supplemental Figures V–VIII and Supplemental Excel File I.

MR-PheWAS of Phenotypes without Known Lp(a) Associations

Of the 1,812 and 1,778 phecodes available in EA and AA, respectively, 1,162 in EA and 842 in AA met the minimum required number of cases or prevalence threshold in the discovery cohort.

European ancestry

Among phenotypes not known to be associated with Lp(a), MR-PheWAS in the eMERGE discovery cohort identified 34 phecodes that had suggestive associations (P-value <0.01) with genetically predicted Lp(a) levels in EA. Of these, 18 were excluded from further analyses because either the P-value (IVW) was >0.05 in both replication cohorts, or the direction of effect was not consistent in one of the replication cohorts compared to the discovery cohort. The remaining 16 phecodes underwent analysis using the combined cohort where 15 of these phecode associations were deemed significant after Bonferroni correction. These included phecodes that fit into 6 categories: arterial thromboembolic disease, non-aortic aneurysmal disease, atrial fibrillation, cardiac conduction disorders, hypertension, and atherosclerosis-related traits (Figure 3A). Results of all four methods in each individual cohort as well as the combined cohort for the 9 significant phenotypes are in Supplemental Excel File II.

Figure 3.

MR-PheWAS estimates for novel associations, thromboembolic disease, and diabetes mellitus. Forest plots (A) of novel associations from the combined (eMERGE, UK Biobank, and MVP) European Ancestry cohort; (B) of arterial and venous thromboembolic disease phenotypes within the eMERGE cohort; and (C) diabetes phenotypes within the eMERGE cohort. For all panels, results are for the inverse variance weighted method. CI: confidence interval, eMERGE: electronic MEdical Records and GEnomics, Lp(a): lipoprotein(a), MR: Mendelian randomization, MVP: Million Veteran Program, OR: odds ratio, SD: standard deviation, UK: United Kingdom.

Although we detected an association with arterial thromboembolic disease, we did not find a significant association with any venous thromboembolism phenotypes using the IVW method (Figure 3B) or other MR methods (Supplemental Table II) in the eMERGE cohort. Given conflicting reports about possible associations between Lp(a) level and type 2 diabetes,^23–26 we looked for, but did not find a significant association with any diabetes-related phecodes with the IVW method (Figure 3C) or other MR methods (Supplemental Table II) with the exception of type 1 diabetes with the weighted median method (P=0.032; OR 1.15 [1.01–1.32]).

African ancestry

MR-PheWAS in the eMERGE discovery cohort identified 7 phecodes that had suggestive associations (P-value <0.01) with genetically predicted Lp(a) levels in AA, however all 7 of these were excluded from further analyses because the P-value (IVW) was >0.05 in both replication cohorts. Power calculations for both EA and AA based on a range of prevalence from 0.01–0.20 in the eMERGE discovery cohort are presented in Supplemental Table III.

Discussion

Ancestry-stratified MR analyses of 804,507 EA and 103,580 AA individuals demonstrated that for many associations of genetically mediated increase in Lp(a) levels with cardiovascular traits observed in EA, no significant association was observed in AA (summarized in Figure 4). In AA the effect estimates of CHD and CVD were less than in EA and nonsignificant, whereas the associations with PAD or AAA were of similar magnitude. Associations with disorders of aortic and mitral valves, congestive heart failure, and chronic kidney disease were absent in AA. We also identified several novel associations with additional cardiovascular traits in EA. These observations of a stronger association between genetically predicted Lp(a) levels and CHD in EA than AA may explain the descrepancies in prior studies where some found associations between Lp(a) levels and CHD in AA while others did not.^10–16

Figure 4.

Summary of strategy and results from MR and MR-PheWAS analyses. ASCVD: atherosclerotic cardiovascular disease, CI: confidence interval, eMERGE: electronic MEdical Records and GEnomics, Lp(a): lipoprotein(a), MR: Mendelian randomization, MVP: Million Veteran Program, OR: odds ratio, SD: standard deviation, SNV: single-nucleotide variant, UK: United Kingdom. The images in the MR-PheWAS box were created with BioRender.com.

While Lp(a) was associated with ASCVD in four major arterial beds in EA, in AA the association was limited to PAD and AAA. Lp(a) level has been previously reported to be associated with AAA in EA and AA,¹⁸ but we are unaware of previous ancestry-stratified reports for association with PAD. Further investigation is needed to better understand the basis of this observation. Endothelial cell ‘phenotypes’ are known to vary across arterial beds²⁷ and it can be speculated that such heterogeneity might influence effects of Lp(a) on different arterial beds in AA. Previous studies have identified differences in how endothelial cells in EA and AA respond to various stimuli^{28, 29} which results in varied responses to different classes of antihypertensive medications.³⁰ It is possible such endothelial heterogeneity influences atherothrombotic effects of Lp(a) in different arterial beds.

