Genome-wide association study highlights APOH as a novel locus for lipoprotein(a) levels

Abstract

Objective:

Lipoprotein(a) (Lp(a]) is an independent risk factor for cardiovascular diseases and plasma levels are primarily determined by variation at the LPA locus. We performed a genome-wide association study (GWAS) in the UK Biobank to determine whether additional loci influence Lp(a) levels.

Approach and Results:

We included 293,274 White British individuals in the discovery analysis. Approximately 93,095,623 variants were tested for association with natural log-transformed Lp(a) levels using linear regression models adjusted for age, sex, genotype batch, and 20 principal components of genetic ancestry. After quality control, 131 independent variants were associated at genome-wide significance (P ≤ 5 × 10⁻⁸). In addition to validating previous associations at LPA, APOE, and CETP, we identified a novel variant at the APOH locus, encoding beta2-glycoprotein I (β2GPI). The APOH variant rs8178824 was associated with increased Lp(a) levels (β [95% CI] (ln nmol/L), 0.064 [0.047, 0.081]; P = 2.8 × 10⁻¹³) and demonstrated a stronger effect after adjustment for variation at the LPA locus (β [95% CI] (ln nmol/L), 0.089 [0.076, 0.10]; P = 3.8 × 10⁻⁴²). This association was replicated in a meta-analysis of 5,465 European-ancestry individuals from the Framingham Offspring Study and Multi-Ethnic Study of Atherosclerosis (β [95% CI] (ln mg/dL), 0.16 [0.044, 0.28]; P = 0.0071).

Conclusions:

In a large-scale GWAS of Lp(a) levels, we identified APOH as a novel locus for Lp(a) in individuals of European ancestry. Additional studies are needed to determine the precise role of β2GPI in influencing Lp(a) levels as well as its potential as a therapeutic target.

Graphical Abstract

Introduction

Lipoprotein(a) (Lp[a]) is an independent risk factor for both coronary artery disease and aortic valve stenosis^1,2. Lp(a) consists of a low-density lipoprotein (LDL)-like particle covalently bound to the glycoprotein apolipoprotein(a) (apo[a]). Levels of Lp(a) are primarily controlled by the size of the apo(a) protein, with smaller apo(a) isoforms leading to higher concentrations of plasma Lp(a). This size polymorphism is caused by a variable number of kringle IV type 2 repeats in the LPA gene. Together with other sequence variation in LPA, these kringle IV type 2 repeats are estimated to explain more than 90% of variability in Lp(a) concentration in individuals of European ancestry^3,4.

Lp(a) plasma concentrations vary widely between populations, with African-ancestry individuals having 2–3 fold higher levels than European-ancestry individuals^5,6. The distribution of Lp(a) is right-skewed across populations, with most individuals having very low levels⁷. While the precise physiological functions of Lp(a) are still unclear, there is evidence that it has proatherogenic and proinflammatory properties⁸. In pathophysiological studies, Lp(a) or apo(a) have been detected in both the lesioned intima of human arteries^9–12 and in aortic valve lesions^13,14.

Several genome-wide association studies (GWAS) of Lp(a) have been performed, highlighting LPA as the major genetic determinant of Lp(a) levels^4,7,15–20. However, these studies have been limited by small sample sizes (N<15,000), sparse genotyping arrays, or a focus on founder populations. In this study, we aimed to identify novel loci for Lp(a) by performing a GWAS in nearly 300,000 individuals from the UK Biobank. The findings could provide further insights into the regulation and clearance of Lp(a) particles and highlight novel targets for Lp(a)-lowering therapies.

Materials and Methods

The data that were analyzed for this study are available from the UK Biobank or from dbGAP in the case of the Multi-Ethnic Study of Atherosclerosis and the Framingham Offspring Study. For details, please see the Major Resources Table in the Supplemental Material.

