Recent advances in understanding the spectrum of genetic determinants of lipoprotein(a) levels

Abstract

Purpose of review

Our understanding of the genetic regulation of lipoprotein(a) [Lp(a)] is hindered by the complex structure of the LPA gene, limited non-European datasets and its elusive cellular receptor(s). This review summarizes recent efforts and advances providing new insights on its genetic architecture, variability across ancestries and regulators beyond the LPA gene.

Recent findings

Impressive advances in DNA sequencing and bioinformatics now resolve LPA variants and kringle IV-type 2 copy number at scale. This provides new reference datasets and enables tools that unlock hidden variation also from already available sequencing datasets. In parallel, genetic studies broaden our understanding of the regulation of Lp(a) across ancestries and improve genetic risk scores. Finally, while recent studies implicate new mechanisms for Lp(a) uptake, upcoming genome-wide gene knockout screens allow comprehensive, agnostic scans for regulators and receptors. Puzzlingly, this still converges on the LDL receptor, whose exact role in Lp(a) uptake remains enigmatic.

Summary

Technological advances establish a foundation for more accurate genetic risk assessment across ancestries. These advances are enhancing our understanding of Lp(a) regulation and build a framework for future integrative genetic studies, which may shed new light on the evolution of the Lp(a) trait, adding important context for its physiological and clinical relevance.

INTRODUCTION

Lipoprotein(a) [Lp(a)] is as a frequent, independent and causal risk factor for multiple cardiovascular diseases (CVDs), with a clear dose-dependent relationship [1,2^▪▪,3^▪,4,5^▪▪] and a wide range of potential pathophysiological effects [6^▪]. On a per-particle basis, Lp(a) is approximately six times more atherogenic than LDL [7], and very high concentrations are associated with a three-fold increase in CVD risk [8^▪,9]. 

Box 1.

no caption available

However, even after 60 years of research, Lp(a) remains an enigmatic trait with many unanswered questions [10^▪,11^▪,12]. Individual Lp(a) concentrations range from less than 0.1 to more than 300 mg/dl (approximately <0.2 to 750 nmol/l), with median levels and distribution showing marked differences between and within ancestries [1,13,14]. Yet, more than 90% of Lp(a) variance is controlled by the LPA gene locus [15,16], making Lp(a) a highly oligogenic, if not effectively monogenic, trait. LPA encodes apolipoprotein(a) [apo(a)], the characteristic structural protein of the Lp(a) particle, and evolved through duplication and remodeling of the plasminogen gene (PLG) [17^▪▪]. It consists of 10 highly similar kringle IV domains (apo(a) domains KIV-1 to KIV-10), one kringle V (KV) and an inactive protease domain [17^▪▪] (Fig. 1). The KIV-2 domain exists in at least three subtypes (KIV-2A, KIV-2B and KIV-2C) distinguished by three exonic single nucleotide polymorphisms (SNPs) and is highly copy number-variable, presenting 1 to ≈40 copies per allele, spanning up to 70% of the LPA coding region and resulting in ≈40 protein isoforms [19].

FIGURE 1.

Domain structure and evolution of the lipoprotein(a) gene LPA from plasminogen. LPA evolved from duplication and remodeling of plasminogen. Each kringle domain consists of two short exons of mostly 160 and 184 bp, spaced by a mostly ≈4 kb long intron, and is linked to the next kringle domain by an ≈1.2 kb intron. All KIV coding sequences are highly similar, with homology extending also into the introns [18]. L: Leader sequence.

The KIV-2 copy number explains ≈30-60% of Lp(a) variance via an inverse correlation with endoplasmatic reticulum (ER) transit time [17^▪▪]. Typically, most circulating Lp(a) originates from the apo(a) isoform with less KIV-2 copies [20], but the relationship between isoform size and Lp(a) plasma concentration is complex [19]. Low-molecular weight (LMW) isoforms with 22 KIV units or less express five to 10 times higher median Lp(a) levels than larger ones [19,20], but isoforms of identical size may still express widely differing individual Lp(a) levels, ranging from ≈0 to more than 200 mg/dl [21]. This pronounced variance is genetically determined [15,21,22], but the causal variants remained elusive for a long time because of the structural complexity and intricate linkage disequilibrium patterns of the LPA gene [19]. Only recently advances in sequencing and bioinformatics uncovered numerous functional SNPs that explain much of the Lp(a) variance in same-sized isoforms but were previously hidden in the KIV-2 region[18,23–27].

