Lipoprotein(a) beyond the kringle IV repeat polymorphism: The complexity of genetic variation in the LPA gene

1Lipoprotein(a) [Lp(a)] has proatherogenic, proinflammatory and possibly prothrombotic properties and represents a major cardiovascular risk factor in the general population. Atherogenic Lp(a) concentrations affect up to 2 billion people worldwide (>30 or >50 mg/dL). Several details of the pathophysiology of Lp(a) are not fully clarified but converging data proposes that the proinflammatory oxidized phospholipids (OxPLs) play a key role in mediating several detrimental effects of Lp(a) as discussed in another review of this series.

2The Lp(a) particle originates from the liver, is found only in old world monkeys and apes and consists of an apolipoprotein(a) [apo (a)] molecule that is bound to the apoB100 moiety of an LDL-like lipoprotein. LPA, the gene encoding apo(a), evolved from a duplication of the plasminogen (PLG) gene about 33–40 million years ago. While plasminogen kringle domains I, II and III (KI, KII, KIII) were lost, the KIV domain expanded and diverged to 10 subtypes (KIV-1 to KIV-10). Kringle V and the protease domain were retained but the protease domain was inactivated by mutations (Fig. 1). The KIV-2 domain is encoded by a copy number variation (CNV) that creates >30 gene alleles, respectively protein isoforms (≈200–800 kDa) in the population. By convention, apo(a) isoform designations report the total number of KIV domains. The KIV-2 number can be deduced by subtracting nine kringles.

3Lp(a) is one of the most heritable human quantitative traits with up to ≈90% heritability. Individual Lp(a) concentrations are relatively stable throughout life-time (albeit some currently underappreciated temporal variability exits). Lp(a) distribution is highly right-skewed in Whites with medians of ≈10–12 mg/dL. The inter-individual Lp(a) concentrations extend for about three orders of magnitude (<0.1 mg/dL to >300 mg/dL) (Fig. 2A) and show considerable cross-ancestry variance (reviewed by Virani and colleagues in this review series). In Black populations, the distribution is more Gaussian and median concentrations are markedly higher. It has been suggested that genetic variability might be responsible for some of the cross-ancestry differences in Lp(a) concentrations.

4While >160 genes are necessary to explain 50–70% of the heritability of other lipoproteins, the LPA gene locus alone explains up to 90% of Lp(a) variance. About 40–70% of Lp(a) variance is explained by the apo(a) isoform size, which shows an inverse relationship with the Lp(a) concentration (Fig. 2A). This is most probably due to a more efficient maturation of smaller apo(a) proteins in the endoplasmatic reticulum. Low molecular weight (LMW) isoforms (10–22 KIV) are associated with ≈4–5 times higher median Lp(a) concentrations (≈40–50 mg/dL) than high molecular weight (HMW; >22 KIV) isoforms (<10 mg/dL) and the concentrations decrease rather suddenly at 23 KIV (Fig. 2B). These considerably higher median Lp(a) concentrations in LMW individuals have been well established since a long time. However, the exact relationship between apo(a) isoform and Lp(a) concentrations is complex, not linear and modified also by several functional single nucleotide polymorphisms (SNPs) (discussed in depth in the following sections). These lead to a large (yet in the field still underappreciated) variance in the individual Lp(a) concentrations within each isoform group (Fig. 2C). Fig. 2D provides an example of such a strongly Lp(a)-modifying SNP (KIV-2 4925G>A). Because of the differing maturation efficiencies and the modifying SNPs, the two LPA alleles of heterozygous individuals are not necessarily equally secreted to plasma. The relative contribution can be visualized by Western blotting and used to apportion the plasma Lp(a) concentration to isoform-specific Lp(a) concentrations, nuancing the Lp(a) trait further.