Study Population

The UK Biobank study recruited over 500,000 individuals aged 40–69 years from 22 recruitment centers across the United Kingdom between 2006 and 2010. Participants provided blood samples for DNA extraction and biomarker analysis and completed a series of questionnaires, as previously described²¹. UK Biobank received ethical approval from the North West Multi-Centre Research Ethics Committee and all participants provided written informed consent. All relevant internal review boards approved this study. Only White British individuals were included in the discovery analysis to reduce confounding by ancestry, where White British ancestry was determined using a combination of self-reported ethnicity and results from a principal component analysis²¹.

Phenotyping

Lp(a) (nmol/L) was measured using an immunoturbidimetric analysis on a Randox AU5800. Measurements were taken at the initial assessment visit (2006–2010) or the first repeat assessment visit (2012–2013). Measurements that returned an error from the analyzer or were outside of the reportable range (3.80–189 nmol/L) were excluded (n=91,426). Additional phenotypes are described in the Supplemental Material.

Genotyping

Genotyping was performed using the Affymetrix UK BiLEVE Axiom array on an initial 50,000 participants, while the remaining 450,000 participants were genotyped using the Affymetrix UK Biobank Axiom array. Quality control and imputation were performed centrally by the UK Biobank, as described previously²¹. Briefly, genetic markers were tested for batch effects, plate effects, departure from Hardy-Weinberg equilibrium, sex effects, array effects, and discordance across control replicates; markers that failed at least one test in a given batch had their genotype calls set to missing. Imputation was performed using only markers present on both the UK BiLEVE UK Biobank Axiom arrays, and markers that failed quality control in more than one batch, had a >5% missing rate, or had a minor allele frequency <0.0001 were removed. Samples with unusually high heterozygosity or >5% missing rate were excluded from analysis.

GWAS for Lp(a)

Associations of 93,095,623 genetic variants with natural log-transformed Lp(a) were tested in linear regression models assuming additive genetic effects in PLINK version 2.0^22,23. All models were adjusted for age, sex, genotype batch, and 20 principal components of ancestry. Variants with minor allele frequency < 0.01 or imputation quality score < 0.3 were removed (n=83,266,620). Clumping was performed on variants reaching genome-wide significance (P ≤ 5 × 10⁻⁸) with PLINK version 1.9^22,24; index variants were chosen greedily starting with the lowest p-value, and variants less than 1Mb away from an index variant with an r² > 0.01 were assigned to that index variant’s clump. The most significant independent variants in each locus (lead variants) were queried for previous associations using PhenoScanner (accessed 02/04/2020)^25,26. In a conditional analysis, all lead variants were tested for association with Lp(a) after additional adjustment for assessment center.

Conditional analysis of the LPA locus

A weighted LPA-region genetic risk score was created using all independent genome-wide significant variants in the LPA region. Lead variants outside the LPA region were tested for association with Lp(a) after adjusting for age, sex, genotype batch, 20 principal components, and the LPA-region genetic risk score.

Replication in Other Populations

The lead variant in each locus was tested for association with Lp(a) in other ethnic groups from the UK Biobank containing at least 1,000 unrelated individuals (South Asians, Black Africans, and Black Caribbeans). Lp(a) (nmol/L) was natural log-transformed and models were adjusted for age, sex, genotype batch, and 20 principal components. Lead variants were also assessed in a fixed-effects meta-analysis of self-reported White individuals from the Multi-Ethnic Study of Atherosclerosis (MESA) and the Framingham Offspring Study. The variant rs1065853 was not available in these cohorts so rs7412 was used as a proxy (linkage disequilbrium r²=0.99 in the UK Biobank). Cohort descriptions and model details are provided in the Supplemental Material.

Statistical Analyses

Two-sided p-values ≤ 5 × 10⁻⁸ were considered significant in the GWAS and two-sided p-values ≤ 0.05 were considered significant in all other analyses. Proportion of variance explained was calculated for independent significant variants in the LPA region independently, when modeled together, and when combined as a weighted genetic risk score.