Yet, deciphering the LPA gene and the Lp(a) trait remains a challenge. This review highlights recent technological advances in mapping variation in the LPA gene, understanding its genetic architecture beyond European populations, and uncovering the genetic determinants of Lp(a) concentrations.

NEW TECHNOLOGIES: LIPIDOLOGY MEETS GENOMICS

Conventional short-read sequencing (NGS) is poorly suited for resolving genome regions like LPA (Supplementary Table 1) and the complex variation patterns in the KIV-2 region also escape current genomic data standards, excluding LPA from benchmark datasets and hindering method development [28,29]. These challenges have sparked considerable interest in LPA within the genomics community, and – largely unnoticed by the lipidology field – LPA has evolved from a cardiovascular curiosity into a locus at the forefront of genomics research [30^▪,31–35] (Fig. 2).

FIGURE 2.

Recent advances in genomics that benefit genetic research on lipoprotein(a). The targeted amplification of all KIV-2 units using universal primers, followed by deep short-read sequencing and mutation detection in read subfractions (termed “KIV-2 batch sequencing” [18]) has long been the cornerstone of LPA mutation analysis (represented top left). This review describes multiple new technological advances that benefit genetic Lipoprotein(a) research. The figure uses icons from NIAID NIH BioArt, Bioicons.com and svgrepo.com. See the Acknowledgements section for attributions.

Long-read sequencing (LRS) technologies generate reads spanning tens or even hundreds of kilobases and have transformed genomics [36^▪], enabling the creation of the first complete human genome reference sequence (T2T-CHM13) [37]. This new reference sequence contains an LPA gene with 23 KIV-2 units, compared with six in the previous reference hg38, and indicates that the KIV-2B units, which complicate KIV-2 variant calling in ≈80% of the individuals [38^▪], cluster at the end of the KIV-2 array rather than being interspersed as in hg38 (confirming an earlier preprint [39]). Additional insights into LPA structure and variability are promised also by hundreds of forthcoming high-quality long-read genomes [40,41^▪,42,43^▪]. For example, Gustafson et al. [41^▪] reported that 93% of their LRS genome assemblies achieved full contiguity in LPA, though without providing orthogonal validation or further details. Similarly, a recent preprint describes a nearly error-free diploid benchmark genome with a likely gap-free, fully resolved assembly of both LPA alleles [44^▪]. The two alleles differ by 55 kb, that is 10 KIV-2 units, and are accessible via a dedicated UCSC Genome browser data hub. These new reference datasets enable new analysis tools for available data, such as the variant calling tool Locityper, which determines KIV-2 SNP haplotypes from any sequencing technology by selecting the haplotype pair in the reference data that best explains the observed sequencing reads [45^▪]. As LRS reference datasets grow in size and population diversity, such tools are poised to augment available short-read datasets. Of note, even LRS reads rarely span the full KIV-2 region at high coverage, particularly in large alleles, so full reconstruction will still require computationally intensive de-novo assembly.