5Understanding the genetics of Lp(a) might be a key to many mysteries of Lp(a) (Key point box 2). Genetic variants associated with certain Lp(a) concentration ranges were already in the past very helpful to support causality between Lp(a) and outcomes. Lp(a) was the first use case for Mendelian randomization studies in the 1990s, long before this term was coined (see review in this series). Nevertheless, the causality of Lp(a) has been debated for a long time until numerous genetic studies underscored the causality of Lp(a) concentrations using genetic variants strongly associated with high Lp(a) concentrations and subsequent cardiovascular disease. Several studies showed also that variants associated with low Lp(a) exert a protective effect on cardiovascular disease. On the other hand, it still takes some efforts to find the right genetic instrument to investigate a causal association between extremely low Lp(a) concentrations and diabetes mellitus. In the latter case, the use of the non-carrier status of rs10455872 is an insufficient genetic instrument, as discussed earlier and in this review series by Lamina and Ward: it comprises around 85% of the population and therefore includes the majority of the Lp(a) concentration distribution, including a substantial fraction of the population with high Lp(a) concentrations as well as large and small apo(a) isoforms.Lp(a) concentrations show pronounced differences across ancestries, which we do not completely understand yet. There is evidence that SNPs with a strong influence on Lp(a) concentrations show a wide frequency variability across ancestries. Others also suggested a role of environmental exposures, respectively differing inflammatory burden. Identifying the unknowns might bring us closer to the full picture of cross-ancestry genetic regulation of the Lp(a) concentration differences, helping to dissect the relative contribution of genetics and environment in determining the Lp(a) trait across ancestries.We have currently a limited understanding of the metabolism and especially the catabolism of Lp(a). Searching for Lp(a)-regulating SNPs outside the LPA gene region might provide new evidence on genes involved in the machinery of Lp(a) catabolism. On the other hand, such studies are complicated by the considerable variance encoded by LPA. A successful search for modulators of Lp(a) metabolism outside of the LPA locus may require identifying individuals with peculiar Lp(a) concentrations that are not caused merely by different LPA SNPs. However, this requires a comprehensive knowledge about the functional LPA SNPs. Interestingly, such focused in-depth genetic studies on peculiar Lp(a) phenotypes are still rare.Genetic variants and how they regulate Lp(a) concentrations could identify targets for future drug interventions. As individuals with low Lp(a) do not present obvious health impairments, LPA is a clear early candidate for the therapeutic intervention using somatic gene editing. On the dawn of these therapies, frequent Lp(a)-lowering variants may indicate possible genome locations for safe gene editing. Conversely, the identification of regulatory polymorphisms that do increase Lp(a) rather than lower it can be highly informative as well, indicating targets that could be addressed by e.g. inhibitors.

6Genetic variants associated with certain Lp(a) concentration ranges were already in the past very helpful to support causality between Lp(a) and outcomes. Lp(a) was the first use case for Mendelian randomization studies in the 1990s, long before this term was coined (see review in this series). Nevertheless, the causality of Lp(a) has been debated for a long time until numerous genetic studies underscored the causality of Lp(a) concentrations using genetic variants strongly associated with high Lp(a) concentrations and subsequent cardiovascular disease. Several studies showed also that variants associated with low Lp(a) exert a protective effect on cardiovascular disease. On the other hand, it still takes some efforts to find the right genetic instrument to investigate a causal association between extremely low Lp(a) concentrations and diabetes mellitus. In the latter case, the use of the non-carrier status of rs10455872 is an insufficient genetic instrument, as discussed earlier and in this review series by Lamina and Ward: it comprises around 85% of the population and therefore includes the majority of the Lp(a) concentration distribution, including a substantial fraction of the population with high Lp(a) concentrations as well as large and small apo(a) isoforms.

7Lp(a) concentrations show pronounced differences across ancestries, which we do not completely understand yet. There is evidence that SNPs with a strong influence on Lp(a) concentrations show a wide frequency variability across ancestries. Others also suggested a role of environmental exposures, respectively differing inflammatory burden. Identifying the unknowns might bring us closer to the full picture of cross-ancestry genetic regulation of the Lp(a) concentration differences, helping to dissect the relative contribution of genetics and environment in determining the Lp(a) trait across ancestries.

8We have currently a limited understanding of the metabolism and especially the catabolism of Lp(a). Searching for Lp(a)-regulating SNPs outside the LPA gene region might provide new evidence on genes involved in the machinery of Lp(a) catabolism. On the other hand, such studies are complicated by the considerable variance encoded by LPA. A successful search for modulators of Lp(a) metabolism outside of the LPA locus may require identifying individuals with peculiar Lp(a) concentrations that are not caused merely by different LPA SNPs. However, this requires a comprehensive knowledge about the functional LPA SNPs. Interestingly, such focused in-depth genetic studies on peculiar Lp(a) phenotypes are still rare.

9Genetic variants and how they regulate Lp(a) concentrations could identify targets for future drug interventions. As individuals with low Lp(a) do not present obvious health impairments, LPA is a clear early candidate for the therapeutic intervention using somatic gene editing. On the dawn of these therapies, frequent Lp(a)-lowering variants may indicate possible genome locations for safe gene editing. Conversely, the identification of regulatory polymorphisms that do increase Lp(a) rather than lower it can be highly informative as well, indicating targets that could be addressed by e.g. inhibitors.

10LPA spans a >130 kb region on chromosome 6 (160,531,483–160,666,375 in the current human genome reference sequence hg38). Its evolutionary history produced a fairly regular and repetitive gene structure (Fig. 1C) with extensive intragenic and intergenic homologies. Each kringle consist of two short exons (160 and 182 bp long, except for KIV-6) highly homologous to the respective exons of the other kringles (>70% base identity between the different KIV; 98–100% between the KIV-2 exons). This homology extends into the intronic sequences with often >60% base identity in the first ±200 intronic bases (Supplemental Fig. 1). Furthermore, LPA is highly homologous to the (i) often neglected liver-expressed pseudogene LPA-like 2 (LPAL2), which flanks LPA upstream and contains sequences homologous to KIV-2, KIV-9, KIV-10 and the protease domain, (ii) plasminogen, which flanks LPA downstream and (iii) the plasminogen pseudogenes PLGLA, PLGLB1, PLGLB2 on chromosome 2. The LPA KIV and KV introns contain large, mostly quite diverged insertions of long interspersed nuclear element-1 retrotransposons (L1 LINEs). L1 LINEs are frequent mobile genetic elements that make up ≈ 20% of the human genome. They may induce structural variation by providing hotspots for recombination events and/or affect gene expression by containing regulatory elements. However, the specific significance of these intronic L1 elements for Lp(a) concentrations (if any) has not been explored yet.