Results

A total of 293,274 individuals with Lp(a) measurements were included in the study. Demographic characteristics of the UK Biobank Lp(a) subset are presented in Table I in the Supplemental Material, with individuals stratified by Lp(a) levels. Individuals with Lp(a) levels greater than or equal to the median were more likely to have coronary artery disease and aortic stenosis but were less likely to be female or diabetic (all P<0.001). These individuals also had higher levels of LDL cholesterol (corrected and uncorrected for Lp[a]) and high-density lipoprotein cholesterol (all P<0.001).

Following quality control, 9,829,003 variants with a minor allele frequency > 0.01 remained for further analysis. The association of these variants with Lp(a) showed no substantial inflation in the test statistics (genomic inflation factor [λ] = 1.03, Figure 1). After clumping, we identified 131 independent variants associated with Lp(a) at the genome-wide significance level of P ≤ 5 × 10⁻⁸. The most significant variant, rs10455872 in LPA, explained 29% of variance in Lp(a) levels (Table II in Supplemental Material). There were 126 other independent variants in the LPA region, explaining an additional 20% of variance.

Figure 1:

Log-transformed p-values for the association of 9,829,003 variants with natural log-transformed Lp(a). The plot has been cropped for better resolution, truncating only the signals at chromosome 6 (LPA region) and chromosome 19 (APOE region).

Outside the LPA region, we identified variants in four loci (Table 1). Variant rs1065853 on chromosome 19, located downstream of APOE, was associated with decreased Lp(a) levels (β [95% CI] (ln nmol/L), −0.11 [−0.12, −0.10]; P = 2.8 × 10⁻⁹⁶), as was variant rs247617 on chromosome 16, located upstream of CETP (β [95% CI] (ln nmol/L), −0.023 [−0.030, −0.017]; P = 1.0 × 10⁻¹³). On chromosome 17, rs8178824 in APOH was associated with an increase in Lp(a) (β [95% CI] (ln nmol/L), 0.064 [0.047, 0.081]; P = 2.8 × 10⁻¹³). Finally, variant rs826128 on chromosome 2, located in the long non-coding RNA AC093639.1, was associated with decreased Lp(a) levels (β [95% CI] (ln nmol/L), −0.039 [−0.053, −0.026]; P = 5.9 × 10⁻⁹). The median Lp(a) level by genotype is shown for all lead variants in Table III of the Supplemental Material. Additional adjustment for assessment center did not materially change the results (Table IV in Supplemental Material).

Table 1:

Association of lead variants with Lp(a).

Variant	CHR	Position	Genes in Locus	Minor Allele (Freq)	β [95% CI] ((ln nmol/L)	P	Variants in Locus*
rs10455872	6	161010118	LPA, ZDHHC14, SNX9†	G (0.076)	1.7 [1.7,1.7]	6.2 × 10−22,136	127
rs1065853	19	45413233	APOE	T (0.080)	−0.11 [−0.12, −0.10]	2.8 × 10−96	1
rs247617	16	56990716	CETP	A (0.32)	−0.023 [−0.030, −0.017]	1.0 × 10−13	1
rs8178824	17	64224775	APOH	T (0.030)	0.064 [0.047,0.081]	2.8 × 10−13	1
rs826128	2	184797074	AC093639.1	A (0.054)	−0.039 [−0.053, −0.026]	5.9 × 10−9	1

^*Number of independent (r² < 0.01), genome-wide significant variants.

^†Additional genes (+/− 3.5 MB from LPA): TULP4, SYTL3, EZR, RP1-155D22.1, RSPH3, RP1-111C20.4, FNDC1, RP11-125D12.1, RP11-125D12.2, RP3-393E18.1, SOD2, ACAT2, PNLDC1, RP1-249F5.3, IGF2R, SLC22A1, SLC22A2, SLC22A3, LPAL2, PLG, RP11-235G24.1, RP11-235G24.2, RP11-235G24.3, RP3-428L16.1, MAP3K4, AGPAT4, PARK2.