For more targeted analyses, we have recently introduced a scalable method for direct KIV-2 sequencing and haplotyping, which combines long-read nanopore sequencing with molecular barcode-based, single molecule-level error correction (UMI-ONT-Seq) [46^▪]. This produced highly accurate consensus sequences for each KIV-2 unit, which confirmed, for example, that the two major European Lp(a)-lowering variants KIV-2 4925G>A [23] and KIV-2 4733G>A [24] occur on different haplotypes [46^▪]. By counting unique haplotypes and adjusting for coverage, UMI-ONT-Seq also estimates total KIV-2 copy number (CN) with accuracy comparable to digital PCR [46^▪], which already outperforms KIV-2 CN quantification by qPCR [47]. While assessing only the total KIV-2 copy number has important limitations [48,49], large biobanks have used sequencing-based KIV-2 copy number quantification as an alternative for apo(a) sizing by Western blot, offsetting its imprecision with statistical power [25,50,51]. Generally, sequencing-based KIV-2 copy number quantification explains similar Lp(a) variance as KIV-2 copy number assessment by qPCR (R² ≈ 20–30%) but showing better parent-offspring concordance [50].

EXISTING DATA, NEW TRICKS: UNLOCKING LPA FROM SHORT-READ SEQUENCING DATA

Since most large biobanks provide only short-read data, various groups have sought to improve the utility of this data for LPA genetics by developing new analysis methods and reanalyzing available NGS data [38^▪,52^▪,53^▪▪].

Behera et al. [53^▪▪] recently identified two intronic KIV-2 SNPs that, when present, occur in every repeat of the gene allele. In heterozygous individuals (40–52%, depending on ancestry) and after proper normalization against diploid genome regions, these SNPs allow determining the KIV-2 copy number of each allele with notable accuracy, as the sequencing coverage on the two SNP alleles reflects the number of KIV-2 units on each allele (Fig. 3) [53^▪▪]. The two SNPs occur in all five major ancestry groups of the 1000 Genomes project, suggesting that they may trace back to very ancestral LPA alleles and that recombination within KIV-2 is rare, which is consistent with the long haplotypes observed by Lanktree et al. [55]. Behera et al. [53^▪▪] successfully reproduced known ancestry-specific allele size distributions, replicated known SNP-isoform associations (e.g. for rs41272110 [27], rs3798220 [56], and rs10455872 [56]) and the allele size estimates in 60 trios correlated strongly (R² > 0.997) with Bionano Optical Mapping data (an optical copy number counting method), supporting validity of the method. This approach has been recently applied also to 8,351 short-read genomes from the GENESIS-HD study (Betschart et al. [52^▪]), again replicating known SNP-isoform associations and complex linkage patterns. Unfortunately, the algorithm has been integrated into Illumina's commercial platforms (BaseSpace, DRAGEN) and submitted as intellectual property [57], limiting its accessibility. Of note, Betschart et al. [52^▪] and Molitor et al. [54^▪] found that other sequencing coverage-based KIV-2 copy number callers perform equally well at least for total KIV 2 copy number determination.

FIGURE 3.

Determination of the KIV-2 copy number (CN) and determination of the allele sizes by sequencing. A. In conventional KIV-2 CN determination, the sequencing coverage on the KIV-2 is normalized to that of one or more unique genome regions. B. When present, the KIV-2 CN phasing SNPs [53^▪▪] occur in every KIV-2 unit of the corresponding gene allele. In heterozygous individuals (≈40–50% of the population), this allows assigning reads to the two alleles. Normalizing read counts for each SNP allele against a unique sequence provides the KIV-2 CN of each allele. The regions used for normalization can be the nonrepetitive part of LPA, as in KILDA [54^▪], or multiple genome regions, as in DRAGEN [53^▪▪]. Note that the figure shows only a naïve calculation model for illustration purposes, while real-world data requires more advanced statistics and normalization.

Accurate variant detection in the KIV-2 from whole-genome or whole-exome data, as provided by large biobanks, is complicated by the high homology between the kringle domains, which causes misaligned reads and spurious variant calls [38^▪]. Our group has observed recently that the best read-mapping strategy strongly depends on presence of KIV-2B units in the analyzed individual genome (see also explanations in Supplementary Table 1) and identified a SNP outside the KIV-2 that predicts KIV-2B presence [38^▪]. Building on this observation, we developed a scalable open-source tool that selectively remaps short-read KIV-2 reads data, choosing the best alignment strategy dynamically based on the individual genotype. This doubled variant calling accuracy in the KIV-2 region and detected more than 700 high confidence KIV-2 mutations in ≈199 000 UK Biobank samples [38^▪]. Unfortunately, this KIV-2B tag SNP approach is currently applicable only to Europeans, highlighting the need for more data from other ancestries.