11The KIV-2 CNV presents ≈40 different alleles, resulting in ≈1600 possible genotypes. This makes up to 70% of the gene hardly accessible for mutation detection and to an uncharted territory on the human genetic map. Each KIV-2 repeat is ≈ 5550 bp long, with some minor variability created by small indels and an intronic short tandem repeat (STR). The haplotype of three synonymous SNPs in KIV-2 exon 1 defines at least three types of KIV-2 units (KIV-2A, KIV-2B and KIV-2C; Supplemental Fig. 2), which differ also by >100 intronic differences. One such difference splits KIV-2A in two subtypes. No effect of these subtypes on Lp(a) concentrations was observed, but they have practical implications for research studies using next generation sequencing (NGS) (explained below). The current human genome reference sequence hg38 contains only six KIV-2 repeats (the third being KIV-2B). This creates considerable alignment difficulties for NGS data, which may contain up to 40 KIV-2 repeats.

12Different ancestries differ in the minor allele frequency (MAF) of these subtypes and in the percentage of KIV-2 repeats being type B or C. The resulting mutation level in the NGS data (also called intra-allelic frequency) ranges from <5% in Africans to >30% in East Asians. Within Europe, the low KIV-2B intra-allelic frequencies in Southern Europe and higher frequencies in Finns correlate well with the known intra-European Lp(a) gradient, likely reflecting differences in the genetic ancestry at the LPA locus.

13The transcriptional regulation of LPA is not fully understood, but involves proximal and distal elements. A 200 bp core promoter region is sufficient to drive LPA expression, but the overall activity of promoter fragments encompassing up to 1.5 kb 5’ region is low. Many transcription factors have been described to regulate LPA transcription but few have been validated in independent studies and/or by identification of the response elements. Additionally, multiple transcription factor modules with opposing effects have been reported in the promoter and functional promoter elements have been identified as far as ≈3.5 kb from the currently annotated transcription start site. In vitro activities may thus strongly depend on the assayed region.

14LPA is regulated also by two enhancer elements located ≈20 kb (region DH-III; chr6:160683862-160685654, hg38) and ≈30 kb (region DH-II; chr6: 160692643–160694671, hg38) upstream. They contain multiple response elements (Sp1, PPAR factors and estrogen) and induce the LPA promoter by 4–8 and 10–15 fold, respectively. Despite DH-II has a smaller effect on LPA transcription than DH-III, early population studies reported that it is more conserved than DH-III, but no systematic studies in large, more recent genetic resources have been reported yet. Intriguingly, the SNP rs186696265, which has the largest independent beta estimate (i.e. the effect on Lp(a) concentrations in mg/dL or nmol/L) identified so far but a low MAF (1.5%), is located between DH-II and DH-III. It increases Lp(a) by 49 mg/dL in an isoform-adjusted model and the odds for coronary artery disease (CAD) by 1.73-fold. This SNP has been identified by multiple GWAS on Lp(a) concentrations, plasma cholesterol phenotypes, triglycerides, cardiovascular phenotypes and even longevity but no function has been assigned yet.

15The isoform alone is not predictive of the Lp(a) concentration in a given individual. At individual level, same-sized isoforms may be associated with 200-fold different Lp(a) values. This can be seen in Fig. 2A, demonstrating that the range in Lp(a) concentrations, e.g. in carriers of small isoforms with 20 KIV repeats, is ranging from below 1 mg/dL to almost 200 mg/dL. This is observed also in individuals expressing only one isoform, indicating that this variance is not caused only by the often neglected contribution of the second isoform. Conversely, the inter-individual variation of alleles that are identical-by-descent within families is markedly smaller (only up to 3-fold). This implies that other genetic variants exist, which dissociate the commonly assumed inverse relationship between apo(a) isoform size and Lp(a) concentrations in a substantial manner. Especially early studies reported and characterized many examples of such “discordant phenotypes”. For example, Cohen et al. described a family where two same-sized alleles (23 KIV) that were distinguishable by an intronic SNP segregated with strikingly different concentrations (1–3 mg/dL vs. ≈22 mg/dL). Such discordant phenotypes can be seen quite often in small apo(a) isoform carriers who have low Lp(a) concentrations despite their small apo(a) isoform. This additional modification of the isoform effect is mirrored across populations and ancestries, with Africans showing much higher Lp(a) concentrations in every isoform group and Finns showing 50% lower Lp(a) in every isoform group than Central Europeans.