The lead variant in each locus was evaluated for association with Lp(a) in the following populations from the UK Biobank: South Asian (n=6,101), Black African (n=2,510), and Black Caribbean (n=3,207). Consistent with results in White British individuals, rs10455872 was associated with increased Lp(a) in South Asians (β [95% CI] (ln nmol/L), 1.01 [0.82,1.19]; P = 8.5 × 10⁻²⁶) and Black Caribbeans (β [95% CI] (ln nmol/L), 0.81 [0.59, 1.03]; P = 1.4 × 10⁻¹²). This variant showed no association in Black Africans, likely owing to its low frequency in this population (minor allele frequency = 6.3 × 10⁻⁴). The variant rs1065853 near APOE was significantly associated with Lp(a) across ethnic groups, with Black African individuals demonstrating the largest decrease in levels per minor allele (T) (β [95% CI] (ln nmol/L), −0.28 [−0.34, −0.21]; P = 9.4 × 10⁻¹⁶). The lead variants in CETP, APOH, and AC093639.1 were not significantly associated in South Asians, Black Africans, or Black Caribbeans (Table V in Supplemental Material).

The lead variants were also tested for association with Lp(a) in a meta-analysis of 5,465 European-ancestry individuals from the Multi-Ethnic Study of Atherosclerosis and the Framingham Offspring Study. Both rs10455872 in LPA and rs8178824 in APOH were significantly associated with increased Lp(a) levels in the meta-analysis (rs10455872: β [95% CI] (ln mg/dL), 2.1 [2.0, 2.2]; P = 6.0 × 10⁻⁵³⁴, and rs8178824: β [95% CI] (ln mg/dL), 0.16 [0.044, 0.28]; P = 0.0071). The lead variants in APOE, CETP, and AC093639.1 showed no significant effects (Table VI in Supplemental Material).

The weighted LPA-region genetic risk score contained 127 variants and explained 44% of the variance in Lp(a) levels. After adjusting for this score, variants in APOE, CETP, and APOH showed stronger effects on Lp(a) (Table VII in Supplemental Material). Conversely, the variant rs826128 on chromosome 2 showed a decreased effect and no longer reached genome-wide significance (β [95% CI] (ln nmol/L), −0.016 [−0.026, −0.0058]; P = 0.0019).

Discussion

We performed a GWAS for plasma Lp(a) levels in 293,274 White British individuals from the UK Biobank. We confirmed the association of loci in the LPA region with Lp(a) levels, as well as APOE and CETP. In addition, we identified APOH as a novel risk locus and replicated this association in a meta-analysis of two independent cohorts.

As expected, our association study identified many significant variants in the LPA gene and the surrounding region. Despite imposing a stringent r² threshold (≤ 0.01), 127 variants were independently associated with Lp(a). Together, the top 4 variants explained 40% of the variance in Lp(a) levels, while the remaining 123 explained an additional 9%. Consistent with previous work^1,20,27, the variant rs10455872 was the most strongly associated with Lp(a), explaining 29% of variation in Lp(a) levels alone.

Outside of the LPA region, we identified variants at four loci, two of which have been previously associated with Lp(a) or Lp(a)-cholesterol: APOE and CETP. Our lead variant near the APOE locus, rs1065853, is in high linkage disequilibrium with the apoE2-defining variant rs7412 (r²=0.99), which has been previously associated with decreased Lp(a)^4,19,28,29. Relative to the apoE3 and apoE4 isoforms, apoE2 has a lower affinity for LDL receptors and LDL receptor-related protein I, potentially leading to less competition for Lp(a) binding and greater clearance of Lp(a)³⁰. Upstream of the CETP locus, the variant rs247617 was also associated with decreased Lp(a) levels. This finding is consistent with clinical studies showing that inhibition of cholesteryl ester transfer protein, the product of CETP, decreases Lp(a) levels^31–33. This variant is also in high linkage disequilibrium with rs247616 (r²=0.99), which has been previously associated with Lp(a)-cholesterol²⁸.