GENETIC EFFECTS ACROSS ANCESTRIES AND THEIR IMPACT ON GENETIC RISKS SCORES

Although elevated Lp(a) confers similar CVD risk across ancestries [58], median levels differ widely from ≈6 mg/dl in East Asians and Finns to ≈40–50 mg/dl in some African populations, reflecting substantial differences in trait distribution [59,60^▪,61^▪]. South Asians exhibit the second-highest median Lp(a) levels after Africans and about one in three individuals worldwide with Lp(a) more than 50 mg/dl is of South Asian descent [61^▪]. The reason for elevated Lp(a) in South Asians remains unclear, as no higher frequency of LMW isoforms has been consistently observed, suggesting population-specific regulatory effects [61^▪], similar to the higher frequency of the regulatory variant rs1800769 in Africans [25,62]. In contrast, a higher frequency of large isoforms explains the lower levels in Chinese [63^▪▪,64].

SNP patterns and SNP-isoform associations differ markedly between ancestries [19]. For instance, rs3798220 was found to not tag short apo(a) isoforms or high Lp(a) in South Asians [65] or Chinese [53^▪▪,63^▪▪,66], and two small studies found also no association of rs3798220 with coronary artery disease in Iraqis and Iranians (recently reviewed in [60^▪]). Conversely, rs10455872 is associated with CVD also in Middle East [60^▪]. However, it is absent in Chinese [63^▪▪,66] and very rare in South Asians [55].

A recent GWAS in more than 18 000 Chinese with follow-up in the UK Biobank provides very interesting insights into the genetics of Lp(a) in East Asia [63^▪▪]. The two Chinese lead variants (rs192717255, rs73596816) explained 11.3 and 10.4% of Lp(a) variance [63^▪▪], which is substantially less than the variance explained in Europeans by rs10455872 alone (24–29% [67,68]) or rs10455872 and rs3798220 together (36%[56]). In line with this, a 28-SNP score derived from Chinese explained only 10% variance in UK Biobank (would be ≈60% for a comparable European SNP score [69]) and adding Chinese SNPs to European SNPs scores, or vice versa, did not improve explained variance in the other ancestry. This highlights considerable differences in the genetic architecture of the Lp(a) trait between Chinese and Europeans, possibly also hinting towards a higher allelic heterogeneity in Chinese. However, the authors also replicated the association of APOE with Lp(a) observed previously in Europeans [67,68,70], confirming still shared biology despite divergent genetic patterns.

These strong ancestry-specific components limit the transferability of SNP-based genetic risk scores (GRS) for Lp(a) across populations [71] (in contrast to the KIV-2 copy number effect [51]), as noted by Privé et al. [72] who observed that the best Lp(a) GRS explains 66% in British, but only 40% in Iranians, 15% in Caribbeans, and 0% in Nigerians. As hardly any high-impact LPA variants for non-European groups are known, more diverse sequencing efforts are urgently required.

PROGRESSES IN IDENTIFYING REGULATORS OUTSIDE OF THE LPA GENE

The role of regulators outside LPA and especially the identity of the Lp(a) receptor(s) remain largely unknown, with no single major receptor identified yet [17^▪▪,73,74]. The LDL receptor (LDLR) is the leading candidate and has been shown to bind Lp(a) in vitro, but in-vivo evidence is inconsistent [17^▪▪]. Some authors proposed that LDLR may contribute noticeably only under conditions of supraphysiological receptor expression and particularly low LDL-C, as achieved by statin therapy plus PCSK9 inhibition [17^▪▪,75].