16Causal variants have been elusive for a long time. However, studies on an STR in the promoter (known as the “pentanucleotide repeat (PNR)”; hg38 chr6:160,665,587–160,665,631) provided interesting insights into the diversity of these phenotypes. The different PNR alleles with ≈6–12 repeat units (PNR6 to PNR12) are associated with very diverse Lp(a) phenotypes. PNR8 alleles are the most frequent ones and recapitulate the full range of Lp(a) and isoform correlations. PNR9 alleles are the human genome reference allele and occur mostly in the HMW isoform range. PNR10 alleles show two different subgroups: one subgroup follows the expected correlation of isoform and Lp(a) across the whole isoform range, while the second subgroup tags LMW isoforms with low Lp(a) (<≈15 mg/dL). Finally, PNR11 alleles tag isoforms with <24 KIV but unexpected low Lp(a) <5 mg/dL. The causal factors for the discordant phenotypes tagged by PNR10 and PNR11 have not been identified yet.

17Similar observations were made also with a DraIII restriction polymorphism found in only some KIV-2 units (KIV-D). The order of KIV-2 and KIV-D units creates nearly 30 individual restriction patterns, which likely mark different background haplotypes. Patterns 3 and 4 were linked to very defined isoform ranges (pattern 3: 27–32 KIV, pattern 4: 27–29 KIV), were associated with very low Lp(a) concentrations, and encompassed 24% and 6% of all null alleles in the study population. Our group recently identified the base change underlying the DraIII restriction polymorphism, but the SNP alone was not able to act as a proxy for these complex restriction patterns.

18Overall, these and other studies indicate the existence of a large diversity of haplotypes that are restricted to certain isoform ranges and are associated with very distinct Lp(a) concentrations. Some causal variants are discussed below.

19Many studies have aimed at identifying SNPs that causally affect Lp(a) concentrations. Since an individual discussion of all interesting LPA SNPs that have been described would go far beyond the focus of this review, Table 1 provides a comprehensive summary for those of particular interest and characterization. Fig. 3 and Fig. 4 report location and MAFs of selected SNPs.

20The relationship between SNPs and Lp(a) concentrations is not always straight-forward. Multiple, partially very recent, studies have provided fascinating insights into the complex genetic entanglements that govern the Lp(a) concentrations. These involve allelic association between SNPs and isoform ranges (Fig. 5), between different SNPs and between SNPs and STRs and span the complete gene body. They can confound associations and may even reverse the direction of an SNP effect (Fig. 6). The following sections will discuss some SNPs that exemplify these entanglements particularly well. Importantly, nearly all these SNPs affect directly Lp(a) concentrations. A noteworthy exception is rs1211014575 (KIV-10 Trp72Arg, p.Trp1685Arg), which abolishes the lysine binding capacity of KIV-10 without affecting Lp(a) concentrations, preventing OxPL accumulation on KIV-10 (Table 1).

21Many LPA SNPs are restricted to specific isoform ranges (Table 1 and Fig. 5A). This can enhance, limit or even mask the effects of functional SNPs. A clear-cut loss-of-function (LOF) mutation on an HMW allele may contribute little to the total plasma Lp(a) concentration since the concentration connected with an HMW allele is already low. Otherwise, a moderate LOF mutation on an LMW allele can have considerable effects. Furthermore, the overall expression level of the isoform may mask an opposite effect of an SNP (Fig. 6). The two regulatory SNPs rs1853021 and rs1800769 exemplify this particularly well. Further prime examples are described in the subsequent sections about the KIV-2 SNPs R21X, 4925G>A and 4733G>A and about the splice donor SNP rs41272114.

22Rs1853021 (also known as +93C/T or c.-49T>C; 5’ UTR SNP) is one of the earliest examples of linkage disequilibrium (LD) between a functional SNP and an isoform range. The T allele (which is the minor allele in the population, but the reference base in the human genome) creates an alternative start codon and reduces reporter gene expression, respectively LPA translation by 30–60%. Accordingly, an association with lower Lp(a) is readily observed in Africans, where the SNP occurs across the isoform range (Fig. 5B). On the contrary, no effect on Lp(a) concentrations is detectable in Whites, where the T-allele is preferentially associated with HMW isoforms (24–34 KIV; Fig. 5B). This markedly diminishes its impact on Lp(a) concentrations in Whites although it is mechanistically present (explained in Fig. 6A).

23In contrast, the 5’UTR SNP rs1800769 (also: 21G>A or +121G/A) has been linked to a 90% increased promoter activity. However, at a first glance, its effect on Lp(a) concentrations appears contradictory. In some studies in Europeans and Mexican Americans, it was found to be associated with lower Lp(a) plasma concentrations, while an association with increased allele-specific Lp(a) plasma concentrations has been reported in Europeans and African Americans by others. This perceived contradictions are caused again by the association of this SNP with very large isoforms in Whites but not in African Americans (>30 KIV in Whites but 24–30 KIV African Americans, Fig. 5B). This leads to a net negative effect in Whites. Accordingly, its negative effect on Lp(a) becomes positive also in Whites if the analyses are adjusted for the apo(a) isoforms. This suggests that rs1800769, while being associated with lower overall Lp(a), may be associated with higher-than-expected Lp(a) in HMW isoforms (Fig. 6B). Unfortunately, the direct investigation of these two interesting regulatory SNPs in contemporary sufficiently powered studies is hampered by the fact that the former was not contained in the 1000 Genomes (1000G) imputation panel, while the latter is not contained in the Haplotype Reference Consortium imputation panel.