The apolipoprotein (a) component of Lp(a) inhibits plasminogen activation and competes for binding of plasminogen to fibrin clots in vitro,^{31, 32} We detected a hitherto unreported association of genetically predicted Lp(a) level with arterial thromboembolic diseases in EA, supporting in vitro observations and suggesting that Lp(a) affects both atherogenesis and thrombosis in EA; however, we did not find an association with venous thromboembolism, consistent with prior studies.^{5, 33} These observations suggest differential effects of Lp(a) in arteries versus veins.

We confirmed previously reported associations in EA including aortic valve disorders (stenosis and regurgitation), mitral valve disorders (regurgitation), congestive heart failure, and chronic kidney disease.^{5, 6} However, these associations were not detected in AA. Pathways that mediate the association of Lp(a) level with aortic valve disorders (stenosis and regurgitation), congestive heart failure, or chronic kidney disease, are unclear although increased propensity to ASCVD may play a role.

Our agnostic phenome-wide analysis to identify associations of Lp(a) with the spectrum of EHR-derived phenotypes using a MR-PheWAS approach identified additional novel associations including aneurysmal disease in vascular beds other than the aorta, atrial fibrillation, cardiac conduction disorders, blood pressure disorders, and other atherosclerosis-related traits. While the association with arterial aneurysmal disease is likely mediated by ASCVD, the basis for the remaining associations is unclear at present. Associations with non-cardiovascular phenotypes were not detected in either ancestry group, suggesting that the pathophysiology of Lp(a) is mostly restricted to cardiovascular traits. In particular, we did not detect an association with type 2 diabetes.

Our findings could have potential clinical implications. The AHA/ACC guideline³⁴ recommends measurement of Lp(a) levels in certain settings whereas the European Atherosclerosis Society recommends routine measurement in all individuals for risk stratification.³⁵ However, modification of these recommendations may be warranted for AA; rather than using a uniform threshold for all ancestry groups, ancestry-specific thresholds might be more appropriate as has also been suggested by others.³⁶ Additionally, new drugs that specifically lower Lp(a) are being evaluated in clinical trials and it is important that these studies include an adequate representation of AA to determine whether any benefit observed differs by ancestry. In a recent clinical trial demonstrating a marked reduction of Lp(a)³⁷ with an antisense therapy, 97% of the participants were EA. Finally, medically lowered Lp(a) levels are unlikely to have unintended effects as associations with non-cardiovascular phenotypes were not detected in either ancestry group.

Strengths and Limitations

Our study examines the causal role of genetically predicted Lp(a) levels in ASCVD and other phenotypes in two ancestry groups from three large cohorts. For many of these phenotypes this is the first report in AA, while for others we examine results from a much larger AA cohort than in previously published studies. We also report results of the first PheWAS for Lp(a) in AA as well as the first PheWAS to examine the entire range of EHR phenotypes in either ancestry group.

Due to the difference in Lp(a) levels between EA and AA, we examined associations with a 1-SD change in ancestry-specific genetically predicted Lp(a) levels instead of absolute values. The differences in associations between the two ancestry groups could have been greater if we used absolute values since a 1-SD change in genetically predicted Lp(a) levels would correspond to a large change in absolute Lp(a) levels in AA. We did not have access to the individual level data for Lp(a) levels within the UK Biobank. In the PheWAS approach, power to detect weaker phenotypic associations may vary depending on the number of cases for a phenotype and, although comparatively large, the sample size of AA was smaller than EA. In both ancestry groups, particularly in AA, we were relatively underpowered to detect any weak phenotypic associations.

Conclusion

In contrast to robust associations of genetically predicted Lp(a) with CHD, CVD, aortic valve disorders, mitral valve disorders, and congestive heart failure, in EA, such associations were weaker or absent in AA. However, associations with PAD and AAA were significant and similar in both EA and AA. Additional novel cardiovascular associations in EA included arterial thromboembolic disease, aneurysmal disease of vascular beds other than the aorta, atrial fibrillation, cardiac conduction disorders, blood pressure disorders, and atherosclerosis-related traits. Our findings have potentially important clinical implications for the use of Lp(a) levels in ASCVD risk stratification in AA and highlight the need to test the effect of Lp(a)-lowering on ASCVD events in AA.

Nonstandard Abbreviations and Acronyms

AA

African ancestry

AAA

abdominal aortic aneurysm

ACC

American College of Cardiology

AHA

American Heart Association

ASCVD

atherosclerotic cardiovascular disease

CHD

coronary heart disease

CI

confidence interval

CVD

cerebrovascular disease

EA

European ancestry

EHR

electronic health record

eMERGE

electronic MEdical Records and Genomics

IVW

inverse variance weighted

Lp(a)

Lipoproetin (a)

MR

Mendelian randomization

MVP

Million Veteran Program

OR

odds ratio

PAD

peripheral artery disease

PheWAS

phenome-wide association study

SD

standard deviation

UK

United Kingdom