Recently, Lp(a) internalization via macropinocytosis mediated by interaction of the apo(a) moiety with plasminogen receptors has been proposed [76], but in a recent follow-up work the macropinocytosis inhibitor imipramine did actually stimulate Lp(a) internalization, which was found to be due to upregulation of the plasminogen receptor PlgRKT and subsequent increased anchoring of Lp(a) to the cell membrane for S100A10- and Annexin A2-induced macropinocytosis [77^▪]. Others noted, however, that hepatocytes are not known to rely on macropinocytosis and suggested that the true Lp(a) receptor remains unidentified [17^▪▪,78^▪▪]. In an effort to scale up the search for Lp(a) receptors in an agnostic way, Khan et al. [79^▪▪] recently performed a genome-wide CRISPR knockout screening in HuH7 cells interrogating the effect of more than 19 000 genes on Lp(a) uptake. Although such an approach is inherently limited by the specific gene expression pattern of the used cell line, it is still noteworthy that it retrieved only LDLR and MYLIP (a negative regulator of LDLR) as significant positive, respectively negative determinants of Lp(a) uptake. Also relaxing the significance threshold brought up only further LDLR regulators, which is in line with a large GWAS that, among all proposed receptor candidates, found only an association signal in LDLR - but its effect on Lp(a) variance was only ≈1% that of rs14055872 [67]. Surely, the Lp(a) metabolism remains a conundrum.

Lambert and Boffa [78^▪▪] recently noted that such endeavors assume that a discrete Lp(a) receptor analogous to the LDL–LDLR system exist. Given the recent evolutionary origin of Lp(a), no specific receptor may have co-evolved and Lp(a) uptake may involve multiple receptors with incidental affinity [78^▪▪]. This would be consistent with the somewhat inconclusive biochemical data implicating many candidates and mechanisms to different extents. This would make identification of an “Lp(a) receptor” difficult and might even question its suitability as a drug target.

CONCLUSION

The very reason for the existence of Lp(a) remains a mystery and the mutation patterns of the LPA gene have been even compared to those of a transcribed pseudogene [17^▪▪,80], raising doubts about whether it really serves a physiological role. The forthcoming technological advances are poised to provide unprecedented insights into the variation patterns in LPA and the genetic architecture of Lp(a). Putting such complete genetic data into a formal evolutionary framework may finally help to answer whether Lp(a) is truly an adaptive trait or just an evolutionary byproduct, with important implications across the field. This may bring us one step closer to resolving one of the many riddles of that Lp(a) poses.

Acknowledgements

Given the focus on developments from the past 18 months and word count limitations, it was necessary to prioritize topics. The author regrets omitting numerous valuable contributions to the field and encourages readers to explore other excellent review articles offering complementary perspectives.

Figure 2 uses icons from Bioicons, NIAID NIH BIOART and svgrepo.com. Attributions Bioicons: Icon nanopore_sequencing icon by DBCLS ( https://togotv.dbcls.jp/en/pics.html ) is licensed under CC-BY 4.0 ( http://creativecommons.org/licenses/by/4.0/ ). Icon img_sequencer_long_read05 by PacBio ( www.pacb.com ) is licensed under CC0 ( https://creativecommons.org/publicdomain/zero/1.0/ ).

Attributions NIAID NIH BIOART: Icon Gene Mutation: NIAID Visual & Medical Arts. (10/7/2024), bioart.niaid.nih.gov/bioart/170. Icon Next Gen Sequencer: NIAID Visual & Medical Arts. (10/7/2024), bioart.niaid.nih.gov/bioart/386. Attribution Svgrepo.com: World map icon under MIT license,https://www.svgrepo.com/svg/402581/world-map.

Financial support and sponsorship

This research was funded in whole or in part by the Austrian Science Fund (FWF) 10.55776/PAT5152823 to S.C.

Conflicts of interest

The author has received honoraria from Novartis AG (Basel) and Silence Therapeutics, PLC (London) for consulting activities related to Lp(a) genetics.

Supplementary Material

colip-37-65-s001.docx ^{(22.2KB, docx)}

REFERENCES AND RECOMMENDED READING

Papers of particular interest, published within the annual period of review, have been highlighted as:

• ▪ of special interest

• ▪▪ of outstanding interest