24On the other hand, the LD of SNPs with isoforms can also be leveraged to ease Lp(a) research. Two SNPs have gained considerable attention (rs10455872 and rs3798220) as they have been reported to tag LMW isoforms and are thus used to circumvent laborious Western blotting in large studies. Despite being very useful at population scale, it is important to note that this correlation is far from perfect at individual level. In a large study with ≈6000 individuals, only about half of the individuals with LMW apo(a) isoforms carried also one of these SNPs.

25Two recent examples illustrated how SNP-SNP LD structures in LPA and, more specifically, sole reliance on R2 as LD measure can be misleading. Fig. 8 summarizes the basic mechanism behind these confounding observations. KIV-2 R21X is a low frequency nonsense SNP in KIV-2 (MAF≈2%). In a study in ≈11,000 individuals, we found for this variant ≈12 mg/dL lower Lp(a) concentrations and a preferential association with HMW isoforms. However, R21X did not provide additional information beyond the genotype of the LPA splice site mutation rs41272114 (discussed in next section). Indeed, we and others found that R21X occurs nearly exclusively on haplotypes that carry also the SNP rs41272114. This latter SNP is considerably more frequent (MAF ≈5%), which creates a misleading low R2 value with R21X. Together with the obvious functional consequence, this could have easily mislead researchers into assuming an independent function of R21X. Intriguingly, also a second very rare frameshift mutation in the KIV-2 has been observed recently on rs41272114-haplotypes.

26Similarly, two methylome-wide studies on Lp(a) independently identified a rare LPA promoter SNP (rs76735376) with a strong effect on LPA expression and Lp(a) concentrations (beta estimate +37 mg/dL/+113 nmol/L). The SNPs was restricted to isoforms 18–21 KIV repeats but R2 with the more frequent rs10455872 was <0.2. The adjustment for rs10455872 and isoforms cut the effect by 8-fold to +5.38 mg/dL. Here the situation is very similar as for rs41272114 and KIV-2 R21X: rs76735376 is located nearly exclusively on rs10455872-haplotypes but the large MAF difference (1% vs 9%) induce a misleadingly low R2 value (Fig. 8).

27Finally, a further layer of complexity is added by the fact that some rare functional SNPs may create subgroups within carriers of a more frequent SNP. These subgroups can present markedly different Lp(a) phenotypes than the parental haplotype. For example, rs10455872 is largely used as proxy for high Lp(a) but ≈5% of all carriers present low Lp(a) (<8.6 nmol/L). Said and colleagues identified a rare missense variant (rs41267813) in some rs10455872 carriers. This SNP lowers the median Lp(a) concentrations in individuals with both SNPs to as little as 7% of the reference group (rs10455872 only), explaining the rs10455872 carriers with the surprisingly low Lp(a) concentrations.

28Up to 30–50% of the population express only one isoform at detectable levels despite being heterozygous at DNA levels. Of course, this depends also on the sensitivity of the electrophoresis protocol since the amount of plasma applied to the SDS agarose gel depends on the Lp(a) concentration measured in plasma: in case of a high Lp(a) concentration with one major band responsible for the majority of Lp(a) in plasma, the second isoform might not be visible when the relative amount of the Lp(a) of this isoform applied to the gel falls under the detection limit. However, for the majority of probands with “null alleles”, two major mechanisms have been identified. On the one hand, large isoforms may fail to mature properly in the endoplasmatic reticulum and are degraded before being secreted. On the other hand, LOF variants can suppress mRNA or protein production. In contrast to several other examples described in this review, such variants may act independently from the background apo (a) isoform; however, depending on the apo(a) isoform with which they occur, the size of the Lp(a) lowering effect might be variable (Fig. 6B). The splice site mutation rs41272114 is the most frequent null allele mutation in Caucasians (MAF ≈3%) and explains ≈25% of all null alleles. It has been largely used as genetic instrument for Mendelian randomization studies to support causal associations between Lp(a) concentrations and multiple outcomes, despite its effect on Lp(a) is overall rather moderate (–5 to –17 mg/dL). This is due to its preferential association with HMW isoforms (like KIV-2 R21X). Generally, splice defects appear to be rather frequent in LPA. At least five different null allele SNPs that abolish splice sites have been described, as well as three splice modifier SNPs that lower Lp(a) by 80–90% (discussed in the next section) (Table 1 and Fig. 3).

29Because of the cysteine-rich structure of the kringle domains, Mooser et al. proposed, already in the mid 1990s, that apo(a) might be particularly susceptible to missense mutations that impair secretion by preventing correct folding. However, no such examples were known until Morgan et al. recently showed that both rs41259144 (p. Arg990Gln) in KIV-4 and rs139145675 (p.Arg1771Cys) in KV impair apo(a) secretion by preventing correct folding. More such SNPs have been proposed by others also in KIV-2, KIV-6 and KIV-9 and the same mechanism has been assumed also for two in-frame deletions in baboons (protease domain) and humans (KIV-2) (Table 1).