Apart from APOE and CETP, no other loci outside the LPA region have been associated with Lp(a) levels at the genome-wide significance level. Here, we identify rs8178824 in APOH as significantly associated with increased Lp(a) and provide independent replication. Relative to rs10455872, the effect size of rs8178824 is small, with individuals homozygous for the minor allele having a median Lp(a) level only 4.7 nmol/L higher than individuals with two major alleles. However, this effect is comparable to those seen for APOE and CETP, where the difference in homozygous genotype classes is 10.5 nmol/L and 1.9 nmol/L, respectively. As demonstrated previously with treatments targeting CETP, which produced reductions in Lp(a) of more than 30%^31–33, therapeutic targeting of the APOH locus could have a more substantial effect on Lp(a) than the effects of a single variant.

The APOH locus encodes beta2-glycoprotein I (β2GPI), a single chain plasma protein with a high affinity for negatively charged surfaces³⁴. Recently, β2GPI has been shown to interact with proprotein convertase subtilisin/kexin-9 (PCSK9)³⁵, whose inhibition leads to reductions in LDL cholesterol³⁶. This evidence is supported by previous studies demonstrating that genetic variation in APOH is associated with decreased levels of LDL cholesterol^37–40 and peak particle diameter⁴¹. In vitro, β2GPI has also been shown to bind to Lp(a), both through the phospholipids of the LDL component and through the kringle IV-domain of apo(a)⁴². Given that apo(a) is a major site for the accumulation of negatively charged oxidized phospholipids⁴³, the interaction of β2GPI and Lp(a) may be primarily mediated through binding of β2GPI to these phospholipids.

The variant we identified in APOH is in perfect linkage disequilibrium with rs1801689 (r²=1.0). Interestingly, the amino acid change caused by rs1801689 (Cys325Gly, also known as Cys306Gly) has been shown to disrupt the ability of β2GPI to bind to phospholipids⁴⁴. This change may reduce β2GPI’s affinity for oxidized phospholipids on apo(a), thereby allowing more free molecules of β2GPI and Lp(a) to circulate in the plasma. Indeed, rs1801689 has also been previously associated with increased levels of plasma β2GPI^45,46. The potential role of β2GPI in lipid metabolism is further supported by the observation that it can accelerate triglyceride clearance in rats⁴⁷. Future studies should investigate whether the presence of β2GPI similarly affects Lp(a) clearance or affects its pathogenicity through other mechanisms.

This study has several strengths and limitations. The UK Biobank discovery sample was larger than any previous Lp(a) GWAS, and thus had more power to detect novel associations. In addition, we were able to replicate our novel finding in APOH in a meta-analysis of two other European-ancestry cohorts. However, the APOH variant showed no association with Lp(a) in other ethnicities from the UK Biobank. The lack of replication observed for this variant and others may reflect reduced power due to smaller sample sizes, different allele frequencies, or different patterns of linkage disequilibrium⁴⁸; nonetheless, additional studies in larger non-European cohorts are warranted. Another limitation of our study is the high percentage of individuals missing Lp(a) measurements (>20%) in the UK Biobank due to the assay’s limited reportable range (3.80–189 nmol/L); our results may therefore not apply to individuals with very high levels of Lp(a).

In summary, we have performed a large-scale GWAS of Lp(a) levels, validating previous loci and identifying APOH as a novel locus. Our findings provide further insight into the regulation of Lp(a) levels and highlight β2GPI as a potential therapeutic target in individuals with elevated Lp(a).

Highlights.

• We have performed a large-scale genome-wide association study of Lp(a) levels.

• We confirmed the association of variants in LPA, APOE, and CETP with Lp(a).

• We identified APOH as a novel risk locus for Lp(a), highlighting β2GPI as a determinant of Lp(a) levels and a potential therapeutic target.

Abbreviations

Lp(a)

Lipoprotein(a)

GWAS

Genome-wide association study

Apo(a)

Apolipoprotein(a)

LDL

Low-density lipoprotein

β2GPI

Beta2-glycoprotein I

MESA

Multi-Ethnic Study of Atherosclerosis