30Taken together, these examples suggests that null alleles might be collectively quite common and may occur throughout the apo(a) protein (Table 1). While SNPs in canonical splice sites are easy to spot, missense variants causing null alleles are harder to identify in-silico. A thoughtful screening approach has been proposed by Morgan et al., who prioritized LPA variants that cause plasminogen deficiency if occurring at homologous positions in plasminogen. Since phylogenetic approaches for variant effect prediction are poorly applicable to LPA due to the fact that LPA is missing in most species, the rationale of Morgan et al. might be a useful approach for further endeavors.

31Moreover, LPA SNPs may cause null alleles also without being clear-cut LOF mutations by simply lowering Lp(a) concentration below the assay detection limit if occurring on an allele with already low basal expression (“operational null alleles”).

32KIV-2 can encompass the majority of the LPA coding region. However, KIV-2 SNPs are not annotated in current SNP reference datasets like GnomAD or TOPMed because sequencing reads do not map uniquely, and the signal of genuine variants is diluted by reads from other KIV-2 units. An approach termed ‘batch sequencing’ has been devised to circumvent the mapping issues and is illustrated in Supplemental Fig. 3. It makes use of the homology between the KIV-2 units to amplify and sequence all repeats as amplicon mixture, align all data to one KIV-2 and detect SNPs alike somatic mutations. Since many of these SNPs are present only in one or few KIV-2 repeats out of up to 80 repeats, this results in a condition that resembles somatic mutations with one or a few KIV-2 repeats carrying the mutation mixed into a vast majority of repeats that do not carry the mutation. Its early practical application was hampered by the limited sensitivity of Sanger sequencing, but ultra-deep next generation sequencing now provides sufficient sensitivity to one mutant KIV-2 in up to 80 KIV-2 repeats. Although technically challenging, this method opened new avenues to study this otherwise almost inaccessible region (see Refs.). The first NGS batch sequencing study readily identified >500 KIV-2 SNPs in 123 individuals, including multiple missense, splice site and nonsense variants that were hiding in plain sight.

33The two splicing mutations KIV-2 SNPs 4925G>A and 4733G>A discovered by this approach stand out as they explain 5% and 10% of isoform-adjusted Lp(a) variance. In terms of variance explained, they thus represent the two most important genetic modifiers of Lp(a) concentrations besides the apo(a) isoform size. Indeed, they are remarkable prime examples that recapitulate many complexities of the genetics of LPA. Both SNPs are very frequent with MAFs of 13% and 22%, meaning that they are found in ≈22% and ≈38% of the European population, respectively. Both show widely varying MAF differences between various ancestries. 4925G>A is found mostly in isoforms at the boundary between LMW and HMW isoforms (≈19–25 KIV) (Fig. 2D). KIV-2 4925G>A decreases Lp(a) by ≈ 30 mg/dL in individuals with LMW apo(a) isoforms explaining ≈19% of isoform-adjusted Lp(a) variance; it decreases Lp(a) by ≈10 mg/dL in individuals with only HMW isoforms explaining ≈1.6% of isoform-adjusted Lp(a) variance. This variant also partially accounts for the astonishing drop in median Lp(a) concentrations at 23–25 KIV repeats (Fig. 2D). In a large German cohort, the median Lp(a) concentration of the 23 KIV isoform group increased from ≈10 to ≈23 mg/dL when 4925G>A carriers were excluded, which underscores the pronounced Lp(a)-lowering effect of this variant. Despite these strong effects on Lp(a) concentrations, its effect on Lp(a) variance at population scale is detectable only in isoform-adjusted regression models (R2 = 0.2% not adjusted vs. R2 = 6.1% isoform-adjusted). This phenomenon is even more notable in the HMW range (R2 = 0.02% vs. 1.6%) and is caused by the fact that within the HMW isoform range the SNP occurs on rather short HMW isoforms. In turn, these present relatively high Lp(a) and 4925G>A reduces their Lp(a) concentrations to a value that is close to the median of the overall HMW group. Therefore, the effect is not visible if the isoform background is not considered (see Fig. 6C).

34The second KIV-2 splicing SNP 4733G>A is associated with a more moderate Lp(a) reduction of –13 mg/dL when adjusted for apo(a) isoforms. However, it is very frequent in Whites and it is found across the whole isoform range with some preferential association to isoforms 24–33 KIV. Its high frequency makes it a major determinant of discordant Lp(a) phenotypes and the most important genetic factor affecting Lp(a) variance in Whites after the apo(a) isoforms. The moderate but lifelong Lp(a) reduction translates into an 9% lower hazard ratio for CAD. Mechanistically, it induces a splicing defect causing an in-frame deletion, which removes a structure-bearing cysteine residue and likely induces an secretion defect caused erroneous folding (alike rs41259144 and rs139145675). In a German population, compound heterozygosity with KIV-2 4925G>A (about 5% of the population) is associated with –32 mg/dL lower median Lp(a) and, as both alleles are blunted, almost no Lp(a) variance (Fig. 7).

35As both SNPs efficiently dissociate Lp(a) concentrations from isoform size (Fig. 1D), they might be interesting genetic tools to better dissect the interplay of Lp(a) concentrations and isoform size. Accordingly, a seminal study in >140,000 Icelanders, has recently used KIV-2 4925G>A to investigate whether LMW isoforms present an independent atherogenic potential. In line with our earlier study, the authors found that the atherogenic potential is conferred through the Lp(a) concentration meaning that subjects with LMW isoforms but low Lp(a) concentrations (caused e.g. by the 4925G>A variant) were not at an increased CAD risk. As discussed in another review of this series, this demonstrates that the Lp(a) concentration has a stronger information content than certain SNPs since it comprises the entire genetic information as well as non-genetic (environmental) factors.

36Ancestry is a major modifier of Lp(a) concentrations and a two-fold variation is observed even within Europe. These differences extend also to MAFs, SNP haplotypes and association with isoform ranges (Table 1, Figs. 4 and 5B). For example, the LOF variant rs41272114 is twice as frequent in Admixed Americans than in Europeans (8% vs. 3% MAF) and a MAF of even 18% was observed in a small sample of 85 Peruvians from the 1000G project. Conversely, the splice site mutation rs143431368 is very rare globally (≪1% MAF) but frequent in Finns (MAF≈5%). The high impact KIV-2 SNPs discussed before range from to 0–22% MAF globally. Also the LMW isoform-tagging SNP rs3798220 presents a particularly pronounced heterogeneity across ancestries. It is absent in Africans, rare in Europeans (MAF≈2%), moderately frequent in South Asians (MAF = 12%) and very frequent in Hispanics (up to 42% MAF). However, it does not tag LMW isoforms in Asians and Hispanics. In Europeans, it is also in partial LD with the strong GWAS hit rs140570886 (associated with +43 mg/dL higher isoform-adjusted Lp(a); Table 1), which in turn presents nine-fold higher MAF in Admixed Americans and Latinos than in all other continental groups of GnomAD.

37Many more such examples exist and leveraging these differences in comparative cross-ancestry genetic studies might help pinpoint functional SNPs. However, it needs to be considered that SNPs may segregate with very different isoform ranges between populations (Fig. 6B and Table 1). As already pointed out by Utermann, this can mislead fine-mapping efforts. In cross-ancestry studies, effect heterogeneity between ancestries is often interpreted as a sign that a GWAS hit is actually rather a proxy SNP than a causal SNP. However, as discussed above, even genuine functional SNPs may present considerably different effects if occurring in different isoform ranges across populations. While intensifying genetic studies in non-Whites will definitely be fruitful, proper care is required. Isoform data is still rare, especially for large non-White populations. Considerable efforts will thus be necessary to map the association between SNPs alleles and apo(a) isoforms across ancestries on a large scale and in a standardized manner. Western blotting is very laborious, which has precluded such endeavors until now. Recent advances in haplotype phasing and imputation algorithms are, however, opening new avenues to approach this shortcoming.

38Many candidate gene, sequencing and GWAS studies have searched for genetic variants that modulate Lp(a) concentrations. Notwithstanding the many entanglements discussed above, these studies have been very successful and have identified dozens of independent SNPs (tagging >2000 significant variants) in a ≈2 megabases region around LPA. These SNPs have been recently used to construct effective genetic risk scores that explain up to ≈70% of Lp(a) concentration variance. They represent valid genetic surrogates for direct Lp(a) quantification, with similar distribution and similar association to cardiovascular outcomes. While for the time being a direct Lp(a) quantification is cheap and easy, these scores may become efficient screening tools in the future as the availability of genomic data in clinical care is constantly increasing.

39Nevertheless, still some gap exists to reach the ≈90% variance that is explained by the complete LPA locus. Further informative variants might be masked by complex associations with isoforms or non-additive epistatic effects. So far, only one isoform-adjusted GWAS (n > 13,000) has been performed but, interestingly, the isoform-adjusted regression model still detected 30 independent hits representing 1961 SNPs in the LPA gene locus. These SNPs might tag further functional SNPs that create discordant phenotypes (e.g. the GWAS hit rs75692336 is a proxy SNP for KIV-2 4925G>A). However, disentangling functional SNPs from simple isoform-tagging SNPs will be a major task as these two roles are not mutually exclusive (as exemplified by KIV-2 4925G>A). They are also modified by the isoform sizes and by non-linear epistatic effects. These are not captured well by standard regression models. For example, Zeng et al. found that the effect of rs140570886 on Lp(a) and CAD depends on the haplotype of rs1800769 (promoter) and rs9458001 (enhancer), while, vice versa, only the minor allele of rs140570886 enables a strong effect of rs1800769 on Lp(a).

40Despite the value of apo(a) isoform information in Lp(a) studies, apo (a) Western blotting is not feasible at very large scale. This limits the scale of in-depth studies. Two very recent studies have provided a major leap forward to address this issue. Both studies estimated the total KIV-2 number at DNA level by using the NGS coverage and apportioned it to the two parental haplotypes using advanced phasing procedures. This gave a “diploid KIV-2 content” representing an imputed KIV-2 genotype (iKIV-2). In Mukamel et al., this iKIV-2 explains 61% of Lp(a) variance in the UK Biobank, which is remarkably close to the upper bound of variance explained by isoforms measured directly by Western blotting (30–70%). This is even more impressive when considering that it does not take into account non-expressed alleles. When comparing the iKIV-2 from whole exome sequencing data to direct KIV-2 sizing using optical mapping (a method similar to Fiber-FISH), the authors observed a good correlation (R2 = 0.67).

41The authors subsequently used the phasing algorithm to assign all LPA SNPs to the background iKIV-2 allele and finally restricted the analyses to 24,969 heterozygous null allele carriers. This removed the confounding effect of the second allele and produced a large haploid Lp(a) dataset. While this concept had been proposed earlier, only current data from the UK Biobank allows to apply it at large scale. Using stepwise conditional analysis the authors finally identified 23 very promising LPA sequence variants with MAFs from 0.01% to 28% that likely causally affect Lp(a) production by the respective allele. These include known null alleles, the functional KIV-2 SNPs 4925G>A, 4733G>A and R21X, and the regulatory SNPs rs1853021 and rs1800769. When accounting also for cis-epistatic and non-linear effects, these SNPs raised the explained Lp(a) variance to an impressive 83%, respectively 90% of heritable variance). 43% of all European haplotypes presented at least one modulator SNP but only 13% of the African alleles. The frequency differences in these 23 SNPs largely explained the cross-ancestry differences in Lp(a). Provided replication of these findings and thorough validation against directly measured isoforms, the wide implementation of these algorithms promise unprecedented opportunities to scale up isoform-adjusted analyses and possibly even accurately predict Lp(a) from genetic data. Unfortunately, to date these algorithms have not been distributed as widely applicable bioinformatic tool.

42Given the strong impact of LPA, other major regulators have been elusive for a long time. An exception is the APOE2 allele which lowers Lp(a) markedly by 3 mg/dL per copy. A GWAS in the UK Biobank finally identified 37 additional loci, but their effects were mostly very small (<5% of the top hit LPA rs10455872). Most intriguingly, among the many candidate receptors for Lp(a) that had been proposed (reviewed in Ref. and in this series), only a minor effect of LDLR (≈2% of the effect of the top hit rs10455872) has been identified by those GWAS, while a previously reported association of SCARB1 variants with high Lp(a) was not replicated. On the other hand the effect size in GWAS may not necessarily reflect the physiological relevance of a gene. Therefore, it will be interesting to see whether these associations will still shed new light on the machinery involved in the metabolism and catabolism of Lp(a). The detection of genes that have been described before (LDLR, APOH) or are known from pharmacological intervention (CETP, PCSK9, LDLR) may be promising. However, it will not be straightforward to disentangle genes which show up in GWAS for both phenotypes, LDL cholesterol and Lp(a), since the cholesterol content of Lp(a) is also included in the LDL cholesterol measurement.

43Advances in genomic technologies, bioinformatics and statistical genetics have generated considerable insights into how SNPs regulate the Lp(a) concentrations beyond the apo(a) isoforms. What do these complex genetic studies tell about Lp(a) biology? The intrinsic biological role of Lp(a) is still unknown. It acts as a preferential carrier of OxPL (reviewed in Ref. and in this series), but this could be rather an acquired function than its native role (given the large number of people with very low to even null Lp(a)). The GnomAD dataset with >120,000 exomes reports a very high mutational burden for LPA, with an LOF ratio (observed LOF number/expected LOF number under neutrality) of even 1.3. It is not clear whether and how this can be reconciled with a critical biological function, which has recently led to the intriguing speculation that LPA might indeed be a large translated pseudogene lacking an intrinsic physiological function. While definitely provocative, this may fit to the genetic data and such a rationale may have important implications on the direction of further Lp(a) research, e.g. the search for specific receptors. On the other hand, given the existence of >15,000 pseudogenes in the human genome, a pseudogene with such a pronounced pathophysiological function and being translated to a large protein might be very unusual.

44It should be pointed out that the majority of genetic data on LPA available to date has been generated in Caucasians. It is unknown whether the LOF frequency is similarly high also in other populations, especially in Africans, which present higher median Lp(a). If not, it would be interesting to interrogate whether this reflects an unknown selective pressure. If yes, it would be intriguing to get to know what other factors are counterbalancing those LOFs.

45Indeed, most SNPs, for which a causal mechanism has been identified, lower Lp(a), with at least ten clear-cut LOF mutations identified so far and many more postulated (Table 1 and references). Every isoform group presents many individuals with Lp(a) that is considerably lower than the group median Lp(a), but a similar amount of people is seen when the isoform-associated Lp(a) deviates at least as much towards high Lp(a). Some SNPs that raise Lp(a) have been found (mostly by association studies), but none present a similarly large effect as the KIV-2 SNPs 4733G>A and 4925G>A. Only for rs1800769 some direct functional impact on high Lp(a) concentrations has been shown. It is fully unclear whether the genetic architecture of high Lp(a) is similar to that of low Lp(a). The complex genetics of Lp(a) may still hide some surprises.